Anda di halaman 1dari 523

UPSC Civil Services Exam - Mathematics

Solutions to Questions - Linear Algebra, Algebra, Calculus, Complex


Analysis
The Union Public Service Commission (UPSC) is a Government of India organization that is responsible for the selection of
personnel for the Central Civil Services such as the Indian Administrative Service (IAS), the Indian Foreign Service (IFS), the
Indian Police Service (IPS) and many others. It conducts several examinations annually to select candidates, the most important of
which is the Civil Services Examination, usually taken by candidates after graduation with a bachelor's or master's degree.
Candidates choose 2 subjects out of a list, and are required to show competence in these at a graduate level.

Here we have provided solutions to questions in the Mathematics Exam over the years. Only solutions to the Linear Algebra,
Algebra, Calculus and Complex Analysis sections are provided. This page is under construction, please check the change log for
updates.

We hope that these solutions will be useful to candidates preparing for the Civil Services, and will also be useful for anyone
studying these subjects. Please drop us an email at sunderlal.chd (at) gmail.com if you found them useful. We would be glad to
hear comments and criticisms, suggestions for improvement, and any error reports. If you have a different or better solution to
any question, we would be glad to incorporate it, and will credit it to you in the notes.

Dr. Sunder Lal Vineet Gupta


Retired Professor of Mathematics Software Engineer
Panjab University Google India Inc.
Chandigarh, India. Bangalore, India.

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License. You
are free to use it for any personal, classroom or other non-commercial purpose, but must seek prior
permission for commercial use.

http://sites.google.com/site/sunderlalchd/home
UPSC Civil Services Main 1979 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 27, 2009

Question 1(a) State and prove Cayley-Hamilton Theorem.

Solution. See 1987, question 5(a).

Question 1(b) Reduce the quadratic expression x2 + 2y 2 + 2z 2 + 2xy + 2xz to the canonical
form.

Solution. Completing the squares: given form = (x + y + z)2 + (y − z)2 . Put X =


x + y + z, Y = y − z, Z = z to get the canonical form = X 2 + Y 2 . The expression is positive
semi-definite.
Alternate solution: See 1981 question 1(b) for an alternate method of canonicalization.

Question 2(a) Find the elements p, q, r such that the product BA of the matrices
   
1 2 1 1 0 0
A =  4 1 2  , B = p 1 0 
−10 2 4 q r 1

is of the form  
a1 b1 c1
BA =  0 b2 c2 
0 0 c3
Hence solve the set of equations Ax = y, where x is the column vector (x1 , x2 , x3 ), and y is
the column vector (0, 8, −4).

1
Solution.
   
1 2 1 a1 b1 c1
BA =  p + 4 2p + 1 p + 2  =  0 b2 c2 
q + 4r − 10 2q + r + 2 q + 2r + 4 0 0 c3

Thus p + 4 = 0, q + 4r − 10 = 0, 2q + r + 2 = 0 ⇒ p = −4, r = 22
7
, q = − 18
7
.
Now solving Ax = y is the same as solving BAx = By because |B| 6= 0.
       
1 2 1 x1 1 0 0 0 0
0 −7 −2 x2  =  −4 1 0  8  =  8 
54
0 0 7
x3 − 18
7
22
7
1 −4 148
7

Thus 54 x = 148
7 3 7
⇒ x3 = 27 74
. −7x2 − 2x3 = 8 ⇒ −7x2 = 2x3 + 8 = 364
27
, so x2 = − 52
27
.
104 74
x1 + 2x2 + x3 = 0 ⇒ x1 = 27 − 27 = 10
9
.
10 52 74
Thus x1 = 9 , x2 = − 27 , x3 = 27 is the required solution.

Question 3(a) If S and T are subspaces of a finite dimensional vector space, then show
that
dim(S + T ) = dim S + dim T − dim(S ∩ T )

Solution. See 1988, question 1(b).

Question 3(b) Determine the value of a for which the following system of equations:

x1 + x2 + x3 = 2
x1 + 2x2 + x3 = −2
x1 + x2 + (a − 5)x3 = a

has (1) a unique solution (2) no solution.



1 1 1

Solution. 1 2 3 = 2a − 10 − 3 − a + 5 + 3 − 1 = a − 6
1 1 a − 5

1. If a − 6 6= 0 i.e. a 6= 6, the system has a unique solution.

2. If a = 6, the system is inconsistent as the third equation becomes x1 + x2 + x3 = 6,


which is inconsistent with the first. So there is no solution.

2
Paper II
Question 4(a) Prove that any two finite dimensional vector spaces of the same dimension
are isomorphic.
Solution. See 1987 question 4(b).

Question 4(b) Define the dual space of a finite dimensional vector space V and show that
it has the same dimension as V.
Solution. Let V ∗ = {f : V −→ R, f a linear transformation}. Then V ∗ is a vector space
for the usual pointwise addition and scalar multiplication of functions: for all v ∈ V and all
α ∈ R, (f + g)(v) = f (v) + g(v), (αf )(v) = αf (v).
Let P a basis for V. Define n linear functionals v1∗ , . . . , vn∗ by vi∗ (vj ) = δij ,
v1 , . . . , vn be P
and vi∗ ( j=1 αj vj ) = nj=1 αj vi∗ (vj ) = αi .
n

Then v1∗ , . . . , vn∗ are linearly independent — ni=1 αi vi∗ = 0 ⇒ ( ni=1 αi vi∗ )(vj ) = αj =
P P
0, 1 ≤ j ≤ n. Pn Pn
∗ ∗ ∗ ∗ ∗ ∗
Pn v1 , . . . , v n generate V — if f ∈ V , then f = i=1 f (vi )v i . Clearly ( i=1 f (vi )vi )(vj ) =

i=1 f (vi )vi (vj ) = f (vj ), so the two sides agree on v1 , . . . , vn , and hence by linearity on all
of V.
Thus v1∗ , . . . , vn∗ is a basis of V ∗ , so dim V ∗ = dim V . V ∗ is called the dual of V.

Question 4(c) Show that every finite dimensional inner product space V over the field of
complex numbers has an orthonormal basis.
Solution. Let w1 , . . . , wn be a basis of V. We will convert it into an orthonormal basis of
V by the Gram-Schmidt orthonormalization process.
Starting with i = 1, define
i−1
X hwi , vj i
v i = wi − vj
j=1
||vj ||2

Each vi is non-zero, as otherwise wi can be written as a linear combination of wj , j < i,


but w1 , . . . , wn are linearly independent.
Now we can prove by induction on i that hvi , vj i = 0 for all j < i — this is enough
because hvi , vj i = hvj , vi i . Suppose it is true for all k < i. Then hvi , vj i = hwi , vj i −
Pi−1 hwi ,vm i hwi ,vj i
m=1 ||vm ||2 hvm , vj i = hwi , vj i − ||vj ||2 hvj , vj i = 0. Thus v1 , . . . , vn are mutually orthog-
onal. They are linearly independent, as ni=1 ai vi = 0 ⇒ h ni=1 ai vi , vj i = aj ||vj ||2 = 0 ⇒
P P
aj = 0 for all i ≤ j ≤ n. Replacing vi by ||vvii || gives us an orthonormal basis of V.

Question 5(a) Define the rank and nullity of a linear transformation. If V and W are
finite dimensional vector spaces over a field, and T is a linear transformation of V into W,
prove that
rank T + nullity T = dim V
Solution. See 1998 question 3(a).

3
Question 5(b) Define a positive definite form. State and prove a necessary and sufficient
condition for a quadratic form to be positive definite.

Solution. See 1992 question 2(c).

Question 5(c) Show that the mapping T : R3 −→ R3 defined by T (x, y, z) = (x − y +


2z, 2x + y, −x − 2y + 2z) is a linear transformation. Find its nullity.

Solution.

T (ax + by) = T (ax1 + by1 , ax2 + by2 , ax3 + by3 )


= (ax1 + by1 − ax2 − by2 + 2ax3 + 2by3 , 2ax1 + 2by1 + ax2 + by2 ,
−ax1 − by1 − 2ax2 − 2by2 + 2ax3 + 2by3 )
= aT (x1 , x2 , x3 ) + bT (y1 , y2 , y3 )

Thus T is a linear transformation.


If (x, y, z) ∈ the null space of T , then x − y + 2z = 0, 2x + y = 0, −x − 2y + 2z = 0 ⇒ y =
−2x, z = − 3x2
. Thus the null space is {(x, −2x, − 3x2
) | x ∈ R} = {(2, −4, −3)x | x ∈ R}.
Thus nullity T = 1.

4
UPSC Civil Services Main 1980 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 27, 2009

Question 1(a) Define the rank of a matrix. Prove that a system of equations Ax = b is
consistent if and only if rank (A, b) = rank A, where (A, b) is the augmented matrix of the
system.

Solution. See 1987 question 3(a).


 
2 1
Question 1(b) Verify the Cayley Hamilton Theorem for the matrix A = , and
1 2
hence find A−1 .

Solution. The Cayley Hamilton theorem is — Every matrix A satisfies


its characteristic
x − 2 −1
equation |xI − A| = 0. In the current problem, |xI − A| = = x2 − 4x + 4 − 1 =
−1 x − 2   
2 2 2 2 1 2 1
x − 4x + 3. Thus we need to show that A − 4A + 3I = 0. Now A = =
          1 2 1 2
5 4 2 5 4 2 1 1 0 0 0
, so A − 4A + 3I = −4 +3 = , verifying the Cayley
4 5 4 5 1 2 0 1 0 0
Hamilton Theorem.  
2 1 1 0
A2 − 4A + 3I = 0 ⇒ A(A − 4I) = −3I ⇒ A−1 = − 3 (A − 4I) = − 3 (
1 1 −4 )=
 2 1 2 0 1
− 13

3 .
− 13 23

Question 2(a) Prove that if P is any non-singular matrix of order n, then the matrices
P−1 AP and A have the same characteristic polynomial.

Solution. The characteristic polynomial of P−1 AP is |xI−P−1 AP| = |xP−1 P−P−1 AP| =
|P−1 ||xI − A||P| = |xI − A| which is the characteristic polynomial of A.

1
 
3 4
Question 2(b) Find the eigenvalues and eigenvectors of the matrix A = .
4 −3
 
3 4 3 − λ 4
Solution. The characteristic equation of A = is = 0 ⇒ −(9 −
4 −3 4 −3 − λ
λ2 ) − 16 = 0 ⇒ λ2 − 25 = 0 ⇒ λ = 5, −5.   
−2 4 x1
If (x1 , x2 ) is an eigenvector for λ = 5, then = 0 ⇒ 2x1 − 4x2 = 0 ⇒
4 −8 x2
x1 = 2x2 . Thus (2x, x), x ∈ R, x 6= 0 gives all eigenvectors for λ = 5, in particular, we can
take (2, 1) as an eigenvector for λ = 5.   
8 4 x1
If (x1 , x2 ) is an eigenvector for λ = −5, then = 0 ⇒ 4x1 + 2x2 = 0 ⇒ x2 =
4 2 x2
−2x1 . Thus (x, −2x), x ∈ R, x 6= 0 gives all eigenvectors for λ = −5, in particular, we can
take (1, −2) as an eigenvector for λ = −5.
Question 3(a) Find a basis for the vector space V = {p(x) | p(x) = a0 + a1 x + a2 x2 } and
its dimension.
Solution. Let f1 = 1, f2 = x, f3 = x2 , then f1 , f2 , f3 are linearly independent, because
α1 f1 + α2 f2 + α3 f3 = 0 ⇒ α1 + α2 x + α3 x2 = 0(zero polynomial) ⇒ α1 = α2 = α3 = 0.
f1 , f2 , f3 generate V because p(x) = a0 + a1 x + a2 x2 = a0 f1 + a1 f2 + a2 f3 for any p(x) ∈ V.
Thus {f1 , f2 , f3 } is a basis for V and its dimension is 3.
Question 3(b) Find the values of the parameter λ for which the system of equations
x + y + 4z = 1
x + 2y − 2z = 1
λx + y + z = 1
will have (i) unique solution (ii) no solution.
 −1  
1 1 4 1
Solution. The system will have the unique solution given by 1 2 −2
   1 if
λ 1 1 1
1 1 4

1 2 −2 = 1(2 + 2) + 4(1 − 2λ) − 1(1 + 2λ) 6= 0. Thus 4 + 4 − 8λ − 1 − 2λ 6= 0 ⇒ λ 6= 7 .
10
λ 1 1
7
When λ = 10
, the system is
x + y + 4z = 1
x + 2y − 2z = 1
7x + 10y + 10z = 10
This system has no solution as it is inconsistent: 4(x+y+4z)+3(x+2y−2z) = 7x+10y+10z =
7, but the third equation says that 7x + 10y + 10z = 10. Thus there is a unique solution if
7 7
λ 6= 10 , and no solution if λ = 10 .

2
Paper II

Question 3(c) If V is a finite dimensional vector space and M is a subspace of V, then


show that each vector x ∈ V can be uniquely expressed as x = y + z, where y ∈ M and
z ∈ M⊥ , the orthogonal complement of M.

Solution. Let P v1 , . . . , vm be any orthonormal basis of M, where m = dim M. Given


m
x ∈ V, let y = i=1 hx, vi i vi , and z =Px − y. Clearly y ∈ M, and x = y + z. Now
hz, vi i = hx, vi i − hy, vi i = hx, vi i − m j=1 hx, vj i hvj , vi i = hx, vi i − hx, vi i = 0. So
hz, vi i = 0, i = 1, . . . , m ⇒ hz, mi = 0 for every m ∈ M, so z ∈ M⊥ .
Now if x = y0 +z0 , then y−y0 = z0 −z. But y−y0 ∈ M, z0 −z ∈ M⊥ , so hy − y0 , z0 − zi =
0 ⇒ hy − y0 , y − y0 i = 0 ⇒ ||y − y0 || = 0 ⇒ y − y0 = 0 ⇒ z0 − z = 0. Thus y = y0 , z = z0
and the representation is unique.

Question 3(d) Find one characteristic value and corresponding characteristic vector for
the operators T on R3 defined as

1. T is a reflection on the plane x = z.

2. T is a projection on the plane z = 0.

3. T (x, y, z) = (3x + y + z, 2y + z, z).

Solution.

1. T (x, y, z) = (z, y, x) because the midpoint of (x, y, z) and (z, y, x) lies on the plane
x = z. T (1, 0, 0) = (0, 0, 1), T (0, 1, 0) = (0, 1, 0), T (0, 0, 1) = (1, 0, 0). Thus it is clear
that 1 is an eigenvalue, and (0, 1, 0) is a corresponding eigenvector.

2. T (1, 0, 0) = (1, 0, 0), T (0, 1, 0) = (0, 1, 0), T (0, 0, 1) = (0, 0, 0). Clearly 1 is an eigen-
value with (1, 0, 0) or (0, 1, 0) as eigenvectors.

3. T (1, 0, 0) = (3, 0, 0), T (0, 1, 0) = (1, 2, 0), T (0, 0, 1) = (1, 1, 1). Clearly (1, 0, 0) is an
eigenvector, corresponding to the eigenvalue 3.

3
UPSC Civil Services Main 1981 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 27, 2009

Question
 1(a) State and prove the Cayley Hamilton theorem and verify it for the matrix

2 3
A= . Use the result to determine A−1 .
3 5

Solution. See 1987 question 5(a) for the


Cayley Hamilton
theorem.
x − 2 −3
The characteristic equation of A is = 0, or (x − 2)(x − 5) − 9 = 0 ⇒
−3 x − 5
x2 − 7x +1 = 0. The
 Cayley
 Hamilton
2
 theorem implies that A − 7A + I = 0.
2 3 2 3 13 21
A2 = = .
3 5 3 5  21 34      
2 13 21 2 3 1 0 0 0
Now A − 7A + I = −7 + =
21 34 3 5 0 1 0 0
2 −1
So thetheorem
 is  verified.
  A − 7A +  I = 0 ⇒ (A − 7I)A = −I ⇒ A = 7I − A. Thus
1 0 2 3 5 −3
A−1 = 7 − = .
0 1 3 5 −3 2

Question 1(b) Let Q be the quadratic form

Q = 5x21 + 5x22 + 2x23 + 8x1 x2 + 4x1 x3 + 4x2 x3

By using an orthogonal change of variables reduce Q to a form without the cross terms i.e.
with terms of the form aij xi xj , i 6= j.
 
5 4 2
Solution. The matrix of the qiven quadratic form Q is A = 4 5 2.
2 2 2

1
The characteristic polynomial of A is

5 − λ 4 2

4
5−λ 2 = 0
2 2 2 − λ
⇒ (5 − λ)(5 − λ)(2 − λ) − 4(5 − λ) − 4(8 − 4λ) + 16 + 16 − 4(5 − λ) = 0
⇒ (λ2 − 10λ + 25)(2 − λ) − 20 + 4λ − 32 + 16λ + 12 + 4λ = 0
⇒ −λ3 + 12λ2 + λ(−25 + 4 + 16 + 4 − 20) + 50 − 20 − 32 + 12 = 0
⇒ λ3 − 12λ2 + 21λ − 10 = 0

Thus the eigenvalues are λ = 1, 1, 10. Let (x1 , x2 , x3 ) be an eigenvector for λ = 10, then
  
−5 4 2 x1 −5x1 + 4x2 + 2x3 = 0 (i)
 4 −5 2  x2  = 0 ⇒ 4x1 − 5x2 + 2x3 = 0 (ii)
2 2 −8 x3 2x1 + 2x2 − 8x3 = 0 (iii)

Subtracting (ii) from (i), we get −9x1 + 9x2 = 0 ⇒ x1 = x2 ⇒ x1 = 2x3 . Thus taking
x3 = 1, we get (2, 2, 1) as an eigenvector for λ = 10.
Let (x1 , x2 , x3 ) be an eigenvector for λ = 1, then
  
4 4 2 x1
4 4 2 x2  = 0 ⇒ 4x1 + 4x2 + 2x3 = 0
2x1 + 2x2 + x3 = 0
2 2 1 x3

Take x3 = 0, x1 = 1 ⇒ x2 = −1 to get (1, −1, 0) as an eigenvector for λ = 1. Take


x1 = x2 = 1 ⇒ x3 = −4 to get (1, 1, −4) as another eigenvector for λ = 1, orthogonal to the
first.
Thus  
√2 √1 √1
 √29 2
− √12
18
√1 
O=

 9 18
√1 0 − √418
9
 
10 0 0
is an orthogonal matrix such that O0 AO = O−1 AO =  0 1 0. If (X1 , X2 , X3 ) are new
    0 0 1
x1 X1
variables, then x2 = O X2  takes Q(x1 , x2 , x3 ) to 10X12 + X22 + X32 .
  
x3 X3
Note: If the orthogonal transformation was not required for the diagonalization, we

2
could do it easily by completing squares:

5x21 + 5x22 + 2x23 + 8x1 x2 + 4x1 x3 + 4x2 x3


8 4
= 5[x21 + x1 x2 + x1 x3 ] + 5x22 + 2x23 + 4x2 x3
5 5
4 2 2 16 4 16
= 5[x1 + x2 + x3 ] + (5 − )x22 + (2 − )x23 + (4 − )x2 x3
5 5 25 5 5
4 2 2 9 2 4 6 2
= 5[x1 + x2 + x3 ] + (x2 + x2 x3 ) + x3
5 5 5 9 5
4 2 2 9 2 2 10 2
= 5[x1 + x2 + x3 ] + [x2 + x3 ] + x3
5 5 5 9 9
9 10
= 5X 2 + Y 2 + Z 2
5 9
where X = x1 + 45 x2 + 52 x3 , Y = x2 + 29 x3 , Z = x3 , or x3 = Z, x2 = Y − 92 Z, x1 = X − 45 Y − 29 Z.

Question 2(a) Define a vector space. Show that the set V of all real-valued functions on
[0, 1] is a vector space over the set of real numbers with respect to the addition and scalar
multiplication of functions.

Solution. See 1984 question 4(a).

Question 2(b) If zero is a root of the characteristic equation of a matrix A, show that the
corresponding linear transformation cannot be one to one.

Solution. If zero is a root of |A − λI| = 0, the characteristic equation of A, then 0 is an


eigenvalue of A, so there is a non-zero eigenvector x such that Ax = 0, thus A is not 1-1.

Question 2(c) Show that a linear transformation T from a Euclidean space V to V is


orthogonal if and only if the matrix corresponding to it with respect to any orthonormal basis
is orthogonal.

Solution. T : V −→ V is said to be orthogonal if hT(u), T(v)i = hu, vi for any u, v ∈ V.


Lemma 1. T is orthogonal iff T takes an orthonormal basis to an orthonormal basis.
Proof: Let {v1 , v2 , . . . , vn } be an orthonormal basis. Then

1. hT(vi ), T(vj )i = hvi , vj i = 0 if i 6= j

2. hT(vi ), T(vi )i = hvi , vi i = 1

3. If ni=1 αi T(vi ) = 0, then h ni=1 αi T(vi ), vj i = αj = 0 for all j, so T(vi ) are linearly
P P
independent.

3
Thus T(v1 ), . . . , T(vn ) form an orthonormal basis. Pn
Pn Conversely, let T(v 1 ), . .
Pn. , T(v n ) be an orthonormal basisPn of V. Let v =
Pn i=1 αi vi , w =
Pi=1 βi vi , then hv, wi = i=1 αi βi and hT(v), T(w)i = h i=1 αi T(vi ), i=1 βi T(vi )i =
n
i=1 αi βi . Thus hT(v), T(w)i = hv, wi , so T is orthogonal.
Lemma 2. Let T∗ be defined by hT(v), wi = hv, T∗ (w)i . Then T∗ is a linear trans-
formation, and T is orthogonal iff T∗ T = TT∗ = I.
Proof: The fact that T∗ is a linear transformation can be easily checked. If T is
orthogonal, then hv, T∗ T(w)i = hT(v), T(w)i = hv, wi , so T∗ T = I. From this and the
fact that T is 1-1, it follows that TT∗ = I.
Lemma 3. If the matrix of T w.r.t. the orthonormal basis {v1 , v2 , . . . , vn } is A = (aij ),
then the matrix of T∗ P is the transpose, i.e. (aji ). P
n ∗ n ∗
Proof: T(vi )P= j=1 aij vj . Let T (vi ) = j=1 bij vj . Now bij = hT (vi ), vj i =
hvi , T(vj )i = hvi , nk=1 ajk vk i = aji . Since TT∗ = I, A0 A = AA0 = I, so A is orthogonal.
The converse is also obvious now.

Question 3(a) Investigate for what values of λ and µ does the following system of equations

x+y+z = 6
x + 2y + 3z = 10
x + 2y + λz = µ

have (1) a unique solution (2) no solution (3) an infinite number of solutions?

Solution.

1 1 1

1. A unique solution exists when 1 2 3 6= 0, whatever µ may be. Thus 2λ−6−(λ−3) 6=
1 2 λ
0 ⇒ λ 6=
 3. Thus−1 λ 6=
forall  3 and for all µ we have a unique solution given by
x 1 1 1 6
y  = 1 2 3  10
z 1 2 λ µ
2. A unique solution does not exist if λ = 3. If µ 6= 10, then the second and third
equations are inconsistent. Thus if λ = 3, µ 6= 10, the system has no solution.

3. If λ = 3, µ = 10, then the system is x+y+z = 6, x+2y+3z = 10. The coefficient matrix
is of rank 2, so the space of solutions is one dimensional. y + 2z = 4 ⇒ y = 4 − 2z,
and thus x = 2 + z. The space of solutions is (2 + z, 4 − 2z, z) for z ∈ R.

4
Question 3(b) Let (xi , yi ), i = 1, . . . , n be n points in the plane, no two of them having the
same abscissa. Find a polynomial f (x) of degree n − 1 which takes the value f (xi ) = yi , 1 ≤
i ≤ n.

Solution. Let f (x) = a0 + a1 x + . . . + an−1 xn−1 . We want to determine a0 , . . . , an−1 such


that       
a0 1 x1 . . . xn−1 1 a 0 y1
 a1  1 x2 . . . xn−1   a1   y2 
2
A  ..  =  ..   ..  =  .. 
      
 .  .  .   . 
n−1
an−1 1 xn . . . xn an−1 yn
   
a0 y1
 a1   y2 
This is possible as |A| 6= 0, as x1 , . . . , xn are distinct.  ..  = A−1  .. .
   
 .  .
an−1 yn
Note: We can also use Lagrange’s interpolation formula from numerical analysis, giving
n Q
j6=i (x − xj )
X
f (x) = yi Q
i−1 j6=i (xi − xj )

The two methods give the same polynomial, which is unique.

Paper II
 
3 0 √0
Question 4(a) Find a set of three orthonormal eigenvectors for the matrix A = 0 √4 3
0 3 6

Solution. The characteristic equation of A is



3 − λ 0 0



|A − λI = 0
4√−λ 3 = 0
0 3 6 − λ

Thus (3 − λ)(4 − λ)(6 − λ) − 3(3 − λ) = 0 ⇒ λ = 3, λ2 − 10λ + 21 = 0. Thus the eigenvalues


of A are 3, 3, 7.
Let (x1 , x2 , x3 ) be an eigenvector for λ = 7. Then
  
−4 0 √0 x1
 0 −3 3 x2  = 0

 
0 3 −1 x3
√ √ √
Thus −4x1 = 0, −3x2 + 3x3 = 0, 3x2 − x3 = 0. Thus √ x1 = 0, x3 = 3x2 with x2 6=1 0 √gives
any eigenvector for λ = 7. Take x2 = 1 to get (0, 1, 3), and normalize it to get (0, 2 , 23 ).

5
Let (x1 , x2 , x3 ) be an eigenvector for λ = 3. Then
  
0 0 √0 x1
0 1 3 x2  = 0

 
0 3 3 x3
√ √
Thus x2 + 3x3 = 0. Thus (x1 , − 3x3 , x3 ) with x1 , x3 ∈ R gives any √ eigenvector for λ = 3.
We can take x1 = 1, x3 = 0, and x1 = 0, x3 = 1 to get (1, 0, 0), (0, − 3, 1) as eigenvectors for
λ = 3 — these are orthogonal √ and therefore span the the eigenspace of λ = 3. Orthonormal
3 1
vectors are (1, 0, 0), (0, − 2 , 2 ). √ √
Thus the required orthonormal vectors are (0, 12 , 23 ), (1, 0, 0), (0, − 23 , 21 ).
In fact
√ 
1 3 0 0√ 1
    
0 2√ 2
3 0 √0 7 0 0
0 − 3 1  0 4
√ 3  √12 − 23 0 = 0 3 0
2 2
0 3 6 3 1 0 0 3
1 0 0 2 2
0

Question 4(b) Show that if A = X0 AX and B = X0 BX are two quadratic forms one of
which is positive definite and A, B are symmetric matrices, then they can be expressed as
linear combinations of squares by an appropriate linear transformation.

Solution. Let B be positive definite. Then there exists an orthogonal real non-singular
matrix H such that H0 BH = In , the unit matrix of order n. A is real-symmetric ⇒ H0 AH is
real symmetric. There exists
 K a real orthogonal
 matrix such that K0 H0 AHK is a diagonal
λ1 0 . . . 0
 0 λ2 . . . 0 
matrix i.e. K0 H0 AHK =  .. 0
..  where λ1 , . . . , λn are the eigenvalues of H AH.
 
..
. . .
0 0 . . . λn
   
x1 X1
0 0 0  ..   .. 
Now K H BHK = K In K = In . Then  .  = HK  .  diagonalizes A, B simulta-
xn Xn
neously.

   
x1 x1
. . . xn A  ...  = λ1 X12 + . . . + λn Xn2 . . . xn B  ...  = X12 + . . . + Xn2
     
x1 x1
xn xn

|A−λB| = 0 because |A−λB| = |H0 ||A−λB||H| =


Note that λ1 , . . . , λn are the roots of Q
|H AH − λH BH| = |H AH − λIn | = ni=1 (λ − λi ).
0 0 0

6
UPSC Civil Services Main 1982 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 27, 2009

Question 1(a) Let V be a vector space. If dim V = n with n > 0, prove that

1. any set of n linearly independent vectors is a basis of V.

2. V cannot be generated by fewer than n vectors.

Solution. From 1983 question 1(a) we get that any two bases of V have n elements.

1. Let v1 , . . . , vn be n linearly independent vectors in V. Then v1 , . . . , vn generate V —


if v ∈ V is such that v is not a linear combination of v1 , . . . , vn , then v, v1 , . . . , vn are
linearly independent, so dim V > n which is not true. Thus v1 , . . . , vn is a basis of V
— here we have used the technique used to complete any linearly independent set to
a basis.

2. V cannot be generated by fewer than n vectors, because then it will have a basis
consisting of less than n elements, which contradicts the fact that dim V = n.

Question 1(b) Define a linear transformation. Prove that both the range and the kernel of
a linear transformation are vector spaces.

Solution. Let V and W be two vector spaces. A mapping T : V −→ W is said to be a


linear transformation if

1. T(v1 + v2 ) = T(v1 ) + T(v2 ).

2. T(αv) = αT(v) for any α ∈ R, v ∈ V.

1
Range of T = T(V), kernel of T = {v | T(v) = 0}. If w1 , w2 ∈ T(V), then w1 =
T(v1 ), w2 = T(v2 ) for some v1 , v2 ∈ V, αw1 + βw2 = αT(v1 ) + βT(v2 ) = T(αv1 + βv2 ).
But αv1 + βv2 ∈ V ∴ αw1 + βw2 ∈ T(V), thus T(V) is a subspace of W. Note that
T(V) 6= ∅ ∵ 0 ∈ T(V) so T(V) is a vector space.
If v1 , v2 ∈ kernel T then T(v1 ) = 0, T(v2 ) = 0. Now T(αv1 + βv2 ) = αT(v1 ) +
βT(v2 ) = 0 ⇒ αv1 + βv2 ∈ kernel T. Thus kernel T is a subspace. kernel T 6= ∅, bf 0 ∈
kernel T so kernel T is a vector space.

Question 2(a) Reduce the matrix


 
2 3 −1 0
1 −1 2 0
1 2 −1 0

to row echelon form.

Solution. Let the given matrix  be called A. 


1 4 −3 0
Operation R1 − R2 ⇒ A ∼ 1 −1 2 0

1 2  −1 0 
1 4 −3 0
Operation R2 − R1 , R3 − R1 ⇒ A ∼ 0 −5 5
 0
 0 −2 2 0
1 4 −3 0
Operation − 15 R2 , − 12 R3 ⇒ A ∼ 0 1 −1 0
 0 1 −1  0
1 4 −3 0
Operation R3 − R2 ⇒ A ∼ 0 1 −1 0
0 0 0 0 
1 0 1 0
Operation R1 − 4R2 ⇒ A ∼ 0 1 −1 0
0 0 0 0  
1 0 1 0
Thus rank A = 2 and the row echelon form is 0 1 −1 0
0 0 0 0

Question 2(b) If V is a vector space of dimension n and T is a linear transformation on


V of rank r, prove that T has nullity n − r.

Solution. See 1998 question 3(a).

2
Question 2(c) Show that the system of equations

3x + y − 5z = −1
x − 2y + z = −5
x + 5y − 7z = 2

is inconsistent.

Solution. From the first two equations, (3x + y − 5z) − 2(x − 2y + z) = −1 − 2(−5) = 9 ⇒
x + 5y − 7z = 9. But this is inconsistent with the third equation, hence the overall system
in inconsistent.

Question 3(a) Prove that the trace of a matrix is equal to the sum of its characteristic
roots.

Solution. The characteristic polynomial of A is |λI − A| = λn + p1 λn−1 + p2 λn−2 + . . . + pn .


Thus the sum of the roots of |λI − A| = −p1 = a11 + a22 + . . . + ann = tr A. Thus the trace
of A = sum of the eigenvalues of A.

Question 3(b) If A, B are two non-singular matrices of the same order, prove that AB
and BA have the same eigenvalues.

Solution. See 1995 question 2(b).


 
cos θ sin θ
Question 3(c) Find the eigenvalues and eigenvectors of the matrix A = .
sin θ − cos θ

Solution. The characteristic equation of A is (cos θ − λ)(− cos θ − λ) − sin2 θ = 0 ⇒


λ2 − 1 = 0 ⇒ λ = ±1.
If (x1 , x2 ) is an eigenvector for λ = 1, then
  
cos θ − 1 sin θ x1
=0
sin θ − cos θ − 1 x2

Thus x1 (cos θ−1)+x2 sin θ = 0, x1 sin θ+x2 (− cos θ−1) = 0. We can take x1 = 1+cos θ, x2 =
sin θ.
Similarly if (x1 , x2 ) is an eigenvector for λ = −1, then
  
cos θ + 1 sin θ x1
=0
sin θ − cos θ + 1 x2

Thus x1 (cos θ+1)+x2 sin θ = 0, x1 sin θ+x2 (− cos θ+1) = 0. We can take x1 = 1−cos θ, x2 =
− sin θ as an eigenvector.

3
Paper II

Question 4(a) If V is finite dimensional and if W is a subspace of V, then show that W


is finite dimensional and dim W ≤ dim V.

Solution. If W = {0} then dim W = 0 ≤ dim V. If W 6= {0}, let v1 ∈ W, v1 6= 0. Let W1


be the space spanned by v1 then W1 is of dimension 1. If W1 = W, then dim W = 1 ≤ dim V.
If W1 6= W, then there exists a v2 ∈ W, v2 6∈ W1 . v1 , v2 are linearly independent —
if av1 + bv2 = 0, then if b 6= 0 then v2 = − ab v1 ⇒ v2 ∈ W1 , which is not true, hence
b = 0 ⇒ a = 0. Now let W2 be the space spanned by v1 , v2 then W2 is of dimension 2. If
W2 = W, then dim W = 2 ≤ dim V.
We continue the same reasoning as above, but this process must stop after at most r
steps where r ≤ n, otherwise we would have found n + 1 linearly independent vectors in V,
which is not possible. After r steps, we would have v1 , . . . , vr which are linearly independent
and span W. Thus dim W ≤ dim V, and W is finite dimensional.

Question 5(a) State and prove the Cayley-Hamilton Theorem when the eigenvalues are all
different.

Solution. See 1987 question 5(a).

Question 5(b) When are two real symmetric matrices said to be congruent? Is congruence
an equivalence relation? Also show how you can find the signature of A.

Solution. Two matrices A, B are said to be congruent to each other if there exists a
nonsingular matrix P such that P0 AP = B.
Congruence is an equivalence relation:
• Reflexive: A ≡ A ∵ A = I0 AI, I is the unit matrix.

• Symmetric: A ≡ B ⇒ P0 AP = B ⇒ A = (P−1 )0 BP−1 ⇒ B ≡ A.

• Transitive: A ≡ B, B ≡ C ⇒ A ≡ C — P0 AP = B, Q0 BQ = C ⇒ Q0 P0 APQ =
C ⇒ A ≡ C because PQ is nonsingular as both P, Q are nonsingular.
Given a symmetric matrix A, we first prove that there exists a nonsingular matrix P
such that P0 AP = diagonal[α1 , α2 , . . . , αr , 0, . . . , 0] where r is the rank of A.
We will prove this by induction on the order n of the matrix A. If n = 1, there is nothing
to prove. Assume that the result is true for all matrices of order < n.
Step 1. We first ensure that we have a11 6= 0. If it is 0, but some other akk 6= 0, we
interchange the k-th row with the first row and the k-th column with the first column, to
get B = P0 AP, where b11 = akk 6= 0. Note that P is the elementary matrix E1k (see 1983
question 2(a)), and is hence nonsingular and symmetric, so B is symmetric.
If all aii are 0, but some aij 6= 0. We add the j-th row to the i-th row and the j-the column
to the i-th column by multiplying A by Eij (1) and its transpose, to get B = Eij (1)AEij (1)0

4
— now bii = aij + aji 6= 0. B is still symmetric, and we can shift bii to the leading place as
above.
(Note that if all aij = 0, we stop.)
Thus we start with a11 6= 0. We subtract aa1k 11
times the first row from the k-th row and
a1k
a11
times the first column from the k-th column, by performing B = Ek1 (− aa1k11
)AEk1 (− aa1k
11
)0
 
a11 0
Repeating this for all k, 2 ≤ k ≤ n, we get P01 AP1 = , where A1 is n − 1 ×
0 A1
0
n − 1 and P1 is nonsingular. Now by induction, ∃P2 , n − 1 × n − 1 such  that  P2 AP2 =
1 0
diagonal[β2 , . . . , βr , 0, . . . , 0], rank A1 = rank A − 1. Now set P = P1 to get the
0 P2
result.
Now that we have P0 AP = diagonal[α1 , α2 , . . . , αr , 0, . . . , 0], let us assume that α1 , . . . , αs
are positive, the rest are negative. Then let αi = βi2 , 1 ≤ i ≤ s, −αj = βj2 , s + 1 ≤ j ≤ r. Set
Q = diagonal[β1−1 , . . . , βr−1 , 1, . . . , 1]. Then x0 Q0 P0 APQx = x21 + . . . + x2s − x2s+1 − x2r . Thus
we can find the signature of A by looking at the number of positive and negative squares of
the RHS.

Question
P 5(c)P Derive a set of necessary and sufficient conditions that the real quadratic
form 3j=1 3i=1 aij xi xj be positive definite.
Is 4x2 + 9y 2 + 2z 2 + 8yz + 6zx + 6xy positive definite?

Solution. For the first part, see 1992 question 2(c).

Q(x, y, z) = 4x2 + 9y 2 + 2z 2 + 8yz + 6zx + 6xy


3 3 9 9 9
= (2x + y + z)2 + 9y 2 + 2z 2 + 8yz + − yz − y 2 − z 2
2 2 2 4 4
3 3 2 27 2 1 2 7
= (2x + y + z) + y − z − yz
2 2 4 4 2
3 3 2 27 2 1 2 14
= (2x + y + z) + (y − z − yz)
2 2 4 27 27
3 3 2 27 7 2 1 2 49 2
= (2x + y + z) + (y − z) − z − z
2 2 4 27 4 108

So set X = 2x + 32 y + 23 z, Y = y − 277
z, Z = z, then Q(x, y, z) is transformed to X 2 + 27
4
Y2−
76
108
Z 2 . Hence Q(x, y, z) is is not positive definite.

5
UPSC Civil Services Main 1983 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 27, 2009

Question 1(a) Let V be a finitely generated vector space. Show that V has a finite basis
and any two bases of V have the same number of vectors.
Solution. Let {v1 , . . . , vm } be a generating set for V, we assume that vi 6= 0, 1 ≤ i ≤ m.
If {v1 , . . . , vm } is linearly independent, then it is a basis of V. Otherwise, there exists a
vk that depends linearly on {vi | 1 ≤ i ≤ m, i 6= k}. This latter set is also a generating
set, and we rename it {u1 , . . . , um−1 }. We now apply the same reasoning to it — either it
is linearly independent and hence a basis, or we can drop an element from it and it still
remains a generating set. In a finite number of steps, we reach {x1 , . . . , xr } ⊆ {v1 , . . . , vm }
such that {x1 , . . . , xr } is linearly independent and a generating set, thus {x1 , . . . , xr } is a
basis of V.
Note: An alternative approach leading to the same result is to pick the maximal linearly
independent subset of {v1 , . . . , vm }. There are only 2m such subsets, so we can do so in a
finite number of steps (in the above procedure we dropped the dependent elements one at
a time to reach the maximal linearly independent subset). Now to be a basis, the maximal
linearly independent subset S = {x1 , . . . , xr } ⊆ {v1 , . . . , vm } needs to generate V. But this
is immediate, as for each vi , either vi ∈ S or S ∪ {vi } is linearly Pdependent — in that case
P r r
j=1 aj xj + bvi = 0, but not all aj , b are 0. Now if b = 0 then j=1 aj xj = 0 ⇒ aj = 0 for
1 ≤ j ≤ r, as S is linearly independent, and this contradicts the statement that not all aj , b
are 0. So b 6= 0, hence vi is a linear combination of S, hence S generates V and is a basis.
Any two bases have the same number of elements: Let {v1 , . . . , vm } and {w1 , . . . , wn }
be two bases of V. Assume wlogP that m ≤ n. Now since w1 ∈ V, w1 is generated by the
m
basis {v1 , . . . , vm }, thus w1 = j=1 aj vj . There must be at least one non-zero ak , as
w1 6= 0. NowP the set {vi | 1 ≤ i ≤ m, i 6= k} ∪ {w1 } generates the set {v1 , . . . , vm } (since
aj
vk = a1k w1 − m j=1,j6=k aP
k
vj ) and hence generates V.
Now we have w2 = m i=1,i6=k ai vi + bw1 . At least one of the ai 6= 0, otherwise we have a
linear equation between w1 and w2 , but these are linearly independent. We replace vi by w2 ,

1
and the result is also a generating set as above. Continuing, after m steps, we
P get a subset
{w1 , . . . , wm } which is a generating set. Now if n > m, we would have wn = m i=0 ai wi , but
this is not possible as the wi were a basis, and thus linearly independent. Hence n = m, and
the two bases have equal number of elements.

Question 1(b) Let V be the vector space of polynomials of degree ≤ 3. Determine whether
the following vectors of V are linearly dependent or independent: u = t3 − 3t2 + 5t + 1, v =
t3 − t2 + 8t + 2, w = 2t3 − 4t2 + 9t + 5.

Solution. Let au + bv + cw = 0. Then

a + b + 2c = 0 (1)
−3a − b − 4c = 0 (2)
5a + 8b + 9c = 0 (3)
a + 2b + 5c = 0 (4)

From (4) - (1) we get b + 3c = 0. Substituting b = −3c in (2), c = −3a ⇒ b = 9a. Now from
(1), a + 9a − 6a = 0 ⇒ a = 0 ⇒ b = c = 0. Thus au + bv + cw = 0 ⇒ a = b = c = 0, so the
vectors are linearly independent.

Question 1(c) For any linear transformation T : V1 → V2 prove that

rank T ≤ min(dim V1 , dim V2 )

Solution. By definition, rank T = dim T (V1 ). Clearly T (V1 ) is a subspace of V2 and


therefore dim T (V1 ) ≤ dim V2 . Let v1 , . . . , vn be a basis of V1 , then T (V1 ) is generated by
T (v1 ), . . . , T (vn ) — If w ∈ T (V1 ), then there
Pn exists v ∈ V1 such that T (v) = w. But v =
P n
i=1 ai vi , ai ∈ R, therefore w = T (v) = i=1 ai T (vi ) ⇒ {T (v1 ), . . . , T (vn )} is a generating
system for T (V1 ) ⇒ dim T (V1 ) ≤ n. Thus rank T = dim T (V1 ) ≤ min(dim V1 , dim V2 ).

Question 2(a) Show that every non-singular matrix can be expressed as a product of ele-
mentary matrices.

Solution. We first list all the elementary matrices:

1. Eij = the matrix obtained by interchanging the i-th and j-th rows (or the i-th and
j-th columns of the unit matrix. For example, if n = 4, then
 
1 0 0 0
0 0 1 0
E23 = 
0 1 0 0

0 0 0 1

2
2. Ei (α) is the matrix obtained by multiplying the i-th row of the unit matrix by α =
the matrix obtained by multiplying the i-th column of the unit matrix by α.

3. Eij (β) = the matrix obtained by adding β times the j-th row to the i-th row of the
unit matrix.

4. (Eij (β))0 = transpose of Eij (β) = the matrix obtained by adding β times the j-th
column to the i-th column of the unit matrix.
All elementary matrices are non-singular. In fact |Eij | = −1, |Ei (α)| = α, |Eij (β)| =
|(Eij (β))0 | = 1.
We now prove the result.
(1) Let C = AB. Then any elementary row transformation
  on AB is equivalent to sub-
R1
 ..  
jecting A to the same row transformation. Let A =  .  and B = C1 . . . Cp n×p .
Rm m×n
 
R1 C1 . . . R1 Cp
 .. .. 
Then AB =  . .  . Thus if any elementary row transformation i.e. (i)
Rm C1 . . . Rm Cp m×p
Intercanging two rows (ii) Multiplying a row by a scalar (iii) Adding a scalar multiple of a
row to another row, is carried out on A, the same will be carried out on AB and vice versa.
Similarly any column transformation on B is equivalent to the same column transformation
on AB.
(2) Multiplying by an elementary matrix Eij , Ei (α), Eij (β) on the left is the same as per-
forming the corresponding elementary row operation on the matrix. Multiplying the matrix
by an elementary matrix to the right is equal to subjecting the matrix to the correspond-
ing column transformation. We write A = IA. Now interchanging the i-th and j-th row
of A is equivalent to doing the same on I in IA (result (1) above), which is the same as
Eij A. Similar results hold for the other two row transformations. Writing A as AI gives
the corresponding result for column transformations.
(3) We now prove that if A is a matrix
 of  rank r > 0, then there exist P, Q products of
Ir 0
elementary matrices such that PAQ = where Ir is the unit matrix of order r. Since
0 0
A 6= 0, A has at least one non-zero element, say aij . By interchanging the i-th row with
the first row and the j-th column with the first column, we get a new matrix Bij = (bij )
such that b11 6= 0. This simply means that there exist elementary matrices P1 , Q1 such
−1
that P1 AQ1 =
  B. We multiply P1 AQ1 by P2 = E1 (b11 ) to obtain P2 P1 AQ1 = C =
1 ∗ ... ∗
∗ 
. Subtracting suitable multiples of the first row from the remaining rows of
 
 ..
. ∗ 

C and suitable multiples of the first column from the remaining columns, we get the new

3
 
1 0 ... 0
0 
matrix D of the form  .. . Thus we have proved that there exist P∗ , Q∗ products
 
. ∗
A 
0
 
1 0 ... 0
0 
∗ ∗
of elementary matrices such that P AQ =  .. . We carry on the same process
 
. ∗
A 
0
 
∗ ∗∗ ∗∗ Ir 0
on A without affecting the first row and column, and in r steps we get P AQ = ,
0 E
where P∗∗ , Q∗∗ are products of elementary matrices. Note that E = 0 because rank A = r.
Now if A is nonsingular, then P∗∗ AQ∗∗ = I. Inverting the elementary matrices (the
inverse of an elementary matrix is elementary), we get that A is a product of elementary
matrices.

Question 2(b) Reduce the matrix A to its normal form, and hence or otherwise determine
its rank.  
0 1 2 1
A = 1 2 3 2
3 1 1 3
 
1 2 3 2
Solution. Interchange of R1 and R2 ⇒ A ∼ 0 1 2 1
  3 1 1 3
1 2 3 2
R3 − 3R1 ⇒ A ∼ 0 1 2 1
0 −5 −8 −3
1 2 3 2
R3 + 5R2 ⇒ A ∼ 0 1 2 1
 0 0 2 2
1 2 3 2
− 21 R3 ⇒ A ∼ 0 1 2 1
0 0 −1 −1 
1 2 3 2
R3 + R2 ⇒ A ∼ 0 1 2 1
0 1 1 0 
1 2 3 2
Interchanging C2 , C4 , A ∼ 0 1 2 1
0 0 1 1 
1 0 0 0
C2 − 2C1 , C3 − 3C1 , C4 − 2C1 ⇒ A ∼ 0 1 2 1
0 0 1 1

4
   
1 0 0 0 1 0 0 0
C3 − 2C2 , C4 − C2 ⇒ A ∼ 0 1 0 0. C4 − C3 ⇒ A ∼ 0 1 0 0
0 0 1 1 0 0 1 0
we have P(3× 3) and Q(4 × 4) both products of elementary matrices such that
Thus 
1 0 0 0
PAQ = 0 1 0 0, which is the normal form of A. Clearly the rank of A is 3.
0 0 1 0

Question 2(c) Show that the equations


x+y+z = 3
3x − 5y + 2z = 8
5x − 3y + 4z = 14
are consistent and solve them.
 
1 1 1
Solution. The coefficient matrix A = 3 −5 2.
5 −3 4
det A = 1(−20
+ 6) − 1(12 − 10) + 1(−9 + 25) = 0, thus rank A < 3. Actually rank A = 2,
1 1
since 6= 0.
3 −5  
1 1 1 3
The augmented matrix B = 3 −5 2 8 .
 5 −3 4 14 
1 1 1 3
R2 − 3R1 , R3 − 5R1 ⇒ B ∼ 0 −8 −1 −1
 0 −8  −1 −1
1 1 1 3
R3 − R2 ⇒ B ∼ 0 −8 −1 −1
0 0 0 0
1 1
Thus rank B = 2, because 6= 0.
0 −8
Since rank A = rank B = 2, the system is consistent, and the space of solutions has
dimension 1.
Now x + y = 3 − z, 3x − 5y = 8 − 2z, subtracting the second from 3 times the first we
get 8y = 1 − z ⇒ y = 1−z 8
. x = 3 − z − 1−z 8
= 23−7z
8
. Thus the solutions are given by
( 23−7z
8
, 1−z
8
, z), z ∈ R.

Question 3(a) Prove that a square matrix satisfies its characteristic equation. Use this
result to find the inverse of  
0 1 2
A = 1 2 3
3 1 1

5
Solution. The first part is the Cayley-Hamilton theorem, see 1987 question 3(a).
The characteristic equation of A is

−λ 1 2

|A − λI| = 1 2 − λ 3 = 0
3 1 1 − λ
⇒ −λ(λ2 − 3λ + 2 − 3) − (1 − λ − 9) + 2(1 − 6 + 3λ) = 0
⇒ λ3 − 3λ2 − 8λ + 2 = 0

Thus A3 − 3A2 − 8A + 2I = 0 ⇒ A(A2 − 3A − 8I) = −2I, or A−1 = − 21 (A2 − 3A − 8I).


    
0 1 2 0 1 2 7 4 5
A2 = 1 2 3 1 2 3 = 11 8 11
3 1 1 3 1 1 4 6 10
        
7 4 5 0 3 6 8 0 0 1 −1 1
1  1
∴ A−1 = − 11 8 11 + 3 6 9 + 0 8 0 = −8 6 −2
2 2
4 6 10 9 3 3 0 0 8 5 −3 1

Note: In this case, we were required to use this method to find the inverse. An alternate
method of finding the inverse by performing elementary row and column operations is shown
in 1985 question 1(c).
Question 3(b) Find the eigenvalues and eigenvectors of
 
8 −6 2
A = −6 7 −4
2 −4 3

Solution.

8 − x −6 2

|A − xI| = −6 7 − x −4 = 0
2 −4 3 − x
⇒ (8 − x)(x2 − 10x + 21 − 16) + 6(6x − 18 + 8) + 2(24 − 14 + 2x) = 0
⇒ −x3 + 18x2 − 85x + 40 + 36x − 60 + 20 + 4x = 0
⇒ x3 − 18x2 + 45x = 0

Thus the eigenvalues are 0, 3, 15.   


8 −6 2 x1
If (x1 , x2 , x3 ) is an eigenvector for the eigenvalue 0, then −6 7 −4
   x2  = 0.
2 −4 3 x3
1
Thus 8x1 − 6x2 + 2x3 = 0, −6x1 + 7x2 − 4x3 = 0, 2x1 − 4x2 + 3x3 = 0 ⇒ x1 = 2 x3 , x2 = x3 .
Thus (1, 2, 2) is an eigenvector for 0, in general (x/2, x, x), x 6= 0 is an eigenvector for 0.

6
  
5 −6 2 x1
If (x1 , x2 , x3 ) is an eigenvector for the eigenvalue 3, then −6 4 −4
   x2  = 0.
2 −4 0 x3
Thus 5x1 − 6x2 + 2x3 = 0, −6x1 + 4x2 − 4x3 = 0, 2x1 − 4x2 = 0 ⇒ x1 = 2x2 , x3 = −2x2 .
Thus (2, 1, −2) is an eigenvector for 3, in general (2x, x, −2x), x 6= 0 is an eigenvector
  for 3.
−7 −6 2 x1
If (x1 , x2 , x3 ) is an eigenvector for the eigenvalue 15, then −6 −8 −4
   x2  = 0.
2 −4 −12 x3
Thus −7x1 −6x2 +2x3 = 0, −6x1 −8x2 −4x3 = 0, 2x1 −4x2 −12x3 = 0 ⇒ x1 = 2x3 , x2 = −2x3 .
Thus (2, −2, 1) is an eigenvector for 15, in general (2x, −2x, x), x 6= 0 is an eigenvector for
15.
Question 3(c) Show that the eigenvalues of an upper or lower triangular matrix are just
the diagonal elements of the matrix.
Solution. Let A = (aij ), such that aij = 0 for i < j, i.e. A is upper triangular. Now
|xI − A| = (x − a11 )(x − a22 ) . . . (x − ann )
showing that |xI − A| = 0 ⇒ x = a11 , a22 , . . . , ann . Thus the eigenvalues of A are
a11 , a22 , . . . , ann .
Similarly for a lower triangular matrix.
Paper II
Question 4(a) Prove that a necessary and sufficient condition that a linear transformation
A on a unitary space is Hermitian is that hAx, xi is real for all x.
Solution. A unitary space is an old name for an inner product space. Let V be an inner
product space over C, and hAv, vi be real for all v ∈ V. Then since
hA(v + w), v + wi = hAv, vi + hAw, wi + hAv, wi + hAw, vi
hAv, wi +hAw, vi is real (because hA(v + w), v + wi −hAv, vi −hAw, wi is real). Hence
hAv, wi + hAw, vi = hw, Avi + hv, Awi (1)
because z real ⇒ z = z.
Also,
hA(v + iw), v + iwi = hAv, vi + hA(iw), iwi − ihAv, wi + ihAw, vi
thus −ihAv, wi + ihAw, vi is real. Thus
−ihAv, wi + ihAw, vi = −ihw, Avi + ihv, Awi (2)
Multiplying (1) by i and adding to (2), we get
2ihAw, vi = 2ihv, Awi

Thus A = A , so A is Hermitian.
Conversely, if hAw, vi = hv, Awi , then hAv, vi = hv, Avi = hAv, vi ⇒ hAv, vi is
real.

7
Question 4(b) If A is a linear transformation on an n-dimensional vector space, then prove
that

1. rank A = rank A0 .

2. nullity A = n − rank A.

Solution. We know that rank A = r if A has a minor of order r different from 0, and all
minors of order > r are 0. Thus rank A = rank A0 .
For the second part, see 1998 question 3(a).

Question 4(c) Show that a real symmetric matrix A is positive definite if and only if there
exists a real non-singular matrix P such that A = PP0 .

Solution. x0 Ax = x0 PP0 x = sum of squares > 0 (because P0 x 6= 0 as P is non-singular).


Conversely: Let x1 , x2 , . . . , xn be a basis of Rn . We will use this to construct a new basis
e1 , e2 , . . . , en which satisfies ei Aej = δij , as follows:
x1
e1 = √
x1 Ax1
y2 = x2 − (x2 Ae1 )e1
y
e2 = √ 2
y2 Ay2
...
i−1
X
yi = xi − (xi Aej )ej
j=1
y
ei = √ i
yi Ayi
...
Pn
e1 , e2 , . . . , en are linearly independent — if i=1 ai ei = 0, then take the largest i such
that ai 6= 0, this allows us to express xi in terms of the other basis vectors, which is not
possible. Inductively we can also verify that ei Aei = 1, and ei Aej = 0 if i 6= j. This is
the Gram-Schmidt orthonormalization process - we exploit the property that any positive
definite matrix A gives rise to an inner product hx,yi = x0 Ay.
Now consider the matrix Q = e1 e2 . . . en . Now if B = Q0 AQ then bij = ei Aej ,
thus B = In . Since Q consists of linearly independent columns, it is invertible, and thus
A = Q0 −1 Q−1 . Setting P = Q0 −1 , we have A = PP0 .

Question 5(a) If S is a skew symmetric matrix of order n and if I + S is non-singular,


then prove that A = (I − S)(I + S)−1 is an orthogonal matrix of order n.

Solution. See 1999, question 2(b).

8
Question 5(b) Under what circumstances will the real n × n matrix
 
x a a ... a
a x a . . . a
 
A=  a a x . . . a 

 ... 
a a a ... x

be (1) positive semidefinite (2) positive definite.

Solution. The eigenvalues of the given matrix can be computed as follows:



x − λ a a . . . a

a
x−λ a ... a
a
a x − λ ... a = 0

...

a a a . . . x − λ

x − λ a − x + λ a − x + λ . . . a − x + λ

a
x−λ−a 0 ... 0

⇒ a 0 x − λ − a ... 0 = 0


...

a 0 0 . . . x − λ − a

x − λ + (n − 1)a 0 0 . . . 0


a x−λ−a 0 ... 0

⇒ a 0 x − λ − a ... 0 = 0


...

a 0 0 . . . x − λ − a
⇒ (x − λ + (n − 1)a)(x − λ − a)n−1 = 0

Thus the eigenvalues are x − a (repeated n − 1 times) and x + (n − 1)a. For positive definite,
λ > 0 ⇒ x > a, x > (n − 1)(−a). If a > 0, this reduces to x > a, if a ≤ 0, this reduces to
x > (n − 1)(−a).
For positive semi-definite, λ ≥ 0. By the same reasoning, if a > 0, then x ≥ a, otherwise
x ≥ (n − 1)(−a).

9
UPSC Civil Services Main 1984 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 27, 2009

Question 1(a) If W1 , W2 are finite dimensional subspaces of a vector space V, then show
that W1 + W2 is finite dimensional and dim W1 + dim W2 = dim(W1 + W2 ) + dim(W1 ∩ W2 ).

Solution. See 1988, question 1(b).

Question 1(b) If A and B are n-rowed square non-zero matrices such that AB = 0, then
show that both A and B are singular. If both A and B are singular, and AB = 0, does it
imply that BA = 0? Justify your answer.

Solution. If A were non-singular, then A−1 AB = 0 ⇒ B = 0. Thus A is singular, and


similarly B is singular.    
1 0 0 0
AB = 0 does not imply that BA = 0. Let A = ,B = . Then AB =
    1 0 0 1
0 0 0 0
, but BA = 6= 0.
0 0 1 0

Question 2(a) Show that row-equivalent matrices have the same rank.

Solution. See 1986 question 3(b).

Question 2(b) A linear transformation T on a vector space V with finite basis α1 , α2 , . . . , αn


is non-singular if and only if the vectors α1 T, α2 T, . . . , αn T are linearly independent in V.
When this is the case, show that T has an inverse T −1 with T T −1 = T −1 T = I.

1
Solution. If T (α1 P ), T (α2 ), . . . , T (αn ) Pare linearly independent, then T P is one-one: Let
n n n
v ∈ V. Then v = a α
i=1 i i , T (v) = a
i=1 i T (α i ). If T (v) = 0, then i=1 ai T (αi ) =
0 ⇒ ai = 0, 1 ≤ i ≤ n, because T (α1 ), T (α2 ), . . . , T (αn ) are linearly independent. Thus
T (v) = 0 ⇒ v = 0, so T is one-one.
T is onto: dim T (v) = dim V = n.
Thus T is invertible, in fact T −1 (T (αi )) = αi .
T −1 is a linear transformation: Let T −1 (v) = u, T−1 (w) = x. Then T (u) = v, T (x) = w.
Let T −1 (av + bw) = z, then T (z) = av + bw = aT (u) + bT (x) = T (au + bx) ⇒ z = au + bx.
Thus T −1 (av + bw) = aT −1 (v) + bT −1 (w), so T −1 is linear. It is obvious that T T −1 =
T −1 T = I, as this is true for the basis elements by definition, and extends to all vectors by
linearity. Pn
Conversely
Pn if T is non-singular, then a 1 T (α1 )+a 2 T (α 2 )+. . .+a n T (α n ) = 0 ⇒ T ( i=1 ai αi ) =
0 ⇒ i=1 ai αi = 0 ⇒ ai = 0, 1 ≤ i ≤ n because α1 , α2 , . . . , αn are linearly independent.
Thus T (α1 ), T (α2 ), . . . , T (αn ) are linearly independent.

Question 2(c) Solve the following system of equations:


3x1 + 2x2 + 2x3 − 5x4 = 8
2x1 + 5x2 + 5x3 − 18x4 = 9
4x1 − x2 − x3 + 8x4 = 7

Solution. Let the coefficient matrix be


 
3 2 2 −5
A = 2 5 5 −18
4 −1 −1 8
Doubling R1 , and subtracting R2 + R3 , we get
 
0 0 0 0
A ∼ 2 5 5 −18
4 −1 −1 8
Thus the rank of A is 2.
The augmented matrix
   
3 2 2 −5 8 0 0 0 0 0
B = 2 5 5 −18 9 ∼ 2 5 5 −18 9
4 −1 −1 8 7 4 −1 −1 8 7
Thus the rank of B is also 2, so the system is consistent.
Since the rank of A is 2, the space of solutions has rank = 4 − 2 = 2. Adding twice
the third equation to the first we get 11x1 + 11x4 = 22 ⇒ x1 = 2 − x4 . Substituting this
in the third equation, we get x2 = 1 − x3 + 4x4 . Thus the required solution system is
(2 − x4 , 1 − x3 + 4x4 , x3 , x4 ), where x3 , x4 take any value in R.

2
Question 3(a) Let V and W be vector spaces over the field F , and let T be a linear trans-
formation from V to W. If V is finite dimensional show that rank T + nullity T = dim V.

Solution. See question 1(a) from 1992.

e = T−1 AT. Show


Question 3(b) Let A be a square matrix and T be non-singular. Let A
that
1. A and A
e have the same eigenvalues.

2. tr A = tr A.
e

3. If x is an eigenvector of A corresponding to an eigenvalue, then T−1 x is an eigenvector


of Ae corresponding to the same eigenvalue.

Solution.
1. The eigenvalues of A are roots of |xI − A| = 0. The eigenvalue of A
e are roots of
−1 −1 −1 −1
0 = |xI − T AT| = |T xIT − T AT| = |T ||xI − A||T| = |xI − A|, so the
eigenvalues are the same.
2. tr AB = tr BA, so tr T−1 AT = tr ATT−1 = tr A.
3. If Ax = λx then T−1 AT(T−1 x) = T−1 (λx) = λT−1 x.

Question 3(c) A 3 × 3 matrix has the eigenvalues 6, 2, -1. The corresponding eigenvectors
are (2, 3, −2), (9, 5, 4), (4, 4, −1). Find the matrix.
 
2 9 4
Solution. Let P =  3 5 4 , and let A be the required matrix. Then AP =
  −2 4 −1 
6 0 0 6 0 0
P 0 2 0 , therefore A = P 0 2 0  P−1 . A simple calculation gives P−1 =
 0 0 −1  0 0 −1     
−21 25 16 2 9 4 6 0 0 12 18 −4
 −5 6 4 , note that |P| = 1. Now  3 5 4  0 2 0  =  18 10 −4.
22 −26  −17  −2 4 −1
 0 0 −1  −12 8 1
12 18 −4 −21 25 16 −430 512 352
Thus A =  18 10 −4   −5 6 4  =  516 620 396 
−12 8 1 22 −26 −17  234 226 −173 
x1 x2 x3 6 0 0
A longer way would be to set A =  y1 y2 y3 . Then AP = P 0 2 0 . This
z1 z2 z3 0 0 −1
yields three systems of linear equations which have to be solved.

3
Paper II

Question 4(a) Let V be the set of all functions from a non-empty set into a field K. For
any function f, g ∈ V and any scalar k ∈ K, let f + g and kf be functions in V defined by
(f + g)(x) = f (x) + g(x), (kf )(x) = kf (x) for every x ∈ X. Prove that V is a vector space
over K.

Solution. V = {f | f : X −→ K}.

1. f, g ∈ V ⇒ (f + g)(x) = f (x) + g(x) ⇒ f + g : X −→ K ⇒ f + g ∈ V.

2. The zero function namely 0(x) = 0∀x ∈ X is the additive identity of V i.e. f + 0 =
0 + f = f ∀f ∈ V.

3. f ∈ V ⇒ −f ∈ V where (−f )(x) = −f (x) anf f + (−f ) = 0 = (−f ) + f .

4. (f + g) + h = f + (g + h) for every f, g, h ∈ V.

5. If f ∈ V, k ∈ K, then (kf )(x) = kf (x)∀x ∈ X, so kf ∈ V and k(f + g) = kf + kg.

6. If k, k 0 ∈ K, f ∈ V, then k(k 0 f ) = (kk 0 )f .

7. If 1 ∈ K is the multiplicative identity, then 1f = f for every f .

8. k, k 0 ∈ K, f ∈ V ⇒ (k + k 0 )f (x) = kf (x) + k 0 f (x) ⇒ (k + k 0 )f = kf + k 0 f .

Thus V is a vector space over K.

Question 4(b) Find the eigenvalues and basis for each eigenspace of the matrix
 
1 −3 3
A = 3 −5 3
6 −6 4
.

Solution. The characteristic equation of A is

 |A − λI|  = 0
1−λ −3 3
⇒  3 −5 − λ 3  = 0
6 −6 4−λ
⇒ (1 − λ)(λ + 5)(λ − 4) + 18(1 − λ) + 3(12 − 3λ − 18) + 3(−18 + 30 + 6λ) = 0
⇒ (1 − λ)(λ2 + λ − 20) + 18 − 18λ − 9λ − 18 + 36 + 18λ = 0
⇒ λ2 + λ − 20 − λ3 − λ2 + 20λ − 9λ + 36 = 0
⇒ λ3 − 12λ − 16 = 0

4
Thus λ = −2, 4, −2. Let (x1 , x2 , x3 ) be an eigenvector for λ = 4.
  
−3 −3 3 x1
3 −9 3 x2  = 0
6 −6 0 x3
Thus −3x1 − 3x2 + 3x3 = 0, 3x1 − 9x2 + 3x3 = 0, 6x1 − 6x2 = 0 ⇒ x1 = x2 , x3 = 2x1 . We
can take (1, 1, 2) as an eigenvector corresponding to λ = 4.
Let (x1 , x2 , x3 ) be an eigenvector for λ = −2.
  
3 −3 3 x1
3 −3 3 x2  = 0
6 −6 6 x3
Thus 3x1 − 3x2 + 3x3 = 0 ⇒ x2 = x1 + x3 . (1, 1, 0), (0, 1, 1) can be taken as eigenvectors for
λ = −2.
Clearly (1, 1, 2) is a basis for the eigenspace for λ = 4. (1, 1, 0), (0, 1, 1) is a basis for the
eigenspace for λ = −2.
Question 4(c) Let a vector space V have finite dimension and let W be a subspace of V
and W 0 the annihilator of W. Prove that dim W + dim W 0 = dim V.
Solution. Let dim V = n, dim W = m, W ⊆ V. Let {v1 , . . . , vn } be a basis of V so chosen
that {v1 , . . . , vm } is a basis of W. Let {v1∗ , . . . , vn∗ } be the dual basis of V ∗ i.e. vi∗ (vj ) = δij .

We shall show that W 0 has {vm+1 , . . . , vn∗ } as a basis.
By definition of the dual basis vi∗ (vj ) = 0 when 1 ≤ i ≤ m and m + 1 ≤ j ≤ n. Since

vj , m + 1 ≤ j ≤ n annihilate the basis of W, it follows that vj (w) = 0 for all w ∈ W. Thus

{vm+1 , . . . , vn∗ } ⊆ W 0 , and are linearly independent, being a subset of a linearly independent
set.
Let f ∈ W 0 , then f = ni=1 ai vi∗ . We shall show that ai = 0 for 1 ≤ i ≤ m, thus f
P
∗ ∗ 0
is a linear P combination of {vP m+1 , . . . , vn }. By definition of W , f (v1 ) = 0, . . . , f (vm ) = 0,
n ∗ n ∗
therefore ( i=1 ai vi )(vj ) = i=1 ai δij = aj = 0 when 1 ≤ j ≤ m. Thus {vm+1 , . . . , vn∗ } is a
basis of W 0 , hence dim W 0 = n − m, hence dim W + dim W 0 = n = dim V.
Question 5(a) Prove that every matrix satisfies its characteristic equation.
Solution. See 1987 question 3(a).
Question 5(b) Find a necessary and sufficient condition that the real quadratic form
Xn X
n
aij xi xj be a positive definite form.
i=1 j=1

Solution. See 1991 question 1(c) and 1992 question 1(c)


Question 5(c) Prove that the rank of the product of two matrices cannot exceed the rank
of either of them.
Solution. See 1987 question 1(b).

5
UPSC Civil Services Main 1985 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 27, 2009

Question 1(a) If W1 and W2 are finite dimensional subspaces of a vector space V, then
show that W1 + W2 is finite dimensional and

dim W1 + dim W2 = dim(W1 + W2 ) + dim(W1 ∩ W2 )

Solution. See 1988 question 1(b).


       
1 0 0 1 0 0 0 0
Question 1(b) Let M1 = , M2 = , M3 = , M4 = . Prove
0 0 0 0 1 0 0 1
that the set {M1 , M2 , M3 , M4 } forms the basis of the vector space of 2 × 2 matrices.

Solution. See 2006 question 1(a).




1 3 3
Question 1(c) Find the inverse of the matrix A = 1 4 3.
1 3 4

Solution.    
1 3 3 1 0 0
1 4 3 = 0 1 0 A
1 3 4 0 0 1
Using operation R2 − R1 , R3 − R1 , we get
   
1 3 3 1 0 0
0 1 0 = −1 1 0 A
0 0 1 −1 0 1

1
Now with operation R1 − 3(R2 + R3 ) we get
   
1 0 0 7 −3 −3
0 1 0 = −1 1 0 A
0 0 1 −1 0 1
 
7 −3 −3
Thus the inverse of A is −1 1 0 .
−1 0 1

Question 2(a) If T : V −→ W is a linear transformation from an n-dimensional vector


space V to a vector space W, then prove that rank(T ) + nullity(T ) = n.

Solution. See 1992 question 1(a).

Question 2(b) Consider the basis S = {v1 , v2 , v3 } of R3 , where v1 = (1, 1, 1), v2 =


(1, 1, 0), v3 = (1, 0, 0). Express (2, −3, 5) in terms of v1 , v2 , v3 . Let T : R3 −→ R2 be
defined as T (v1 ) = (1, 0), T (v2 ) = (2, −1), T (v3 ) = (4, 3). Find T (2, −3, 5).

Solution. Let (2, −3, 5) = a(1, 1, 1) + b(1, 1, 0) + c(1, 0, 0). Then a + b + c = 2, a + b =


−3, a = 5 ⇒ a = 5, b = −8, c = 5. Thus (2, −3, 5) = 5v1 − 8v2 + 5v3 .
T (2, −3, 5) = 5T (v1 ) − 8T (v2 ) + 5T (v3 ) = 5(1, 0) − 8(2, −1) + 5(4, 3) = (9, 23).
 
6 3 −4
Question 2(c) Reduce the following matrix into echelon form: −4 1 −6.
1 2 −5
   
6 3 −4 1 2 −5
Solution. A = −4 1 −6 ∼ −4 1 −6 by exchanging R1 and R3 .
1 2 −5 6 3 −4 
1 2 −5
Now R2 + 4R1 , R3 − 6R1 ⇒ A ∼ 0 9 −26.

  0 −9 26
1 2 −5
R3 + R2 ⇒ A ∼ 0 9 −26.
0 0 0  
1 2 −5
Multiply R2 by 91 to get A ∼ 0 1 − 269
.
0 0 0
1 0 97

Now R1 − 2R2 ⇒ A ∼ 0 1 − 26 9
 which is the required form.
0 0 0

2
Question 3(a) Show that if λ is an eigenvalue of matrix A, then λn is an eigenvalue of
An , where n is a positive integer.
Solution. If x is an eigenvector for λ, then An x = An−1 Ax = λAn−1 x. Repeating this
process, we get the result.
Question 3(b) Determine if the vectors (1, −2, 1), (2, 1, −1), (7, −4, 1) are linearly indepen-
dent in R3 .
Solution. If possible, let a(1, −2, 1) + b(2, 1, −1) + c(7, −4, 1) = 0. Then a + 2b + 7c =
0, −2a + b − 4c = 0, a − b + c = 0. Adding the last two we get a = −3c, and from the third
we then get b = −2c. These values satisfy the first equation also, hence letting c = −1 we
get 3(1, −2, 1) + 2(2, 1, −1) − (7, −4, 1) = 0. Thus the vectors are linearly dependent.
Question 3(c) Solve
2x1 + 3x2 + x3 = 9 (1)
x1 + 2x2 + 3x3 = 6 (2)
3x1 + x2 + 2x3 = 8 (3)
Solution. 2(2) − (1) ⇒ x2 + 5x3 = 3 ⇒ x2 = 3 − 5x3 . Substituting x2 in (2), x1 = 7x3 .
5
Now substituting x1 , x2 in (3), we get 21x3 + 3 − 5x3 + 2x3 = 8 ⇒ x3 = 18 , x2 = 29
18
, x1 = 35
18
,
which is the required solution.
(Using Cramer’s rule would have been lengthy.)
Paper II
Question 4(a) Let V be the vector space of all functions from R into R. Let Ve be the
subset of all even functions f, f (−x) = f (x), and Vo be the subset of all odd functions
f, f (−x) = −f (x). Prove that
1. Ve and Vo are subspaces of V
2. Ve + Vo = V
3. Ve ∩ Vo = {0}
Solution.
1. Let f, g ∈ Ve , then αf + βg ∈ Ve for all α, β ∈ R, because (αf + βg)(−x) = αf (−x) +
βg(−x) = αf (x) + βg(x) = (αf + βg)(x), thus Ve is a subspace of V. Similarly, if
f, g ∈ Vo , then αf + βg ∈ Vo for all α, β ∈ R, because (αf + βg)(−x) = αf (−x) +
βg(−x) = −αf (x) − βg(x) = −(αf + βg)(x), thus Ve is a subspace of V.
2. Let f (x) ∈ V. Define F (x) = f (x)+f
2
(−x)
, G(x) = f (x)−f
2
(−x)
. Then F (−x) = F (x) ⇒
F ∈ Ve , G(x) = −G(x) ⇒ G ∈ Vo and f (x) = F (x) + G(x). Thus Ve + Vo = V.
3. If f ∈ Ve ∩ Vo , then f (−x) = f (x) ∵ f ∈ Ve , f (−x) = −f (x) ∵ f ∈ Vo . Thus
2f (−x) = 0 for all x ∈ R, so f = 0 ⇒ Ve ∩ Vo = {0}.

3
Question 4(b) Find the dimension and basis of the solution space S of the system

x1 + 2x2 + 2x3 − x4 + 3x5 = 0


x1 + 2x2 + 3x3 + x4 + x5 = 0
3x1 + 6x2 + 8x3 + x4 + 5x5 = 0

Solution.    
1 2 2 −1 3 1 2 2 −1 3
A = 1 2 3 1 1 ∼ 1 2 3 1 1
3 6 8 1 5 0 0 0 0 0
by performing R3 − R1 − 2R2 .
Thus rank A < 3. Actually rank A = 2, because if A = (C1 , C2 , C3 , C4 , C5 ), where Ci are
columns, then C1 and C3 are linearly independent.
Adding the first two equations, we get 4x5 = −2x1 − 4x2 − 5x3 . Subtracting 3 times
the second from the first, we get 4x4 = −2x1 − 4x2 − 7x3 . From these we can see that
X1 = (2, 0, 0, −1, −1), X3 = (0, 1, 0, −1, −1), X3 = (0, 0, 4, −5, −7) are three independent
solutions. Since rank A = 2, the dimension of the solution space S is 3, and {X1 , X2 , X3 }
is its basis.

Question 4(c) Let W1 and W2 be subspaces of a finite dimensional vector space V. Prove
that (W1 + W2 )0 = W10 ∩ W20 .

