MATH3705 Notes - By Eric Hua Contents

I LAPLACE TRANSFORM
1.1 Denition of LT . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.2 Existence . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.3 Laplace Transform of Periodic Functions . . . . . . . . . 1.1.4 The Gamma Function . . . . . . . . . . . . . . . . . . . DIFFERENTIATION, INTEGRATION AND LT . . . . . . . . 1.2.1 LT of the derivative of a function . . . . . . . . . . . . . 1.2.2 LT of the integral of a function . . . . . . . . . . . . . . 1.2.3 The Derivative of LT of a Function . . . . . . . . . . . . 1.2.4 The Integral of LT of a Function . . . . . . . . . . . . . SHIFTING THEOREMS . . . . . . . . . . . . . . . . . . . . . . 1.3.1 The First Shifting Theorem . . . . . . . . . . . . . . . . 1.3.2 The Second Shifting Theorem . . . . . . . . . . . . . . . CONVOLUTION AND THE DIRAC DELTA DISTRIBUTION 1.4.1 Convolutions . . . . . . . . . . . . . . . . . . . . . . . . 1.4.2 Dirac delta function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3
3 3 4 5 6 7 7 9 10 13 14 14 14 18 18 20

1.2

1.3

1.4

II Series Solutions of ODEs

2.1 2.2 2.3 2.4 2.5 2.6 2.7 Taylor Series . . . . . . . . Analytic functions . . . . . Singular point and ordinary Power Series Solution . . . . Cauchy-Euler Equations . . The Frobenius Method . . . Bessels Equation . . . . . . . . . . . . . . point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22
22 22 23 24 26 31 36

3 III 3.1 3.2 3.3 3.4

Fourier Series
Pre-knowledge . . . . . . . (Full) Fourier series . . . . 2 -periodic function . . . Points of discontinuity and . . . . . . . . . . . . . . . . . . . . . convergence 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

38
38 39 40 42

3.5 3.6 3.7 3.8

Geometric interpretation of Fourier series . . . . . . . . . . . . . . . . . . General 2L-periodic function . . . . . . . . . . . . . . . . . . . . . . . . . Fourier cosine and sine Series on interval (0, L): Even and odd extensions (Full) Fourier Series on Interval [a, b] . . . . . . . . . . . . . . . . . . . .

. . . .

43 43 44 47

4 IV Partial Dierential Equations 4.1 What is a Partial Dierential Equation? . . . . . . . . . . . . . . . 4.2 Heat Equation or Diusion Equation . . . . . . . . . . . . . . . . . 4.2.1 Homogeneous Boundary Conditions: Separation of Variables 4.2.2 Non-homogeneous Boundary Conditions . . . . . . . . . . . 4.3 The Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 Dirichlet Problem and Separation of Variables . . . . . . . . 4.4 Laplace Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.1 Solve the Laplaces equation inside a rectangle . . . . . . . 4.4.2 Laplace equation inside a circle . . . . . . . . . . . . . . . . 4.4.3 Laplace equation outside a circle . . . . . . . . . . . . . . . 4.4.4 Laplace equation within an annulus . . . . . . . . . . . . . . 5

49
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 49 50 53 54 54 59 59 63 65 66

V Sturm-Liouville Problems
5.1 5.2 5.3 Sturm-Liouville Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . Regular and Singular Sturm-Liouville Problems . . . . . . . . . . . . . . . Bessels equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

68
68 73 74

VI Fourier Transform
6.1 6.2 6.3 6.4 6.5 Fourier transform and its inverse Linearity . . . . . . . . . . . . . Shifting theorem . . . . . . . . . Fourier transform and derivatives Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

78
78 79 80 81 82

1
1.1
1.1.1

I LAPLACE TRANSFORM
Denition of LT
Denition

The Laplace Transform is widely used in engineering applications, such as solving linear ordinary dierential equations. It transforms the equation in t-space to one in s-space. This makes the problem much easier to solve. Let f (t) be a function dened on [0, ). We may assume f (t) = 0 when t < 0. The Laplace transform (LT) of f (t) is the function F (s), dened by: F (s) = L {f (t)} =
0

est f (t) dt.

