Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

Linear Algebra in Sage

Linear Algebra Tutorial
Sage Days 15
May 16, 2009
Robert A. Beezer
University of Puget Sound

Graphic: Gidon Eshel, Univ. of Chicago

Solving Systems of Equations

Create a matrix and a vector.

A is a 3 3 nonsingular matrix.
1 of 14

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

b is a 3-slot vector.
A = matrix([[2,1,1/3],[-1,6,2],[1/2,1,8]])
A

2
B
@ 1

1
3

1
6
1

1
2

C
2A
8

b = vector([14/3, 2, -6])
b

( 14
3 ; 2; 6)
Solve Ax = b.
Solve commands ("right" is location of solution vector).

A.solve_right(b)

(2; 1; 1)

print A.det()
print A.det() == 0
299/3
False
A.inverse()

B
@

6
13
27
299
12
299

1
13

95
598
9
598

1 C
23
A
3
23

Vectors are rows or columns as appropriate, compute A1 b

A.inverse()*b

(2; 1; 1)
Can "divide" by a matrix
b/A
770
( 299
;

44
20
897 ; 23 )

b/A.transpose()

2 of 14

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

(2; 1; 1)
Augment and row-reduce
R = A.augment(matrix(b).transpose())
R

1
3

1
6
1

2
B
@ 1

1
2

14
3

2
8

R.echelon_form()

1
@0
0

0
1
0

0
0
1

C
2A
6

1
2
1A
1

Properties of Matrices
B is a 6 5 matrix of rank 4 over the rationals Q
B = matrix(QQ,[
[10,0,3,8,7],
[-16,-1,-4,-10,-13],
[-6,1,-3,-6,-6],
[0,2,-2,-3,-2],
[3,0,1,2,3],
[-1,-1,1,1,0]
])
B

10
B
B 16
B
B 6
B
B
0
B
B
@
3
1

0
1
1
2
0
1

3
4
3
2
1
1

8
10
6
3
2
1

B.right_kernel()

RowSpanQ 1

23

1
2

B.right_kernel().basis()

3 of 14

1
7
C
13 C
C
6 C
C
2 C
C
C
3A
0
1

21

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

[(1; 23 ; 21 ; 1; 21 )]

B.left_kernel().basis()

[(1; 0; 3; 1; 3; 1) ; (0; 1; 2; 1; 1; 1)]

B.row_space()

1 0 0 0
B
B0 1 0 0
RowSpanQ B
@0 0 1 0
0 0 0 1

B.column_space()

1 0 0 0
B
B0 1 0 0
RowSpanQ B
@0 0 1 0
0 0 0 1

print
print
print
print
print
print
print

1
2
C
3 C
C
1A
2

41
41
41
0

41

3
4
49

C
C
C
A

"Rank", B.rank()
"Right Nullity", B.right_nullity()
"Columns", B.ncols()
"Rank", B.rank()
"Left Nullity", B.left_nullity()
"Rows", B.nrows()

Rank 4
Right Nullity 1
Columns 5
Rank 4
Left Nullity 2
Rows 6

C is a random 50 50 matrix over Q

C = random_matrix(QQ, 50, num_bound=10, den_bound=10)
C

4 of 14

06/13/2011 11:13 AM

B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B

Linear Algebra in
BSage -- Sage

http://localhost:8000/home/admin/21/print

7
8
73
4
3

2
5

10
89

2
3
4
5
7

3
5
25

( 52 ; 4;

3
5

2
5
45
45

10
2
31
6
1
1
94

1
5

4
1

7
2
9
21
1
4
4

1
8
25

Column 5 of the matrix.

C[4]

1
35
2

3 1
10 ; 5 ;

3
59
53
65

3
8

8
4
9

5
53
32

9
10

8
25
9

1
34
5
0
9
2
7
65
5
1

4
3
1
3

4
1
2
7
2

10

9
7
8

4
2
2
59
43
1
25
8
7
10
5
1
1
6
1
5
3

1
51
5
87
1
0
4
5
5
2

2
2

2
5

10

4; 51 ; 23 ; 54 ; 61 ; 81 ; 0; 10; 0; 0; 76 ; 1; 2; 25 ; 6; 1;

Python slicing for entries, show() for looks

show(C[20:30, 25:35])

5 of 14

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@

8
9
7
4
9
2
3
8
1
3
1
4

http://localhost:8000/home/admin/21/print

3
5
4
5

8
2

10
41
67

1
2

1
3
1
2
98
32
61
45

0
3
0
89

25
2
71
41

4
7
91
1
3

1
4
21

95
47
41
9
25
1
9
2

3
7
9

5
2

10
7

1
21
10

1
2
2
5

1
5
2
7
2
10
7

0
21

8
5
5
6

34
76
0
6
56
1
1
51
9
10
9

7
6
31
58

1
4
7

3
5

1
4
4
5

91

1
1
5
1
2

3
2

10

0
7
10
2
0
1
76
1

7
6
51

0
0
0

2
7
21

C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A

C.det()