Solution. Let V ∗ be the dual of V i.e. V ∗ = {f | f : V −→ R, f linear}. Then


W 0 = {f | f ∈ V ∗ , ∀w ∈ W.f (w) = 0}. W 0 is a vector subspace of V ∗ of dimension
dim V − dim W.
If W1 ⊆ W2 , then W20 ⊆ W10 , because if f ∈ W20 , f (w) = 0∀w ∈ W2 , and therefore
f (w) = 0∀w ∈ W1 , so f ∈ W10 .
Now W1 ⊆ W1 + W2 and W2 ⊆ W1 + W2 , so (W1 + W2 )0 ⊆ W10 and (W1 + W2 )0 ⊆ W20 ,
thus (W1 + W2 )0 ⊆ W10 ∩ W20 .
Conversely, if f ∈ W10 ∩ W20 , then f (w1 ) = 0, f (w2 ) = 0 for all w1 ∈ W1 , w2 ∈ W2 . Now
any w ∈ W1 + W2 is of the form w = w1 + w2 , so f (w) = f (w1 ) + f (w2 ) = 0, because f is
linear. Thus f ∈ (W1 + W2 )0 .
Thus (W1 + W2 )0 = W10 ∩ W20 .
 
1 1+i 2i
Question 5(a) Let H = 1 − i 4 2 − 3i. Find P so that P0 HP is diagonal. Find
−2i 2 + 3i 7
the signature of H.

Solution.      
1 1+i 2i 1 0 0 1 0 0
1 − i 4 2 − 3i = 0 1 0 H 0 1 0
−2i 2 + 3i 7 0 0 1 0 0 1

4
Subtracting (1 − i)R1 from R2 , and adding 2iR1 to R3 , we get
     
1 1 + i 2i 1 0 0 1 0 0
0 2 −5i = −1 + i 1 0 H 0 1 0
0 5i 3 2i 0 1 0 0 1

Subtracting (1 + i)C1 from C2 , and adding −2iC1 to C3 , we get


     
1 0 0 1 0 0 1 −1 − i −2i
0 2 −5i = −1 + i 1 0 H 0 1 0 
0 5i 3 2i 0 1 0 0 1

Subtracting 25 iR2 from R3 , and adding 52 iC2 to C3 we get

1 −1 − i 52 − 92 i
     
1 0 0 1 0 0
0 2 0  = −1 + i 1 0 H 0 5
1 2
i 
19 5 9 5
0 0 −2 2
+ 2i −2i 1 0 0 1

1 −1 + i 52 + 29 i
 

Thus P = 0 1 − 52 i 
0 0 1
Index = Number of positive entries = 2. Signature = Number of positive entries - Number
of negative entries = 1.

Question 5(b) Prove that every matrix is a root of its characteristic polynomial.

Solution. This is the Cayley Hamilton theorem, proved in Question 5(a), 1987.

Question 5(c) If B = AP, where P is nonsingular and A orthogonal, show that PB−1 is
orthogonal.

Solution. B−1 = P−1 A−1 , so PB−1 = PP−1 A−1 = A−1 . Now (A−1 )0 A−1 = (AA0 )−1 = I.
Similarly A−1 (A−1 )0 = (A0 A)−1 = I, so PB−1 is orthogonal.

5
UPSC Civil Services Main 1986 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 27, 2009

Question 1(a) If A, B, C are three n × n matrices, show that A(BC) = (AB)C. Show by
an example that matrix multiplication is non-commutative.

Solution. Let A = (aij P),nB = (bij ), C = (cij ). Let


Pn BC = (βij ), AB = (αij ). Then the ij-th
element
P of P the RHS = k=1 αik ckj . But αik = l=1 ail blk , so the ij-th element of the RHS
= nk=1 nl=1 ail blk ckj . Pn Pn Pn
Similarly, the ij-th element of the LHS = l=1 ail βlj = l=1 ail k=1 blk ckj . Thus
A(BC) = (AB)C.        
1 1 0 0 1 1 0 0 0 1
Let A = ,B = , then AB = = . But BA =
  0 1  0 1 0 1 0 1 0 1
0 0 1 1 0 0
= . Thus AB 6= BA.
0 1 0 1 0 1

Question 1(b) Examine the correctness or otherwise of the following statements:

1. The division law is not valid in matrix algebra.

2. If A, B are square matrices each of order n, and I is the corresponding unit matrix,
then the equation
AB − BA = I
can never hold.

Solution.
       
1 0 0 0 0 0 0 0
1. True. Let A = ,B = ,C = . Then AB = = AC, but
0 0 0 1 0 0 0 0
B 6= C.

1
2. True. We have proved in 1987 question 5(c) that AB and BA have the same eigen-
values. Trace of AB − BA = trace of AB - trace of BA = sum of the eigenvalues of
AB - sum of the eigenvalues of BA = 0. But trace of In = n, thus AB − BA = I can
never hold.

Question 1(c) Find a 3 × 3 matrix X such that


   
1 2 −2 1 0 1
−1 3 0  X = 0 1 0
0 −2 1 0 1 1

1
2 −2
Solution. |A| = −13 0 = 3 − 2(−1) − 2(2) = 1, so A is non-singular. Hence
 0
−2 1  
1 0 1 3 2 6
−1
X=A 0 1 0. Asimple calculation gives A−1 = 1 1 2. Thus
0 1 1 2 2 5
    
3 2 6 1 0 1 3 8 9
X = 1 1 2 0 1 0 = 1 3 3
2 2 5 0 1 1 2 7 7

Question 2(a) If M, N are two subspaces of a vector space S, then show that their dimen-
sions satisfy
dim M + dim N = dim (M ∩ N ) + dim (M + N )

Solution. See 1998 question 1(b).

Question 2(b) Find a maximal linearly independent subsystem of the system of vectors
v1 = (2, −2, −4), v2 = (1, 9, 3), v3 = (−2, −4, 1), v4 = (3, 7, −1).

Solution. v1 , v2 are linearly independent because av1 +bv2 = (2a+b, −2a+9b, −4a+3b) =
0 ⇒ a = b = 0.
v3 is dependent on v1 , v2 . If v3 = av1 + bv2 , then (2a + b, −2a + 9b, −4a + 3b) =
7 6
(−2, −4, 1) ⇒ a = − 10 , b = − 10 .
Similarly v4 is dependent on v1 , v2 . If v4 = av1 + bv2 , then (2a + b, −2a + 9b, −4a + 3b) =
(3, 7, −1) ⇒ a = b = 1.
Thus the maximally linearly independent set is {v1 , v2 }.

2
Question 2(c) Show that the system of equations

4x + y − 2z + w = 3
x − 2y − z + 2w = 2
2x + 5y − w = −1
3x + 3y − z − 3w = 1

although consistent is not uniquely solvable. Determine a general solution using x as a


parameter.
 
4 1 −2 1
1 −2 −1 2 
Solution. The coefficient matrix A =  .
2 5 0 −1
3 3 −1 −3
 
4 1 −2 1 3
1 −2 −1 2 2
The augmented matrix B =  .
2 5 0 −1 −1
3 3 −1 −3 1
Add R2 to R4 , and subtract R1 , to get
   
4 1 −2 1 4 1 −2 1 3
1 −2 −1 2 
 , B ∼ 1 −2 −1 2 2

A∼ 
2 5 0 −1   2 5 0 −1 −1
0 0 0 0 0 0 0 0 0

1 −2 1

Since −2 −1 2 = 1 − 16 + 5 6= 0, it follows that rank A = rank B = 3, so the system
5 0 −1
is consistent. Since rank A = 3, the space of solutions is of dimension 1.
Subtracting the second equation from the fourth, we get 2x + 5y − 5w = −1. But
2x + 5y − w = −1, so w − 5w = 0 ⇒ w = 0.
Now y − 2z = 3 − 4x, −2y − z = 2 − x ⇒ −5z = 8 − 9x ⇒ z = 9x−8 5
. Now y =
−2x−1 −2x−1 9x−8
3 − 4x + 2 9x−8
5
= 5
. Thus the space of solutions is (x, 5
, 5
, 0). The system does
not have a unique solution.

Question 3(a) Show that every square matrix satisfies its characteristic equation. Using
this result or otherwiseshow that if
 
1 0 2
A = 0 −1 1
0 1 0

then A4 − 2A3 − 2A2 + 6A − 2I = A, where I is the 3 × 3 identity matrix.

3
Solution. The first part is the Cayley Hamilton theorem. See 1987 Question 5(a).
The characteristic equation of A is |A − xI| = 0, thus

1 − x 0 2

0
−1 − x 1 = (1 − x)(x2 + x − 1) = −x3 + 2x − 1 = 0
0 1 −x
By the Cayley Hamilton Theorem, A3 − 2A + I = 0.
Thus A4 = 2A2 − A, and 2A3 = 4A − 2I. Hence A4 − 2A3 − 2A2 + 6A − 2I =
2A2 − A − 4A + 2I − 2A2 + 6A − 2I = A as required.

Question 3(b) 1. Show that a square matrix is singular if and only if at least one of its
eigenvalues is 0.
2. The rank of an n×n matrix A remains unchanged if it is premultiplied or postmultiplied
by a nonsingular matrix, and that rank(XAX−1 ) = rank(A).
Solution.
1. The characteristic polynomial of A is |A−xI|. Putting x = 0, we see that the constant
term in the characteristic polynomial is |A|. Thus if A has 0 as an eigenvalue iff 0 is
a root of the characteristic polynomial iff |A| = 0.
 
R1
2. Let A =  ... , where each Ri is 1×n, i.e. A is m×n. Now rank(A) is the dimension
 
Rm
of the row space of A, i.e. the space generated by  R1 , . . . , Rm . Let P = (pij ) be 
an
p11 R1 + p12 R2 + . . . + p1m Rm
 p21 R1 + p22 R2 + . . . + p2m Rm 
m × m nonsingular matrix. Then B = PA =  .
 
..
 . 
pm1 R1 + pm2 R2 + . . . + pmm Rm
Thus the rows of PA ⊂ the row space of A, being linear combinations of rows of
A. Writing A = P−1 B, we get that the row space of A ⊂ the row space of B, so
rank(A) = rank(B).
Let Q be non-singular n × n, and C = AQ. It can be proved as above that the column
space of A = the column space of C, thus rank(A) = rank(C).
Now by using the above results, rank(XAX−1 ) = rank(XA) = rank(A).

Paper II
Question 4(a) If V1 and V2 are subspaces of a vector space V, then show that dim(V1 +V2 ) =
dim(V1 ) + dim(V2 ) − dim(V1 ∩ V2 ).
Solution. See 1998, question 1(b).

4
Question 4(b) Let V and W be vector spaces over the same field F and dim V = n. Let
{e1 , . . . , en } be a basis of V. Show that a map f : {e1 , . . . , en } −→ W, can be uniquely
extended to a linear transformation T : V −→ W whose restriction to the given basis is f
i.e. T (ei ) = f (ei ).
Pn Pn
Solution.
Pn If v =Pn i=1 a i e i , define T (v) = i=1 ai f (ei ). Clearly T (ei ) = f (ei ). If
v = i=1 ai ei , w = i=1 bi ei , then
n
X
T (αv + βw) = T ( (αai + βbi )ei )
i=1
n
X
= (αai + βbi )f (ei )
i=1
n
X n
X
= α ai f (ei ) + β bi f (ei )
i=1 i=1
= αT (v) + βT (w)

Thus T is a linear transformation.


PnIf U is any other linear transformation
Pn satisfying
Pn U (ei ) P= f (ei ), then P
n
for any v =
n
a e
i=1 i i , by linearity, T (v) = T ( a e
i=1 i i ) = a
i=1 i T (e i ) = a
i=1 i f (e i ) = i=1 ai U (ei ) =
U (v). Since this is true for every v, we have T = U .

Question 5(a) 1. If A and B are two linear transformations and if A−1 and B −1 exist,
show that (AB)−1 exists and (AB)−1 = B −1 A−1 .

2. Prove that similar matrices have the same characteristic polynomial and hence the
same eigenvalues.

3. Prove that the eigenvalues of a Hermitian matrix are real.

Solution.

1. Clearly (AB)(B −1 A−1 ) = AIA−1 = AA−1 = I, (B −1 A−1 )(AB) = B −1 A−1 AB =


B −1 B = I. Thus AB is invertible and its inverse is B −1 A−1 .

2. If B = P−1 AP then |λI−B| = |λP−1 P−P−1 AP| = |P−1 ||λI−A||P| = |λI−A|. Thus
A and B have the same characteristic polynomial and therefore the same eigenvalues.

3. See 1993 question 2(c).

5
1
Question 5(b) Reduce 2x2 + 4xy + 5y 2 + 4x + 13y − 4
= 0 to canonical form.

Solution.
1
LHS = 2(x + y + 1)2 − 2y 2 − 2 + 5y 2 + 9y −
4
9
= 2(x + y + 1)2 + 3(y 2 + 3y) −
4
3 27 9
= 2(x + y + 1)2 + 3(y + )2 − −
2 4 4
3
= 2X 2 + 3Y 2 − 9 where X = x + y + 1, Y = y +
2
X2 Y2
2X 2 + 3Y 2 − 9 = 0 ⇒ 9/2
+ 3
= 1. Thus the given equation is an ellipse.
 
0 1 1
Question 5(c) Find the reciprocal of the matrix T = 1 0 1. Then show that the
  1 1 0
b+c c−a b−a
transform of the matrix A = 2 c − b c + a a − b by T i.e. TAT−1 is a diagonal matrix.
1

b−c a−c a+b


Determine the eigenvalues of the matrix A.

Solution. |T| = −1(−1) + 1(1) = 2. So


   
A11 A21 A31 −1 1 1
1 1
T−1 = A12 A22 A32  =  1 −1 1 
2 2
A13 A23 A33 1 1 −1

Here Aij denotes the cofactor of aij . Now

    
0 1 1 b+c c−a b−a −1 1 1
1 1
TAT−1 = 1 0 1 c − b c + a a − b  1 −1 1 
2 2
1 1 0 b−c a−c a+b 1 1 −1
  
0 2a 2a −1 1 1
1
= 2b 0 2b   1 −1 1 
4
2c 2c 0 1 1 −1
   
4a 0 0 a 0 0
1
= 0 4b 0  = 0 b 0
4
0 0 4c 0 0 c

Thus TAT−1 is diagonal. Now the eigenvalues of A and TAT−1 are the same, so the
eigenvalues of A are a, b, c.

6
UPSC Civil Services Main 1987 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 27, 2009

 
7 −3
Question 1(a) 1. Find all the matrices which commute with the matrix .
5 −2
2. Prove that the product of two n × n symmetric matrices is a symmetric matrix if and
only if the matrices commute.

Solution.

1.      
a b 7 −3 7 −3 a b
=
c d 5 −2 5 −2 c d

⇒ 7a + 5b = 7a − 3c (i)
−3a − 2b = 7b − 3d (ii)
7c + 5d = 5a − 2c (iii)
−3c − 2d = 5b − 2d (iv)
(i) and (iv) ⇒ 5b = −3c. From (ii) we get d = a + 3b, and from (iii) we get the
same
  − 9c = 5a + 15b = 5d, or d = a + 3b. Thus the required matrices are
thing: 5a
a b
, a, b arbitrary.
− 53 b a + 3b

2. Given A0 = A, B0 = B. Suppose AB = BA, then (AB)0 = B0 A0 = BA = AB ⇒ AB


is symmetric. Let AB be symmetric. Then AB = (AB)0 = B0 A0 = BA, so A and B
commute. Thus AB is symmetric ⇔ AB = BA when A0 = A, B0 = B.

1
Question 1(b) Show that the rank of the product of two square matrices A, B each of order
n satisfies the inequality
rA + rB − n ≤ rAB ≤ min(rA , rB )
where rC stands for the rank of C, a square matrix.
 
G
Solution. There exists a non-singular matrix P such that PA = , where G is a
  0
G
rA × n matrix of rank rA . Now PAB = B has at most rA non-zero rows obtained on
0
multiplying rA non-zero rows of G with B. Thus rPAB , which is the same as rank rAB as P
is non-singular, ≤ rA . 
Similarly there exists a non-singular matrix Q such
 that BQ = H 0 , where H is a
n × rB matrix of rank rB . Now ABQ = A H 0 has at most rB non-zero, columns, so
rABQ ≤ rB . Now rABQ = rAB as |Q| 6= 0, so rAB ≤ rB , hence rAB ≤ min(rA , rB ).
Let S(A) denote the space generated by the vectors r1 , . . . , rn where ri is the ith row
of A, then dim(S(A)) = rA , similarly dim(S(B)) = rB . Let S denote the space generated
by the rows of A and B. Clearly dim(S) ≤ dim(S(A)) + dim(S(B)) = rA + rB . Clearly
S(A + B) ⊆ S. Therefore rA+B ≤ dim(S) ≤ rA + rB .    
IrA 0 −1 IrA 0
Now there exist non-singular matrices P, Q such that PAQ = or A = P Q−1 .
    0 0 0 0
−1 0 0 −1 −1 IrA 0
Let C = P Q . Then A + C = P Q−1 = P−1 Q−1 , so A + C
0 In−rA 0 In−rA
is nonsingular.
Now rank B = rank((A + C)B) ≤ rank(AB) + rank(CB). But rank(CB) ≤ rank(C) =
n−rA . Thus rB ≤ rAB +n−rA ⇒ rA +rB −n ≤ rAB . Hence rA +rB −n ≤ rAB ≤ min(rA , rB ).

Question 1(c) If 1 ≤ a ≤ 5, find the rank of the matrix


 
1 1 1 1
1 3 −2 a 
A= 2 2a − 2 −a − 2 3a − 1

3 a+2 −3 2a + 1

1 0 0 0

1 2 −3 a − 1
Solution. |A| = by carrying out the operations C2 −C1 , C3 −
2 2a − 4 −a − 4 3a − 3
3 a − 1 −6 2a − 2
2
−3 1

0
0 1

C1 , C4 − C1 . Thus |A| = (a − 1) 2a − 4 −a − 4 3 = (a − 1) 2a − 10 −a + 5 3 =
a−1 −6 2 a−5 0 2
(a − 1)(a − 5)2 .

2
Thus |A| 6= 0 when a 6= 1, a 6= 5. So for 1 < a < 5, rank A = 4.
If a = 5,
 
1 1 1 1
1 3 −2 5 
A =  2 8 −7 14

3 7 −3 11
 
1 0 0 0
1 2 −3 4 
=   (C2 − C1 , C3 − C1 , C4 − C1 )
2 6 −9 12
3 4 −6 8
 
1 0 0 0
0 2 −3 4 
=   (R2 − R1 , R3 − 2R1 , R4 − 3R1 )
0 6 −9 12
0 4 −6 8
 
1 0 0 0
0 2 −3 4
=  0 0 0 0
 (R3 − 3R2 , R4 − 2R2 )
0 0 0 0

1 0
which has rank 2, as 6= 0, showing that rank of A is 2 when a = 5.
0 2
If a = 1,
 
1 1 1 1
1 3 −2 1
A =  2 0 −3 2

3 3 −3 3
 
1 0 0 0
1 2 −3 0
=  2 −2 −5 0
 (C2 − C1 , C3 − C1 , C4 − C1 )
3 0 −6 0

1 0 0

which has rank 3 since 1 2 −3 =
6 0, showing that rank of A is 3 when a = 1.
2 −2 −5

Question 2(a) If the eigenvalues of a matrix A are λj , j = 1, 2, . . . n and if f (x) is a


polynomial in x, show that the eigenvalues of the polynomial f (A) are f (λj ), j = 1, 2, . . . n.

3
Solution. Let xr be an eigenvector of λr . Then Ak xr = Ak−1 (Axr ) = λr Ak−1 xr = . . . =
λkr xr . Thus the eigenvalues of Ak are λkj , j = 1, 2, . . . , n.
Let f (x) = a0 + a1 x + . . . + am xm . Then (a0 I + a1 A + . . . + am Am )xr = (a0 + a1 λr +
. . . + am λmr )xr = f (λr )xr . Thus the eigenvalues of f (A) are f (λj ), j = 1, 2, . . . n.

Question 2(b) If A is skew-symmetric, then show that (I − A)(I + A)−1 , where I is the
corresponding identity matrix, is orthogonal.
0 ab
 
Hence construct an orthogonal matrix if A = .
− ab 0
−1
Solution. For
 theaorthogonality
  − A)(I
of (I
a
+ A) , see question 2(a)
 of 1999.

1 −b 1 b −1 b b −a
I−A= a , and I + A = ⇒ (I + A) = a2 +b2 .
b
1 − ab 1 a b !
b2 −a2 −2ab
    2 
−1 1 b −a b −a 1 b − a2 −2ab a2 +b2 a2 +b2
Thus (I−A)(I+A) = a2 +b2 = a2 +b2 = b2 −a2 ,
a b a b 2ab b 2 − a2 2ab
a2 +b2 a2 +b2
which is the required orthogonal matrix.

Question 2(c) 1. If A and B are arbitrary square matrices of which A is non-singular,


show that AB and BA have the same characteristic polynomial.
2. Show that a real matrix A is orthogonal if and only if |Ax| = |x| for all x.

Solution.
1. BA = A−1 ABA. Thus the characteristic polynomial of BA is |xI−BA| = |xA−1 A−
A−1 ABA| = |A−1 ||xI − AB||A| = |xI − AB| which is the characteristic polynomial
of AB.
√ √
2. If A is orthogonal, i.e. A0 A = I, then |Ax| = x0 A0 Ax = x0 x = |x|.
Conversely |Ax| = |x| ⇒ x0 A0 Ax = x0 x ⇒ x0 (A0 A − I)x = 0 for all x. Thus
A0 A − I = 0, so A is orthogonal.
Note that is A = (aij ) is symmetric, and ni,j=1 aij xi xj = 0 for all x, then choose x = ei to
P
get e0i Aei = aii = 0, and choose x = ei +ej to get 0 = x0 Ax = aii +2aij +ajj = 2aij ⇒ aij = 0.
(Here ei is the i-th unit vector.) Thus A = 0.

Question 3(a) Show that a necessary and sufficient condition for a system of linear equa-
tions to be consistent is that the rank of the coefficient matrix is equal to the rank of the
augmented matrix. Hence show that the system
x + 2y + 5z + 9 = 0
x − y + 3z − 2 = 0
3x − 6y − z − 25 = 0
is consistent and has a unique solution. Determine this solution.

4
Solution. Let the system be Ax = b where A is m × n, x is n × 1 and b is m × 1. Let
rank A = r. A = [c1 , c2 , . . . , cn ] where each cj is an m × 1 column. We can assume without
loss of generality that c1 , c2 , . . . , cr are linearly independent, r = rank A. The system is now
x1 c1 + x2 c2 + . . . + xn cn = b
, where x0 = (x1 , . . . , xn ). Suppose rank([A b]) = r. This means that out of n + 1 columns,
exactly r are independent. But by assumption, c1 , c2 , . . . , cr are linearly independent, there-
fore these vectors form a basis for the column space of [A b]. Consequently there exist
α1 , . . . , αr such that α1 c1 + α2 c2 + . . . + αr cr = b. This gives us the required solution
{α1 , . . . , αr , 0, . . . , 0} to the linear system.
Conversely, let the system be consistent. Let A = [c1 , c2 , . . . , cn ] as before, with c1 , c2 , . . . , cr
linearly independent, r = rank A. Since the column space of A, i.e. the space generated
by c1 , c2 , . . . , cn has dimension r, each cj for r + 1 ≤ j ≤ n is linearly dependent on
c1 , c2 , . . . , cr . Since there exist α1 , . . . , αn such that α1 c1 + α2 c2 + . . . + αn cn = b, b is a
linear combination of c1 , c2 , . . . , cn . But each cj for r + 1 ≤ j ≤ n is a linear combination of
c1 , c2 , . . . , cr , therefore b is a linear combination of c1 , c2 , . . . , cr . Thus the space generated
by {c1 , c2 , . . . , cn , b} also has dimension
 r, sorank([A b]) = r = rank A.
1 2 5
The coefficient matrix A = 1 −1 3 . |A| = 24 6= 0, so rank A = 3. The aug-
 3 −6  −1
1 2 5 −9 1 2
5

mented matrix B = 1 −1 3 2  has rank ≤ 3, but since 1 −1 3 6= 0, it has
3 −6 −1 25 3 −6 −1
rank 3. Thus the given system is consistent.
Subtracting the second equation from the first we get 3y + 2z + 11 = 0. Subtracting 3
times the second equation from the third, we get 3y + 10z + 19 = 0. Clearly z = −1, y =
−3 ⇒ x = 2. Thus (2, −3, −1) is the unique solution. In fact the only solution of the system
is      
x −9 2
y  = A−1  2  = −3
z 25 −1

Question 3(b) In an n-dimensional vector space the system of vectors xj , j = 1, . . . , r are


linearly independent and can be expressed linearly in terms of the vectors yk , k = 1, . . . , s.
Show that r ≤ s.
Find a maximal linearly independent subsystem of the linear forms
f1 = x + 2y + z + 3t
f2 = 4x − y − 5z − 6t
f3 = x − 3y − 4z − 7t
f4 = 2x + y − z

5
Solution. Let W be the subspace spanned by yk , k = 1, . . . , s. Then dim W ≤ s. Since
xj ∈ W, j = 1, . . . , r because xj is a linear combination of yk , k = 1, . . . , s, and xj , j =
1, . . . , r are linearly independent, dim W ≥ r ⇒ r ≤ s.
Clearly f1 and f4 are linearly independent. f2 is linearly expressible in terms of f1 and f4
because f2 = af1 + bf4 ⇒ a + 2b = 4, 2a + b = −1, a − b = 5, 3a = −6 ⇒ a = −2, b = 3 satisfy
all four, hence f2 = −2f1 + 3f4 . Similarly f3 = − 37 f1 + 53 f4 . Thus {f1 , f4 } is a maximally
independent subsystem.

Paper II

Question 4(a) Let T : V −→ W be a linear transformation. If V is finite dimensional,


show that
rank T + nullity T = dim V

Solution. See question 1(a) of 1992.

Question 4(b) Prove that two finite dimensional vector spaces V, W over the same field F
are isomorphic if they are of the same dimension n.

Solution. Let dim V = dim W = n. Let v1 , . . . , vn be a basis of V,Pand w1 , . . . , wn be a


W. Define T : V −→ W by T (vi ) = wi and if v ∈ V, v = ni=1 ai vi , ai ∈ R then
basis of P
T (v) = ni=1 ai T (vi ). Then

1. T is a linear transformation. If v = ni=1 ai vi , u = ni=1 bi vi then


P P

n
X
T (αv + βu) = T ( (αai + βbi )T (vi )
i=1
n
X
= (αai + βbi )T (vi )
i=1
n
X n
X
= α ai T (vi ) + β bi T (vi )
i=1 i=1
= αT (v) + βT (u)

T is 1-1. Let T (v) = 0, where v = ni=1 ai vi . Then 0 = T (v) = ni=1 ai T (vi ) =


P P
2. P
n
i=1 ai wi ⇒ ai = 0, i = 1, . . . , n, because w1 , . . . , wn are linearly independent. Thus
T (v) = 0 ⇒ v = 0.

3. T is onto. If w ∈ W and w = ni=0 bi wi , then T (v) = w where v = ni=1 bi vi .


P P

6
Note: The converse of 4(b) is also true i.e. if T : V −→ W is an isomorphism i.e. V, W are
isomorphic, then dim V = dim W.
Let v1 , . . . , vn be a basis of V. Then {w1 = T P is a basis of W.
(v1 ), . . . , wn = T (vn )}P
n n
w1 , . . . , wn are linearly independent. If i=0 bi wi = 0, then i=0 bi T vi = 0 ⇒
T ( ni=0 bi vi ) = 0 ⇒ ni=0 bi vi = 0 ⇒ bi = 0 for 1 ≤ i ≤ n, because v1 , . . . , vn are linearly
P P
independent.
w1 , . . . , wn generate W.P If w ∈ W, then there exists a P v ∈ V such that
Pn T (v) = w,
n n
because
Pn T is onto. Let v = i=0 bi vi , then w = T (v) = T ( i=0 bi vi ) = i=0 bi T (vi ) =
i=0 bi wi .

Question 5(a) Prove that every square matrix is the root of its characteristic polynomial.

Solution. This is the Cayley Hamilton Theorem. Let A be a matrix of order n. Let

|A − xI| = ao + a1 x + . . . + an xn

Then we wish to show that

ao I + a1 A + . . . + an A n = 0

Suppose the adjoint of A − xI is B0 + B1 x + . . . + Bn−1 xn−1 , where Bi are matrices of order


n. Then by definition of the adjoint,

(A − xI)(B0 + B1 x + . . . + Bn−1 xn−1 ) = |A − xI|I

Substituting for |A − xI| the expression ao + a1 x + . . . + an xn and equating coefficients


of like powers, we get

AB0 = ao I
AB1 − B0 = a1 I
AB2 − B1 = a2 I
...
ABn−1 − Bn−2 = an−1 I
−Bn−1 = an I

Multiplying these equations successively by I, A, A2 , . . . , An on the left and adding, we get


0 = ao I + a1 A + . . . + an An , which was to be proved.

7
Question 5(b) Determine the eigenvalues and the corresponding eigenvectors of
 
2 2 1
A = 1 3 1
1 2 2

Solution.

λ − 2 −2 −1

|λI − A| = −1 λ − 3 −1 = 0

−1 −2 λ − 2

⇒ (λ − 2)2 (λ − 3) − 2(λ − 2) + 2(−λ + 2) − 2 − 2 − (λ − 3) = 0


⇒ (λ2 − 4λ + 4)(λ − 3) − 5λ + 7 = 0
⇒ λ3 − 7λ2 + 11λ − 5 = (λ − 1)(λ2 − 6λ + 5) = 0
⇒ λ = 1, 5, 1

Let (x1 , x2 , x3 ) be an eigenvector for λ = 5. Then


  
3 −2 −1 x1
−1 2 −1 x2  = 0
−1 −2 3 x3

Thus 3x1 − 2x2 − x3 = 0, −x1 + 2x2 − x3 = 0, −x1 − 2x2 + 3x3 = 0 ⇒ x1 = x2 = x3 . Thus


(1, 1, 1) is an eigenvector for λ = 5. In fact (x, x, x) with x 6= 0 are eigenvectors for λ = 5.
Let (x1 , x2 , x3 ) be an eigenvector for λ = 1. Then
  
−1 −2 −1 x1
−1 −2 −1 x2  = 0
−1 −2 −1 x3

Thus x1 + 2x2 + x3 = 0. We can take x1 = (1, 0, −1) and x2 = (0, 1, −2) as eigenvectors for
λ = 1. These are linearly independent, and all eigenvectors for λ = 1 are linear combinations
of x1 , x2 .    
1 1 0 5 0 0
Let P = 1 0 1 . Then P−1 AP = 0 1 0.
1 −1 −2 0 0 1

Question 5(c) Let A and B be n square matrices over F . Show that AB and BA have
the same eigenvalues.

Solution. If A is non-singular, then

BA = A−1 ABA ⇒ |xI − BA| = |xA−1 A − A−1 ABA| = |A−1 ||xI − AB||A| = |xI − AB|

8
Thus the characteristic polynomials of AB and BA are the same, so they have the same
eigenvalues.
If A is singular,
 then let rank(A) = r. Then there  existP, Q non-singular such that
Ir 0 I 0
PAQ = . Now PABP−1 = PAQQ−1 BP−1 = r Q−1 BP−1 . Let Q−1 BP−1 =
 0 0 0 0
B1 B2
, where B1 is r × r, B2 is r × n − r, B3 is n − r × r, B4 is n − r × n − r. Then
B3 B4     
−1 Ir 0 B1 B2 B1 B2
PABP = = , so the characteristic roots of AB are the
0 0 B3 B4 0 0
same as those of B1 , along with 0 repeated n − r times.
   
−1 B 1 B 2 Ir 0 B 1 0
Now Q−1 BAQ = Q−1 BP PAQ = = so the character-
B3 B4 0 0 B3 0
istic roots of BA are the same as those of B1 , along with 0 repeated n − r times. Thus BA
and AB have the same characteristic roots.

9
UPSC Civil Services Main 1988 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 27, 2009

Question 1(a) Show that a linear transformation of a vector space Vm of dimension m


into a vector space Vn of dimension n over the same field can be represented as a matrix.
If T is a linear transformation of V2 into V4 such that T(3, 1) = (4, 1, 2, 1) and T(−1, 2) =
(3, 0, −2, 1), then find the matrix of T.

Solution. Let vi , i = 1, . . . , m be a basis of Vm and wj , j = 1, . . . , n be a basis of Vn . If


n
X
T(vi ) = aji wj , i = 1, . . . , m
j=1

then T corresponds to the n × m matrix A whose (i, j)’th entry is aij . In fact (v1 , . . . , vm ) =
(w1 , . . . , wn )A.
It can be easily seen that
2 1
e1 = (1, 0) = (3, 1) − (−1, 2)
7 7
1 3
e2 = (0, 1) = (3, 1) + (−1, 2)
7 7

1
and therefore
2 1
(4, 1, 2, 1) − (3, 0, −2, 1)
T(e1 ) =
7 7
1
= (5, 2, 6, 1)
7
1 ∗
= (5e + 2e∗2 + 6e∗3 + e∗4 )
7 1
1 3
T(e2 ) = (4, 1, 2, 1) + (3, 0, −2, 1)
7 7
1
= (13, 1, −4, 4)
7
1
= (13e∗1 + e∗2 − 4e∗3 + 4e∗4 )
7
 
5 13
2 1 
Thus T corresponds to the matrix 17 
6 −4 w.r.t. the standard basis.

7 4

Question 1(b) If M, N are finite dimensional subspaces of V, then show that dim(M +
N ) = dim M + dim N − dim(M ∩ N ).

Solution. Let {u1 , u2 , . . . , ur } be a basis of M ∩ N where dim(M ∩ N ) = r. Complete


{u1 , u2 , . . . , ur } to a basis {u1 , u2 , . . . , ur , v1 , . . . , vm } of M, where dim M = m + r. Com-

B
plete {u1 , u2 , . . . , ur } to a basis {u1 , u2 , . . . , ur , w1 , . . . , wn } of N , where dim N = n+r. We
shall show that = {u1 , u2 , . . . , ur , v1 , . . . , vm , w1 , . . . , wn } is a basis of M + N , proving
the result.
If u ∈ M + N , then u = v + w for some v ∈ M, w ∈ N . Since B B is a superset of the

written as a linear combination of elements of


We now show that the set
B
bases of M, N , v, w can be written as linear combination of elements of

B . Thus B
is linearly independent. If possible let
generates M + N .
⇒ u can be

n
X m
X r
X
αi vi + βi w i + γi u i = 0
i=1 i=1 i=1
Pn Pm Pr Pn
Since
Pn i=1 αi v i = − i=1 β i w i − i=1 γi u i it follows that i=1 αi vi ∈ N .P Therefore
n r
means that ni=1 αi vi −
P P
α
Pri=1 i iv ∈ M ∩ N ⇒ α
i=1 i i v = η
i=1 i i u for ηi ∈ R. This
i=1 ηi ui = 0. But {u1 , u2 , . . . , ur , v1 , . . . , vm } are linearly independent, so αi = 0, 1 ≤
i ≤ n. Similarly we can show that βi = 0, 1 ≤ i ≤ m. Then the linear indepen-
dence of {u1 , u2 , . . . , ur } shows that γi = 0, 1 ≤ i ≤ r. Thus the vectors in
early independent and form a basis of M + N , showing that the dimension of M + N is
B
are lin-

m + n + r = (m + r) + (n + r) − r, which completes the proof.

2
Question 1(c) Determine a basis of the subspace spanned by the vectors v1 = (1, 2, 3), v2 =
(2, 1, −1), v3 = (1, −1, −4), v4 = (4, 2, −2).
Solution. v1 , v2 are linearly independent because if αv1 + βv2 = 0 then α + 2β =
0, 2α + β = 0, 3α − β = 0 ⇒ α = β = 0. If v3 = αv1 + βv2 , then the three linear equations
α + 2β = 1, 2α + β = −1, 3α − β = −4 should be consistent — clearly α = −1, β = 1 satisfy
all three, showing v3 = v2 − v1 . Again suppose v4 = αv1 + βv2 , then the three linear
equations α + 2β = 4, 2α + β = 2, 3α − β = −2 should be consistent — clearly α = 0, β = 2
satisfy all three, showing v4 = 2v2 .
Hence {v1 , v2 } is a basis for the vector space generated by {v1 , v2 , v3 , v4 }.
 
a1 b
Question 2(a) Show that it is impossible for S = , b 6= 0 to have identical eigen-
b a2
values.
 
0 λ1 0
Solution. We know given S symmetric ∃O orthogonal so that O SO = , where
0 λ2
λ1 , λ
2 are eigenvalues
 of S. If λ1 = λ2 , then we have S = O0 −1 (λI)O−1 = λ(OO0 )−1 = λI ⇒
λ 0
S= . Thus if b 6= 0, S cannot have identical eigenvalues.
0 λ

Question 2(b) Prove that the eigenvalues of a Hermitian matrix are all real and the eigen-
values of a skew-Hermitian matrix are either zero or pure imaginary.
Solution. See question 2(a), year 1998.

Question 2(c) If x0 Ax > 0 for all x 6= 0, A symmetric, then for all y 6= 0 y0 A−1 y > 0.
If λ is the largest eigenvalue of A, then
x0 Ax
λ = sup
x∈Rn x0 x
x6=0
.
Solution. Clearly A = A0 A−1 A ∴ x0 Ax = x0 A0 A−1 Ax = y0 A−1 y where y = Ax for any
x ∈ Rn , x 6= 0. Since |A| 6= 0, any vector y can be written as Ax, by taking x = A−1 y.
Thus x0 Ax > 0 ⇒ y0 A−1 y > 0 for all y 6= 0.  
λ1 0 . . . 0
 0 λ2 . . . 0 
0
0
Let M = supx∈Rn xxAx . Let O be an orthogonal matrix such that O AO = .. .
 
0x  ..
x6=0 . .
0 0 . . . λn
Let 0 6= x = Oy, then x x = y O Oy = y y. Now x Ax = y O AOy = i λi yi2 ≤ λy0 y
0 0 0 0 0 0 0
P
0 0
where λ is the largest eigenvalue of A. Thus λ ≥ xyAx 0y = xxAx 0 x , so λ ≥ M . On the other
0
hand, if x 6= 0 is an eigenvector corresponding to λ, then x Ax = λx0 x ⇒ λ = xxAx
0
0x ≤ M .

Thus λ = M as required.

3
Question 3(a) By converting A to an echelon matrix, determine its rank, where
 
0 0 1 2 8 9
0 0 4 6 5 3
 
A= 0 2 3 1 4 7 
0 3 0 9 3 7
0 0 5 7 3 1

Solution. Consider  
0 0 0 0 0
0 0 2 3 0
 
0
1 4 3 0 5
A =
2

 6 1 9 7
8 5 4 3 3
9 3 7 7 1
Interchange the first row with the third, then third with fourth, fourth with fifth and fifth
with sixth to get  
1 4 3 0 5
0 0 2 3 0
 
0
2 6 1 9 7
A ∼ 8 5

 4 3 3
9 3 7 7 1
0 0 0 0 0
Now perform R3 − 2R1 , R4 − 8R1 , R5 − 9R1 to get
 
1 4 3 0 5
0 0 2 3 0 
 
0
0 −2 −5 9 −3 
A ∼ 0 −27 −20 3 −37

 
0 −33 −20 7 −44
0 0 0 0 0

Interchange the second and the third row, and perform − 12 R2 , 21 R3 to get
 
1 4 3 0 5
5
0 1
 2
− 92 3 
2 
0 0 3
0 1 0 
A ∼  2 
 0 −27 −20 3 −37 

0 −33 −20 7 −44
0 0 0 0 0

4
Perform R4 + 27R2 , R5 + 33R2 to get
 
1 4 3 0 5
5
0
 1 2
− 92 3
2


0 3
0 0 1 0
A ∼ 95
2 
0
 0 2
− 237
2
7 
2 
125
0 0 2
− 283
2
11 
2
0 0 0 0 0
95 125
Operation R4 − 2
R3 , R5 − 2
R3
yields
 
1 4 3 0 5
5
0
 1 2
− 92 

3
2
0 3
0 0 1 0
A ∼ 2 
0
 0 0 − 759
4
7 
2 
0 0 0 − 941
4
11 
2
0 0 0 0 0
4
Now multiply R4 with − 759
 
1 4 3 0 5
5
0
 1 2
− 92 3

2

0 3
0 0 1 0 
A ∼ 2 
14 
0
 0 0 1 − 759 
0 0 0 − 941
4
11 
2
0 0 0 0 0
941
Performing R5 + 4
R4 results in
 
1 4 3 0 5
5
0
 1 2
− 29 3
2


0 3
0 0 1 0 
A ∼ 2
14

0
 0 0 1 − 759  
11
0 0 0 0 2
− 941×7
1882

0 0 0 0 0

which can be converted to  


1 4 3 0 5
5
0
 1 2
− 92 3
2


0 3
0 0 1 0 
A ∼ 2 
14 
0
 0 0 1 − 759 
0 0 0 0 1 
0 0 0 0 0
which is an echelon matrix. Its rank is clearly 5, so the rank of A = 5.

5
Question 3(b) Given AB = AC does it follow that B = C? Can you provide a counterex-
ample?

Solution. It does not follow that B = C.


   
1 0 0 1
A= , B= ⇒ AB = 0
0 0 0 0
C = 0 ⇒ AC = 0, but B 6= C.
 
0 −1 0
Question 3(c) Find a nonsingular matrix which diagonalizes A = −1 −1 1 , B =
  0 1 0
2 1 −2
1 2 −2 simultaneusly. Find the diagonal form of A.
−2 −2 3

Solution.

−2λ −1 − λ 2λ −2λ −1 − λ 0

|A − λB| = 0 ⇒ −1 − λ −1 − 2λ 1 + 2λ = 0 ⇒ −1 + λ 0 0 = 0
2λ 1 + 2λ −3λ 2λ 1 + 2λ −λ

Thus λ = 0, 1, −1. This shows that the matrices are diagonalizable simultaneously.
We now determine x1 , x2 , x3 such that (A − λB)xi = 0, i = 1, 2, 3. For λ = 0, let
x1 0 = (x1 , x2 , x3 ) be such that (A − λB)x1 = 0. Thus
  
0 −1 0 x1
−1 −1 1 x2  = 0
0 1 0 x3

Thus −x2 = 0, −x1 − x2 + x3 = 0, x2 = 0. Thus x1 0 = (1, 0, 1).


For λ = 1, let x2 0 = (x1 , x2 , x3 ) be such that (A − λB)x2 = 0. Thus
  
−2 −2 2 x1
−2 −3 3  x2  = 0
2 3 −3 x3

Thus −2x1 −2x2 +2x3 = 0, −2x1 −3x2 +3x3 = 0, 2x1 +3x2 −3x3 = 0 ⇒ x2 −x3 = 0 ⇒ x1 = 0.
Thus we may take x2 0 = (0, 1, 1).
For λ = −1, let x3 0 = (x1 , x2 , x3 ) be such that (A − λB)x3 = 0. Thus
  
2 0 −2 x1
0 1 −1   x2  = 0
−2 −1 3 x3

6
Thus 2x1 − 2x3 = 0, x2 − x3 = 0, −2x1 − x2 + 3x3 = 0 ⇒ x1 = x2 = x3 . Thus we may take
x3 0 = (1, 1, 1).
 
1 0 1
Let P = 0 1 1 so that
1 1 1
        
1 0 1 0 −1 0 1 0 1 1 0 1 0 −1 −1 0 0 0
0
P AP = 0 1 1
  −1 −1 1   0 1 1 = 0
  1 1 0 0 −1 = 0 1 0 
1 1 1 0 1 0 1 1 1 1 1 1 0 1 1 0 0 −1

7
UPSC Civil Services Main 1989 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

 
3 1 −1
Question 1(a) Find a basis for the null space of the matrix A = .
0 1 2

Solution. A is a linear transformation from R3 to R2 defined by A(e1 ) = 3e∗1 , A(e2 ) =


e∗1 + e∗2 , A(e3 ) = −e∗1 + 2e∗2 , where e1 , e2 , e3 is the standard basis of R3 and e∗1 , e∗2 is the
standard basis of R2 . Thus A(a, b, c) = e∗1 (3a + b − c) + e∗2 (b + 2c). Consequently, (a, b, c) ∈
null space of A if and only if 3a + b − c = 0, b + 2c = 0 ⇒ b = −2c, a = c. Thus null space
of A is {(c, −2c, c) | c ∈ R}. Note that rank A = 2, so the null space has dimension 1.
A basis for the null space is (1, −2, 1), any other multiple of this can also be regarded as a
basis.

Question 1(b) If W is a subspace of a finite dimensional vector space V then prove that
dim V/W = dim V − dim W.

Solution. Let v1 , . . . , vr be a basis of W, dim W = r. Let vr+1 , . . . , vn be n−r vectors in V


so chosen that v1 , . . . , vn is a basis of V, dim V = n. We will show that vi + W, r + 1 ≤ i ≤ n
is a basis of V/W ⇒ dim V/W = n − r.

1
First we show linear independence:
n
X
αi (vi + W) = 0
i=r+1
Xn
⇒ αi vi + W = 0 + W
i=r+1
Xn
⇒ αi vi ∈ W
i=r+1
Xn r
X
⇒ αi vi = −αi vi (say)
i=r+1 i=1
Xn
⇒ αi vi = 0
i=1
⇒ αi = 0, 1 ≤ i ≤ n (vi are linearly independent.)

Thus vi + W, r + 1 ≤ i ≤ n are linearly independent.


Pn
PnIf v + W is anyPnelement of V/W, Pnthen v = i=1 αi vi as v ∈ V. Therefore v + W =
i=1 αi vi +W = i=1 αi (vi +W) = i=r+1 αi (vi +W) because v1 +W = . . . = vr +W = W.
Thus vi + W, r + 1 ≤ i ≤ n generate V/W. Hence dim V/W = n − r = dim V − dim W

Question 1(c) Show that all vectors (x1 , x2 , x3 , x4 ) in the vector space V4 (R) which obey
x4 −x3 = x2 −x1 form a subspace V. Show further that V is spanned by ξ1 = (1, 0, 0, −1), ξ2 =
(0, 1, 0, 1), ξ3 = (0, 0, 1, 1).

Solution. If y = (y1 , y2 , y3 , y4 ), z = (z1 , z2 , z3 , z4 ) ∈ V then αy + βz = (a1 , a2 , a3 , a4 ) ∈ V


because

a4 − a3 = (αy4 + βz4 ) − (αy3 + βz3 )


= α(y4 − y3 ) + β(z4 − z3 )
= α(y2 − y1 ) + β(z2 − z1 ) ∵ y4 − y3 = y2 − y1 , z4 − z3 = z2 − z1
= a2 − a1

Thus V is a subspace of V4 (R). Note that V 6= ∅.


Clearly ξ1 , ξ2 , ξ3 are linearly independent ⇒ dim V ≥ 3. But V =
6 V4 (R) because
(1, 0, 0, 0) 6∈ V ∴ dim V < 4 ⇒ dim V = 3.
Hence ξ1 , ξ2 , ξ3 is a basis of V and therefore span V.

Question 2(a) Let P be a real skew-symmetric matrix and I the corresponding unit matrix.
Show that I − P is non-singular. Also show that Q = (I + P)(I − P)−1 is orthogonal.

2
Solution. We have proved (question 2(a), year 1998) that the eigenvalues of a skew-
Hermitian and therefore of a skew-symmetric matrix are zero or pure imaginary. This means
|I − P| =
6 0 because 1 cannot be an eigenvalue of P.
Q Q = [(I − P)−1 ]0 (I + P)0 (I + P)(I − P)−1 = (I + P)−1 (I − P)(I + P)(I − P)−1 . But
0

(I − P)(I + P) = I − P2 = (I + P)(I − P), therefore Q0 Q = I. Similarly QQ0 = I ⇒ Q is


orthogonal.
Related Results:

1. If S is skew-Hermitian, then A = (I+S)(I−S)−1 is unitary. Conversely, if A is unitary,


then A can be written as A = (I + S)(I − S)−1 for some skew-Hermitian matrix S
provided −1 is not an eigenvalue of A.
Proof:
0 0 0 0
A = ((I − S)−1 )0 (I + S) = (I − S )−1 (I + S )
= (I + S)−1 (I − S)
0
∴ AA = (I + S)(I − S)−1 (I + S)−1 (I − S)
= (I + S)(I + S)−1 (I − S)−1 (I − S) = I
∵ (I − S)−1 (I + S)−1 = (I − S2 )−1 = (I + S)−1 (I − S)−1
0
Similarly A A = I, so A is unitary.
Now A(I − S) = I + S ⇒ A − I = (A + I)S ⇒ S = (A + I)−1 (A − I). It can be checked
as above that S is skew-Hermitian. Note that |A + I| =
6 0.

2. If H is Hermitian, then A = (H + iI)−1 (H − iI) is unitary and every unitary matrix


can be thus represented provided it does not have −1 as its eigenvalue.

3. If S is real, S0 = −S and S2 = −I, then S is orthogonal and of even order, and there
exist non-null vectors x, y such that x0 x = y0 y = 1, x0 y = 0, Sx + y = 0, Sy = x.
Proof: S0 S = −SS = I, so S is orthogonal, |S| 6= 0 ⇒ S is of even order.
Choose y such that y0 y = 1. Then y0 Sy = (y0 Sy)0 = y0 S0 y = −y0 Sy ⇒ y0 Sy = 0.
Set x = Sy, then y0 x = 0, Sx + y = 0. In addition, x0 x = y0 S0 Sy = y0 y = 1.

Question 2(b) Show that an n × n matrix A is similar to a diagonal matrix if and only if
the set of eigenvectors of A includes a set of n linearly independent vectors.

Solution. See question 2(c) of 1998.

Question 2(c) Let r1 , r2 be distinct eigenvalues of a matrix A and let ξi be an eigenvector


0
corresponding to ri , i = 1, 2. If A is Hermitian, show that ξ1 ξ2 = 0.

Solution. See question 2(c) of 1993.

3

1 1 , and B =
Question
 3(a) Find the roots of the equation |xA − B| = 0 where A = 1 4
0 3 . Use the result to show that the real quadratic forms F = x2 + 2x x + 4x2 , G = 6x x
3 0 1 1 2 2 1 2
can be simultaneously reduced by a non-singular linear substitution to y12 + y22 , y12 − 3y22 .

x x − 3
= 4x2 − (x − 3)2 ⇒ ±2x = x − 3 ⇒ x = −3, 1.
Solution. |xA − B| =
x − 3 4x
Let x1 = (x1 , x2 ) be a row vector such that (A − B) xx12 = 0.


  
1 −2 x1
= 0 ⇒ x1 − 2x2 = 0
−2 4 x2

We take x1 = 2, x2 = 1, so x1 = (2, 1).


Let x2 = (x1 , x2 ) be a row vector such that (−3A − B) xx12 = 0.


  
−3 −6 x1
= 0 ⇒ x1 + 2x2 = 0
−6 −12 x2

We take x1 = −2, x2 =1, so  x2 = (−2,1).


x1 Ax01 = (2, 1) 11 14 21 =(2, 1) 36 = 12.
x2 Ax02 = (−2, 1) 11 14 −2 1 = (−2, 1) −12 = 4.
0
Note that x1 Ax2 = 0. 
x1 Bx01 = (2, 1) 03 30 21 =(2, 1) 36 = 12.


x2 Bx02 = (−2, 1) 03 30 −2 1 = (−2, 1) −63


= −12.
0
Note that x1 Bx2 = 0.
√1 0
Thus if P = [x1 0 , x2 0 ], then P0 AP = 12 0 , and P0 BP = 12 0
 
0 4 0 −12 . Let Q = 12
0 12
,
then Q0 P0 APQ = 10 01 and Q0 P0 BPQ = 10 −3
 0

as desired. Thus the required non-singular
linear transformation is PQ.
−1
− tan 2θ tan 2θ
   
cos θ − sin θ 1 1
Question 3(b) Show that = .
sin θ cos θ tan 2θ 1 − tan 2θ 1

Solution.
− tan 2θ cos2 2θ − sin 2θ cos 2θ
  
1
R.H.S = θ
tan 2 1 sin 2θ cos 2θ cos2 2θ
cos2 2θ − sin2 2θ −2 sin 2θ cos 2θ
 
= = L.H.S
2 sin 2θ cos 2θ − sin2 2θ + cos2 2θ

4
 
0 1
Question 3(c) Verify the Cayley-Hamilton theorem for A = .
−2 3

−λ 1
Solution. The characteristic equation for A is = 0 ⇒ −3λ + λ2 + 2 = 0
−2 3 − λ
2
Thus according
     theorem A − 3A + 2I = 0.
to the Cayley-Hamilton
0 1 0 1 −2 3
A2 = =
 −2
 3  −2 3  −6 7  
−2 3 0 1 1 0 0 0
−3 +2 =
−6 7 −2 3 0 1 0 0  
0 1
Thus the Cayley Hamilton theorem is verified for A = .
−2 3

5
UPSC Civil Services Main 1990 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

1 Linear Algebra
Question 1(a) State any definition of the determinant of an n×n matrix and show that the
determinant function is multiplicative i.e. det AB = det A det B for any two n × n matrices
A, B. You may assume the matrices to be real.
Solution. Let π be a permutation of 1, . . . , n. Define sign(π) as follows: count the number
of pairs of numbers that need to be interchanged to get to π from the identity permutation.
If this is even, the sign is 1, and if it is odd, the sign is −1. Now if Π is the set of all
permutations of 1, . . . , n, define
X Y
det A = sign(π) aiπ(i)
π∈Π i

where aij are the elements of A.


Note that the det A is n-linear i.e. if we perform any row or column operation on A the
determinant is unchanged. Also, if any two rows are swapped, the sign of the determinant
changes. These are simple consequences of the above definition.
Consider the 2n × 2n matrix

a11 a12 . . . a1n 0 . . . 0

 a21 a22 . . . a2n 0 . . . 0 
 . .. .. .. .. 
 .
 . . . . . 

a a . . . ann 0 . . . 0 
 
P =  n1 n2
 −1 0 . . . 0 b11 . . . b1n 

 0 −1 . . . 0 b21 . . . b2n 
 
 . .. .. .. .. 
 .. . . . . 
0 0 . . . −1 bn1 . . . bnn

1
Then det P = det A det B, because if for any permutation π, π(i) > n for i ≤ n, then
the corresponding element of the sum is 0 as aiπ(i) = 0. Thus π(i) ≤ n if i ≤ n, and
consequently π(j) > n if j > n. So each permutation consists of a permutation of 1, . . . , n
and a permutation of n + 1, . . . , 2n, consequently we can factor the sum, to get det P =
det A det B.
Now we perform a series of column operations to P — add b11 C1 + . . . + bn1 Cn to Cn+1 ,
to get 
a11 a12 . . . a1n c11 0 . . . 0

 a21 a22 . . . a2n c21 0 . . . 0 
 . .. .. .. .. 
 .
 . . . . . 

an1 an2 . . . ann cn1 0 . . . 0 
 
 −1 0 . . . 0 0 b12 . . . b1n 
 
 0 −1 . . . 0 0 b22 . . . b2n 
 
 . .. .. .. .. .. 
 .. . . . . . 
0 0 . . . −1 0 bn2 . . . bnn
where C = AB = (cij ). Similarly add b12 C1 + . . . + bn2 Cn to Cn+2 , . . ., b1n C1 + . . . + bnn Cn
to C2n to get  
 A  C
 −1 0 . . . 0 
 
 0 −1 . . . 0  
  0
 ...  
0 0 . . . −1
We can now verify that det P = det C. Any permutation π that leads to a non-zero term
in the determinant sum must have π(j) = j − n for j > n, thus piπ(i) = −1, i > n. Also
π(j) > n for j ≤ n, so any such π can be written as a permutation of 1, . . . , n followed by a
series of swaps of the i-th number with the (n + i)-th number, which is n + i. Also sign(π)
is the same as the sign of the corresponding permutation π 0 of 1, . . . , n — we first do π 0 by
exchanges and then additionally swap the i-th element with the (i + n)-th element, for each
i ≤ n. Now if n is even, this involves an even number of additional swaps, and multiply by
(−1)n corresponding to piπ(i) for i > n, otherwise we get an odd number of additional swaps,
flipping the sign, but we still multiply by (−1)n = −1.
Thus det P = det C = det A det B.

Question 1(b) Prove Laplace’s formula for simulataneous expansion of the determinant by
the first row and column; that given an (n+1)×(n+1) matrix in the block form M = αγ D β

,
where α is a scalar, β is a 1 × n matrix (a row vector), γ is a n × 1 matrix (a column vector),
and D is an n × n matrix, then det M = α det D − βD0 γ 0 , where D0 is the matrix of cofactors
of D and βD0 γ 0 stands for the matrix product of size 1 × 1.

Solution. Let M = (aij ), 1 ≤ i, j ≤ n + 1. Thus α = a11 , β = (a12 . . . a1,n+1 ),

2
   
a21 a22 . . . a2,n+1
γ =  ...  and D =  ... ..
.
   
.
an+1,1 an+1,2 . . . an+1,n+1
det M = a11 |A11 | − a12 |A12 | + . . . + (−1)n a1,n+1 |A1,n+1 | where Aij is the minor corre-
sponding to aij (formed Pby deleting the i-th row and j-th column of A). Clearly D = A11 ,
n+1
so det M = α det D − j=2 (−1)j a1j det A1j . Now

a21 a 22 . . . a 2,j−1 a 2,j+1 . . . a 2,n+1


a31 a32 . . . a3,j−1 a3,j+1 . . . a3,n+1
|A1j | = ..

.. .. .. ..
. . . . .

an+1,1 an+1,2 . . . an+1,j−1 an+1,j+1 . . . an+1,n+1

Let Bij be the minor of aij in D. Expanding |A1j | in terms of the first column, we get

|A1j | = a21 |B2j | − a31 |B3j | + . . . + (−1)n+1 an+1,1 |Bn+1,1 |

n+1 X
X n+1
det M = α det D − a1j ai1 |Bij |(−1)i (−1)j
j=2 i=2
 
a21
= α det D − (a12 a13 . . . a1,n+1 )(cij )  ... 
 
an+1,1
= α det D − βD0 γ

where cij = (−1)i+j |Bij |, thus D0 = (cij ) is the matrix of cofactors of D.

Question 1(c) For M as in 1(b), if D is invertible, show that det M = det D(α − βD−1 γ).

Solution. If D is invertible, then DD0 = D0 D = (det D)I ⇒ D0 = D−1 det D. So


det M = α det D − βD0 γ = α det D − βD−1 det Dγ = det D(α − βD−1 γ).

Question 2(a) Write the definition of the characteristic polynomial, eigenvalues and eigen-
vectors of a square matrix. Also say briefly something about the importance and/or applica-
tions of these notions.

Solution. Let A be an n × n real or complex matrix. The polynomial |xIn − A| is called


the characteristic polynomial of A. The roots of this polynomial are called the eigenvalues
of A. If λ is an eigenvalue of A, then all the non-zero vectors x such that Ax = λx are
called eigenvectors of A corresponding to λ.
Many problems in mathematics and other sciences require finding eigenvalues and eigen-
vectors of an operator.

3
• Eigenvalues can be used to find a very simple matrix for an operator — either diagonal
or a block diagonal form. This can be used to compute powers of matrices quickly.

• If one wishes to solve a linear differential system like x0 = Ax, or study the local
properties of a nonlinear system, finding the diagonal form of the matrix can give us
a decoupled form of the system, allowing us to find the solution or understand its
qualitative behavior, like its stability and oscillatory behavior.

• The calculation of Google’s Pagerank is essentially the computation of the principal


eigenvector (corresponding to the eigenvalue with the largest absolute value) of a very
large matrix (the adjacency matrix of the web graph) — this is used to find the relative
importance of documents on the World Wide Web. Similar calculations are used to
compute the stationary distribution of a Markov system.

• In mechanics, the eigenvectors of the inertia tensor are used to define the principal
axes of a rigid body, which are important in analyzing the rotation of the rigid body.

• Eigenvalues can be used to compute low rank approximations to matrices, which help
in reducing the dimensionality of various problems. This is used in statistics and
operations research to explain a large number of observables in terms of a few hidden
variables + noise.

• Eigenvalues can help us determine the form of a quadric or higher dimensional surface
— see the relevant section in year 1999.

• In quantum mechanics, states are represented by unit vectors, while observable quan-
tities (like position and energy) are represented by Hermitian matrices. The basic
problem in any quantum system is the determination of the eigenvalues and eigenvec-
tors of the energy matrix. The eigenvalues are the observed values of the observable
quantity, and discreteness of the eigenvalues leads to the quantization of the observed
values.

Question 2(b) Show that a Hermitian matrix possesses a set of eigenvectors which form
an orthonormal basis. State briefly how or why a general n × n complex matrix may fail to
possess n linearly independent eigenvectors.

Solution. Let H be Hermitian, and λ1 , . . . , λn its eigenvalues, not necessarily distinct. Let
x1 with norm 1 be an eigenvector corresponding to λ1 . Then there exists (from a result
analogous to the result used in question 3(a), year 1995) a unitary matrix U such that x1 is
its first column. Therefore
 
−1 0 λ1 L
U1 HU1 = U1 HU1 =
0 H1

4
0
where H1 is (n − 1) × (n − 1) and L is (n − 1) × 1. Since U1 HU1 is Hermitian, it follows
that L = 0. Consequently  
0 λ1 0
U1 HU1 =
0 H1
Now H1 is Hermitian with eigenvalues λ2 , . . . , λn . Repeating the above argument, we find
U∗2 an (n − 1) × (n − 1) unitary matrix such that
 
0 ∗ λ 2 0
U∗2 H1 U2 =
0 H2
 
1 0
If U2 = then U2 is unitary, and
0 U∗2
 
λ1 0 0
0 0
U2 U1 HU1 U2 =  0 λ2 0 
0 0 H2
Repeating this process or by induction, we can get U unitary such that
 
λ1 · · · 0
0  .. .. 
U HU =  . .
0 ··· λn

If U = [C1 , C2 , . . . , Cn ], then C1 , C2 , . . . , Cn are eigenvectors of H and form an orthonormal


system.
A complex matrix A would fail to have n eigenvectors which are linearly independent
if A is not diagonalizable i.e. we cannot find P such that P−1 AP is a diagonal matrix.
For example if A = 10 1c , c 6= 0, and x1, x2 are two independent eigenvectors of A, then
−1 1 0 1 0

P = [x1 x2 ] would lead to P AP = 0 1 ⇒ A = 0 1 which is false.

Question 2(c) Define the minimal polynomial and show that a complex matrix is diago-
nalizable (i.e. conjugate to a diagonal matrix) if and only if the minimal polynomial has no
repeated root.

Solution. Given a complex n × n complex matrix A, if f (x) is a nonzero polynomial with


complex coefficients of least degree such that f (A) = 0, then f (x) is called the minimal
polynomial of A. The Cayley-Hamilton therem tells us that any n × n complex matrix A
satisfies the degree n polynomial equation |A − xI| = 0, so the minimal polynomial exists
and is of degree ≤ n.
A complex n × n matrix can be thought of as a linear transformation from Cn to Cn . Let
T : V −→ V, dim V Q = n. Let the minimal polynomial of T be p(x), having distinct roots
c1 , . . . , ck , so p(x) = kj=1 (x − cj ). We shall show that T is diagonalizable.
If k = 1, then the minimal polynomial is x−c, thus T−cI = 0, so T = cI is diagonalizable.
So assume k > 1.

5
k
x−ci
Q
Consider the polymonials pj = cj −ci
. Clearly pj (ci ) = 0 for i 6= j, and pi (ci ) = 1. This
i=1
i6=j
implies that the polynomials p1 , . . . , pk are linearly independent, and each one is of degree
k − 1 < k. Thus these form a basis of the space Pof polynomials of degree ≤ k − 1. Thus
k
given any polynomial g of degree ≤ k − 1, g = i=1 αi pi , where αi = g(ci ). In particular,
1 = ki=1 pi , x = ki=1 ci pi . Thus
P P

k
X k
X
I= pi (T), T = ci pi (T)
i=1 i=1

Moreover pi (T)pj (T) = 0, i 6= j because pi (x)pj (x) is divisible by the minimal polynomial
of T. Also pj (T) 6= 0, 1 ≤ j ≤ k, because the degree of pj is less than k, the degree of the
minimal polynomial of T.
Set Vi = pi (T)V, then V = I(V) = ki=1 pi (T)V = V1 + . . . + Vk . We shall now show that
P
Vi = Vci , the eigenspace of T with respect to ci .
v ∈ Vi ⇒ v = pi (T)w for some w ∈ V. Since (x−ci )pi is divisible by p, (T−ci I)v = 0, so
Tv = ci v so v ∈ Vci . Conversely, if v ∈ Vci , then Tv = ci v, or (T−ci I)v = 0 ⇒ pj (T)v = 0
for j 6= i. Since
Pv = pi (T)v + . . . + pk (T)v, we get v = pi (T)v ⇒ v ∈ Vi .
Thus V = ki=1 Vci so V has a basis consisting of eigenvectors, so T is diagonalizable.
Conversely let T be diagonalizable, then we shall show that the minimal polynomial of
T has distinct roots. Let  
λ1 0 . . . 0
 0 λ2 . . . 0 
−1
P TP =  ..
 
.. 
. .
0 0 . . . λn
and out of λ1 , . . . , λn , let λ1 , . . . , λk be distinct. Let g(x) = (x − λ1 ) . . . (x − λk ). Then
v ∈ V ⇒ v = v1 + . . . + vk where vi ∈ Vλi , the eigenspace of λi . Thus g(T)(v) = 0, so
g(T) = 0. Thus g(x) is divisible by the minimal polynomial of T. Since g(x) has all distinct
roots, it immediately follows that the minimal polynomial also has all distinct roots.

a b is expressible in the form LDU, where
Question 3(a) Show  that a 2 × 2 matrix M = c d 
L has the form α1 01 , D is diagonal and U has the form 10 β1 If and only if either a 6= 0
or a = b = c = 0. Also show that when a 6= 0 the factorization M = LDU is unique.

Solution. Given M = ac db , suppose M = LDU = α1 01 a01 a02 10 β1 = aa11α a02 01 β1 =


     
a1 a1 β

a1 α a1 αβ+a2 . Thus M = LDU ⇒ a1 = a, a1 β = b, a1 α = c, a αβ + a = d. Thus if a = 0,
 010  1 β 2
0 0 1 0
then b = c = 0 and d = a2 , In this case, M = 0 d = α 1 0 d 0 1 whatever α, β may
be, i.e. M can be represented as LDU in infinitely many ways.
If a 6= 0, then a1 = a, β = ab , α = ac , a2 = d − bca are uniquely determined. Thus
 a 0  b
M = ac db = 1ac 01 0 d− bca 10 1a and has a unique representation.


6
   
Conversely, if M = 00 d0 , i.e. a = b = c = 0, then M = α1 01 00 d0 10 β1 for any
a 0  b 
α, β ∈ R. If M = ac db , a 6= 0, then M = LDU with L = 1ac 01 , D = 0 d− bca , U = 10 1a
 

as shown above.

Question 3(b) Suppose a real matrix has eigenvalue λ, possibly complex. Show that there
exists a real eigenvector for λ if and only if λ is real.

Solution. If λ is real, then the n × n matrix A − λI defines a linear transformation from


Rn to Rn . Since |A − λI| = 0, the rows are linearly dependent, so there exists x ∈ Rn , x 6= 0
such that (A − λI)x = 0 ⇒ Ax = λx. Thus there exists a real eigenvector for λ.
Conversely, suppose Ax = λx, x ∈ Rn , x 6= 0. then λx = Ax = Ax = λx = λx ⇒
(λ − λ)x = 0 ⇒ λ − λ = 0 ∵ x 6= 0 ⇒ λ = λ i.e. λ is real.

n
Question 3(c)  If a 2×2 matrix
 A has order n, i.e. A = I2 , then show that A is conjugate
cos θ sin θ
to the matrix where θ = 2πm for some integer m.
− sin θ cos θ n

Solution. Note: A has to be real, otherwise the result  is false: if α1 , α2 are two distinct
α1 0
n-th roots of unitysuch that α1 6= α2 , then A = 0 α2 has order n, but A is not conjugate
cos θ sin θ
to whose eigenvalues are complex conjugates of each other.
− sin θ cos θ
An = I ⇒ eigenvalues of A are n-th roots of unity. If A has repeated eigenvalues, then
these can be 1 or −1, because eigenvalues of real matrices are complex conjugates of each
other, so the repeated eigenvalues must be real, and they also must be roots of 1.
−1

1 c .
Case 1: A has eigenvalues 1, 1. There exists P non-singular such that P AP = 0 1
Now (P−1 AP)n = 10 nc = P−1An P = P−1 I2 P = I2 , so nc = 0 ⇒ c = 0. Thus A is

1
cos θ sin θ
conjugate to I2 = , θ = 2πn .
− sin θ cos θ n

Case 2: A has eigenvalues −1, −1. There exists P non-singular such that P−1 AP =
−1 c −1 n −1 nc
 1 nc
 n
0 −1 . Now P A P = 0 −1 or 0 1 , according as nis odd or even.  But A = I2 ,
cos θ sin θ
therefore n is even and c = 0. Thus A is conjugate to −I2 = , θ = 2πm ,m =
− sin θ cos θ n
n
2
.
Case 3: A hasdistinct eigenvalues
 λ1 , λ2 . Then λ1 = λ2 . If λ1 = cos θ + i sin θ, with
2πm cos θ sin θ cos θ − λ sin θ
θ = n , set B = . The eigenvalues of B are roots of =
− sin θ cos θ − sin θ cos θ − λ
0 ⇒ λ = cos θ ± i sin θ. Since A and B have the same eigenvalues λ1 , λ2 distinct, both are
λ1 0
conjugate to 0 λ2 and are therefore conjugate to each other.

7
UPSC Civil Services Main 1991 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

Question 1(a) Let V(R) be the real vector space of all 2×3 matrices with real entries. Find
a basis of V(R). What is the dimension of V(R).
     
1 0 0 0 1 0 0 0 1
Solution. Let A1 = , A2 = , A3 =
  0 0 0  0 0 0  0 0 0
0 0 0 0 0 0 0 0 0
and B1 = , B2 = , B3 = . Clearly Ai , Bi , i = 1, 2, 3 ∈
1 0 0 0 1 0 0 0 1
V(R). These generate V(R) because
 
a1 a2 a3
A= = a1 A1 + a2 A2 + a3 A3 + b1 B1 + b2 B2 + b3 B3
b1 b2 b3

for any arbitrary element A ∈ V(R).


 They  are linearly independent because if the RHS in the above equation was equal to
0 0 0
, then ai = 0, bi = 0 for i = 1, 2, 3. Thus Ai , Bi , i = 1, 2, 3 is a basis for V(R) and
0 0 0
the dimension of V(R) is 6.

Question 1(b) Let C be the field of complex numbers and let T be the function from C3 to
C3 defined by

T(x1 , x2 , x3 ) = (x1 − x2 + 2x3 , 2x1 + x2 , −x1 − 2x2 + 2x3 )

1. Verify that T is a linear transformation.

2. If (a, b, c) ∈ C3 , what are the conditions on a, b, c so that (a, b, c) is in the range of T?


What is the rank of T?

1
3. What are the conditions on a, b, c so that (a, b, c) is in the null space of T? What is
the nullity of T?

Solution. T(e1 ) = (1, 2, −1), T(e2 ) = (−1, 1, −2), T(e3 ) = (2, 0, 2). Clearly T(e1 ) and
T(e3 ) are linearly independent. If

(−1, 1, 2) = α(1, 2, −1) + β(2, 0, 2)

then α+2β = −1, 2α = 1, −α+2β = −2, so α = 12 , β = − 43 , so T(e2 ) is a linear combination


of T(e1 ) and T(e3 ). Thus rank of T is 2, nullity of T is 1.
If (a, b, c) is in the range of T, then (a, b, c) = α(1, 2, −1) + β(2, 0, 2). Thus α + 2β =
a− b
a, 2α = b, −α + 2β = c. From the first two equations, α = 2b , β = 2 2 . The equations would
be consistent if − 2b + a − 2b = c, or a = b + c. So the condition for (a, b, c) to belong to the
range of T is a = b + c.
If (a, b, c) ∈ null space of T, then a − b + 2c = 0, 2a + b = 0, −a − 2b + 2c = 0. Thus
3a + 2c = 0, so a = − 2c 3
, b = 4c
3
. Thus the conditions for (a, b, c) to belong to the null space of
T are 3a + 2c = 0, 3b = 4c. Thus the null space consists of the vectors {(− 2c , 4c , c) | c ∈ R},
3 3
showing that the nullity of T is 1.

Question 1(c) If A = 1 2
1 3 , express A6 − 4A5 + 8A4 − 12A3 + 14A2 as a linear polynomial
in A.

Solution. Characteristic polynomial of A is 1−λ 2 2
1 3−λ = (λ − 3)(λ − 1) − 2 = λ − 4λ + 1.
By the Cayley Hamilton theorem, A2 − 4A + I = 0. Dividing the given polynomial by
A2 − 4A + I, we have

A6 − 4A5 + 8A4 − 12A3 + 14A2


= A4 (A2 − 4A + I) + 7A4 − 12A3 + 14A2
= (A4 + 7A2 )(A2 − 4A + I) + 16A3 + 7A2
= (A4 + 7A2 + 16A)(A2 − 4A + I) + 71A2 − 16A
= (A4 + 7A2 + 16A + 71I)(A2 − 4A + I) + 268A − 71I

Since A2 − 4A + I = 0, A6 − 4A5 + 8A4 − 12A3 + 14A2 = 268A − 71I.

Question 2(a) Let T : R2 −→ R2 be a linear transformation defined by T(x1 , x2 ) =


(−x2 , x1 ).

1. What is the matrix of T in the standard basis of R2 ?

2. What is the matrix of T in the ordered basis B = {α1 , α2 } where α1 = (1, 2), α2 =
(1, −1)?

2
Solution.  T(e1 ) = (0, 1) = e2 , T(e2 ) = (−1, 0) = −e1. Thus (T(e1 ), T(e2 )) =
(e1 e2 ) 01 −1 0 −1
0 . So the matrix of T in the standard basis is 1 0 .
T(α1 ) = (−2, 1), T(α2 ) = (1, 1). If (a, b) = xα1 + yα2 , then x + y = a, 2x − y = b, so
x = a+b
3
, y = 2a−b
3
. This shows that
T(α1 ) = (−2, 1) = − 13 α1 − 35 α2
T(α2 ) = (1, 1) = 23 α1 + 13 α2
 1 2
−3 3
Thus (T(α1 ) T(α2 )) = (α1 α2 ) 5 1 . Consequently the matrix of T in the ordered
 1 2 − 3 3
−3 3
basis B is .
− 53 31

Question 2(b)
 Determine
 a non-singular matrix P such that P0 AP is a diagonal matrix,
0 1 2
where A = 1
 0 3. Is the matrix congruent to a diagonal matrix? Justify your answer.
2 3 0
Solution. The quadratic form associated with A is Q(x, y, z) = 2xy + 4xz + 6yz. Let
x = X, y = X + Y, z = Z (thus X = x, Y = y − x, Z = z). Then
Q(X, Y, Z) = 2X 2 + 2XY + 4XZ + 6XZ + 6Y Z
= 2X 2 + 2XY + 10XZ + 6Y Z
Y 5 2 Y 2 25 2
= 2(X + + Z) − − Z +YZ
2 2 2 2
Y 5 2 1
= 2(X + + Z) − (Y − Z)2 − 12Z 2
2 2 2
Put
Y 5 x y 5z
ξ = X+ + Z= + +
2 2 2 2 2
η = Y − Z = −x + y − z
ζ = Z=z
   1 1 5 −1   
1 − 12 −3
 
x 2 2 2
ξ ξ
y  = −1 1 −1 η  = 1 1 −2 η 
2
z 0 0 1 ζ 0 0 1 ζ
Q(x, y, z) transforms to 2ξ 2 − 21 η 2 − 12ζ 2 . Thus
 
2 0 0
P0 AP = 0 − 12 0 
0 0 −12
1 − 12 −3
 

with P = 1 21 −2 Clearly A is congruent to a diagonal matrix as shown above.


0 0 1

3
Question 2(c) Reduce the matrix
 
1 3 4 −5
−2 −5 −10 16 
 
5 9 33 −68
4 7 30 −78

to echelon form by elementary row transformations.