This is named for Pierre-Simon Laplace, one of the best French mathematicians in the mid-to-late 18th century.The LT transforms functions of t to functions of another variable s. If F (s) is the Laplace transform of f (t), then f (t) is the inverse Laplace transform of F (s): f (t) = L1 {F (s)}. Proposition 1. The LT and L1 are linear: L {af (t) + bg (t)} = aL {f (t)} + bL {g (t)} L1 {aF (s) + bG(s)} = aL1 {F (s)} + bL1 {G(s)}

Examples of Laplace Transforms f (t) for t 0 L(f ) 1 eat , s>a tn sin at cos at n! sn+1 1 s 1 sa (n = 0, 1, . . .) a + a2 s 2 s + a2 s2 } { 1 1 st = . dt = e s s 0

For example, L {1} = Since

st

ebt sin atdt =

ebt (b sin at a2 +b2 st

a cos at), we have { } est a = 2 (s sin at a cos at) . 2 2 a +s s + a2 0

L {sin at} =
0

e {

sin at dt =

L e 1.1.2 Existence

at

=
0

(as)t

} { 1 1 (as)t e = . dt = sa sa 0

For which functions f is the LT actually dened on? We want the indenite integral to converge, of course. A function f (t) is piecewise continuous on a nite interval [a, b] if [a, b] can be subdivided into a nite number of subintervals such that f is continuous in each of the subintervals, and it approaches a nite limit when t approaches an end of any of the subintervals. Function f is piecewise continuous on an innite interval if it is piecewise continuous on any nite subinterval of its domain. A function f (t) is of exponential order if there exist constants t0 and M such that |f (t)| < M et , for all t > t0 .
2

For example, tn and et are of exponential order, et is not of exponential order. 4

Theorem 1. Suppose that f (t) is piecewise continuous and of exponential order with |f (t)| < M et , for all t > t0 .

Then the Laplace transform of f (t) exists for all s > . Proof. |F (s)| M
0

e(s)t dt =

M < . s

Remark. The condition in this theorem is sucient but not necessary. For example, the function f (t) = tn , 1 n < 0 is not piecewise continuous. But the Laplace transform of f (t) exists. Remark. If f (t) satises the conditions in the theorem above, then F (s) 0 when s . { } 2 Example 1. L et does not exist. Proof. { 2} t s2 /4 L e e
0 (ts/2)2 s2 /4

dt e

1 dt = .

1.1.3

Laplace Transform of Periodic Functions

A function f (t) is periodic if there exists > 0 such that f (t + ) = f (t) for all t. is a period of f . The smallest positive period of f is called the fundamental period of f , and f is called -periodic. Theorem 2. If f is called -periodic for t 0, then 1 L {f (t)} = est f (t) dt. 1 es 0 Proof. L {f (t)} =
0 n=0 n=0 0

st

(n+1)

f (t) dt =

st

f (t) dt =

es(x+n) f (x) dx,

where x = t n . Example 2. If f (t) =

t, 0 t < 1; and 2-periodic, then 0, 1 t < 2. L {f (t)} = 1 (1 + s)es . s2 (1 e2s ) 5

{ Example 3. Half-rectied wave: f (t) = L {f (t)} =

sin t, 2k t < (2k + 1) ; Then 0, (2k + 1) t < (2k + 2) .

1 . (s2 + 1)(1 es ) { sin t, 2k t < (2k + 1) ; Example 4. Full-rectied wave: f (t) = Then sin t, (2k + 1) t < (2k + 2) . L {f (t)} = 1.1.4 The Gamma Function (s2 1 s coth . + 1) 2

To nd Laplace transform of tn where n is not an integer, we need the Gamma function: (p) = ex xp1 dx (p > 0).
0

Properties: (1) = 1. It is divergent when p = 0. (p + 1) = p(p) for any p by integration by parts, except for 0, -1, -2, -3, . (n + 1) = n! )= Example 5. ( 1 2 .
(2k1)!! . 2k

+1 Example 6. ( 2k2 )= 1 ( 2k2 )=

2k . (2k1)!!
(p+1) . sp+1

Theorem 3. For any p > 1, L{tp } = Proof. Let x = st. Example 7. L{t0.5 } = L{t0.5 } =
(0.5+1) s0.5+1 (0.5+1) s0.5+1

1 . 2s0.5+1

1.2
1.2.1

DIFFERENTIATION, INTEGRATION AND LT

LT of the derivative of a function
0

In the denition of the LT, replace f (t) by its derivative f (t): L {f (t)} = Now integrate by parts (u = e
0 st

est f (t) dt.