187558143411975607125910504689634916293080048459801797630721278449241962022353955141284605117745
668930993248348814342449533378685145998341315304441572331718197554296221857483527

C.trace()

15937
1260

C.norm()

40:7778652883

EigenStu
D is a 4 4 matrix, not diagonalizable
D = matrix(QQ, [
[-2,1,-2,-4],
[12,1,4,9],
[6,5,-2,-4],
[3,-4,5,10]
])
show(D)

2
B
B 12
B
@ 6
3

6 of 14

1
1
5
4

2
4
2
5

1
4
C
9C
C
4 A
10

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

Build a characteristic polynomial for D (then simplify)

var('t')
S = D - t*identity_matrix(4)
print S
print
p(t)=S.det()
p(t)
[ -t - 2
[
12
[
6
[
3

1
-t + 1
5
-4

-2
-4]
4
9]
-t - 2
-4]
5 -t + 10]

t4 7 t3 + 18 t2 20 t + 8

p.find_root(-10, 10)

1:9999948114
show(p.factor())

(t 2)3 (t 1)
Easier, but less instructive: : :
q(t) = D.charpoly('t')
q

t 7! (((t 7)t + 18)t 20)t + 8

D.eigenvalues()

[1; 2; 2; 2]
diag,
print
print
print
print
print

evecs = D.eigenmatrix_right()
"Diagonal matrix with eigenvalues"
diag
"Matrix with eigenvectors as columns"
evecs

Diagonal matrix with eigenvalues

[1 0 0 0]
[0 2 0 0]
[0 0 2 0]
[0 0 0 2]

7 of 14

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

Matrix with eigenvectors as columns

[ 1 1 0 0]
[-3 -2 0 0]
[-3 1 0 0]
[ 0 -2 0 0]
Jordan Canonical Form
D.jordan_form()

1
0 0 0
0
1C
2 1 0 C
C
@0 2 1AA
0 0 2

1
B0 1
B 0
B
@ @0A
0

J, P = D.jordan_form(transformation = True)
print J
print
print P
[1|0 0 0]
[-+-----]
[0|2 1 0]
[0|0 2 1]
[0|0 0 2]
[ 1
5
[ -3 -10
[ -3
5
[ 0 -10

-8
21
2
11

1]
0]
0]
1]

Check the results

P^(-1)*D*P

1
B
B0
B
@0
0

0
2
0
0

1
0 0
C
1 0C
C
2 1A
0 2

Manipulate the basis for the Jordan form representation

Q = P.transpose().gram_schmidt()[0].transpose()
Q

8 of 14

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

75
19
130
19
155
19

1
B 3
B
B
@ 3
0

20
29
70
87
50
87
65
87

10

15
14

0C
C
C
5 A
14
5
7

Orthogonal?
Q.transpose()*Q

19
B 0
B
B
@ 0
0

4350
19

0
0

1
0
0C
C
C
0A

0
0

175
87

25
14

Resulting representation?
Q^(-1)*D*Q

1
B
B0
B
@0
0

20
19

2
0
0

1
535
133
698 C
609
C
C
1A

2915
1653
1
2
0

Decompositions
LU, QR, SVD, Jordan Canonical Form, Smith Normal Form, Cholesky
Decomposition, : : :
Convert D to a matrix E over the reals (double precision), R, to obtain QR
decomposition
E = D.change_ring(RDF)
E

2:0
B
B 12:0
B
@ 6:0
3:0

1:0
1:0
5:0
4:0

2:0
4:0
2:0
5:0

1
4:0
C
9:0 C
C
4:0 A
10:0

ortho, triangular = E.QR()

print "Orthogonal"

9 of 14

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

print ortho
print
print "Upper Triangular"
print triangular
print
Orthogonal
[ -0.14396315015 -0.206755085285
0.83647166171 0.486664263392]
[ 0.863778900898
0.11873886424 0.366775046298 -0.324442842262]
[ 0.431889450449 -0.661782341253 -0.372388950068 0.486664263392]
[ 0.215944725225 0.710772502024 -0.164674510583 0.648885684523]
Upper Triangular
[ 13.8924439894
[
-0.0
[
-0.0
[
-0.0