Solution. Let the given matrix be A. Operations R2 + 2R1 , R3 − 5R1 , R4 − 4R1 ⇒


 
1 3 4 −5
0 1 −2 6 
A≈ 0 −6 13

−43
0 −5 14 −58

Operations R3 + 6R2 , R4 + 5R2 ⇒


 
1 3 4 −5
0 1 −2 6 
A≈ 
0 0 1 −7 
0 0 4 −28

Operations R4 − 4R3 ⇒  
1 3 4 −5
0 1 −2 6 
A≈ 
0 0 1 −7
0 0 0 0
Operation R1 − 3R2 ⇒  
1 0 10 −23
0 1 −2 6 
A≈ 
0 0 1 −7 
0 0 0 0
Operations R1 − 10R3 , R2 + 2R3 ⇒
 
1 0 0 47
0 1 0 −8
A≈ 
0 0 1 −7
0 0 0 0

which is the required row echelon form. The rank of A is 3.

4
Question 3(a) U is an n-rowed unitary matrix such that |I − U| 6= 0, show that the matrix
H defined by iH = (I + U)(I − U)−1 is Hermitian. If eiα1 , . . . , eiαn are the eigenvalues of U
then cot α21 , . . . , cot α2n are eigenvalues of H.
Solution.
(iH)(I − U) = (I + U)
0 0 0
⇒ (I − U )(iH) = (I + U )

0 0 0 0
Substituting I = U U, we have from the second equation that U (U − I)(iH) = U (U + I).
0 0 0
So (iH) = −iH = −(I + U)(I − U)−1 = −iH, so H = H, thus H is Hermitian.
If an eigenvalue of a nonsingular matrix A is λ, then λ−1 is an eigenvalue of A−1 ∵ Ax =
λx ⇒ λ−1 x = A−1 x, note that λ 6= 0 ∵ |A| 6= 0. Thus the eigenvalues of H are
1 1 + eiαj
,1 ≤ j ≤ n
i 1 − eiαj
eiαj /2 + e−iαj /2
= −i −iαj /2 ,1 ≤ j ≤ n
e − eiαj /2
eiαj /2 +e−iαj /2
2
= ,1 ≤ j ≤ n
e−iαj /2 −eiαj /2
2i
cot αj
= ,1 ≤ j ≤ n
2

Question 3(b) Let A be an n × n matrix with distinct eigenvalues λ1 , . . . , λn . Show that if


A is non-singular then there exist 2n matrices X such that X2 = A. What happens in case
A is a singular matrix?
Solution. There exists √P non-singular
√ such that P−1 AP = diagonal[λ1 , . . . , λn ].
Let Y1 = diagonal[ λ1 , . . . , λn ], and let X = PYP−1 . Then X2 = PYP−1 PYP−1 =
−1
PY2 P = A. Thus any of the 2n matrices √ formed
√ by choosing a sign for each of the
−1
diagonal entries from X = P diagonal[± λ1 , . . . , ± λn ] P has the same property (note
that they are all distinct).
If one of the eigenvalues is zero, the number of matrices X would become 2n−1 , since we
would have one less choice.
Question 3(c) Show that a real quadratic x0 Ax is positive definite if and only if there exists
a non-singular matrix B such that A = B0 B.
Solution. If A = B0 B, then x0 Ax = x0 B0 Bx = X0 X, where X = Bx. Now if x 6= 0, then
Bx 6= 0, as B is nonsingular, and 0 is not its eigenvalue. Thus x0 Ax = X0 X > 0, so x0 Ax
is positive definite.
Conversely, see the result used in the solution of question 2(c), year 1992.

5
UPSC Civil Services Main 1992 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

Question 1(a) Let U and V be vector spaces over a field K and let V be of finite dimension.
Let T : V −→ U be a linear transformation, prove that dim V = dim T(V) + dim nullity T.

Solution. See question 3(a), year 1998.

Question 1(b) Let S = {(x, y, z) | x + y + z = 0, x, y, z ∈ R}. Prove that S is a subspace


of R3 . Find a basis of S.

Solution. S = 6 ∅ because (0, 0, 0) ∈ S. If (x1 , y1 , z1 ), (x2 , y2 , z2 ) ∈ S then α1 (x1 , y1 , z1 ) +


α2 (x2 , y2 , z2 ) ∈ S because (α1 x1 + α2 x2 ) + (α1 y1 + α2 y2 ) + (α1 z1 + α2 z2 ) = α1 (x1 + y1 + z1 ) +
α2 (x2 + y2 + z2 ) = 0. Thus S is a subspace of R3 .
Clearly (1, 0, −1), (1, −1, 0) ∈ S and are linearly independent. Thus dim S ≥ 2. However
(1, 1, 1) 6∈ S, so S = 6 R3 . Thus dim S = 2 and {(1, 0, −1), (1, −1, 0)} is a basis for S.

Question 1(c) Which of the following are linear transformations?

1. T : R −→ R2 defined by T(x) = (2x, −x).

2. T : R2 −→ R3 defined by T(x, y) = (xy, y, x).

3. T : R2 −→ R3 defined by T(x, y) = (x + y, y, x).

4. T : R −→ R2 defined by T(x) = (1, −1).

Solution.

1
1.

T(αx + βy) = (2αx + 2βy, −αx − βy)


= (2αx, −αx) + (2βy, −βy)
= αT(x) + βT(y)

Thus T is a linear transformation.

2. T(2(1, 1)) = T(2, 2) = (4, 2, 2) 6= 2T(1, 1) = 2(1, 1, 1) Thus T is not a linear transfor-
mation.

3.

T(α(x1 , y1 ) + β(x2 + y2 )) = T(αx1 + βx2 , αy1 + βy2 )


= (αx1 + βx2 + αy1 + βy2 , αy1 + βy2 , αx1 + βx2 )
= α(x1 + y1 , y1 , x1 ) + β(x2 + y2 , y2 , x2 )
= αT(x1 , y1 ) + βT(x2 , y2 )

Thus T is a linear transformation.

4. T(2(0, 0)) = T(0, 0) = (1, −1) 6= 2T(0, 0) Thus T is not a linear transformation.

Question 2(a) Let T : M2,1 −→ M2,3 be a linear transformation defined by (with the usual
notation)        
1 2 1 3 1 6 1 0
T = ,T =
0 4 1 5 1 0 0 2
 
x
Find T .
y

Solution.
       
x 1 1 1
= x −y +y
y 0 0 1
       
x 2 1 3 6 1 0 2x + 4y x 3x − 3y
T = (x − y) +y =
y 4 1 5 0 0 2 4x − 4y x − y 5x − 3y

2
Question 2(b) For what values of η do the following equations

x+y+z = 1
x + 2y + 4z = η
x + 4y + 10z = η 2

have a solution? Solve them in each case.


1 1 1 
Solution. Since the determinant of the coefficient matrix 1 2 4 is 0, the system has to
1 4 10
be consistent to be solvable.
Clearly x + 4y + 10z = 3(x + 2y + 4z) − 2(x + y + z). Thus for the system to be consistent
we must have η 2 = 3η − 2, or η = 1, 2.
If η = 1, then x + y + z = 1, x + 2y + 4z = 1 so y + 3z = 0, or y = −3z, x = 1 + 2z. Thus
the space of solutions is {(1 + 2z, −3z, z) | z ∈ R}. Note that the rank of the coefficient
matrix is 2, and consequently the space of solutions is one dimensional.
If η = 2, then x + y + z = 1, x + 2y + 4z = 2, so y + 3z = 1 or y = 1 − 3z, hence x = 2z.
Consequently, the space of solutions is {(2z, 1 − 3z, z) | z ∈ R}.

Question 2(c) Prove that a necessary and sufficient condition of a real quadratic form
x0 Ax to be positive definite is that the leading principal minors of A are all positive.

Solution. Let all the principal minors be positive. We have to prove that the quadratic
form is positive definite. We prove the result by induction.
If n = 1, then a11 x2 > 0 ⇔ a11 > 0. Suppose as induction hypothesis the result is true
B B1 
for n = m. Let S = B01 k be a matrix of a quadratic form in m + 1 variables, where
B is m × m, B1 is m × 1 and k is a single element. Since all principle minors of B are
leading principal minors of S, and are hence positive, the induction hypothesis gives that B
is positive definite. This means that there exists a non-singular m × m matrix P such that
P0 BP = Im (We shall prove this presently). Let C be an m-rowed column to be determined
soon. Then
 0 
P0 BP P0 BC + P0 B1
   
P 0 B B1 P C
=
C0 1 B0 1 k 0 1 C0 B0 P + B01 P C0 BC + C0 B1 + B0 1 C + k

Let C be so chosen that BC + B1 = 0, or C = −B−1 B1 . Then


 0     0 
P 0 B B1 P C P BP 0
=
C0 1 B0 1 k 0 1 0 B0 1 C + k

Taking determinants, we get |P0 ||S||P| = B0 1 C + k, because P0 BP = Im , and B0 1 C + k is


0 0 2
a single element. Since |S|
 > 0,itfollows that
 B 1 C + k > 0, so let B 1 C + k = α . Then
P C Im 0
Q0 SQ = Im+1 with Q = . Thus the quadratic forms of S and Im+1 take
0 1 0 α−1
the same values. Hence S is positive definite, so the condition is sufficient.

3
The condition is necessary - Since x0 Ax is positive definite, there is a non-singular matrix
P such that P0 AP = I ⇒ |A||P|2 = 1 ⇒ |A| > 0.
Let 1 ≤ r < n. Let xr+1 = . . . = xn = 0, then we obtain a quadratic form in r variables
which is positive definite. Clearly the determinant of this quadratic form is the r×r principal
minor of A which shows the result.
Proof of the result used: Let A be positive definite, then there exists a non-singular P
such that P0 AP = I.
We will prove this by induction. If n = 1, then the form corresponding to A is a11 x2 and

a11 > 0, so that P = ( a11 ).
Take  
1 −a−1 11 a12 0 ... 0
0 
P1 =  ..
 

. (n − 1) × (n − 1) 
0
then  
a11 0 a13 ... a1n
 0 
 
P01 AP1 =  a13
 

 .. 
 . (n − 1) × (n − 1) 
a1n
Repeating this process, we get a non-singular Q such that
 
a11 0 ... 0
Q0 AQ =  ...
 
(n − 1) × (n − 1) 
0

Given the (n − 1) × (n − 1) matrix on the lower right, we get by induction P∗ s.t. P∗ 0 ((n −
1) × (n − 1) matrix)P∗ is diagonal. Thus ∃P, |P| 6= 0, P0 AP = [α1 , . . . , αn ] say. Take R =
√ √
diagonal[ α1 , . . . , αn ], then R0 P0 APR = In .

2 1

Question 3(a) State the Cayley-Hamilton theorem and use it to find the inverse of 4 3 .

Solution. Let A be an n × n matrix. If |λI − A| = λn + a1 λn−1 + . . . + an = 0 is the


characteristic equation of A, then the Cayley-Hamilton theorem says that An + a1 An−1 +
. . . + an I = 0 i.e. a matrix satisfies its characteristic
 equation.
The characteristic equation of A = 4 3 is2 1


2 − λ 1
= λ2 − 5λ + 2 = 0
4 3 − λ

4
By the Cayley-Hamilton theorem, A2 − 5A + 2I = 0, so A(A − 5I) = −2I, thus A−1 =
− 12 (A − 5I). Thus
  3
− 21
   
−1 1 2 1 5 0
A =− − = 2
2 4 3 0 5 −2 1

Question 3(b) Transform the following into diagonal form


x2 + 2xy, 8x2 − 4xy + 5y 2
and give the transformation employed.
8 −2
 
Solution. Let A = 11 10 , B = −2 5

1 − 8λ 1 + 2λ
Let 0 = |A − λB| = = −5λ + 40λ2 − 4λ2 − 4λ − 1
1 + 2λ −5λ

Thus 36λ2 − 9λ − 1 = 0, so λ = 9± 81+144 72
= 13 , − 12
1
. 
x1 1 5 5
Let (x1 , x2 ) be the vector such that (A − λB) x2 = 0 with λ = 3 . Thus − 3 x1 + 3 x2 =
0 ⇒ x1 = x2 . We take x1 = 11 so that (A − λB)x1 = 0 with λ = 31 . Similarly, if (x1 , x2 ) is


the vector such that  (A − λB) xx12 = 0 with λ = − 12 1


, then 53 x1 + 56 x2 = 0, so 2x1 + x2 = 0.
1
We take x2 = −2 .
Now
x01 Ax1 = ( 1 1 ) 11 10 11 = ( 1 1 ) 21 = 3
  

x02 Ax2 = ( 1 −2 ) 11 10 −2 = ( 1 −2 ) −1
 1  
1 = −3
x01 Ax2 = ( 1 1 ) 11 10 −2 = ( 1 1 ) −1
 1
 
1 =0
If P = (x1 x2 ), then P0 AP = 30 −3
 
0
, thus x2 + 2xy ≈ 3X 2 − 3Y 2 by P = 11 −2 1
.
Similarly
x01 Bx1 = ( 1 1 ) −2 8 −2
 1 6

5 1 = (1 1) 3 = 9
x02 Bx2 = ( 1 −2 ) −2 8 −2
 1  12

= ( 1 −2 ) −12 = 36
0 8 −2
5
 1−2 12

x1 Bx2 = ( 1 1 ) −2 5 −2 = ( 1 1 ) −12 = 0

Thus P0 BP = 90 36 0 , so 8x2 − 4xy + 5y 2 is transformed to 9X 2 + 36Y 2 by X = P x


  
Y y

Question 3(c) Prove that the characteristic roots of a Hermitian matrix are all real, and
the characteristic roots of a skew Hermitian matrix are all zero or pure imaginary.
Solution. For Hermitian matrices, see question 2(c), year 1995.
0 0
If H is skew-Hermitian, then iH is Hermitian, because (iH) = iH = −iH = iH as
0
H = −H . Thus the eigenvalues of iH are real. Therefore the eigenvalues of H are −ix
where x ∈ R. So they must be 0 (if x = 0) or pure imaginary.

5
UPSC Civil Services Main 1993 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

Question 1(a) Show that the set S = {(1, 0, 0), (1, 1, 0), (1, 1, 1), (0, 1, 0)} spans the vector
space R3 but is not a basis set.

Solution. The vectors (1, 0, 0), (0, 1, 0), (1, 1, 1) are linearly independent, because α(1, 0, 0)+
β(1, 1, 0) + γ(1, 1, 1) = 0 ⇒ α + γ = 0, β + γ = 0, γ = 0 ⇒ α = β = γ = 0.
Thus (1, 0, 0), (1, 1, 0), (1, 1, 1) is a basis of R3 , as dimR R3 = 3.
Any set containng a basis spans the space, so S spans R3 , but it is not a basis because
the four vectors are not linearly independent, in fact (1, 1, 0) = (1, 0, 0) + (0, 1, 0).

Question 1(b) Define rank and nullity of a linear transformation. If V is a finite dimen-
sional vector space and T is a linear operator on V such that rank T2 = rank T, then prove
that the null space of T is equal to the null space of T2 , and the intersection of the range
space and null space of T is the zero subspace of V.

Solution. The dimension of the image space T(V) is called rank of T. The dimension of
the vector space kernel of T = {v | T(v) = 0} is called the nullity of T.
Now v ∈ null space of T ⇒ T(v) = 0 ⇒ T2 (v) = 0 ⇒ v ∈ null space of T2 . Thus null
space of T ⊆ null space of T2 . But we are given that rank T = rank T2 , so therefore nullity
of T = nullity of T2 , because of the nullity theorem — rank T + nullity T = dim V. Thus
null space of T = null space of T2 .
Finally if v ∈ range of T, and v ∈ null space of T, then v = T(w) for some w ∈ V. Now

T2 (w) = T(v) ⇒ w ∈ null space of T2


⇒ w ∈ null space of T
⇒ 0 = T(w) = v

Thus range of T∩ null space of T = {0}.

1
2
Question 1(c) If the  matrix of a linear operator T on R relative to the standard basis
1 1
{(1, 0), (0, 1)} is 1 1 , find the matrix of T relative to the basis B = {(1, 1), (−1, 1)}.

Solution. Let v1 = (1, 1), v2 = (−1, 1). Then T(v1 ) = (11) 11 11 = (2, 2) = 2v1 .
T(v2 ) = (−11) 11 11 = (0, 0) = 0. So (T(v1 ), T(v2 ) = (v1 ) v2 ) 20 00 , so the matrix of T
relative to the basis B is 20 00 .

A−1
   
A 0 0
Question 2(a) Prove that the inverse of is where A, C are
B C −C−1 BA−1 C−1
 
1 0 0 0
1 1 0 0
nonsingular matrices. Hence find the inverse of 
1 1 1 0.

1 1 1 1

Solution.
A−1
    
A 0 0 I 0
−1 = = Identity matrix.
B C −C−1 BA C−1  BA−1 − BA−1 I 
A−1
  
0 A 0 I 0
−1 −1 −1 = −1 −1 = Identity matrix, which shows
−C BA C B C −C B + C B I
the result.
B = 11 11  . Then A−1 = C−1 = −1 and C−1 BA−1 =
1 0
  1 0

Let A = C =  1 1 and
 1
1 0 1 1 1 0 1 0 0 1 = 0 1 Thus
−1 1 1 1 −1 1 = −1 1 0 1 0 0

 −1  
1 0 0 0 1 0 0 0
1 1 0 0
  −1 1 0 0

1 =
 0 −1 1

1 1 0 0
1 1 1 1 0 0 −1 1

Question 2(b) If A is an orthogonal matrix with the property that −1 is not an eigenvalue,
then show that A = (I − S)(I + S)−1 for some skew symmetric matrix S.

Solution. We want S skew symmetric such that A(I + S) = I − S i.e. A + AS = I − S


or AS + S = I − A or (I + A)S = I − A. Let S = (I + A)−1 (I − A), note that I + A is
invertible because if |I + A| = 0, then −1 will be an eigenvalue of A.
Note that the two factors of S commute, because (I+A)(I−A) = I−A2 = (I−A)(I+A),
so (I − A)(I + A)−1 = (I + A)−1 (I − A).

2
Now
S0 = (I − A)0 ((I + A)−1 )0
= (I − A0 )(I + A0 )−1
= (AA0 − A0 )(A0 A + A0 )−1
−1
= (A − I)A0 A0 (A + I)−1
= −(I − A)(I + A)−1
= −(I + A)−1 (I − A)
= −S
Thus S is skew symmetric, so A = (I − S)(I + S)−1 where S = (I + A)−1 (I − A)

Question 2(c) Show that any two eigenvectors corresponding to distinct eigenvalues of (i)
Hermitian matrices (ii) unitary matrices are orthogonal.
Solution. We first prove that the eigenvalues of a Hermitian matrix, and therefore of a
symmetric matrix, are real.
Let H be Hermitian, and λ be one of its eigenvalues. Let x 6= 0 be an eigenvector
0 0
corresponding to λ. Thus Hx = λx, so x0 Hx = x0 λx. But (x0 Hx) = (x0 Hx)0 = x0 H x =
0
x0 Hx, because H = H. Note that (x0 Hx)0 = x0 Hx, since it is a single element, therefore
0
x0 Hx is real. Similarly x0 x 6= 0 is real, so λ = xxHx 0 x is real.

Let H be Hermitian, Hx1 = λ1 x1 , Hx2 = λ2 x2 with λ1 6= λ2 . Clearly x02 Hx1 = λ1 x02 x1 ,


0 0
x01 Hx2 = λ2 x01 x2 . But (x02 Hx1 )0 = x01 H x2 = λ1 x01 x2 . So λ2 x01 x2 = x01 Hx2 = x01 H x2 =
0
λ1 x01 x2 because H = H. Since λ1 6= λ2 , x01 x2 = 0, so x1 , x2 are orthogonal.
Let U be unitary, Ux1 = λ1 x1 , Ux2 = λ2 x2 , where λ1 , λ2 are distinct eigenvalues of
0 0
U with corresponding eigenvectors x1 , x2 . Thus x02 U Ux1 = λ2 x02 λ1 x1 . Since U U = I,
λ2 x02 λ1 x1 = x02 x1 , so (1 − λ2 λ1 )(x02 x1 ) = 0. But 1 − λ2 λ1 = λ2 λ2 − λ2 λ1 = λ2 (λ2 − λ1 ) 6= 01 .
Thus x02 x1 = 0, so x1 , x2 are orthogonal.

Question 3(a) A matrix B of order n is of the form λA, where λ is a scalar and A has 1
everywhere except the diagonal, which has µ. Find λ, µ so that B may be orthogonal.
 
µ 1 ... 1
 1 µ . . . 1
Solution. A =  . . .
. B = λA. Thus
... 
1 1 ... µ
  
λµ λ . . . λ λµ λ . . . λ
 λ λµ . . . λ   λ λµ . . . λ 
B0 B =  . . .
  = BB0 = B2
...  . . . ... 
λ λ . . . λµ λ λ . . . λµ
1
We used here the fact that all eigenvalues of a unitary matrix have modulus 1. If Ux = λx, then
0 0
x U = λx0 . Thus x0 U Ux = λλx0 x, so x0 x = λλx0 x. Now x0 x 6= 0, so λλ = 1.
0

3
Clearly each diagonal element of BB0 is λ2 µ2 + (n − 1)λ2 , and each nondiagonal element is
2λ2 µ + (n − 2)λ2 . Thus B will be orthogonal if 2λ2 µ + (n − 2)λ2 = 0, λ2 µ2 + (n − 1)λ2 = 1.
Since λ 6= 0, µ = 2−n
2
= 1 − n2 , and λ2 = (1− n )12 +n−1 = 1
n2
= n42 , thus λ = ± n2 .
2 1−n+ 4
+n−1

Question 3(b) Find the rank of the matrix


 
1 −1 3 6
A = 1 3 −3 −4
5 3 3 11

by reducing it to its normal form.

Solution.  
    1 0 0 0
1 −1 3 6 1 0 0 0 1 0 0
A = 1 3 −3 −4 = 0 1 0 A 
0

0 1 0
5 3 3 11 0 0 1
0 0 0 1
Operation C2 + C1 , C3 − 3C1 , C4 − 6C1 ⇒
 
    1 1 −3 −6
1 0 0 0 1 0 0 0
1 4 −6 −10 = 0 1 0 A  1 0 0
0 0 1 0
5 8 −12 −19 0 0 1
0 0 0 1

Operation R2 − R1 ⇒
 
    1 1 −3 −6
1 0 0 0 1 0 0 0
0 4 −6 −10 = −1 1 0 A  1 0 0
0 0 1 0
5 8 −12 −19 0 0 1
0 0 0 1

Operation R3 − 2R2 ⇒
 
  1
  1 −3 −6
1 0 0 0 1 0 0 0
0 4 −6 −10 = −1 1 0 A  1 0 0
0 0 1 0
5 0 0 1 2 −2 1
0 0 0 1

Interchanging C3 and C4 we get


 
  1
  1 −6 −3
1 0 0 0 1 0 0 0
0 4 −10 −6 = −1 1 0 A  1 0 0
0 0 0 1
5 0 1 0 2 −2 1
0 0 1 0

4
R3 − 5R1 ⇒
 
   1  1 −6 −3
1 0 0 0 1 0 0 0
0 4 −10 −6 = −1 1 0 A  1 0 0
0 0 0 1
0 0 1 0 −3 −2 1
0 0 1 0

Operation 41 R2 ⇒
 
    1 1 −6 −3
1 0 0 0 1 0 0
0 1 − 5
0 1 0 0
2
− 32  = − 14 41 0 A 
0

0 0 1
0 0 1 0 −3 −2 1
0 0 1 0

Operation C3 + 52 C2 , C4 + 32 C2 ⇒

1 − 27 − 23
 
    1
1 0 0 0 1 0 0
0 1 0 0 = − 1 1 0 A 
0 1 52 3 
2 
4 4 0 0 0 1 
0 0 1 0 −3 −2 1
0 0 1 0
   
1 0 0 0 1 0 0
Thus the normal form of A is 0 1 0 0 so rank A = 3. P = − 14 14 0 and
0 0 1 0 −3 −2 1
7 3
 
1 1 −2 −2
0 1 5 3 
Q= 2 2  and PAQ is the normal form.
0 0 0 1 
0 0 1 0

Question 3(c) Determine the following form as definite, semidefinite or indefinite

2x21 + 2x22 + 3x23 − 4x2 x3 − 4x3 x1 + 2x1 x2

Solution. Completing the squares of the given form (say Q(x1 , x2 , x3 )):
1 3
Q(x1 , x2 , x3 ) = 2(x1 + x2 − x3 )2 + x22 + x23 − 2x2 x3
2 2
1 1
= 2(x1 + x2 − x3 )2 + (x3 − x2 )2 + x22
2 2
Thus Q can be written as the sum of 3 squares with positive coefficients, so it is positive
definite.

5
UPSC Civil Services Main 1994 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

Question 1(a) Show that f1 (t) = 1, f2 (t) = t − 2, f3 (t) = (t − 2)2 forms a basis of P3 =
{Space of polynomials of degree ≤ 2}. Express 3t2 −5t+4 as a linear combination of f1 , f2 , f3 .

Solution. If α1 f1 + α2 f2 + α3 f3 ≡ 0, then α3 being the coefficient of t2 is equal to 0. Then


coefficient of t is α2 so it must be 0, hence α1 = 0. Thus f1 , f2 , f3 are linearly independent.
Since {1, t, t2 } is a basis for P3 , its dimension is 3, hence f1 , f2 , f3 is a basis of P3 .
Now by Taylor’s expansion p(t) = 3t2 − 5t + 4 = p(2) + p0 (2)(t − 2) + p”(2) 2!
(t − 2)2 =
6 + 7(t − 2) + 3(t − 2)2 = 6f1 + 7f2 + 3f3 .

Question 1(b) Let T : R4 −→ R3 be defined by

T(a, b, c, d) = (a − b + c + d, a + 2c − d, a + b + 3c − 3d), a, b, c, d ∈ R

Verify that rank(T) + nullity(T) = dim(V4 (R).

Solution. Let

T(1, 0, 0, 0) = (1, 1, 1) = v1
T(0, 1, 0, 0) = (−1, 0, 1) = v2
T(0, 0, 1, 0) = (1, 2, 3) = v3
T(0, 0, 0, 1) = (1, −1, −3) = v4

T(R4 ) is generated by v1 , v2 , v3 , v4 and therefore a maximal independent subset of v1 , v2 , v3 , v4


will form a basis of T(R4 ). v1 , v2 are linearly independent because if αv1 + βv2 = 0, then
α − β = 0, α = 0 so α = β = 0.
v3 is dependent on v1 , v2 , because if v3 = αv1 + βv2 , then α − β = 1, α = 2, α + β =
3 ⇒ α = 2, β = 1 ∴ v3 = 2v1 + v2 .

1
v4 is dependent on v1 , v2 , because if v4 = αv1 + βv2 , then α − β = 1, α = −1, α + β =
−3 ⇒ α = −1, β = −2 ∴ v4 = −v1 − 2v2 .
Thus v1 , v2 is a basis of T(R4 ), so rank T = 2.
Now (a, b, c, d) ∈ ker T ⇔ a − b + c + d = 0, a + 2c − d = 0, a + b + 3c − 3d = 0
Choosing particular values of a, b, c, d, we see that (1, 2, 0, 1), (−1, 1, 1, 1) ∈ ker T and are
linearly independent, so dim ker T ≥ 2. But (1, 2, 0, 1), (−1, 1, 1, 1) generate ker T, because
if (a, b, c, d) ∈ ker T, and (a, b, c, d) = α(1, 2, 0, 1) + β(−1, 1, 1, 1), then α − β = a, 2α + β =
b, β = c, α + β = d, so α = a + c, β = c and these satisfy the remaining equations 2α + β =
b, α + β = d, because (a, b, c, d) ∈ ker T and therefore a − b + c + d = 0, a + 2c − d = 0. Thus
(a, b, c, d) = (a + c)(1, 2, 0, 1) + c(−1, 1, 1, 1), so dim ker T = nullity T = 2
Hence rank T + nullity T = 4 = dim(R4 ), as required.

Question 1(c) If T is an operator on R3 whose basis is B = {(1, 0, 0), (0, 1, 0), (0, 0, 1)}
such that  
0 1 1
[T : B] =  1 0 −1
−1 −1 0
find a matrix of T w.r.t. a basis B1 = {(0, 1, −1), (1, −1, 1), (−1, 1, 0)}.

Solution. The basis B is the standard basis, hence the representation of B1 in this basis is
as given. (Note that if B were some other basis, we would write B1 in that basis, and then
continue as below.) Let v1 = (0, 1, −1), v2 = (1, −1, 1), v3 = (−1, 1, 0). Then

T(v1 ) = (0, 1, −1) = v1


T(v2 ) = (0, 0, 0) = 0
T(v3 ) = (1, −1, 0) = −v3

Thus  
1 0 0
[T : B1 ] = 0 0 0 
0 0 −1
Note: The main idea behind the above solution is to express T(vi ) = ni=0 αii vi . Now
P
we solve for αii to get the matrix for T in the new basis.
An alternative is to compute P−1 [T : B]P, where P is given by [v1 , . . . , vn ] = [e1 , . . . , en ]P
or P = [v1 0 , . . . , vn 0 ]. Show that this is true.

Question 2(a) If A = haij i is an n × n matrix such that aii = n, aij = r if i 6= j, show that

[A − (n − r)I][A − (n − r + nr)I] = 0

Hence find the inverse of the n × n matrix B = hbij i where bii = 1, bij = ρ, i 6= j and
1
ρ 6= 1, ρ 6= 1−n .

2
Solution. Let C = A − (n − r)I, then every entry of C is r. Let D = A − (n − r + nr)I =
C − nrI. Thus CD = C2 − nrC. Each entry of C2 is nr2 , which is the same as each entry
of nrC, so CD = 0 as required.
The given equation implies

A[A − (2n − 2r + nr)I] = −(n − r)(n − r − nr)I

Let A = nB, where r = ρn. Thus A satisfies the conditions for the equation to hold, so
substituting A and r in the above equation

nB[nB − (2n − 2nρ + n2 ρ)I] = −(n − nρ)(n − nρ − n2 ρ)I


B[B − (2 − 2ρ + nρ)I] = −(1 − ρ)(1 − ρ − nρ)I
1
B−1 = − [B − (2 − 2ρ + nρ)I]
(1 − ρ)(1 − ρ − nρ)
1−2ρ+nρ
Thus the diagonal elements of B−1 are all (1−ρ)(1−ρ−nρ)
, while the off-diagonal elements are
2−3ρ+nρ
all (1−ρ)(1−ρ−nρ) .

Question 2(b) Prove that the eigenvectors corresponding to distinct eigenvalues of a square
matrix are linearly independent.

Solution. Let x1 , x2 , . . . , xr be eigenvectors corresponding to distinct eigenvalues λ1 , . . . , λr


of a matrix A.Q Let a1 x1 + . . . + ar xr = 0, ai ∈ R. We shall show that ai = 0, 1 ≤ Q i ≤ r.
Let L1 = ri=2 (A − λi I). Note that the factors of L1 commute. Thus L1 x2 = ri=3 (A −
λi I)(A − λ2 I)x2 = 0 because Ax2 = λ2 x2 . Similarly L1 x3 = . . . = L1 xr = 0. Moreover
L1 x1 = (λ1 − λ2 ) . . . (λ1 − λr )x1 .
Consequently

0 = L1 (a1 x1 + . . . + ar xr )
= a1 L 1 x 1
= a1 (λ1 − λ2 ) . . . (λ1 − λr )x1

λ1 − λi 6= 0, 2 ≤ i ≤ r, and Q
x1 6= 0 so a1 = 0.
Similarly taking Li = rj=1 (A − λj I), we show that ai = 0 for 1 ≤ i ≤ r. Thus
i6=j
x1 , x2 , . . . , xr are linearly independent.

Question 2(c) Determine the eigenvalues and eigenvectors of the matrix


 
3 1 4
A = 0 2 6
0 0 5

3
Solution. The characteristic polynomial of A is |λI − A| = (λ − 3)(λ − 2)(λ − 5).1 Thus
the eigenvalues of A are 3, 2, 5.
If x = (x1 , x2 , x3 ) is an eigenvector corresponding to λ = 3 then
  
0 1 4 x1
(A − 3I)x = 0 −1 6
   x2  = 0
0 0 2 x3

Thus x2 + 4x3 = 0, −x2 + 6x3 = 0, 2x3 = 0, take x1 = 1, x2 = 0 to get (1, 0, 0) as an


eigenvector for λ = 3. All the eigenvectors are (x1 , 0, 0), x1 6= 0.
If x = (x1 , x2 , x3 ) is an eigenvector corresponding to λ = 2 then
  
1 1 4 x1
(A − 2I)x = 0 0 6
   x2  = 0
0 0 3 x3

Thus x1 + x2 + 4x3 = 0, 6x3 = 0, 3x3 = 0, take x1 = 1 to get (1, −1, 0) as an eigenvector for
λ = 2. All the eigenvectors are (x1 , −x1 , 0), x1 6= 0.
If x = (x1 , x2 , x3 ) is an eigenvector corresponding to λ = 5 then
  
−2 1 4 x1
(A − 5I)x =  0 −3 6   x2  = 0
0 0 0 x3

Thus −2x1 + x2 + 4x3 = 0, −3x2 + 6x3 = 0, take x3 = 1 to get (3, 2, 1) as an eigenvector for
λ = 5. All eigenvectors are (3x3 , 2x3 , x3 ), x3 6= 0.

Question 3(a) Show that the matrix congruent to a skew symmetric matrix is skew sym-
metric. Use the result to prove that the determinant of a skew symmetric matrix of even
order is the square of a rational function of its elements.

Solution. Let B = P0 AP, be congruent to A, and A0 = −A. Then B0 = P0 A0 P =


−P0 AP = −B, so B is skew symmetric.
We prove the second result by induction on m, where n = 2m  is the order of the skew
0 a 2
symmetric matrix under consideration. If m = 1 then A = −a 0 , |A| = a , so the result is
true for m = 1.
Assume by induction that the result is true for all skew symmetric matrices of even order
< 2m. If A ≡ 0, there is nothing to prove. Otherwise there exists at least one non-zero
element aij . Changing the first row with row j, we move aij in the first row. Changing
1
Note that the determinant of an upper diagonal or a lower diagonal matrix is just the product of the
elements on the main diagonal.

4
column 1 and column j, we get −aij in the symmetric position. Now by multiplying the new
matrix by suitable elementary matrices on the left and right, we get
 
0 aij ∗ ∗ ... ∗
−aij 0 ∗ ∗ ... ∗
0
 
P AP =  ∗ ∗ 

 ... A2m−2 
∗ ∗
Now we can find P∗ a product of elementary matrices such that
 
0 aij 0 0 ... 0
−aij 0 0 0 ... 0
0
P∗ P0 APP∗ = 
 
 0 0 

 ... A2m−2 
0 0
Thus det A = determinant of a skew symmetric matrix of order 2 × determinant of a
skew symmetric matrix of order 2(m − 1). The induction hypothesis now gives the result.

Question 3(b) Find the rank of the matrix

c −b a0
 
0
 −c 0 a b0 
A=  b −a 0 c0 

−a0 −b0 −c0 0

where aa0 + bb0 + cc0 = 0, a, b, c positive integers.


0 c
Solution. The rank of a skew symmetric matrix is always even2 . Since −c 2
0 = c >

0, rank A ≥ 2.
If c0 6= 0, then
0
0
0 c −b a
0
cc 0 ac0 b0 c0
−c |A| =
0
b 0 −a0 0 0 c
−a −b −c 0
Adding b0 R3 − aR4 to R2 , all the entries of the second row become 0, so −c0 |A| = 0 ⇒
|A| = 0 ⇒ rank A < 4 ⇒ rank A = 2.
2
We can prove this by induction on the order of the skew symmetric matrix S. It is true if S is a 1 × 1
 be the 0-matrix, thus has rank 0. Now given an (n + 1) × (n + 1) matrix S, we can
matrix, since it must
C b
write it as −b 0
0 , where C is skew symmetric, and hence has even rank by the induction hypothesis. If b
is linearly dependent on
 the columns of C, then by a series of elementary operations on S, we can transform
it into P0 SP = 0C0 00 , so rank S = rank P0 SP = rank C. If b is linearly independent of the columns of
C b 0

C, then b’ is linearly independent of the rows of C, so rank S = rank −b 0
0 = rank[C b] + 1(∵ [−b 0] is
independent of the rows of [C b]) = rank C + 2, which is also even.

5
If c0 = 0,
0
c −b a0 a
c 0 a b0 a
a|A| =
b0 −a 0 0
−a a −b0 a 0 0
Adding −b0 R3 to R4 , we see that the fourth row has all 0’s, hence rank A = 2 as before.
Alternate solution:
c −b a0
 
0
 −c 0 a b0 
−a|A| =   b −a 0 c0 

aa0 ab0 ac0 0


Add −c0 R2 and b0 R3 to R4 , then all entries of the last row become 0. So rank A < 4, and
by the reasoning above, rank A ≥ 2, rank A 6= 3 so rank A = 2.

Question 3(c) Reduce the following symmetric matrix to a diagonal form and interpret the
results in terms of quadratic forms.
 
3 2 −1
A= 2 2 3 
−1 3 1

Solution.
x
(x y z)A y
z
= 3x2 + 2y 2 + z 2 + 4xy − 2xz + 6yz
2 1 2 2 22
= 3(x + y − z)2 + y 2 + z 2 + yz
3 3 3 3 3
2 1 2 2 11 2 117 2
= 3(x + y − z) + (y + z) − z
3 3 3 2 6
2 117 2
= 3X 2 + Y 2 − Z
3 6
where X = x − 23 y − 13 z, Y = y + 112
z, Z = z. This implies z = Z, y = Y − 11 2
Z, x =
2 11 1 2
X + 3 (Y − 2 Z) − 3Z = X + 3 Y − 4Z.
1 32 −4
 
3 0 0
Then if P = 0 1 − 11 2
, we have P0 AP = 0 2
3
0 .
0 0 1 0 0 − 117 6
The quadratic form associated with A is indefinite as it takes both positive and negative
values. Note that x0 Ax and x0 P0 APx0 take the same values.

6
UPSC Civil Services Main 1995 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

1 Linear Algebra
Question 1(a) Let T(x1 , x2 , x3 ) = (3x1 + x3 , −2x1 + x2 , −x1 + 2x2 + 4x3 ) be a linear trans-
formation on R3 . What is the matrix of T w.r.t. the standard basis? What is a basis of the
range space of T? What is a basis of the null space of T?

Solution.

T(e1 ) = T(1, 0, 0) = (3, −2, −1) = 3e1 − 2e2 − e3


T(e2 ) = T(0, 1, 0) = (0, 1, 2) = e2 + 2e3
T(e3 ) = T(0, 0, 1) = (1, 0, 4) = e1 + 4e3
 
3 0 1
T ⇐⇒ A = −2 1 0
−1 2 4

Clearly T(e2 ), T(e3 ) are linearly independent. If (3, −2, −1) = α(0, 1, 2) + β(1, 0, 4), then
β = 3, α = −2, but 2α + 4β 6= −1, so T(e1 ), T(e2 ), T(e3 ) are linearly independent. Thus
(3, −2, −1), (0, 1, 2), (1, 0, 4) is a basis of the range space of T.
Note that T(x1 , x2 , x3 ) = 0 ⇔ x1 = x2 = x3 = 0, so the null space of T is {0}, and the
empty set is a basis. Note that the matrix of T is nonsingular, so T(e1 ), T(e2 ), T(e3 ) are
linearly independent.

Question 1(b) Let A be a square matrix of order n. Prove that Ax = b has a solution
⇔ b ∈ Rn is orthogonal to all solutions y of the system A0 y = 0.

1
Solution. If x is a solution of Ax = b and y is a solution of A0 y = 0, then b0 y = x0 A0 y = 0,
thus b is orthogonal to y.
Conversely, suppose b0 y = 0 for all y ∈ Rn which is a solution of A0 y = 0. Let
W = A(Rn ) = the range space of A, and W ⊥ its orthogonal complement. If A0 y = 0 then
x0 A0 y = 0 ⇒ (Ax)0 y = 0 for every x ∈ Rn ⇒ y ∈ W ⊥ . Conversely y ∈ W ⊥ ⇒ ∀x ∈
Rn .(Ax)0 y = 0 ⇒ x0 A0 y = 0 ⇒ A0 y = 0. Thus W ⊥ = {y | A0 y = 0}. Now b0 y = 0 for all
y ∈ W ⊥ , so b ∈ W ⇒ b = Ax for some x ∈ Rn ⇒ Ax = b is solvable.
Question 1(c) Define a similar matrix and prove that two similar matrices have the same
characteristic equation. Write down a matrix having 1, 2, 3 as eigenvalues. Is such a matrix
unique?
Solution. Two matrices A, B are said to be similar if there exists a matrix P such that
B = P−1 AP. If A, B are similar, say B = P−1 AP, then characteristic polynomial of B is
|λI − B| = |λI − P−1 AP| = |P−1 λIP − P−1 AP| = |P−1 ||λI − A||P| = |λI − A|. (Note that
|X||Y| = |XY|.) Thus the characteristic polynomial of B is the same as that of A.
1 0 0
Clearly the matrix A = 0 2 0 has eigenvalues 1,2,3. Such a matrix is not unique, for
0 0 3
1 1 0
example B = 0 2 0 has the same eigenvalues, but B 6= A.
0 0 3

Question 2(a) Show that  


5 −6 −6
A = −1 4 2
3 −6 −4
is diagonalizable and hence determine A5 .
Solution.
|A − λI| = 0
5 − λ −6 −6

⇒ −1 4 − λ 2 = 0

3 −6 −4 − λ
⇒ (5 − λ)[(4 − λ)(−4 − λ) + 12] + 6[4 + λ − 6] − 6[6 − 3(4 − λ)] = 0
⇒ (5 − λ)[λ2 − 4] + 6[λ − 2 − 3λ + 6] = 0
⇒ −λ3 + 5λ2 + 4λ − 20 − 12λ + 24 = 0
⇒ λ3 − 5λ2 + 8λ − 4 = 0
Thus λ = 1, 2, 2.
If (x1 , x2 , x3 ) is an eigenvector for λ = 1, then
  
4 −6 −6 x1
−1 3 2   x2  = 0
3 −6 −5 x3
⇒ 4x1 − 6x2 − 6x3 = 0
−x1 + 3x2 + 2x3 = 0
3x1 − 6x2 − 5x3 = 0

2
Thus x1 = x3 , x3 = −3x2 , so (−3, 1, −3) is an eigenvector for λ = 1.
If (x1 , x2 , x3 ) is an eigenvector for λ = 2, then
  
3 −6 −6 x1
−1 2 2  x2  = 0
3 −6 −6 x3
⇒ 3x1 − 6x2 − 6x3 = 0
−x1 + 2x2 + 2x3 = 0
3x1 − 6x2 − 6x3 = 0

Thus x1 − 2x2 − 2x3 = 0, so taking x1 = 0, x2 = 1, (0, 1, −1) is an eigenvector for λ = 2.


Taking x1 = 4, x2 = 1, (4, 1, 1) is another eigenvector for λ = 2, and these two are linearly
independent.   
−3 0 4 2 −4 −4
Let P =  1 1 1. A simple calculation shows that P−1 = 12 −4 9 7 .
−3 −11  2 −3 −3
1 0 0
−1
Clearly P AP = 0 2 0.

0 0 2  
1 0 0
Now P−1 A5 P = (P−1 AP)5 = 0 32 0 .
0 0 32
 
1 0 0
A5 = P 0 32 0  P−1
0 0 32
   
−3 0 4 1 0 0 2 −4 −4
1
= 1 1 1   0 32 0  −4 9 7
2
−3 −1 1 0 0 32 2 −3 −3
  
−3 0 128 2 −4 −4
1
= 1 32 32   −4 9 7
2
−3 −32 32 2 −3 −3
 
125 −186 −186
= −31 94 62 
93 −186 −154

Note: Another way of computing A5 is given below. This uses the characteristic poly-
nomial of A : A3 = 5A2 − 8A + 4I and not the diagonal form, so it will not be permissible
here.

3
A5 = A2 (5A2 − 8A + 4I)
= 5A(5A2 − 8A + 4I) − 8(5A2 − 8A + 4I) + 4A2
= 25(5A2 − 8A + 4I) − 76A2 + 84A − 32I
= 49A2 − 116A + 68I

Now calculate A2 and substitute.

Question 2(b) Let A and B be matrices of order n. If I − AB is invertible, then I − BA


is also invertible and
(I − BA)−1 = I + B(I − AB)−1 A
Show that AB and BA have the same characteristic values.

Solution.
(I + B(I − AB)−1 A)(I − BA)
= I − BA + B(I − AB)−1 A − B(I − AB)−1 ABA
= [I + B(I − AB)−1 A] − B[I + (I − AB)−1 AB]A (1)
−1 −1 −1
Now (I − AB) (I − AB) = (I − AB) − (I − AB) AB = I
∴ (I − AB)−1 = I + (I − AB)−1 AB
Substituting in (1) (I + B(I − AB)−1 A)(I − BA)
= I + B(I − AB)−1 A − B(I − AB)−1 A = I

Thus I − BA is invertible and (I − BA)−1 = I + B(I − AB)−1 A as desired.


We shall show that λI − AB is invertible if and only if λI − BA is invertible. This means
that if λ is an eigenvalue of AB, then |λI − AB| = 0 ⇒ |λI − BA| = 0 so λ is an eigenvalue
of BA.
If λI − AB is invertible, then

(I + B(λI − AB)−1 A)(λI − BA)


= λI − BA + λB(λI − AB)−1 A − B(λI − AB)−1 ABA
= λ[I + B(λI − AB)−1 A] − B[I + (λI − AB)−1 AB]A (2)
Now (λI − AB)−1 (λI − AB) = λ(λI − AB)−1 − (λI − AB)−1 AB = I
∴ λ(λI − AB)−1 = I + (λI − AB)−1 AB
Substituting in (2) (I + B(λI − AB)−1 A)(λI − BA)
= λI + λB(λI − AB)−1 A − λB(λI − AB)−1 A = λI

Thus λI − BA is invertible if λI − AB is invertible. The converse is obvious as the situation


is symmetric, thus AB and BA have the same eigenvalues.
We give another simple proof of the fact that AB and BA have the same eigenvalues.
1. Let 0 be an eigenvalue of AB. This means that AB is singular, i.e. 0 = |AB| =
|A||B| = |BA|, so BA is singular, hence 0 is an eigenvalue of BA.

4
2. Let λ 6= 0 be an eigenvalue of AB and let x 6= 0 be an eigenvector corresponding to
λ, i.e. ABx = λx. Let y = Bx. Then y 6= 0, because Ay = ABx = λx 6= 0 as λ 6= 0.
Now BAy = BABx = B(ABx) = λBx = λy. Thus λ is an eigenvalue of BA.

Question 2(c) Let a, b ∈ C, |b| = 1 and let H be a Hermitian matrix. Show that the
eigenvalues of aI + bH lie on a straight line in the complex plane.

Solution. Let t be as eigenvalue of H, which has to be real because H is Hermitian. Clearly


a + tb is an eigenvalue of aI + bH. Conversely, if λ is an eigenvalue of aI + bH, then λ−ab
(note b 6= 0 as |b| = 1) is an eigenvalue of H.
Clearly a + tb lies on the straight line joining points a and a + b:

z = (1 − x)a + x(b − a), x ∈ R

For the sake of completeness, we prove that the eigenvalues of a Hermitian matrix H are
real. Let z 6= 0 be an eigenvector corresponding to the eigenvalue t.

Hz = tz
⇒ z0 Hz = tz0 z
0
⇒ z0 Hz = tz0 z
0
But z0 Hz = z0 H z = z0 Hz = tz0 z
⇒ tz0 z = tz0 z
⇒t = t ∵ z0 z 6= 0

Question 3(a) Let A be a symmetric matrix. Show that A is positive definite if and only
if its eigenvalues are all positive.

Solution. A is real symmetric so all eigenvalues of A are real. Let λ1 , λ2 , . . . , λn be


eigenvalues of A, not necessarily distinct. Let x1 be an eigenvector corresponding to λ1 .
Since λ1 and A are real,
p x1 is also real. Replacing x1 if necessary by µx1 , µ suitable, we can
assume that ||x1 || = x01 x1 = 1.
Let P1 be an orthogonal matrix with x1 as its first column. Such a P1 exists, as will be
shown at the end of this result. Clearly the first column of the matrix P1 −1 AP1 is equal
λ
to P1 −1 Ax = λ1 P1 −1 x = 00 , because P1 −1 x is the first column of P1 −1 P = I. Thus
0
P1 −1 AP1 = λ01 B = P01 AP1 where B is (n − 1) × (n − 1) symmetric. Since P01 AP1 is
L


symmetric, it follows that P1 −1 AP1 = P01 AP1 = λ01 B0 . Induction now  gives that there
λ2 0 . . . 0
exists an (n − 1) × (n − 1) orthogonal matrix Q such that Q0 BQ = . . . 
0 0 . . . λn

5
where λ2 , λ3 , . . . , λn are eigenvalues of B. Let P2 = 01 Q 0

, then P2 is orthogonal and
P02 P01 AP1 P2 = diagonal[λ 1 , . . . , λ n ]. Set P = P P
1 2 . . . P 0
n and (y1 , . . . , yn )P = x then
,
n
x0 Ax = y0 P0 APy = i=0 λ2i yi2 .
P
0
Pn 2 2
Since P is non-singular, quadratic forms x Ax and i=0 λi yi assume the same values.
Hence A is positive definite if and only if ni=0 λ2i yi2 is positive definite if and only if λi > 0
P
for all i. p
Result used: If x1 is a real vector such that ||x1 || = x01 x1 = 1 then there exists an
orthogonal matrix with x1 as its first column.
Proof: We have to find real column vectors x2 , . . . , xn such that ||xi || = 1, 2 ≤ i ≤ n
and x2 , . . . , xn is an orthonormal system i.e. x0i xj = 0, i 6= j. Consider the single equation
x01 x = 0, where x is a column vector to be determined. This equation has a non-zero solution,
in fact the space of solutions is of dimension n − 1, the rank of the coefficient matrix being
1. If y2 is a solution, we take x2 = ||yy22 || so that x01 x2 = 0.
We now consider the two equations x01 x = 0, x02 x = 0. Again the number of unknowns
is more than the number of equations, so there is a solution, say y3 , and take x3 = ||yy33 || to
get x1 , x2 , x3 mutually orthogonal.
Proceeding in this manner, if we consider n − 1 equations x01 x = 0, . . . , x0n−1 x = 0, these
will have a nonzero solution yn , so we set xn = ||yynn || . Clearly x1 , x2 , . . . , xn is an orthonormal
system, and therefore P = [x1 , . . . , xn ] is an orthogonal matrix having x1 as a first column.

Question 3(b) Let A and B be square matrices of order n, show that AB − BA can never
be equal to the identity matrix.

Solution. Let A = haij i and B = hbij i. Then

tr AB = Sum of diagonal elements of AB


Xn Xn n X
X n
= aik bki = bki aik = tr BA
i=1 k=1 k=1 i=1

Thus tr(AB − BA) = tr AB − tr BA = 0. But the trace of the identity matrix is n, thus
AB − BA can never be equal to the identity matrix.
n
X
Question 3(c) Let A = haij i, 1 ≤ i, j ≤ n. If |aij | < |aii |, then the eigenvalues of A lie
j=1
i6=j
in the disc n
X
|λ − aii | ≤ |aij |
j=1
i6=j

6
Solution. See the solution to question 2(c), year 1997. We showed that if |λ − aii | >
n
X
|aij | then |λI − A| =
6 0, so λ is not an eigenvalue of A. Thus if λ is an eigenvalue, then
j=1
i6=j
n
X
|λ − aii | ≤ |aij |, so λ lies in the disc described in the question.
j=1
i6=j

7
UPSC Civil Services Main 1996 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

Question 1(a) In R4 let W1 be the space generated by {(1, 1, 0, −1), (2, 4, 6, 0)} and let W2
be space generated by {(−1, −2, −2, 2), (4, 6, 4, −6), (1, 3, 4, −3)}. Find a basis for the space
W1 + W2 .

Solution. Let v1 = (1, 1, 0, −1), v2 = (2, 4, 6, 0), v3 = (−1, −2, −2, 2), v4 = (4, 6, 4, −6), v5 =
(1, 3, 4, −3). Since w ∈ W1 + W2 can be written as w = w1 + w2 , and w1 = α1 v1 + α2 v2
and w2 = α3 v3 + α4 v4 + α5 v5 , it follows that w is a linear combination of vi ⇒ W1 + W2
is generated by {vi , 1 ≤ i ≤ 5}. Thus a maximal independent subset of {vi , 1 ≤ i ≤ 5} will
be a basis of W1 + W2 .
Clearly v1 and v2 are linearly independent. If possible, let v3 = λ1 v1 + λ2 v2 , then the
four equations

λ1 + 2λ2 = −1
λ1 + 4λ2 = −2
0λ1 + 6λ2 = −2
−λ1 + 0λ2 = 2

should be consistent and provide us λ1 , λ2 . Clearly the third and fourth equations give us
λ1 = −2, λ2 = − 31 which do not satisfy the first two equations. Thus v1 , v2 , v3 are linearly
independent.
If possible let v4 = λ1 v1 + λ2 v2 + λ3 v3 . Then

λ1 + 2λ2 − λ3 =4 (1a)
λ1 + 4λ2 − 2λ3 =6 (1b)
0λ1 + 6λ2 − 2λ3 =4 (1c)
−λ1 + 0λ2 + 2λ3 = −6 (1d)

1
Adding (1b) and (1d) we get 4λ2 = 0, so λ2 = 0. Solving (1a) and (1b) we get λ3 = −2, λ1 =
2. These values satisfy all the four equations, so v4 = 2v1 − 2v3 .
If possible let v5 = λ1 v1 + λ2 v2 + λ3 v3 . Then

λ1 + 2λ2 − λ3 =1 (2a)
λ1 + 4λ2 − 2λ3 =3 (2b)
0λ1 + 6λ2 − 2λ3 =4 (2c)
−λ1 + 0λ2 + 2λ3 = −3 (2d)

Adding (2b) and (2d) we get 4λ2 = 0, so λ2 = 0. (2c) then gives us λ3 = −2, and
(2a) now gives λ1 = −1, which satisfies all equations. Thus v5 = −v1 − 2v3 . Hence
{(1, 1, 0, −1), (2, 4, 6, 0), (−1, −2, −2, 2)} is a basis of W1 + W2 .

Question 1(b) Let V be a finite dimensional vector space and v ∈ V, v 6= 0. Show that
there exists a linear functional f on V such that f (v) 6= 0.

Solution. Complete v to a basis of V, say {v1 = v, v2 , . . . , vn }, where dim V = n. Define


n
X n
X
f (vj ) = δ1j and f ( aj v j ) = aj f (vj ).
j=1 j=1
functional over V, and f (v) = f (v1 ) = 1. Note that f (vj ) = 0, j > 1
Clearly f is a linear P
and if any w ∈ V, w = i ai vi , f (w) = a1 .

Question 1(c) Let V = R3 , v1 , v2 , v3 be a basis of V. Let T : V −→ V be such that


T(vi ) = v1 + v2 + v3 , 1 ≤ i ≤ 3. By writing the matrix of T w.r.t. another basis show that
the matrices    
1 1 1 3 0 0
A = 1 1 1 and B = 0 0 0
1 1 1 0 0 0
are similar.

Solution. Clearly A is the matrix of T w.r.t. the basis v1 , v2 , v3 . Note that

[T(v1 ), T(v2 ), T(v3 )] = (v1 , v2 , v3 )A

Let

w1 = v1 + v 2 + v 3
w2 = v1 − v2
w3 = v2 − v3
⇒ T(w1 ) = 3w1 , T(w2 ) = T(w3 ) = 0

We now show that w1 , w2 , w3 is a basis for V, i.e. these are linearly independent.

2
Let αw1 + βw2 + γw3 = 0, then (α + β)v1 + (α − β + γ)v2 + (α − γ)v3 = 0. But
v1 , v2 , v3 are linearly independent, therefore α + β = 0, α − β + γ = 0, α − γ = 0 ⇒ α =
β = γ = 0 ⇒ w1 , w2 , w3 are linearly independent.
The matrix of T w.r.t. the basis w1 , w2 , w3 is clearly B. Note that the choice of
w1 , w2 , w3 is suggested by the shape of B.
6 0 then B = P−1 AP, so A and B are similar.
If (w1 , w2 , w3 ) = (v1 , v2 , v3 )P, |P| =

Question 2(a) Let V = R3 and T : V −→ V be a linear map defined by

T(x, y, z) = (x + z, −2x + y, −x + 2y + z)

What is the matrix of T w.r.t. the basis (1, 0, 1), (−1, 1, 1), (0, 1, 1)? Using this matrix write
down the matrix of T with respect to the basis (0, 1, 2), (−1, 1, 1), (0, 1, 1).

Solution. Let v1 = (1, 0, 1), v2 = (−1, 1, 1), v3 = (0, 1, 1). T(x, y, z) = (x+z, −2x+y, −x+
2y+z) = αv1 +βv2 +γv3 , say. This means α−β = x+z, β+γ = −2x+y, α+β+γ = −x+2y+
z. This implies α = x + y + z, β = y, γ = −2x. Thus T(x, y, z) = (x + y + z)v1 + yv2 − 2xv3 .
Hence  
2 1 2
[T(v1 ) T(v2 ) T(v3 )] = [v1 v2 v3 ]  0 1 1
−2 2 0
Let w1 = (0, 1, 2), w2 = (−1, 1, 1), w3 = (0, 1, 1). Then
 
1 0 0
[w1 w2 w3 ] = [v1 v2 v3 ] 1 1 0
0 0 1

Hence

[T(w1 ) T(w2 ) T(w3 )] = [T(v1 ) T(v2 ) T(v3 )]P


= [v1 v2 v3 ]AP
= [w1 w2 w3 ]P−1 AP

where    
2 1 2 1 0 0
A =  0 1 1 , P = 1 1 0
−2 2 0 0 0 1
Thus the matrix of T w.r.t. basis w1 , w2 , w3 is
     
1 0 0 2 1 2 1 0 0 3 1 2
−1
P AP = −1  1 0   0 1 1 1 1 0 = −2 0 −1
0 0 1 −2 2 0 0 0 1 0 2 0

3
Question 2(b) Let V and W be finite dimensional vector spaces such that dim V ≥ dim W.
Show that there is always a linear map of V onto W.
Solution. Let w1 , w2 , . . . , wm be a basis of W, and v1 , v2 , . . . , vn be a basis of V, n ≥ m.
Define
T(vi ) = wi , i = 1, 2, . . . , m
T(vi ) = 0, i = m + 1, . . . , n
and for any v ∈ V, v = i=1 αi vi , T(v) = m
Pn P
i=1 αi T(vi ). Pm
Clearly
Pm T : V
Pm−→ W is linear. T is onto, since if w ∈ W, w = i=1 ai wi , then
T( i=1 ai vi ) = i=1 ai T(vi ) = w, proving the result.

Question 2(c) Solve by Cramer’s rule


x + y − 2z = 1
2x − 7z = 3
x+y−z = 5
Solution.
1 1 −2 −1 −1 −2

D = 2 0 −7 = −5 −7 −7 = −2
1 1 −1 0 0 −1
˛ ˛ ˛ ˛
˛1
˛
˛
1 −2˛˛˛ ˛1
˛
˛
1 −2˛˛˛
˛3
˛
˛
0 −7˛˛˛ ˛3
˛
˛
0 −7˛˛˛
˛5
˛
1 −1˛˛ ˛4
˛
0 1 ˛˛ −31 31
x= D
= D
= −2
= 2
˛ ˛ ˛ ˛
˛1
˛
˛
1 −2˛˛˛ ˛1
˛
˛
0 0 ˛˛˛
˛2
˛
˛
3 −7˛˛˛ ˛2
˛
˛
1 −3˛˛˛
˛1
˛
5 −1˛˛ ˛1
˛
4 1 ˛˛ 13
y= D
= D
= −2
= − 13
2
˛ ˛ ˛ ˛
˛1 1 1˛˛˛ ˛1 1 1˛˛˛
˛ ˛
˛ ˛
˛2 0 3˛˛˛ ˛2 0 3˛˛˛
˛ ˛
˛ ˛
˛1 1 5˛˛ ˛0 0 4˛˛
˛ ˛
−8
z= D
= D
= −2
=4

Question 3(a) Find the inverse of the matrix


 
0 1 0 0
0 0 1 0
A= 0 0

0 1
1 0 0 0
by computing its characteristic polynomial.

4
Solution. The characteristic polynomial of A is

−λ 1 0 0

0 −λ 1 0
|A − λI| =
0 0 −λ 1
1 0 0 −λ
= −λ[−λ3 ] − 1[1] = λ4 − 1 = 0

Thus by the Cayley-Hamilton theorem, A4 = I, so A−1 = A3 .


    
0 1 0 0 0 1 0 0 0 0 1 0
 0 0 1 0 0 0
  1 0 0 0 0 1
A2 = 0 0 0 1 0 0
= 
0 1 1 0 0 0
1 0 0 0 1 0 0 0 0 1 0 0
    
0 0 1 0 0 1 0 0 0 0 0 1
 0 0 0 1 0 0 1 0 1 0 0 0
 = A−1
A3 = 
1
 =
0 0 0 0 0 0 1 0 1 0 0
0 1 0 0 1 0 0 0 0 0 1 0

Question 3(b) If A and B are n × n matrices such that AB = BA, show that AB and
BA have a common characteristic vector.

Solution. Let λ be any eigenvalue of A and let Vλ be the eigenspace of A corresponding to λ.


We show that B(Vλ ) ⊆ Vλ . Let v ∈ Vλ , then A(Bv) = B(Av) = B(λv) = λBv ⇒ Bv ∈ Vλ .
Consider B∗ : Vλ −→ Vλ such that B∗ (v) = B(v) — note that B∗ is a restriction of B
to Vλ and we have already shown that B(Vλ ) ⊆ Vλ .
Let µ be an eigenvalue of B∗ , then µ is also an eigenvalue of B (because a basis of Vλ can be
∗ C

extended to a basis of V, and in this basis B = B0 D for some matrices C, D). Let v ∈ Vλ
be an eigenvector of B∗ corresponding to µ, by definition v 6= 0. Then Bv = B∗ v = µv.
Thus A and B have a common eigenvector v, note that Av = λv as v ∈ Vλ .

Question 3(c) Reduce to canonical form the orthogonal matrix


2
− 23 13

3
O =  32 13 − 23 
1 2 2
3 3 3

Solution. Before solving this particular problem, we present a general discussion about
orthogonal matrices. An orthogonal matrix satisfies O0 O = I, so its determinant is 1 or -1,
here we focus on the case where |O| = 1. If λ is an eigenvalue of O and x a corresponding
eigenvector, then |λ|2 x0 x = (Ox)0 Ox = x0 O0 Ox = x0 x, so |λ| = 1. Since the characteristic

5
polynomial has real coefficients, the eigenvalues must be real or in complex conjugate pairs.
Thus for a matrix of order 3, at least one eigenvalue is real, and must be 1 or -1. Since
|O| = 1, one real value must be 1, and the three possibilities are {1, 1, 1}, {1, −1, −1} and
{1, eiθ , e−iθ }. √
Here we consider the third case, as the given matrix has 1 and 13 ± i 2 3 2 as eigenvalues,
proved later.
Let Z = X1 + iX2 be an eigenvector corresponding to the eigenvalue eiθ . Let X3 be
the eigenvector corresponding to the eigenvalue 1. Since Z and X3 correspond to different
eigenvalues, these are orthogonal, i.e. Z0 X3 = (X01 + iX02 )X3 = 0 ⇒ X01 X3 = 0, X02 X3 = 0.
Note that X1 , X2 , X3 are real vectors. Since OZ = eiθ Z = (cos θ + i sin θ)(X1 + iX2 ).
Equating real and imaginary parts we get

OX1 = X1 cos θ − X2 sin θ


OX2 = X1 sin θ + X2 cos θ
0 0
∴ X1 O OX1 = (X01 cos θ − X02 sin θ)(X1 cos θ − X2 sin θ)
⇒ X01 X1 = X01 X1 cos2 θ − X02 X1 cos θ sin θ − X01 X2 sin θ cos θ + X02 X2 sin2 θ
⇒ 0 = X01 X1 sin2 θ − X02 X2 sin2 θ + 2X01 X2 cos θ sin θ
⇒ 0 = X01 X1 sin θ − X02 X2 sin θ + 2X01 X2 cos θ (1)

(Note that sin θ 6= 0 since we are considering the case where eiθ is complex.) Similarly

X02 O0 OX1 = (X01 sin θ + X02 cos θ)(X1 cos θ − X2 sin θ)


⇒ X02 X1 = X01 X1 sin θ cos θ − X01 X2 sin2 θ − X02 X2 sin θ cos θ + X02 X1 cos2 θ
⇒ 0 = X01 X1 cos θ − X01 X2 cos θ − 2X01 X2 sin θ (2)

Multiplying (1) by sin θ and (2) by cos θ and adding, we get X01 X1 − X02 X2 = 0 or X01 X1 =
X02 X2 , so from (2), X1 X2 = 0, i.e. X1 , X2 are orthogonal.
Thus X1 , X2 , X3 are mutually orthogonal. We can assume that X01 X1 = X02 X2 = 1,
replacing Z by λZ, λ ∈ R if necessary. Similarly we can take X03 X3 = 1. Let P = [X1 X2 X3 ]
so that P0 P = I. Now

O[X1 X2 X3 ] = [X1 cos θ − X2 sin θ, X1 sin θ + X2 cos θ, X3 ]


 
cos θ sin θ 0
= [X1 X2 X3 ] − sin θ cos θ 0
0 0 1
 
cos θ sin θ 0
−1 0
⇒ P OP = P OP =  − sin θ cos θ 0
0 0 1

which is the canonical form of O when the eigenvalues are 1, eiθ , e−iθ .

6
Solution of given problem.
2
− 23 13

3
O =  32 13 − 32 
1 2 2
2 3 3 3

− λ −2 1 2 − 3λ −2 1
3 3 3 1
|O − λI| = 23 1
3
− λ 2
−3 = 2 1 − 3λ −2
1 2 2 27
3 3 3
−λ 1 2 2 − 3λ
1
= [(2 − 3λ)2 (1 − 3λ) + 4(2 − 3λ) + 1(3 + 3λ) + 2(6 − 6λ)]
27
1
= − [27λ3 − 45λ2 + 45λ − 27]
27
1
= − [(λ − 1)(3λ2 − 2λ − 3)]
3

Thus λ = 1, 13 ± i 2 3 2 are eigenvalues of O. √
2 2
Thus the canonical form of O is derived from above, where cos θ = 13 , sin θ = 3
:

1 2 2
 
3√ 3
0
− 2 2 1
0
3 3
0 0 1

The matrix P can be determined as follows (this is not needed for this problem, but is
given for completeness):

1. Let (x1 , x2 , x3 ) be an eigenvector for λ = 1, then


1 2 1
− x1 − x2 + x3 = 0
3 3 3
2 2 2
x1 − x2 − x3 = 0
3 3 3
1 2 1
x1 + x2 − x3 = 0
3 3 3
Thus x2 = 0, x1 − x3 = 0, so we can take (1, 0, 1) as an eigenvector.

2. The vectors X1 , X2 in the above discussion are determined by the requirements

OX1 = X1 cos θ − X2 sin θ


OX2 = X1 sin θ + X2 cos θ

7

2 2
where cos θ = 13 , sin θ = 3
. This gives us the following equations

2x11 − 2x12 + x13 = x11 − x21 2 2 (3)

2x11 + x12 − 2x13 = x12 − x22 2 2 (4)

x11 + 2x12 + 2x13 = x13 − x23 2 2 (5)

2x21 − 2x22 + x23 = x11 2 2 + x21 (6)

2x21 + x22 − 2x23 = x12 2 2 + x22 (7)

x21 + 2x22 + 2x23 = x13 2 2 + x23 (8)

Adding the √ last 3 equations, we get 2x21 √ = x11 + x12 + x13 . Subtracting equation (6)
from (8), 2x22 = x13 − x11 , and from (7) 2x23 = x11 − x12 + x13 . Substituting these
in the first 3 equations and simplifying, we get x11 = −x13 . Setting x11 = 0, x12 = 1,
we get (0, 1, 0), ( √12 , 0, − √12 ) as a possible solution for X1 , X2 .

Putting these together we get


0 √12 1
 

P = 1 0 0
0 − √12 1

2 2
 1 
3√ 3
0
We can now verify that OP = P − 2 3 2 1
3
0
0 0 1

8
UPSC Civil Services Main 1997 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

1 Linear Algebra
Question 1(a) Let V be the vector space of polynomials over R. Find a basis and the
dimension of W ⊆ V spanned by

v1 = t3 − 2t2 + 4t + 1
v2 = 2t3 − 3t2 + 9t − 1
v3 = t3 + 6t − 5
v4 = 2t3 − 5t2 + 7t + 5

Solution. v1 and v2 are linearly independent, because if αv1 + βv2 = 0, then α + 2β =


0, −2α − 3β = 0, 4α + 9β = 0, α − β = 0 ⇒ α = β = 0.
v3 depends linearly on v1 , v2 — if αv1 + βv2 = v3 , then α + 2β = 1, −2α − 3β =
0, 4α + 9β = 6, α − β = −5 ⇒ α = −3, β = 2 which satisfy all the equations. Thus
v3 = −3v1 + 2v2 .
v4 depends linearly on v1 , v2 — if αv1 + βv2 = v4 , then α + 2β = 2, −2α − 3β =
−5, 4α + 9β = 7, α − β = 5 ⇒ α = 4, β = −1 which satisfy all the equations. Thus
v4 = 4v1 − v2 .
Thus dimR W = 2 and v1 , v2 is a basis of W.

Question 1(b) Verify that T(x1 , x2 ) = (x1 + x2 , x1 − x2 , x2 ) is a linear transformation from


R2 to R3 . Find its range, rank, null space and nullity.

1
Solution. Let x = (x1 , x2 ), y = (y1 , y2 ). Then
T(αx + βy) = T(αx1 + βy1 , αx2 + βy2 )
= (αx1 + βy1 + αx2 + βy2 , αx1 + βy1 − αx2 − βy2 , αx2 + βy2 )
= (α(x1 + x2 ), α(x1 − x2 ), αx2 ) + (β(y1 + y2 ), β(y1 − y2 ), βy2 )
= αT(x1 , x2 ) + βT(y1 , y2 )
Thus T is linear.

T(e1 ) = T(1, 0) = (1, 1, 0)


T(e2 ) = T(0, 1) = (1, −1, 1)
Clearly T(e1 ), T(e2 ) are linearly independent. Since T (R2 ) is generated by T(e1 ) and T(e2 ),
the rank of T is 2.
The range of T = {αT(e1 ) + βT(e2 ), α, β ∈ R}
= {α(1, 1, 0) + β(1, −1, 1)}
= {(α + β, α − β, β) | α, β ∈ R}
To find the null space of T, if T(x1 , x2 ) = (0, 0, 0), then x1 + x2 = 0, x1 − x2 = 0, x2 = 0,
so x1 = x2 = 0. Thus the null space of T is {0}, and nullity T = 0.

Question 1(c) Let V be the space of 2×2 matrices over R. Determine whether the matrices
A, B, C ∈ V are dependent where
     
1 2 3 −1 1 −5
A= B= C=
3 1 2 2 −4 0
Solution. If αA + βB + γC = 0, then
α + 3β + γ = 0 (1)
2α − β − 5γ = 0 (2)
3α + 2β − 4γ = 0 (3)
α + 2β = 0 (4)
From (4), we get α = −2β. This, together with (3) gives γ = −β. These satisfy (1) and
(2) also, so taking β = 1, α = −2, γ = −1 gives us −2A + B − C = 0. Thus A, B, C are
dependent.

Question 2(a) Let A be an n × n matrix such that each diagonal entry is µ and each
off-diagonal entry is 1. If B = λA is orthogonal, determine λ, µ.

Solution. Clearly A is symmetric. Let A = (aij ). B0 B = BB0 = λ2 A2 = I =⇒


P n 2
k=1 λ aik akj = δij
Taking i = j = 1, we get λ2 (µ2 +n−1) = 1 Taking i = 1, j = 2, we get λ2 (2µ+n−2) = 0.
Thus µ = −(n−2)/2 and λ2 [(n−2)2 /4+n−1] = 1. Simplifying, λ2 [n2 −4n+4+4n−4]/4 = 1,
which means λ2 = n42 , or λ = ± n2 .

2
Question 2(b) Show that  
2 −1 0
A =  −1 2 0 
2 2 3
is diagonalizable over R. Find P such that P−1 AP is diagonal and hence find A25 .

Solution. Characteristic equation of A is



2 − x −1 0

−1 2 − x 0 = 0

2 2 3−x
⇒ (2 − x)(2 − x)(3 − x) + 1(−(3 − x)) = 0
(3 − x)(4 − 4x + x2 − 1) = 0

Thus the eigenvalues are 3, 3, 1.


Let (x1 , x2 , x3 ) be an eigenvector for λ = 1.
  
1 −1 0 x1
 −1 1 0   x2  = 0
2 2 2 x3

Thus x1 − x2 = 0, −x1 + x2 = 0, 2x1 + 2x2 + 2x3 = 0. Take x1 = 1, then x2 = 1, x3 = −2, so


(1, 1, −2) is an eigenvector with eigenvalue 1.
Let (x1 , x2 , x3 ) be an eigenvector for λ = 3.
  
−1 −1 0 x1
 −1 −1 0   x2  = 0
2 2 0 x3

Thus x1 + x2 = 0. Take x1 = 1, x3 = 0, then x2 = −1, so (1, −1, 0) is an eigenvector


with eigenvalue 3. Take x1 = 0, x3 = 1, then x2 = 0, so (0, 0, 1) is also an eigenvector for
eigenvalue 3.     
1 1 0 1 0 0 1 0 0
Let P =  1 −1 0  then AP = P  0 3 0  or P−1 AP =  0 3 0 
−2 0 1  0  0 3  0 0 3 
1 0 0 1 0 0
−1 25 −1
Now P A P = (P AP) = 25  0 3 25
0  25
. Thus A = P 0 3  25
0  P−1
25
 1 1 0 0 3 0 0 325
2 2
0
−1 1 1
|P| = −2, so P = 
2
− 2 0 .
1 1 1

3
1 1
   
1 1 0 1 0 0 2 2
0
A25 =  1 −1 0   0 3 25
0  1
2
− 12 0 
−2 0 1 0 0 325 1 1 1
   1 1 
1 325 0 2 2
0
=  1 −325 0   12 − 21 0 
−2 0 325 1 1 1
 1+325 1−325

2 2
0
25 1+325
=  1−32 2
0 
−1 + 325 −1 + 325 325

Question 2(c) Let A = [aij ] be a square matrix of order n such that |aij | ≤ M . Let λ be
an eigenvalue of A, show that |λ| ≤ nM .

Solution. We first prove the following:


X n
Lemma: If A = [aij ] and |aij | ≤ aii then |A| 6= 0.
j=1
i6=j
If |A| = 0 then there exist x1 , . . . , xn ∈ C not all zero such that

a11 x1 + a12 x2 + . . . + a1n xn = 0


...
ai1 x1 + ai2 x2 + . . . + ain xn = 0
...
an1 x1 + an2 x2 + . . . + ann xn = 0

x
Let |xi | = max(|x1 |, |x2 |, . . . , |xn |), so | xji | ≤ 1 for all j.

x 1 x 2 x n
0 = aii − (−ai1 − ai2 − . . . − ain )
xi xi xi

x1 x2 xn
≥ |aii | − ai1 + ai2 + . . . + ain

xi xi xi
≥ |aii | − |ai1 | − |ai2 | − . . . − |ain |
n
X
which contradicts the premise |aij | ≤ aii Thus |A| 6= 0.
j=1
i6=j

4
n
X
Now the lemma tells us that if |λ − aii | > |aij | then |λI − A| =
6 0, so λ is not an
j=1
i6=j
n
X
eigenvalue of A. Thus |λ| ≤ |λ − aii | + |aii | ≤ |aij | ≤ nM as desired.
j=1

Question 3(a) Define a positive definite matrix and show that a positive definite matrix is
always non-singular. Show that the converse is not always true.