, dv = f (t) dt):
0

st

f (t) dt =

f (t)est | 0

f (t) (s) est dt = f (0) + sL {f (t)} .

Therefore, if F (s) is the LT of f (t) then sF (s) f (0) is the LT of f (t): L {f (t)} = sL {f (t)} f (0). Replace f by f in (1), L {f (t)} = sL {f (t)} f (0), and apply (1) again: L {f (t)} = s2 L {f (t)} sf (0) f (0), This is called a derivative theorem for the LT. By induction, we have Theorem 4. Suppose f and its derivatives of order up to n 1 are continuous and of exponential order with |f (j ) (t)| Keat for all t > M , 0 j n 1, and that f (n) is piece-wise continuous, then L{f (n) } exists for s > a and { } L f (n) (t) = sn L {f (t)} sn1 f (0) sn2 f (0) f (n1) (0). Using (1) and (3), the LT of any constant coecient ODE ax (t) + bx (t) + cx(t) = f (t) is
a(s2 L {x(t)} sx(0) x (0)) + b(sL {x(t)} x(0)) + cL {x(t)} = F (s),

(1)

(2)

(3)

where F (s) = L {f (t)}. In particular, the LT of the solution, X (s) = L {x(t)}, satises 7

F (s) + asx(0) + ax (0) + bx(0) . as2 + bs + c Note that the denominator is the characteristic polynomial of the DE. X (s) = Example 8. Using the Laplace transform solve the dierential equation f 4f + 3f = 1 with boundary conditions f (0) = f (0) = 0. Solution: Take the Laplace transform of the equation. Since f (0) = f (0) = 0, if L(f ) = F (s) then L(f ) = sF (s) and L(f ) = s2 F (s). Thus, s2 F 4sF + 3F = and so (s2 4s + 3)F = 1 s 1 1 F = 2 s s 4s + 3 1 s

(4)

and, since s2 4s + 3 = (s 3)(s 1), this gives F = 1 s(s 3)(s 1)

Before we can invert this, we need to do a partial fraction expansion. 1 A B C = + + s(s 3)(s 1) s s3 s1 1 = A(s 3)(s 1) + Bs(s 1) + Cs(s 3)

(5)

So substituting in s = 0 we get A = 1/3, s = 3 gives B = 1/6 and s = 1 gives C = 1/2. Hence 1 1 1 F = + 3s 6(s 3) 2(s 1) and so f (t) = 1 1 3t 1 t + e e. 3 6 2

Example 9. Using the Laplace transform solve the dierential equation f 4f + 3f = 0 with boundary conditions f (0) = 1 and f (0) = 1. 8

Solution: In this example there are non-zero boundary conditions. Since L(f ) = sF f (0) L(f ) = s2 F sf (0) f (0) the subsidiary equation in this case is s2 F s 1 4sF + 4 + 3F = 0 so (s2 4s + 3)F = s 3. Hence F = and f (t) = et 1.2.2 LT of the integral of a function 1 s1 (6) (7)

Theorem 5. If f is continuous and L{f } exists, then { t } L{f } L f (x)dx = . s 0 Equivalently, L Proof. Let g (t) = t
0 1

L {f } s

} =

f (x)dx.

f (x)dx. Then

L{f (t)} = L{g (t)} = sL{g (t)} g (0) = sL{g (t)}. Example 10. } sin at 1 = L , 2 2 s +a a { } t 1 cos at 1 sin ax 1 L dx = , = 2 2 s(s + a ) a a2 0 { } t 1 at sin at 1 cos ax 1 L dx = . = 2 2 2 2 s (s + a ) a a3 0 t Example 11. Solve the integral equation f (t) + 0 f (x)dx = 1. {
1

Solution: Taking the Laplace transform we get 1 1 F (s) = , F (s) = , f (t) = L1 F (s) + s s s+1 9

1 s+1

} = et .