2.0154841021
3.95898662912
8.78175215913]
-6.24001793541
5.76589282015
11.6505245045]
0.0 -0.284437791007 -0.202100535715]
-0.0
-0.0 -0.324442842262]

Checks
(ortho.transpose()*ortho - identity_matrix(4)).norm()

5:57101754397 1016

(ortho*triangular - E).norm()

5:59659361322 1015

Vector Spaces
Sage is so much more than numerical computation.
Can work naturally with vector spaces and modules over a variety of elds and
rings.
F.<a> = FiniteField(3^2)
F

F32
V is a 3-dimensional vector space over F
V=F^3
V

F332
10 of 14

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

V.list()

[(0; 0; 0) ; (2a; 0; 0) ; (a + 1; 0; 0) ; (a + 2; 0; 0) ; (2; 0; 0) ; (a; 0; 0) ; (2a + 2; 0

"Generator" of all 2-D subspaces of V
subs = V.subspaces(2)

Python "list comprehension"

all_subs = [U for U in subs]

Grab one of the subspaces, #42

all_subs[42]

How many such subspaces?

How many basis matrices in echelon_form?
all_subs[86]
len(all_subs)

= (32 )2 + (32 )1 + 1

11 of 14

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

Accuracy
Octave and Matlab emphasize numerical results - everything is a oating point
number.
Their rational form is deceptive. Example by William Stein:
octave:1> format rat;
octave:2> a = [-86/17,40/29,-68/43,-20/11;-24/17,-1/38,-2/25,49/17]
a=
-86/17
40/29
-68/43
-20/11
-24/17
-1/38
-2/25
49/17
octave:3> rref(a)
ans =
1
0 155/2122 -725/384
0
1 -152/173 -6553/795
and in Matlab:
>> format rat;
>> a = [-86/17,40/29,-68/43,-20/11;-24/17,-1/38,-2/25,49/17]
a=
-86/17
-24/17

40/29
-1/38

-68/43
-2/25

-20/11
49/17

>> rref(a)
ans =
1
0

0
1

13/178
-152/173

-725/384
-1426/173

Now in Sage:
F = matrix(2,[-86/17, 40/29, -68/43, -20/11, -24/17, -1/38,
-2/25, 49/17])
show(F.echelon_form())

Entry in lower right corner:

12 of 14

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

print N(-6553/795, digits = 9), " Octave"

print N(-1426/173, digits=9), " Matlab"
print N(-30037214/3644069, digits = 9), " Sage"

Speed
Matrices with symbolic entries
var('x y')
n=6
entries = [x^i-y^j+i+j for i in range(1,n+1) for j in
range(1,n+1)]
G = matrix(SR, n, entries)
G

xy+2
B 2
Bx y+3
B 3
Bx y+4
B
B x4 y + 5
B
B 5
@ x y+6
x6 y + 7

G.det()

y 2 + x + 3
x2 y 2 + 4
x3 y 2 + 5
x4 y 2 + 6
x5 y 2 + 7
x6 y 2 + 8

y 3 + x + 4
x2 y 3 + 5
x3 y 3 + 6
x4 y 3 + 7
x5 y 3 + 8
x6 y 3 + 9

y 4 + x + 5
x2 y 4 + 6
x3 y 4 + 7
x4 y 4 + 8
x5 y 4 + 9
x6 y 4 + 10

y 5 + x + 6
x2 y 5 + 7
x3 y 5 + 8
x4 y 5 + 9
x5 y 5 + 10
x6 y 5 + 11

0
G.det().simplify_full()

0
time G.det()

0
Time: CPU 0.00 s, Wall: 0.00 s
Reals, 53-bit precision
H = random_matrix(RR, 10)
time H.det()

3:07894731367045

Time: CPU 0.12 s, Wall: 0.15 s

Reals, 200-bit precision

13 of 14

06/13/2011 11:13 AM

Linear Algebra in Sage -- Sage

http://localhost:8000/home/admin/21/print

J = random_matrix(RealField(200), 10)
time J.det()

9:1432891763792026925313110748493443825766112531483542162293
Time: CPU 0.14 s, Wall: 0.14 s
Reals, Interval Arithmetic
K = (1/17.0)*random_matrix(RIF, 10, bound = 10)
time K.det()
Traceback (click to the left of this block for traceback)
...
TypeError: random_element() got an unexpected keyword argument
'bound'
Reals, Double Precision
L = random_matrix(RDF, 300)
timeit("L.det()")
5 loops, best of 3: 61.3 ms per loop
Rationals (Exact)
M = random_matrix(QQ, 800, num_bound = 10, den_bound = 10)
time M.det()

Time: CPU 145.71 s, Wall: 146.92 s

14 of 14

06/13/2011 11:13 AM