Solution. Let A be an n × n real symmetric matrix. A is said to be positive definite if


the associated quadtratic form
 
x1
  x2 
x1 x2 . . . xn A 
. . .  > 0

xn

for all (x1 , x2 , . . . , xn ) 6= (0, 0, . . . , 0) in Rn .


If |A| = 0 then rank A < n, which means that columns of A i.e. c1 , c2 , . . . , cn are
linearly dependent i.e. there exist real numbers x1 , x2 , . . . , xn not all zero such that
   
x1 x1
 x2    x2 
x1 c1 + x2 c2 + . . . + xn cn = A  . . . = 0 =⇒ x1 x2 . . . xn A . . . = 0
  

xn xn

where (x1 , x2 , . . . , xn ) 6= (0, 0, . . . , 0), which means that A is not positive definite. Thus A
is positive definite =⇒ |A| 6= 0.
The converse is not true. Take
 
1 0 0
A = 0 1 0 
0 0 −1

then |A| = −1, but  


 0
0 0 1 A 0 = −1

1
so A is not positive definite.

5
Question 3(b) Find the eigenvalues and their corresponding eigenvectors for the matrix
 
6 −2 2
−2 3 −1
2 −1 3

Solution. The characteristic equation for A is

0 = |A − xI|
 
6 − x −2 2
=  −2 3 − x −1 
2 −1 3 − x
= (6 − x)((3 − x)2 − 1) + 2(−6 + 2x + 2) + 2(2 − 6 + 2x)
= (6 − x)(9 − 6x + x2 ) − 6 + x − 8 + 4x − 8 + 4x
0 = x3 − 12x2 + 36x − 32
= (x − 2)(x2 − 10x + 16)

Thus the eigenvalues are 2, 2, 8.


Let (x1 , x2 , x3 ) be an eigenvector for λ = 2.
  
4 −2 2 x1
−2 1 −1  x2  = 0
2 −1 1 x3
Thus 4x1 − 2x2 + 2x3 = 0, −2x1 + x2 − x3 = 0, 2x1 − x2 + x3 = 0. Take x1 = 1, x2 = 0, then
x3 = −2, so (1, 0, −2) is an eigenvector with eigenvalue 2. Take x1 = 0, x2 = 1, then x3 = 1,
so (0, 1, 1) is an eigenvector with eigenvalue 2.
Let (x1 , x2 , x3 ) be an eigenvector for λ = 8.
  
−2 −2 2 x1
−2 −5 −1  x2  = 0
2 −1 −5 x3
Thus −2x1 − 2x2 + 2x3 = 0, −2x1 − 5x2 − x3 = 0, 2x1 − x2 − 5x3 = 0. From the last two, we
get x2 + x3 = 0, and from the first we get x1 = 2x3 . Take x3 = 1, then x2 = −1, x1 = 2, so
(2, −1, 1) is an eigenvector with eigenvalue 8.

Question 3(c) Find P invertible such that P reduces Q(x, y, z) = 2xy + 2yz + 2zx to its
canonical form.

Solution. The matrix of Q(x, y, z) is


 
0 1 1
A = 1 0 1
1 1 0

6
which has all diagonal entries 0, so we cannot complete squares right away.
     
0 1 1 1 0 0 1 0 0
1 0 1 = 0 1 0 A 0 1 0
1 1 0 0 0 1 0 0 1

Add the second row to the first and the second column to the first.
     
2 1 2 1 1 0 1 0 0
1 0 1 = 0 1 0 A 1 1 0
2 1 0 0 0 1 0 0 1

Subtract 12 R1 from R2 and 1


C
2 1
from C2 .

1 − 12 0
     
2 0 2 1 1 0
0 − 21 0 = − 12 21 0 A 1 12 0
2 0 0 0 0 1 0 0 1

Subtract R1 from R3 and C1 from C3 .

1 − 12 −1
     
2 0 0 1 1 0
0 − 1 0  = − 1 1 0 A 1 1 −1
2 2 2 2
0 0 −2 −1 −1 1 0 0 1

1 − 12 −1
   
2 0 0
Thus P = 1 21 −1 and P0 AP = 0 − 12 0 
0 0 1 0 0 −2
2 1 2 2
So Q(x, y, z) −→ 2X − 2 Y − 2Z .
Alternative Solution. Let x = X, y = X + Y, z = Z

Q(x, y, z) = 2X 2 + 2XY + 2ZX + 2ZY + 2ZX


= 2[X 2 + XY + 2ZX + ZY ]
Y Y2
= 2[(X + + Z)2 − − Z 2]
2 4
Put ξ = X + Y /2 + Z, η = Y, ζ = Z, so X = ξ − η/2 − ζ, Y = η, Z = ζ.

1 − 12 −1 1 − 21 −1
           
x 1 0 0 X 1 0 0 ξ ξ
y  = 1 1 0  Y  = 1 1 0 0 1 0  η  = 1 12 −1 η 
z 0 0 1 Z 0 0 1 0 0 1 ζ 0 0 1 ζ

1 − 21 −1
 

Thus Q(x, y, z) −→ 2ξ 2 − η 2 /2 − 2ζ 2 , and P = 1 12 −1 as before. Note that we put


0 0 1
x = X, y = X + Y, z = Z to create one square term to complete the squares.

7
UPSC Civil Services Main 1998 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

Question 1(a) Given two linearly independent vectors (1, 0, 1, 0) and (0, −1, 1, 0) of R4 ,
find a basis of R4 which includes them.

Solution. Let v1 = (1, 0, 1, 0), v2 = (0, −1, 1, 0). Clearly these are linearly independent.
Let e1 , e2 , e3 , e4 be the standard basis. Then v1 , v2 , e1 , e2 , e3 , e4 generate R4 . We have to
find four vectors out of these which are linearly independent and include v1 , v2 .
If αv1 + βv2 + γe1 = 0, then α + γ = 0, −α = 0, α + β = 0 ⇒ α = β = γ = 0. Therefore
v1 , v2 , e1 are linearly independent.
We now show that v1 , v2 , e1 , e4 are linearly independent. Let αv1 + βv2 + γe1 + δe4 = 0
then δ = 0, and therefore α = β = γ = 0 because v1 , v2 , e1 are linearly independent.
Thus v1 , v2 , e1 , e4 is a basis of R4 .
Note that e2 = v1 − v2 − e1 , e3 = v1 − e1 .

Question 1(b) If V is a finite dimensional vector space over R and if f and g are two
linear transformations from V to R such that f (v) = 0 implies g(v) = 0, then prove that
g = λf for some λ ∈ R.

Solution. If g = 0, take λ = 0, so g(v) = 0 = 0f (v) for all v ∈ V.


If g 6= 0, then f 6= 0. Thus ∃v ∈ V such that f (v) 6= 0 ⇒ ∃w ∈ V such that f (w) = 1
v
(Note that f ( f (v) ) = 1).
Thus V/ ker f w R, or dim(ker f ) = n − 1. Similarly ker g has dimension n − 1. In fact,
ker f = ker g ∵ ker f ⊆ ker g and dim(ker f ) = dim(ker g). Let {v2 , . . . , vn } be a basis of
ker f and extend it to {v1 , v2 , . . . , vn } a basis of V. Then g = λf with λ = g(v1 )/f (v1 ) ∵
if v = α1 v1 + . . . + αn vn , then g(v) = α1 g(v1 ) = α1 λf (v1 ) = λf (v).

1
Question 1(c) Let T : R3 −→ R3 be defined by T(x1 , x2 , x3 ) = (x2 , x3 , −cx1 − bx2 − ax3 )
where a, b, c are fixed real numbers. Show that T is a linear transformation of R3 and that
A3 + aA2 + bA + cI = 0 where A is the matrix of T w.r.t. the standard basis of R3 .
Solution. Let x = (x1 , x2 , x3 ), y = (y1 , y2 , y3 ). Then
T(αx + βy) = (αx2 + βy2 , αx3 + βx3 , −c(αx1 + βy1 ) − b(αx2 + βy2 ) − a(αx3 + βy3 ))
= α(x2 , x3 , −cx1 − bx2 − ax3 ) + β(y2 , y3 , −cy1 − by2 − ay3 )
= αT(x) + βT(y)
Thus T is linear.
Clearly
T(1, 0, 0) = (0, 0, −c)
T(0, 1, 0) = (1, 0, −b)
T(0, 0, 1) = (0, 1, −a)
 
0 1 0
A= 0 0 1 
−c −b −a
The characteristic equation of A is |A − λI| = 0.

−λ 1 0

0 λ 1 = 0

−c −b −a − λ
−λ2 (a + λ) − bλ − c = 0
λ3 + aλ2 + bλ + c = 0
Now by the Cayley-Hamilton theorem A3 + aA2 + bA + cI = 0.
Question 2(a) If A and B are two matrices of order 2 × 2 such that A is skew-Hermitian
and AB = B then show that B = 0.
Solution. We first of all prove that eigenvalues of skew-Hermitian matrices are 0 or pure
0
imaginary. Let A be skew-Hermitian, i.e. A = −A and let λ be its characteristic root. If
x is an eigenvector of λ, then
Ax = λx
⇒ x0 λx = x0 Ax
0
= −x0 A x
0
= −Ax x
= −λx0 x
Thus λ = −λ ∵ x0 x 6= 0, showing that the real part of λ is 0.
Now if B 6= 0 and c1 , c2 are the columns of B, then c1 6= 0 or c2 6= 0. AB = B means
that Ac1 = c1 and Ac2 = c2 . Since either c1 6= 0 or c2 6= 0, 1 must be an eigenvalue of A,
which is not possible. Hence c1 = 0 and c2 = 0, which means B = 0.

2
Question 2(b) If T is a complex matrix of order 2 × 2 such that tr T = tr T2 = 0, then
show that T2 = 0.

Solution. Let λ1 , λ2 be the eigenvalues of T, then λ21 , λ22 are the eigenvalues of T2 . Given
that

tr T = λ1 + λ2 = 0
tr T2 = λ21 + λ22 = 0

0 = λ21 + λ22 = λ21 + (−λ1 )2 ⇒ λ1 = 0 and from λ1 + λ2 = 0 we get λ1 = λ2 = 0. The


characteristic equation of T is (x − λ1 )(x − λ2 ) = 0, or x2 = 0. By Cayley-Hamilton
theorem, we immediately get T2 = 0.

Question 2(c) Prove that a necessary and sufficient condition for an n × n real matrix A
to be similar to a diagonal matrix is that the set of characteristic vectors of A includes a set
of n linearly independent vectors.

Solution.
Necessity: By hypothesis there exists a nonsingular matrix P such that
 
λ1 0 . . . 0
 0 λ2 . . . 0 
P−1 AP = D =   ... ... ... ... 

0 0 . . . λn

Let P = [c1 , c2 , . . . , cn ], where each ci is an n-row column vector.

A[c1 , c2 , . . . , cn ] = [c1 , c2 , . . . , cn ]D = [λ1 c1 , λ2 c2 , . . . , λn cn ]

so Aci = λi ci for i = 1, . . . , n. Thus c1 , c2 , . . . , cn are characteristic vectors of A corre-


sponding to the eigenvalues λ1 , . . . , λn . Since P is nonsingular, c1 , c2 , . . . , cn are linearly
independent. Thus the set of characteristic vectors of A includes a set of n linearly indepen-
dent vectors.
Sufficiency: Let c1 , c2 , . . . , cn be n linearly independent eigenvectors of A corresponding
to eigenvalues λ1 , . . . , λn . Thus Aci = λi ci for i = 1, . . . , n. Let P = [c1 , c2 , . . . , cn ], then
P is nonsingular (otherwise 0 is an eigenvalue of P, so ∃x = (x1 , . . . , xn ) 6= 0 such that
Px = 0 ⇒ x1 c1 + . . . + xn cn = 0 ⇒ c1 , c2 , . . . , cn are not linearly independent.). Clearly
 
λ1 0 . . . 0
 0 λ2 . . . 0 
P−1 AP = D =   ... ... ... ... 

0 0 . . . λn

3
Question 3(a) Let A be a m × n matrix. Show that the sum of the rank and nullity of A
is n.

Solution. The matrix A can be regarded as a linear transformation A : F n −→ F m where


F is the field to which the entries of A belong, and the bases for F n , F m are standard bases.
Let T : V −→ W be a linear transformation, where dim(V) = n, dim(W) = m. We shall
show that dim(T(V)) + dim(kernel T) = n.
Take vn−r+1 , . . . , vn to be any basis of kernel T, where dim(kernel T) = r. Complete it
to a basis v1 , . . . , vn−r+1 . . . , vn of V. We shall show that T(v1 ), . . . , T(vn−r ) are linearly
independent and generate T(V), thus dim(T(V)) = n − r.
If w ∈ T(V), then ∃v ∈ V such that T(v) = w. If v = α1 v1 + . . . + αn vn , αi ∈ F, then
w = T(v) = α1 T(v1 ) + . . . + αn−r T(vn−r ) because T(vi ) = 0 for i > n − r. Thus T(V) is
generated by T(v1 ), . . . , T(vn−r ).
If α1 T(v1 ) + . . . + αn−r T(vn−r ) = 0, then T(α1 v1 + . . . + αn−r vn−r ) = 0. This implies
α1 v1 +. . .+αn−r vn−r ∈ kernel T ⇒ α1 v1 +. . .+αn−r vn−r = αn−r+1 vn−r+1 +. . .+αn vn . But
v1 , . . . , vn are linearly independent, so αi = 0 for i = 1, . . . , n. Hence T(v1 ), . . . , T(vn−r ) are
linearly independent, so they form a basis for T(V). Thus dim(T(V)) + dim(kernel T) = n.

Question 3(b) Find all real 2 × 2 matrices A with real eigenvalues which satisfy AA0 = I.

Solution. Since AA0 = I, |A| = ±1. If |A| = 1, then


    
a b a c 1 0
=
c d b d 0 1

so a2 + b2 = 1, c2 + d2 = 1, ac + bd = 0, ad − bc = 1. Let a = cos θ, b = sin θ. Then


c cos θ + d sin θ = 0 c cos θ sin θ + d sin2 θ = 0
⇒ ⇒ d = cos θ, c = − sin θ
−c sin θ + d cos θ = 1 −c sin θ cos θ + d cos2 θ = cos θ
 
cos θ sin θ
Thus A = , θ is real.
− sin θ cos θ
Now the eigenvalues of A are given by

cos θ − λ sin θ
|A − λI| = =0
− sin θ cos θ − λ

So (cos θ − λ)2 + sin2 θ = 0, or λ2 − 2λ cos θ + 1 = 0. Thus



2 cos θ ± 4 cos2 θ − 4
λ= = cos θ ± i sin θ
2
Since the eigenvalues of A are real, sin θ = 0, so cos θ = ±1. Thus
   
1 0 −1 0
A= ,
0 1 0 −1

4
 
0 1
If |A| = −1, J = , then |JA| = 1. Also JA(JA)0 = JAA0 J0 = JJ0 = I. Thus
1 0
 
cos θ sin θ
JA =
− sin θ cos θ
      
−1 cos θ sin θ 0 1 cos θ sin θ − sin θ cos θ
A=J = =
− sin θ cos θ 1 0 − sin θ cos θ cos θ sin θ
Now the eigenvalues of A are given by

λ + sin θ − cos θ
0 = |λI − A| = = λ2 − sin2 θ − cos2 θ = λ2 − 1
− cos θ λ − sin θ

Hence λ = ±1, so the eigenvalues are always real. Thus the possible values of A are
     
1 0 −1 0 − sin θ cos θ
, , for all real θ
0 1 0 −1 cos θ sin θ

Question 3(c) Reduce to diagonal matrix by rational congruent transformation the sym-
metric matrix  
1 2 −1
A= 2 0 3 
−1 3 1

Solution. The corresponding quadratic form is

x2 + z 2 + 4xy − 2xz + 6yz


= (x + 2y − z)2 − 4y 2 + 10yz
5 25
= (x + 2y − z)2 − 4(y − z)2 + z 2
4 4
25
= X 2 − 4Y 2 + Z 2
4
where X = x + 2y − z, Y = y − 5z/4, Z = z. From this we get z = Z, y = Y + 5Z/4, x =
X − 2Y − 32 Z. Thus

1 −2 − 32
     
1 0 0 1 0 0 1 2 −1
 0 −4 0  =  −2 1 5 
0  2 0 3  0 1 4
0 0 25 4
− 23 54 1 −1 3 1 0 0 1

5
UPSC Civil Services Main 1999 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

Question 1(a) Let V be the vector space of functions from R to R. Show that f, g, h ∈ V
are linearly independent where f (t) = e2t , g(t) = t2 and h(t) = t.

Solution. Let a, b, c ∈ R and let ae2t + bt2 + ct = 0 for all t. Setting t = 0 shows that
a = 0. From t = 1 we get b + c = 0, and t = −1 gives b − c = 0, hence b = c = 0. Thus
f, g, h are linearly independent.

Question 1(b) If the matrix  linear transformation T on V2 (R) with respect to the
 of the
1 1
basis B = {(1, 0), (0, 1)} is , then what is the matrix of T with respect to the ordered
1 1
basis B1 = {(1, 1), (1, −1)}.

Solution. T(e1 ) = e1 + e2 , T(e2 ) = e1 + e2 . Let v1 = (1, 1) = e1 + e2 , v2 = (1, −1) = e1 −


e2 . Then T(v1 ) = T((1, 1)) = (2, 2) = 2e1 + 2e2 = 2v1 . T(v2 ) = T((1, −1)) =(0, 0) =  0.
2 0
Thus the matrix of T with respect to the ordered basis B1 = {(1, 1), (1, −1)} is .
0 0
 
4 2 2
Question 1(c) Diagonalize the matrix A = 2 4 2.
2 2 4

1
Solution. The characteristic equation is

4 − x 2 2

0 = 2
4−x 2
2 2 4 − x
= (4 − x)((4 − x)2 − 4) + 2(4 − 8 + 2x) − 2(8 − 2x − 4)
= (4 − x)(12 − 8x + x2 ) − 8 + 4x − 8 + 4x
= 48 − 32x + 4x2 − 12x + 8x2 − x3 − 16 + 8x
= −(x3 − 12x2 + 36x − 32)
= −(x − 2)(x2 − 10x + 16) ∵ 2 is a root
= −(x − 2)(x − 2)(x − 8)
The characteristic roots are 2, 2, 8.
If (x1 , x2 , x3 ) is an eigenvector for λ = 8, then
    
−4 2 2 x1 0
 2 −4 2   x2  =  0 
2 2 −4 x3 0
Thus x1 = x2 = x3 , so (1, 1, 1) is an eigenvector for λ = 8.
Similarly for λ = 2,     
2 2 2 x1 0
 2 2 2   x2  =  0 
2 2 2 x3 0
Thus x1 + x2 + x3 = 0. Take x1 = 1, x2 = 0, so (1, 0, −1) is an eigenvector. Take x1 = 0, x2 =
1, so (0, 1, −1)
 is an eigenvector
 for λ = 2. These
 eigenvectors
 are linearly independent.
1 1 0 8 0 0
Thus if P =  1 0 1 , then P−1 AP =  0 2 0 
1 −1 −1   0 0 2
8 0 0
To check, verify that AP = P 0 2 0 

0 0 2
   
1 0 0 i
Question 2(a) Test for congruency the matrices A = and B = . Prove
0 −1 −i 0
that A2n = B2m = I where m, n are positive integers.
Solution. A and B are not congruent, because A is symmetric and B is not. If A ≡ B
then ∃P non-singular such that P0 AP = B which implies that B should be symmetric.
    
2 1 0 1 0 1 0
A = =
0 −1 0 −1 0 1
    
0 i 0 i 1 0
B2 = =
−i 0 −i 0 0 1

2
Hence A2n = (A2 )n = I, and B2m = (B2 )m = I.

Question 2(b) If A is a skew symmetric matrix of order n then prove that (I−A)(I+A)−1
is orthogonal.

Solution.

O = (I − A)(I + A)−1
OO0 = (I − A)(I + A)−1 ((I + A)−1 )0 (I − A)0
= (I − A)(I + A)−1 (I − A)−1 (I + A) as A0 = −A
= (I − A)[(I − A)(I + A)]−1 (I + A)
= (I − A)[I − A2 )]−1 (I + A)
= (I − A)[(I + A)(I − A)]−1 (I + A)
= (I − A)(I − A)−1 (I + A)−1 (I + A)
= I

Similarly it can be shown that O0 O = I. Hence O is orthogonal.

Question 2(c) Test for positive definiteness the quadratic form 2x2 + y 2 + 2z 2 + 2xy − 2zx.

Solution. The given form is


1
2
(4x2 + 3y 2 + 4z 2 + 4xy − 4zx)
= 21 ((2x + y − z)2 + y 2 + 3z 2 + 2yz)
= 21 ((2x + y − z)2 + (y + z)2 + 2z 2 )
Now 2x + y − z = 0, y + z = 0, z = 0 implies x = y = z = 0. Hence the form is positive
definite.

Question 2(d) Reduce the equation

x2 + y 2 + z 2 − 2xy − 2yz + 2zx + x − y − 2z + 6 = 0

into canonical form and determine the nature of the quadric.

Solution. Consider x2 + y 2 + z 2 − 2xy − 2yz + 2zx. Its matrix is


 
1 −1 1
S =  −1 1 −1 
1 −1 1

Its characteristic equation is



1 − λ −1 1

−1 1 − λ −1 =0

1 −1 1 − λ

3
or λ3 − 3λ2 = 0. Thus λ = 0, 0, 3.
We next determine the characteristic vectors. For λ = 0, we get
    
1 −1 1 x1 0
 −1 1 −1   x2  =  0 
1 −1 1 x3 0

Thus x1 − x2 + x3 = 0. Take (1, 1, 0) and (−1, 1, 2) as orthogonal characteristic vectors


corresponding to λ = 0.
For λ = 3, we get     
−2 −1 1 x1 0
 −1 −2 −1   x2  =  0 
1 −1 −2 x3 0
This yields x1 = −x2 = x3 . Take (1, −1, 1) as the characteristic vector for λ = 3.
Thus if  1 
√ √1 − √1
 3 2 6
O =  − √13 √12 √16 

√1 0 √2
3 6
 
3 0 0
Then O0 SO =  0 0 0 .
0 0 0
Let    
X x
O  Y  =  y 
Z z
or
X Y Z
x = √ +√ −√
3 2 6
X Y Z
y = −√ + √ + √
3 2 6
X 2Z
z = √ +√
3 6
Thus the given equation can be transformed to
X Y
3X 2 + √ + √ − √Z6 + √X3 − √Y2 − √Z6 − 2X
√ − 4Z
√ +6=0
3 2
2
√ 3 6
⇒ 3X − Z 6 + 6 = 0
q √
⇒ 32 X 2 = Z − 6

√ 2
q
2
Shifting the origin to (0, 0, 6), we get X = 3
Z, showing that the equation is a parabolic
cylinder.

4
1 Reduction of Quadrics
For the sake of completeness, we give the complete theoretical discussion for the above
question.
Let

F (x, y, z) = ax2 + by 2 + cz 2 + 2f yz + 2gzx + 2hxy + 2ux + 2vy + 2wz + d = 0

It can be expressed in matrix form as


  
a h g u x
 h b f v  y
  
x y z 1 
 g
=0
f c w  z 
u v w d 1

Let the 4 × 4 matrix be Q.


Step I. Consider
    
 a h g x  x
x y z  h b f  y  = x y z S y 
g f c z z

1. Find the characteristic roots of S — λ1 , λ2 , λ3 .

2. Find characteristic vectors v1 , v2 , v2 corresponding to λ1 , λ2 , λ3 which are orthogonal.


These on normalization give us
 
O = ||vv11 || ||vv22 || ||vv33 ||

Thus we get O0 SO =diagonal(λ1 , λ2 , λ3 ). Let


   
X x
O  Y  =  y 
Z z

This gives us three equations expressing x, y, z in term of X, Y, Z. Substituting in F , we get

F (X, Y, Z) = λ1 X 2 + λ2 Y 2 + λ3 Z 2 + 2U X + 2V Y + 2W Z + d = 0

(Note that d is unaffected.)


  
X x
Note 1 Since O is orthogonal, the transformation O  Y  =  y  is just a rotation
Z z
of the axes, and therefore the nature of the quadric is unaffected.

5
Step II. We now consider 3 possibilities (ρ is rank of the matrix):
1. ρ(S) = 3 ⇒ λ1 λ2 λ3 6= 0. Shift the origin to (− λU1 , − λV2 , − W
λ3
), i.e. x = X + λU1 , y =
Y + λV2 , z = Z + W
λ3
. (Actually we are just completing the squares.) F gets transformed
to
λ1 x2 + λ2 y 2 + λ3 z 2 + d2 = 0
2. ρ(S) = 2. One characteristic root, say λ3 = 0. Shift the origin to (− λU1 , − λV2 , 0), and F
gets transformed to
λ1 x2 + λ2 y 2 + 2w2 z + d2 = 0
3. ρ(S) = 1. Two characteristic roots, say λ2 = λ3 = 0. Shift the origin to (− λU1 , 0, 0),
and F gets transformed to
λ1 x2 + 2v2 y + 2w2 z + d2 = 0

Note that ρ(S) = 0 ⇒ λ1 = λ2 = λ3 = 0. Then F (x, y, z) is no longer a quadric, it is a


plane.
Step III. Observe that ρ(S) ≤ ρ(Q) ≤ 4, ρ(S) ≤ 3.
1. Let ρ(Q) = 4, ρ(S) = 3. As shown above, F (x, y, z) = 0 is transformed to
λ1 x2 + λ2 y 2 + λ3 z 2 + d2 = 0
|Q| = λ1 λ2 λ3 d2 ⇒ d2 = |Q||S|
. Thus the quadric is λ1 x2 + λ2 y 2 + λ3 z 2 = − |Q|
|S|
, which
is a central quadric i.e. a quadric surface with a center, e.g., a sphere, ellipsoid, or
hyperboloid, depending upon the signs and magnitudes of the eigenvalues. If the right
hand side has positive sign (maybe by multiplying the equation with -1), then look at
the signs of the coefficients of the l.h.s. If all are positive, it is an ellipsoid, further
if all are equal, it is a sphere. If 1 or 2 are negative, it is a hyperboloid. If all 3 are
negative, the surface is the empty set.
2. ρ(Q) = 3, ρ(S) = 3. |Q| = λ1 λ2 λ3 d2 = 0 ⇒ d2 = 0, because λ1 λ2 λ3 6= 0. Thus the
quadric becomes λ1 x2 + λ2 y 2 + λ3 z 2 = 0, which is a cone.
3. ρ(Q) = 4, ρ(S) = 2
F (x, y, z) = λ1 x2 + λ2 y 2 + 2w2 z + d2 = 0
 
λ1 0 0 0
 0 λ2 0 0 
Q=  0 0 0 w2 

0 0 w 2 d2
d2
|Q| = −λ1 λ2 w22 . Since ρ(Q) = 4, w2 6= 0. Shifting the origin to (0, 0, − 2w 2
) we get

F (x, y, z) = λ1 x2 + λ2 y 2 + 2w2 z = 0
where w22 = −|Q|/λ1 λ2 . The surface is a paraboloid.

6
4. ρ(Q) = 3, ρ(S) = 2

F (x, y, z) = λ1 x2 + λ2 y 2 + 2w2 z + d2 = 0

ρ(Q) = 3 ⇒ |Q| = −λ1 λ2 w22 ⇒ w2 = 0. Since


 
λ1 0 0 0
 0 λ2 0 0 
Q=  0 0

0 0 
0 0 0 d2

and ρ(Q) = 3, d2 6= 0. Thus

F (x, y, z) = λ1 x2 + λ2 y 2 + d2 = 0

The quadric is a hyperbolic or elliptic cylinder.

5. ρ(Q) = 2, ρ(S) = 2  
λ1 0 0 0
 0 λ2 0 0 
Q=
 0 0

0 0 
0 0 0 d2
ρ(Q) = 2 ⇒ d2 = 0, and F (x, y, z) = λ1 x2 + λ2 y 2 = 0. The quadric is a pair of distinct
planes or a point, if λ1 = λ2 6= 0.

6. ρ(Q) = 4, ρ(S) = 1

F (x, y, z) = λ1 x2 + 2v2 y + 2w2 z + d2 = 0


 
λ1 0 0 0
 0 0 0 v2 
Q=  0 0 0 w2 

0 v2 w2 d2
which shows that ρ(Q) = 4 is not possible.

7. ρ(Q) = 3, ρ(S) = 1

F (x, y, z) = λ1 x2 + 2v2 y + 2w2 z + d2 = 0

ρ(Q) = 3 ⇒ both v2 and w2 cannot be 0. Suppose v2 6= 0. Shift the origin to


d2
(0, − 2v2
, 0).
F (x, y, z) = λ1 x2 + 2v2 y + 2w2 z = 0

7
Rotate the axes by

x = X
v2 w2
y = p 2 2
Y −p 2 Z
v2 + w2 v2 + w22
w2 v
z = p 2 Y + p 2 Z
v2 + w22 v22 + w22
q
2
F (x, y, z) = λ1 X + 2v3 Y = 0 v3 = v22 + w22
Thus the quadric is a parabolic cylinder.

8. ρ(Q) = 2, ρ(S) = 1 ρ(Q) = 2 ⇒ v2 = w2 = 0, d2 6= 0.

F (x, y, z) = λ1 X 2 + d2 = 0
The quadric is two parallel planes.

9. ρ(Q) = 1, ρ(S) = 1 The quadric immediately reduces to F (x, y, z) = λ1 X 2 = 0, so it


represents two coincident planes x = 0.
ρ(Q) ρ(S) Surface Canonical form
4 3 central quadric λ1 x + λ2 y 2 + λ3 z 2 = − |Q|
2
|S|
2 2 2
3 3 cone λ1 x + λ2 y + λ3 z = 0
4 2 paraboloid λ1 x + λ2 y 2 + 2w2 z = 0, w22 = − λ|Q|
2
1 λ2
3 2 elliptic or hyperbolic cylinder λ1 x2 + λ2 y 2 + d2 = 0
pair of distinct planes if λ1 λ2 < 0
2 2 λ1 x2 + λ2 y 2 = 0
point if λ1 λ2 > 0
4 1 Not possible p
3 1 parabolic cylinder λ1 X 2 + 2v3 Y = 0, v3 = v22 + w22
2 1 pair of parallel planes λ1 X 2 + d2 = 0
1 1 Two coincident planes λ1 X 2 = 0

8
UPSC Civil Services Main 2000 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

Question 1(a) Let V be a vector space over R and let

T = {(x, y) | x, y ∈ V}

Define (x, y) + (x1 , y1 ) = (x + x1 , y + y1 ) in T and (α + iβ)(x, y) = (αx − βy, βx + αy)


for every α, β ∈ R. Show that T is a vector space over C.

Solution.

1. v1 , v2 ∈ T ⇒ v1 + v2 ∈ T

2. (0, 0) is the additive identity where 0 is the zero vector in V.

3. If (x, y) ∈ T , then (−x, −y) ∈ T , and (x, y) + (−x, −y) = (0, 0)

4. Clearly v1 +v2 = v2 +v1 and (v1 +v2 )+v3 = v1 +(v2 +v3 ) as addition is commutative
and associative in V.

5. z ∈ C, v ∈ T ⇒ zv ∈ T

6. 1v = (1 + i0)(x, y) = (x, y)

7.

(α + iβ)((x1 , y1 ) + (x2 , y2 ))
= (α(x1 + x2 ) − β(y1 + y2 ), β(x1 + x2 ) + α(y1 + y2 ))
= (αx1 − βy1 , βx1 + αy1 ) + (αx2 − βy2 , βx2 + αy2 )
= (α + iβ)(x1 , y1 ) + (α + iβ)(x2 , y2 )

1
8.

((α + iβ)(γ + iδ))(x, y)


= (αγ − βδ + i(αδ + βγ))(x, y)
= ((αγ − βδ)x − (αδ + βγ)y, (αγ − βδ)y + (αδ + βγ)x)
= (α(γx − δy) − β(δx + γy), β(γx − δy) + α(δx + γy))
= (α + iβ)((γx − δy), (δx + γy))
= (α + iβ)((γ + iδ)(x, y))

Thus T is a vector space over C.

Question 1(b) Show that if λ is a characteristic root of a non-singular matrix A, then λ−1
is a characteristic root of A−1 .

Solution.
Av = λv v 6= 0
−1 −1
⇒ A Av = λA v
⇒ A−1 v = λ−1 v
Thus λ−1 is a characteristic root of A−1 .

Question 2(a) Prove that a real symmetric matrix A is positive definite


 if and only if
1 2 3
A = BB0 for some non-singular B. Show also that A = 2 5 7  is positive definite
3 7 11
0 0
and find B such that A = BB . (Here B is the transpose of B.)

Solution. If A = BB0 for some non-singular B, then x0 Ax = x0 BB0 x where x 6= 0 is a


column vector. Since |B| 6= 0, B0 x 6= 0 =⇒ x0 B.(B0 x) is the sum on n squares, at least one
of which is non-zero. Thus x0 Ax > 0 whenever x 6= 0, showing that A is positive definite.
Conversely, if A is positive definite, then ∃P non-singular such that P0 AP = In . Thus
A = P0−1 P−1 . Letting B = P0−1 we get A = BB0 as required.
The existence of P can be found by induction on n. Let
 
a11 a12 . . . a1n
 a21 a22 . . . a2n 
A= 
 ... 
an1 an2 . . . ann

Define
1 − aa11 . . . − aa1n
 12

11
 0 1 ... 0 
Q= 
 ... 
0 0 ... 1

2
 
0 a11 0
Then Q is non-singular, and Q AQ = , where S is (n − 1) × (n − 1) positive
0 S
definite. Let Q∗ be a (n − 1) ×  (n − 1) non-singular
 matrix such that Q∗ 0 SQ∗ is diagonal,
1 0
by induction. Then let Q1 = , and let P = Q1 Q. Then P0 AP is diagonal
0 Q∗
(b11 , b22 , . . . , bnn ). Let B = diagonal ( √b111 , . . . , √b1nn ). Then B0 P0 APB = In .
The quadratic form Q(x, y, z) associated with the given matrix A is given by
  
 1 2 3 x
x y z  2 5 7   y  = x2 + 5y 2 + 11z 2 + 4xy + 6xz + 14yz
3 7 11 z

Completing the squares we get Q(x, y, z) = (x + 2y + 3z)2 + (y + z)2 + z 2 , so A is positive


definite, as z = 0, y + z = 0, x + 2y + 3z = 0 =⇒ x = y = z = 0.
If B is a 3 × 3 matrix such that
   
x x + 2y + 3z
B0  y  =  y+z 
z z

then x0 BB0 x = Q = x0 Ax, so A = BB0 as A and BB0 are both symmetric. Clearly
 
1 0 0
B=  2 1 0 
3 1 1

and it can easily be verified that A = BB0 .

Question 2(b) Prove that a system Ax = B of non-homogeneous equations in n unknowns


has a unique solution provided the coefficient matrix is non-singular.

Solution. If A is non-singular, then the system is consistent because the rank of the
coefficient matrix A = n = rank of the n × n + 1 augmented matrix (A, B). If x1 , x2 are
two solutions, then
Ax1 = B = Ax2
=⇒ A(x1 − x2 ) = 0
=⇒ A−1 A(x1 − x2 ) = 0
=⇒ x1 = x2
Thus the unique solution is given by the column vector x = A−1 B.

3
Question 2(c) Prove that two similar matrices have the same characteristic roots. Is the
converse true? Justify your claim.

Solution. Let B = P−1 AP then characteristic polynomial of B is |λI − B| = |λI −


P−1 AP| = |P−1 λIP − P−1 AP| = |P−1 ||λI − A||P| = |λI − A|. (Note that |X||Y| = |XY|.)
Thus the characteristic polynomial of B is the same as that of A, so both A and B have the
same characteristic roots.
The converse is not true. Let
   
1 1 1 0
A= , B=
0 1 0 1

Then A and B have the same characteristic polynomial (λ − 1)2 and thus the same charac-
teristic roots. But B can never be similar to A because P−1 BP = B whatever P may be.

4
UPSC Civil Services Main 2001 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

Question 1(a) Show that the vectors (1, 0, −1), (0, −3, 2) and (1, 2, 1) form a basis of the
vector space R3 (R).
Solution. Since dimR (R3 ) = 3, it is enough to prove that these are linearly independent.
If possible, let
a(1, 0, −1) + b(0, −3, 2) + c(1, 2, 1) = 0
This implies
a + c = 0, −3b + 2c = 0, −a + 2b + c = 0
Solving for c, c + 34 c + c = 0, so c = 0, hence a = b = 0. (Note that if these linearly
independent vectors were not a basis, they could be completed into one, but in R3 any four
vectors are linearly dependent, so this is a maximal linearly independent set, hence it is a
basis.
Alternate Solution. Since dim(R3 ) = 3, to show that (1, 0, −1), (0, −3, 2) and (1, 2, 1)
form a basis it is enough to show that these vectors generate R3 . In fact, given (x1 , x2 , x3 ),
we can always find a, b, c s.t. (x1 , x2 , x3 ) = a(1, 0, −1) + b(0, −3, 2) + c(1, 2, 1) as follows:
a + c = x1 , −3b + 2c = x2 , −a + 2b + c = x3 . Thus (c − x1 ) + 2(2c − x2 )/3 + c = x3 ,
so c + 34 c + c = x1 + 23 x2 + x3 . Thus c = 3x1 +2x 10
2 +3x3
, a = x1 − c = 7x1 −2x10
2 −3x3
, and
2c−x2 x1 −x2 +x3
b= 3 = 5
.
|A|
Question 1(b) If λ is a characteristic root of a non-singular matrix A, then prove that λ
is a characteristic root of Adj A.
Solution. If µ is a characteristic root of A, then aµ is a characteristic root of aA for a
constant a, because if Av = µv, v 6= 0 a vector, then aAv = aµv. Hence the result.
If λ is the characteristic root of A, |A| 6= 0, then λ 6= 0, and λ−1 is a characteristic root
of A−1 , because Av = λv =⇒ A−1 v = λ−1 v.
Since Adj A = A−1 |A|, it follows that |A| λ
is a characteristic root of Adj A.

1
 
1 0 0
Question 2(a) If A =  1 0 1  show that for all integers n ≥ 3, An = An−2 + A2 − I.
0 1 0
50
Hence determine A .

Solution. Characteristic equation of A is



λ−1 0 0

1 λ 1 = 0

0 1 λ

or (λ−1)(λ2 −1) = λ3 −λ2 −λ+1 = 0. From the Cayley-Hamilton theorem, A3 −A2 −A+I =
0 ⇒ A3 = A + A2 − I. Thus the result is true for n = 3. Suppose the theorem is true for
n = m i.e. Am = Am−2 + A2 − I. We shall prove it for m + 1.

Am+1 = Am A
= (Am−2 + A2 − I)A
= Am−1 + A3 − A
= Am−1 + A2 + A − A − I
= Am−1 + A2 − I

The result follows by induction.


Let n = 2m. Using successively An = An−2 + A2 − I, we get A2m = mA2 − (m − 1)I.
Now     
1 0 0 1 0 0 1 0 0
A2 =  1 0 1  1 0 1  =  1 1 0 
0 1 0 0 1 0 1 0 1
so

A50 = 25A

2
− 24I   
25 0 0 24 0 0
=  25 25 0  −  0 24 0 
25 0 25 0 0 24
 
1 0 0
=  25 1 0 
25 0 1

Question 2(b) When is a square matrix A said to be congruent to a square matrix B?


Prove that every matrix congruent to a skew-symmetric matrix is skew-symmetric.

Solution. A ≡ B if ∃P nonsingular, s.t. P0 AP = B. If S0 = −S then (P0 SP)0 = P0 S0 P =


−(P0 SP), so P0 SP is also skew-symmetric.

2
Question
 2(c) Determine
 the orthogonal matrix P such that P−1 AP is diagonal where
7 4 −4
A=  4 −8 −1 .
−4 −1 −8

Solution. The characteristic equation is



λ − 7 −4 4

−4 λ + 8 1 = 0

4 1 λ + 8
(λ − 7)((λ + 8)2 − 1) + 4(−4 − 4λ − 32) + 4(−4 − 4λ − 32) = 0
λ3 + 9λ2 − 81λ − 729 = 0
(λ + 9)(λ2 − 81) = 0

Thus λ = 9, −9, −9.

1. λ = 9. If (x1 , x2 , x3 ) is the eigenvector corresponding to λ = 9, we get

2x1 − 4x2 + 4x3 = 0


−4x1 + 17x2 + x3 = 0
4x1 + x2 + 17x3 = 0

From the second and third we get 18x2 +18x3 = 0. Take x2 = 1. Then x3 = −1, x1 = 4,
so (4, 1, −1) is an eigenvector for λ = 9.

2. λ = −9. If (x1 , x2 , x3 ) is the eigenvector corresponding to λ = −9, we get

−16x1 − 4x2 + 4x3 = 0


−4x1 − x2 + x3 = 0
4x1 + x2 − x3 = 0

There is only one equation 4x1 + x2 − x3 = 0. Take x1 = 0, x2 = 1, then x3 = 1,


so (0, 1, 1) is an eigenvector. Take x1 = −1, x2 = 2, then x3 = −2, so (−1, 2, −2) is
another eigenvector. These two are orthogonal to each other and are eigenvectors for
λ = −9. Note that to make the second vector orthogonal to the first, we needed to
ensure x2 = −x3 , then the equation suggested values for x1 , x2 .

Let  
0 − 13 √4
18
P= √1 2 √1
 
2 3 18 
√1 − 32 − √118
2

3
Clearly P0 P = I, since the columns of P are mutually orthogonal unit vectors.
 Moreover from

−9 0 0
Ax = xλ for the eigenvalues and eigenvectors, it follows that AP = P  0 −9 0 .
  0 0 9
−9 0 0
−1
Thus P AP =  0 −9 0 , which is diagonal as required.
0 0 9

Question 2(d) Show that the real quadratic form

Φ = n(x21 + x22 + . . . + x2n ) − (x1 + x2 + . . . + xn )2

in n variables is positive semi-definite.

Solution. Consider the expression

E = (X − x1 )2 + . . . + (X − xn )2
= nX 2 − 2X(x1 + . . . + xn ) + (x21 + x22 + . . . + x2n )

Clearly E being the sum of squares is non-negative, i.e. E ≥ 0. Let

(x1 + x2 + . . . + xn ) (x21 + x22 + . . . + x2n )


A= B=
n n
Then E = n(X 2 − 2AX + B) = n((X − A)2 + B − A2 ). When X = A, E = n(B − A2 ) = Φ,
and since E ≥ 0, Φ ≥ 0.
If x1 = x2 = . . . = xn = 1, then Φ = 0 showing that Φ is actually positive semi-definite.
Alternate solution. By Cauchy’s inequality
n
! n ! n
!2
X X X
a2i b2i ≥ ai bi
i=1 i=1 i=1

Setting b1 = b2 = . . . = bn = 1, we get
n
! n
!2
X X
n a2i − ai ≥0
i=1 i=1

showing that Φ is positive semi-definite.

4
UPSC Civil Services Main 2002 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

June 14, 2007

Question 1(a) Show that the mapping T : R3 −→ R3 where T(a, b, c) = (a − b, b − c, a + c)


is linear and non-singular.

Solution.

T(λa, λb, λc) = (λ(a − b), λ(b − c), λ(a + c))


= λ(a − b, b − c, a + c)
= λT(a, b, c)

T(a, b, c) + T(x, y, z) = (a − b, b − c, a + c) + (x − y, y − z, x + z)
= (a − b + x − y, b − c + y − z, a + c + x + z)
= T(a + x, b + y, c + z)

Thus T is linear.
Now we show that

T(1, 0, 0) = (1, 0, 1)
T(0, 1, 0) = (−1, 1, 0)
T(0, 0, 1) = (0, −1, 1)

are linearly independent.

a1 (1, 0, 1) + a2 (−1, 1, 0) + a3 (0, −1, 1) = 0


⇒ a1 − a2 = 0, a2 − a3 = 0, a1 + a3 = 0
⇒ a1 = a2 = a3 = 0

1
Thus (1, 0, 1), (−1, 1, 0), (0, −1, 1) are linearly independent.
Since (1, 0, 0), (0, 1, 0), (0, 0, 1) generate R3 , (1, 0, 1), (−1, 1, 0), (0, −1, 1) generate T(R3 ),
hence dim(T(R3 )) = 3. Thus T is non-singular.
Alternatively,

T(a, b, c) = (0, 0, 0) ⇐⇒ a − b = 0, b − c = 0, a + c = 0 =⇒ a = b = c = 0

Thus T is 1-1, therefore it is onto, which shows it is nonsingular.

Question 1(b) Prove that a square matrix A is non-singular if and only if the constant
term in its characteristic polynomial is different from 0.

Solution. Let  
a11 ... a1n
 a21 ... a2n 
A=
. . .

... ...
an1 ... ann
Then characteristic polynomial of A = Det(xI − A). I = n × n unit matrix. Clearly
n
X
Det(xI − A) = xn − aii xn−1 + . . . + (−1)n DetA
i=1

Thus A is nonsingular iff the constant term in the characteristic polynomial of A 6= 0.

Question 2(a) Let T : R5 −→ R5 be a linear mapping given by T (a, b, c, d, e) = (b − d, d +


e, b, 2d + e, b + e). Obtain bases for its null space and range space.

Solution. Clearly
T (1, 0, 0, 0, 0) = (0, 0, 0, 0, 0)
T (0, 1, 0, 0, 0) = (1, 0, 1, 0, 1)
T (0, 0, 1, 0, 0) = (0, 0, 0, 0, 0)
T (0, 0, 0, 1, 0) = (−1, 1, 0, 2, 0)
T (0, 0, 0, 0, 1) = (0, 1, 0, 1, 1)
are generators of the range space of T . In fact, if v1 = (1, 0, 1, 0, 1), v2 = (−1, 1, 0, 2, 0), v3 =
(0, 1, 0, 1, 1) then v1 , v2 , v3 generate T (R5 ). We now show that v1 , v2 , v3 are linearly inde-
pendent. Let α1 v1 +α2 v2 +α3 v3 = 0. Then α1 −α2 = 0, α2 +α3 = 0, α1 = 0 ⇒ α2 = α3 = 0.
Thus v1 , v2 , v3 are linearly independent over R ⇒ T (R5 ) is of dimension 3 with basis
v 1 , v2 , v3 .
Thus the null space is of dimension 2, because dim(null space) + dim(range space) =
dim(given vector space = R5 ) = 5. Since e1 = (1, 0, 0, 0, 0) and e3 = (0, 0, 1, 0, 0) belong to
the null space of T , and both are linearly independent over R, e1 , e3 is a basis of the null
space of T .

2
Question 2(b) Let A be a 3 × 3 real symmetric matrix with eigenvalues 0, 0, 5. If the
corresponding eigenvectors are (2, 0, 1), (2, 1, 1), (1, 0, −2) then find the matrix A.
     
2 2 1 0 0 0 0 0 0
Solution. Let P = 0 1 0 , then P−1 AP = 0 0 0, so A = P 0 0 0 P−1 .
1 1 −2 2 0 0 5 0 0 5
1

5
−1 5
−1
A simple calculation shows that P = 0 1  0 , therefore
1
5
0 − 25
 2
−1 15
    
2 2 1 0 0 0 5
1 0 −2
A = 0 1 0  0 0 0  0 1 0 = 0 0 0 
1
1 1 −2 0 0 5 5
0 − 25 −2 0 4
 
1 0 −2
Thus  0 0 0  is the required symmetric matrix with 0, 0, 5 as eigenvalues.
−2 0 4

Question 2(c) Solve the following system of linear equations:

x1 − 2x2 − 3x3 + 4x4 = −1


−x1 + 3x2 + 5x3 − 5x4 − 2x5 = 0
2x1 + x2 − 2x3 + 3x4 − 4x5 = 17

Solution. There are three equations


in 5 unknowns, therefore the rank of the coefficient
1 −2 −3

matrix ≤ 3. Since −1 3 5 = 1(−6 − 5) + 2(2 − 10) + (−3)(−1 − 6) = −6, the rank
−2 1 −2
of the coefficient matrix is 3. Using Cramer’s rule we solve the system

x1 − 2x2 − 3x3 = −1 − 4x4 (1)


−x1 + 3x2 + 5x3 = 5x4 + 2x5 (2)
2x1 + x2 − 2x3 = 17 − 3x4 + 4x5 (3)

3

−1 − 4x4 −2 −3
1
x1 = − 5x4 + 2x5 3 5
6
17 − 3x4 + 4x5 1 −2
1
= − [(−1 − 4x4 )(−11) − 3(5x4 + 2x5 − 51 + 9x4 − 12x5 ) + 2(−10x4 − 4x5 − 85 + 15x4 − 20x5 )]
6
1
= − [−6 + 44x4 − 42x4 + 10x4 + 30x5 − 48x5 ]
6
= 1 − 2x4 + 3x5

1 −1 − 4x 4 −3
1
x2 = − −1 5x4 + 2x5 5
6
2 17 − 3x4 + 4x5 −2
1
= − [−10x4 − 4x5 − 85 + 15x4 − 20x5 − 8 − 32x4 + 51 − 9x4 + 12x5 + 30x4 + 12x5 ]
6
1
= − [−42 − 6x4 ]
6
= 7 + x4

1 −2 −1 − 4x 4

1
x3 = − −1 3 5x4 + 2x5
6
2 1 17 − 3x4 + 4x5
1
= − [51 − 9x4 + 12x5 − 5x4 − 2x5 − 34 + 6x4 − 8x5 − 20x4 − 8x5 + 7 + 28x4 ]
6
1
= − [24 − 6x5 ]
6
= −4 + x5

The solution space is (1 − 2x4 + 3x5 , 7 + x4 , −4 + x5 , x4 , x5 ), where x4 , x5 ∈ R (arbitrarily).


Note that the vector space of solutions is of dimension 2.
Alternate Method.

x2 + 2x3 = −1 + x4 + 2x5 adding (1) and (2) (4)


7x2 + 8x3 = 17 + 7x4 + 8x5 adding 2×(2) and (3) (5)
6x3 = −24 + 6x5 using 7×(4) - (5) (6)
x2 = 7 + x4 from (4) and (6) (7)
x1 = 1 − 2x4 + 3x5 using (1), (6), (7) (8)

The solution space is as shown above.

4
Question 2(d) Use Cayley-Hamilton theorem to find the inverse of the following matrix
 
0 1 2
A= 1  2 3
3 1 1

Solution. Characteristic polynomial is given by |xI − A| = 0, where I is the 3 × 3 unit


matrix.

x
−1 −2

−1 x − 2 −3 = 0

−3 −1 x − 1
x[x2 − 3x + 2 − 3] + 1[−x + 1 − 9] − 2[1 + 3x − 6] = 0
x3 − 3x2 − 8x + 2 = 0

By Cayley-Hamilton theorem, A3 − 3A2 − 8A + 2I = 0, or A(A2 − 3A − 8I) = −2I. Thus


1
A−1 = − (A2 − 3A − 8I)
2      
7 4 5 0 3 6 8 0 0
1
= − 11 8 11 − 3 6 9 − 0 8 0
2
4 6 10 9 3 3 0 0 8
 
−1 1 −1
1
= −  8 −6 2 
2
−5 3 −1
 1 1 1

2
− 2 2
= −4 3 −1
5
2
− 23 12

Check A−1 A = I.

5
UPSC Civil Services Main 2003 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

September 16, 2007

Question 1(a) Let S be any non-empty subset of a vector space V over the field F . Show
that the set {a1 x1 + . . . + an xn | a1 , . . . , an ∈ F, x1 , . . . xn ∈ S, n ∈ N} is the subspace
generated by S.

Solution. Let W be the subset mentioned above. Let w1 , w2 ∈ W and a, b ∈ F . Then


w1 = a1 x1 + . . . ar xr , where a1 , . . . , ar ∈ F, x1 , . . . xr ∈ S and w2 = b1 y1 + . . . bs ys where
b1 , . . . bs ∈ F, y1 , . . . , ys ∈ S. Now αw1 +βw2 = c1 z1 +. . .+cr+s zr+s , where ci = αai , 1 ≤ i ≤
r, cj+r = βbj , 1 ≤ j ≤ s, and zi = xi , 1 ≤ i ≤ r, zj+r = yj , 1 ≤ j ≤ s. Clearly cj ∈ F, zj ∈ S
for 1 ≤ j ≤ r + s, showing that for any w1 , w2 ∈ W, α, β ∈ F, αw1 + βw2 ∈ W, moreover
W 6= ∅ as S ⊆ W and S = 6 ∅. Thus W is a subspace of V.

2 1 1
Question 1(b) If A = 0 1 0, then find the matrix represented by 2A10 − 10A9 +
1 1 2
14A − 6A − 3A + 15A − 21A4 + 9A3 + A − I.
8 7 6 5

Solution. The characteristic equation of A is



2 − x 1 1

|A − xI| = 0 1−x 0 = (2 − x)2 (1 − x) − (1 − x) = 0
1 1 2 − x

or (1 − x)(4 − 4x + x2 ) − 1 + x = 3 − 7x + 5x2 − x3 = 0, or x3 − 5x2 + 7x − 3 = 0. By the

1
Cayley-Hamilton theorem, we get A3 − 5A2 + 7A − 3I = 0. Now

2A10 − 10A9 + 14A8 − 6A7 − 3A6 + 15A5 − 21A4 + 9A3 + A − I


= 2A7 [A3 − 5A2 + 7A − 3I] − 3A3 [A3 − 5A2 + 7A − 3I] + A − I
= 2A7 0 − 3A 3
 0 + A−I
1 1 1
= A − I = 0 0 0
1 1 1

which is the required value.

Question 2(a) Prove that the eigenvectors corresponding to distinct eigenvalues of a square
matrix are linearly independent.

Solution. Let x1 , x2 , . . . , xk be eigenvectors corresponding to distinct eigenvalues λ1 , . . . , λk


of the square matrix A.
We will show that if any subset of the vectors x1 , x2 , . . . , xk is linearly dependent, then
we can find a smaller set that is also linearly dependent — but this leads to a contradiction
as the eigenvectors are all non-zero.
Suppose, without loss of generality, that x1 , x2 , . . . , xr are linearly dependent. Thus there
exist α1 , . . . αr ∈ R, not all zero, such that

α 1 x1 + . . . + α r x r = 0 (1)

Thus A(α1 x1 + . . . + αr xr ) = 0 ⇒ α1 λ1 x1 + . . . + αr λr xr = 0. Multiplying (1) by λ1 and


subtracting, we have α2 (λ2 − λ1 )x2 + . . . + αr (λr − λ1 )xr = 0. Now αi 6= 0 ⇒ αi (λi − λ1 ) 6=
0, so not all αi (λi − λ1 ) can be zero, so we have a smaller set x2 , . . . , xr which is also
linearly dependent. This leads us to the contradiction mentioned above, hence the vectors
x1 , x2 , . . . , xk must be linearly independent.

Question 2(b) If H is a Hermitian matrix, then show that (H + iI)−1 (H − iI) is a unitary
matrix. Also show that every unitary matrix A can be written in this form provided 1 is not
an eigenvalue of A.

Solution. See related results of 1989, question 2(b).


 
6 −2 2
Question 2(c) If A = −2
 3 −1 then find a diagonal matrix D and a matrix B
2 −1 3
such that A = BDB where B0
0
denotes the transpose of B.

2
 
 x1
Solution. Let Q(x1 , x2 , x3 ) = x1 x2 x3 A x2  be the quadratic form associated with

x3
A. Then

Q(x1 , x2 , x3 ) = 6x21 + 3x22 + 3x23 − 4x1 x2 + 4x1 x3 − 2x2 x3


1 1 7 7 2
= 6[x1 − x2 + x3 ]2 + x22 + x23 − x2 x3
3 3 3 3 3
1 1 2 7 1 2 16 2
= 6[x1 − x2 + x3 ] + [x2 − x3 ] + x3
3 3 3 7 7
 
6 0 0
Let X1 = x1 − 31 x2 + 31 x3 , X2 = x2 − 71 x3 , X3 = x3 and D = 0 73 0 . Then
0 0 16 7
     
x1 x1 X1
0 
  
x1 x2 x3 A x2 = x1 x2 x3 BDB x2 = X1 X2 X3 D X2 
  
x3 x3 X3

1 − 31 31
        
X1 x1 x1 6 0 0
1  
where X2 = 0 1 − 7
   x2 = B x2 . Thus A = BDB where D = 0 73
0  0  0
16
X3 0 0 1 x3 x3 0 0 7
1 0 0
1
and B = − 3 1 0

1
3
− 17 1

Question 2(d) Reduce the quadratic form given below to canonical form and find its rank
and signature:
x2 + 4y 2 + 9z 2 + u2 − 12yx + 6zx − 4zy − 2xu − 6zu

Solution. Let

Q(x, y, z, u) = x2 + 4y 2 + 9z 2 + u2 − 12yx + 6zx − 4zy − 2xu − 6zu


= (x − 6y + 3z − u)2 − 32y 2 + 32yz − 12yu
3
= (x − 6y + 3z − u)2 − 32(y 2 − yz + yu)
8
1 3 2
= (x − 6y + 3z − u) − 32(y − z + u) + 8z 2 + 18u2 − 24uz
2
2 4
1 3 2 3
= (x − 6y + 3z − u) − 32(y − z + u) + 8(z − u)2
2
2 4 2

3
Put

X = x − 6y + 3z − u
1 3
Y = y− z+ u
2 4
3
Z = z− u
2
U = u
2
√ that ∗Q(x,√y, z, u)∗ is transformed∗2 to X∗2 − 32Y
so 2
+ 8Z 2 . We now put X ∗ = X, Y ∗ =
32Y, Z = 8Z, U = U to get X − Y + Z ∗2 as the canonical form of Q(x, y, z, u).
Rank of Q(x, y, z, u) = 3 = rank of the associated matrix. Signature of Q(x, y, z, u) =
number of positive squares - number of negative squares = 2 − 1 = 1.

4
UPSC Civil Services Main 2004 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

September 13, 2007

Question 1(a) Let S be the space generated by the vectors {(0, 2, 6), (3, 1, 6), (4, −2, −2)}.
What is the dimension of S? Find a basis for S.

Solution. (0, 2, 6), (3, 1, 6) are linearly independent, because α(0, 2, 6) + β(3, 1, 6) = 0 ⇒
3β = 0, 2α + β = 0 ⇒ α = β = 0. Thus dim S ≥ 2.
If possible let (4, −2, −2) = α(0, 2, 6)+β(3, 1, 6), then 4 = 3β, −2 = 2α+β, −2 = 6α+6β
should be consistent. Clearly β = 43 , α = 12 (−2 − 43 ) = − 53 from the first two equations, and
these values satisfy the third. Thus (4, −2, −2) is a linear combination of (0, 2, 6) and (3, 1, 6).
Hence dim S = 2 and {(0, 2, 6), (3, 1, 6)} is a basis of S, being a maximal linearly inde-
pendent subset of a generating system.

Question 1(b) Show that f : R3 −→ R where f (x, y, z) = 3x + y − z is a linear transfor-


mation. What is the dimension of the kernel? Find a basis for the kernel.

Solution.

f (α(x1 , y1 , z1 ) + β(x2 , y2 , z2 )) = f (αx1 + βx2 , αy1 + βy2 , αz1 + βz2 )


= 3(αx1 + βx2 ) + αy1 + βy2 − (αz1 + βz2 )
= α(3x1 + y1 − z1 ) + β(3x2 + y2 − z2 )
= αf (x1 , y1 , z1 ) + βf (x2 , y2 , z2 )

Thus f is a linear transformation.


Easy solution for this particular example. Clearly (1, 0, 0) does not belong to the
kernel, therefore the dimension of the kernel is ≤ 2. A simple look at f shows that (0, 1, 1)
and (1, −1, 2) belong to the kernel and are linearly independent, thus the dimension of the
kernel is 2 and {(0, 1, 1), (1, −1, 2)} is a basis for the kernel.

1
General solution. Clearly f : R3 −→ R is onto, thus the dimension of the range of
f is 1. From question 3(a) of 1998, dimension of nullity of f + dimension of range of f =
dimension of domain of f , so the dimension of the nullity of f = 2. Given this, we can pick
a basis for the kernel by looking at the given transformation.

Question 2(a) Show that T the linear transformation from R3 to R4 represented by the
matrix  
1 3 0
 0 1 −2
 
2 1 1
−1 1 2
is one to one. Find a basis for its image.

Solution. {e1 , e2 , e3 } be the standard basis for R3 . Then

T(e1 ) = (1, 0, 2, −1) = v1


T(e2 ) = (3, 1, 1, 1) = v2
T(e3 ) = (0, −2, 1, 2) = v3

By linearity, if T(a, b, c) = av1 + bv2 + cv3 = 0, then a + 3b = 0, b − 2c = 0, 2a + b + c =


0, −a + b + 2c = 0 ⇒ a = b = c = 0. Thus T is one-one. Also {v1 , v2 , v3 } forms a basis for
the image, since {e1 , e2 , e3 } generates R3 , and {v1 , v2 , v3 } is a linearly independent set.

Question 2(b) Verify whether the following system of equations is consistent:

x + 3z = 5
−2x + 5y − z = 0
−x + 4y + z = 4

Solution. The first equation gives x = 5 − 3z, the second now gives 5y = z + 10 − 6z = 10 −
5z ⇒ y = 2−z. Putting these values in the third equation we get 4 = −5+3z+8−4z+z = 3,
hence the given system is inconsistent.
   
1 0 3 1 0 3 5
Alternative. Let A = −2 5 −1 be the coefficient matrix and B = −2 5 −1 0
−1 4 1 −1 4 1 4
be the augmented matrix, then it can be shown that rank A = 2 and rank B = 3, which
implies that the system is inconsistent. For consistency the ranks should be equal. This
procedure will be longer in this particular case.

2
 
1 1
Question 2(c) Find the characteristic polynomial of the matrix A = . Hence find
−1 3
A−1 and A6 .

x − 1 −1
Solution. The characteristic polynomial of A is given by |xI − A| = =
1 x − 3
(x − 1)(x − 3) + 1 = x2 − 4x + 4.
2
The Cayley-Hamilton theorem states that A satisfies its characteristic equation
 i.e. A −
−3 1
4A + 4I = 0 ⇒ (A − 4I)A = A(A − 4I) = −4I. Thus A−1 = − A−4I = − 14 =
3
4 −1 −1
− 41

4
1 1
4 4
From A2 − 4A + 4I = 0 we get

A2 = 4A − 4I
A3 = 4A2 − 4A = 4(4A − 4I) − 4A = 12A − 16I
A6 = (12A − 16I)2 = 144A2 − 384A + 256I = 144(4A − 4I) − 384A + 256I
 
−128 192
= 192A − 320I =
−192 256

Question 2(d) Define a positive definite quadratic form. Reduce the quadratic form x21 +
x23 + 2x1 x2 + 2x2 x3 to canonical form. Is this quadratic form positive definite?

Solution. If Q(x1 , . . . , xn ) = ni=1 aij xi xj , aij = aji is a quadratic form in n variables with
P
j=1
aij ∈ R, thenPit is said to be positive definite if Q(α1 , . . . , αn ) > 0 whenever αi ∈ R, i =
1, . . . , n and i αi2 > 0.
Let the qiven be Q(x1 , x2 , x3 ). Then

Q(x1 , x2 , x3 ) = x21 + x23 + 2x1 x2 + 2x2 x3


= (x1 + x2 )2 + x23 + 2x2 x3 − x22
= (x1 + x2 )2 + (x2 + x3 )2 − 2x22

Let X1 = x1 + x2 , X2 = x2 , X3 = x2 + x3 i.e.
    
x1 1 −1 0 X1
x2  = 0 1 0 X2 
x3 0 −1 1 X3

then Q(x1 , x2 , x3 ) is transformed to X12 − 2X22 + X32 . Since Q(x1 , x2 , x3 ) and the transformed
quadratic form assume the same values, Q(x1 , x2 , x3 ) is an√indefinite form. The canonical
form of Q(x1 , x2 , x3 ) is Z12 − Z22 + Z32 where Z1 = X1 , Z2 = 2X2 , Z3 = X3 .

3
UPSC Civil Services Main 2005 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

September 13, 2007

Question 1(a) Find the values of k for which the vectors (1, 1, 1, 1), (1, 3, −2, k), (2, 2k −
2, −k − 2, 3k − 1) and (3, k − 2, −3, 2k + 1) are linearly independent in R4 .

Solution. The given vectors are linearly independent if the matrix


 
1 1 1 1
1
 3 −2 k  
2 2k − 2 −k − 2 3k − 1
3 k+2 −3 2k + 1

is non-singular.
Now

1 1 1 1 1 0 0 0
2 −3 k − 1
1 3 −2 k 1 2 −3 k − 1
2 2k − 2 −k − 2 3k − 1 = 2 2k − 4 −k − 4 3k − 3 = 2k − 4 −k − 4 3k − 3

k−1 −6 2k − 2
3 k + 2 −3 2k + 1 3 k − 1 −6 2k − 2

2
−3 k − 1
= 2k − 4 −k − 4 3k − 3 = (k − 5)[−9k + 9 + (k − 1)(k + 4)] 6= 0
k−5 0 0

Clearly (k − 5)[−9k + 9 + (k − 1)(k + 4)] = 0 ⇔ k = 1, 5. Thus the vectors are linearly


independent when k 6= 1, 5.

1
Question 1(b) Let V be the vector space of polynomials in x of degree ≤ n over R. Prove
that the set {1, x, x2 , . . . , xn } is a basis for V. Extend this so that it becomes a basis for the
set of all polynomials in x.

Solution. {1, x, x2 , . . . , xn } are linearly independent over R — If a0 + a1 x + . . . + an xn = 0


where ai ∈ R, 0 ≤ i ≤ n, then we must have ai = 0 for every i because the non-zero
polynomial a0 +a1 x+. . .+an xn can have at most n roots in R whereas a0 +a1 x+. . .+an xn = 0
for every x ∈ R.
Every polynomial in x of degree ≤ n is clearly a linear combination of 1, x, x2 , . . . , xn

S
with coefficients from R. Thus {1, x, x2 , . . . , xn } is a basis for V.
We shall show that = {1, x, x2 , . . . , xn , xn+1 , . . .} is a basis for the space of all polyno-
mials.
S
(i) Linear Independence: Let {xi1 , . . . , xir } be a finite subset of . Let n = max{i1 , . . . , ir },

S
then {xi1 , . . . , xir } being a subset of the linearly independent set {1, x, x2 , . . . , xn } is linearly
independent, which shows the linear independence of .

S S
(ii) Let f be any polynomial. If degree of f is m, then f is a linear combination of
{1, x, x2 , . . . , xm }, which is a subset of . Thus is a basis of W, the space of all
polynomials over R.
3
Question 2(a) Let T bea linear transformation on R whose matrix relative to the standard
3
2 1 −1
basis of R is 1 2 2 . Find the matrix of T relative to the basis

3 3 4
B
= {(1, 1, 1), (1, 1, 0), (0, 1, 1)}.

Solution.  Let the vectors


  of the given
 basis be v1 , v2 , v3 . (T(v1 ), T(v2 ), T(v3 )) =
2 1 −1 1 1 0 2 3 0
1 2 2  1 1 1 =  5 3 4.
3 3 4 1 0 1 10 6 7
If (a, b, c) = αv1 + βv2 + γv3 , then α + β = a, α + β + γ = b, α + γ = c therefore
α = a − b + c, β = b − c, γ = b − a. Consequently
T(v1 ) = 7v1 − 5v2 + 3v3
T(v1 ) = 6v1 − 3v2 + 0v3
T(v1 ) = 3v1 − 3v2 + 4v3

B
 
7 6 3
This shows that the matrix of T with respect to given basis is −5 −3 −3
3 0 4

Question 2(b) If S is a skew-Hermitian matrix, then show that A = (I + S)(I − S)−1 is a


unitary matrix. Show that a unitary matrix A can be expressed in the above form provided
−1 is not an eigenvalue of A.

Solution. See related results of question 2(a) year 1989.

2
Question 2(c) Reduce the quadratic form

6x21 + 3x22 + 3x23 − 4x1 x2 − 4x2 x3 + 4x1 x3

to a sum of squares. Also find the corresponding linear transformation, index and signature.

Solution.

Q(x1 , x2 , x3 ) = 6x21 + 3x22 + 3x23 − 4x1 x2 − 4x2 x3 + 4x1 x3


2 2 1 1 2
= 6[x21 − x1 x2 + x1 x3 + x22 + x23 − x2 x3 ]
3 3 9 9 9
7 2 7 2 8
+ x2 + x3 − x2 x3
3 3 3
1 1 2 7 4 33
= 6[x1 − x2 + x3 ] + [x2 − x3 ]2 + x23
3 3 3 7 21
Put X1 = x1 − 13 x2 + 31 x3 , X2 = x2 − 47 x3 , X3 = x3 , so that
   1
1 3 − 17
 
x1 X1
x2  = 0 1 4  X2  (1)
7
x3 0 0 1 X3

and Q(x1 , x2 , x3 ) is transformed to 6X12 + 37 X22 + 33 2


X .
21 3
Let

√1 Z1
 
 
X1 q 6
 3 
X2  =  7 Z2 
 q 
X3 7
Z
11 3

then Q(x1 , x2 , x3 ) is transformed to Z12 + Z22 + Z32 , which is its canonical form. Thus
Q(x1 , x2 , x3 ) is positive definite. The Index of Q(x1 , x2 , x3 ) = Number of positive squares in
its canonical form = 3. The signature of Q(x1 , x2 , x3 ) = Number of positive squares - the
number of negative squares in its canonical form = 3.
The required linear transformation which transforms Q(x1 , x2 , x3 ) to sums of squares is
given by (1), and the linear transformation which transforms it to its canonical form is given
by
 √1
 
   1 1 0 0  
x1 1 3 −7  6 q  Z1
3
 x2  =  0 1 4   0 7
0  Z2 
7  q 
x3 0 0 1 0 0 7 Z3
11

3
UPSC Civil Services Main 2006 - Mathematics
Linear Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

Question 1(a) Let V be a vector space of all 2 × 2 matrices over the field F. Prove that V
has dimension 4 by exhibiting a basis for V.
   
Solution. Let M1 = 10 00 , M2 = 00 10 , M3 = 01 00 , M4 = 00 01 . We will show that
{M1 , M2 , M3 , M4 } is a basis of V over F. 
{M1 , M2 , M3 , M4 } generate V. Let A = ac db ∈ V. Then A = aM1 + bM2 + cM3 +
dM4 , where a, b, c, d ∈F. Thus {M1 , M2 , M3 , M4 } is a set of generators for V over F.
{M1 , M2 , M3 , M4 } are linearly independent over F. If aM1 + bM2 + cM3 + dM4 =

a b = 0 for a, b, c, d ∈F, then clearly a = b = c = d = 0, showing that {M , M , M , M }
c d 1 2 3 4
are linearly independent over F.
Hence {M1 , M2 , M3 , M4 } is a basis of V over F and dim V = 4.

1 3

Question 1(b) State the Cayley-Hamilton theorem and using it find the inverse of 2 4 .

Solution. Let A be an n × n matrix and let In be the n × n identity matrix. Then the n-
degree polynomial |xIn −A| is called the charateristic polynomial of A. The Cayley-Hamilton
theorem states that every matrix is a root of its characteristic polynomial:

if |xIn − A| = xn + a1 xn−1 + . . . + an
then An + a1 An−1 + . . . + an In = 0

|xIn − A| = 0 is called the characteristic equation of A.


−3
Let A = 12 34 . The characteristic equation of A is 0 = x−1
−2 x−4 = (x − 1)(x − 4) − 6 =
x2 − 5x − 2.
By the Cayley-Hamilton Theorem, A2 −5A−2I2 = 0 ⇒ A(A−5I2 ) = (A−5I2 )A = 2I2 .
−2 3 
Thus A is invertible and A−1 = 21 (A − 5I2 ), so A−1 = 12 12 34 − 50 05 = 1 −21
  
2

1
B
Question 2(a) If T : R2 −→ R2 is defined by T(x, y) = (2x − 3y, x + y), compute the
matrix of T with respect to the basis = {(1, 2), (2, 3)}.

Solution. It is obvious that T : R2 −→ R2 is a linear transformation. Clearly

T(v1 ) = T(1, 2) = (−4, 3)


T(v2 ) = T(2, 3) = (−5, 5)

Let (a, b) = αv1 +βv2 , where a, b, α, β ∈ R, then α+2β = a, 2α+3β = b ⇒ α = 2b−3a, β =


2a − b. Thus = 18v1 − 11v2 , T(v2 ) = 25v1 − 15v1 , so (v1 , v2 )T = (T(v
T(v1 )   1 ), T(v2 ))=
B

18 25 18 25
(v1 , v2 ) . Thus the matrix of T with respect to the basis is
−11 −15 −11 −15
 
3 −2 0 −1
0 2 2 1
Question 2(b) Using elementary row operations, find the rank of A =  
1 −2 −3 −2
0 1 2 1

Solution. Operations R1 − 2R3 , R2 − R4 give


 
1 2 6 3
0 1 0 0
A∼ 1 −2 −3 −2

0 1 2 1

Operation R3 − R1 gives  
1 2 6 3
0 1 0 0
A∼ 
0 −4 −9 −5
0 1 2 1
Operations R3 + 4R2 , R4 − R2 ⇒
 
1 2 6 3
0 1 0 0
A∼ 
0 0 −9 −5
0 0 2 1

R4 + 92 R3 ⇒  
1 2 6 3
0 1 0 0 
A∼ 
0 0 −9 −5 
0 0 0 − 91
Clearly |A| = 1 ⇒ rank A = 4.

2
Question 2(c) Investigate for what values of λ and µ the equations

x+y+z = 6
x + 2y + 3z = 10
x + 2y + λz = µ

have (1) no solution (2) a unique solution (3) infinitely many solutions.

Solution.
 (2) The  equations will have a unique solution
for all values of µ if the coefficient
1 1 1 1 1 1 1 0
0
matrix 1 2 3  is non-singular. i.e. 1 2 3 = 1 1 2 = λ − 1 − 2 6= 0 i.e. λ 6= 3.
1 2 λ 1 2 λ 1 1 λ − 1
Thus for λ 6= 3 and for all µ we have a unique solution which can be obtained by Cramer’s
rule or otherwise.
(1) If λ = 3, µ 6= 10 then the system is inconsistent and we have no solution.
(3) If λ = 3, µ = 10, the system will have infinitely many solutions obtained by solving
x + y = 6 − z, x + 2y = 10 − 3z ⇒ x = 2 + z, y = 4 − 2z, z is any real number.

Question 2(d) Find the quadratic form q(x, y) corresponding to the symmetric matrix
 
5 −3
A=
−3 8

Is this quadratic form positive definite? Justify your answer.

Solution. The quadratic form is


  
 5 −3 x
q(x, y) = x y
−3 8 y
= 5x2 − 6xy + 8y 2
6 8
= 5[x2 − xy + y 2 ]
5 5
3 2 31 2
= 5[(x − y) + y ]
5 25
6 (0, 0), (x, y) ∈ R2 . Thus q(x, y) is positive definite. In
Clearly q(x, y) > 0 for all (x, y) =
3
fact, q(x, y) = 0 ⇒ x − 5 y = 0, y = 0 ⇒ x = y = 0.

3
UPSC Civil Services Main 1979 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

March 12, 2009

Question 1(a) Define the centre of a group and show that a group of prime power has a
non-trivial centre. Hence or otherwise prove that a group of order p2 is abelian for every
prime p.