1.2.3

The Derivative of LT of a Function

0

Dierentiate the denition of the LT with respect to s: F (s) = Repeating this: dn F (s) = (1)n dsn This gives: Theorem 6. If f (t) is piecewise continuous and of exponential order, and L{f (t)} = F (s), then dn L {tn f (t)} = (1)n n F (s). ds Equivalently, { (n) } 1 F (s) 1 nL L {F (s)} = (1) . tn These formulas are used in the following cases: (i) Find the Laplace transforms of function with the forms tn f (t), when the transform of f (t) is easy to nd. (ii) Find the inverse transform of F (s) if the inverse transform of the derivative of F (s) is easy to nd. As we know the LT of f (t) = eat is F (s) = (s a)1 . By the theorem we have Example 12. L {teat } = F (s) = (s a)2 , L {t2 eat } = F (s) = 2 (s a)3 , L {t3 eat } = F (s) = 2 3 (s a)4 , ... , and in general { } L tn eat = (1)n F (n) (s) = n! (s a)n1 . Example 13. Using the Laplace transform solve the dierential equation f 4f + 3f = 2et with boundary conditions f (0) = f (0) = 0. (9)
0

est tf (t) dt.

est tn f (t) dt.

(8)

10

Solution: This time we have L(2et ) = 2/(s 1) on the right hand side. This means that the subsidiary equation is (s2 4s + 3)F = so F = 2 s1

2 (s 1)2 (s 3)

We need to do partial fractions again, but this is one of those cases with a repeated root: 1 A B C = + + 2 2 (s 1) (s 3) s 1 (s 1) s3 and multiplying across 1 = A(s 1)(s 3) + B (s 3) + C (s 1)2 so s = 1 gives B = 1/2 and s = 3 gives C = 1/4. No value of s gives A on its own, so wee try s = 2: 1 1 1 = A + + 2 4 which means that A = 1/4. Hence F = and 1 1 1 + 2 2(s 1) (s 1) 2(s 3) 1 1 f = et tet + e3t 2 2

Example 14. Using the Laplace transform solve the dierential equation y 2ay + a2 y = 0 with boundary conditions y (0) = 1 and y (0) = 0. a is some real constant. Solution: Taking the Laplace transform we get s2 Y 1 2aY + a2 Y = 0 and hence Y = which means that y = teat 11 1 (s a)2

Example 15. Using LT to solve the DE x + x = t100 et , x(0) = 0.

Note this would be highly impractical to solve using undetermined coecients. (You would have 101 undetermined coecients to solve for!) Solution: First, we compute the LT of the solution to the DE. The LT of the LHS: by (1), L {x + x} = sX (s) + X (s), where F (s) = L {f (t)} (s). For the LT of the RHS, let { } F (s) = L et = By (8), { 100 t } d100 d100 1 F ( s ) = L t e = . ds100 ds100 s + 1 The rst several derivatives of
1 s+1

1 . s+1

are as follows:

d 1 1 d2 1 1 d3 1 1 = , = 2 , = 62 , 2 2 3 3 ds s + 1 (s + 1) ds s + 1 (s + 1) ds s + 1 (s + 1)4 and so on. Therefore, the LT of the RHS is: d100 1 1 = 100! . 100 ds s + 1 (s + 1)101 Consequently, X (s) = 100! Using (9), we can compute the ILT of this: { 100! 1 (s + 1)102 } { } 1 1 1 1 101 t = L 101! = t e . 101 (s + 1)102 101 1 . (s + 1)102

x(t) = L

{X (s)} = L

12

1.2.4

The Integral of LT of a Function

Theorem 7. Let f (t) be piecewise continuous and of exponential order, and L{f (t)} = f (t) F (s). If lim exists, then t +
t0

{ L Equivalently, L Proof. Let g (t) =

f ( t) . t 1

f (t) t

} =

F (x)dx. {
s

{F (s)} = tL

} F (x)dx .

Then f (t) = tg (t), F (s) = L{f } = G (s).

These formulas are used in the following cases: t) (i) Find the Laplace transforms of function with the forms f ( , when the transform of t f (t) is easy to nd. (ii) Find the inverse transform of F (s) if the inverse transform of the integral of F (s) is easy to nd. Example 16. { L Example 17. { L
1

sin at t

} =

a dx = arctan(s/a), x2 + a2 2 {

a > 0.

1 (s 1)3

} = = = = =

} 1 tL dx (x 1)3 s { } 1 1 t/2L (s 1)2 { + } 1 1 (t/2)tL dx (x 1)2 s { } 1 2 1 t /2L s1 2 t te 2

1 +

13

1.3
1.3.1

SHIFTING THEOREMS
The First Shifting Theorem

Theorem 8. If F(s) exists for s > c 0, then, for any constant a < s c, { } L eat f (t) = F (s a). Equivalently, L1 {F (s a)} = eat L1 {F (s)} . Example 18. { } L ebt sin at = { a . (s b)2 + a2

Example 19.

} } { s s 1 L =L s2 + 6s + 10 (s + 3)2 + 1 { } { } s+33 s3 1 3t 1 =L =e L = e3t (cos t 3 sin t). (s + 3)2 + 1 s2 + 1

1

1.3.2

The Second Shifting Theorem 1 u(t a) = 0

Dene the unit step (Heaviside) function by for t a for t < a,

where a > 0. In some books, u(t a) = Ha (t). Example 20. Compute the Laplace transform of the unit step function: L[u(t a)] = =

est dt {a st } e = s a as e = , s for s > 0.

est u(t a) dt

14

Assuming f (t) = 0, t < 0. The transform of a function f (t), t 0, to u(t a)f (t a) is called a shift of f (t) by a units. f (t a) for t a u(t a)f (t a) = 0 for t < a, Theorem 9. (The Second Shift Theorem) If F(s) exists for s > c 0, then, for any constant a 0, L {u(t a)f (t a)} = eas L{f (t)}. Equivalently, { } [ ] L1 eas F (s) = u(t a) L1 {F (s)} ta . If f (t) = (t 1)2 u(t 1), then L[f (t)] = es L[t2 ] = 0 for t < 0, for 0 t 2, for t > 2. 2es . s3

Example 21.

Let f (t) =

1 1

This can be expressed as f (t) = u(t) + 2u(t 2), so

L[f (t)] = L[u(t)] + 2L[u(t 2)] 1 2e2s = + . s s Let f (t) = sin(t)u(t ). Then L[f (t)] = es L[sin(t + )] = es L[ sin(t)] = es 1 . +1

s2

If a function is dened as f ( t 1 ) , t t1 ; f (t), t < t t ; 2 1 2 f (t) = , ; fn (t), tn1 < t. Then f (t) = [1 u(t t1 )]f1 (t) + [u(t t1 ) u(t t2 )]f2 (t) + [u(t t2 ) u(t t3 )]f3 (t) + + [u(t tn2 ) u(t tn1 )]fn1 (t) + u(t tn1 )fn (t). 15

Example 22. Solve the dierential equation: 0 t<3 f + f 6f = 2 3t<5 4 5t with f (0) = f (0) = 0. Solution: 0 t<3 2u(t 3) 2u(t 5) + 4u(t 5) = 2 3t<5 4 5t

The dierential equation is f + f 6f = 2u(t 3) + 2u(t 5) with f (0) = f (0) = 0 and we take the Laplace transform of both sides: s2 F + sF 6F = or F = Now, partial fractions has ( ) 2 1 2 1 1 1 1 2 3t 1 2t = + + =L + e + e s(s + 3)(s 2) 3s 15 s + 3 5 s 2 3 15 5 Now, in the expression for F this gets multiplied by various exponential factors, the eect of this is to delay the answer: ) ( ) ( 1 2 3t+9 1 2t6 1 2 3t+15 1 2t10 + e u(t 3) + + e + e u(t 5) f= + e 3 15 5 3 15 5 Example 23. Solve the dierential equation: 0 t<3 f + 4f + 7f = 2 3t<5 4 5t with f (0) = f (0) = 0. ) 2 ( 3 s e + e5s s

( 3s ) 2 e + e5s s(s + 3)(s 2)

16

Solution:

0 t<3 2u(t 3) 2u(t 5) + 4u(t 5) = 2 3t<5 4 5t

The dierential equation is f + 4f + 7f = 2u(t 3) + 2u(t 5) with f (0) = f (0) = 0 and we take the Laplace transform of both sides: s2 F + 4sF + 7F = or F = Now, partial fractions has s(s2 2 s+2 2 1 1 1 = 2 + 4s + 7) 7s 7 (s + 2) + 3 7 (s + 2)2 + 3 ( ) 1 1 3t 2 3t =L e cos( 3t) e sin( 3t) 7 7 7 3 s(s2 ) 2 ( 3s e + e5s s

( 3s ) 2 e + e5s + 4s + 7)

Now, in the expression for F this gets multiplied by various exponential factors, the eect of this is to delay the answer.