Solution. See theorem 2.11.2 Page 86 of Algebra by Herstein. See also the corollary to the
theorem.

Question 1(b) Show that any two cyclic groups of the same order are isomorphic.

Solution. Let G1 and G2 be two cyclic groups, G1 = hai , G2 = hbi .


Case 1. G1 and G2 are of infinite order. Define T : G1 −→ G2 by T (ar ) = br .

1. T is a homomorphism. Let α, β ∈ G1 , then α = ar , β = as for some r, s ∈ Z.


T (αβ) = T (ar+s ) = br+s = br bs = T (α)T (β).

2. T is one-one. α 6= β ⇒ r 6= s ⇒ T (α) 6= T (β).

3. T is onto is obvious.

Thus G1 ' G2 .
Case 2. ord(G1 ) = ord(G2 ) = n. Again define T (ar ) = br . Let α, β ∈ G1 ⇒ α =
a , β = as ⇒ T (αβ) = T (ar+s ). If r + s ≡ t mod n, then T (αβ) = T (at ) = bt = br+s =
r

T (α)T (β), so T is a homomorphism. If α 6= β, then r 6≡ s mod n, therefore T (α) 6= T (β).


T is onto is obvious, so G1 ' G2 .

1
√ √ √ √
Question 1(c) If Q is the field of rational numbers, prove that Q( 2, 3) = Q( 2 + 3),
where Q(a1 , a2 ) is the smallest subfield of the real field containing Q, a1 , a2 .
√ √ √ √ √ √
Solution. Clearly Q( √2, 3) √ ⊇ Q( 2 + 3). √ We√ shall prove
√ that√ the degree of√Q( 2, √ 3)
over Q =√degree√of Q( 2 + 3) = 4 ⇒ Q( 2, 3) = Q( √ 2 + 3). Let x = 2 +√ 3 so
that x − 2 = 3. Squaring both sides, we get x2 − 2 2x + 2 = 3 ⇒ x2 − 1 = 2 2x ⇒
x4 − 2x2 + 1 = 8x2 ⇒ x4 − 10x2 + 1 = 0. Now x4 − 10x2 + 1 is irreducible over Q — it has no
rational roots, as ±1 are not roots of the polynomial. Another way√to see √ this is to observe
that t2 −√10t + 1 √ has no rational roots,
√ so is irreducible.
√ Thus (Q( 2 + √3) : Q) = 24.
2
Now 3 6∈ Q( √2), because if 3 √ = a√ + b 2, a,√b ∈ Q, then 3 = √a √ + 2 2ab + 2b , which
would imply that 2 ∈ Q. Thus (Q( 2, 3) : Q( 2) = 2 ⇒ (Q( 2, 3) : Q) = 4.

2
UPSC Civil Services Main 1980 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 18, 2009

Question 1(a) Let R be a relation on a non-empty set A and for every y ∈ A let

By = {x | (x, y) ∈ R}

Show that the collection {By | y ∈ A} is a partition if and only if R is an equivalence


relation.

Solution. Let R be an equivalence relation. We need to show that {By | y ∈ A} is a


partition of A. Let a, b ∈ A. It is enough to show that if x ∈ Ba ∩ Bb , then Ba = Bb . Let
y ∈ Ba , then (y, a) ∈ R. As x ∈ Ba , (x, a) ∈ R, and x ∈ Bb , (x, b) ∈ R. Now by symmetry
and transitivity of R, (x, a) ∈ R ⇒ (a, x) ∈ R, and (y, a) ∈ R, (a, x) ∈ R ⇒ (y, x) ∈ R, and
now (x, b) ∈ R ⇒ (y, b) ∈ R ⇒ y ∈ Bb so Ba ⊆ Bb . SSimilarly Bb ⊆ Ba ⇒ Ba = Bb . So
Ba ∩ Bb = ∅ or Ba = Bb for all a, b ∈ A. Since A = a∈A Ba , the set {By | y ∈ A} is a
partition of A.
The converse is false. Let A = {a, b} and let R = {(a, b), (b, a)}. Then Ba = {b}, Bb =
{a}, which is a partition of A, but R is not an equivalence relation, as it is not reflexive or
transitive.

Question 1(b) Let I be the set of integers and let R be a relation on I × I defined by
(m, n)R(p, q) if and and only if mq = np. Show that R is an equivalence relation and
identify the partition generated by R.

Solution. The question as stated is not correct. (2, 3)R(0, 0) and (0, 0)R(1, 2), but
(2, 3)R(1, 2) is false, as 2 × 2 6= 3 × 1. However if we define I ∗ = I − {0}, and R is de-
fined as a relation over I × I ∗ with the above definition, then we can show that it is an
equivalence relation.

1
Reflexive: Clearly (m, n)R(m, n), as mn = nm. Symmetric: (m, n)R(p, q) ⇒ (p, q)R(m, n)
because mq = np ⇒ qm = pn. Transitive: If (m, n)R(p, q), then mq = np. If (p, q)R(r, s)
then ps = qr. Thus mqs = nps = nqr ⇒ ms = nr because q 6= 0. Thus (m, n)R(r, s).
Clearly B(m,n) = {(a, b) | mb/n ∈ I, a = mb/n}. In fact this is isomorphic to the set of
all rational numbers.

Question 1(c) Let A and B be any two non-empty sets. Show that the collections of all
mappings from A to B is a proper subset of the collection of all relations from A to B.

Solution. Let f be a function from A to B. Then the set {(a, f (a)) | a ∈ A} is a relation
from A to B. Thus the set of all functions is a subset of the set of all relations. It is a proper
subset if B has more than one element — the R = A × B is a relation from A to B, but is
not a function.

Question 2(a) Show that the set of all transformations x −→ ax+b


cx+d
, a, b, c, d ∈ R, ad−bc = 1
is a group. Examine whether the restricted set of transformations where a, b, c, d are integers
subject to the same constraint will form a group.

Solution. Let G be the set of all transformations, and H be the set of all transformations
where a, b, c, d are integers.

• G is nonempty. I : x −→ x is in G, with a = d = 1, b = c = 0.
a x+b
a1 x+b1 a2 x+b2 a2 c 1x+d1 +b2 (a1 a2 +b2 c1 )x+a2 b1 +b2 d1
• If T1 (x) = c1 x+d1
, T2 (x) = c2 x+d2
, then (T2 ◦ T1 )(x) = 1
a x+b
1
= (c2 a1 +d2 c1 )x+c2 b1 +d1 d2
c2 c 1x+d1 +d2
1 1
is an element of G because (a1 a2 + b2 c1 )(c2 b1 + d1 d2 ) − (a2 b1 + b2 d1 )(c2 a1 + d2 c1 ) =
(a1 d1 − b1 c1 )(a2 d2 − b2 c2 ) = 1.

• T ◦ I = I ◦ T = T for every T ∈ G.

• Given T (x) = ax+b


cx+d
, T −1 (x) = −cx+a
dx−b
is the inverse of T . In fact T ◦ T −1 = I = T −1 ◦ T ,
this can be checked by substituting in the above formula and setting ad − bc = 1.

• The operation is clearly associative.


−1
Thus G is a group. H is clearlyclosed,  and T ∈ H ⇒ T ∈ H, so H is a group.
a b
In fact G is isomorphic to { | ad − bc = 1, a, b, c, d ∈ R} and H is isomorphic to
  c d
a b
{ | ad − bc = 1, a, b, c, d ∈ Z}.
c d

2
Question 2(b) Show that a group of order 15 is cyclic.

Solution. By Cauchy’s theorem, a group of order 15 has an element a such that the order of
a = o(a) = 5 and G has an element b such that o(b) = 3. Let H = hai , the group generated
by a, and K = hbi , the group generated by b. Then H ∩ K = {e}, because x ∈ H ⇒ x = e
or o(x) = 5, and x ∈ K ⇒ x = e or o(x) = 3.
Since G has a unique subgroup of order 5, and a unique subgroup of order 3, by Sylow’s
theorems, H, K are normal subgroups of G. Then ab = ba because aba−1 ∈ H, ba−1 b−1 ∈
K ⇒ aba−1 b−1 ∈ H ∩ K ⇒ ab = ba. Hence o(ab) = 15 ⇒ G is cyclic.
Here we have used the fact that if in a group G, o(x) = l, o(y) = m, (l, m) = 1, xy =
yx ⇒ o(xy) = lm.

Question 2(c) Let G be a finite group and F = {f : G −→ C} be the set of all complex
valued functions on G. If addition and multiplication in F are defined for f, g ∈ F by

(f + g)(x) = f (x) + g(x)


X
(f g)(x) = f (xy −1 )g(y)
y∈G

for every x ∈ G, show that F is a ring with identity.

Solution. F is a commutative group for addition is obvious, as the 0 function is the additive
identity, −f is the inverse of f , and + is commutative and associative in C.
Let e∗ : F −→ C be defined by e∗ (x) = 1 if x = e, 0 otherwise. Then
X
(f.e∗ )(x) = f (xy −1 )e∗ (y) = f (xe−1 ) = f (x)
y∈G

and X
(e∗ .f )(x) = e∗ (xy −1 )f (y) = f (x)
y∈G

because e∗ (xy −1 ) = 1 ⇔ xy −1 = e ⇔ x = y, and all other terms vanish.

1. Thus e∗ is the multiplicative identity of F .

2. f (g + h) = y∈G f (xy −1 )(g + h)(y) = f g + f h


P

3. (f + g)h = f h + gh

Thus F is a ring with identity element.

3
UPSC Civil Services Main 1981 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

March 12, 2009

Question 1(a) Let f : X −→ Y be a bijective mapping and let A ⊆ X, B ⊆ X, show that


f (A ∩ B) ⊆ f (A) ∩ f (B) and f (A ∪ B) = f (A) ∪ f (B). Examine whether f −1 (A ∩ B) is
properly contained in f −1 (A) ∩ f −1 (B).

Solution.

A ∩ B ⊆ A ⇒ f (A ∩ B) ⊆ f (A)
A ∩ B ⊆ B ⇒ f (A ∩ B) ⊆ f (B)
⇒ f (A ∩ B) ⊆ f (A) ∩ f (B)

A ⊆ A ∪ B ⇒ f (A) ⊆ f (A ∪ B)
B ⊆ A ∪ B ⇒ f (B) ⊆ f (A ∪ B)
⇒ f (A) ∪ f (B) ⊆ f (A ∪ B)

Conversely, let y ∈ f (A ∪ B) ⇒ ∃x ∈ A ∪ B.f (x) = y ⇒ x ∈ A or x ∈ B ⇒ f (x) ∈


f (A) or f (x) ∈ f (B) ⇒ f (x) ∈ f (A) ∪ f (B). This shows that f (A ∪ B) ⊆ f (A) ∪ f (B),
hence f (A ∪ B) = f (A) ∪ f (B).
f −1 (A∩B) is not properly contained in f −1 (A)∩f −1 (B), as f −1 (A∩B) = f −1 (A)∩f −1 (B).
x ∈ f −1 (A) ∩ f −1 (B) ⇒ f (x) ∈ A, f (x) ∈ B ⇒ f (x) ∈ A ∩ B ⇒ x ∈ f −1 (A ∩ B).
Conversely x ∈ f −1 (A ∩ B) ⇒ f (x) ∈ A ∩ B ⇒ f (x) ∈ A, f (x) ∈ B ⇒ x ∈ f −1 (A), x ∈
f (B) ⇒ x ∈ f −1 (A) ∩ f −1 (B).
−1

Thus the two sides are contained in each other, hence must be equal.

1
Question 1(b) Define a binary relation on a set A. Give examples of relations which are

1. reflexive, symmetric but not transitive.

2. reflexive, transitive but not symmetric.

3. symmetric, transitive but not reflexive.

Solution.

1. Reflexive, symmetric but not transitive. A = {a, b, c}, and R = {(a, a), (b, b), (c, c), (a, b),
(b, c), (b, a), (c, b)}.

2. reflexive, transitive but not symmetric. A = {a, b, c}, and R = {(a, a), (b, b), (c, c), (a, b),
(b, c), (a, c)}.

3. symmetric, transitive but not reflexive. A = {a}, R = ∅. Note that if (x, y) ∈ R, then
by symmetry (y, x) ∈ R, and then by transitivity (x, x) ∈ R, (y, y) ∈ R. Hence for any
element a ∈ A, (a, a) 6∈ R ⇒ (a, b) 6∈ R, (b, a) 6∈ R for all b ∈ A.

Question 2(a) Examine whether the set of rational numbers { 1+2m


1+2n
| m, n ∈ Z} forms a
group under multiplication.

Solution. Let G be the set under consideration.


Note that the set of non-zero rational numbers forms a group under multiplication. To
prove that G is its subgroup, all we need to show is that given x, y ∈ G, x−1 y ∈ G. Let y =
1+2m
1+2n
1+2q
, x = 1+2p , then x−1 y = 1+2(p+m+2mp)
1+2(q+n+2qn)
. x−1 y ∈ G because m + p + 2mp, q + n + 2qn ∈ Z.
Thus G is a group.

Question 2(b) Show that the set of matrices


   
±1 0 0 ±1
{ , }
0 ±1 ±1 0

is a group under multiplication. Examine whether it has any proper subgroup.


       
1 0 −1 0 1 0 −1 0
Solution. Let A1 = , A2 = , A3 = , A4 = , A5 =
   0 1  0 1  0 −1 0 −1
0 1 0 −1 0 1 0 −1
, A6 = , A7 = , A8 = . These are all the elements of
1 0 1 0 −1 0 −1 0
the given set G.

1. A1 is the identity element of G.

2
2. G is closed w.r.t. multiplication, as shown in the following table.

A1 A2 A3 A4 A5 A6 A7 A8
A1 A1 A2 A3 A4 A5 A6 A7 A8
A2 A2 A1 A4 A3 A6 A5 A8 A7
A3 A3 A4 A1 A2 A7 A8 A5 A6
A4 A4 A3 A2 A1 A8 A7 A6 A5
A5 A5 A7 A6 A8 A1 A3 A2 A4
A6 A6 A8 A5 A7 A2 A4 A1 A3
A7 A7 A5 A8 A6 A3 A1 A4 A2
A8 A8 A6 A7 A5 A4 A2 A3 A1

3. Every element of G is invertible.

4. Multiplication is associative, as it is associative for all matrices.

Thus G is a group. It has several proper subgroups — {A1 }, {A1 , A2 }, {A1 , A2 , A3 , A4 }, as


is clear from the above table.

3
UPSC Civil Services Main 1982 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

November 1, 2008

Question 1(a) Write if each of the following statements is true or false.


1. If ∗ is any binary operation on any set S, then a ∗ a = a for all a ∈ S.
2. If ∗ is any commutative binary operation on any set S, then a ∗ (b ∗ c) = (b ∗ c) ∗ a.
3. If ∗ is any associative binary operation on any set S, then a ∗ (b ∗ c) = (b ∗ c) ∗ a.
4. Every binary operation defined on a set having exactly one element is both commutative
and associative.
5. A binary operation on a set S assigns at least one element to each ordered pair of
elements of S.
6. A binary operation on a set S assigns at most one element to each ordered pair of
elements of S.
7. A binary operation on a set S may associate more than one element to some ordered
pair of elements of S.
8. A binary operation on a set S may assign exactly one element to each ordered pair of
elements of S.

Solution.
1. If ∗ is any binary operation on any set S, then a ∗ a = a for all a ∈ S.
False. Take S = Z, ∗ = +, 2 + 2 6= 2.
2. If ∗ is any commutative binary operation on any set S, then a ∗ (b ∗ c) = (b ∗ c) ∗ a.
True.

1
3. If ∗ is any associative binary operation on any set S, then a ∗ (b ∗ c) = (b ∗ c) ∗ a.
   
1 0 1 0
False. Take S = 2×2 matrices, ∗ = multiplication. Let a = ,b = ,c =
  0 0 1 0
1 0
. Then a ∗ (b ∗ c) = a, (b ∗ c) ∗ a = b 6= a.
0 1
4. Every binary operation defined on a set having exactly one element is both commutative
and associative.
True.
5. A binary operation on a set S assigns at least one element to each ordered pair of
elements of S.
True, since a binary operation assigns exactly one element to each ordered pair.
6. A binary operation on a set S assigns at most one element to each ordered pair of
elements of S.
True, same reason as above.
7. A binary operation on a set S may associate more than one element to some ordered
pair of elements of S.
False.
8. A binary operation on a set S may assign exactly one element to each ordered pair of
elements of S.
True.

Question 1(b) Show that N is a normal subgroup of G if and only if ∀g ∈ G. gN g −1 = N .


Does it imply that if N is a normal subgroup, then ∀n ∈ N.∀g ∈ G.g −1 ng = n? Give an
example.
Solution. For the first part see Lemma 2.6.1, Page 50 of Algebra by Herstein.
The second statement is false. Let G = S3 , the symmetric group on 3 symbols. Let
N = {Identity permutation, (123), (132)}, then [G : N ] = Index of N in G = 2, so N is
normal in G.
Clearly (123)(12) = (13) and (12)(123) = (23). Thus (12)−1 (123)(12) = (12)(123)(12) =
(132) 6= (123).
Question 1(c) Let φ be a homomorphism of a group G into G with kernel K. Show that
Gφ = φ(G) is a group and there is an isomorphism of Gφ with G/K.
Solution. φ(e) = e ⇒ φ(G) 6= ∅. Let y1 , y2 ∈ φ(G). Then there exist x1 , x2 ∈ G, φ(x1 ) =
y1 , φ(x2 ) = y2 . Thus φ(x−1 −1 −1
1 x2 ) = φ(x1 ) φ(x2 ) = y1 y2 ∈ φ(G). Hence φ(G) is a subgroup
of G, so is a group.
See Question 1(a) year 1985 for the second part.

2
Question 1(d) Let I be an ideal in a ring R and let the coset of the element x ∈ R be
defined by x + I = {x + i | i ∈ I}. Then the distinct cosets form a partition of R, and if
addition and multiplication are defined by (x+I)+(y +I) = x+y +I, (x+I)(y +I) = xy +I,
then the cosets constitute a ring R/I in which the 0 element is 0 + I and the inverse of x + I
is (−x) + I.

Solution. We first show that x + I = y + I if x − y ∈ I and x + I ∩ y + I = ∅ if x − y 6∈ I.


Let j = x − y ∈ I. Let t ∈ x + I. Then t = x + i, i ∈ I. t = x + i = y + i − j ∈ y + I
because i − j ∈ I, so x + I ⊆ y + I. Since x − y ∈ I ⇒ y − x ∈ I ⇒ y + I ⊆ x + I, hence
x + I = y + I.
Conversely if t ∈ (x + I) ∩ (y + I), then t = x + i = y + j ⇒ x − y = j − i ∈ I. Hence
if x − y 6∈ I, then x + I ∩ y + I = ∅. Note that z ∈ z + I as 0 ∈ I. So the cosets form a
partition of R.

1. 0 + I = I ∈ R/I ⇒ R/I 6= ∅. R/I is clearly closed for addition. Note that addition is
well-defined i.e. if x + I = x1 + I, y + I = y1 + I ⇒ x + y + I = x1 + y1 + I.

2. (x + I) + (0 + I) = (x + 0) + I = x + I = (0 + I) + (x + I) for every x ∈ R, so 0 + I is
the additive identity of R/I.

3. (x + I) + (−x + I) = (x + (−x)) + I = 0 + I = (−x + I) + (x + I) so every element in


R/I has an additive inverse.

4. Addition is associative and commutative, follows from the fact that this is so in R.

5. R/I is closed for multiplication. Note that multiplication is well defined — if x + I =


x1 + I, y + I = y1 + I ⇒ xy + I = x1 y1 + I, because x = x1 + i, y = y1 + j, i, j ∈ I, so
xy = (x1 + i)(y1 + j) = x1 y1 + x1 j + y1 i + ij ⇒ xy − x1 y1 ∈ I because x1 j + y1 i + ij ∈ I.

6. (x + I)[(y + I) + (z + I)] = (x + I)(y + I) + (x + I)(z + I) as x(y + z) = xy + xz.

Thus R/I is a ring as desired.

3
UPSC Civil Services Main 1983 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 31, 2008

Question 1(a) Show that the set I × I of integers is a commutative ring with unity with
respect to addition and multiplication defined as under:

(a, b) + (c, d) = (a + c, b + d)
(a, b)(c, d) = (ac, bd)

Solution. Let (a, b), (c, d), (e, f ) be any elements of I × I.

1. Clearly I × I is closed with respect to the addition operation.

2. (0, 0) + (a, b) = (a, b) + (0, 0) = (a, b). Thus (0, 0) is the additive identity of I × I.

3. (a, b)+(−a, −b) = (0, 0) = (−a, −b)+(a, b), thus every element has an additive inverse.

4. (a, b) + (c, d) = (a + c, b + d) = (c, d) + (a, b).

5. [(a, b) + (c, d)] + (e, f ) = (a + c + e, b + d + f ) = (a, b) + [(c, d) + (e, f )]. Thus I × I is


an additive abelian group.

6. Clearly I × I is closed with respect to the multiplication operation.

7. [(a, b) + (c, d)](e, f ) = (ae + ce, bf + cf ) = (a, b)(c, d) + (a, b)(e, f ).

8. (a, b)(c, d) = (ac, bd) = (ca, db) = (c, d)(a, b).

9. (1, 1)(a, b) = (a, b) = (a, b)(1, 1).

Thus I × I is a commutative ring with unity.

1
Question 1(b) Prove that the relation of isomorphism on the collection of groups is an
equivalence relation.

Solution. See 1985 Question 1(c)(4).

Question 1(c) Prove that a polynomial domain K[x] over a field K is a principal ideal
domain.

Solution. Let A be an ideal of K[x], A 6= h0i , A 6= K[x]. Let f (x) be a polynomial of the
smallest degree in A − {0}. Then f (x) is a generator of A, proved as follows: if g(x) 6= 0 be
any element of A. If f (x) does not divide g(x), then since K[x] is a Euclidean domain (proof
below), there exist polynomials q(x) and r(x) in K[x] such that g(x) = f (x)q(x) + r(x),
where deg(r(x)) < deg(f (x)). Since g(x), f (x) ∈ A and A is an ideal, it follows that
r(x) = g(x) − f (x)q(x) ∈ A, which contradicts the fact that f (x) has smallest degree in A.
Thus any element of A is divisible by f (x) ⇒ A = hf (x)i , the ideal generated by f (x) (note
that 0 = f (x)0).
Proof of the fact that K[x] is a Euclidean domain. The Euclidean function d : K[x] −
{0} −→ Z is the degree of a polynomial.

1. d(f (x)) ≥ 0 for every f (x) ∈ K[x] − {0}.

2. d(f (x)) ≤ d(f (x)g(x)) for any f (x) 6= 0, g(x) 6= 0.

3. Given g(x) and f (x) 6= 0, we show that there exist q(x), r(x) such that g(x) =
q(x)f (x) + r(x) where r(x) = 0 or d(r(x)) < d(f (x)). The proof is by induction
on the degree n of g(x). If n < d(f (x)) = m, then q(x) = 0, r(x) Pn = g(x). Suppose
i
the result
P holds ifor all polynomials of degree < n. Let g(x) = i=0 bi x , bn 6= 0 and
f (x) = m a
i=0 i x . Let h(x) = g(x) − (bn /am )xn−m
f (x). Then degree of h(x) < n, and
therefore there exist q1 (x) and r(x) such that h(x) = q1 (x)f (x) + r(x) where r(x) = 0
or d(r(x)) < m. Hence g(x) = (q1 (x) + (bn /am )xn−m )f (x) + r(x).

Thus K[x] is a Euclidean domain.

2
UPSC Civil Services Main 1984 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 31, 2008

Question 1(a) Prove that a non-void subset S of a ring R is a subring of R if an only if


a − b ∈ S and ab ∈ S for all a, b ∈ S.

Solution. Let a, b ∈ S.

• a − a = 0 ∈ S, 0 − b ∈ S, and therefore a − (−b) = a + b ∈ S, so S is closed w.r.t.


addition.

• a + b = b + a as this is true in R also.

• 0 ∈ S.

• We have seen a ∈ S ⇒ −a ∈ S, and a + (−a) = 0 = (−a) + a.

• + is associative, as it is so in R. Thus S is an additive subgroup.

• ab ∈ S, given.

• Multiplication distributes over addition in S, as it does so in R.

Thus if the given requirements are satisfied, S is a subring of R. Conversely, if S is a


subring of R, then a, b ∈ S ⇒ ab ∈ S, −b ∈ S ⇒ a + (−b) ∈ S ⇒ a − b ∈ S.

Question 1(b) Prove that an integral domain can be imbedded in a field.

Solution. See Theorem 3.6.1 Page 140 of Algebra by Herstein.

1
Question 1(c) Prove that for any two ideals A and B of a ring R, the product AB is an
ideal of R.
P
Solution. Note first that AB = { i∈F ai bi | ai ∈ A, bP
i ∈ B, F finite}.
AB isPa subring of R. If x ∈ AB, y ∈ AB, then x = i∈F ai bi | ai ∈ A, bi ∈ B, F finite
and y = i∈G ci di | ci ∈ A, di ∈ B, G finite. Clearly
X X
x−y = ai bi + ci di ∈ AB
i∈F i∈G
X  X  X
xy = ai bi ci di = xi yi where xi ∈ A, yi ∈ B
i∈F i∈G i∈H

Note that ai bi cj dj = ai (bi cj )dj , and ai (bi cj ) ∈ A as A is an ideal, cj ∈ A, bi ∈ R ⇒ bi cj ∈ A.


Thus xy can be expressedP as above.
For any r ∈ R, xr = i∈F ai (bi r) = i∈F ai b0i , b0i ∈ B as B is an ideal. So xr ∈ AB.
P
Similarly rx ∈ AB. Thus AB is an ideal.

2
UPSC Civil Services Main 1985 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 31, 2008

Question 1(a) State and prove the fundamental theorem of homomorphisms of groups.

Solution. Theorem. Let f be a homomorphism from a group G onto a group G0 with


kernel N . Then N is a normal subgroup of G, and G/N is isomorphic to G0 .
Proof: N = {x | x ∈ G, f (x) = e0 }, where e0 is the identity of G0 . Clearly N 6= ∅, as
e ∈ N . If x, y ∈ N , then f (x) = e0 , f (y) = e0 , so f (xy −1 ) = f (x)f (y)−1 = e0 ⇒ xy −1 ∈ N ⇒
N is a subgroup of G.
Now let g ∈ N , then for any x ∈ G, f (xgx−1 ) = f (x)e0 f (x)−1 = e0 ⇒ xgx−1 ∈ N , so N
is a normal subgroup of G.
Let φ : G/N −→ G0 defined by φ(gN ) = f (g).

• φ is well defined: We need to show that φ does not depend on the choice of coset
representative i.e. if g1 N = g2 N then φ(g1 N ) = φ(g2 N ). Now g1 N = g2 N ⇒ g2−1 g1 ∈
N ⇒ f (g2−1 g1 ) = e0 ⇒ f (g1 ) = f (g2 ) ⇒ φ(g1 N ) = φ(g2 N ).

• φ is a homomorphism: φ(g1 N )φ(g2 N ) = f (g1 )f (g2 ) = f (g1 g2 ) = φ(g1 g2 N ).

• φ is onto: Let y ∈ G0 , then f being onto, there exists x ∈ G such that f (x) = y.
Clearly φ(xN ) = f (x) = y.

• φ is 1-1. If g1 N 6= g2 N , then g2−1 g1 6= N , so f (g2−1 g1 ) 6= e0 ⇔ f (g1 ) 6= f (g2 ) ⇔


φ(g1 N ) 6= φ(g2 N ).

Thus φ is an isomorphism from G/N onto G0 , so G/N ' G0 .


Note 1: If η : G −→ G/N is the natural homomorphism i.e. η(g) = gN , then f = φ ◦ η.
Note 2: If f is not assumed to be onto, we can only say that G/N ' f (G).

1
Question 1(b) Prove that the order of each subgroup of a finite group divides the order of
the group.

Solution. See Theorem 2.4.1 Page 41 of Algebra by Herstein.

Question 1(c) Is each of the following statements true or false?

1. If a is an element of a ring (R, +, .) and m, n ∈ Z, then (am )n = amn .

2. Every subgroup of an abelian group is not necessarily abelian.

3. A semigroup (G, .) is which the equations ya = b, ax = b are solvable for any a, b, is a


group.

4. The relation of isomorphism in the class of all groups is not an equivalence relation.

5. There are only two abstract groups of order 6.

Solution.

1. If a is an element of a ring (R, +, .) and m, n ∈ Z, then (am )n = amn .


m
True. (am )n = a . . .am} = a.a.a.a.
| . .{z |
mn
{z . . . .a} = a .
n times mn times
2. Every subgroup of an abelian group is not necessarily abelian.
False. If G is abelian, and H is a subgroup of G, then for any a, b ∈ H, a, b ∈ G ⇒
ab = ba ⇒ H is abelian.

3. A semigroup (G, .) is which the equations ya = b, ax = b are solvable for any a, b, is a


group.
True. Let a ∈ G, then ax = a is solvable ⇒ there exists an element e ∈ G such that
ae = a. Now let b ∈ G, then there exists y ∈ G, ya = b. Then be = yae = ya = b.
Thus e is the right identity for G. For any a ∈ G, the equation ax = e is solvable, thus
a has a right inverse.
Since G has a right identity and a every element of G has a right inverse, G is a group.
Note that since G is a semigroup, it is already closed the semigroup operation, and the
operation is associative.

4. The relation of isomorphism in the class of all groups is not an equivalence relation.
False.

• The relation ' is reflexive, as the identity map is an isomorphism from any group
to itself.

2
• If σ : G −→ G0 is an isomorphism, then σ −1 : G0 −→ G is an isomorphism, so '
is symmetric.
• If G1 ' G2 and σ1 : G1 −→ G2 , and G2 ' G3 , σ2 : G2 −→ G3 , then σ2 ◦ σ1 :
G1 −→ G3 is also an isomorphism, so G1 ' G3 , thus ' is transistive.

5. There are only two abstract groups of order 6.


True. The two groups are the cyclic group of order 6 and S3 , the symmetric group on
3 symbols.
If G is abelian, then G has x, an element of order 2 and y, an element of order 3. Since
xy = yx, o(xy) = 6 so G is cyclic.
If G is non-abelian, let a, b ∈ G, where o(a) = 2, o(b) = 3 (such elements exist because
of Cauchy’s theorem), then G = {e, a, b, b2 , ab, ba} ' S3 .

3
UPSC Civil Services Main 1986 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 27, 2008

Question 1(a) Prove that a map f : X −→ Y is injective if and only if f can be left
cancelled in the sense that f ◦ g = f ◦ h ⇒ g = h. Prove that f is surjective if and only if it
can be right cancelled in the sense that g ◦ f = h ◦ f ⇒ g = h.
Solution. Suppose that f is injective (1-1). Let f ◦ g = f ◦ h, where g, h are mappings
from Z −→ X, where Z can be arbitrary. We have to prove g(z) = h(z) ∀z ∈ Z. If
g(z) 6= h(z) then f (g(z)) 6= f (h(z)), since f is injective, consequently f ◦ g 6= f ◦ h. Thus
g(z) = h(z)∀z ∈ Z, so f ◦ g = f ◦ h ⇒ g = h.
Conversely let f ◦ g = f ◦ h ⇒ g = h. We have to prove that f is 1-1. Suppose it is
not 1-1. Then there exist x1 , x2 ∈ X, x1 6= x2 , f (x1 ) = f (x2 ). Define g, h : X −→ X by
g(x) = x1 , h(x) = x2 for all x ∈ X. Then ∀x ∈ X f (g(x)) = f (x1 ) = f (x2 ) = f (h(x)), so
f ◦ g = f ◦ h, but g 6= h, which is a contradiction. Thus f must be 1-1.
Suppose that f is onto. Given (g ◦ f )(x) = (h ◦ f )(x), we need to show that g(y) = h(y)
for all y ∈ Y . For any y ∈ Y , there exists x ∈ X such that f (x) = y, since f is onto. Thus
g(y) = (g ◦ f )(x) = (h ◦ f )(x) = h(y), showing that g ◦ f = h ◦ f ⇒ g = h.
Conversely let g ◦ f = h ◦ f ⇒ g = h. We have to prove that f is onto. Suppose it is
not onto. Then there exists y0 ∈ Y such that there is no x ∈ X such that f (x) = y0 . Let
g, h : Y −→ Z (for any Z) be defined by g(y) = h(y) if y 6= y0 , and g(y0 ) 6= h(y0 ). Clearly
(g ◦ f )(x) = (g ◦ f )(x) for every x ∈ X, but g 6= h, which is a contradiction. Thus f is onto.

Question 1(b) The product HK of two subgroups H, K of a group G is a subgroup of G if


and only if HK = KH.
Solution. See Lemma 2.5.1 Page 44 of Algebra by Herstein.

Question 1(c) Prove that a finite integral domain is a field.


Solution. See lemma 3.2.2 Page 127 of Algebra by Herstein.

1
UPSC Civil Services Main 1987 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 27, 2008

Question 1(a) Let f : X −→ Y, g : Y −→ Z be bijections, prove that g ◦ f is a bijection


and (g ◦ f )−1 = f −1 ◦ g −1 .

Solution. g ◦ f : X −→ Z. Let x1 6= x2 then then f (x1 ) 6= f (x2 ) because f is 1-1. Thus


g(f (x1 )) 6= g(f (x2 )) because g is 1-1. Thus g ◦ f is 1-1.
For any z ∈ Z, ∃y ∈ Y such that g(y) = z, because g is onto. Now y ∈ Y , so there exists
x ∈ X such that f (x) = y, because f is onto. Thus for any z ∈ Z, we have determined
x ∈ X such that (g ◦ f )(x) = g(f (x)) = g(y) = z, thus g ◦ f is onto. Thus g ◦ f is invertible.

(g ◦ f ) ◦ (f −1 ◦ g −1 )(z) = (g ◦ f )f −1 (g −1 (z))
= (g ◦ f )(f −1 (g −1 (z)))
= g(f (f −1 (g(z))))
= g(g −1 (z)) = z ∀z ∈ Z

Thus (g ◦ f ) ◦ (f −1 ◦ g −1 ) = IZ .

(f −1 ◦ g −1 ) ◦ (g ◦ f )(x) = (f −1 ◦ g −1 )(g(f (x))


= f −1 (g −1 (g(f (x))))
= f −1 (f (x)) = x ∀x ∈ X

Thus (f −1 ◦ g −1 ) ◦ (g ◦ f ) = IX , so f −1 ◦ g −1 = g ◦ f .

Question 1(b) Prove that H ∗ K is a subgroup of (G, ∗) if and only if H ∗ K = K ∗ H.

Solution. See Lemma 2.5.1, Page 44 of Algebra by Herstein.

1
Question 1(c) If G is a finite group of order g and H is a subgroup of G of order h , then
prove that h is a factor of g.

Solution. See Theorem 2.4.1, Page 41 of Algebra by Herstein.

2
UPSC Civil Services Main 1988 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 26, 2008

Question 1(a) If H and K are normal subgroups of a group G such that H ∩ K = {e},
show that hk = kh for h ∈ H, k ∈ K.

Solution. Consider hkh−1 k −1 . Since K is a normal subgroup of G, hkh−1 ∈ K ⇒


hkh−1 k −1 ∈ K. Similarly, H is normal in G, so khk −1 ∈ H ⇒ hkh−1 k −1 ∈ H. Thus
hkh−1 k −1 ∈ H ∩ K ⇒ hkh−1 k −1 = e ⇒ hk = kh as required.

Question 1(b) Show that the set of even permutations on n symbols, n > 1, is a normal
subgroup of the symmetric group Sn and has order n!/2.

Solution. Let An be the set of all even permutations, then An 6= ∅ because the identity
permutation is in An . If α, β ∈ An , then αβ ∈ An because α even, β even ⇒ αβ is an even
permutation. α ∈ An ⇒ α−1 ∈ An , because α even ⇒ α−1 even. Thus An is a subgroup of
Sn .
An is a normal subgroup. Let β ∈ An , then αβα−1 is even for all α ∈ Sn , because if α is
odd, then α−1 is odd, so αβα−1 is even, and if α is even, then α−1 is even, so αβα−1 is even.
Thus αβα−1 ∈ An for all α ∈ Sn , β ∈ An , so An is normal in Sn .
Since n > 1, Sn has an odd permutation, for example (1, 2). Then Sn = An ∪ An (1, 2) is
the coset decomposition of Sn modulo An : Let α ∈ Sn . If α is even, then α ∈ An . If α is
odd, then β = α(1, 2) is even, so β ∈ An . Now α = β(1, 2) ∈ An (1, 2), so Sn = An ∪ An (1, 2).
It is obvious the An ∩ An (1, 2) = ∅, so the index of An in Sn is 2, i.e. the order of An = n!/2
as required.

1
Question 1(c) Prove that the set of inner automorphisms of a group G is a normal subgroup
of the group of automorphisms of G.

Solution. Let A be the group of automorphisms of G and let I be the set of all inner
automorphisms (which are automorphisms of the form αa (x) = axa−1 for a ∈ G).

1. I 6= ∅ as the identity automorphism is in I, because if α is the identity automorphism,


then α(x) = x = exe−1 = αe (x), so α ∈ I.

2. If αa , αb ∈ I, then αa ◦ αb (x) = αa (αb (x)) = a(bxb−1 )b−1 = αab (x) ⇒ αa ◦ αb = αab ∈ I.

3. αa ∈ I ⇒ αa−1 ∈ I and αa αa−1 = αaa−1 = αe . Thus αa−1 is the inverse of αa in I.

Thus I is a subgroup of G.
Let σ ∈ A be arbitrary. Then for any αa ∈ I,

σαa σ −1 (x) = σ(aσ −1 (x)a−1 )


= σ(a)σ(σ −1 (x))σ(a−1 )
= σ(a)xσ(a)−1 = ασ(a) (x)

Thus σαa σ −1 ∈ I. Hence I is a normal subgroup of A.

Question 2(a) Show that the numbers 0, 2, 4, 6, 8 with addition and multiplication modulo
10 form a field isomorphic to J5 the ring of integers modulo 5. Give the isomorphism explic-
itly.

Solution. Let F = {0, 2, 4, 6, 8}.

1. F is closed w.r.t. addition modulo 10. The addition table is


+ 0 2 4 6 8
0 0 2 4 6 8
2 2 4 6 8 0
4 4 6 8 0 2
6 6 8 0 2 4
8 8 0 2 4 6

2. Addition is commutative and associative, as it is so in J10 and F ⊂ J10 .

3. Clearly 2, 8 and 4, 6 are inverses of each other, so F is an abelian group under addition.

2
4. F is closed under multiplication. The multiplication table is

× 0 2 4 6 8
0 0 0 0 0 0
2 0 4 8 2 6
4 0 8 6 4 2
6 0 2 4 6 8
8 0 6 2 8 4

5. 6 is the multiplicative identity, and multiplication is commutative.

6. Every non-zero element has an inverse 2 × 8 = 4 × 4 = 6 × 6 = 6.

Thus F ∗ = F − {0} is a multiplicative group. Thus F is a field.


Isomorphism. For α ∈ Z, let [α] be the residue class of α modulo 5. Let σ : F −→ J5
be defined by σ(0) = [0], σ(6) = [1], σ(2) = σ(6 + 6) = σ(6) + σ(6) = [1] + [1] = [2], σ(4) =
σ(6 + 6 + 6 + 6) = 4σ(6) = [4], σ(8) = σ(6 + 6 + 6) = 3σ(6) = [3]. It can be easily checked
that σ is an isomorphism.

Question 2(b) If R is a commutative ring with identity and U is an ideal of R. Show that
U is maximal if and only if R/U is a field.

Solution. Let U be a maximal ideal. We already know that R/U is a commutative ring
with identity given by 1 + U . R/U will be a field if we show that every nonzero element
a + U ∈ R/U is invertible. Now a + U 6= U ⇒ a 6∈ U . Thus the ideal generated by a
and U contains U properly ⇒ ha, U i = R as U is maximal. Thus 1 = ab + u for some
b ∈ R, u ∈ U ⇒ (a + U )(b + U ) = 1 + U ⇒ a + U is invertible. Thus R/U is a field.
Conversely, let R/U be a field, and M ⊇ U be an ideal. We shall show that if M 6= U ,
then M = R ⇒ U is maximal. Now if M ) U ⇒ ∃a ∈ M, a 6= U . Thus a + U is a
non-zero element of R/U and therefore invertible i.e. there exists b + U ∈ R/U such that
(a + U )(b + U ) = 1 + U , or ab − 1 ∈ U ⇒ ab − 1 ∈ M . But ab ∈ M as a ∈ M , therefore
1 ∈ M ⇒ M = R ⇒ U is maximal.
See theorem 3.5.1 Page 138 of Algebra by Herstein for a different proof.

Question 2(c) Determine all the ideals U in J12 , the ring of integers modulo 12. In each
case describe J12 /U by finding a familiar ring with which the quotient ring is isomorphic.
Which of these ideals is maximal?

Solution. Let h[α]i denote the ideal generated by [α] ∈ J12 = {[0], [1], . . . [11]}. Since J12

3
is a principal ideal domain, we have to determine the distinct principal ideals. These are

h[0]i = {[0]}
h[1]i = J12 = h[11]i = h[5]i = h[7]i
h[2]i = {[0], [2], [4], [6], [8], [10]} = h[10]i
h[3]i = {[0], [3], [6], [9]} = h[9]i
h[4]i = {[0], [4], [8]} = h[8]i
h[6]i = {[0], [6]}

If U = h[2]i , then R/U = {[0]+U, [1]+U } ' Z/2Z, the isomorphism being σ([0]+U ) = 0
mod 2, σ([1] + U ) = 1 mod 2.
If U = h[3]i , then R/U = {[0] + U, [1] + U, [2] + U } ' Z/3Z, the isomorphism being
σ([0] + U ) = 0 mod 3, σ([1] + U ) = 1 mod 3, σ([2] + U ) = 2 mod 3.
If U = h[4]i , then R/U = {[0] + U, [1] + U, [2] + U, [3] + U } ' Z/4Z, the isomorphism
being σ([0] + U ) = 0 mod 4, σ([1] + U ) = 1 mod 4, σ([2] + U ) = 2 mod 4, σ([3] + U ) = 3
mod 4.
If U = h[6]i , then R/U = {[α] + U | 0 ≤ α ≤ 5} ' Z/6Z, the isomorphism being
σ([α] + U ) = α mod 6.
Clearly ideals generated by [2], [3] are maximal as R/h[2]i ' Z/2Z, and R/h[3]i ' Z/3Z,
which are fields.
Thus J12 has 6 ideals, namely h[0]i , h[1]i , h[2]i , h[3]i , h[4]i , h[4]i of which h[2]i , h[3]i are
maximal.

4
UPSC Civil Services Main 1989 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

September 21, 2008

Question 1(a) Let G be a finite group of order 2p, p a prime. Show that G has a normal
subgroup of order p.

Solution. Assume that G has a subgroup H of p elements. We shall show that H is normal
in G. Clearly [G : H] i.e. the index of H in G is 2. Let G = H ∪ Hx, where H and Hx
are distinct right cosets i.e. x 6∈ H. Consider xH, xH 6= H because x 6∈ H ⇒ xH ∩ H =
∅ ⇒ xH ⊆ Hx. Similarly, Hx ⊆ xH. Thus if x 6∈ H, then Hx = xH. If x ∈ H, then
xH = H = Hx. Thus xHx−1 = H for every x ∈ G, so H is normal in G.
Existence of H: State Cauchy’s theorem, or better yet, prove it (See theorem 2.11.3 Page
87 of Algebra by Herstein). Let a be an element of G of order p, then H, the subgroup
generated by a is of order p.

Question 1(b) Give an example of an infinite group in which every element is of finite
order.

Solution. Let Ωn = group of n-th roots of unity.


Let G = ∪∞ n
n=1 Ωn = {α | α ∈ C, α = 1 for some n}. G is a subgroup of C − {0}. If
α ∈ G, β ∈ G, then αm = 1, β n = 1 for some m, n ⇒ (αβ)mn = 1 ⇒ αβ ∈ G. α ∈ G ⇒
α−1 ∈ G ∵ αn = 1 ⇒ α−n = 1. Clearly every element of G is of finite order. If G were finite,
2πi
say order M , then αM = 1 for every α ∈ G. But β = e M +1 ∈ G, β M 6= 1. Thus G is not
finite.
Another example: Consider the set of all infinite sequences of bits, under the operation
bitwise exclusive or: 0 ⊕ 0 = 1 ⊕ 1 = 0, 0 ⊕ 1 = 1 ⊕ 0 = 1. The identity element is the all 0
sequence, every element is its own inverse, and the operation is associative and commutative.
The group is clearly infinite, but every element has order 2.

1
Question 1(c) Let G be a group and let H be the smallest group containing elements of the
form x−1 y −1 xy, x, y ∈ G. Show that H is normal in G and the factor group G/H is abelian.

Solution. Let x ∈ G, h ∈ H, then x−1 hx = x−1 hxh−1 h. But x−1 hxh−1 ∈ H by definition,
therefore x−1 hx = x−1 hxh−1 h ∈ H ⇒ x−1 Hx = H for every x ∈ G. Thus H is normal in G.
Now in the factor group G/H, xH.yH = xyH. Since x−1 y −1 xyH = H as x−1 y −1 xy ∈ H,
it follows that xyH = yxH = yH.xH, thus G/H is abelian.

Question 2(a) If each element of a ring is idempotent, show that the ring is commutative.

Solution. See question 2(a), 1997.

Question 2(b) If a finite field F has q elements, then show that q = pn , where p is the
characteristic of F .

Solution. Let e be the multiplicative identity of F . Consider the map φ : Z −→ F defined


by φ(n) = ne. Then φ is a homomorphisms of rings as φ(m + n) = (m + n)e = me + ne =
φ(m) + φ(n) and φ(mn) = mne = mne2 = me.ne = φ(m)φ(n). Now ker φ = {n | φ(n) =
ne = 0 ⇔ p | n} = hpi , the ideal generated by p. Thus the field Z/pZ is isomorphic to a
subfield of F . In other words, F contains a subfield say Λ containing p elements. Now F is
finite, therefore F as a vector space over Λ is of finite dimension. Let (F : Λ) = n, and let
{v1 , . . . , vn } be a basis of F over Λ. Then F = {a1 v1 + . . . + an vn | a1 , . . . an ∈ Λ}. Since
each ai has p values, F has pn elements. Actually, F is isomorphic to Λn as a vector space.

Question 2(c) Let A be a ring and I be a two-sided ideal generated by the subset of all
elements of the form ab − ba, a, b ∈ A. Prove that the residue class ring A/I is commutative.

Solution.

A/I is commutative ⇔ (a + I)(b + I) = (b + I)(a + I)∀a, b ∈ A


⇔ ab + I = ba + I
⇔ ab − ba ∈ I which is true.

Hence A/I is commutative.

2
UPSC Civil Services Main 1990 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 28, 2008

Question 1(a) Let G be a group having no proper subgroup. Show that G should be a finite
group of order which is a prime, or unity.

Solution. See question 1(a), 1991. Once we have proved that G is finite, then we observe
that G has exactly one element if and only if the order of G is 1. If the order of G > 1, then
we show that it is a prime number.

Question 1(b) If the order of a group is 20, show that its 5-Sylow subgroup is a normal
subgroup. Also prove that a group of order 16 has a proper normal subgroup.

Solution. We know from various Sylow theorems that the numbr of 5-Sylow subgroups
≡ 1 mod 5 and is a divisor of 20 and therefore 4. Thus G, a group of order 20, has exactly
one Sylow subgroup of order 5, say H. Now aHa−1 for any a ∈ G is also a subgroup of order
5, therefore by uniqueness, aHa−1 = H. Thus H is normal in G.
For the second part, we prove a general theorem of which this is a special case.
Theorem. Let G be a group of order pr , p a prime, then G has a normal subgroup of
order ps for every s, 0 ≤ s < r.
Proof: By induction on r. If r = 1, then G is cyclic of prime order, hence the result
is true. Assume true for groups of order pm , m < r. Since G is a group of order pr , the
power of a prime, its center is non-trivial. Since the order of the center is pn , n ≥ 1, the
center has an element, say a, of order p (Cauchy’s theorem, Theorem 2.11.3 of Algebra
by Herstein). Let H = hai be the group generated by a. Since a ∈ center of G, H is a
normal subgroup of G. Now G/H is a group of order pr−1 . Using the induction hypothesis,
we see that G/H has a normal subgroup N ∗ of order ps−1 , 0 ≤ s − 1 < r − 1. Let η :
G −→ G/H be the natural homomorphism. Set N = η −1 (N ∗ ), we show that N is a
normal subgroup of G of order ps . η −1 (N ∗ ) 6= ∅. If x, y ∈ N , then η(x), η(y) ∈ N ∗ ,

1
then η(x)(η(y))−1 ∈ N ∗ ⇒ η(xy −1 ) ∈ N ∗ ⇒ xy −1 ∈ N , so N is a subgroup of G. For
x ∈ N, a ∈ G, η(x) ∈ N ∗ ⇒ η(a)η(x)η(a)−1 = η(axa−1 ) ∈ N ∗ as N ∗ is a normal subgroup
of G/H. Thus axa−1 ∈ N , so N is a normal subgroup of G. N ⊇ H is immediate as
∀h ∈ H.η(h) = H, the identity element of G/H. Consider η : N −→ N ∗ , then η is a
homomorphism with kernel H ⇒ N/H ' N ∗ ⇒ o(N ) = o(N ∗ )o(H) = ps .
Now for a group of order 16, p = 2, r = 4, and the above theorem shows that it has
normal groups of order 2, 4, and 8.

Question 1(c) If C is the center of a group G, and G/C is cyclic, prove that G is abelian.

Solution. See question 1(c), 1991.

Question 2(a) Show that the set of Gaussian integers is a Euclidean ring. Find an HCF
of 5i and 3 + i.

Solution. Z[i] = {a + bi | a, b ∈ Z} is an integral domain as it is a subring of the field of


complex numbers.
An integral domain R is said to be a Euclidean domain if there exists a function N :
R → Z (the ring of integers) such that
1. N (a) ≥ 0

2. N (ab) ≥ N (a) where a, b 6= 0

3. Given a, b ∈ R, b 6= 0, there exist q, r ∈ R such that a = bq + r where r = 0 or


N (r) < N (b).
For Z[i], let N (α) = N (a + ib) = a2 + b2 . Clearly
1. N (α) ≥ 0 for every α ∈ Z[i].

2. N (αβ) ≥ N (α) for all α, β ∈ Z[i] because N (αβ) = N (α)N (β) and N (β) ≥ 1 if β 6= 0.
a+ib
3. Let α = a + ib, β = m + ni, β 6= 0. Then f racαβ = m+ni = x + iy, x ∈ Q, y ∈ Q.
Determine p, q ∈ Z such that |x−p| ≤ 2 , |y−q| ≤ 2 (take p = [x] if x = [x]+θ, 0 ≤ θ ≤ 21
1 1

and p = [x] + 1 if x = [x] + θ, 21 < θ < 1).


α
Now β
− (p + qi) = x − p + i(y − q). Thus N ( αβ − (p + qi)) = (x − p)2 + (y − q)2 < 1.
Now α = (p + qi)(m + ni) + γ where γ = (x − p + i(y − q))(m + ni). Clearly γ ∈ Z[i]
and N (γ) = N (β)((x − p)2 + (y − q)2 ) < N (β), which is what we wanted to prove.
Thus Z[i] is a Euclidean ring.
Now 5i = (3 + i)(2i) + (2 − i), and 3 + i = (2 − i)(1 + i) ⇒ (5i, 3 + i) = 2 − i.
Note: In this case writing the division algorithm was easy, otherwise N (5) = 25, N (3 +
i) = 10 ⇒ GCD is a factor of 5 = (25, 10). Thus the GCD can be 1, 2 − i, 2 + i, 5. We rule
out 2 + i, 5 by showing that 2 + i 6 | 3 + i. 2 − i then fits the bill.

2
Question 2(b) If K is a finite extension of a field F of degree n, prove taht any element
of K is algebraic over F with degree m where m divides n.

Solution. Let α ∈ K, then the n + 1 elements 1, α, α2 , . . . , αn are linearly dependent over


F , because (K : F ) = degree of K over F = n. Thus there exist aP 0 , a1 , . . . , an ∈ F , not all
0, such that a0 + a1 α + . . . + an αn = 0 ⇒ α is a root of f (x) = ni=0 ai xi ∈ F [x] ⇒ α is
algebraic over F .
Let p(x) be the minimal polynomial of α over F , deg p(x) = m. Then (F (α) : F ) = m
— first of all 1, α, . . . , αm−1 are linearly independent over F , because otherwise α will be
the root of a non-zero polynomial of degree less than m. We know that α algebraic over F
implies F (α) = F [α] as F (α) is the smallest field containing F and α, and F [α] is a field1 .
Now any element of F [α] is a linear combination of 1, α, . . . αm−1 . Take f (α) again.
f (x) = q(x)p(x) + r(x) where r(x) = 0 or deg r(x) < m. Thus f (α) = r(α), hence (F (α) :
F ) = m. We also know that (K : F ) = (K : F (α))(F (α) : F ) (See 2(c), 1993 — if
{v1 , . . . , vr } is a basis of K over F (α), and {w1 , . . . , wm } is a basis of F (α) over F , then
{vi wj | 1 ≤ i ≤ r, 1 ≤ j ≤ m} is a basis for K over F ).
Thus m divides n.
p √
Question√ 2(c) Find the minimum polynomial over Q (the field of rationals) of 5 − 2
and i + 3.

Solution. Let x = i + 3, then (x − i)2 = 3 ⇒ x2 − 2ix + i2 = 3 ⇒ x2 − 4 = 2ix ⇒
(x2 − 4)2 = −4x2 ⇒ x4 − 4x2 + 16 = 0. We shall show that x4 − 4x2 + 16 is irreducible over
Q. If possible, let x4 − 4x2 + 16 = (x2 + ax + b)(x2 + cx + d), then a + c = 0, ac + b + d =
−4, ad + bc = 0, db = 16. Using a + c = 0, ac + bd = 0, we get c(b − d) = 0. If c = 0, then
a = 0, so b + d = −4, bd = 16 so b, d are roots of x2 + 4x + 16, thus b, d are not real numbers.
Thus b = d ⇒ b = d = ±4 ⇒ ac = −12 or ac = 0 (not possible). Thus a, c are roots of
x2 − 12 = 0, thus are not rationals. Hence x4 − 4x2 + 16 is not reducible.
A simpler way of seeing the above is that t2 −4t+16 has non-real roots, hence is irreducible
4
over Q, so xp − 4x2 + 16 is not reducible over Q.
√ 2
√ 4
p Let√x = 5 − 2. Then x − 5 = − 2 ⇒ p x −√ 10x2 + 23 = 0 is a polynomial satisfied by
5 − 2. It is the minimal polynomial of 5 − 2 because it is irreducible over Q, since
2
t − 10t + 23 has non real p roots.
√ √
Hence the degree of 5 − 2 and i + 3 is 4.

1
Let f (α) = a0 + a1 α + . . . + ar αr be any non-zero element of F [α]. Then the polynomial p(x) 6 | f (x) ⇒
(f (x), p(x)) = 1 ⇒ there exist b(x), c(x) ∈ F [x] such that p(x)b(x) + f (x)c(x) = 1 ⇒ f (α)c(α) = 1 ⇒ f (α)
is invertible

3
UPSC Civil Services Main 1991 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

September 13, 2008

Question 1(a) If the group G has no non-trivial subgroups, show that G must be finite of
prime order.

Solution. Here we assume that G has more than one element.


G is cyclic: Let a ∈ G, a 6= e. Let H be the cyclic group generated by a. Then H 6= {e},
therefore H = G, so G is cyclic.
G has finite order: If order of G is infinite, then the group K generated by a2 is a
non-trivial subgroup of G, because K 6= {e}, K 6= G as a 6∈ K — note that a ∈ K, a = (a2 )m
for some m shows that a is of finite order ⇒ G is of finite order. This is a contradiction,
hence order of G is finite.
The order of G is a prime number: If the order is pq, p > 1, q > 1, then order of ap
or equivalently the order of the group generated by ap is q ⇒ G has a nontrivial subgroup,
which is a contradiction. Hence order of G is a prime number.

Question 1(b) Show that a group of order 9 must be abelian.

Solution. We first prove that if G is a group with centre C such that G/C is cyclic, then
G is abelian. Let G/C be generated by the coset aC. Let x, y ∈ G, then xC = (aC)r
and yC = (aC)s for some integers r, s. This means that x ∈ ar C, y ∈ as C and therefore
x = ar c1 , y = as c2 , c1 , c2 ∈ C. Now xy = ar c1 as c2 = ar as c1 c2 since c1 ∈ C, so it commutes
with every element of G. Similarly, c2 ∈ C so it commutes with ar , so
xy = ar+s c1 c2 = as+r c2 c1 = as c2 ar c1 = yx
Hence G is abelian.
Now we prove that a group G of order p2 , p prime, is abelian. In particular, a group of
order 9 will be abelian. Let C be the center of G. Then C is of order p or p2 as the center
of a prime power group is non-trivial (Theorem 2.11.2 page 86 of Algebra by Herstein).

1
If the order of C is p2 , and G = C so G is abelian.
If order of C is p, then G/C is of order p and therefore is a cyclic group. Thus G must
be abelian as shown above. In either case G is abelian.

Question 1(c) If the characteristic of an integral domain D is finite, show that it is a prime
number.

Solution. If possible let m be the chararacteristic of D, where m = pq, p, q > 1. Let


a ∈ D, a 6= 0. Then 0 = ma2 = pa.qa. But D is an integral domain, therefore either pa = 0
or qa = 0. Suppose without loss of generality that pa = 0. If b ∈ D is arbitrary, then
0 = (pa)b = (pb)a. But a 6= 0, therefore pb = 0 ⇒ m is not the smallest positive integer such
that ma = 0 for every a ∈ D. Thus the assumption m has a proper factorization is wrong,
hence m is a prime number.

Question 2(a) Find the greatest common divisor (GCD) in J[i], the ring of Gaussian in-
tegers of (i) 3 + 4i and 4 − 3i (ii) 11 + 7i and 18 − i.

Solution. (i) 4 − 3i = (−i)(3 + 4i), and −i is a unit in J[i] as i(−i) = 1. It follows that
4 − 3i and 3 + 4i are associates of each other. Thus the GCD of 4 − 3i and 3 + 4i can be
taken to be either of them.
(ii) N (11 + 7i) = (11 + 7i)(11 − 7i) = 170, N (18 − i) = 325. Since (170, 325) = 5, we can
find integers x, y such that 170x + 325y = 5, or

(11 + 7i)[(11 − 7i)x] + (18 − i)[(18 + i)y] = 5

showing that if α divides 11 + 7i, 18 − i in J[i], then α divides 5. Therefore the GCD of
11 + 7i, 18 − i is a factor of 5, i.e. 1, 2 − i, 2 + i, 5.
Now 11+7i
2+i
= (11+7i)(2−i)
5
= 295
+ 35 i. Thus 2 + i 6 | 11 + 7i.
11+7i
2−i
= (11+7i)(2+i)
5
= 3 + 5i. Thus 2 − i | 11 + 7i. 18−i 2−i
= (18−i)(2+i)
5
= 37
5
+ 16
5
i, so
2 − i 6 | 18 − i.
Thus the GCD of 11 + 7i and 18 − i is 1.
Note: We could have got this by Euclid’s Algorithm also.

18 − i = (11 + 7i) + 7 − 8i N (7 − 8i) < N (11 + 7i)


11 + 7i = (7 − 8i)i + 3 N (3) < N (7 − 8i)
7 − 8i = (2 − 3i)3 + (1 + i) N (1 + i) < N (3)
3 = (1 + i)(1 − i) + 1 N (1) < N (1 + i)

Thus the GCD of 11 + 7i and 18 − i is 1.

2
Question 2(b) Show that every maximal ideal of a commutative ring R with unit element
is a prime ideal.

Solution. Let M be a maximal ideal. Let ab ≡ 0 mod M , i.e. ab ∈ M . Suppose that


a 6∈ M i.e. a 6≡ 0 mod M . We shall show that b ≡ 0 mod M , proving that M is a
prime ideal. Consider hM, ai , the ideal generated by M and a. Clearly M ⊆ hM, ai and
M 6= hM, ai as a 6∈ M , therefore hM, ai = R as M is maximal. Thus e ∈ hM, ai , where
e is the unit element of R. Thus e = m + xa where m ∈ M, x ∈ R, so b = mb + xab.
mb ∈ M, xab ∈ M because ab ∈ M . Hence mb + xab = b ∈ M , which was to be proved,
showing that M is a prime ideal.
Remark. The converse of the above statement is not true. Let R = Z[x], P = h2i , the
ideal generated by 2, then P is prime but not maximal — in fact h2i ( h2, xi ( R.

Question 2(c) The field K is an extension of the field F . If α, β ∈ K are both algebraic
over F , show that α ± β, αβ, α/β (if β 6= 0) are all algebraic over F .

Solution. Let p(x) be the minimal polynomial of α over F , then F [x]/hp(x)i ' F [α], the
homomorphism from F [x] to F [α] being f (x) = f (α) with kernel hp(x)i . Thus F [α] = F (α)
(the smallest field containing F and α in K). If deg p(x) = n, then 1, α, . . . , αn−1 are linearly
independent over F and generate F (α). Hence (F (α) : F ) = n ⇒ if γ ∈ F (α), γ is algebraic
over F as 1, γ, . . . , γ n are linearly dependent over F , so γ is a root of a polynomial of degree
≤ n.
Now β being algebraic over F , is algebraic over F (α) ⇒ F (α.β) is a finite extension of
F (α), and (F (α, β) : F (α)) = degree of the minimal polynomial of β over F (α) ≤ degree
of the minimal polynomial of β over F . Since (F (α, β) : F ) = (F (α, β) : F (α))(F (α) : F )
(see question 2(c) of 1993), it follows that F (α, β) is an algebraic extension over F . In fact
if (F (α, β) : F ) = m and ζ ∈ F (α, β), then 1, ζ, ζ 2 , . . . , ζ m are linearly dependent, so ζ is a
root of a polynomial of degree ≤ m. Thus α ± β, αβ, α/β, being elements of F (α, β) are all
algebraic over F .

3
UPSC Civil Services Main 1992 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

September 13, 2008

Question 1(a) If H is a cyclic normal subgroup of a group G, then show that every subgroup
of H is a normal subgroup of G.

Solution. Let K be a normal subgroup of H. Let H = hai, and let K = har i, where r is
the least positive integer such that ar ∈ K.
Then k ∈ K ⇒ k = (ar )m for some m.

gkg −1 = g(ar )m g = gam g · gam g . . . gam g


| {z }
r times
Now H is normal in G, so gam g −1 ∈ H ⇒ gam g −1 = at for some t. Thus gkg −1 = (ar )t =
(a ) ⇒ gkg −1 ∈ K. Thus K is normal in G.
r t

Note: Cyclic subgroups need not be normal. G = S3 , H = {I, (1, 2)} is cyclic but not
normal in S3 .

Question 1(b) Show that a group of order 30 is not simple.

Solution. o(G) = 3 · 2 · 5.
n5 = number of Sylow groups of order 5 is 1 or 6 because n5 ≡ 1 mod 5 and n5 | 30.
n3 = number of Sylow groups of order 3 is 1 or 10 because n3 ≡ 1 mod 3 and n3 | 30.
If G has 6 Sylow groups of order 5, then G has 24 elements of order 5, because if H and
K are two subgroups of order 5, then H ∩ K {e} when H 6= K. Thus each Sylow subgroup
of order 5 gives rise to 4 distinct elements of order 5.
If G has 10 subgroups of order 3, then G has 20 elements of order 3. Thus either n3 = 1
or n5 = 1. So G has a unique Sylow subgroup of order 3 or 5, which has to be a normal
subgroup of G. Thus G is not simple.
Note that n5 > 1, n3 > 1 means that G must have at least 45 elements.

1
Question 1(c) Let p be the smallest prime factor of the order of a group G, then prove that
any subgroup of index p is normal in G.
Solution. Let G/H = {x1 H, x2 H, . . . , xp H}. For any x ∈ G consider the mapping πx :
G/H −→ G/H defined by πx (xj H) = xxj H = xk H for some k, 1 ≤ k ≤ p. Clearly πx is
one-one and therefore gives rise to a permutation on p symbols. Let Sp denote the symmetric
group on p symbold. Define φ : G −→ Sp by φ(x) = πx . Then φ is a homomorphism as
πxy (xj H) = xy(xj (H)) = x(yxj H) = πx (πy (xj H) ⇒ φ(xy) = φ(x)φ(y)
Thus by the fundamental theorem of homomorphisms G/K is isomorphic to a subgroup of
Sp , where K is the kernel of φ.
K ⊆ H. Proof: Let x ∈ K. Then πx is the identity permutation in Sp i.e. πx (xj H) =
xxj H = xj H for every j, 1 ≤ j ≤ p. Let xr be such that xr H = H, such an xr exists then
xH = xxr H = xr H = H ⇒ x ∈ H. Thus K ⊆ H.
(G : K) = (G : H)(H : K) — This follows immediately from (G : K) = o(G)/o(K).
(Note that all groups are of finite order here. This statement also holds for groups of infinite
order).
Let (H : K) = r. Then (G : K) = pr and therefore pr | p!, because G/K is isomorphic
to a subgroup of Sp , so order of G/K = (G : K) divides o(Sp ) = p!. Thus r | (p − 1)!. But
r divides o(G) also, because K is a subgroup of H which is a subgroup of G. Consequently
r divides ((p − 1)!, o(G)). But ((p − 1)!, o(G)) = 1 as p is the smallest prime factor of o(G).
Thus r = 1 ⇒ K = H. Hence H being a kernel of a homomorphism φ : G −→ Sp is a
normal subgroup of G.
Remark: We don’t need it in the above proof, but it is worth noticing that
K = ∩a∈G aHa−1
For x ∈ K ⇔ xxj H = xj H ∀j.1 ≤ j ≤ p
⇔ x ∈ xj Hx−1
j ∀j.1 ≤ j ≤ p
−1
⇔ x ∈ aHa ∀a ∈ G
(Note that aHa = xj Hx−1
−1
j if a = xj H).
Proof of (G : K) = (G : H)(H : K). Let G/H = {x1 H, x2 H, . . . , xn H} and H/K =
{y1 K, . . . , ym K}. Then we will show that G/K = {xi yj K | 1 ≤ i ≤ m, 1 ≤ j ≤ n}.
xi y j ≡ xk y l mod K ⇒ yl−1 x−1
k xi y j ∈ K
⇒ yl−1 x−1
k xi y j ∈ H
−1
⇒ xk xi ∈ H (∵ yl , yj ∈ H)
⇒ x k H = xi H ⇒ k = i
⇒ yl−1 yj ∈ K
⇒ yl K = yj K ⇒ l = k
Given x ∈ G, xH = xj H for some j, 1 ≤ j ≤ n. Since x−1 −1
j x ∈ H, xj xK = yk K for some
k, 1 ≤ k ≤ m. Therefore xK = xj yk K, so {xi yj K | 1 ≤ i ≤ m, 1 ≤ j ≤ n} is a complete
system of representation of cosets of G/K. This implies (G : K) = mn = (G : H)(H : K).

2
Question 2(a) If R is a unique factorization domain, then prove that any f ∈ R[x] is an
irreducible element of R[x] if and only if either f is an irreducible element of R or f is an
irreducible polynomial in R[x].
Solution. We first observe that units of R and R[x] are the same — let f, g ∈ R[x] be such
that f g = 1 then deg f + deg g = 0 ⇒ deg f = 0, deg g = 0 ⇒ f, g ∈ R and both are units
in R.
If f is an irreducible element of R, then f is an irreducible element of R[x] — if f = gh
then deg g + deg h = 0 ⇒ deg g = 0, deg h = 0 ⇒ g, h ∈ R, but since f is irreducible in R,
either g is a unit in R or f is a unit in R, and therefore in R[x].
Conversely, if f is an irreducible element in R[x] and f ∈ R, then f has to be irreducible
in R also, because if f = gh is a proper factorization of f ∈ R, then this would be a proper
factorization of f in R[x] also, because units of R and R[x] are the same, so g, h cannot be
units in R[x].
Now let f ∈ R[x] be an irreducible element of R[x] and f 6∈ R, then f is an irreducible
polynomial. But an irreducible polynomial need not be an irreducible element of R[x]. For
example, 2x2 + 2 is an irreducible polynomial in Z[x] but is not an irreducible element. Thus
the correct question would be — f ∈ R[x] is an irreducible element of R[x] if and only if
either f is an irreducible element of R or f is an irreducible primitive polynomial in R[x].

Question 2(b) Prove that the polynomials x2 + 1 and x2 + x + 4 are irreducible over F , the
field of integers modulo 11. Prove that F [x]/hx2 + 1i and F [x]/hx2 + x + 4i are isomorphic
fields each having 121 elements.

Solution. For irreducibility of the polynomial x2 + x + 4 see question 2(c), 1996.


If possible let x2 + 1 ≡ (x + a)(x + b) mod 11 where a, b are integers. This implies that
a + b ≡ 0 mod 11, ab ≡ 1 mod 11 ⇒ a2 ≡ −1 mod 11, which is not possible, since the
only quadratic residues of 11 are 0, 1, 4, 9, 5 and 3. Thus x2 + 1 has no linear factors modulo
11 i.e. x2 + 1 is irreducible modulo 11.
Let p(x) be an irreducible polynomial over a field F and α be a root of p(x) in some
extension of F . Then the field F [x]/hp(x)i is isomorphic to F [α]. Proof: Consider the
mapping ρ : F [x] −→ F [α] defined by ρ(f (x)) = f (α). It can be easily seen that ρ is a
homomorphism, onto with kernel hp(x)i . If deg p(x) = n, then (F [α] : F ) = n. Clearly
1, α, α2 , . . . , αn−1 are independent over F , otherwise α will be the root of a polynomial of
degree < n. Let β ∈ F (α) = F [α], then β = a0 +a1 α+. . . ar αr , let f (x) = a0 +a1 x+. . .+ar xr ,
then there exist q(x), s(x) such that f (x) = q(x)p(x) + r(x) where s(x) = 0 or deg s(x) <
deg p(x). Thus β = f (α) = s(α) as p(α) = 0, showing that β is a linear combination of
1, α, α2 , . . . , αn−1 .
In case p(x) = x2 + 1, F = field of integers modulo 11, then F [x]/hx2 + 1i ' F [α] with
α2 + 1 = 0. Now (F [α] = F (α) : F ) = 2 with 1, α as its basis. Thus F (α) = {a0 + a1 α |
a0 , a1 ∈ F }. Clearly F (α) has 121 elements. Similarly, F [x]/hx2 + x + 4i has 121 elements.
Consider the mapping σ : F [x] −→ F [x] defined by σ(x) = x − 5 and σ(a) = a for a ∈ F .
It is obvious that σ is an isomorphism. Now σ(x2 + 1) = (x − 5)2 + 1 = x2 − 10x + 26 ≡

3
x2 + x + 4 mod 11. This shows that σ gives rise to a map from K1 = F [x]/hx2 + 1i to
K2 = F [x]/hx2 + x + 4i . Any typical element of K1 is of the form a0 + a1 x + hx2 + 1i where
a0 , a1 ∈ F . Then σ(a0 + a1 x + hx2 + 1i ) = a0 + a1 (x − 5) + hx2 + x + 4i .
We now check that σ is an isomorphism. We write αx + β = α + βx + hx2 + 1i . Then

σ(αx + β + γx + δ) = σ((α + γ)x + β + δ)


= (α + γ)x + β + δ − 5((α + γ) + hx2 + x + 4i
= σ(αx + β) + σ(γx + δ)

σ((αx + β)(γx + δ)) = σ(αγx2 + (αδ + βγ)x + βδ)


= σ((αδ + βγ)x + βδ − αγ) as αγx2 ≡ −αγ mod x2 + 1
= (αδ + βγ)x − 5(αδ + βγ) + βδ − αγ + hx2 + x + 4i

Now

(αx + β − 5α + hx2 + x + 4i )(γx + δ − 5γ + hx2 + x + 4i )


= αγx2 + αδx − 5αγx + βγx + βδ − 5βγ − 4αγx − 5αδ + 25γα + hx2 + x + 4i
= αγ(−x − 4) + αδx − 5αγx + βγx + βδ − 5βγ − 4αγx − 5αδ + 3γα + hx2 + x + 4i
= x[−αγ + αδ + βγ − 5αγ − 5αγ] + βδ − 5βγ − 5αδ − αγ + hx2 + x + 4i
≡ x[αδ + βγ] + βδ − 5βγ − 5αδ − αγ + hx2 + x + 4i mod 11

Thus σ((αx + β)(γx + δ)) = σ((αx + β))σ((γx + δ)) showing that σ is a homomorphism.
σ is 1 − 1: The kernel of σ is an ideal of K1 , but K1 is a field, therefore the only ideals
of K1 are the trivial ideal h0i and K1 . Since σ is not a zero map, it follows that the kernel
of σ is h0i , thus σ is 1 − 1.
σ is onto: Since K1 and K2 have 121 elements each, and sigma is one-one, σ(K1 ) = K2 .
Thus σ is an isomorphism from K1 to K2 .

Question 2(c) Find the degree of the splitting field of f (x) = x5 − 3x3 + x2 − 3 over Q, the
field of rationals.

Solution. f (x) has -1 as a root, so f (x) = (x + 1)(x4 − x3 − 2x2 + 3x − 3). It does not
have any other linear factors as −1, 1, 3, −3 are not roots of x4 − x3 − 2x2 + 3x + 3.
Let x4 − x3 − 2x2 + 3x + 3 = (x2 + bx + c)(x2 + dx + e), where b, c, d, e ∈ Z. Then
b + d = −1, c + e + bd = −2, be + dc = 3, ce = −3. From ce = −3, we get c = −1, e = 3 or
c = 1, e = −3 (the other choices are symmetric). Using c = 1, e = −3, we get −3b + d = 3,
and now from b + d = −1, we get b = −1, d = 0. Thus we get

f (x) = (x + 1)(x2 − x + 1)(x2 − 3)

4
√ √
Consequently, the splitting field of f (x) over Q is the smallest field containing ± 3, 1±i2 3 ,
namely the √roots of x2 − 3 and x2 − x + 1. √ √ √
Thus Q( 3, i)√ is the required
√ splitting field. Since Q( 3, i) ⊇ Q( 3) ⊇
√ Q, and (Q( 3) :
Q) = 2 and (Q( 3, i) : Q( 3)) = 2 it follows that the splitting field Q( 3, i) of f (x) has
degree 4 over Q.

5
UPSC Civil Services Main 1993 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 28, 2008

Question 1(a) Let G be a cyclic group of order n and p | n. Prove that there exists a
homomorphism of G onto a cyclic group of order p. What is the kernel of the homomorphism?

Solution. Let G = hai = {a, a2 , . . . , an } and G0 = hbi = {b, b2 , . . . , bp }.


Define φ : G −→ G0 by φ(ar ) = bt where r ≡ t mod p, t = 1, 2, . . . p.
φ(ar · as ) = bu where r + s ≡ u mod p. If φ(ar ) = bx where r ≡ x mod p and φ(as ) = by
where s ≡ y mod p, then x + y ≡ r + s ≡ u mod p ⇒ bx+y = bu . So φ(ar · as ) = φ(ar )φ(as ),
thus φ is a homomorphism.
ker φ = {ar | φ(ar ) = bp ⇔ r ≡ p mod p}. Thus the kernel of φ is {ap , a2p , . . . , amp , mp =
n}.

Question 1(b) Show that a group G of order 56 cannot be simple.