17

1.4
1.4.1

CONVOLUTION AND THE DIRAC DELTA DISTRIBUTION

Convolutions

The convolution of f (t) and g (t) is dened by (f g ) =

0 t

f (u)g (t u) du =
0

f (t u)g (u) du.

Remark. f g = g f . Example 24. Find the convolution (f g )(t) when f (t) = t, g (t) = e2t (t 0). Solution: From the denition of convolutions t t (f g )(t) = f ( )g (t ) d = e2(t ) d 0 0 t t 2t 2 2t = e e d = e e2 d
0 0

Use integration by parts with

dv = e2 d 1 du = d, v = e2 2 ( u = ,
t

= e = = = = = Convolution theorem

v du ([ ) t 2 ]t 1 2 2t e e d e 2 2 0 0 ( ) t 2t 1 t 2 2t e e +0+ e d 2 2 0 [ ]t t e2t 1 2 e + 2 2 2 0 ( ) 2t t e 1 2t 1 + e + 2 2 2 2 t 1 1 + e2t 2 4 4

0

2t

u dv = e

2t

[uv ]t 0

L{f g (t)} = F (s)G(s) = L{f }L{g }. The LT of the convolution is the product of the LTs. (Or, equivalently, the inverse LT of the product is the convolution of the inverse LTs.) Proof: Do a change-of-variables in the following double integral: 18

 L{f g (t)} = =

est f (u)g (t u) dt du 0 u su = e f (u) es(tu) g (t u) dt du u 0 su = e f (u) du esv g (v ) dv

0 0

e 0

st

f (u)g (t u) du dt

= L{f }L{g }. Example 25. Calculate the inverse Laplace transform of

1 . s3 s2

Solution: This can be computed using partial fractions and LT tables. However, it can also be computed using convolutions. First we factor the denominator, as follows 1 1 1 = 2 . 2 s s s1 We know the inverse Laplace transforms of each term: s3 [ L
1

] 1 = t, s2

[ L
1

] 1 = et s1

We apply the convolution theorem: [ L

1

] t 1 1 = uetu du s2 s 1 0 t [ ] t u t t = e ue 0 e eu du
0

= t 1 + e Therefore, [ L
1

] 1 1 (t) = et t 1. 2 s s1

Example 26. Find the convolution f (t) = 1 2 3 4 5.

19

Solution: Take the LT. Since the LT of the convolution is the product of the LTs: 5! L{1 2 3 4 5} = 5!(1/s)5 = 5 = F (s). s [ ] 4! 4 We know from LT tables that L1 s 5 (t) = t , so [ f (t) = L 1.4.2 Dirac delta function { k (t) = Then k (t a) = Properties:
1 k (t a) = ku(t a) ku(t (a + k )). k (t a)dt = 1. 1 k, 0 < t k ; 0, otherwise. 1

[F (s)] (t) = 5L

] 4! (t) = 5t4 . s5

Consider the function

1 ; k, a < t a + k 0, otherwise.

k (t a)f (t)dt = 1 a+ k .

a+1/k
a

kf (t)dt = f (t ), where t is a number between a and

L{k (t a)} = eas k (1 es/k ). s lim L{k (t a)} = eas .

k

Denition 1. Dirac delta distribution, or Dirac delta (generalized) function, is dened as (t a) = lim k (t a).
k

The Dirac delta function (t) is technically not a function. Roughly speaking, it may be thought of as being analogous to a radar ping: if a continuous function f (t) represents an objects trajectory then the delta function samples its value at t = 0. Properties: L{ (t)} = 1. L{ (t a)} = eas . 20

 a (f ) =

(t a)f (t)dt = f (a), which is a linear functional. If a > 0 then 0 (t a)f (t)est dt = f (a)eas .
0

Example 27. Solve x + x = (t ), x(0) = x (0) = 0. Remark. This models a unit mass attached to an undamped spring suspended from a board with no initial displacement or initial velocity. At time t , the board is hit very hard (say with a hammer blow) in the upward direction. As we will see, this will start the mass to oscillate. Solution: Taking Laplace transforms gives s2 X (s) + X (s) = es , so X (s) = The inverse LT is x(t) = sin(t)u(t ). Example 28. Solve f + f 6f = (t 4) with f (0) = f (0) = 0. Solution: Using L[ (t a)] = eas this gives s2 F + sF 6F = e4s or F = By partial fractions we have 1 1 1 1 1 = =L (s + 3)(s 2) 5s2 5s+3 ( 1 2t 1 3t e e 5 5 ) e4s (s + 3)(s 2) 1 es . +1

s2

so, the e4s causes a delay of four and we have ) ( 1 2t8 1 3t+12 e e u(t 4) f= 5 5