Solution. The number of 7-Sylow subgroups of ≡ 1 mod 7 and divides 56, so can be 1 or
8. If 1, then the 7-Sylow group is normal in G. If the number is 8, then G has 48 elements of
order 7, because if H, K are different Sylow subgroups of order 7, then H ∩ K = {e} because
7 is a prime.
Thus the Sylow subgroups of order 8 can come from the remaining elements which are
only 8 in number (including e). Thus there is a unique Sylow subgroup of order 8, which
has to be normal.
Thus any group of order 56 has a normal subgroup of order 7 or of order 8, so it cannot
be simple.

1
Question 1(c) Let H and K be normal subgroups of G (finite), with H a normal subgroup
of K. If P = K/H, S = G/H, show that G/K ' S/P .

Solution. Define φ : G/H −→ G/K by φ(aH) = aK.


• φ is well-defined: aH = bH ⇔ b−1 a ∈ H ⇒ b−1 a ∈ K(∵ H ⊆ K) ⇒ aK = bK ⇒
φ(aH) = φ(bH).

• φ is a homomorphism: φ(aH · bH) = φ(abH) = abK = aK · bK = φ(aH)φ(bH).

• φ is onto: Given xK ∈ G/K, φ(xH) = xK, xH ∈ G/H.


ker φ = {xH | xK = K}. But xK = K ⇔ x ∈ K. Thus ker φ = K/H. By the fundamental
G/H
theorem of homomorphisms, S/P = K/H ' G/K.
√ √
Question 2(a) If Z is the set of integers, then show that Z[ −3] = {a + b −3 | a, b ∈ Z}
is not a UFD.
√ √ 2 2
Solution. Let α = a + b −3 ∈ Z[ −3]. Then α is a unit √ iff N (α) = a + 3b = 1, because
if N (α) = 1, then αα = 1 ⇒ α is a unit with α = a − b −3 as its inverse. Conversely, if
√ N (αβ) = N (α)N (β) = 1 ⇒ N (α) = 1. In fact N (α) = 1 ⇒ α = ±1, the only
αβ = 1, then
units of Z[ −3].
2 is irreducible. Let 2 = αβ. We will prove that either α or β is a unit. N (2) = 4 ⇒
N (α) = 1, 2, 4. But N (α) = a2 + 3b2 = 2 is not possible for a, b ∈ Z. If N (α) = 1, then α is
a unit, otherwise N (α) = 4 ⇒ N (β) = 1√⇒ β is a √ unit. Thus 2 is irreducible. √
Similarly it can √ that 1 + −3, 1 − −3 are irreducible (N (1 + −3) √
√ be shown = 4).
Moreover, 2, 1 + −3, 1 − −3 are √ not associates
√ of each other as the only units in Z[ −3]
are ±1. Now 4 = 2 ·√2 = (1 + −3)(1 − −3) are two different factorizations of 4 into
irreducibles, hence Z[ −3] is not a UFD.

Question 2(b) Construct the addition and mutiplication table for Z3 [x]/hx2 + 1i, where Z3
is the set of integers modulo 3 and hx2 + 1i is the ideal generated by 1 + x2 .

Solution. Let f (x) = a0 + a1 x + . . . + an xn with ai ∈ Z3 . Since

(−1)r/2

r
x ≡ r−1 mod x2 + 1
(−1) 2 x

it follows that

f (x) ≡ a0 + a1 x − a2 − a3 x + a4 + . . . mod x2 + 1
= [bo ] + [b1 ][x]

where [x] is the residue class of x modulo x2 + 1 and [b0 ], [b1 ] are residue classes in Z3 .
Conversely, [bo ] + [b1 ][x] always belongs to Z3 [x]/hx2 + 1i.

2
Thus Z3 [x]/hx2 + 1i has 9 elements, namely

{0, 1, 2, x, x + 1, x + 2, 2x, 2x + 1, 2x + 2}

Note that we have listed representative elements of distinct residue classes modulo x2 + 1.
Now addition is simply: (a0 + a1 x) + (b0 + b1 x) = c0 + c1 x where ci ≡ ai + bi mod 3 for
i = 0, 1.
Multiplication is defined by (a0 + a1 x)(b0 + b1 x) = c0 + c1 x where c0 ≡ a0 b0 − a1 b1 mod 3
and c1 ≡ a0 b1 +a1 b0 mod 3. The reader can now expand these into the appropriate addition
and multiplication tables.
Notice the strong resemblance between the addition and multiplication rules derived
above and the corresponding rules for complex numbers. This is a consequence of the fact
that i is a root of x2 + 1, in fact the ring of Gaussian integers is isomorphic to Z[x]/hx2 + 1i.

Question 2(c) Let Q be the rational number field, and Q(21/2 , 21/3 ) by the smalled extension
field containing 21/2 , 21/3 . Find a basis of Q(21/2 , 21/3 ) over Q.

Solution. If K ⊇ L ⊇ k are fields such that (K : L) = m, (L : k) = n then (K : k) = mn.


In fact if {v1 , . . . , vm } is a basis of K over L, {w1 , . . . , wn } is a basis of L over k, then
{vi wj | 1 ≤ i ≤ m, 1 ≤ j ≤ n} is a basis of K over k.
X m X n X m X n m X
X n
Proof: Let aij vi wj = 0 with aij ∈ k. Then aij vi wj = ( aij wj )vi =
i=1 j=1 i=1 j=1 i=1 j=1
n
X Xn
0. But aij wj ∈ L, and as v1 , . . . , vm are linearly independent over L, aij wj = 0 for
j=1 j=1
each i, 1 ≤ i ≤ m. However w1 , . . . , wn are linearly independent over k, thus aij = 0 for
1 ≤ i ≤ m, 1 ≤ j ≤ n.
Xm Xn
Let α ∈ K. Then α = αi vi , αi ∈ L. Now let αi = aij wj with aij ∈ k, then
i=1 j=1
m X
X n
α= aij vi wj ⇒ {vi wj | 1 ≤ i ≤ m, 1 ≤ j ≤ n} generate K over k. This completes
i=1 j=1
the proof. √
Now Q(21/2 ) has {1, 2} as a basis over Q, and (Q(21/2 , 21/3 ) : Q(21/2 )) = 3 with x3 −2 as
the irreducible polynomial of 21/3 over Q(21/2 ). Thus 1, 21/3 , 22/3 is a basis of (Q(21/2 , 21/3 )
over Q(21/2 ). Thus by the above result, {1, 21/2 , 21/3 , 21/2+1/3 , 22/3 , 22/3+1/2 } is a basis for
Q(21/2 , 21/3 ) over Q.

3
UPSC Civil Services Main 1994 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 28, 2008

Question 1(a) If G is a group such that (ab)n = an bn for three consecutive integers for all
a, b ∈ G, then show that G is abelian.

Solution. We are given that (ab)i = ai bi , (ab)i+1 = ai+1 bi+1 , (ab)i+2 = ai+2 bi+2 .
Now (ab)i+1 = abai bi = abai bi = ai+1 bi+1 . Thus ai b = bai .
Also, (ab)2 (ab)i = ai+2 bi+2 = a2 ai b2 bi = a2 ai bbbi = a2 bai bbi = a2 b2 ai bi , because ai b = bai .
But (ab)i = ai bi , hence (ab)2 = a2 b2 ⇒ abab = a2 b2 ⇒ ba = ab. Thus G is abelian.
Note that the result is false if we only have two consecutive integers e.g. G = S3 has
(ab)6 = e = a6 b6 , and (ab)7 = (ab)6 ab = ab = a7 b7 .

Question 1(b) Can a group of order 42 be simple? Justify your claim.

Solution. By Sylow theorems, the number of 7-Sylow groups is ≡ 1 mod 7, and divides
42, and therefore divides 6 ⇒ there is only 1 Sylow group of order 7, which has to be normal,
thus a group of order 42 cannot be simple.

Question 1(c) Show that the additive group of integers modulo 4 is isomorphic to the mul-
tiplicative group of the non-zero elements of integers modulo 5. State the two isomorphisms.

Solution.

Z/(4) = {[0], [1], [2], [3]} = h[1]i


Z/h5i = {[1], [2], [3], [4]} = h[2]i
= {[2], [2]2 = [4], [2]3 = [3], [2]4 = [1]}

Two cyclic groups of the same order are isomorphic. φ : Z/(4) −→ Z/h5i:

1
φ([1]) = [2]
φ([1] + [1]) = φ([2]) = [2]2 = [4]
φ([3]) = φ(3.[1]) = [2]3 = [3]
φ([4]) = φ(4.[1]) = [2]4 = [1]
f : Z/h5i −→ Z/(4).

f ([2]) = [1]
f ([4]) = f ([2]2 ) = f ([2]) + f ([2]) = [2]
f ([3]) = f ([2]3 ) = f ([2]) + f ([2]) + f ([2]) = [3]
f ([1]) = f ([2]4 ) = f ([2]) + f ([2]) + f ([2]) + f ([2]) = [4]

Question 2(a) Find all the units of the integral domain of Gaussian integers.
Solution. Let Z[i] = {a + ib | a, b ∈ Z}. Let N (a + ib) = a2 + b2 . We will show that
α ∈ Z[i] is a unit ⇔ N (α) = 1.
If α is a unit then αβ = 1 for some β ∈ Z[i] ⇒ N (αβ) = N (α)N (β) = 1 ⇒ N (α) = 1
because N (α), N (β) are positive integers.
Conversely, N (α) = 1 ⇒ a2 + b2 = 1 ⇒ (a + ib)(a − ib) = 1 ⇒ α is a unit.
Now the only integer solutions to N (α) = a2 + b2 = 1 are a = ±1, b = 0 or a = 0, b = ±1.
Thus the only units are {±1, ±i}.

Question 2(b) Prove or disprove: The polynomial ring I[x] over the ring of integers is a
Principal Ideal Domain (PID).
Solution. It is not a PID. The ideal generated by 2 and x is not a principal ideal. Suppose
h2, xi = hf (x)i. Then 2 ∈ hf (x)i ⇒ f (x)g(x) = 2 for some g(x). This means that f (x) is a
constant and divides 2, so f (x) = 1or2.
f (x) = 2 ⇒ x 6∈ hf (x)i ∵ 2g(x) = x is not possible for any g(x) ∈ I[x].
f (x) = 1 ⇒ 1 ∈ h2, xi ⇒ 1 = 2p(x) + xq(x) ⇒ 2× the constant term of a(x) = 1, which
is not possible. Thus h2, xi is not a principal ideal.

Question 2(c) Let R be an integral domain (not necessarily a unique factorization domain),
and F its field of quotients. Show that any element f (x) ∈ F [x] is of the form f (x) = f0a(x)
where f0 (x) ∈ R[x] and a ∈ R.
Solution.Qf (x) = a0 + a1 x + . . . am xm , where ai ∈ F . Now ai = bi /ci , where bi , ci ∈ R.
Then f (x) i ci = A0 + A1 x + . . . + Am xm where Ai ∈ R.
Thus f (x) = f0a(x) , where f0 (x) = A0 + A1 x + . . . + Am xm , and a = i ci .
Q

2
UPSC Civil Services Main 1995 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 28, 2008

Question 1(a) Let G be a finite set closed under an associative binary operation such that
ab = ac ⇒ b = c, ba = ca ⇒ b = c for all a, b, c ∈ G. Prove that G is a group.

Solution. Let G = {a1 , a2 , . . . , an }. Consider {a1 a1 , a2 a1 , . . . , an a1 } and {a1 a1 , a1 a2 , . . . , a1 an }.


These sets have distinct elements because aj ai = ak ai ⇒ aj = ak , and ai aj = ai ak ⇒ aj = ak .
Thus G = {a1 , a2 , . . . , an } = {a1 a1 , a2 a1 , . . . , an a1 } = {a1 a1 , a1 a2 , . . . , a1 an }. Thus there ex-
ists r, 1 ≤ r ≤ n such that a1 = a1 ar . Now for any aj ∈ G, aj = as a1 for some s, therefore
aj ar = as a1 ar = as a1 = aj . Hence we have proved that G has a right identity. As seen above,
for any aj ∈ G, the set {aj a1 , aj a2 , . . . , aj an } = G, hence therefore there exists k, 1 ≤ k ≤ n
such that aj ak = ar , thus every element has a right inverse.
Similarly, we can show that G has a left identity and every element in G has a left
inverse. Let as be the left identity. Then ar = as ar = as , so the left identity is the same as
the right identity. If ai aj = ar and ak ai = ar , then ak = ak ar = ak ai aj = ar aj = aj (using
associativity), hence the left inverse is the same as the right inverse. Thus G has an identity,
every element of G has an inverse, and the operation is associative, so G is a group.
Alternatively, let x ∈ G and let xy = e, where e is the right identity. Then exy = ee =
e = xy ⇒ ex = x, so e is also the left identity. Now yxy = ye = ey ⇒ yx = e, thus the right
inverse is the same as the left inverse.

Question 1(b) Let G be a subgroup of order pn , where p is a prime number and n > 0. Let
H be a proper subgroup of G and N (H) = {x ∈ G | x−1 hx ∈ H for every h ∈ H} = {x ∈
G | x−1 Hx = H}. Prove that N (H) 6= H.

Solution. The proof is by induction over n. If n = 1, then H = {e} is the only possibility
for a proper subgroup, since G is cyclic. N (H) = G 6= H. If n = 2, it is well known that G
is abelian, and therefore for any proper subgroup H of G, N (H) = G 6= H.

1
Assume as induction hypothesis that the result is true for all groups of order pm where
m < n.
Let G be a group of order pn and let H be a proper subgroup of G. We consider the
following two possible cases
Case (i): H does not contain C, the center of G, then there exists an element z ∈ C − H.
Clearly z ∈ N (H) and therefore N (H) ⊃ H properly.
Case (ii): H ⊇ C. In this case H = H/C is a proper subgroup of G = G/C. Since G
is a prime power group, it is known that the center C of G is nontrivial, therefore |G| =
order of G = pm where m < n. Thus by the induction hypothesis the normalizer of H in G
contains H properly, i.e. there exists an element b ∈ G such that b 6∈ H and b ∈ N (H) i.e.
−1
b Hb = H. It is now obvious that b 6∈ H and b−1 Hb ⊆ HC = H i.e. b ∈ N (H). Hence
N (H) 6= H.
Alternative presentation: Let Co = {e}, C1 = center of G. If C1 6= G, let Z1 be
the center of G/C1 . Let C2 = η −1 (Z1 ), where η : G −→ G/C1 is the natural map. Thus
C2 /C1 = Z1 . If C2 6= G, we define C3 = η −1 (center of G/C2 ), where η is now the natural
map from G ont G/C2 .
Clearly C0 ( C1 ( C2 ( . . . because the center of a prime power group is non-trivial.
Since G is finite, we have Cr = G for some r. Thus C0 ( C1 ( C2 ( . . . ( Cr = G. Now
each Ci is normal in G, because Z1 is normal in G/C1 ⇒ η −1 (Z1 ) = C2 is normal in G and
so on.
Since C0 ⊆ H, and Cr 6⊆ H, there is a k, 0 ≤ k < r such that Ck ⊆ H, Ck+1 6⊆ H. Let
x ∈ Ck+1 , x 6∈ H. For any g ∈ G, x−1 g −1 xg ∈ Ck , because xCk ∈ center of G/Ck , x ∈ Ck+1 ,
which means that xgCk = xCk gCk = gCk xCk = gxCk . Thus x−1 g −1 xg ∈ Ck .
In particular x−1 h−1 xh ∈ Ck ∀h ∈ H. Thus x−1 h−1 xh ∈ H because Ck ⊆ H, or
x−1 h−1 x ∈ H for all h ∈ H. Thus x ∈ N (H). But x 6∈ H, so N (H) 6= H.

Question 1(c) Show that a group of order 112 is not simple.

Solution. Let G be a group of order 112.


If the Sylow 2-subgroup, which is of order 16, is unique, then it is automatically a normal
subgroup of G and we have nothing to prove.
Let us therefore assume that G has more than one Sylow 2-subgroups. By one of Sylow
theorems, the number of such subgroups is ≡ 1 mod 2, and is a divisor of 112 and therefore
of 7. Thus G has 7 subgroups say H1 , H2 , . . . , H7 such that |Hi | = 16, 1 ≤ i ≤ 7.
Observe that Hi ∩ Hj for i 6= j must have at least 4 elements because if not |Hi Hj | ≥ 128
|H ||H |
as |Hi Hj | = |Hii∩Hjj | , which is not possible.
We now consider the following two cases.
Case 1: Suppose (without loss of generality) that |H| = |H1 ∩ H2 | = 8. This means
that H is a normal subgroup of H1 as well as H2 and therefore N (H) contains H1 H2 . But
|H1 ||H2 |
|H1 H2 | = |H 1 ∩H2 |
= 32, therefore |N (H)| ≥ 32 and 16 divides |N (H)| as N (H) ⊃ H.
Consequently |N (H)| = 112 i.e. N (H) = G. Thus H is a normal subgroup of G showing
that G is not simple.

2
Case 2: Let |Hi ∩ Hj | = 4 for i 6= j. Let H = H1 ∩ H2 , then |H| = 4. We have proved
in question 1(b) that NH1 (H) (the normalizer of H in H1 ) contains H properly and also
NH2 (H) contains H properly. Thus each of NH1 (H) and NH2 (H) have 8 or 16 elements.
Case 2(a): One of the normalizers has 16 elements — suppose without loss of generality
that NH1 (H) = H1 , then NG (H) contains H1 and NH2 (H) and therefore NG (H) contains
at least 16 × 8/4 elements, and 16 divides |NG (H)| as H1 ⊂ NG (H) — note that |Hi | =
16, |NH2 (H)| ≥ 8 and H1 ∩ NH2 (H) being a subgroup of H1 ∩ H2 has at most 4 elements.
Thus as in case 1, we get NG (H) = G, so H is a normal subgroup of G, showing that G is
not simple.
Case 2(b): NH1 (H) 6= H1 and NH2 (H) 6= H2 , then |NH1 (H)| = |NH2 (H)| = 8. In this
case NG (H) contains at least 8 × 8/4 elements and 8 divides |NG (H)|. Thus |NG (H)| = 16
or 56. If |NG (H)| = 16, then it is one of the Hi , say NG (H) = H3 , in this case |H1 ∩ H3 | = 8,
which contradicts the precondition for case 2 i.e. |Hi ∩Hj | = 4 for i 6= j. Thus |NG (H)| = 56,
and in this case G is not simple as NG (H) is a proper normal subgroup of G.
This completes the proof.
Alternative Presentation. o(G) = 24 · 7. The number of 7-Sylow subgroups ≡ 1
mod 7 and divides o(G). Thus the number of 7-Sylow subgroups is 1 or 8. If 1, then the
7-Sylow subgroup of G is normal in G, and G is not simple. Otherwise we will show that G
has a unique 16-Sylow subgroup, which will be normal in G and hence G will not be simple.
Let the number of 7-Sylow subgroups be 8. This accounts for 49 elements, 48 of order 7,
and the identity. Note that if H and K are Sylow subgroups of order 7, then H ∩ K = {e}
if H 6= K because order of H is prime.
We are now left with 63 elements + identity. The number of 2-Sylow groups is ≡ 1
mod 2 and divides 7. Thus out of these 64 elements we should get 7 16-Sylow subgroups
(because if there is only one 16-Sylow subgroup, it is normal, hence G is not simple). These
7 subgroups of order 16 will have a unique subgroup of order 8, which would be normal in
G.
Thus in all cases, G is not simple.

Question 2(a) Let R be a ring with identity. If an element of R has more than one right
inverse, show that it has infinitely many right inverses.

Solution. Let ax = e, ay = e, x 6= y, then xa 6= e (because xa = e ⇒ xay = x ⇒ ey = x ⇒


y = x). Consider x, (xa − e) + x, (ya − e) + x. Then

ax = a((xa−e)+x) = axa−a+ax = a−a+ax = ea((ya−e)+x) = aya−a+ax = a−a+ax = e

Thus we get three distinct right inverses (if xa − e + x = ya − e + x then xax = yax ⇒
x = y). So given n inverses a1 , a2 , . . . , an of a, by considering a1 , a1 a − e + a1 , a2 a − e +
a1 , . . . , an a − e + a1 we can get n + 1 distinct right inverses. Hence there must be infinitely
many right inverses.

3
Question 2(b) Let hp(x)i be an ideal generated by an irreducible polynomial in F [x], F a
field. Prove that it is a maximal ideal.

Solution. Let hp(x)i ( M ⊆ F [x]. We will show that M = F [x].


Let g(x) ∈ M, g(x) 6∈ hp(x)i ⇒ p(x) 6 | g(x). Thus (g(x), p(x)) = 1 i.e. g(x) and p(x) are
coprime. Then there exist a(x), b(x) ∈ F [x] such that a(x)g(x) + p(x)b(x) = 1 ⇒ 1 ∈ M ⇒
M = F [x].
Note that F [x] is a principal ideal domain. Therefore hp(x), g(x)i is a principal ideal and
it has to be generated by 1, because p(x) has no other divisors.

Question 2(c) Let F be a field of characteristic p > 0. Let f (x) = a0 + a1 x + . . . + an xn ∈


F [x]. Define f 0 (x) = a1 + 2a2 x + . . . + nan xn−1 . If f 0 (x) = 0, then prove that there exists
g(x) = F [x] such that f (x) = g(xp ) = g(x)p .

Solution. f 0 (x) = 0 ⇔ rar = 0 ⇔ ar = 0 when r 6≡ 0 mod p. Thus f (x) = tm=0 amp xmp
P
where t = [n/p]. Let g(y) = a0 + ap y + . . . atp y t . Then g(xp ) = f (x) = (g(x))p .

4
UPSC Civil Services Main 1996 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 28, 2008

Question 1(a) Let R be the set of all real numbers and G = {(a, b) | a, b ∈ R, a 6= 0}. Let
x : G × G −→ G be defined by (a, b) ∗ (c, d) = (ac, bc + d). Show that (G, ∗) is a group. Is it
abelian? Is (H, ∗) a subgroup of (G, ∗) where H = {(1, b) | b ∈ R}?

Solution.
1. α = (a, b) ∈ G, β = (c, d) ∈ G ⇒ α ∗ β ∈ G ∵ ac 6= 0 when a 6= 0, c 6= 0.

2. (a, b) ∗ (1, 0) = (a, b.1 + 0) = (a, b). (1, 0) ∗ (a, b) = (a, 0.a + b) = (a, b). Hence (1, 0) is
identity of G.

3. (a, b) ∗ (1/a, −b/a) = (1, b/a − b/a) = (1, 0). (1/a, −b/a) ∗ (a, b) = (1, (−ab/a) + b) =
(1, 0). Thus the inverse exists for every (a, b) ∈ G.

4. (a, b) ∗ [(c, d) ∗ (e, f )] = (a, b) ∗ (ce, de + f ) = (ace, bce + de + f ). [(a, b) ∗ (c, d)] ∗ (e, f ) =
(ac, bc + d) ∗ (e, f ) = (ace, bce + de + f ). Thus ∗ is associative.
Hence G is a group.
G is not abelian: (c, d) ∗ (a, b) = (ac, da + b). Thus if (a, b) ∗ (c, d) = (c, d) ∗ (a, b), then
bc + d = da + b. This need not be true, for example if a = d = 1, b = 0.
If (1, a), (1, b) ∈ H, then (1, a) ∗ (1, b) = (1, a + b) ∈ H. (1, 0) ∈ H. Finally (1, a)−1 =
(1, −a) ∈ H, hence H is a subgroup of G.

Question 1(b) Let f be a homomorphism of a group G onto a group G0 with kernel H. For
each subgroup K 0 of G0 , define K as K = {x | x ∈ G, f (x) ∈ K 0 }. Prove that
1. K 0 is isomorphic to K/H.

2. G/K is isomorphic to G0 /K 0 .

1
Solution.

1. Let f ∗ : K −→ K 0 , f ∗ is a restriction of f . f ∗ is a homomorphism onto K 0 (onto


because given y ∈ K 0 , y ∈ G0 ⇒ ∃x ∈ G, f (x) = y. This x ∈ K.). ker(f 0 ) =
{x | x ∈ K, f (x) = e0 , the identity of G0 }. ker(f ∗ ) ⊆ H, but H ⊆ K, therefore
f (x) = f ∗ (x) = e0 for x ∈ H. Thus H ⊆ ker(f ∗ ), that is H = ker(f ∗ ). Thus
K/H ' K 0 by the fundamental theorem of homomorphisms.

2. Let φ : G −→ G0 /K 0 defined by φ(x) = f (x)K 0 . Now

• φ is a homomorphism: φ(xy) = f (xy)K 0 = f (x)f (y)K 0 = f (x)K 0 f (y)K 0 =


φ(x)φ(y).
• φ is onto: Let yK 0 ∈ G0 /K 0 . f is onto ⇒ there exists x ∈ G such that f (x) = y.
Note that y ∈ G0 . Then φ(x) = f (x)K 0 = yK 0 . ker(φ) = K ∵ x ∈ ker(φ) ⇔
f (x)K 0 = K 0 ⇔ f (x) ∈ K 0 ⇔ x ∈ K. Thus G/K ' G0 /K 0 .

Question 1(c) Prove that a normal subgroup H of a group G is maximal ⇔ the quotient
G/H is simple.

Solution. Let G/H be simple. Let K be a normal subgroup of G such that H ⊆ K,


K 6= G. Then K/H is a normal subgroup of G/H. G/H is simple so K/H is identity or
K/H = G/H. If K/H is identity, then K = H. If K/H = G/H then K = G. Hence H is
maximal.
Conversely, let H be maximal. Let H 0 be a normal subgroup of G/H. Assume H 0 is
different from the identity of G/H, i.e. H 0 contains at least one element different from the
identity of G/H. We shall show that H 0 = G/H, showing that G/H is simple.
Let η : G −→ G/H be the natural homomorphism. Then

η −1 (H 0 ) = {x | x ∈ G, η(x) = Hx ∈ H 0 }

is a normal subgroup of G. η −1 (H 0 ) ⊇ H. By assumption, there exists x ∈ G, xH 6= H such


that xH ∈ H 0 ⇒ x ∈ η −1 (H 0 ) but x 6∈ H. Since H is maximal, η −1 (H 0 ) = G ⇒ H 0 = G/H.

Question 2(a) In a ring R, prove that the cancellation law holds in R ⇔ R has no zero
divisors.

Solution. Let ab = ac ⇒ b = c, a 6= 0. Then R has no zero divisors because ab = 0 =


a0, a 6= 0 ⇒ b = 0.
Conversely ab = ac ⇒ a(b − c) = 0, a 6= 0 ⇒ b − c = 0 ⇒ b = c.

2
Question 2(b) If S is an ideal of R and T any subring of R then prove that S is an ideal
of S + T = {s + t | s ∈ S, t ∈ T }.

Solution. The only thing we have to check is α ∈ S + T, a ∈ S ⇒ αa ∈ S (the other


condition a, b ∈ S ⇒ a−b ∈ S is true). α = s+t, s ∈ S, t ∈ T , thus αa = (s+t)a = sa+ta ∈ S
because s ∈ S, a ∈ S ⇒ sa ∈ S, t ∈ T ⇒ t ∈ R, a ∈ S ⇒ ta ∈ S ⇒ sa + ta ∈ S.

Question 2(c) Prove that the polynomial x2 + x + 4 is irreducible over the field of integers
modulo 11.

Solution. If x2 + x + 4 were reducible modulo 11, then it would have a linear factor i.e. it
would have a root in the field Z11 . But

x=0 ⇒ 0+0+4 6≡ 0 mod 11


x=1 ⇒ 1+1+4 6 ≡ 0 mod 11
x=2 ⇒ 4+2+4 6 ≡ 0 mod 11
x=3 ⇒ 9+3+4 6 ≡ 0 mod 11
x=4 ⇒ 16 + 4 + 4 6 ≡ 0 mod 11
x=5 ⇒ 25 + 5 + 4 6 ≡ 0 mod 11
x=6 ⇒ 36 + 6 + 4 6 ≡ 0 mod 11
x=7 ⇒ 49 + 7 + 4 6 ≡ 0 mod 11
x=8 ⇒ 64 + 8 + 4 6 ≡ 0 mod 11
x=9 ⇒ 81 + 9 + 4 6 ≡ 0 mod 11
x = 10 ⇒ 100 + 10 + 4 6 ≡ 0 mod 11

Therefore x2 + x + 4 has no linear factor in Z11 [x] so it is irreducible in Z11 [x].

3
UPSC Civil Services Main 1997 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

Question 1(a) Show that a necessary and sufficient condition for a subset H of a group G
to be a subgroup is HH −1 = H.
Solution. Let H be a subgroup. Clearly H ⊆ HH −1 because h ∈ H can be written as
h = he where h ∈ H, e ∈ H −1 ⇒ h ∈ HH −1 . If x ∈ HH −1 , then x = hk −1 where h, k ∈ H.
But H is a group, so hk −1 ∈ H, thus HH −1 ⊆ H ⇒ HH −1 = H.
Conversely, let H = HH −1 and assume H 6= ∅.
1. a ∈ H ⇒ a−1 ∈ H −1 ⇒ aa−1 ∈ HH −1 = H ⇒ e ∈ H.
2. x, y ∈ H ⇒ xy −1 ∈ H. Thus x ∈ H ⇒ x−1 = ex−1 ∈ H. x, y ∈ H ⇒ y −1 ∈ H ⇒ y ∈
H −1 ⇒ xyinH.
Thus H is a subgroup of G.

Question 1(b) Show that the order of each subgroup of a finite group is a divisor of the
order of the group.
Solution. Lagrange’s theorem, see Theorem 2.4.1 page 41 of Algebra by Herstein.

Question 1(c) In a group G, the commutator of (a, b), a, b ∈ G is the element aba−1 b−1
and the smallest subgroup containing all commutators is called the commutator subgroup of
G. Show that a quotient group G/H is abelian ⇔ H contains the commutator subgroup of
G.
Solution. Let G/H be abelian. Then HaHb = HbHa ⇒ Hab = Hba ⇒ Haba−1 b−1 =
H ⇒ aba−1 b−1 ∈ H. This means H contains all the commutators, and therefore contains
the group generated by them (i.e. the commutator subgroup).
Conversely, if H contains the commutator subgroup, then for any a, b ∈ G, aba−1 b−1 ∈
H ⇒ Haba−1 b−1 = H ⇒ Hab = Hba ⇒ HaHb = HbHa ⇒ G/H is abelian.

1
Question 2(a) If x2 = x for all x in a ring R, show that R is commutative. Give an
example to show that the converse is not true.

Solution. a + b = (a + b)2 = (a + b)(a + b) = a2 + ab + ba + b2 = a + ab + ba + b. Thus


ab + ba = 0. Setting a = b, we get 2b2 = 0 ⇒ 2b = 0. Thus ab = −2ba + ba ⇒ ab = ba. Thus
R is commutative.
Converse is not true — Z is commutative but n2 6= n for n 6= 0, 1.

Question 2(b) Show that an ideal S of the ring of integers Z is a maximal ideal ⇔ S is
generated by a prime integer.

Solution. Let S be maximal. Since Z is a PID, we have S = hqi for some q ∈ Z, q 6= 0, 1, −1.
We will prove that if q | ab, q - a then q | b showing that q is prime. Since q - a, we
have a 6∈ S. Consider the ideal generated by S and a. It is Z, because S is maximal.
hS, ai = Z ⇒ 1 = α + ta, α ∈ S. Thus 1 = xq + ta, α = xq. Hence b = xbq + tab. Clearly
q | RHS, so q | b ⇒ q is a prime.
Conversely let S = hpi where p is a prime. We wish to show that S is maximal. Let A
be an ideal, A ⊃ S and A 6= S, then we shall show that A = Z. Since A ⊃ S, ∃a ∈ A, a 6∈ S.
Now a 6∈ S ⇔ p - a ⇔ (a, p) = 1 ⇔ xa + yp = 1 for some x, y ∈ Z ⇒ 1 ∈ A(∵ a ∈ A ⇒ xa ∈
A, p ∈ A ⇒ yp ∈ A). Hence Z = A, so S is a maximal ideal.

Question 2(c) Show that in an integral domain every prime element is irreducible. Give
an example to show that the converse is not true.

Solution. Let R be an integral domain with unity. Let p be a prime element of R i.e.
if p | ab then p | a or p | b. We have to show that if a | p then either a is an associate
of p or a is a unit. If a | p, then p = ab for some b ∈ R. But p is a prime, therefore
p = ab ⇒ p | ab ⇒ p | a or p | b. If p | a, then p is an associate of a as a | p. If p | b, then
b = px for some x ∈ R. Thus p = pax ⇒ ax = 1 as R is an integral domain, thus a is a unit.
Hence a prime element is irreducible.
The converse is not
√ true. Let R be an √integral domain which is not a unique factorization
domain e.g. R = Z[ −5]. For α = a + b −5, N (α) = a2 + 5b2 . 2 is an irreducible element
of R — 2√= αβ ⇒ N (α)N (β) = 4 ⇒ N (α) = 1, 2, 4. N (α) = 1 ⇒ α is a unit, because if
α = a + b −5 then a2 + 5b2 = 1 ⇒ b = 0, a = ±1 ⇒ a is a unit. If N (α) = 4, then N (β) = 1
2 2
so β is a unit. N (α) = 2 is impossible
√ as a +√5b = 2 is not possible.
√ √
Now 2 is not
√ prime — 2 | (1 + −5)(1 − −5). But 2 - 1 + −5 because
√ (1 + −5) =
2α = 2(a + b −5) ⇒ 2a = 1, which is not possible. Similarly 2 - 1 − −5. So 2 is not
prime.

2
UPSC Civil Services Main 1998 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

Question 1(a) Prove that if a group has only 4 elements then it must be abelian.
Solution. Let G be a group of order 4. If it has an element of order 4, then G is cyclic
and therefore abelian. If G has no elements of order 4, then the order of all elements other
than identity is 2 because the order of an element must be a divisor of 4. Let x, yinG, then
(xy)2 = xyxy = e ⇒ yx = x−1 ey −1 = x−1 y −1 = xy because x−1 = x, y −1 = y. Hence
xy = yx for every x, y ∈ G so G is abelian.
Question 1(b) If H and K are subgroups of G then show that HK is a subgroup of G if
and only if HK = KH.
Solution. See Lemma 2.5.1 page 44 of Algebra by Herstein.
Question 1(c) Show that every group of order 15 has a normal subgroup of order 5.
Solution. By Sylow’s theorem a group G of order 15 has a subgroup of order 5. Again by
one of Sylow’s theorems the number of subgroups is ≡ 1 mod 5, and this number divides 3.
Therefore there is exactly 1 subgroup of order 5, say H. Now aHa−1 is also a subgroup of G
of order 5, but H is the only such subgroup, so aHa−1 = H, hence H is a normal subgroup.
Hence every group of order 15 has a normal subgroup of order 5.
Question 2(a) Let (R, +, .) be a system satisfying all the axioms for a ring with unity with
the possible exception of a + b = b + a. Prove that (R, +, .) is a ring.
Solution. Let e denote unity of R. Then (a+b)(e+e) = a(e+e)+b(e+e) = ae+(a+b)e+be.
Also (a+b)(e+e) = (a+b)e+(a+b)e = ae+be+ae+be. Thus ae+be = be+ae ⇒ a+b = b+a.
Thus R is a ring.
A similar question is the following. Let (R, +, .) be a system satisfying all the axioms for
a ring with the possible exception of a + b = b + a. If there is an element c ∈ R such that
ac = bc ⇒ a = b for every a, b ∈ R, then show that R is a ring.

1
Question 2(b) If p is a prime then prove that Zp is a field. Discuss the case when p is not
a prime.

Solution. Zp is a commutative ring with unity. Let [a] ∈ Zp such that a 6≡ 0 mod p
i.e. [a] 6= [0]. Let {[x1 ], . . . , [xp ]} = Zp . Then [a][x1 ], . . . , [a][xp ] are all distinct, since
[a][xi ] = [a][xj ] ⇒ a(xi − xj ) ≡ 0 mod p ⇒ xi ≡ xj mod p because a 6≡ 0 mod p. Thus
there exists k such that [a][xk ] = [1] ⇒ every non-zero element in Zp has an inverse. Thus
Z∗p = Zp − {[0]} is a group, so Zp is a field.
If p is not prime, then Zp is not even an integral domain — if p = n1 n2 , n1 > 1, n2 > 1,
then [n1 ][n2 ] = [0], but [n1 ] 6= [0], [n2 ] 6= [0] in Zp .
See corollary to Lemma 3.2.2 page 128 of Algebra by Herstein.

Question 2(c) Let D be a principal ideal domain. Show that every element that is neither
0 nor a unit in D is a product of irreducible elements.

Solution.

1. If A1 ⊆ A2 ⊆ . . . ⊆ Ak ⊆ Ak+1 ⊆ . . . is an ascending chain of ideals, then there exists


an integer m thus that Am = Am+1 = . . ..
Proof: Let A = ∞
S
i=1 Ai , then we will show that A is an ideal — If a, b ∈ A, then
a ∈ Ar for some r, and b ∈ As for some s. Hence a, b ∈ As if s ≥ r (say), thus a−b ∈ As
because As is an ideal ⇒ a − b ∈ A. Let a ∈ A, d ∈ D ⇒ a ∈ Ar ⇒ ra ∈ Ar because
Ar is an ideal ⇒ ra ∈ A. Thus A is an ideal. Since D is a PID, A = hai, i.e.a generates
A. By definition of A, there exists m s.t. a ∈ Am . Thus A = Am = Am+1 = . . . ⊂= A.

2. Every nonzero, non-unit element in D is divisible by an irreducible element.


Proof: Let a ∈ D, a 6= 0, a non-unit. If a is not irreducible then we have nothing to
prove. If a is irreducible, then a has a proper divisor, say a1 ⇒ ha1 i ⊂ hai. Continuing
this process, we have a2 , a3 , . . . , such that as divides as−1 for s = 1, 2, . . ., where a0 = a.
But this sequence must terminate i.e. ∃m such that ham i = ham+1 i = . . . because of
step 1. But this means that am has no proper factors i.e.am is irreducible.

3. Let a ∈ D, a non-unit. If a is irreducible, there is nothing to prove. If not, by step


2, a = p1 a1 where p1 is irreducible, and a1 | a properly. If a1 is a unit, then a is a
product of irreducible factors. If not, then a1 = p2 a2 where a2 | a1 properly. But this
process cannot go on forever, by the same argument as in step 2. Thus we must have
an integer k such that a = p1 p2 . . . pk ak where ak is a unit. Thus a is a product of
irreducible elements.

2
UPSC Civil Services Main 1999 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

Question 1(a) If φ is a homomorphism of G into H with kernel K, then show that K is a


normal subgroup of G.

Solution. See Lemma 2.7.3 page 57 of Algebra by Herstein.

Question 1(b) If p is a prime number pα | o(G), then prove that G has a subgroup of order
pα .

Solution. This is Sylow’s theorem. See theorem 2.12.1 page 92 of Algebra by Herstein.

Question 1(c) Let R be a commutative ring with unit element whose only ideals are the
zero ideal and R itself. Show that R is a field.

Solution. See Lemma 3.5.1 Page 137 of Algebra by Herstein.

1
UPSC Civil Services Main 2000 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

October 25, 2007

Question 1(a) Let n be a fixed positive integer and let Zn be the ring of integers modulo n.
Let
G = {[a] ∈ Zn | a 6= 0, (a, n) = 1}
Show that G is a group under multiplication defined in Zn . Hence or otherwise show that
aφ(n) ≡ 1 mod n for all integers a relatively prime to n, where φ(n) denotes the number of
positive integers that are less than n and relatively prime to n.

Solution. The only thing we have to show is that every element in G is invertible as we
already know that G is multiplicatively closed, and has identity element namely [1]. Let
a1 , . . . , am , m = φ(n) be representatives of prime residue classes modulo n. Let a be any
integer such that (a, n) = 1, i.e. a is coprime with n. Then [aa1 ], [aa2 ], . . . , [aam ] are all
distinct because aai ≡ aaj mod n ⇒ a(ai − aj ) ≡ 0 mod n, but (a, n) = 1, therefore
ai − aj ≡ mod n, which is not true. Thus there exists a j such that aaj ≡ 1 mod n, note
that G = {[a1 ], [a2 ], . . . , [am ]} = {[aa1 ], [aa2 ], . . . , [aam ]} and [1] ∈ G. Consequently [a] is
invertible, in fact [a][aj ] = [1].
We know that if G is a group of order n, then x ∈ G ⇒ xn = e for every x ∈ G, where e
is the identity of G — consider H the cyclic subgroup of G generated by x. Using Lagrange’s
theorem, which says that the order of a subgroup divides the order of a group if the group
is finite, we get o(x) = o(H) | o(G) = n. Thus n = o(x)k, so xn = xo(x)k = ek = e.
Thus if a is any integer such that (a, n) = 1, then [a] ∈ G ⇒ [a]φ(n) = [1] because
o(G) = φ(n). Hence aφ(n) ≡ 1 mod n.

Question 1(b) Let M be a subgroup and N a normal subgroup of a group G. Show that
M N is a subgroup of G and M N/N is isomorphic to M/M ∩ N .

Solution.

1
1. M N 6= ∅

2. x, y ∈ M N ⇒ x = m1 n1 , y = m2 n2 where m1 , m2 ∈ M, n1 , n2 ∈ N . Then xy =
m1 n1 m2 n2 = m1 m2 m−1 −1
2 n1 m2 n2 . Since N is a normal subgroup, m2 n1 m2 ∈ N , there-
fore xy = m1 m2 n∗1 n2 where n∗1 = m−12 n1 m2 , showing that xy ∈ M N .

3. x ∈ M N ⇒ x−1 = n−1 −1 −1 −1 −1
1 m1 = m1 m1 n1 m1 ∈ M N

Thus M N is a subgroup of G.
Consider the function f : M −→ M N/N defined by f (m) = mN . Then

1. f (m1 m2 ) = m1 m2 N = m1 N m2 N = f (m1 )f (m2 ) as N is a normal subgroup.

2. f is onto. If xN is any element of M N/N where x = mn, then xN = mnN = mN =


f (m).

3. ker f = {m | f (m) = mN = N ⇔ m ∈ N } = M ∩ N .

Thus f is a homomorphism, and by the fundamental theorem of homomorphisms, M/M ∩


N ' M N/N .

Question 2(a) Let F be a finite field. Show that the characteristic of F must be a prime
integer p and the number of elements in F must be pm for some positive integer m.

Solution. Let characteristic F = n. Let n = λµ and let a ∈ F, a 6= 0, 0 = na2 = λaµa =


0 ⇒ λa = 0 or µa = 0. Suppose λa = 0. Then for any b ∈ F, b 6= 0, λab = a.λb = 0 ⇒ λb = 0
because a 6= 0. Thus λx = 0 for every x ∈ F , so λ = n because n is the smallest such integer.
Thus if n = λµ, then λ = n or µ = n, so n is prime, say p.
Consider the mapping f : Z −→ F defined by f (n) = ne where e is the multiplicative
identity of F . It is obvious that f is a homomorphism, and that ker f is hpi, the ideal
generated by p. Thus Z/hpi is isomorphic to a subfield of F . In other words F contains a
field Λ having p elements. If (F : Λ) = m, then F has pm elements. For details see question
2(c)(ii) year 2002.

Question 2(b) Let F be a field and F [x] denote the set of all polynomials defined over F .
If f (x) is an irreducible polynomial in F [x], show that the ideal hf (x)i generated by f (x) in
F [x] is maximal and F [x]/hf (x)i is a field.

Solution. Let A by an ideal, A ⊃ hf (x)i. Since F [x] is a principal ideal domain, let
A = hg(x)i. Then A ⊃ hf (x)i ⇒ f (x) = g(x)h(x). But f (x) is irreducible, so either g(x) is
a unit or g(x) is an associate of f (x). Thus hg(x)i = F [x] or hg(x)i = hf (x)i ⇒ hf (x)i is
maximal.
In order to show that F [x]/hf (x)i is a field, the only thing we have to show is that any
non-zero element in F [x]/hf (x)i is invertible. Let g(x) + hf (x)i be any non-zero element in
F [x]/hf (x)i i.e.f (x) - g(x). This means that f (x), g(x) are comprime, therefore there exist

2
a(x), b(x) such that a(x)g(x) + b(x)f (x) = 1. Consequently a(x)g(x) ≡ 1 mod f (x), thus
g(x) + hf (x)i has a(x) + hf (x)i as an inverse in F [x]/hf (x)i.
Alternately, let g(x) be as above. Consider M = ideal generated by f (x), g(x). Since
f (x) - g(x), M 6= hf (x)i, and as hf (x)i is maximal, M = F [x]. Thus there exists a(x), b(x) ∈
F [x] such that a(x)g(x) + b(x)f (x) = 1 and we get the same conclusion as above.

Question 2(c) Show that a finite commutative ring with no zero divisors must be a field.

Solution. See Lemma 3.2.2 page 127 of Algebra by Herstein.

3
UPSC Civil Services Main 2001 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

Question 1(a) Let K be a field and G be a finite subgroup of the multiplicative group of
the non-zero elements of K. Show that G is a cyclic group.
Solution. Let a ∈ G be chosen that o(a) ≥ o(b) ∀b ∈ G — this is possible because G is
finite. We shall show that G = hai.
Step 1. For any b ∈ G, o(b) | o(a). If not, there exists an element b ∈ G s.t. o(b) =
pl r, (p, r) = 1, o(a) = pm s, (p, s) = 1 where l > m ≥ 0, because if all primes occurring in
m
o(b) have power less than that occurring in o(a), then o(b) | o(a). Define x = br , y = ap ⇒
o(x) = pl , o(y) = s ⇒ o(xy) = pl s(∵ (o(x), o(y)) = 1, xy = yx) ⇒ o(xy) > o(a) which is a
contradiction. Hence o(b) | o(a).
Step 2. If o(a) = n, then bn = 1 ∀b ∈ G. Thus all elements of G are roots of xn − 1 = 0.
But this equation has at most n roots in K, thus |G| ≤ n. But o(a) = n ∴ 1, a, . . . , an−1 are
all distinct in G. Therefore o(G) ≥ n.
Thus o(G) = n ⇒ hai = G so G is cyclic.
Question 1(b) Prove that the polynomial 1 + x + . . . xp−1 , where p is a prime number, is
irreducible over the field of rational numbers.
xp −1
Solution. f (x) is irreducible ⇐⇒ f (1 + x) is irreducible. f (x) = x−1
. Thus
(x + 1)p − 1
f (1 + x) =
x
x + p1 xp−1 + p p
p
 
2
xp−2 + . . . + p−1
x
=
  x     
p−1 p p−2 p p−3 p p−r−1 p
= x + x + x + ... + x + ... +
1 2 r p−1
p p p!
 
Now p | r for r = 1, 2, . . . , p − 1, as r = r!(p−r)! , and p | p!, but p - r!, p - (p − r)!. Thus
the Eisenstein criterion gives the result.

1
Question 2(a) Let N be a normal subgroup of a group G. Show that G/N is abelian ⇔ for
all x, y ∈ G, xyx−1 y −1 ∈ N .

Solution. Let G/N be abelian, then xN yN = yN xN ⇒ xyN = yxN ⇒ x−1 y −1 xyN =


N ⇒ x−1 y −1 xy ∈ N .
Conversely, xyx−1 y −1 ∈ N ⇒ xyx−1 y −1 N = N ⇒ x−1 y −1 N = y −1 x−1 N ⇒ x−1 N y −1 N =
y −1 N x−1 N . Thus G/N is abelian.

Question 2(b) If R is a commutative ring with unit element and M is an ideal of R, show
that M is a maximal ideal of R if and only if R/M is a field.

Solution. Theorem 3.51, page 139 of Algebra by Herstein.

Question 2(c) Prove that every finite extension of a field is an algebraic extension. Give
an example to show that the converse is not true.

Solution. Let K | k be a finite extension. Let (K : k) = n i.e. dimension of K as a vector


space over k is n. Let α ∈ K, α 6= 0, then {1, α, α2 , . . . , αn } are linearly dependent, i.e. there
exist a0 , a1 , . . . , anP∈ k with at least one ai 6= 0 such that a0 + a1 α + . . . + an αn = 0 i.e. α is
a root of f (x) = ni=0 ai xi with coefficients from k. Thus K | k is an algebraic extension of
k as every element of K is algebraic over k.
Example: Let K = Q(21/n , n = 2, 3, 4, . . .). K | Q is algebraic but (K : Q) is not finite.
If (K : Q) = r then 21/n for n > r + 1 is a root of the polynomial of degree ≤ r + 1, which
is not possible because 21/n is a root of xn − 2 = 0 which is an irreducible polynomial over
Q, showing that 21/n cannot be a root of a polynomial of degree < n.
K | Q is algebraic because every element is contained in a field L such that Q ⊆ L ⊂ K
and (L : Q) < ∞ ⇒ α is algebraic over Q.

2
UPSC Civil Services Main 2002 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

Question 1(a) Show that a group of order 35 is cyclic.

Solution. Let G be a group of order 35. By Sylow’s theorem, G has a subgroup of order 7
and the number of such groups is ≡ 1 mod 7 and divides 35. If the number is 7t + 1, then
7t + 1 | 35 ⇒ 7t + 1 | 5 ⇒ t = 0. Thus G has a unique Sylow subgroup H of order 7, which
must be a normal subgroup of G. Similarly the number of 5-Sylow subgroups is ≡ 1 mod 5
and divides 35. If this number is 5r + 1, then 5r + 1 | 35 ⇒ 5r + 1 | 7 ⇒ r = 0. Thus G has
a unique 5-Sylow group say K which is normal in G.
This means HK is a subgroup of G. o(H) = 7, o(K) = 5 ⇒ H ∩ K = {e}. Let
x ∈ H, o(x) = 7, y ∈ K, o(y) = 5. Now xyx−1 y −1 = x · yx−1 y −1 ∈ H, xyx−1 · y −1 ∈ K ⇒
xyx−1 y −1 = e ⇒ xy = yx ⇒ o(xy) = 35. Thus G is cyclic, xy being a generator.
More generally, o(G) = pq, p < q, p - q − 1 ⇒ G is cyclic of order pq, using a similar
argument.

Question 1(b) Show that the polynomial 25x4 + 9x3 + 3x + 3 is irreducible over the field of
rational numbers.

Solution. Eisenstein’s irreducibility criterion states that if f (x) = a0 + a1 x + . . . + an xn


is a polynomial with integer coefficients and if there is a prime p such that p | ai , 0 ≤ i <
n, p2 - a0 , p - an then f (x) is irreducible over the rationals. (Proof: see theorem 3.10.2 page
160 of Topics in Algebra by Herstein). In the present case p = 3 does the trick.

Question 2(a) 1. Show that a group of order p2 is abelian where p is a prime number.

2. Prove that a group of order 42 has a normal subgroup of order 7.

Solution.

1
1. o(G) = p2 . Let C be the center of G. Then the center of G is nontrivial ∵ o(G) is a
power of a prime. If o(C) = p2 , then G = C ⇒ G is abelian. If o(C) = p, then G/C is
cyclic of order p. Let G/C be generated by aC. Let x, y ∈ G ⇒ xC = ar C, yC = as C
for some r, s, 1 ≤ r, s < p. Then

⇒ x = ar c1 , y = as c2 for some c1 , c2 ∈ C
⇒ xy = ar c1 as c2
⇒ xy = ar as c1 c2 ∵ c1 as = as c1
⇒ xy = as c2 ar c1 = yx

Thus G is abelian.

2. By Sylow’s theorem, G has a subgroup of order 7. The number of Sylow subgroups is


≡ 1 mod 7 and divides 42. If this number is 7r + 1, then 7r + 1 | 42 ⇒ 7r + 1 | 6 ⇒
r = 0. Thus this subgroup H is unique. Consider aHa−1 , a ∈ G. o(aHa−1 ) = o(H) =
7 ∴ aHa−1 = H as H is the unique subgroup of order 7. Thus H is a normal subgroup
of order 7.

Question 2(b) Prove that in the ring F [x] of polynomials over a field F , the ideal I = [p(x)]
is maximal ⇐⇒ p(x) is irreducible over F .

Solution. Let I be maximal. If g(x) is any divisor of p(x), then [g(x)] ⊇ I ⇒ [g(x)] = F [x]
or [g(x)] = [p(x)]. Thus g(x) is a unit or g(x) is an associate of p(x). Thus p(x) is irreducible.
Conversely let p(x) be irreducible. Let M be an ideal, M ⊇ [p(x)]. Since F [x] is Euclidean
and therefore a Principal Ideal Domain, M = [f (x)] say. Then p(x) ∈ [f (x)] ⇒ f (x) | p(x).
Thus f (x) is a unit or f (x) is an associate of p(x) ⇒ M = F [x] or M = [p(x)] ⇒ I is
maximal.

Question 2(c) 1. Show that every finite Integral domain is a field.

2. Let F be a field with q elements and let E be an extension of F of degree n over F .


Show that E has q n elements.

Solution.

1. See Lemma 3.2.2 page 127 of Algebra by Herstein.

2. E as a vector space over F has degree n. If x1 , . . . , xn is a basis of E over F , then


α ∈ E ⇒ α = α1 x1 + . . . + αn xn where α1 , . . . , αn ∈ F are uniquely determined by α.
Thus E = {α1 x1 + . . . + αn xn | αi ∈ F } ' F n =⇒ |E| = q n , as each αi has q choices.

2
UPSC Civil Services Main 2003 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

Question 1(a) If H is a subgroup of a group G such that x2 ∈ H for every x ∈ G then


prove that H is a normal subgroup of G.
Solution. Let h ∈ H, g ∈ G. Then h(h−1 g −1 )2 g 2 (g −1 hg)2 = hh−1 g −1 h−1 g −1 g 2 g −1 hgg −1 hg =
g −1 hg. Now x ∈ G ⇒ x2 ∈ H, therefore (h−1 g −1 )2 , g 2 , (g −1 hg)2 ∈ H Consequently for any
h ∈ H, g −1 hg ∈ H. Thus H is a normal subgroup of G.
Alternative solution. We shal prove that Hx = xH for every x ∈ G. Clearly for
any h ∈ H, xh = xh.xh.h−1 x−1 x−1 x = h1 x, where h1 = (xh)2 h−1 x−2 ∈ H, this shows that
xH ⊆ Hx. Similarly hx = xx−1 x−1 h−1 hxhx = xh1 with h1 = x−2 h−1 (hx)2 ∈ H. Thus
Hx ⊆ xH, so xH = Hx for every x ∈ G. Hence H is a normal subgroup of G.

Question 1(b) Show that the ring Z[i] = {a + bi | a, b ∈ Z, i = −1} of Gaussian integers
is a Euclidean domain.
Solution. For α = a + ib ∈ Z[i], we define N (α) = a2 + b2 . Clearly (i) N (α) > 0 for α 6= 0,
(ii) For α 6= 0, β 6= 0, N (αβ) = N (α)N (β). Let α = a + ib, β = c + id 6= 0. We shall find
γ, δ ∈ Z[i] such that α = βγ + δ where δ = 0 or N (δ) < N (β). This will prove Z[i] is a
Euclidean domain for the Euclidean function N (α).

α a + ib (a + ib)(c − id)
= = = p + iq
β c + id c2 + d2
where p, q are rational numbers. We determine integers x, y so that |p−x| ≤ 21 , |q −y| ≤ 1
2
— x, y are the integers nearest to p, q respectively. Let γ = x + iy. Then
α
= γ + (p − x) + (q − y)i ⇒ α = βγ + βη = βγ + δ
β
where δ = βη. Clearly δ = α − βγ is a Gaussian integer, and if δ 6= 0, then N (δ) =
N (β)[(p − x)2 + (q − y)2 ] ≤ N (β)[ 41 + 41 ] < N (β). This completes the proof.

1
Question 2(a) 1. Let R be the ring of all real-valued continuous functions on the closed
interval [0, 1]. Let M = {f (x) ∈ R | f ( 31 ) = 0}. Show that M is a maximal ideal of
R.

2. Let M, N be two ideals of a ring R. Show that M ∪ N is an ideal of R if and only of


either M ⊆ N or N ⊆ M .

Solution.
1. M is an ideal. M 6= ∅ since the function f (x) = 0 clearly belongs to M .
Let f, g ∈ M then the function (f − g)(x) = f (x) − g(x) is continuous everywhere on
[0, 1] and (f − g)( 13 ) = f ( 13 ) − g( 31 ) = 0, so f − g ∈ M . Thus M is a subgroup of the
group (R, +).
If g ∈ M and f ∈ R, then the function (f g)(x) = f (x)g(x) is continuous everywhere
on [0, 1] and (f g)( 31 ) = f ( 13 )g( 13 ) = 0 as g( 31 ) = 0, thus f g ∈ M . Thus M is an ideal of
R. Note that R is a commutative ring with unity I, where I(x) = 1.
Let M ⊆ A ⊆ R where A is an ideal of R. If M 6= A, we shall show that A = R. Let
β ∈ A − M , thus β( 13 ) = c 6= 0. Define α : [0, 1] −→ [0, 1] by α(x) = c for all x ∈ [0, 1].
Then the function µ = β − α ∈ M ⊂ A as µ( 31 ) = 0. Thus α = β − µ ∈ A as β, µ ∈ A.
Now consider γ : [0, 1] −→ [0, 1] defined by γ(x) = 1c for all x. Clearly γ ∈ R. Since A
is an ideal, γα ∈ A. But γα(x) = 1c c = 1, thus γα = I ∈ A. Since I is unity in R, it
follows that A = R, hence M is a maximal ideal of R.
Note: The converse of the above statement is also true i.e. if M is a maximal ideal
of R, then there exists number r ∈ [0, 1] such that M = {f | f ∈ R, f (r) = 0}. The
proof needs compactness of [0, 1] which is not an algebraic concept.

2. If M ⊆ N , then M ∪ N = N and if N ⊆ M , then M ∪ N = M , so in both cases M ∪ N


is an ideal of R.
Conversely, let M ∪ N be an ideal of R. If possible, let us assume that M * N and
N * M , this means there exist x ∈ M − N, y ∈ N − M . Now x ∈ M, y ∈ N ⇒ x, y ∈
M ∪ N . But M ∪ N is an ideal, thus x − y ∈ M ∪ N , hence x − y ∈ M or x − y ∈ N . If
x − y ∈ M , then x − (x − y) = y ∈ M as M is an ideal, but this is a contradiction. If
x − y ∈ N , then (x − y) + y = x ∈ N , which is also not possible. Thus our assumption
that M * N and N * M is incorrect, so if M ∪ N is an ideal, either M ⊆ N or
N ⊆ M.


Question 2(b) 1. Show that Q( 3, i) is the splitting field for x5 − 3x3 + x2 − 3 where Q
is the field of rational numbers.

2. Prove that x2 + x + 4 is irreducible over F , the field of integers modulo 11 and prove
further that F [x]/hx2 + x + 4i is a field with 121 elements.

2
Solution.
1. x5 − 3x3 + x2 − 3 = x3 (x2 − 3) + x2 − 3 = (x2 − 3)(x3 +√1) = (x2 − 3)(x + 1)(x2 − x + 1).

Thus the roots of x5 − 3x3 + x2 − 3 are −1, ± 3, 1±i2 3 . Consequently all the roots of

the given polynomial lie in the field Q( 3, i). Conversely, if K√ is any field containing
√ the roots of the √given polynomial, then 3 ∈ K, and therefore
Q, which contains
i ∈ K, thus Q( 3, i) ⊆ K. Thus √ Q( 3, i) is the smallest field containing all the roots
5 3 2
of x − 3x + x − 3. Thus Q( 3, i) is the splitting field of the given polynomial over
Q.
2. See question 2(c) from 1996 for the irreducibility of x2 + x + 4 over F .
See question 2(b) from 1992 for the second part.

Question 2(c) If R is a unique factorization domain (UFD), then prove that R[x] is also
a UFD.
Solution. Let F denote the field of quotients of R.
Result 1. If f (x) ∈ R[x] is irreducible, then f (x) remains irreducible in F [x]. (Note that
the converse is obvious as R[x] ⊆ F [x].) Let f (x) be reducible in F [x] i.e. f (x) = g(x)h(x),
where deg g(x) < deg f (x), deg h(x) < deg f (x) and g(x), h(x) ∈ F [x]. We can write g(x) =
a1 b−1 −1
1 g1 (x), h(x) = a2 b2 h1 (x), where g1 (x), h1 (x) ∈ R[x] and are primitive and a1 , b1 , a2 , b2 ∈
R (b1 is the LCM of all the denominators of g(x), and a1 is the GCD of the numerators). Thus
b1 b2 f (x) = a1 a2 g1 (x)h1 (x). But by Gauss Lemma, the product of two primitive polynomials
is primitive, therefore g1 (x)h1 (x) is primitive. Since f (x) is irreducible in R[x], therefore it
is also primitive. Consequently b1 b2 = content of b1 b2 f (x) = a1 a2 = content of a1 a2 g(x)h(x)
and therefore we get f (x) = g1 (x)h1 (x), thus f (x) is reducible in R[x]. Hence if f (x) is
irreducible in R[x] then it is irreducible in F [x].
Result2. Factorization exists in R[x]. Let f (x) ∈ R[x], f (x) 6= 0 and f (x) not a unit.
Let a = c(f ) = content of f then f = af ∗ where f ∗ is a primitive polynomial in R[x] of the
same degree as f . Since F [x] is a UFD (being a Euclidean domain), we can write f ∗ (x) =
p1 (x) . . . pr (x), where each pi (x) is an irreducible element of F [x]. Let pi (x) = ai b−1i qi (x),
where ai , bi ∈ R, and qi (x) ∈ R[x] is a primitive polynomial. Thus we get
b1 . . . br f ∗ (x) = a1 . . . ar q1 (x) . . . qr (x)
But the product q1 (x) . . . qr (x) is again primitive (Gauss Lemma), therefore equating the
contents of both sides (note that f ∗ (x) is primitive), we get b1 . . . br = a1 . . . ar , therefore
f ∗ (x) = q1 (x) . . . qr (x)
where each qi (x) ∈ R[x] and is irreducible in F [x] and therefore irreducible in R[x]. Since R
is a UFD, a = π1 . . . πt , where π1 , . . . , πt are irreducible in R. Thus
f (x) = π1 . . . πt q1 (x) . . . qr (x)

3
where π1 , . . . , πt , q1 (x), . . . , qr (x) are irreducible elements of R[x]. Note that π1 , . . . , πt being
constants cannot have a proper factorization in R[x] if they do not have one in R. Hence
the result is established.
Result 3. Uniqueness. If possible, let

π1 . . . πt q1 (x) . . . qr (x) = π10 . . . πu0 g1 (x) . . . gs (x)

where π1 , . . . , πt , π10 , . . . , πu0 are irreducible elements in R and q1 (x) . . . qr (x), g1 (x) . . . gs (x) are
irreducible elements of R[x]. Using Gauss Lemma, we get that the products q1 (x) . . . qr (x),
g1 (x) . . . gs (x) are primitive. Comparing the contents of both sides, we get π1 . . . πt =
π10 . . . πu0 . But R is a UFD, so t = u, and we can reorder the πi0 to ensure that each πi
is an associate of πi0 . Thus we are left with q1 (x) . . . qr (x) = g1 (x) . . . gs (x). We consider this
equation in F [x]. By the first result each qi , gj , 1 ≤ i ≤ r, 1 ≤ j ≤ s is irreducible in F [x].
Since F [x] is a UFD, we get r = s and by reordering, we get that qi (x) is an associate of
gi (x) in F [x]. We can assume w.l.o.g. that qi (x) = (unit in F [x])gi (x), 1 ≤ i ≤ r. Since
units in F [x] are non-zero constants, these are of the form cd−1 where c, d ∈ R. Thus we
get di qi (x) = ci gi (x). Using contents, we conclude that di = ci , thus qi (x) is an associate of
gi (x) in R[x], so the factorization is unique.
Thus R[x] is a UFD.

4
UPSC Civil Services Main 2004 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

Question 1(a) If p is a prime number of the form 4n + 1, n a natural number, then show
that the congruence x2 ≡ −1 mod p is solvable.
Solution. Consider the multiplicative group G of non-zero residue classes modulo p. In this
group [1] and [p − 1] are the only two elements which are there own inverses as the equation
x2 = [1] has exactly two solutions in the field Z/pZ. Since order of G is φ(p) = p − 1 = 4n,
the remaining 4n − 2 elements form 2n − 1 pairs, in each pair each element is the inverse of
the other. Thus Y
[r] = [1]
1<r<p−1

as each one of the 2n − 1 pairs when multiplied would give us [1]. Consequently
Y
[r] = [p − 1] =⇒ (p − 1)! ≡ p − 1 ≡ −1 mod p
1≤r≤p−1

This is Wilson’s theorem. Now

p − 1 p + 1 
(p − 1)! = 1 · 2 · ... · · . . . · (p − 1)
2 2
p − 1 p − 1 p − 3
= 1 · 2 · ... · p− p− . . . (p − 1))
2 2 2
p−1 p − 1 2
≡ (−1) 2 ( )! mod p
2
p−1
Since p ≡ 1 mod 4, (−1) = 1 and we get
2

p − 1 2
( )! ≡ (p − 1)! ≡ −1 mod p
2
showing that the congruence x2 ≡ −1 mod p is solvable.

1
Question 1(b) Let G be a group and let a, b ∈ G. If ab = ba and (O(a), O(b)) = 1 then
show that O(ab) = O(a)O(b).

Solution. Let O(a) = l, O(b) = m and O(ab) = k. Now (ab)lm = alm blm because ab = ba.
But alm = (al )m = e, blm = (bm )l = e therefore (ab)lm = e and consequently k divides lm.
Also e = (ab)k = ak bk ⇒ ak = b−k ⇒ akm = b−km = e ⇒ l | km, but (l, m) = 1, therefore
l | k. Considering e = akl = b−kl , we get m | k. Since (l, m) = 1, we get lm | k. Hence
k = lm completing the proof.

Question 2(a) Verify that the set E of the four roots of x4 − 1 = 0 forms a multiplicative
group. Also prove that a transformation T, T (n) = in is a homomorphism from I+ (group of
all integers with addition) onto E under multiplication.
2πi 4πi 6πi 8πi πi 3πi
Solution. Clearly E = {e 4 , e 4 , e 4 , e 4 } = {e 2 , eπi , e 2 , e2πi } = {1, −1, i, −i}. The
following multiplication table shows that E is a multiplication group.

1 −1 i −i
1 1 −1 i −i
−1 −1 1 −i i
i i −i −1 1
−i −i i 1 −1

In fact

1. α, β ∈ E ⇒ αβ ∈ E

2. 1 is the multiplicative identity of E.

3. Each element of E has an inverse in E.

4. The operation of multiplication is associative in E as it is so in C − {0}.

Now T (n) = in = i, −1, −i, 1 according as n = 1, 2, 3, 0 mod 4. Thus T is a mapping


from I+ to E and it is clearly onto (note that T (0) = 1, T (1) = i, T (2) = −1, T (3) = −i).
Moreover T is a homomorphism is obvious as T (m + n) = im+n = im in = T (m)T (n).

Question 2(b) Prove that if the cancellation law holds for a ring R then a(6= 0) ∈ R is not
a zero divisor and conversely.

Solution. Assume the cancellation law holds. If a 6= 0 and ab = 0 for some b ∈ R, then we
get ab = a0 and since a 6= 0, the cancellation law gives us b = 0, showing that a is not a zero
divisor. Conversely assume R has no zero divisors. Let a 6= 0 and ax = ay ⇒ a(x − y) = 0.
This should imply x − y = 0 because otherwise a will be a zero divisor. Thus ax = ay, a 6=
0 =⇒ x = y i.e. cancellation law holds. This completes the proof.

2
Question 2(c) Show that the residue class ring Z/(m) is a field if m is a prime number.

Solution. We first show that Z/(m) is a commutative ring for all m ≥ 1.

1. Let [a] = {x | x ∈ Z, x ≡ a mod m}, [b] ∈ Z/(m), then [a] + [b] = [a + b] as x ≡ a


mod m, y ≡ b mod m ⇒ x + y ≡ a + b mod m.

2. [a] + [b] = [a + b] = [b + a] = [b] + [a] for every [a], [b] ∈ Z/(m).

3. [a] + [0] = [a + 0] = [a] for every [a] ∈ Z/(m).

4. If [a] ∈ Z/(m) then [−a] ∈ Z/(m) and [a] + [−a] = [a + (−a)] = [0], hence [−a] is the
additive inverse of [a].

5. [a] + ([b] + [c]) = [a] + [b + c] = [a + b + c] = [a + b] + [c] = ([a] + [b]) + [c] showing the
operation is additive.

6. For [a], [b] ∈ Z/(m), [a][b] = [ab] = [b][a] as x ≡ a mod m, y ≡ b mod m ⇒ xy ≡ ab


mod m. This shows that Z/(m) is closed with respect to the operation of multiplication
of residue classes modulo m. Moreover this operation is commutative.

7. Clearly [a]([b][c]) = [a][bc] = [abc] = ([a][b])[c], so multiplication is associative.

8. [a]([b] + [c]) = [a][b + c] = [a(b + c)] = [ab + ac] = [a][b] + [a][c] and ([a] + [b])[c] =
[a][c] + [b][c].

9. [a][1] = [1][a] = [a].

Thus Z/(m) is a commutative ring with identity. To show that it is a field, we have to show
that every non-zero element is invertible.
Let [a] ∈ Z/(m), [a] 6= [0]. This means that a 6≡ 0 mod m. Since m is a prime, it
follow that (a, m) = 1, and therefore there exist integers b and c such that ab + cm = 1.
Consequently ab ≡ 1 mod m, or [a][b] = [1] i.e. [a] is invertible and [b] is its inverse.
Hence Z/(m) is a field when m is a prime.
Note: Z/(m) is not even an integral domain when m > 1 and is not prime. m = bc where
1 < b < m, 1 < c < m and therefore [b], [c] 6= [0] but [b][c] = [bc] = [0], showing that Z/(m)
has zero divisors and is not an integral domain.

Question 2(d) Define an irreducible element and a prime element in an integral domain
D with unit. Prove that every prime element in D is irreducible, but the converse of this is
not in general true,

Solution. Irreducible element. An element a 6= 0 which is not a unit in D is said to be


an irreducible element if a = bc implies that either b or c is a unit (consequently either b or
c is an associate of a).

3
Prime. An element a 6= 0, a not a unit is is said to be a prime element if a | bc ⇒ a | b
or a | c.
Every prime is irreducible. Let a be a prime element in D. If possible let a = bc, we
shall show that either b or c is a unit. Since a | bc and a is a prime element, a | b or a | c. If
a | b then there exists x ∈ D such that b = xa =⇒ a = xac. But D is an integral domain
and therefore cancellation holds. Thus a = xac ⇒ 1 = xc i.e. c is a unit. Similarly we can
show that if a | c then b is a unit. Thus a is an irreducible element i.e. it has no proper
divisors.
Example where an irreducible need not be prime.
√ √
D = Z[ −5] = {a + b −5 | a, b ∈ Z}

We define for α ∈ D, N (α) = a2 + 5b2 where α = a + b −5. Clearly for α, β ∈ D,
N (αβ) = N (α)N (β). Moreover α ∈ D is a unit if and only if N (α) = 1. Thus D has only
two units namely ±1.
Now we show that 2 is irreducible but not a prime. If possible, let 2 = αβ, then
N (2) = 4 = N (α)N (β), showing that N (α) = 1, 2, 4 ⇒ √ N (β) = 4, 2, 1. If N (α) = 1, then α
is a unit, and if N (α) = 4, then β is a unit. If α = a + b −5, then N (α) = 2 ⇒ a + 5b2 = 2,
which is not possible for√a, b ∈ Z. Hence
√ 2 is irreducible.
√ √
However√2 · 3 = (1 + √ −5)(1 − −5), i.e. 2 | (1 + −5)(1 √ − −5) but 2 does not divide
either (1 + −5) or (1 − −5), √ because N (2) = 4, N (1 ± −5) = 6 and 4 - 6. Hence 2 is
irreducible but not prime in Z[ −5].

4
UPSC Civil Services Main 2005 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

Question 1(a) If M and N are normal subgroups of a group G such that M ∩ N = {e},
show that every element of M commutes with every element of N .

Solution. Let x ∈ M, y ∈ N . We consider the element α = xyx−1 y −1 . Now x−1 ∈ M


and y ∈ N ⊆ G, and M is a normal subgroup of G, therefore yx−1 y −1 ∈ M , consequently
α ∈ M . Similarly since N is a normal subgroup of G and y ∈ N , xyx−1 ∈ N , hence
α = xyx−1 y −1 ∈ N . Thus α ∈ M ∩ N , which means that α = xyx−1 y −1 = e ⇒ xy = yx i.e.
every element of M commutes with every element of N .

Question 1(b) Show that (1 + i) is a prime element in the ring R of Gaussian integers.

Solution. The ring of Gaussian integers is a Euclidean domain with Euclidean function
N (a + ib) = a2 + b2 , therefore any two elements α, β ∈ R have a GCD (greatest common
divisor). If d is the GCD of α, β, then there exist γ, δ ∈ R such that αγ + βδ = d. Moreover
α is a unit in R if and only if N (α) = 1, because if N (α) = 1 then αα = 1, implying that
α is a unit, and conversely, if α is a unit, then there exist β ∈ R such that αβ = 1, and
therefore N (αβ) = N (α)N (β) = 1 ⇒ N (α) = N (β) = 1 as both are positive integers.
First of all we prove that 1 + i is an irreducible element (note that it is not a unit as
N (1 + i) = 2). Let 1 + i = αβ. Taking norm of both sides, we get N (αβ) = N (α)N (β) =
2 ⇒ N (α) = 1 or N (β) = 1, so either α is a unit or β is a unit. Thus 1 + i is an irreducible
element.
Let 1 + i divide αβ and assume that 1 + i does not divide α. We shall show that 1 + i
divides β. Since the only divisors of 1 + i are 1 + i and units, and 1 + i does not divide α, it
follows that GCD of α and 1 + i is 1. Thus there exists γ, δ ∈ R such that γ(1 + i) + δα = 1
or γβ(1 + i) + δαβ = β. Since (1 + i) divides the left hand side of this equation, it follows
that 1 + i divides β. Hence 1 + i is a prime element in R.

1
Question
p 2(a) p H and K be two subgroups of a finite group G, such that |H| >
1. Let
|G| and |K| > |G|. Prove that H ∩ K 6= {e}.

2. If f : G −→ G0 is an isomorphism, prove that the order of a ∈ G is equal to the order


of f (a).

Solution.
|H||K|
1. We prove that |HK| = |H∩K|
.
If H ∩ K = {e}, then hk = h1 k1 ⇔ h−1 1 h = k1 k
−1
⇔ h−1 1 h, k1 k
−1
∈ H ∩ K ⇔ h−11 h =
k1 k −1 = e ⇔ h = h1 , k = k1 . Thus there are no repetitions in HK = {hk | h ∈
|H||K|
H, k ∈ K}, so |HK| = |H||K| = |H∩K| . (This is sufficient to prove the result, but for
completeness we show the result when H ∩ K 6= {e}.)
If H ∩ K 6= {e}, then hk = h1 k1 ⇔ h−1 1 h, k1 k
−1
∈ H ∩ K ⇔ h−11 h = k1 k
−1
=u∈
−1
H ∩ K ⇔ h = h1 u, k = u k1 with u ∈ H ∩ K. Thus hk is duplicated at least |H ∩ K|
times as hk = (hu)(u−1 k) with u ∈ H ∩K. It is duplicated no more than |H ∩K| times,
|H||K|
because hk = h1 k1 ⇒ h = h1 u, k = u−1 k1 with u ∈ H ∩ K. Hence |HK| = |H∩K| .
√ √
|H||K| |G| |G|
Now |G| ≥ |HK| = |H∩K| ≥ |H∩K| Thus |H ∩ K| > 1, so |H ∩ K| 6= {e}.

2. Let o(a) = order of a = m and order of f (a) = o(f (a)) = n. Then e0 = f (am ) = f (a)m ,
where e0 is the identity of G0 , showing that n divides m. Conversely, f (e) = e0 =
f (a)n = f (an ) ⇒ an = e as f is one-one. This means that m divides n. Thus m = n,
which was to be proved.

Question 2(b) Prove that any polynomial ring F [x] over a field F is a UFD.

Solution. We know that F [x] is a Euclidean domain with the Euclidean function being
the degree of the polynomial — the algorithm being: given f (x), g(x) 6= 0 belonging to
F [x], there exist q(x), r(x) ∈ F [x] such that f (x) = q(x)g(x) + r(x) where r(x) = 0 or
deg r(x) < deg g(x).
Step 1. If f (x), g(x) ∈ F [x], both not 0, then they have a GCD d(x), and there exist
λ(x), µ(x) ∈ F [x] such that d(x) = f (x)λ(x) + g(x)µ(x). Let S = {f (x)a(x) + g(x)b(x) |
a(x), b(x) ∈ F [x]}. Then S 6= ∅, as f (x), g(x) ∈ S. Let d(x) be a non-zero polynomial is S
with minimal degree, i.e. deg d(x) ≤ deg h(x) for every nonzero h(x) ∈ S. Clearly if any d0 (x)
divides f (x) and g(x), then d0 (x) divides d(x) because d(x) is of the form f (x)a(x)+g(x)b(x).
Moreover d(x) divides both f (x) and g(x), otherwise we have q(x), r(x) ∈ F [x] such that
f (x) = d(x)q(x) + r(x) where deg r(x) < deg d(x), but this is not possible as r(x) ∈ S as it is
of the form f (x)a(x) + g(x)b(x) so deg r(x) ≥ deg d(x). So d(x) divides f (x), and similarly
d(x) divides g(x).

2
Step 2. An irreducible element of F [x] is a prime element i.e. if f (x) is irreducible and
f (x) | g(x)h(x) and f (x) - g(x) then f (x) | h(x).
If f (x) - g(x), then f (x) is irreducible implies its only divisors are units or associates of
f (x). Therefore the GCD of f (x) and g(x) is 1. By Step 1, we have 1 = f (x)a(x) + g(x)b(x)
for some a(x), b(x) ∈ F [x]. Thus h(x) = h(x)f (x)a(x) + h(x)g(x)b(x). Clearly f (x) divides
the right hand side, so f (x) | h(x), as required.
Step 3. Every non-zero non-unit element in F [x] can be written as the product of
irreducible elements in F [x].
The proof is by induction on the degree of f (x). If deg f (x) = 0, then f (x) is a non-zero
constant, therefore a unit in F [x], so we have nothing to prove.
Let the result be true for all polynomials whose degree is < deg f (x). If f (x) is irre-
ducible, we have nothing to prove. If f (x) is not irreducible, then there exist g(x), h(x),
1 ≤ deg g(x), deg h(x) < deg f (x) such that g(x)h(x) = f (x). Now by induction both g(x)
and h(x) are products of irreducible elements, therefore f (x) is the product of irreducible
elements.
Step 4: Uniqueness. If possible let

f (x) = cf1 (x) . . . fr (x) = dg1 (x) . . . gs (x)

where f1 , . . . , fr , g1 , . . . , gs are irreducible, and c, d ∈ F . We will show that r = s and that


the gi ’s can be reordered such that each fi is the associate of gi .
Now f1 (x) divides g1 (x) . . . gs (x), therefore by step 2, f1 (x) must divide one of g1 (x), . . . , gs (x).
Let us assume without loss of generality that f1 (x) | g1 (x), but g1 (x) is also irreducible and
f1 (x) is not a unit, therefore f1 (x) and g1 (x) are associates. Thus we get

c0 f2 (x) . . . fr (x) = d0 g2 (x) . . . gs (x)

If r < s, then after r steps we shall get gr+1 (x) . . . gs (x) = 1, which is not possible, hence
r ≥ s, similarly s ≥ r so r = s. Now by relabelling g1 , . . . , gr we get each fi (x) is an associate
of gi (x), 1 ≤ i ≤ r. Hence F [x] is a UFD.

3
UPSC Civil Services Main 2006 - Mathematics
Algebra
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

December 16, 2007

Question 1(a) Let S be the set of all real numbers except −1. Define ∗ on S by

a ∗ b = a + b + ab

Is (S, ∗) a group? Find the solution of the equation

2∗x∗3=7

in S.

Solution. Clearly S 6= ∅.

1. S is closed for the operation (∗). If a+b+ab = −1, then a+b+ab+1 = (a+1)(b+1) =
0 ⇒ a = −1 or b = −1. Thus a, b ∈ S ⇒ a 6= −1, b 6= −1 ⇒ a+b+ab 6= −1 ⇒ a∗b ∈ S.

2. a ∗ 0 = 0 ∗ a = a + 0 + a.0 = a, showing that 0 is the identity for S.


a a a2
3. a 6= −1, then b = − 1+a 6= −1 and a ∗ b = b ∗ a = a − 1+a
− 1+a
= 0, thus S is closed
with respect to inverses for the operation (∗).

4. a ∗ b = b ∗ a for every a, b ∈ S.

5. (a ∗ b) ∗ c = (a + b + ab) ∗ c = a + b + ab + c + ac + bc + abc and a ∗ (b ∗ c) = a ∗ (b + c + bc) =


a + b + c + bc + ab + ac + abc. Thus (a ∗ b) ∗ c = a ∗ (b ∗ c) thus the operation (∗) is
associative.

Hence (S, ∗) is an abelian group.


2 ∗ x ∗ 3 = 2 + x + 3 + 2x + 3x + 6 + 6x. Therefore we want 12x + 11 = 7, so x = − 31 .

1
Question 1(b) If G is a group of real numbers under addition and N is the subgroup of
G consisting of the integers, prove that G/N is isomorphic to the group H of all complex
numbers of absolute value 1 under multiplication.

Solution. Let f : G −→ H be defined by f (α) = e2iπα . Then f is an onto homomorphism.

1. f (α + β) = e2iπ(α+β) = e2iπα e2iπβ = f (α)f (β).

2. Let z be any complex number with |z| = 1, then z 6= 0. Let θ = arg z, then

θ
f( ) = eiθ = z

Moreover kernel f = N, because α ∈ kernel f if and only if e2iπα = 1 ⇔ α ∈ N. Thus by the


fundamental theorem of homomorphisms G/N is isomorphic to H.
Alternative solution. Let f be as defined above. Define φ : G/N −→ H by φ(α) =
φ(α + N) = f (α) for α ∈ G. Then

1. φ is well defined i.e.. if α = β then φ(α) = φ(β) i.e. φ does not depend on the choice
of representative in the coset. Clearly α = β ⇔ α − β ∈ N ⇒ e2iπα = e2iπβ ⇒ f (α) =
f (β).

2. φ is a homomorphism. φ(α + β) = φ(α + β) = f (α + β) = f (α)f (β) = φ(α)φ(β).

3. φ is 1-1. If α 6= β, then α − β 6∈ N and therefore e2iπ(α−β) 6= 1 ⇒ f (α) 6= f (β) ⇒


φ(α) 6= φ(β).

4. φ is onto. If z is any complex number with |z| = 1 and α ∈ G is so determined that


f (α) = z (as above) then φ(α) = f (α) = z.

Thus φ is an isomorphism from G/N onto H i.e. G/N is isomorphic to H.

Question 2(a) 1. Let O(G) = 108. Show that there exists a normal subgroup of order
27 or 9.

2. Let G be the set of all those ordered pairs (a, b) of real numbers for which a 6= 0 and
define in G an operation ⊗ as follows:

(a, b) ⊗ (c, d) = (ac, bc + d)

Examine whether G is a group with respect to the operation ⊗. If it is a group, is G


abelian?

Solution.

2
1. According to one of the Sylow theorems, the number of subgroups of G of order 27 is
≡ 1 (modulo 3) and is a divisor of 108 and therefore of 4, thus the number of such
subgroups is 1 or 4. If G has a unique Sylow group H of order 27, then it has to be a
normal subgroup because O(a−1 Ha) = 27 and therefore a−1 Ha = H for every a ∈ G.
Let us therefore assume that G has more than one subgroup of order 27. Then G has
four subgroups of order 27, say H1 , H2 , H3 , H4 .
We first of all observer that Hi ∩ Hj must have at least 9 elements, because if not, then
|H ||H |
|Hi Hj |, the number of elements in Hi Hj , would be at least 243 as |Hi Hj | = |Hii∩Hjj | ,
and this is not possible. Let H = Hi ∩ Hj , i 6= j, then O(H) = 9, because Hi 6= Hj .
Now NHi (H), the normalizer of H in Hi , contains H properly (see 1995 question 1(b)),
showing that NHi (H) = Hi and similarly NHj (H) = Hj . Thus NG (H) ⊇ Hi as well as
Hj and therefore O(NG (H)) ≥ 81 and is divisor of 108. Hence NG (H) = G and H is
a normal subgroup of G. Thus G has a normal subgroup of order 27 or of order 9.

2. We observe that G 6= ∅ and

(a) G is closed with respect to the operation ⊗ i.e. (a, b), (c, d) ∈ G ⇒ (a, b) ⊗ (c, d) ∈
G.
(b) (1, 0) is identity of G w.r.t. ⊗ as (a, b)(1, 0) = (a, b) = (1, 0) ⊗ (a, b)
(c) If (a, b) ∈ G, then (a−1 , −ba−1 ) ∈ G as a 6= 0, and (a, b) ⊗ (a−1 , −ba−1 ) = (1, 0) =
(a−1 , −ba−1 )(a, b). Thus every element of G has an inverse w.r.t. the operation
⊗ and it belongs to G.
(d) (a, b) ⊗ ((c, d) ⊗ (e, f )) = (a, b) ⊗ (cd, de + f ) = (ace, bce + de + f ) = ((a, b) ⊗
(c, d)) ⊗ (e, f )

Thus G is a subgroup w.r.t. operation ⊗. G is not an abelian group, as (a, b) ⊗ (2, 0) =


(2a, 2b) whereas (2, 0) ⊗ (a, b) = (2a, b) showing that (2, 0) ⊗ (a, b) 6= (a, b) ⊗ (2, 0) when
b 6= 0.

√ √
Question 2(b) Show that Z[ 2] = {a + 2b | a, b ∈ Z} is a Euclidean domain.

Solution. Definition: An integral domain R 6= {0} is called a Euclidean domain if there


exists a function g : R − {0} −→ Z (ring of integers) such that
1. g(a) ≥ 0 for every a ∈ R∗ = R − {0}.

2. For every a, b ∈ R∗ , g(ab) ≥ g(a).

3. Euclid’s Algorithm: For every a ∈ R, b ∈ R∗ , there exist q, r ∈ R such that a = bq + r,


where r = 0 or g(r) < g(b).
√ √
For α ∈ Z[ 2], α = a + b 2, a, b ∈ Z, we define N (α) = a2 − 2b2 and g(α) = |N (α)|. Clearly

3

1. g(α) ≥ 0 for every α ∈ Z[ 2], α 6= 0.

2. For α, β ∈ Z[ 2], α 6= 0, β 6= 0, g(αβ) = g(α)g(β) ≥ g(α) because g(β) ≥ 1.
√ √
Note that if α = a + b 2, β = c + d 2, then

N (α)N (β) = (a2 − 2b2 )(c2 − 2d2 )


= a2 c2 + 4b2 d2 − 2a2 d2 − 2b2 c2
= (ac + 2bd)2 − 2(ad + bc)2

= N (ac + 2bd + 2(ad + bc)
= N (αβ)
√ √ √ √
3. Let α = a + b 2 ∈ Z[ 2] and β = c + d 2 ∈ Z[ 2] and β 6= 0. Clearly
√ √
α (a + b 2)(c − d 2) √
= √ √ =p+q 2
β (c + d 2)(c − d 2)

where p = cac−2bd bc−ad


2 −2d2 , q = c2 −2d2 are rational numbers. Let m, n be the integers nearest to

p, q respectively i.e. |p − m| ≤ 12 , |q − n| ≤ 12 . Note that if p = [p] + θ, where 0 ≤ θ < 1


and [p] is the integral part of p, then m = [p] if θ ≤ 21 and m = [p] + 1 if θ > 12 .
Let p − m = r, q − n = s, then |r| ≤ 12 , |s| ≤ 12 . Now
√ √ √
α = a + b 2 = (c + d 2)(p + q 2)
√ √
= (c + d 2)((m + r) + (n + s) 2)
√ √ √ √
= (c + d 2)(m + n 2) + (c + d 2)(r + s 2)
√ √ √ √
Let γ = m + n √2, δ = (c + d 2)(r + s 2), then α = βγ + δ, where γ ∈ Z[ 2] and
δ = α − βγ ∈ Z[ 2].
Now either δ = 0 or g(δ) = |N (β)||r2 − 2s2 |. But |r2 − 2s2 | ≤ 41 + 24 < 1, therefore
√ √
g(δ) < g(β). Thus given α, β ∈ Z[ 2], β 6= 0, we have found γ, δ ∈ Z[ 2] such that
α = βγ + δ where δ = 0 or g(δ) < g(β).

This shows that Z[ 2] is a Euclidean domain.

4
UPSC Civil Services Main 1979 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 16, 2010

Question 1(a) If f (x) ∈ C n+1 for |x − a| ≤ h, then show that


n Z x (n+1)
X f (k) (a) k f (t)(x − t)n
f (x) = (x − a) + dt
k=0
k! a n!

Further show that the remainder after n + 1 terms can be expressed as


f (n+1) (X)(x − X)n
(x − a)
n!
where a < X < x.
Solution. Carrying out integration by parts on the given integral repeatedly, we get
Z x (n+1) x Z x (n)
f (t)(x − t)n f (n) (t)(x − t)n f (t)(x − t)n−1
dt = + dt
a n! n! a a (n − 1)!
Z x (n)
f (n) (a)(x − a)n f (t)(x − t)n−1
= − + dt
n! a (n − 1)!
Z x (n−1)
f (n) (a)(x − a)n f (n−1) (a)(x − a)n−1 f (t)(x − t)n−2
= − − + dt
n! (n − 1)! a (n − 2)!
= ...
n Z x
X f (r) (a)(x − a)r
= − + f 0 (t) dt
r=1
r! a
n
X f (r) (a)(x − a)r
= − + f (x) − f (a)
r=1
r!
n
X f (r) (a)(x − a)r
= − + f (x)
r=0
r!

1
Thus n x
f (r) (a)(x − a)r f (n+1) (t)(x − t)n
X Z
f (x) = + dt
r=0
r! a n!
as required. Now we use the result — if f (x) is continuous in [a, b], then
Z b
f (x) dx = f (µ)(b − a) for some µ ∈ (a, b)
a
Rb
This is obvious as m(b − a) ≤ a f (x) dx ≤ M (b − a), where m, M are respectively the
min and max of f (x) over [a, b]. and every value between m and M is attained (by the
intermediate value theorem).
Since f ∈ C n+1 , f (n+1) is continuous and we can use the above result. Therefore
Z x (n+1)
f (t)(x − t)n f (n+1) (X)(x − X)n
dt = (x − a)
a n! n!
where a < X < x.

Question 1(b) If  2 2

2xy x − y ,

(x, y) 6= (0, 0)
f (x, y) = x2 + y 2
0, (x, y) = (0, 0)

∂ 2f ∂ 2f
show that 6= . Explain this result.
∂x ∂y ∂y ∂x
Solution.
f (0, k) − f (0, 0) 0
fy (0, 0) = lim = lim = 0
k→0 k k→0 k
2 −k 2
f (h, k) − f (h, 0) 2hk hh2 +k 2
fy (h, 0) = lim = lim = 2h
k→0 k k→0 k
∂ 2f fy (h, 0) − fy (0, 0) 2h
(0, 0) = lim = lim =2
∂x ∂y h→0 h h→0 h

f (h, 0) − f (0, 0) 0
fx (0, 0) = lim = lim = 0
h→0 h h→0 h
2 2
f (h, k) − f (0, k) 2hk hh2 −k
+k2
fx (0, k) = lim = lim = −2k
h→0 h h→0 h
∂ 2f fx (0, k) − fx (0, 0) −2k
(0, 0) = lim = lim = −2
∂y ∂x k→0 k k→0 k

∂ 2f ∂ 2f
Thus 6= at (0, 0). The reason for this is that neither fxy nor fyx is continuous
∂x ∂y ∂y ∂x
at (0, 0). This implies that neither fx (x, y) nor fy (x, y) is differentiable at (0, 0). Thus the
criteria of Young’s or Schwartz theorem are not satisfied.

2
R1
Question 2(a) If B(x, y) = 0
tx−1 (1 − t)y−1 dt, x > 0, y > 0, then show that B(x, x) =
21−2x B(x, 12 ).

Solution. See 1982, question 2(b), where we proved

Γ( 12 )Γ(x) 22x−1 [Γ(x)]2


=
Γ(x + 12 ) Γ(2x)

Γ( 21 )Γ(x) 1 [Γ(x)]2
Since = B(x, ) and = B(x, x), we get B(x, x) = 21−2x B(x, 12 ) as required.
Γ(x + 12 ) 2
Γ(2x)

Paper II

Question 3(a) Find the maximum of x21 x22 . . . x2n under the restriction x21 +x22 +. . .+x2n = 1.
Using the result derive the inequality
1 a1 + . . . + an
(a1 a2 . . . an ) n ≤
n
for positive real numbers a1 , . . . , an .

Solution. Let F (x1 , . . . , xn ) = x21 x22 . . . x2n + λ( ni=1 x2i − 1), where λ is Lagrange’s unde-
P
termined multiplier. For extreme values,
∂F
= 2x21 x22 . . . x2i−1 xi x2i+1 . . . x2n + 2λxi = 0, 1 ≤ i ≤ n
∂xi
Since x21 +x22 +. . .+x2n = 1, (x1 , . . . , xn ) 6= (0, . . . , 0), it follows that λ = −x21 x22 . . . x2i−1 x2i+1 . . . x2n
for 1 ≤ i ≤ n. Thus x21 = x22 = . . . = x2n = n1 , and λ = − nn−1 1
.

∂ 2F
= 2x21 x22 . . . x2i−1 x2i+1 . . . x2n + 2λ = 0 for 1 ≤ i ≤ n
∂x2i
n
∂ 2F Y 4 1
= 4xi xj x2r = at x21 = x22 = . . . = x2n =
∂xi ∂xj r=1,r6=i,j
nn−1 n
8 X
⇒ d2 F = dxi dxj (only the cross terms appear)
nn−1 1≤i<j≤n

Now x21 + x22 + . . . + x2n = 1, so 2 ni=1 xi dxi = 0 ⇒ ni=1 dxi = 0 ⇒ dxn = −(dx1 + . . . +
P P
dxn−1 ).
n−1
2 8 h X X 2 i
Thus d F = n−1 dxi dxj − dxi . Clearly d2 F < 0, so we have a
n 1≤i<j≤n−1 i=1
maximum when x21 = x22 = . . . = x2n = n1 .

3
Derivation of inequality:
 1 n  x2 + x2 + . . . + x2  n
1 2 n
x21 x22 . . . x2n ≤ = as x21 + x22 + . . . + x2n = 1
n n
2 2 2
1 x + x2 + . . . + xn
⇒ (x21 x22 . . . x2n ) n ≤ 1
n
Let x2i = ai , 1 ≤ i ≤ n, to get
1 a1 + . . . + an
(a1 a2 . . . an ) n ≤
n
as required.
x 2 1 2 2
e−x (t +1)
Z Z
−t2
Question 4(a) Define f (x) = e dt , g(x) = 2
dt. Show that g 0 (x) +
0 0 tZ + 1 √
x
0 π −t2 π
f (x) = 0 for all x. Deduce that (i) g(x) + f (x) = 4 (ii) lim e dt = .
x→∞ 0 2
Solution.
Z x 
0 −t2
2
f (x) = 2 e dt e−x
0
1 2 2
∂ e−x (t +1)
Z  
0
g (x) = dt
0 ∂x t2 + 1
1 2 (t2 +1)
e−x (−2x(t2 + 1))
Z
= dt
0 t2 + 1
2 Rx 2
Let xt = y in the second equation, then dy = x dt, g 0 (x) = −2e−x 0
e−y dy = −f 0 (x),
showing that f 0 (x) + g 0 (x) = 0 for all x.
R1
1. f 0 (x) + g 0 (x) = 0 ⇒ f (x) + g(x) = C, a constant. Clearly f (0) = 0, g(0) = dt
0 t2 +1
=
i1
tan−1 t = π4 . So C = f (0) + g(0) = π4 , thus g(x) + f (x) = π4 .
0
2 2
1
e−x t
Z 
2
2. lim g(x) = lim 2
dt e−x .
x→∞ x→∞ 0 t +1
Z 1 −x2 t2 Z 1
e dt π 2 2 2
But 2
dt ≤ 2
= , ∵ e−x t ≤ 1. Thus 0 ≤ g(x) ≤ π4 e−x ⇒
0 t +1 0 1+t 4
limx→∞ g(x) = 0.
Z x 2 Z x √
−t2 π −t2 π
Thus lim f (x) = lim e dt = =⇒ lim e dt = , as both sides are
x→∞ x→∞ 0 4 x→∞ 0 2
positive.

4
UPSC Civil Services Main 1980 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 16, 2010

Question 1(a) Let f be a real continuous function on the interval [a, b] which is differen-
tiable in the open interval (a, b). Prove that there exists at least one point x ∈ (a, b) at which
f (b) − f (a) = (b − a)f 0 (x).
Solution. (This is the mean value theorem of calculus.)
Consider the function φ(x) = f (x)−f (a)−(x−a)A, where A is a constant so determined
f (b) − f (a)
that φ(b) = 0 ⇒ A = . Thus φ(b) = φ(a) = 0, φ(x) is continuous in [a, b] as
b−a
it is the sum of functions continuous over [a, b], and φ(x) is differentiable over (a, b) as
f (x), f (a), (x − a)A are all differentiable over (a, b). Thus φ(x) satisfies the conditions of
Rolle’s theorem, so there is at least one point c ∈ (a, b) such that φ0 (c) = 0 i.e. f 0 (c) − A =
f (b) − f (a)
0 ⇒ f 0 (c) = , thus f (b) − f (a) = (b − a)f 0 (c).
b−a
Question 1(b) Find the volume of the solid of revolution formed by rotating the area cut
off from y 2 = 4ax by the line y = x through 2π radians about the x-axis.
Solution. The curve y 2 = 4ax intersects y = x at (0, 0) and (4a, 4a). The required volume
V = Volume of rotating y 2 = 4ax about the x-axis - Volume of rotating y = x about the
x-axis. Thus
Z 4a Z 4a
2
V = πy1 dx − πy22 dx
Z0 4a 0
Z 4a
= π4ax dx − πx2 dx
0 0
2 4a 3 4a
 
x x
= 4aπ −π
2 0 3 0
2
16a 64a3 32πa3
= 4aπ −π =
2 3 3

1
Question 2(a) A rectangular box open at the top is to have a capacity of 4m3 . Find the
dimensions of the box requiring least material for its construction.

Solution. Let x be the length, y be the breadth, and z the height of the box. Then the
capacity of the box is xyz = 4. The surface area of the box is S = 2xz + 2yz + xy =
4
xy + 2(x + y) xy = xy + x8 + y8 . For extreme values

∂S 8 ∂S 8
= y − 2 = 0, =x− 2 =0
∂x x ∂y y
x4
Thus y = x82 , substituting in the second equation we get x = 8
⇒ x = 0, 2. Since x = 0 is
inadmissible, x = 2 ⇒ y = 2, z = 1.

∂ 2S 16 ∂ 2 S 16 ∂ 2 S
= , = , =1
∂x2 x3 ∂y 2 y 3 ∂x ∂y
 2 2
∂ 2S ∂ 2S ∂ S ∂ 2S ∂ 2S
Thus − = 4 − 1 > 0 when x = 2, y = 2, and = 2 > 0, = 2 > 0.
∂x2 ∂y 2 ∂x ∂y ∂x2 ∂y 2
So x = 2, y = 2 is a minimum for S.
Thus the dimensions of the required box are 2 meters × 2 meters × 1 meter.

2
UPSC Civil Services Main 1981 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 16, 2010

|x|
Question 1(a) Let f (x) = . Discuss the value of limx→0 f (x) and f (0). Justify your
x
answer. Discuss the differentiability of f and f 2 at x = 0.

Solution. Clearly (
1, x>0
f (x) =
−1, x < 0
Thus limx→0+ f (x) = 1, limx→0− f (x) = −1, showing that limx→0 f (x) does not exist. Thus
no matter what value is given to f (x) at 0, f (x) cannot be continuous at x = 0.
Since f (x) is not continuous at x = 0, the question of its differentiability does not arise., as
f (x0 + h) − f (x0 )
differentiability implies continuity — note that f (x0 + h) − f (x0 ) = h, so if
h
f (x0 + h) − f (x0 )
f (x) is differentiable at x0 , then limh→0 f (x0 +h)−f (x0 ) = limh→0 limh→0 h =
0
h
f (x0 ) · 0 = 0, so limh→0 f (x0 + h) = f (x0 ).
Clearly f 2 (x) = 1 for every x 6= 0, therefore f 2 would become differentiable at x = 0 if
we define f (x) = 1 or −1, making f 2 (x) = 1 for all x. In any other case f 2 (x) would not be
differentiable at x = 0, as it would not be continuous at x = 0.

x
Question 2(a) Suppose a manufacturer can sell x items at a price P = 200 − paisa
100
per item, and it costs y = 50x + 20000 paise to produce the x items. What is the production
level for maximum profits, and the selling price per item?
 x 
Solution. Selling price of x items = 200 − x
100
Cost price of x items = 50x + 20000.

1
 x 
Profict P on the sale of x items = 200 − x − (50x + 20000).
100
For extreme values,
dP x x x
= 200 − − − 50 = 150 − = 0 =⇒ x = 7500
dx 100 100 50
d2 P 1
2
=− <0
dx 50
showing that the profit is maximum when 7500 items are sold at 125 paisa per item.
Z ∞Z ∞
2 2
Question 2(b) Evaluate the double integral e−(x +y ) dx dy.
0 0

Solution. See 2002 question 1(b).


1
Question 2(c) State the mean value theorem. If f 0 (x) = for all x, and f (0) = 0,
1 + x2
show that 0.4 < f (2) < 2.0.
Solution. If f (x) is a real valued function defined on the closed interval [a, b], such that (i)
f (x) is continuous on the closed interval [a, b], (ii) f (x) is differentiable on the open interval
(a, b), then there exists a point c ∈ (a, b) such that f (b) − f (a) = (b − a)f 0 (c).
We apply the mean value theorem to f (x) on [0, 2] and get f (2) = f (2) − f (0) =
(2 − 0)f 0 (c) = 1+c
2
2 for some c ∈ (0, 2).
2 2 2
Now 0 < c < 2 so < 2
< , thus 0.4 < f (2) < 2.
1+4 1+c 1+0

Question 3(a) The region bounded by the curves y = x3 and y = x is revolved about the
x-axis. Find the volume generated.
Solution. The curves intersect each other at (0, 0 and (1, 1). The required volume V =
Volume obtained by rotating the arc from (0, 0) to (1, 1) of y 2 = x about the x-axis - Volume
obtained by rotating the arc from (0, 0) to (1, 1) of y = x3 about the x-axis. Thus
Z 1 Z 1
2
V = πy1 dx − πy22 dx
0 0
Z 1 Z 1
= π x dx − π x6 dx
0 0
π π 5π
= − =
2 7 14

Γ(p)Γ(q)
Question 3(b) Show that B(p, q) = .
Γ(p + q)
Solution. See 1991 question 2(c).

2
UPSC Civil Services Main 1982 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 16, 2010

Question 1(a) A function f (x) is defined for x > 0 as follows:


(
0, x irrational
f (x) = 2
q2
, x rational, x = pq in lowest terms

Show that f (x) has a removable discontinuity at every rational point, but is continuous for
irrational values.

Solution.

1. f (x) is continuous at x = c, c irrational — Let  > 0 be given. Since the number


of positive integers q for which q 2 ≤ 2 is finite, we can find a positive number δ > 0
such that (c − δ, c + δ) does not contain any rational number pq for which q 2 ≤ 2 , i.e.
x ∈ (c − δ, c + δ), x = pq , (p, q) = 1 ⇒ q 2 > 2 ⇒ q22 < . Thus for x ∈ (c − δ, c + δ),

• If x is irrational, |f (x) − f (c)| = 0 < 


• If x is rational, x = pq , (p, q) = 1, then |f (x) − f (c)| = | q22 − 0| < .

Thus f is continuous at all irrational points.

2. f (x) is discontinuous at x = c = pq , where (p, q) = 1. Let  > 0,  < q22 , then


whatever δ > 0 we take, we can find an irrational number x ∈ (c − δ, c + δ), and
|f (x) − f (c)| = |0 − q22 | > . Thus f (x) is not continuous at any rational point.

3. To show that the discontinuity is removable at x = c, c ∈ Q, we need to show that


limx→c f (x) exists, because if limx→c f (x) = l, we can define f (c) = l, and make f
continuous at c.

1

Let  > 0, and let δ > 0 be chosen so that 0 < δ < . Let x ∈ R, 0 < |x − pq | < δ.
2q
Then
|f (x) − 0| = |0 − 0| <  if x 6∈ Q
m p
If x ∈ Q, let x = , x 6= , then
n q
p m p qm − pn  qm − pn 
x − = − = < ⇒ <

q n q qn 2q n 2
2 qm − pn  m p
Now |f (x) − 0| = | n22 − 0| ≤ < because |qm − pn| ≥ 1 as 6= .
n n n n q

Hence we have shown that 0 < |x − pq | < δ with δ < ⇒ |f (x) − 0| < , i.e.
2q
limx→ pq f (x) = 0, showing that f (x) has a removable discontinuity at all rational
points.

Question 1(b) If f (x) is differentiable three times in (a − h, a + h), prove that there is a
point c in the interval such that

f (a + h) + f (a − h) h2
− f 0 (a) = f 000 (c)
2h 6
Solution. In order to prove this we prove the following particular case.
If f (x) is thrice differentiable and f (a) = 0, f (a − h) = 0, f (a + h) = 0, f 0 (a) = 0 then
there exists a point c ∈ (a − h, a + h) such that f 000 (c) = 0.

1. f (a − h) = f (a) = 0 and f 0 (x) exists, therefore Rolle’s theorem tells us that there
exists ξ1 ∈ (a − h, a) such that f 0 (ξ1 ) = 0.

2. f (a + h) = f (a) = 0 and f 0 (x) exists, therefore Rolle’s theorem tells us that there
exists ξ2 ∈ (a, a + h) such that f 0 (ξ2 ) = 0.

3. f 0 (ξ1 ) = f 0 (a) = 0 and f 00 (x) exists in [ξ1 , a], Rolle’s theorem gives a point η1 ∈ (ξ1 , a)
such that f 00 (η1 ) = 0.

4. f 0 (a) = f 0 (ξ2 ) = 0 and f 00 (x) exists in [a, ξ2 ], Rolle’s theorem gives a point η2 ∈ (a, ξ2 )
such that f 00 (η2 ) = 0.

5. Finally f 00 (η1 ) = f 00 (η2 ) = 0 and f 000 (x) exists in [η1 , η2 ], then Rolle’s theorem gives a
point c ∈ (η1 , η2 ) such that f 000 (c) = 0. Clearly, c ∈ (η1 , η2 ) ⊆ (a − h, a + h).

2
Now to prove the main result, let
Q(x) = f (a) + (x − a)f 0 (a) + B(x − a)2 + A(x − a)3
where A, B are so determined that Q(a + h) = f (a + h), Q(a − h) = f (a − h). Thus
f (a + h) − Q(a + h) = 0 ⇒ f (a + h) − f (a) − hf 0 (a) − h2 B − h3 A = 0 (1)
f (a − h) − Q(a − h) = 0 ⇒ f (a − h) − f (a) + hf 0 (a) − h2 B + h3 A = 0 (2)
Subtracting (2) from (1) we get
f (a + h) − f (a − h) − 2hf 0 (a) − 2h3 A = 0 (3)
We now take φ(x) = f (x) − Q(x), then φ(a) = 0, φ0 (a) = 0, φ(a − h) = 0, φ(a + h) = 0,
so by the result proved above, we get a point c ∈ (a − h, a + h) such that φ000 (c) = 0. But
000
φ000 (c) = f 000 (c) − Q000 (c) = f 000 (c) − 6A ⇒ A = f 6(c) . Substituting in (3) we have
h3 000
f (a + h) − f (a − h) − 2hf 0 (a) = f (c)
3
or
f (a + h) + f (a − h) h2
− f 0 (a) = f 000 (c)
2h 6
as required.
Note: In such questions the general tendency is to take φ(t) = f (a + th) − f (a − th), 0 ≤
t ≤ 1 and apply Taylor’s theorem to φ(t) in [0, 1]. We get
1 00
φ(1) = φ(0) + φ0 (0) + φ (θ), 0 < θ < 1
2!
φ(1) = f (a + h) − f (a − h)
φ(0) = 0
φ0 (0) = h(f 0 (a) − (−h)f 0 (a) = 2hf 0 (a)
φ00 (0) = h2 f 00 (a) − h2 f 00 (a) = 0
φ000 (θ) = h3 f 000 (a + θh) − (−h3 )f 000 (a − θh) = h3 [f 000 (a + θh) + f 000 (a − θh)]
Thus
f (a + h) + f (a − h) h2
− f 0 (a) = [f 000 (a + θh) + f 000 (a − θh)]
2h 6
i.e. we get two points in (a − h, a + h) and not one, so such a solution does not work in our
case.
Question 1(c) The arc of the parabola y 2 = 4ax from the vertex to one extremity of the
latus rectum is√revolved about the corresponding chord. Prove that the volume of the spindle
2 5 3
so formed is πa .
75
Solution.

3
Arc OL joining (0, 0) to (a, 2a) is rotated
about the chord OL. Let P be the point L
(x, y), M the foot of the the perpendicular on (a, 2a)
|y − 2x| P
OL i.e. y = 2x. Therefore P M = √ .
5 M
y 2 − 4xy + 4x2 O
OM 2 = OP 2 − P M 2 = x2 + y 2 −
5
4y 2 + 4xy + x2
=
5 √
2y + x 4 ax + x y 2 = 4ax
OM = √ = √
5 5
The required volume is V , given by
Z a
V = π(P M )2 d(OM )
0

y 2 − 4xy + 4x2 4ax − 4ax · 4x + 4x2
(P M )2 = =

5 5
a √ √
2 √x + 1 2 a+ x
d(OM ) = √ dx = √ dx
5 5x
Z a √ √
4  √ 2 a + x
⇒V = π x a − 2 ax + x √ dx
0 5 5x
Z a
4π √ √ √ 3
= √ x2 + ax − 2x ax + 2a ax − 4ax + 2 ax 2 dx
5 5 0
a
4π x3 x2

3 2 3
= √ − 3a + 2a 2 x 2
5 5 3 2 3 0
√  3 3 3
 √
4 5π a 3a 4a 2 5 3
= − + = πa
25 3 2 3 75
as required.

Question 2(a) If x cos u + y sin u = 1 and v = x sin u − y cos u, prove that


∂ 2u ∂u ∂u
v2 +v = cos 2u
∂x ∂y ∂x ∂y
Solution. Differentiate x cos u + y sin u = 1 with respect to x to get
∂u ∂u
cos u − x sin u + y cos u =0
∂x ∂x
or
∂u ∂u ∂u
cos u − (x sin u − y cos u) = cos u − v =0 ⇒ v = cos u (1)
∂x ∂x ∂x

4
Differentiate x cos u + y sin u = 1 with respect to y to get
∂u ∂u
−x sin u + sin u + y cos u =0
∂y ∂y
or
∂u ∂u ∂u
sin u − (x sin u − y cos u) = sin u − v =0 ⇒ v = sin u (2)
∂y ∂y ∂y
Differentiate (1) with respect to y to get

∂u ∂ 2u ∂u ∂v
− sin u =v +
∂y ∂x ∂y ∂x ∂y
But
∂v ∂u ∂u ∂u
= x cos u − cos u + y sin u = − cos u
∂y ∂y ∂y ∂y
as x cos u + y sin u = 1. Therefore

∂u ∂ 2u ∂u ∂u ∂u
− sin u =v + − cos u
∂y ∂x ∂y ∂x ∂y ∂x
Multiplying by v, and substituting (1), (2), we get

∂ 2u ∂u ∂u
− sin u(sin u) = v 2 +v − cos u(cos u)
∂x ∂y ∂x ∂y
or
∂ 2u ∂u ∂u
v2 +v = cos2 u − sin2 u = cos 2u
∂x ∂y ∂x ∂y
as required.

Question 2(b) Prove that

√  1  √   1  2
1
π2 Γ 3= 3 Γ
6 3

Solution. We know that


π
Z 1 Z
Γ(m)Γ(n) m−1 n−1
2
= B(m, n) = x (1 − x) dx = 2 sin2m−1 θ cos2n−1 θ dθ
Γ(m + n) 0 0

on putting x = sin2 θ.
1
1. We take n = 2
to get
π
Γ( 12 )Γ(m)
Z
2
=2 sin2m−1 θ dθ
Γ(m + 21 ) 0

5
2. We take m = n to get
π π
π
[Γ(n)]2
Z Z Z
2
2n−1 1 2
2n−1 1
=2 (sin θ cos θ) dθ = (sin 2θ) dθ = (sin φ)2n−1 dφ
Γ(2n) 0 22n−2 0 22n−1 0

by putting 2θ = φ. But
π
Z π Z
2
2n−1
(sin φ) dφ = 2 (sin φ)2n−1 dφ
0 0

so π
[Γ(n)]2
Z
1 2
= 2n−2 (sin θ)2n−1 dθ
Γ(2n) 2 0

Using both these, we have

Γ( 12 )Γ(n) 22n−1 [Γ(n)]2


=
Γ(n + 12 ) Γ(2n)
or √
1  Γ( 12 )Γ(2n) π
Γ(n)Γ n + = 2n−1
= 2n−1 Γ(2n)
2 2 2
1
Put n = 3
to get
√ 1 2
 2 2
 1


1 π 23 Γ 3 1 1 Γ
3
Γ 3
Γ = =⇒ Γ = π 23
3 Γ( 56 ) 3 Γ( 56 )

π 2
 1
 2π 5
 1

Now Γ(n)Γ(1 − n) = , so Γ 3
Γ 3
= √ and Γ 6
Γ 6
= 2π. Thus we get
sin nπ 3

Γ( 61 )
2 √


1 1 3
Γ = π 23
3 2π
so   2
√ 1 √ 1 1
3 Γ = π 23 Γ
3 6
as required.

Question 2(c) Find a point within a triangle such that the sum of its distances from the
angular points may be a minimum.

Solution.

6
Let the point be P = (x, y), and the A
vertices of the triangle ABC be (xi , yi ), i =
1,
P2, 3.p Then u = P A + P B + P C =
3 2 2
r1
i=1 (x −
pxi ) + (y − yi ) = r1 + r2 + r3 , a c
where ri = (x − xi )2 + (y − yi )2 , i = 1, 2, 3. P γ
For extreme values α
∂u
3
x − xi β
X r2 r3
= =0
∂x ri
i=1 B b C
3
∂u X y − yi
= =0
∂y i=1
ri

Thus
x − x1 x − x2 x − x3 y − y1 y − y2 y − y3
+ =− , + =−
r1 r2 r3 r1 r2 r3
Squaring and adding, we get
2 
2+ (x − x1 )(x − x2 ) + (y − y1 )(y − y2 ) = 1
r1 r2
or
1  1
(x − x1 )(x − x2 ) + (y − y1 )(y − y2 ) = −
r1 r2 2
Now by the cosine formula, r12 + r22 − 2r1 r2 cos α = a2 or

(x − x1 )2 + (y − y1 )2 + (x − x2 )2 + (y − y2 )2 − 2r1 r2 cos α = (x1 − x2 )2 + (y1 − y2 )2

Thus
1
cos α = (2x2 − 2xx1 − 2xx2 + 2x1 x2 + 2y 2 − 2yy1 − 2yy2 + 2y1 y2 )
2r1 r2
1  1
= (x − x1 )(x − x2 ) + (y − y1 )(y − y2 ) = −
r1 r2 2

Thus α = 2π 3
. By symmetry, β = γ = 2π 3
, i.e. the sides of the triangle subtend the same
angle at P .
Note that this solution is inadmissible when any angle of the triangle is > 2π3
. In this case,
the partial derivatives of u do not vanish anywhere inside the triangle, hence the minimum
must lie at one of the vertices. A simple check shows that the minimum is at the vertex
whose angle is > 2π3
.

7
3 3
∂ 2u X ∂ 2 ri X ∂ x − xi
= =
∂x2 i=1
∂x2 i=1
∂x ri
3  
X 1 x − xi x − xi
= −
i=1
r i ri2 ri
3 3
(x − xi )2 (y − yi )2
  X
X 1
= 1− 2
= 3
>0
i=1
ri ri i=1
r i

∂2u
Similarly ∂y 2
> 0.

3 3 3 3
∂ 2u X ∂ x − xi X x − xi ∂ri X x − xi y − y i X (x − xi )(y − yi )
= =− 2
=− 2
=−
∂x ∂y i=1
∂y ri i=1
ri ∂y i=1
ri ri i=1
ri3
 2 2
∂ 2u ∂ 2u ∂ u
Now we need to check that 2 2
− > 0. To make the calculations easier,
∂x ∂y ∂x ∂y
let x − xi = ri cos αi , y − yi = ri sin αi . Then
2 3 3 3 2
∂ 2u ∂ 2u ∂ 2u sin2 αi cos2 αi
 X X  X
sin αi cos αi
− = −
∂x2 ∂y 2 ∂x ∂y i=1
ri i=1
ri i=1
ri
1
= [sin2 α1 cos2 α2 + cos2 α1 sin2 α2 − 2 sin α1 cos α2 cos α1 sin α2 ] + . . .
r1 r2
1 1 1
= [sin α1 cos α2 − cos α1 sin α2 ]2 + [. . .]2 + [. . .]2 > 0
r1 r2 r1 r3 r2 r3
Thus u = P A + P B + P C is at a minimum when P is the point at which the sides subtend
the same angle.
Note: This is called the Steiner’s problem, and the point P is called the Fermat Point or
the Fermat-Torricelli Point of the triangle.

Paper II

Question 3(a) Find the extreme values of f (x1 , x2 , x3 ) = x21 +x22 +x23 subject to 3i,j=1 aij xi xj =
P
1, where aij = aji .

Solution. Let F (x1 , x2 , x3 ) = x21 + x22 + x23 − λ 3i,j=1 aij xi xj − 1 . For extreme values,
P 

3
∂F X
= 2xi − 2λ aij xj = 0, 1 ≤ i ≤ 3 (∗)
∂xi j=1

8
3 3
X ∂F X
This implies 0 = xi = 2f − 2λ aij xi xj = 2f − 2λ ⇒ f = λ at the extreme values.
i=1
∂xi i,j=1
Now the equations (*) are a set of linear equations, which can be written as 2(I−λA)x =
0, where I is a 3 × 3 identity matrix,
P3 and A = (aij ) is a 3 × 3 symmetric matrix, so has
real eigenvalues. Since x 6= 0 as i,j=1 aij xi xj = 1, I − λA must be singular ⇒ λ 6= 0, so
A − λ−1 I is singular. Thus λ−1 is an eigenvalue of A.
Hence the extreme values of f are λ−1 , where λ is an eigenvalue of A = (aij ).

9
UPSC Civil Services Main 1983 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 16, 2010

Question 1(a) Consider the function

log(2 + x) − x2n sin x


f (x) = lim
n→∞ 1 + x2n
in the interval 0 ≤ x ≤ π2 . Show that the function does not vanish anywhere in this interval
and explain why is so although f (0), f ( π2 ) differ in sign.

Solution. Since x2n → 0 as n → ∞ for 0 ≤ x < 1, it follows that


log 3 − sin 1
f (x) = log(2 + x), 0 ≤ x < 1, f (1) =
2
1 log(2+x)
For x > 1, x2n
→ 0 and x2n
→ 0 as n → ∞, therefore
log(2+x)
x2n
− sin x
f (x) = lim 1 = − sin x
n→∞
x2n
+1

Thus 
log(2 + x), 0 ≤ x < 1

f (x) = log 3−sin
2
1)
, x=1

− sin x, 1 < x ≤ π2

Clearly f (0) = log 2 > 0, f ( π2 ) = − sin π2 = −1 < 0, and f (x) 6= 0 for any x ∈ [0, π2 ].
The reason why this can happen is that f (x) is not continuous in the interval [0, π2 ], so
the intermediate value theorem does not apply. The function is discontinuous at x = 1 as
limx→1+ f (x) = − sin 1, limx→1− f (x) = log 3 and both of these are different from f (1).

1
Question 1(b) Prove that there is a point a ∈ (−h, h) such that
Z h
2h3 00
f (x) dx = h[f (h) + f (−h)] − f (a)
−h 3
Solution. Integrating by parts, we get
Z h ih Z h Z h
0
f (x) dx = xf (x) − xf (x) dx = h[f (h) + f (−h)] − xf 0 (x) dx
−h −h −h −h
R0
Now put x = −t in −h
xf 0 (x) dx to get
Z 0 Z 0 Z h
0 0
xf (x) dx = (−t)f (−t)(−dt) = − tf 0 (−t) dt
−h h 0

Thus Z h Z h
x f 0 (x) − f 0 (−x) dx

f (x) dx = h[f (h) + f (−h)] −
−h 0
0 0
f (x) − f (−x) f (x) + f 00 (−x)
00
f 0 (x) − f 0 (−x)
We now note that lim = lim = 2f 00 (0), showing that
x→0 x x→0 1 x
is Riemann integrable in [0, h]. Since x2 keeps the same sign, the first mean value theorem
of integral calculus gives us
Z h 0 0
2 f (x) − f (−x) f 0 (ξ) − f 0 (−ξ) h 2
Z
x dx = x dx
0 x ξ 0

for some ξ ∈ (0, h). Consequently,


Z h
f 0 (ξ) − f 0 (−ξ) 2h3
f (x) dx = h[f (h) + f (−h)] −
−h 2ξ 3
We now apply Lagrange’s mean value theorem to the function f 0 (x) in [−ξ, ξ] and get
f 0 (ξ) − f 0 (−ξ)
= f 00 (a) for some a ∈ (−ξ, ξ)

Thus
h
2h3 00
Z
f (x) dx = h[f (h) + f (−h)] − f (a)
−h 3
Note that 0 < ξ < h ⇒ a ∈ [−h, h].
Corollary: If f 000 (x) exists in (−h, h) prove that
f (h) − f (−h) f 0 (h) + f 0 (−h) h3 000
=h − f (a)
2 2 3
for some a ∈ (−h, h).
Proof: Use the above result for f 0 (x) instead of f (x). The left hand side now becomes
f (h) − f (−h), and dividing by two, we get the above equation.

2
Question 1(c) Find the least perimeter of an isoceles triangle in which a circle of radius r
can be inscribed.

Solution.

Let α be the semi-vertical angle. D is the A


midpoint of BC. E is the point of contact α
of AC and the circle or radius r inscribed
OE
in ABC. Therefore = tan α, or AE =
AE F E
OE cot α = r cot α. Similarly, we find out r r
that OA = r csc α, and AD = AO + OD = O
r(1 + csc α), consequently BD = AD tan α =
r(1 + csc α) tan α. Since 4ABC is isoceles,
BD = DC, DC = CE and AE = AF . B D C

The perimeter

P = BD + DC + CE + AE + BF + AF
= 4BD + 2AE = 4r(1 + csc α) tan α + 2r cot α
= 4r tan α + 4r sec α + 2r cot α
dP
= 4r sec2 α + 4r sec α tan α − 2r csc2 α = 0

2 2 sin α 1
⇒0 = + −
cos2 α cos2 α sin2 α
⇒ 0 = 2 sin α + 2 sin α − cos2 α
2 3

⇒ 0 = 2 sin3 α + 3 sin2 α − 1
⇒ 0 = (sin α + 1)2 (2 sin α − 1)

Thus dP

= 0 ⇒ sin α = 21 , since sin α = −1 is not possible as 2α is the angle of a triangle.
Thus α = π6 .

d2 P
= 8r sec2 α tan α + 4r(sec α tan2 α + sec3 α) − 4r csc α(− csc α cot α) > 0
dα2
for α = π6 . Therefore we have a minimum when α = π6 . The required perimeter is 4r √13 +
√ √
4r √23 + 2r 3 = 6r 3.

Question 2(a) If f (0) = 0 and f 0 (x) = 1+x


1
2 , prove without using the method of integration

that  x+y 
f (x) + f (y) = f
1 + xy

3
x+y
Solution. Let u = f (x) + f (y), v = 1+xy
. Then

∂v (1 − xy)(1) − (x + y)(−y) 1 + y2
= =
∂x (1 − xy)2 (1 − xy)2
∂v (1 − xy)(1) − (x + y)(−x) 1 + x2
= =
∂y (1 − xy)2 (1 − xy)2

∂(u, v) 12 1
1+x 1+y 2
= 1+y2 1+x = 0
2
∂(x, y) (1−xy)2 (1−xy) 2

 x+y 
Therefore u, v are functionally dependent. Let u = φ(v) ⇒ f (x) + f (y) = φ .
1 − xy
Substituting y =  x0+gives us f (x) + f (0) = φ(x) or φ(x) = f (x) as f (0) = 0. Hence
y
f (x) + f (y) = f , which was to be proved.
1 − xy

Question 2(b) Show that under the transformation u = x2 − y 2 , v = 2xy the equation

∂ 2H 2
2∂ H ∂H ∂H
y2 2
− x 2
=x −y
∂x ∂y ∂x ∂y
 
∂ ∂ ∂H
becomes u −v = 0.
∂v ∂u ∂v

Solution.
∂H ∂H ∂u ∂H ∂v ∂H ∂H
= + = 2x + 2y
∂x ∂u ∂x ∂v 2∂x ∂u  ∂v  2
∂ 2H 2
∂ 2H

∂H ∂ H ∂ H ∂ H
= 2 + 2x 2x + 2y + 2y 2x + (2y)
∂x2 ∂u ∂u2 ∂u ∂v ∂v ∂u ∂v 2
∂H ∂ 2H ∂ 2H ∂ 2H
= 2 + 4x2 2 + 8xy + 4y 2 2 (1)
∂u ∂u ∂u ∂v ∂v
∂H ∂H ∂u ∂H ∂v ∂H ∂H
= + = −2y + 2x
∂y ∂u ∂y ∂v ∂y ∂u ∂v
2
 2 2
 2
∂ 2H
 
∂ H ∂H ∂ H ∂ H ∂ H
= −2 − 2y (−2y) + 2x + 2x (−2y) + (2x)
∂y 2 ∂u ∂u2 ∂u ∂v ∂v ∂u ∂v 2
∂H ∂ 2H ∂ 2H ∂ 2H
= −2 + 4y 2 2 − 8xy + 4x2 2 (2)
∂u ∂u ∂u ∂v ∂v
∂H ∂H ∂H ∂H ∂H ∂H
x −y = 2x2 + 2xy + 2y 2 − 2xy
∂x ∂y ∂u ∂v ∂u ∂v
∂H
= 2(x2 + y 2 ) (3)
∂u

4
Using (1), (2), (3) we get
2 2
 
2∂ H 2∂ H ∂H ∂H
0 = y −x − x −y
∂x2 ∂y 2 ∂x ∂y
2 2 2
2 ∂H 2 2∂ H 3 ∂ H 4∂ H
= 2y + 4x y + 8xy + 4y
∂u ∂u2 ∂u ∂v ∂v 2
2 2
∂H ∂ H ∂ H ∂ 2H ∂H
+2x2 − 4x2 y 2 2 + 8x3 y − 4x4 2 − 2(x2 + y 2 )
∂u ∂u ∂u ∂v ∂v ∂u
2 2
∂ H ∂ H
= 8xy(x2 + y 2 ) + 4(y 4 − x4 ) 2
∂u ∂v ∂v
2 2
∂ H ∂ H
⇒0 = (y 2 − x2 ) 2 + 2xy
∂v ∂u ∂v
∂ 2H ∂ 2H
⇒0 = −u 2 + v
 ∂v ∂u
 ∂v
∂ ∂ ∂H
⇒0 = u −v
∂v ∂u ∂v
as required.

Question 2(c) Prove that the volume of the solid generated by the revolution of the tractrix
t
x = a cos t + a log tan , y = a sin t
2
about the asymptote is equal to half the volume of a sphere of radius a.
Solution. The x-axis is the asymptote, as t → 0 ⇒ x → ∞, y → 0. The volume required
is twice the volume in the first quadrant.

Z π
2 dx
V = 2 πy 2 dt
0 dt
Z π
2 a
= 2π a2 sin2 t(−a sin t + ) dt
0 sin t
Z π
2
3
= 2πa (sin t − sin3 t) dt
0
Z π
2
3
= 2πa sin t cos2 t dt
0

cos3 t 2 2πa3

3
= 2πa − =
3 0 3
3
1 4πa 1
= = × Volume of a sphere of radius a
2 3 2

5
Paper II

Question 3(a) Obtain a set of sufficient conditions such that for a function f (x, y)

∂ 2f ∂ 2f
=
∂x ∂y ∂y ∂x

Solution. Euler’s Theorem: If the partial derivatives fxy and fyx are continuous at (a, b)
then fxy (a, b) = fyx (a, b).
Proof: Let Ψ(h, k) = f (a + h, b + k) − f (a + h, b) − f (a, b + k) + f (a, b) where (a + h, b +
k), (a + h, b), (a, b + k) all belong to a neighborhood N of (a, b) — we can take for N an open
disc with center (a, b) in which fxy and fyx exist.
Let G(x) = f (x, b + k) − f (x, b) for x ∈ Ih where Ih = [a, a + h] or [a + h, a] according
as h > 0 or h < 0. Clearly G0 (x) = fx (x, b + k) − fx (x, b) for x ∈ Ih . We apply Lagrange’s
Mean Value Theorem to G(x) and obtain:

G(a+h)−G(a) = Ψ(h, k) = hG0 (a+θh) = h[fx (a+θh, b+k)−fx (a+θh, b)] (∗)

where 0 < θ < 1.


Now we consider F (t) = fx (a + θh, t) for t ∈ Ik where Ik = [b, b + k] or [b + k, b] according
as k > 0 or k < 0. We apply Lagrange’s Mean Value Theorem to F (t) and obtain:

F (b + k) − F (b) = fx (a + θh, b + k) − fx (a + θh, b) = kF 0 (b + θ1 k)

where 0 < θ1 < 1. But F 0 (t) = ∂


f (a
∂y x
+ θh, t), so

∂ 2f
F (b + k) − F (b) = k (a + θh, b + θ1 k)
∂y ∂x

Using (*) we get


∂ 2f
Ψ(h, k) = hk (a + θh, b + θ1 k)
∂y ∂x
Since fyx is continuous at (a, b), we get

Ψ(h, k) ∂ 2f ∂ 2f
lim = lim (a + θh, b + θ1 k) = (a, b)
h→0,k→0 hk h→0,k→0 ∂y ∂x ∂y ∂x

Now instead of G(x), we start with H(y) = f (a+h, y)−f (a, y) for y ∈ Ik , and proceeding
exactly as above, we get
Ψ(h, k) ∂ 2f
lim = (a, b)
h→0,k→0 hk ∂x ∂y
∂ 2f ∂ 2f
Hence (a, b) = (a, b), which completes the proof.
∂x ∂y ∂y ∂x

6
Question 3(b) Find the maximum and minimum values of x2 + y 2 + z 2 subject to the
conditions x + y + z = 1, xyz + 1 = 0.

Solution. Let F (x, y, z) = x2 + y 2 + z 2 + λ1 (x + y + z − 1) + λ2 (xyz + 1) where λ1 , λ2 are


Lagrange’s undetermined multipliers. The extreme values are given by
∂F
= 2x + λ1 + λ2 yz = 0
∂x
∂F
= 2y + λ1 + λ2 xz = 0
∂y
∂F
= 2z + λ1 + λ2 xy = 0
∂z
Subtracting the first two, 2(x − y) + λ2 z(y − x) = 0 ⇒ x = y or λ2 = z2 (Note that
x 6= 0, y 6= 0, z 6= 0 because xyz + 1 = 0). Similarly from the other pairs of equations, we
get y = z or λ2 = x2 , and x = z or λ2 = y2 .
Since xyz = −1, x + y + z = 1, it follows that x, y, z cannot be all positive or all negative,
moreover two must be positive and one negative. In particular, x = y = z is not possible.
Suppose x 6= y, then λ2 = z2 . Now ∂F ∂x
= 0 ⇒ 2x + λ1 + 2y = 0.
Substituting the values of λ1 , λ2 in ∂F
∂z
= 0, we get

2xy
2z − 2(x + y) + =0
z
But x + y = 1 − z, xy = − z1 , so we get 2z − 2(1 − z) − z22 = 0 ⇒ 2z 3 − z 2 − 1 = 0 ⇒
(z − 1)(2z 2 + z + 1) = 0. But 2z 2 + z + 1 has no real roots, so the only real root is z = 1.
Thus we get λ2 = 2, z = 1, λ1 = 0(∵ x + y + z = 1 ⇒ x + y = 0 ⇒ λ1 = 0), x = ±1, y =
∓1(∵ x + y = 0, xy = 1 ⇒ x2 = 1 ⇒ x = ±1, y = ∓1). Hence the stationary values are
(1, −1, 1), (−1, 1, 1) and f (x, y, z) = 3 at these points.
On taking y 6= z we shall get (1, 1, −1), (1, −1, 1) as stationary points (by symmetry),
and for x 6= z, we get (1, 1, −1), (−1, 1, 1). f (x, y, z) is always equal to 3, hence we cannot
say whether it is a maximum or minimum without checking d2 F .
Considering the point (−1, 1, 1),

d2 F = 2(dx)2 + 2(dy)2 + 2(dz)2 + 4 dx dy + 4 dx dz − 4 dy dz

Now x+y+z = −1 ⇒ dx+dy+dz = 0 ⇒ dz = −dx−dy. xyz = −1 ⇒ yz dx+zx dy+xy dz =


0 ⇒ dx − dy − dz = 0 ⇒ dx = 0, dz = −dy.
Thus d2 F = 8(dz)2 > 0, so f has a minimum at all these stationary points.
Note: The question can be treated as that of one variable as y, z can be eliminated, but
the calculation becomes quite messy.

7
UPSC Civil Services Main 1984 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 16, 2010

Question 1(a) 1. Show that the function


(
x2 sin2 x1 , x 6= 0
f (x) =
0, x=0

is continuous at 0.
π
2. tan x is not continuous at x = 2

Solution.

1. Given  > 0, let δ =√ , then |x| ≤ δ =⇒ |x2 sin2 x1 | ≤ |x2 | < , because | sin2 x1 | ≤ 1.
Thus |x − 0| < δ =  ⇒ |f (x) − f (0)| < , showing that f (x) is continuous at x = 0.
1
2. Since limx→ π2 sin x = 1, given  = 2
there exists δ1 > 0 such that 0 < |x − π2 | < δ1 ⇒
| sin x − 1| < 21 ⇒ sin x > 12 .
Since limx→ π2 cos x = 0, given any real number G > 0 there exists δ2 > 0 such that
0 < |x − π2 | < δ2 ⇒ | cos x| < 2G
1
.
Let δ = min(δ1 , δ2 ). Then δ > 0 and 0 < |x − π2 | < δ ⇒ | tan x| > 12 · 2G = G. This
shows that tan x is not bounded in any neighborhood of π2 , therefore limx→ π2 tan x does
not exist, so tan x is not continuous at x = π2 .

Note that if limx→a f (x) = l, then f (x) is bounded in a neighborhood of a, because given
 > 0, there exists δ > 0 such that 0 < |x − a| < δ ⇒ l −  < f (x) < l + .

1
Question 1(b) Find the volume of the torus generated by revolving a disc of radius r about
a line at a distance a > r from the center of the circle.

Solution. Let the line be the x-axis, and let the circle have center (0, a).√ The circle’s
equation is x2 + (y − a)2 = r2√
. The upper semicircle is given by f1 (x) = a + r2 − x2 , and
the lower one by f2 (x) = a − r2 − x2 , and the desired volume is given by
Z r
V = π (f12 (x) − f22 (x)) dx
Z−rr
= π (f1 (x) − f2 (x))(f1 (x) + f2 (x)) dx
−r
Z r √
= π (2 r2 − x2 )(2a) dx
−r
Z r√
= 8aπ r2 − x2 dx
0
Put x = r sin θ
Z π
2
= 8aπ r cos θ · r cos θ dθ
0
Z π
2
2
= 8ar π cos2 θ dθ
0
1 π 
= 8ar2 π = 2π 2 ar2
22
We could get the same result by applying Pappus’ Theorem — the volume of a solid of
revolution generated by rotating a plane figure about an external axis is equal to the product
of the area of the figure and the distance traveled by its geometric centroid during revolution.
Thus V = πr2 · 2aπ = 2π 2 ar2 .

Question 1(c) Let f (x, y) = (x + y) sin( x1 + y1 ), when x 6= 0, y 6= 0, and f (x, 0) =


f (0, y) = 0. Examine whether (i) f (x, y) is continuous, and (ii) limx→0 f (x, y) for y 6= 0 and
limy→0 f (x, y) for x 6= 0 exist.

Solution. (i) Given  > 0, let δ1 = δ2 = 2 . Then |x| < δ1 , |y| < δ2 ⇒ |f (x, y) − f (0, 0)| =
|(x + y) sin( x1 + y1 )| ≤ |x| + |y| < , as | sin( x1 + y1 )| ≤ 1. Thus f (x, y) is continuous at (0, 0).
(ii) Since |x sin( x1 + y1 )| ≤ |x|, it follows that limx→0 x sin( x1 + y1 ) = 0. Thus if limx→0 f (x, y)
exists for y 6= 0, then limx→0 y sin( x1 + y1 ) should also exist, for y 6= 0, because y sin( x1 + y1 ) =
(x + y) sin( x1 + y1 ) − x sin( x1 + y1 ).
Let g(x) = y sin( x1 + y1 ). Suppose limx→0 g(x) = l for y = π2 . Then given  > 0,  < π2 ,
there exists δ > 0 such that 0 < |x| < δ ⇒ |g(x) − l| < 2 . Let x1 = 2nπ 1 1
, x2 = (2n+1)π , n large

2
so that |x1 | < δ, |x2 | < δ. Now

|g(x1 ) − g(x2 )| = |g(x1 ) − l + l − g(x2 )|


≤ |g(x1 ) − l| + |g(x2 ) − l| < 
2 π 2
g(x1 ) = sin(2nπ + ) =
π 2 π
2 π 2
g(x2 ) = sin((2n + 1)π + ) = −
π 2 π
4
⇒ |g(x1 ) − g(x2 )| = >
π
Thus we have a contradiction. Hence limx→0 y sin( x1 + y1 ) does not exist for y 6= 0, so
limx→0 f (x, y) for y 6= 0 does not exist. Similarly it can be seen that limy→0 f (x, y) for x 6= 0
does not exist, by symmetry.
ZZ
Question 2(a) Evaluate xy dx dy over the area given by the boundary y = 0(0 ≤ x ≤
3); y = (x − 3)2 (2 ≤ x ≤ 3); y = 1(1 ≤ x ≤ 2); y = x(0 ≤ x ≤ 1).

Solution.

The region of integration consists of three


y = (x − 3)2
parts
y=x
1. The triangle 4 bounded by
y = 0, x = 1, y = x.
2. The square  bounded by
x = 1, x = 2, y = 0, y = 1. 4  R
3. The region R bounded by O x=1
x = 2, y = 0, y = (x − 3)2 , (2 ≤ x ≤ 3).

1 x 1
y 2 ix 1 1 3
ZZ Z Z Z Z
1
I1 = xy dx dy = xy dy dx = x dx = x dx =
4 x=0 y=0 0 2 0 2 0 8
Z 2 Z 1
x2 i 2 y 2 i 1 3 1
ZZ
3
I2 = xy dx dy = xy dy dx = · = · =
 x=1 y=0 2 1 2 0 2 2 4
Z 3 Z (x−3)2 Z 3
y 2 i(x−3)2
ZZ
I3 = xy dx dy = yx dy dy = x dx
R x=2 0 x=2 2 0
1 3 1 0 4 1 h u6 u5 i0
Z Z
4 1  3 1  13
= x(x − 3) dx = u (u + 3) du = +3 = − =
2 2 2 −1 2 6 5 −1 2 5 6 60
1 3 13 131
Thus the given integral = I1 + I2 + I3 = + + =
8 4 60 120

3
Question 2(b) If B(p, q) is the Beta function, show that
pB(p, q) = (q − 1)B(p + 1, q − 1)
where p, q are real, p > 0, q > 1. Hence or otherwise find B(p, n) where n is an integer > 0.
Z 1
Solution. By definition, B(p, q) = xp−1 (1 − x)q−1 dx. Integrating by parts, we get
0
p
1 1
xp
Z
x
B(p, q) = (1 − x)q−1 + (q − 1)(1 − x)q−2 dx
p 0 0 p
Since p > 0, (q − 1) > 0, we get
Z 1
pB(p, q) = (q − 1) xp+1−1 (1 − x)(q−1)−1 dx = (q − 1)B(p + 1, q − 1)
0

In particular, pB(p, n) = (n − 1)B(p + 1, n − 1). Repeating this formula, we get


n−1
B(p, n) = B(p + 1, n − 1)
p
(n − 1)(n − 2)
= B(p + 2, n − 2)
p(p + 1)
(n − 1)!
= B(p + n − 1, 1)
p(p + 1) . . . (p + n − 1)
Z 1
(n − 1)!
= xp+n−2 (1 − x)1−1 dx
p(p + 1) . . . (p + n − 2) 0
(n − 1)! Γ(p)Γ(n)
= =
p(p + 1) . . . (p + n − 1) Γ(p + n)

x+y ∂(u, v)
Question 2(c) If u = and v = tan−1 x + tan−1 y, find . Are u and v func-
1 − xy ∂(x, y)
tionally related? If so, find the relationship.
Solution.
∂u (1 − xy) · 1 − (x + y)(−y) 1 + y2
= =
∂x (1 − xy)2 (1 − xy)2
∂u (1 − xy) · 1 − (x + y)(−x) 1 + x2
= =
∂y (1 − xy)2 (1 − xy)2
∂v 1
=
∂x 1 + x2
∂v 1
=
∂y 1 + y2
∂u ∂u
∂(u, v) ∂x ∂y ∂u ∂v ∂u ∂v
= ∂x ∂y − ∂y ∂x = 0
∂v ∂v =
∂(x, y)
∂x ∂y

4
This shows that u, v are functionally related. Let x = tan θ, y = tan φ. Then v = θ + φ.
tan θ + tan φ
u= = tan(θ + φ) = tan v.
1 − tan θ tan φ
Alternatively, usingv = tan−1 x + tan−1 y we write
tan v − tan(tan−1 y) tan v − y
x = tan(v − tan−1 y) = −1
=
1 + tan v tan(tan y) 1 + y tan v
tan v−y
1+y tan v
+y (tan v)(1 + y 2 )
u = tan v−y = = tan v
1 − y 1+y tan v
1 + y2

as before.

Paper II

Question 3(a) Show that the maximum and minimum values of the function u = x2 +y 2 +xy
where ax2 + by 2 = ab, a > b > 0 are given by 4(u − a)(u − b) = ab.

Solution. Let F (x, y) = x2 +y 2 +xy +λ(ax2 +by 2 −ab) where λ is Lagrange’s undetermined
multiplier. The extreme values are obtained from
∂F ∂F
= 2x + y + 2λax = 0, = 2y + x + 2λby = 0
∂x ∂y
∂F ∂F
0 = x +y = 2(x2 + y 2 + xy) + 2λ(ax2 + by 2 ) = 2u + 2λab
∂x ∂y
u
Thus λ = − ab . Consequently 2x + y − 2ux b
= 0 ⇒ 2x(b − u) + yb = 0, and 2y + x − 2uy a
=
0 ⇒ 2y(a − u) + ax = 0.
Since ax2 + by 2 = ab > 0, (x, y) 6= (0, 0), so the coefficient matrix of the above linear
2(b − u) b
equations must be singular i.e. = 0 or 4(a − u)(b − u) − ab = 0.
a 2(a − u)
Thus the maximum and minimum √ values are √ given by 4(a − u)(b − u) − ab = 0.
Note: We can substitute x = b cos t, y = a sin t to make u a function of one variable,
and proceed accordingly.

Question 3(b) Discuss the continuity and differentiability of the function


(
x2 tan−1 xy − y 2 tan−1 xy , x 6= 0, y 6= 0
f (x, y) =
0, x=y=0

Also examine if fxy and fyx are equal at (0, 0).

Solution.
1. The function is continuous at (0, 0), because |f (x, y) − f (0, 0)| = 0 or ≤ π2 (x2 + y 2 ).

5
2. fxy (0, 0) 6= fyx (0, 0). For this we calculate the following:

(a) If (x, y) 6= (0, 0),


 
−1 y 2 1 y 1 1
fx (x, y) = 2x tan +x y 2 − 2 − y2 2
x 1+ 2 x
x 1 + xy2 y
y x2 y y3 y
= 2x tan−1 − 2 2
− 2 2
= 2x tan−1 − y
x x +y x +y x

(b)
f (h, k) − f (0, k) h2 tan−1 k
h
− k 2 tan−1 h
k
fx (0, k) = lim = lim
h→0 h h→0 h
h
tan−1
Now limh→0 h tan−1 k
h
= 0, limh→0 h
k
= 1
k
— if θ = tan−1 hk , hk = tan θ, then
h
tan−1 θ
h
= k tan
k
θ
, and tanθ θ → 1 as θ → 0.
Thus fx (0, k) = −k, in particular fx (0, 0) = 0.
(c) By symmetry, fy (x, y) = x − 2y tan−1 x
y
is x 6= 0, y 6= 0, and fy (h, 0) = h.

fx (0, k) − fx (0, 0) −k − 0
fyx (0, 0) = lim = lim = −1
k→0 k k→0 k
fy (h, 0) − fy (0, 0) h−0
fxy (0, 0) = lim = lim =1
h→0 h h→0 h

Thus fxy (0, 0) 6= fyx (0, 0).

3. The function is differentiable at (0, 0) as both fx (x, y) and fy (x, y) are continuous at
(0, 0). Note that limx→0,y→0 fx (x, y) = limx→0,y→0 fy (x, y) = 0 and fx (0, 0) = 0 =
fy (0, 0).

6
UPSC Civil Services Main 1985 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 16, 2010

Question 1(a) If (
xy
x2 +y 2
, (x, y) 6= (0, 0)
f (x, y) =
0, (x, y) = (0, 0)
Show that both the partial derivatives fx , fy exist at (0, 0), but the function is not continuous
there.

Solution. See 2004, question 2(d).

Question 1(b) If for all values of the parameter λ and some constant n, F (λx, λy) =
∂F ∂F
λn F (x, y) identically, where F is assumed to be differentiable, prove that x +y =
∂x ∂y
nF (x, y).

Solution. See 1996, question 2(a).

Question 1(c) Prove the relation between the beta and gamma functions

Γ(m)Γ(n)
β(m, n) =
Γ(m + n)

Solution. See 1991, question 2(c).

1
Question 2(a) If a function f defined on [a, b] is continuous on [a, b] and differentiable on
(a, b), and f (a) = f (b), then prove that there exists at least one real number c, a < c < b
such that f 0 (c) = 0.

Solution. This is Rolle’s theorem. Since f is continuous on [a, b], it is bounded


and attains its maximum and minimum. Since f (a) = f (b) and we assume that f is not a
constant function, it follows that there exists a real number c such that f (c) = M , maximum
of f (without loss of generality) and a < c < b. Then the right hand derivative of f at c is
f (c + h) − f (c)
lim+ and clearly it is ≤ 0, because f (c + h) − f (c) ≤ 0 and h > 0. However
h→0 h
f (c + h) − f (c)
the left hand derivative of f at c is lim− ≥ 0 because f (c + h) − f (c) ≤ 0 and
h→0 h
0
h < 0. Since f (c) exists, the right hand derivative of f at c equals the left hand derivative
of f at c, which is possible only when both are zero, i.e. f 0 (c) = 0.

Question 2(b) Use Maclaurin’s expansion to show that

x2 x3 x4
log(1 + x) = x − + − + ...
2 3 4
Hence find the value of log(1 + x + x2 + x3 + x4 ).

Solution. Since log(1 + x) possesses continuous derivatives of all orders for every value of
x for which 1 + x > 0 i.e. x > −1, we have

x2 00 xn−1 (n−1)
f (x) = log(1 + x) = f (0) + xf 0 (0) + f (0) + . . . + f (0) + Rn
2! (n − 1)!

where
xn (n) xn (n − 1)!
Rn = f (θx) = (−1)n−1 , 0<θ<1
n! n! (1 + θx)n
(−1)n−1 (n − 1)!
as f (n) (x) for f (x) = log(1 + x) is .
(x + 1)n
x 1
Let 0 < x ≤ 1, so that 0 < < 1, then |Rn | < n
and therefore Rn → 0 as n → ∞.
1 + θx
Let −1 < x ≤ 0, we consider the expansion

x2 00 xn−1 (n−1)
f (x) = log(1 + x) = f (0) + xf 0 (0) + f (0) + . . . + f (0) + Rn
2! (n − 1)!

where
n−1
xn

n−1 (n) n n 1−θ 1
Rn = (1 − θ) f (θx) = (−1) x , 0<θ<1
(n − 1)! 1 + θx 1 + θx

2
— this is Cauchy’s form of the remainder — note that we consider Cauchy’s form of the
remainder as Lagrange’s form of the remainder does not enable us to prove that Rn → 0 as
n → ∞. when −1 < x < 0.
1−θ 1 1
Since 0 < θ < 1 and |x| < 1, 0 < < 1. Also, < . Now since xn → 0
1 + θx 1 + θx 1 − |x|
as n → ∞, we get Rn → 0 as n → ∞.
Now using the fact that f (n) (0) = (−1)n−1 (n − 1)!, we get the required result
x2 x3 x4
log(1 + x) = x − + − + ...
2 3 4

1 − x5
log(1 + x + x2 + x3 + x4 ) = log( ) = log(1 − x5 ) − log(1 − x)
1−x
x10 x15 x2 x3 x4
= −(x5 + + + . . .) + (x + + + + . . .)
2 3 2 3 4
x2 x3 x4 1  x6 x7 x8 x9 x10  1 
= x+ + + + x5 −1 + + + + + − 1 + ...
2 3 4 5 6 7 8 9 2 5
which is the required expansion.

Question 2(c) Find the volume generated by revolving y 2 = 4ax about the latus rectum.

Solution.

Let P be the point (x, y), M be the foot of


the perpendicular on the latus rectum x = a. P M
Then P M = a − x. SM = y, where S is
the focus (a, 0). Because of symmetry we S (a, 0)
can confine ourselves to the first quadrant, x=a
0 ≤ y ≤ 2a.
y 2 = 4ax

Z Z 2a
2
V = 2 π(P M ) d(SM ) = 2π (a2 − 2ax + x2 ) dy
0
2a 
y2 y4 
Z
= 2π + a2 − dy
0 2 16a2
2a
y3 y5 8a3 32a5
  
2 3
= 2π a y − + = 2π 2a − +
6 80a2 0 6 80a2
30 − 20 + 6 32πa3
= 2πa3 =
15 15

3
Paper II
Question 3(a) Show that for the function
( 2
x y2
x2 +y 2
, (x, y) 6= (0, 0)
f (x, y) =
0, x = 0, y = 0
1. fx is not differentable at (0, 0).
2. fyx is not continuous at (0, 0).
3. fxy (0, 0) = fyx (0, 0).
Solution.
1. For (x, y) 6= (0, 0),
(x2 + y 2 )2xy 2 − x2 y 2 (2x) 2xy 4
fx (x, y) = = 2
(x2 + y 2 )2 (x + y 2 )2
2.
h2 y 2
f (h, y) − f (0, y) h2 +y 2
−0 hy 2
fx (0, y) = lim = lim = lim =0
h→0 h h→0 h h→0 h2 + y 2

Thus fx (0, y) = 0 for y 6= 0.


3.
f (x + h, 0) − f (x, 0)
fx (x, 0) = lim =0
h→0 h
In particular, fx (0, 0) = 0.
fx (0,k)−fx (0,0)
4. fyx (0, 0) = limk→0 k
=0
f (x,k)−f (x,0) x2 k
5. fy (x, 0) = limk→0 k
= limk→0 x2 +k2
= 0 when x 6= 0.
f (0,k)−f (0,0)
6. fy (0, 0) = limk→0 k
= 0.
fy (h,0)−fy (0,0)
7. fxy (0, 0) = limh→0 h
= 0.
Thus we have shown that fxy (0, 0) = fyx (0, 0) = 0.
8. For (x, y) 6= (0, 0),
∂ ∂ 2xy 4
fyx (x, y) = fx (x, y) =
∂y ∂y (x2 + y 2 )2
8xy 3 (x2 + y 2 )2 − 2xy 4 · 2(x2 + y 2 ) · 2y
=
(x2 + y 2 )4
3
8xy 3 (x2 + y 2 ) − 8xy 5 8x3 y 3

xy
= = 2 =8 2
(x2 + y 2 )3 (x + y 2 )3 x + y2
Now from question 1(a) above, it follows that fyx is not continuous at (0, 0).

4
9. If fx were differentiable at (0, 0), we would have

fx (h, k) = fx (0, 0) + hfxx (0, 0) + kfyx (0, 0) + h2 + k 2 φ(h, k)

where limh→0,k→0 φ(h, k) = 0. Substituting the known values, we get

2hk 4 1
φ(h, k) = √
(h2 + k 2 )2 h2 + k 2
2m4
Now we can see that limh→0,k→0 φ(h, k) does not exist — if k = mh, φ(h, mh) = 5 ,
(1+m2 ) 2
so limh→0 φ(h, mh) is different for different values of m (say 0, 1).

Thus we have shown all the three results required.

5
UPSC Civil Services Main 1986 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 16, 2010

Question 1(a) 1. A function f (x) is defined as follows:


( 1
e1− x2 sin x1 , x 6= 0
f (x) =
0, x=0

Examine whether or not f (x) is differentiable at x = 0.

2. If f 0 (x) exists and is continuous, find the value of

1 x
Z
lim (x − 3y)f (y) dy
x→0 x2 0

Solution.
f (x) − f (0)
1. By definition, f 0 (0) = lim , if this limit exists.
x→0 x
f (x) − f (0) 1 1 1
lim = lim e1− x2 sin
x→0 x x→0 x x
1 − 12 1
= e lim 2 e x lim x sin
x→0 x x→0 x
1 1
= e lim te−t · 0 Letting t = 2
, also lim x sin = 0
t→∞ x x→0 x
= 0 ∵ lim te−t = 0
t→∞

Thus f (x) is differentiable at x = 0, and f 0 (0) = 0.

1
2. Define
Z x
F (x) = (x − 3y)f (y) dy
0
Z x Z x
= x f (y) dy − 3 yf (y) dy
0 0
Z x Z x
0
F (x) = f (y) dy + xf (x) − 3xf (x) = f (y) dy − 2xf (x)
0 0
F 00 (x) = f (x) − 2f (x) − 2xf 0 (x) = −f (x) − 2xf 0 (x)
Note that F 00 (x) exists because f 0 (x) exists.
F (x)
The required limit is lim 2 . Since lim F (x) = 0, lim x2 = 0 and both are differen-
x→0 x x→0 x→0
tiable, we can apply L’Hospital’s rule to get
F (x) F 0 (x)
lim = lim
x→0 x2 x→0 2x

Again L’Hospital’s rule applies, so


F (x) F 00 (x) −f (x) − 2xf 0 (x) f (0)
lim 2
= lim = lim =−
x→0 x x→0 2 x→0 2 2
since f 0 (x) exists and is continuous, so limx→0 xf 0 (x) = 0.
Thus Z x
1 f (0)
lim 2 (x − 3y)f (y) dy = −
x→0 x 0 2

Question 1(b) Use Rolle’s theorem to establish that under suitable conditions (to be stated)

f (a) f (b) f (a) f 0 (ξ)
g(a) g(b) = (b − a) g(a) g 0 (ξ) , a < ξ < b

Hence or otherwise deduce the inequality


nbn−1 (a − b) < an − bn < nan−1 (a − b)
where a > b and n > 1.
Solution. Let f (x), g(x) be continuous in the closed interval [a, b] and differentiable in the
open interval (a, b). Let

f (a) f (x) x − a f (a) f (b)
F (x) = −
g(a) g(x) b − a g(a) g(b)
x−a
= [f (a)g(x) − g(a)f (x)] − [f (a)g(b) − g(a)f (b)]
b−a
Thus F (x) is

2
1. continuous in the closed interval [a, b].
2. differentiable in the open interval (a, b)
3. F (a) = F (b) = 0.
Thus F (x) satisfies the requirements of Rolle’s theorem, consequently, there exists ξ, a <
ξ < b such that F 0 (ξ) = 0. But
1
F 0 (ξ) = [f (a)g 0 (ξ) − g(a)f 0 (ξ)] − [f (a)g(b) − g(a)f (b)]
b−a
f (a) f 0 (ξ)
− 1 f (a) f (b) = 0

= 0
g(a) g (ξ) b − a g(a) g(b)

therefore
f (a) f (b) 0
= (b − a) f (a) f 0 (ξ) , a < ξ < b


g(a) g(b) g(a) g (ξ)
Let f (x) = xn , g(x) = 1, then the above result implies
n n n
b a b nξ n−1
1 1 = (a − b) 1
, b < ξ < a
0
or an − bn = (a − b)nξ n−1 . Now since bn−1 < ξ n−1 < an−1 , we get
nbn−1 (a − b) < an − bn < nan−1 (a − b), as required.
(ax3 + by 3 )n ∂ 2u ∂ 2u ∂ 2u
Question 1(c) If u = + xf ( xy ), find the value of x2 2 + 2xy + y2 2 .
3n(3n − 1) ∂x ∂x ∂y ∂y
Solution. We use the result proved in 2006, question 2(b): If f (x, y) is a homogeneous
function of degree n possessing continuous partial derivatives of degree 2,
∂ 2f ∂ 2f 2
2∂ f
x2 + 2xy + y = n(n − 1)f
∂x2 ∂x ∂y ∂y 2
(ax3 + by 3 )n
Let v = , then v is homogeneous of degree 3n, so
3n(3n − 1)
2
2∂ v ∂ 2v 2
2∂ v (ax3 + by 3 )n
x 2
+ 2xy +y 2
= 3n(3n − 1) = (ax3 + by 3 )n
∂x ∂x ∂y ∂y 3n(3n − 1)
Let w = xf ( xy ), then w is homogeneous of degree 1, so
∂ 2w ∂ 2w 2
2∂ w
x2 + 2xy + y = 1(1 − 1)w = 0
∂x2 ∂x ∂y ∂y 2
Now u = v + w, so adding the above equations we have
∂ 2u ∂ 2u 2
2∂ u
x2 + 2xy + y = (ax3 + by 3 )n
∂x2 ∂x ∂y ∂y 2

3
Question 2(a) 1. Without evaluating the involved integrals, show that
1
Z x Z
t dt x dt
+ =0
1 1 + t2 1 t(1 + t2 )
Z a+T
2. If f (x) is periodic of period T , show that f (t) dt is independent of a.
a

Solution.
du t 1 u2 u
1. Let t = u1 , so that dt = − 2
and 2
= 2
= , implying that
u 1+t u 1+u 1 + u2
Z x Z 1   Z 1
t dt x u du x du
2
= 2
− 2
= − 2
1 1+t 1 1+u u 1 u(1 + u )

Thus 1
Z x Z
t dt x dt
+ =0
1 1 + t2 1 t(1 + t2 )
Z a+T
2. Define F (a) = f (t) dt. We shall prove that F (a) = F (b) for any b. In particular
Ra T
F (a) = F (0) = 0
f (t) dt which is independent of a.

Z b+T Z a+T
F (b) − F (a) = f (t) dt − f (t) dt
b a
Z b+T Z a+T Z b
= f (t) dt − f (t) dt − f (t) dt
b b a
Z b+T Z b Z b
= f (t) dt + f (t) dt − f (t) dt
b a+T a
Z b+T Z b
= f (t) dt − f (t) dt
a+T a
Z b Z b
= f (u + T ) du − f (t) dt ∵u+T =t
a a
Z b Z b
= f (u) du − f (t) dt = 0 ∵ f (u + T ) = f (u)
a a

Thus F (b) = F (a) for all b.

4
Question 2(b) Find the volume of the solid generated by revolving one arc of x = a(t −
sin t), y = a(1 − cos t) about its base.
Solution. One arc is given by 0 ≤ t ≤ 2π and the base is the x-axis. Thus
Z 2π
dx
V = π y 2 dt
dt
Z0 2π Z 2π 
2 2 3 t 3
= π a (1 − cos t) a(1 − cos t) dt = πa 2 sin2 dt
0 0 2
Z π
3 t
= 8πa 2 sin6 θ dθ (θ = )
0 2
Z π
2
= 32πa3 sin6 θ dθ (sin(π − θ) = sin θ, so double the integral and half the limit)
0
5·3·1 π
= 32πa3 = 5π 2 a3
6·4·2 2

Z 2 Z x − 21
Question 2(c) Evaluate (x−y)2 +2(x+y)+1 dx dy by using the transformation
0 0
x = u(1 + v), y = v(1 + u). Assume u, v are positive in the region concerned.
Solution.
1. The Jacobian of the transformation is

∂(x, y) 1 + v u
= =1+u+v >0
∂(u, v) v 1 + u
because u ≥ 0, v ≥ 0.
2. The region of integration in the xy-plane is the triangle bounded by the lines y = 0, y =
x, x = 2.
3. The region of integration in the uv-plane lies in the first quadrant as u ≥ 0, v ≥ 0.
Clearly
x=y ⇔ u=v
x = 2 ⇔ u(1 + v) = 2
y = 0 ⇔ v = 0 (∵ 1 + u > 0)

(1, 1)
v-axis
u(1 + v) = 2

(1, 0) (2, 0) u-axis

5
The curve u(1 + v) = 2 meets the u-axis at (2, 0) and the line u = v at (1, 1). The
region of integration is bounded by v = 0, v = u, u(1 + v) = 2 and therefor consists
of the triangle with vertices (0, 0), (1, 0), (1, 1) and the portion bounded by u = 1, v =
0, u(1 + v) = 2, in which u varies from 1 to 2, and v varies from 0 to 2−u u
.

4.
 − 12  − 21
(x − y)2 + 2(x + y) + 1 = (u − v)2 + 2(2uv + u + v) + 1
1
= (u2 + v 2 + 2uv + 2u + 2v + 1)− 2 = (1 + u + v)−1

Thus the product of the Jacobian and the integrand is 1.

Thus
Z 2 Z x − 21
2
I = (x − y) + 2(x + y) + 1 dx dy
0 0
2
Z 1 Z u Z 2 Z
u
−1
= dv du + dv du
0 0 1 0
Z 2
1 2
= + du − 1
2 1 u
1
= 2 log 2 −
2

Paper II

Question 3(a) Find the maximum and minimum values of f (x, y) = 7x2 + 8xy + y 2 where
x, y are constrained by the relation x2 + y 2 = 1.

Solution. Let F (x, y) = 7x2 +8xy +y 2 +λ(x2 +y 2 −1), where λ is Lagrange’s undetermined
multiplier. For extreme values
∂F ∂F
= 14x + 8y + 2λx = 0, = 8x + 2y + 2λy = 0
∂x ∂y

Thus (7 + λ)x + 4y = 0, 4x + (1 + λ)y = 0. Since (x, y) 6= (0, 0) because x2 + y 2 = 1, it


follows that

7 + λ 4
= (7 + λ)(1 + λ) − 16 = λ2 + 8λ − 9 = 0 ⇒ λ = −9, 1
4 1 + λ

∂F ∂F
x +y = 14x2 + 8xy + 2λx2 + 8xy + 2y 2 + 2λy 2 = 2f + 2λ(x2 + y 2 ) = 0
∂x ∂y

6
Since x2 + y 2 = 1, we get f = −λ ⇒ at stationary points f = 9, −1. Thus the maximum
value of f is 9, minimum value is −1.
Check: We have found the maximum and minimum values without finding the stationary
points.
1 2
λ = −9 ⇒ x = 2y, x2 + y 2 = 1 ⇒ 5y 2 = 1 ⇒ y = ± √ , x = ± √
5 5
1 2
λ = 1 ⇒ y = −2x, x2 + y 2 = 1 ⇒ 5x2 = 1 ⇒ x = ± √ , y = ∓ √
5 5

In case (1), f (x, y) = 28


5
+ 165
+ 15 = 9. In case (2) f (x, y) = 57 − 16
5
+ 54 = −1, confirming the
above.
Note: The question could also be done by substituting x = cos t, y = sin t, and then
f (x, y) = 7 cos2 t+8 cos t sin t+sin2 t, which is now a function of one variable. Differentiating
and letting the derivative be 0, we get −14 cos t sin t + 8(cos2 t − sin2 t) + 2 sin t cos t = 0.
Let z = tan t, then 2z 2 + 3z − 2 = 0 ⇒ z = −2, 21 ⇒ (x, y) = (± √15 , ∓ √25 ), (± √25 , ± √15 ) ⇒
f (x, y) = 9, −1.

7
UPSC Civil Services Main 1987 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 14, 2010

1 9 1 9
Question 1(a) If x1 = x+ and for n > 0, xn+1 = xn + , find the value of
2 x 2 xn
limn→∞ xn (x > 0 is assumed).

Solution. x > 0 ⇒ x1r > 0. By induction, clearly xn > 0 for all n ≥ 1.


x + x9 9
Now x1 = ≥ x · = 3, because the arithmetic mean is always ≥ the geometric
2 x
mean of two positive numbers. Similarly xn ≥ 3 for all n ≥ 1, thus the sequence {xn } is
bounded below. Moreover
1 9 1
xn+1 − xn = xn + − xn = (9 − x2n ) ≤ 0
2 2xn 2xn
as x2n ≥ 9. Thus {xn } is a monotonically decreasing sequence. Let l be the greatest lower
bound of {xn } — then
1 9  1 9  1 9
l = lim xn = lim xn−1 + = lim xn−1 + = l+
n→∞ n→∞ 2 xn−1 2 n→∞ limn→∞ xn−1 2 l
Thus 2l2 = l2 + 9 ⇒ l = 3 (l = −3 is not admissible as all elements of the sequence are ≥ 3).
Thus limn→∞ xn = 3.
1
Question 1(b) 1. If x = −a, h = 2a, f (x) = x 3 , find θ from the mean value theorem:
f (x + h) = f (x) + hf 0 (x + θh).

2. If u = x + y − z, v = x − y + z, w = x2 + (y − z)2 , examine whether or not there is a


functional relationship between u, v, w and find the relationship, if any.

Solution.

1
1. Substituting the given values, we get f (a) = f (−a) + 2af 0 (−a + 2aθ). Now f (x) =
1 2
x 3 ⇒ f 0 (x) = 13 x− 3 , so

1 12a 2
a 3 = (−a) 3 + (−a + 2aθ)− 3
3
1 2a − 2 2
= −a 3 + a 3 (2θ − 1)− 3
3
2 2
⇒2 = (2θ − 1)− 3
3
2
⇒3 = (2θ − 1)− 3
⇒ 27 = (2θ − 1)−2
1
⇒ ±√ = 2θ − 1
27

So θ = 21 (1 ± 1

3 3
).

2. To establish that u, v, w have a functional relation, we have to show that the Jacobian
∂(u, v, w)
= 0. Clearly
∂(x, y, z)

1 1 −1 1 1 0
∂(u, v, w)
= 1 −1 1 = 1
−1 0 = 0
∂(x, y, z) 2x 2(y − z) −2(y − z) 2x 2(y − z) 0

Thus u, v, w are functionally dependent.


Now u2 = x2 + (y − z)2 + 2x(y − z), v 2 = x2 + (y − z)2 − 2x(y − z), thus u2 + v 2 = 2w,
which is the desired functional relationship.

 x n1 +1 + y n1 +1  12
Question 1(c) If u = csc−1 , show that
x+y

∂ 2u ∂ 2u 2
2∂ u 1
x2 2
+ 2xy + y 2
= 2 tan u(2n + sec2 u)
∂x ∂x ∂y ∂y 4n
1
Solution. Let v = csc u. Then v is a homogeneous function of degree 2n
, because
 (ax) n1 +1 + (ay) n1 +1  12 1
 x n1 +1 + y n1 +1  12 1
v(ax, ay) = = a 2n = a 2n v(x, y)
a(x + y) x+y

2
Applying Euler’s theorem to v,
∂v ∂v 1
x +y = v
∂x ∂y 2n
∂u ∂u 1 ∂v ∂u
−x csc u cot u − y csc u cot u = csc u ∵ v = csc u, = − csc u cot u
∂x ∂y 2n ∂x ∂x
∂u ∂u 1
x +y = − tan u (1)
∂x ∂y 2n
2 2
∂u ∂ u ∂ u 1 ∂u
+x 2 +y = − sec2 u Differentiating (1) w.r.t. x (2)
∂x ∂x ∂x ∂y 2n ∂x
2
∂ u ∂u ∂ 2u 1 ∂u
x + +y 2 = − sec2 u Differentiating (1) w.r.t. y (3)
∂x ∂y ∂y ∂y 2n ∂y
2 2
∂ 2u
 
∂ u ∂ u 1 ∂u ∂u
x2 2 + 2xy + y2 2 = (−1 − 2
sec u) x +y (2) × x + (3) × y
∂x ∂x ∂y ∂y 2n ∂x ∂y
1 1
= − (2n + sec2 u)(− tan u) From (1)
2n 2n
1
= tan u(2n + sec2 u)
4n2

Question 2(a) 1. Show by means of a suitable substitution that


π

tx−1
Z Z
2
2x−1 2y−1 1
sin θ cos θ dθ = dt, x, y > 0
0 2 0 (1 + t)x+y

2. Establish the inequality Z 1


1 dx π
< 1 <
2 0 (4 − x2 − x3 ) 2 6

Solution.
Z 1
1. By definition B(y, x) = z y−1 (1 − z)x−1 dz.
0
2
Put z = cos θ, dz = −2 cos θ sin θ dθ, to get
Z 0
B(y, x) = cos2y−2 θ sin2x−2 θ (−2 cos θ sin θ) dθ
π
2
Z π
2
= 2 sin2x−1 θ cos2y−1 θ dθ
0

3
1
Now, put z = t+1
in B(y, x),
Z 1
B(y, x) = z y−1 (1 − z)x−1 dz
0
Z 0
1 y−1  1 x−1 −dt
= 1−
t+1 t+1 (1 + t)2
Z∞∞
tx−1
= dt
0 (1 + t)x+y

From these two we have


Z π
1 ∞ tx−1
Z
2
2x−1 2y−1
sin θ cos θ dθ = dt
0 2 0 (1 + t)x+y

as required.

2. Since 4 − x2 + x3 = 4 − x2 (1 − x), and x2 (1 − x) > 0 for 0 < x < 1, it follows that


1 1
4 − x2 + x3 < 4 ⇒ < 1 .
2 (4 − x2 − x3 ) 2
1 1
On the other hand, 4 − x2 + x3 > 4 − x2 for x > 0, so (4 − x2 + x3 )− 2 < (4 − x2 )− 2 .
Hence
Z 1 Z 1 Z 1
1 dx dx
dx < 1 < √
0 2 4 − x2
2 3
0 (4 − x − x ) 2 0
Z 1 1
1 dx −1 x π
< 1 < sin =
2 2 3
0 (4 − x − x ) 2 2 0 6

x3
Question 2(b) Find the volume of the solid generated by revolving the curve y 2 = ,a >
2a − x
0 about its asymptote x = 2a.

Solution. Consider a thin vertical strip of thickness dx, and rotate it about the asymptote

4
r
x3
— its volume is 2πrh dx = 2π(2a − x)2 dx. Thus
2a − x
s
Z 2a
x3
V = 4π (2a − x) dx
0 2a − x

Z 2a
3
= 4π 2a − x x 2 dx
0
Put x = 2a sin2 θ ⇒ dx = 4a sin θ cos θ dθ
Z π√
2 3
= 4π 2a cos θ(2a) 2 sin3 θ 4a sin θ cos θ dθ
0
Z π
2
3
= 64πa sin4 θ cos2 θ dθ
0
3·1·1 π
= 64πa3 = 2π 2 a3
6·4·2 2

ZZ
1 1 2
Question 2(c) Evaluate x 2 y 3 (1 − x − y) 3 dx dy where D is the domain bounded by the
D
lines x = 0, y = 0, x + y = 1.

Solution. We convert this to a Dirichlet integral using the standard transformation. Put
x + y = u, y = uv, so that dx dy = u du dv, (see 1989, question 3(a) for example).
Z 1Z 1
1 1 1 1 2
I = u 2 (1 − v) 2 u 3 v 3 (1 − u) 3 u du dv
Z0 1 Z0 1
11 2 1 1
= u 6 (1 − u) 3 v 3 (1 − v) 2 du dv
0 0
Γ( 17
6
)Γ( 53 ) Γ( 34 )Γ( 32 ) 2
3
Γ( 23 ) 31 Γ( 13 )Γ( 32 ) 16  2   1 
= = = Γ Γ
Γ( 276
) Γ( 17 6
) 753
222
Γ( 32 ) 945 3 3

But Γ( 23 )Γ( 13 ) = π
sin π3
= 2π

3
. Thus I = 32π
√ .
945 3

Paper II

Question 3(a) Let f (x) = x if x is rational, and 1 − x if x is irrational. Show that f is


continuous only at x = 12 .

Solution. See 2001, question 1(a).

5
Question 4(a) Find the maximum and minimum value of f (x, y) = xy subject to the con-
dition that x2 + y 2 + xy = a2 .

Solution. Let F (x, y) = xy − λ(x2 + y 2 + xy − a2 ), where λ is Lagrange’s undetermined


multiplier. For extreme values,
∂F ∂F y x
= y − 2λx − λy = 0, = x − 2λy − λx = 0 ⇒ λ = =
∂x ∂y 2x + y 2y + x

as (x, y) 6= (0, 0). Thus 2x2 + xy = 2y 2 + xy ⇒ (x + y)(x − y) = 0 ⇒ x = y, x = −y. Using


x2 + y 2 + xy = a2 we get
a2
1. x = y ⇒ 3x2 = a2 ⇒ x = y = ± √a3 ⇒ f (x, y) = 3

2. x = −y ⇒ x2 = a2 ⇒ x = a, y = −a or x = −a, y = a. In either of these cases


f (x, y) = −a2 .
2
Thus the required maximum value is a3 and the required minimum is −a2 .
Note: In this problem there was no need to check the nature of the critical points, as the
maximum amd minimum values occur at these points. If it were required, it could be done
as follows.
d2 F = −2λ(dx)2 − 2λ(dy)2 + 2(1 − λ)dx dy
2x+y
Now x2 + y 2 + xy = a2 ⇒ 2x dx + x dy + y dx + 2y dy = 0, or dy = − x+2y dx. Thus
 2
2 2 2x + y 2x + y
d F = −2λ(dx) − 2λ (dx)2 − 2(1 − λ)(dx)2
x + 2y x + 2y

Case 1: x = a, y = −a, λ = −1 or x = −a, y = a, λ = −1

d2 F = 2(dx)2 − 4(−1)(dx)2 + 2(−1)2 (dx)2 > 0

so we have a local minimum at (a, −a) or (−a, a).


Case 2: x = y = ± √a3 , λ = 31 .

2 4 2
d2 F = − (dx)2 − (dx)2 − (dx)2 < 0
3 3 3
so we have a local maximum at x = y = ± √a3 .

6
UPSC Civil Services Main 1988 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 14, 2010

Question 1(a) If f (x) = tan x prove that


   
(n) n (n−2) n (n−4) nπ
f (0) − f (0) + f (0) − . . . = sin
2 4 2

Solution. Clearly sin x = f (x) cos x. Using Leibnitz’s formula for the derivative of the
product of two functions, we get
n  
dn sin x X n (n−r)
= f (x) cos(r) (x)
dxn r=0
r

dr rπ dr rπ dr
Now cos x = cos(x + 2
) and sin x = sin(x + 2
). Therefore, cos x = 0 when
dxr r
dxr n
dxr
d n d
x = 0, n odd, and r
cos x = (−1) 2 when x = 0, n even. Also, sin x = sin nπ
2
when
dx dxn
x = 0.
Thus    
nπ (n) n (n−2) n (n−4)
sin = f (0) − f (0) + f (0) − . . .
2 2 4
.

Question 1(b) Find the minimum value of x2 + y 2 + z 2 when x + y + z = k.

Solution. Let F (x) = x2 + y 2 + z 2 + λ(x + y + z − k), where λ is Lagrange’s undetermined


multiplier. For stationary values,
∂F ∂F ∂F
= 2x + λ = 0, = 2y + λ = 0, = 2z + λ = 0
∂x ∂y ∂z

1
Thus λ = −2x = −2y = −2z ⇒ x = y = z. From x + y + z = k we have x = y = z = k3 .
Now
∂ 2F ∂ 2F ∂ 2F ∂ 2F ∂ 2F ∂ 2F
d2 F = (dx)2
+ (dy)2
+ (dz)2
+ dx dy + dy dz + dz dx
∂x2 ∂y 2 ∂z 2 ∂x ∂y ∂y ∂z ∂z ∂x
= 2(dx)2 + 2(dy)2 + 2(dz)2
Now dx + dy + dz = 0 from x + y + z = k, so substituting dz = −dx − dy, we get
d2 F = 2(dx)2 + 2(dy)2 + 2(dx + dy)2 = 4(dx)2 + 4(dy)2 + 4dx dy
 
dy 2 3 2
= 4 dx + + (dy) > 0
2 4
k k2 2 2 k2
Thus x2 +y 2 +z 2 is minimum when x = y = z = 3
and the minimum value of 9
+ k9 + k9 = 3
.

Question 1(c) Find the asymptotes of the cubic


x3 − xy 2 − 2xy + 2x − y = 0
and show that they cut the curve again in points which lie on the line 3x − y = 0.
Solution. Since the coefficient of the highest degree term of y in the equation, i.e.y 2 is −x,
it follows that x = 0 is an asymptote parallel to the y axis.
There is no asymptote parallel to the x-axis, as the coefficient of x3 , the highest degree
term of x peresent in the equation, is a constant.
If y = mx + c is an asymptote, then m is given by Φ3 (m) = 1 − m2 = 0 ⇒ m = ±1. c
Φ2 (m)
is given by − 0 provided it is not indeterminate, where Φr are homogeneous terms of
Φ3 (m)
degree r present in the equation. Thus c = − −2m−2m
= −1. Thus the asymptotes of the type
y = mx + c are y = x − 1, y + x = −1.
The joint equation of asymptotes is given by x(y − x + 1)(y + x + 1) = 0 or x(y 2 − x2 +
2y + 1) = xy 2 − x3 + 2yx + x = 0, or P3 = x3 − xy 2 − 2yx − x = 0.
Thus the point of intersection of the curve and the asymptotes lie on the line 3x − y = 0
as the equation of the given curve is P3 + 3x − y = 0.
Note: The asymptotes parallel to the y-axis can also be found as follows:
Let x = my + d be an asymptote. Dividing the equation by y 3 and letting y → ∞, we
get m3 − m = 0, as xy → m when y → ∞. Thus m = 0, ±1.
If m = 0, then x = d, so d3 − dy 2 − 2dy + 2d − y = 0. Dividing by y 2 and letting y → ∞
we get d = 0, so x = 0 is an asymptote.
Similarly we can find d = 1 when m = ±1. This will determine all three asymptotes.

Question 2(a) Find the center of gravity of one loop of the lemniscate of Bernoulli r2 =
a2 cos 2θ.
Solution.

2
Because of symmetry, y = 0. The area of
the lemniscate is
Z π Z π  π4
1 4 2 4 sin 2θ a2
r dθ = a2 cos 2θ dθ = a2 =
2 − π4 0 2 0 2
π
Now θ= 4
Z π
2 4
3
θ = − π4
x = r cos θ dθ
3a2 − π4
Z π
4a 4 3
= (1 − 2 sin2 θ) 2 cos θ dθ
3 0
√ cos t dt
Let 2 sin θ = sin t ⇒ cos θ dθ = √
2
Z π Z π √
4a 2
2 3 4a 2
4 4a 3 · 1 · π π 2a
= √ (1 − sin t) 2 cos t dt = √ cos t dt = √ =
3 2 0 3 2 0 3 24·2·2 8

the centroid is π 82a , 0 .

Thus
ZZ
3
Question 2(b) Evaluate x 2 y 2 (1 − x2 − y 2 ) dx dy over the positive quadrant of the circle
x2 + y 2 = 1.
dX dY
Solution. Substitute x2 = X, y 2 = Y ⇒ dx dy = 1 1 and the region of integration is
4X 2 Y 2
transformed to X ≥ 0, Y ≥ 0, X + Y ≤ 1. The required integral becomes
ZZ ZZ
3 dX dY 1 1 1
I= X Y (1 − X − Y )
4
1 1 = X 4 Y 2 (1 − X − Y ) dX dY
X≥0,Y ≥0 4X 2 Y 2 4 X≥0,Y ≥0
X+Y ≤1 X+Y ≤1

Put X + Y = u, Y = uv ⇒ X = u(1 − v).



∂(X, Y ) 1 − v −u
= =u
∂(u, v) v u
Y
Also, X ≥ 0, Y ≥ 0, X + Y ≤ 1 ⇒ 0 ≤ u ≤ 1, 0 ≤ v ≤ 1(∵ v = X+Y
).

1 1 1 1
Z Z
1 1 1
I = u 4 (1 − v) 4 u 2 v 2 (1 − u)u du dv
4 0 0
1 1 7
Z Z 1
1 1
= u 4 (1 − u) du v 2 (1 − v) 4 dv
4 0 0
1 11 3 5 1 Γ( 114
)Γ(2) Γ( 32 )Γ( 54 ) 1 Γ( 32 )Γ( 54 )
= B( 4 , 2)B( 2 , 4 ) = =
4 4 Γ( 19 4
) Γ( 114
) 4 Γ( 19 4
)

The above integral actually becomes Dirichlet’s integral.

3
Question 2(c) Show that the volume of the solid obtained by revolving the curve (a−x)y 2 =
π 2 a3
a2 x about its asymptote is .
2
Solution. The asymptote is x = a, because a − x is the coefficient of the highest degree
term in y present in the equation of the curve. Thus
Z ∞
V = π(a − x)2 dy
−∞
Z ∞ Z ∞
ay 2 2 6 dy
= 2π a− 2 2
dy = 2πa
0 a +y 0 (a + y 2 )2
2

Put y = a tan θ, dy = a sec2 θ dθ


Z π Z π
6
2 a sec2 θ dθ 3
2
2 31 π π 2 a3
V = 2πa 4 2 2
= 2πa cos θ dθ = 2πa =
0 a (1 + tan θ) 0 22 2

Paper II

Question 3(a) Evaluate


x2 y 2  32
ZZ 
1− − 2 dx dy
a2 b
x2 y 2
over the area of the ellipse + 2 = 1.
a2 b
Solution.
ZZ Let x = aX, y = bY , so dx dy = ab dX dY . The given integral becomes
3
ab (1 − X − Y 2 ) 2 dX dY over the circle X 2 + Y 2 ≤ 1. Changing to polar coordinates,
2

X = r cos θ, Y = r sin θ, we get


π
Z
2
Z 1
3
I = 4ab (1 − r2 ) 2 r dr dθ
0 0
Z 1
3
= 2πab (1 − r2 ) 2 r dr
0
Put r2 = t, 2r dr = dt
Z 1
3 2 5 1
i 2πab
= πab (1 − t) dt = −πab (1 − t)
2 2 =
0 5 0 5

4
Question 4(a) Show that a local extreme value of f given by
f (x) = xk1 + . . . + xkn , x = (x1 , . . . , xn )
subject to the condition x1 + . . . + xn = a is ak n1−k .
Solution. Let F (x) = xk1 +. . .+xkn −λ(x1 +. . .+xn −a) where λ is Lagrange’s undetermined
multiplier. For extreme values, for all i
∂F λ
= kxk−1
i − λ = 0 ⇒ xk−1
i =
∂xi k
λ a
Thus xk−1
1 = . . . = xk−1
n = , or x1 = . . . = xn . Using x1 + . . . + xn = a we get xi = , so
n
k n
X a k nak
the extreme value is = k = ak n1−k .
i=0
n n
Note: To decide the nature of the extreme value, we consider
d2 F = k(k − 1)xk−2 2 2
1 [(dx1 ) + . . . + (dxn ) ]

using x1 = . . . = xn . Thus d2 F > 0 if k < 0 or k > 1 (assuming a > 0), so we have a


minimum. If 0 < k < 1, d2 F < 0, and we have a maximum. This is not required for this
question.
Question 4(b) The function f : R2 → R is given by
( 2
x y
x2 +y 2
, (x, y) 6= (0, 0)
f (x, y) =
0, (x, y) = (0, 0)
Prove that at (0, 0) f is continuous and possesses all directional derivatives, but is not dif-
ferentiable.
Solution. Let  > 0 and δ = , then
x2 y r2 cos2 θ r sin θ p
|f (x, y) − f (0, 0)| = 2 = ≤ r = x2 + y 2

x + y2 r2

Thus for any (x, y) belonging to the open disc with center (0, 0) and radius δ =  i.e. for
x2 + y 2 < 2 , we get |f (x, y) − f (0, 0)| < , so f (x, y) is continuous at (0, 0).
Let v = (cos θ, sin θ), then by definition the directional derivative at (a, b) in the direction
of v is given by
f (a + t cos θ, b + t sin θ) − f (a, b)
lim
t→0 t
when this limit exists, and it is denoted by fv (a, b).
Here (a, b) = (0, 0), so
f (t cos θ, t sin θ) t2 cos2 θ t sin θ
fv (0, 0) = lim = lim = cos2 θ sin θ
t→0 t t→0 t3
Thus all directional derivatives exist.
However for f to be differentiable, all directional derivatives should be equal — this is
clearly not the case here, as the directional derivative for θ = 0 is 0, but for θ = π4 it is 2√1 2 .

5
UPSC Civil Services Main 1989 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 14, 2010

Question 1(a) If f is at least thrice continuously differentiable, then show that

h2 00
f (a + h) = f (a) + hf 0 (a) + f (a + θh)
2!
where θ lies between 0 and 1, and prove that limh→0 θ = 13 .

Solution. Let
(a + h − x)2
φ(x) = f (a + h) − f (x) − (a + h − x)f 0 (x) − A
2!
where A is so determined that φ(a + h) = φ(a). Clearly φ(a + h) = 0, and φ(a) = f (a + h) −
2 2
f (a) − hf 0 (a) − h2! A, so φ(a) = 0 ⇒ h2! A = f (a + h) − f (a) − hf 0 (a).
Now φ satisfies the requirements of Rolle’s theorem in [a, a + h], therefore there exists
θ ∈ (0, 1) such that φ0 (a + θh) = 0, note that a + θh ∈ (a, a + h). But

φ0 (a + θh) = −(h − θh)f 00 (a + θh) + (h − θh)A = 0 ⇒ A = f 00 (a + θh)

Now from φ(a) = 0, we get

h2 00
f (a + h) = f (a) + hf 0 (a) + f (a + θh)
2!
as required.
Now consider the equality
h2 00
f (a + h) − f (a) − hf 0 (a) − 2
f (a) h2 f 00 (a + θh) − f 00 (a)
=
h3 2 h3

1
Taking the limit as h → 0 on both sides, we get
2
f (a + h) − f (a) − hf 0 (a) − h2 f 00 (a)
LHS = lim
h→0 h3
f (a + h) − f (a) − hf 00 (a)
0 0
= lim L’Hospital’s rule
h→0 3h2
f 00 (a + h) − f 00 (a)
= lim L’Hospital’s rule
h→0 6h
1 000
= f (a)
6
f 00 (a + θh) − f 00 (a) θ f 00 (a + θh) − f 00 (a)
RHS = lim = lim
h→0 2h h→0 2 θh
θ 000
= f (a)
2
Thus θ = 31 as h → 0, provided f 000 (a) 6= 0.
The question above is a particular case of the following:
Taylor’s Theorem: Let f (n−1) , the (n − 1)-th derivative of a real valued function be
continuous in the closed interval [a, a + h], and let f (n) exist in the open interval (a, a + h),
then there exists a real number θ, 0 < θ < 1, such that
hn−1 (n−1)
f (a + h) = f (a) + hf 0 (a) + . . . + f (a) + Rn
(n − 1)!
where
hn (1 − θ)n−p (n)
Rn = f (a + θh), 0 < θ < 1
(n − 1)!p
Proof: The condition that f (n−1) (x) is continuous in [a, a + h] ⇒ f, f 0 , f 00 , . . . , f (n−2) are
continuous in [a, a + h]. Let
(a + h − x)n−1 (n−1)
φ(x) = f (x) + (a + h − x)f 0 (x) + . . . + f (x) + A(a + h − x)p
(n − 1)!
where A is a constant to be determined so that φ(a + h) = φ(a) i.e.

0 (hn−1 (n−1)
f (a + h) = f (a) + hf (a) + . . . + f (a) + Ahp
(n − 1)!
Now (i) φ(x) is continuous in [a, a + h] (ii) φ(x) is differentiable in (a, a + h) and (iii)
φ(a + h) = φ(a). Thus all requirements of Rolle’s theorem are satisfied,so there exists a real
number θ, 0 < θ < 1 such that φ0 (a + θh) = 0 (note that any real number c ∈ (a, a + h) can
be written as c = a + c−a
h
h, and since a < c < a + h, we get 0 < θ = c−a
h
< 1). But

(a + h − x)n−1 (n)
φ0 (x) = f (x) − pA(a + h − x)p−1
(n − 1)!

2
so
(h(1 − θ))n−1 (n) hn−p (1 − θ)n−p (n)
φ0 (a+θh) = f (a+θh)−pAhp−1 (1−θ)p−1 = 0 ⇒ A = f (a+θh)
(n − 1)! (n − 1)!p

Thus
hn−1 (n−1) hn (1 − θ)n−p (n)
f (a + h) = f (a) + hf 0 (a) + . . . + f (a) + f (a + θh)
(n − 1)! (n − 1)!p

If we put n = p we get

hn−1 (n−1) hn
f (a + h) = f (a) + hf 0 (a) + . . . + f (a) + f (n) (a + θh)
(n − 1)! n!

The given question is for n = 2, so

h2 00
f (a + h) = f (a) + hf 0 (a) + f (a + θh)
2!
Now we prove the second part: If f (n+1) (x) is continuous from the right at x = a, then
1
θ → n+1 as h → 0 provided f (n+1) (a) 6= 0.
In the given question, n = 2, and f (3) (x) is continuous in [a, a + h] so θ → 13 .
f (n+1) (x) is continuous from the right at x = a ⇒ there exists k, 0 < k < h, such that
f (n+1) (x) exists in [a, a + k]. Thus by the above theorem:

0 (k n−1 (n−1) k n (n)


f (a + k) = f (a) + kf (a) + . . . + f (a) + f (a + θk)
(n − 1)! n!
n n+1
(k (n) k
f (a + k) = f (a) + kf 0 (a) + . . . + f (a) + f (n+1) (a + φk)
n! (n + 1)!

where 0 < θ < 1, 0 < φ < 1. Subtracting the first from the second,
k
f (n) (a + θk) − f (n) (a) = f (n+1) (a + φk)
n+1

Now we use Lagrange’s mean value theorem for f (n) (x) in [a, a + θk] and obtain

k
θkf (n+1) (a + θθ1 k) = f (n+1) (a + φk), 0 < θ1 < 1
n+1
1
Taking limit as h → 0 i.e. k → 0, we get limh→0 θ = n+1 as f (n+1) (a+θθ1 k) and f (n+1) (a+φk)
both tend to f (n+1) (a) as k → 0, and f (n+1) (a) 6= 0.

Question 1(b) Prove that the√volume of a right circular cylinder of greatest volume which
can be inscribed in a sphere is 33 times that of a sphere.

3
Solution. Let the radius of the sphere be a, and the radius and height of the cylinder be r
and h respectively. Then r2 + ( h2 )2 = a2 . Letting r = a cos θ, h = 2a sin θ, the volume of the
cylinder is 2πa3 cos2 θ sin θ.
dV
For extreme values, = 2πa3 [cos3 θ − 2 cos θ sin2 θ] = 0. Now cos θ = 0 ⇒ θ = π2 ⇒

r = 0, which is not admissible. So cos2 θ − 2 sin2 θ = 0 ⇒ tan θ = √12 as 0 < θ < π2 . Thus
q
sin θ = √13 , cos θ = 23 .
d2 V 3
q
2
= 2πa [−3 cos θ sin θ + 2 sin θ − 4 cos θ sin θ] < 0 for sin θ = 3 , cos θ = 23 . Thus
3 2 2 √1

V is maximum when tan θ = √12 .

V = 2πa3 23 · √1 = 4 πa3 · 3
√ 3 3 3
3
Thus V is 3
times the volume of the sphere.

∂ 2z
Question 1(c) Show that at the point of the surface xx y y z z = c where x = y = z, =
∂x ∂y
−[x log ex]−1 .

Solution. Taking logs on both sides of xx y y z z = c, we get

z log z + y log y + x log x = log c

Differentiating with respect to x,


∂z ∂z 1 + log x log ex
(1 + log z) + (1 + log x) = 0 ⇒ =− =− (∗)
∂x ∂x 1 + log z log ez
∂z log ey
Similarly, =− .
∂y log ez
Differentiating (*) w.r.t. y,

∂ 2z 1 ∂z ∂z ∂ 2z 1 ∂z ∂z log ex log ey
(1 + log z) + =0⇒ =− =−
∂x ∂y z ∂y ∂x ∂x ∂y z log ez ∂y ∂x z(log ez)3

∂ 2z log ex log ex
Letting x = y = z, we get =− = −[x log ex]−1 as required.
∂x ∂y x(log ex)3

Question 2(a) Find the surface of the solid generated by revolving x = a cos3 t, y = a sin3 t
about the x-axis.

4
Solution. We confine ourselves to the first quadrant because of symmetry. The curve is
2 2 2
x3 + y 3 = a3 .
Z π
2 ds
S = 2 2πy dt
0 dt
Z π r
2
3 dx 2  dy 2
= 4π a sin t + dt
0 dt dt
Z π
2 1
sin3 t (−3a cos2 t sin t)2 + (3a sin2 t cos t)2 2 dt

= 4πa
0
Z π
2
= 4πa 3a sin4 t cos t dt
0
5
 π2
2 sin t 12 2
= 12πa = πa
5 0 5

Question 2(b) If for a curve x sin θ + y cos θ = f 0 (θ) and x cos θ − y sin θ = f 00 (θ) then show
that S = f (θ) + f 00 (θ) + C.

Solution. Solving for x and y, we get

x = f 0 (θ) sin θ + f 00 (θ) cos θ


y = f 0 (θ) cos θ − f 00 (θ) sin θ
dx
= f 0 (θ) cos θ + f 00 (θ) sin θ − f 00 (θ) sin θ + f 000 (θ) cos θ

= f 0 (θ) + f 000 (θ) cos θ


dy
= −f 0 (θ) sin θ + f 00 (θ) cos θ − f 00 (θ) cos θ − f 000 (θ) sin θ

= − f 0 (θ) + f 000 (θ) sin θ

Z r 2  2 Z
dx dy  0
f (θ) + f 000 (θ) dθ

S = + dθ =
dθ dθ
00
= f (θ) + f (θ) + C

as required.
ZZZ
1
Question 2(c) Evaluate (1 − z) 2 dx dy dz over the interior of the tetrahedron with
faces x = 0, y = 0, z = 0, x + y + z = 1.

5
Solution.
Z 1 Z 1−z Z 1−y−z
1
I = (1 − z) 2 dx dy dz
z=0 y=0 x=0
Z1 Z1−z
1
= (1 − z) 2 (1 − y − z) dy dz
z=0 y=0
1
y2
Z h i1−z
1
= (1 − z) 2 y − − zy dz
z=0 2 0
Z 1
1
h (1 − z)2 i
= (1 − z) 1 − z −
2 − z(1 − z) dz
z=0 2
Z 1
3
h (1 − z) i
= (1 − z) 2 1 − − z dz
z=0 2
Z 1 1
1 5 1 −2 7 1
= (1 − z) 2 dz = (1 − z) 2 =
z=0 2 2 7 0 7

Paper II

Question 3(a) Show that the value of


ZZ
1 1
(1 − x − y)3 x 2 y 2 dx dy

taken over the interior of the triangle whose vertices are the origin and the points (0, 1) and
π
(1, 0) is 480 .

Solution. We will convert this into a Dirichlet integral. Let x + y = u, y = uv, so


∂(x, y) 1 − v −u
= u. u varies from 0 to 1, and v = y = y also
x = u(1 − v), and = u x+y
∂(u, v) v u
varies from 0 to 1.

Z 1 Z 1
1 1 1 1
I = (1 − u)3 u 2 (1 − v) 2 u 2 v 2 u du dv
Z0 1 Z0 1
3 3
= (1 − u)4−1 u3−1 (1 − v) 2 −1 v 2 −1 du dv
0 0
Γ(4)Γ(3) Γ( 23 )Γ( 32 )
=
Γ(7) Γ(3)
3! π π
= =
6! 4 480
as Γ( 12 ).

6
Question 3(b) Obtain the largest and the least values of 2(x + y + z) − xyz on the closed
ball x2 + y 2 + z 2 ≤ 9.

Solution. Let F (x, y, z) = 2(x + y + z) − xyz − λ(x2 + y 2 + z 2 − 9), where λ is Lagrange’s


undetermined multiplier. For extreme values:
∂F ∂F ∂F
= 2 − yz + 2λx = 0, = 2 − xz + 2λy = 0, = 2 − yx + 2λz = 0
∂x ∂y ∂z
Subtracting the first two, z(x − y) + 2λ(x − y) = 0 ⇒ x = y or z = −2λ. Similarly from the
other pairs, we get y = z or x = −2λ and x = z or y = −2λ.
We now explicitly find extreme vaues to get the greatest and least values of f (x, y, z) =
2(x + y + z) − xyz

1. x = y = z: This gives us x2 + x2 + x2 = 9, so x = ± 3.
√ √ √ √
(a) x = y = z = 3: f (x, y, z) = 6 3 − 3 3 = 3 3.
√ √ √ √
(b) x = y = z = − 3: f (x, y, z) = −6 3 + 3 3 = −3 3.

2. x = y, y = −2λ: 4λ2 + 4λ2 + z 2 = 9 ⇒ z 2 = 9 − 8λ2 . Using 2 − xy + 2λz = 0, we have


2λ2 − 1
2 − 4λ2 + 2λz = 0 ⇒ z = . Thus
λ
 2λ2 − 1 2
= 9 − 8λ2
λ
⇒ 4λ4 − 4λ2 + 1 = 9λ2 − 8λ4
⇒ 12λ4 − 13λ2 + 1 = 0
1
⇒ λ2 = 1,
12
λ = 1 ⇒ x = y = −2, z 2 = 1
λ = −1 ⇒ x = y = 2, z 2 = 1
1 1 25
λ = √ ⇒ x = y = − √ , z2 =
2 3 3 3
1 1 25
λ = − √ ⇒ x = y = √ , z2 =
2 3 3 3
We examine each of these 8 cases:

(a) x = y = −2, z = 1: f (x, y, z) = −6 − 4 = −10.


(b) x = y = −2, z = −1: f (x, y, z) = −10 + 4 = −6.
(c) x = y = 2, z = 1: f (x, y, z) = 10 − 4 = 6.
(d) x = y = 2, z = −1: f (x, y, z) = 6 + 4 = 10.
(e) x = y = − √13 , z = √5 :
3
f (x, y, z) = √6
3
− 5

3 3
= 13
√ .
3 3

7
(f) x = y = − √13 , z = − √53 : f (x, y, z) = − √143 + 5

3 3
= − 337
√ .
3

(g) x = y = √1 , z = √5 : f (x, y, z) = 14
√ − 5
√ = 37
√ .
3 3 3 3 3 3 3

(h) x = y = √1 , z = − √53 : f (x, y, z) = − √63 + 5


√ = − 313
√ .
3 3 3 3

We need not consider the other possibilities like y = z, z = −2λ as the situation is symmetric
and no new values will result. The greatest value of f (x, y, z) is 10, and the least value is
-10.
We don’t consider d2 F , as it is not needed. We use the fact the all the maximum and
minimum values occur at the extreme values. Calculation of d2 F is a very lengthy process.

8
UPSC Civil Services Main 1990 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 14, 2010

Question 1(a) If a function f (x) of the real variable x has the first five derivatives 0 at a
given value x = a, show that it has a maximum or a minimum at x = a according as the
sixth derivative is negative or positive. What happens if only the first four derivatives are 0,
but not the fifth?

Solution. We shall take up the general case, which would cover both the cases —
Let f (n) (a) exist and f (n) (a) 6= 0. Let f (r) (a) = 0, 1 ≤ r ≤ n − 1. Then f (a) is not an
extreme value when n is odd, and if n is even, then f has a maximum at x = a if f (n) (a) < 0
and a minimum if f (n) (a) > 0.
Thus the above result is proved when n = 6. When n = 5, f (x) has neither maximum
nor minimum at x = a.
Proof: The existence of f (n) (a) imlies that f 0 (x), f 00 (x), . . . , f (n−1) (x) all exist in a certain
neighborhood of a, say (a − δ1 , a + δ1 ).
Case (1): If f (n) (a) > 0, then f (n−1) (x) is increasing at a i.e. there exists 0 < δ < δ1
such that f (n−1) (x) < f (n−1) (a) = 0 for a − δ < x < a, and f (n−1) (x) > f (n−1) (a) = 0 for
a < x < a + δ.
Case (2): If f (n) (a) < 0, then f (n−1) (x) is decreasing at a i.e. there exists 0 < δ < δ1
such that f (n−1) (x) > f (n−1) (a) = 0 for a − δ < x < a, and f (n−1) (x) < f (n−1) (a) = 0 for
a < x < a + δ.
Let h be such that |h| < δ, then by Taylor’s theorem

hn−1 (n−1)
f (a + h) = f (a) + hf 0 (a) + . . . + f (a + θh), 0 < θ < 1
(n − 1)!

hn−1 (n−1)
Thus f (a + h) − f (a) = f (a + θh) as f (r) (a) = 0, 1 ≤ r ≤ n − 2. Since
(n − 1)!
a + θh ∈ (a − δ, a + δ), we have the following conclusions:

1
1. n even and f (n) (a) > 0: If h < 0, then hn−1 < 0 and f (n−1) (a + θh) < 0 (by Case
(1) above, as a − δ < a + θh < a. If h > 0, then hn−1 > 0 and f (n−1) (a + θh) > 0 (as
a < a + θh < a + δ.
hn−1 (n−1)
In either case f (a + θh) > 0 for all h with |h| < δ, i.e. f (a + h) > f (a)
(n − 1)!
for |h| < δ, thus f (x) has a minimum at x = a.
hn−1 (n−1)
2. n even and f (n) (a) < 0: Using Case (2), we get f (a + θh) < 0 for all h
(n − 1)!
with |h| < δ, i.e. f (a + h) < f (a) for |h| < δ, thus f (x) has a maximum at x = a.

3. n odd and f (n) (a) > 0: If h < 0, then hn−1 > 0 and f (n−1) (a + θh) < 0, so f (a +
h) < f (a) for h < 0, |h| < δ. If h > 0, then hn−1 > 0 and f (n−1) (a + θh) > 0, so
f (a + h) > f (a) for 0 < h < δ. Thus f has neither minimum nor maximum at x = a.
The case for n odd and f (n) (a) < 0 is similar — h < 0 ⇒ f (a + h) > f (a), h > 0 ⇒
f (a + h) < f (a), so there is no extreme value at x = a.

This completes the proof.

Question 2(a) Show that

f (x) = (x − 2)2 (x2 + 2bx + c)(x + 3)3

has a critical point at x = −1 ⇔ 2b + 5c = 7.

Solution.

f 0 (x) = 2(x − 2)(x2 + 2bx + c)(x + 3)3 + (x − 2)2 (2x + 2b)(x + 3)3
+ 3(x − 2)2 (x2 + 2bx + c)(x + 3)2
f 0 (−1) = 2(−3)(1 − 2b + c)8 + 72(−2 + 2b) + 108(1 − 2b + c)
= −48 + 96b − 48c − 144 + 144b + 108 − 216b + 108c = −84 + 24b + 60c

For −1 to be a critical point, f 0 (−1) = 0 ⇔ −84 + 24b + 60c = 0 ⇔ 2b + 5c = 7.

Question 2(b) Assuming that the above condition is satisfied, examine the nature of f (x)
at its three critical points.

Solution.

f 0 (x) = 2(x−2)(x2 +2bx+c)(x+3)3 +(x−2)2 (2x+2b)(x+3)3 +3(x−2)2 (x2 +2bx+c)(x+3)2

so it follows that the three critical points are x = 2, x = −3 and x = −1 when 2b + 5c = 7.

2
1. x = 2.
(x − 2)2 (x2 + 2bx + c)(x + 3)3 = (x − 2)2 g(x)
 
f (x) =
f 0 (x) = 2(x − 2)g(x) + (x − 2)2 g 0 (x)
f 00 (x) = 2g(x) + 4(x − 2)g 0 (x) + (x − 2)2 g 00 (x)
f 00 (2) = 2g(2) = 250(4 + 4b + c)
= 250(4 + 14 − 10c + c) = 250 · 9(2 − c)
There is a maximum at x = 2 if 2 < c, and a minimum at x = 2 if 2 > c. The test
fails when c = 2. Considering f 000 (x) when x = 2, c = 2, 2b = −3,
f 000 (x) = 2g 0 (x) + 4g 0 (x) + 4(x − 2)g 00 (x) + (x − 2)2 g 000 (x) + 2(x − 2)g 00 (x)
f 000 (2) = 6 (4 + 2b)(2 + 3)3 + 3(4 + 4b + c)(2 + 3)2
 
 
= 6 125 + 75(4 − 6 + 2) 6= 0
so there is no maximum or minimum at x = 2 when c = 2.
2. x = −3
(x + 3)3 (x − 2)2 (x2 + 2bx + c) = (x + 3)3 h(x)
 
f (x) =
f 0 (x) = 3(x + 3)2 h(x) + (x + 3)3 h0 (x)
f 00 (x) = 6(x + 3)h(x) + 3(x + 3)2 h0 (x) + 3(x + 3)2 h0 (x) + (x + 3)3 h00 (x)
f 00 (−3) = 0
f 000 (x) = 6h(x) + 12(x + 3)h0 (x) + (x + 3)2 [some polynomial in x]
f 000 (−3) = 6h(−3) = 150(9 − 6b + c) = 150(9 − 21 + 15c + c) = 600(4c − 3)
So if c 6= 34 , f (x) does not have a maximum or minimum at x = −3. If c = 34 , then

f (4) (x) = 18h0 (x) + (x + 3)[some polynomial in x]


 
so f (4) (−3) = 18 2(−5)(9 − 6b + c) + 25(−6 + 2b) = 18(−240 + 110b − 10c) < 0. Hence
if c = 43 , b = 13
8
, f (x) has a maximum at x = −3.
3. x = −1. Let y = x + 3. Then we need to consider the value y = 2.
f (y) = y 3 (y − 5)2 (y 2 + (2b − 6)y + 9 − 6b + c)
= y 3 (y 2 − 10y + 25)(y 2 + (1 − 5c)y − 12 + 16c)
= y 7 + (1 − 5c − 10)y 6 + (−12 + 16c + 25 − 10 + 50c)y 5
+(120 − 160c + 25 − 125c)y 4 + (400c − 300)y 3
= y 7 − (9 + 5c)y 6 + (3 + 66c)y 5 + (145 − 285c)y 4 + (400c − 300)y 3
f 00 (y) = 42y 5 − 30(9 + 5c)y 4 + 20(3 + 66c)y 3 + 12(145 − 285c)y 2 + 6(400c − 300)y
f 00 (2) = 1344 − 480(9 + 5c) + 160(3 + 66c) + 48(145 − 285c) + 12(400c − 300)
= 864 − 720c

3
So if c > 65 , then f 00 (−1) < 0 and f (x) has a maximum at x = −1. If c < 56 , f (x) has
a minimum at x = −1. If c = 56 , then this test fails, we consider higher derivatives at
y = 2 with c = 65 :

f 000 (y) = 210y 4 − 1800y 3 + 4932y 2 − 4728y + 1080

Now it is clear that f 000 (2) 6= 0, hence there is no maximum or minimum at x = −1


when c = 65 .

Question 2(c) Show that f (x) has at least three points of inflection irrespective of the
condition 2b + 5c = 7.

Solution. From the calculation of f 00 (−3) = 0 whether 2b + 5c = 7 or not. Since f (−3) =


f (2) = 0, there exists a real number b between −3 and 2 such that f 0 (b) = 0. Now f 0 (2) =
f 0 (b) = f 0 (−3) = 0, thus we get a1 ∈ (−3, b), a2 ∈ (b, 2) where f 00 (a1 ) = 0, f 00 (a2 ) = 0. Thus
there are least 3 points where f 00 (x) = 0.
Note: This question is not complete: A point of inflection is a point where the curvature
of a curve changes sign (the curve goes from convex to concave or vice versa). Any point a
is a point of inflection if f 00 (a) = 0 and the lowest order non-zero derivative at a is of odd
order, or equivalently, f 00 (a + ) and f 00 (a − ) have opposite signs in the neighborhood of a.
Thus f 00 (x) = 0 is a necessary condition for a point to be an inflection point, but is not
sufficient. f 00 (x) is a 5th degree curve. Take two of the roots and equate them, and two other
roots and equate them too. We can thus get two equations in b, c, which can be solved to get
their values. Now f 00 (x) = A(x − B)(x − C)2 (x − D)2 , where B, C, D are distinct. This curve
has only one point of inflection, B. For example, the similar curve g(x) = 35x3 − 21x5 + 5x7
has only one inflection point, although g 00 (x) has 3 roots.

Question 3(a) Prove that the n-th derivative of f (x)g(x) equals


n  
X n (n−i)
f (x)g (i) (x)
i=0
i

n

where f (m) denotes the m-th derivative of f (x) and i
are the binomial coefficients.

Solution. The proof is by induction on n. It is true for n = 1, as (f g)0 = f 0 g + f g 0 =


1
f (1) (x)g (0) (x) + 11 f (0) (x)g (1) (x) where f (0) (x) = f (x).
 
0
Induction hypothesis: Assume that the result is true for n = m.
m  
(m)
X m (m−i)
(f g) = f (x)g (i) (x)
i=0
i

4
Differentiating both sides,
m  
(m+1)
X m 
(f g) (x) = f (m−i+1) (x)g (i) (x) + f (m−i) (x)g (i+1) (x)
i=0
i
  m h   
m (m+1) (0)
X m m i (m−i+1)
= f (x)g (x) + + f (x)g (i) (x)
0 i=1
i i − 1
 
m (0)
+ f (x)g (m+1) (x)
m
  m  
m + 1 (m+1) (0)
X m + 1 (m−i+1)
= f (x)g (x) + f (x)g (i) (x)
0 i=1
i
 
m + 1 (0)
+ f (x)g (m+1) (x)
m+1
m+1
X m + 1
= f (m+1−i) (x)g (i) (x)
i=0
i

We used m0 = 1 = m+1
  m m+1
 m m
 m+1

0
, m
= 1 = m+1
, i
+ i−1
= i
. The result is now
established for n = m + 1, and hence by induction for all n.

Question 3(b) Show that the m-th derivative gm (x) of g(x) = tan−1 x satisfies

(1 + x2 )gm+1 (x) + 2mxgm (x) + m(m − 1)gm−1 (x) = 0

and hence is of the form


(−1)m−1 (m − 1)!φm (x)
gm (x) =
(1 + x2 )m
where φm (x) is a polynomial of dergree m given by φ1 (x) = 1, φ2 (x) = 2x and the recursion
φm+1 (x) = 2xφm (x) − (1 + x2 )φm−1 (x).

1
Solution. g1 (x) = or (1 + x2 )g1 (x) = 1.
1 + x2
Using Leibnitz’s theorem proved above and differentiating the above equation m times,
we get      
m 2 m m
(1 + x )gm+1 (x) + 2xgm (x) + 2gm−1 (x) = 0
0 1 2
or
(1 + x2 )gm+1 (x) + 2mxgm (x) + m(m − 1)gm−1 (x) = 0
which is the required relation.
1 (−1)1−1 (1 − 1)!φ1 (x)
Now g1 (x) = = as φ1 (x) = 1.
1 + x2 (1 + x2 )1

5
−2x (−1)2−1 (2 − 1)!φ2 (x)
g2 (x) = = as φ2 (x) = 2x.
(1 + x2 )2 (1 + x2 )2
Assume that the result is true for all n ≤ m. From the relation proved above:

(−1)m−1 (m − 1)!φm (x) (−1)m−2 (m − 2)!φm−1 (x)


(1 + x2 )gm+1 (x) + 2mx + m(m − 1) =0
(1 + x2 )m (1 + x2 )m−1
or
(−1)m (m − 1)!φm (x) (−1)m−1 (m − 2)!φm−1 (x)
gm+1 (x) = 2mx + m(m − 1)
(1 + x2 )m+1 (1 + x2 )m
m
(−1) m! h i
= 2 m+1
2xφm (x) − (1 + x2 )φm−1 (x)
(1 + x )
(−1)m m!
= φm+1 (x)
(1 + x2 )m+1

as φm+1 (x) = 2xφm (x) − (1 + x2 )φm−1 (x). This completes the proof.

Question 3(c) Find φm (x) for x ≤ 6. Can you find a general formula?

Solution.

φ1 (x) = 1
φ2 (x) = 2x
φ3 (x) = 2x · 2x − (1 + x2 ) = 3x2 − 1
φ4 (x) = 2x(3x2 − 1) − (1 + x2 )(2x) = 4x3 − 4x
φ5 (x) = 2x(4x3 − 4x) − (1 + x2 )(3x2 − 1) = 5x4 − 10x2 + 1
φ6 (x) = 2x(5x4 − 10x2 + 1) − (1 + x2 )(4x3 − 4x)
= 6x5 − 20x3 + 6x

Writing these down using binomial coefficients:


 
1 0
φ1 (x) = x
0
 
2 1
φ2 (x) = x
1
   
3 2 3 0
φ3 (x) = x − x
2 0
   
4 3 4 1
φ4 (x) = x − x
3 1

6
     
5 4 5 2 5 0
φ5 (x) = x − x + x
4 2 0
     
6 5 6 3 6 1
φ6 (x) = x − x + x
5 3 1
     
n n−1 n n−3 n
φn (x) = x − x + xn−5 − . . .
n−1 n−3 n−5
2r+1≤n   2r+1≤n  
X
r n n−2r−1
X
r n
= (−1) x = (−1) xn−2r−1
r=0
n − 2r − 1 r=0
2r + 1

Question 3(d) Show that the n-th derivative of ex tan−1 x at x = 0 equals


     
n n n
n − 2! + 4! − 6! + ...
3 5 7

Solution. Let g(x) = tan−1 (x). Using Leibnitz’s theorem, we get


n   n  
dn x −1
X n dn−r ex X n
(e tan x) = gr (x) n−r = gr (x)ex
dxn r=0
r dx r=0
r

Thus n   n  
dn x

−1
X n X n
(e tan x) = g r (0) = (−1)r−1 (r − 1)!φr (0)
dxn x=0 r=0
r r=1
r
as g(0) = 0 and gr (x) was computed above in terms of φr (x). Now φ2r (0) = 0, and φ2r+1 =
(−1)r . Thus
2r+1≤n
dn x
 X  n 
−1
(e tan x) = (2r)!(−1)r
dxn x=0 r=0
2r + 1
or
dn x
        
n n n n
(e tan−1 x) = 0! − 2! + 4! − 6! + ...
dxn x=0 1 3 5 7
as required.

Paper II
Z π
2
Question 4(a) Discuss the convergence of the integral log(sin x) dx and if convergent,
0
evaluate it.

7
Solution.π The integrand log sin x has a discontinuity only at x = 0. We consider the
Z
2
integral − log(sin x) dx so that the integrand is positive.
0
Let g(x) = x−m , 0 < m < 1, then

− log sin x − cos x 1 xm cos x x


lim = lim+ = lim =0
x→0+ x−m x→0 sin x −mx−m−1 x→0+ m sin x
as L’Hospital’s rule applies.
Z π Z π
2 2
But g(x) dx is convergent as 0 < m < 1, therefore − log(sin x) dx converges. Now
0 0
Z π Z π Z π Z π
2 2 2 2
log(sin 2x) dx = log 2 dx + log(cos x) dx + log(sin x) dx
0 0 0 0
Z π Z π
2 2
But log(cos x) dx = log(sin x) dx, so
0 0
Z π Z π
2 π 2
2 log(sin x) dx = − log 2 + log(sin 2x) dx
0 2 0

By putting 2x = t. π
Z Z π
2 1
log(sin 2x) dx = log(sin t) dt
0 2 0
Now substituting t = π − y, we have
π
Z π Z 0 Z
2
log(sin t) dt = log(sin y) (−dy) = log(sin y) dy
π π
2 2
0
Z π Z π
2 2
∴ log(sin 2x) dx = log(sin y) dy
0 0
Z π
2 π
⇒ log(sin x) dx = − log 2
0 2

2
Question 4(b) Find the point on the parabola
√ y = 2x, z = 0 which is nearest to the plane
z = x + 2y + 8. Show that this distance is 6.

Solution. Let the point (x, y, 0) by on the parabola. Then the distance d from the plane
z = x + 2y + 8 is given by

|x + 2y − 0 + 8| |x + 2y + 8|
d= √ = √
1+4+1 6

8
⇒ 6d2 = x2 + 4xy + 4y 2 + 16x + 32y + 64.
Put 2x = y 2 , and let F = 6d2 to get

y4 2
F = 6d = + 2y 3 + 4y 2 + 8y 2 + 32y + 64
4
We need to minimize F . The critical points are given by
dF
= y 3 + 6y 2 + 24y + 32 = 0
dy
⇒ (y + 2)(y 2 + 4y + 16) = 0

Thus y = −2 is the only real value.

d2 F
= 3y 2 + 12y + 24 = 12 − 24 + 24 > 0
dy 2
at y = −2. Hence y = −2 is a minimum. √
When y = −2, x = 2, and d = |2−4+8|

6
= 6 as required.

Question 4(c) Show that


ZZZ
1
(a2 b2 c2 − b2 c2 x2 − c2 a2 y 2 − a2 b2 z 2 ) 2 dx dy dz

x2 y 2 z 2 π 2 a2 b2 c2
over the volume bounded by 2 + 2 + 2 = 1 is equal to .
a b c 4
Solution. Let
ZZZ
1
I = 2
(a2 b2 c2 − b2 c2 x2 − c2 a2 y 2 − a2 b2 z 2 ) 2 dx dy dz
x2 2
{(x,y,z) | + y2 + z2 ≤1}
a2 b c

x2 y 2 z 2  21
ZZZ 
= 8abc 1 − 2 − 2 − 2 dx dy dz
{(x,y,z) | x2 2 2
+ y2 + z2 ≤1,x≥0,y≥0,z≥0} a b c
a2 b c

Let x = aX, y = bY, z = cZ, so that dx dy dz = abc dX dY dZ


ZZZ
2 2 2 2 2 2 12
I = 8a b c 2 2 2
(1 − X − Y − Z ) dX dY dZ
X +Y +Z ≤1
X≥0,Y ≥0,Z≥0

du dv dw
Let X 2 = u, Y 2 = v, Z 2 = w, so that dX = √ , dY
2 u
= √ , dZ
2 v
= √
2 w
and
ZZZ
1 1 1 1
2 2 2
I=a b c (1 − u − v − w) 2 u− 2 v − 2 w− 2 du dv dw
u+v+w≤1
u≥0,v≥0,w≥0

9
We now convert this to a Dirichlet integral — let u + v + w = α, v + w = αβ, w = αβγ,
∂(u, v, w)
so that = −α2 β and
∂(α, β, γ)
Z 1Z 1Z 1
1 1 1 1
I = abc 2 2 2
(1 − α) 2 (α − αβ)− 2 (αβ − αβγ)− 2 (αβγ)− 2 α2 β dα dβ dγ
Z0 1 Z0 1 Z0 1
1 1 1 1 1
= a2 b2 c2 (1 − α) 2 α 2 (1 − β)− 2 (1 − γ)− 2 γ − 2 dα dβ dγ
0 0 0
3 3
2 2 2 Γ( 2 )Γ( 2 ) Γ( 21 )Γ(1) Γ( 21 )Γ( 21 )
= abc
Γ(3) Γ( 32 ) Γ(1)
1 1 4
2 2 2 2 Γ( 2 ) π 2 a2 b2 c2
= abc =
2 4

as required, as Γ( 12 ) = π.

10
UPSC Civil Services Main 1991 - Mathematics
Calculus
Sunder Lal
Retired Professor of Mathematics
Panjab University
Chandigarh

January 14, 2010

Question 1(a) Sketch the curve (x2 − a2 )(y 2 − b2 ) = a2 b2 .

Solution. We note the following points:


1. The curve is symmetrical about both the coordinate axes. Enough to consider when
x ≥ 0, y ≥ 0.
2. (0, 0) lies on the curve.
3. The coefficient of the highest power of x is y 2 − b2 . Thus y = ±b are two asymptotes
parallel to the x-axis. The coefficient of the highest power of y is x2 − a2 , thus x = ±a
are two asymptotes parallel to the y-axis. For other asymptotes of the form y =
mx + c, we know that limx→∞ xy = m. We divide by x4 and let x → ∞ to obtain
 a2  y 2 b2  a2 b2
lim 1− 2 − − 4 = m2 = 0 so the only asymptotes of the form y = mx+c
x→∞ x x2 x2 x
have m = 0, and we have already found those. Thus the curve has only 4 asymptotes.
a2 b 2 a2 b2 b2 x 2 bx
4. y 2 − b2 = 2 2
⇒ y 2
= 2 2
+ b 2
= 2 2
⇒ y = ±√ . Since we need
x −a x −a x −a x 2 − a2
bx
to trace only in the first quadrant, we let y = √ .
x 2 − a2
dy b bx2 −a2 b
5. = √ − 3 = 3 , showing that the curve has no critical
dx x 2 − a2 (x2 − a2 ) 2 (x2 − a2 ) 2
points.
6. If x2 − a2 < 0 , then (x2 − a2 )y 2 = b2 x2 ⇒ x = 0, y = 0. Hence x2 − a2 > 0, so the
curve lies beyond the line x = ±a. Similarly it lies beyond the lines y = ±b. (0, 0) is
an isolated point.

1
d2 y 5
7. 2
= 32 a2 b(x2 − a2 )− 2 2x > 0 for x > a, so the curve is downwards convex.
dx
A sketch is shown here:

x = −a x=a
y=b

y = −b

Question 1(b) Find the cubic curve which has the same asymptotes as the curve

x3 − 6x2 y + 11xy 2 − 6y 3 + x + y + 1 = 0

and which passes through the points (0, 0), (1, 0) and (0, 1).

Solution. Consider the curve x3 − 6x2 y + 11xy 2 − 6y 3 + ax + by + c = 0. Since it has the


same terms of degree 3 and 2 as the given curve, it has the same asymptotes. Now since
it must pass through (0, 0), c = 0. Since it passes through (1, 0), 1 + a = 0 ⇒ a = −1,
and since it passes through (0, 1), −6 + b = 0 ⇒ b = 6. Hence the required curve is
x3 − 6x2 y + 11xy 2 − 6y 3 − x + 6y = 0.

Question 1(c) Show that the function f (x, y) = y 2 + x2 y + x4 has (0, 0) as the only critical
point and f (x, y) has a minimum at that point.

∂f ∂f
Solution. The critical points are given by = 2xy + 4x3 = 0 ⇒ x = 0, y = −2x2 , =
∂x ∂y
x2
2y + x2 = 0 ⇒ y = − .
2 2
When x = 0, y = − x2 ⇒ y = 0.
2
y = −2x2 , y = − x2 ⇒ x = 0, y = 0. Hence (0, 0) is the only critical point.

2
∂ 2f 2 ∂ 2f ∂ 2f
Now = 2y + 12x = 0 at (0, 0). = 2 and = 2x = 0 at (0, 0). Thus
∂x2 ∂y 2 ∂x ∂y
∂ 2 f ∂ 2 f  ∂ 2 f 2
− = 0 at (0, 0)
∂x2 ∂y 2 ∂x ∂y
so we cannot say whether it is maximum or minimum. However f (x, y) = (x2 + y2 )2 + 34 y 2 ≥ 0
for all (x, y) ∈ R2 , so f (x, y) has a minimum at (0, 0).

Question 1(d) Find the percentage error in the volume of a right circular cone when an
error of 1% is made in measuring the height and an error of 0.5% is made in measuring the
base radius.

Solution.
π 2
V = r h
3
π
log V = log + 2 log r + log h
3
dV dr dh
= 2 +
V r h
dr dh dV
Now = 0.5% = 0.005, = 1% = 0.01. Thus = 0.02 = 2%. Hence the error in
r h V
measuring the volume is 2%.
ZZ
Question 2(a) Evaluate F (x + y)xm−1 y n−1 dx dy with F (u) = (1 − u)l−1 where D is
D
the interior of the triangle formed by x = 0, y = 0, x + y = 1 and l, m, n are all positive.

Solution. (Note: This is Dirichlet’s integral.)


∂(x, y) v u
Put u = x + y, uv = x ⇒ y = u − uv, so that = = −u.
∂(u, v) 1 − v −u
Z 1Z 1
I = F (u)(uv)m−1 un−1 (1 − v)n−1 u du dv
Z0 1 Z0 1
= F (u)um+n−1 v m−1 (1 − v)n−1 du dv
0 0
Γ(m)Γ(n) 1
Z
= F (u)um+n−1 du
Γ(m + n) 0
Γ(m)Γ(n) 1
Z
= (1 − u)l−1 um+n−1 du
Γ(m + n) 0
Γ(m)Γ(n) Γ(l)Γ(m + n) Γ(m)Γ(n)Γ(l)
= =
Γ(m + n) Γ(m + n + l) Γ(m + n + l)

3
x2 y 2 z 2
Question 2(b) Find the center of gravity of the solid bounded by the ellipsoid 2 + 2 + 2 =
a b c
1 and lying in the positive octant.
ZZZ
Solution. Let S be the positive octant of the ellipsoid. Let M = dx dy dz.
ZZZ ZZZ Z Z ZS
1 1 1
Then x = x dx dy dz, y = y dx dy dz, z = z dx dy dz.
M S M S M S
Now put x = aX, y = bY, z = cZ so that
ZZZ
π
M = abc dX dY dZ = abc
X≥0,Y ≥0,Z≥0,
2 2 2
6
X +Y +Z ≤1

because the volume of a sphere of radius 1 is 4π


3
.

6a2 bc
ZZZ
x= X dX dY dZ
πabc X≥0,Y ≥0,Z≥0,
2 2 2
X +Y +Z ≤1

Put X 2 = u, Y 2 = v, Z 2 = w. Then

ZZZ
6a 1 1
−1 1 −1 1 −1
x = u≥0,v≥0,w≥0, u 2 u 2 v 2 w 2 du dv dw
π D: 2 2 2
u+v+w≤1
ZZZ ZZZ
3a 1 1 3a 1 1
= u0 v − 2 w− 2 du dv dw = u1−1 v 2 −1 w 2 −1 du dv dw
4π D 4π D
1 1
3a Γ(1)Γ(