21

II Series Solutions of ODEs

We have fully investigated solving second order linear dierential equations with constant coecients: Ay + By + Cy = 0, where A,B,C are constants. Now we will explore how to nd solutions to second order linear dierential equations whose coecients are not necessarily constant: P (x)y + Q(x)y + R(x)y = 0.

2.1

Taylor Series
f (n) (x0 ) n=0

Denition 2. The Taylor series about x0 of a function f (x) is the series (x x0 )n .

n!

There exists R 0 such that the series is convergent in |x x0 | < R and divergent in |x x0 | > R. The number R is called Radius of Convergence. We have f (x) =
f (n) (x0 ) n=0

n!

(x x0 )n ,

|x x0 | < R.

Remark. We use Ratio Test to nd R. Example 29.

1 32x 1 32x 1 1x

= =
1 3

n=0

xn ,

|x| < 1, R = 1.

= =

1 12(x1) 1 1 3 1 2 x 3

n=0

2n (x 1)n . R = 1 . 2 R= 3 . 2

2 n n n=0 ( 3 ) x .

2.2

Analytic functions

Denition 3. A function f (x) is analytic at x0 if f has Taylor series about x0 which converges to f (x) in an interval containing x0 .
1 Example 30. f (x) = 1 is analytic at x = 0; f (x) = x does not exist; f (x) = ex is analytic at any x.

x is not analytic at 0, since f (0)

22

Remark: If f and g are analytic at x0 , then cf , f g , f g , f /g (if g (x0 ) = 0) are analytic at x0 . Remark: If f is analytic at x0 , then its Taylor series about x0 is unique.

2.3

Singular point and ordinary point

P (x)y + Q(x)y + R(x)y = 0.

Consider the equation If we divide two sides by P (x), then the equation is changed to y + p(x)y + q (x)y = 0. (1)

Denition 4. If both p(x) and q (x) are analytic at a point x0 , then x0 is called an ordinary point. Otherwise, it is called a singular point. Remark. If p(x) and q (x) are polynomials, then any point is an ordinary point. Example 31. The following equation has singular points x = 1, 2: y + x + ex x + 1 y + y = 0. x1 x2

Theorem 10. If x0 is an ordinary point of the ODE (1), then the general solution of (1) is analytic at x0 , and is therefore given by y=
n=0

an (x x0 )n

in an open interval containing x0 with two arbitrary coecients (usually a0 and a1 ). The radius R is at least the distance from x0 to the closest singular point of the equation. Example 32. Consider the following equation with two singular points x = 1, 2: y + x + ex x + 1 y + y = 0. x1 x2
3 4

Then the series solution about x0 = 0 has R 1; the series solution about x0 = . R 1 4 Example 33. Consider the following equation: y + 1 y + xy = 0. x2 + 1

has

Then the series solution about x0 = 0 has R 1; the series solution about x0 = 2 has R 5. 23

Denition 5. Let x0 be a singular point. If both (x x0 )p(x) and (x x0 )2 q (x) are analytic at x0 , then x0 is called a regular singular point (non-essential singular point). If at least one of (x x0 )p(x) and (x x0 )2 q (x) is not analytic at x0 , then x0 is called an irregular singular point (essential singular point). Example 34. Find all singularities and classify them: 1 y = 0. (1) y + x(xx y + x2 (xx 2) 2)3 (2) (x sin x)y + (cos x)y + ex y = 0. Solution: (1) It has one regular singular point x = 0 and one irregular singular point x = 2. (2) We change the equation to y + cos x ex y + y = 0. x sin x x sin x

x x and xcos are not analytic at zeros of the denominator x sin x. Let Note that xcos sin x sin x x sin x = 0, we imply that x = k , k = 0, 1, 2, ....They are all singular points, in which x = 0 is irregular, and others are regular.

2.4

Power Series Solution

Now we solve an equation at ordinary points.

Example 35. Let y = a0 + a1 x + a2 x2 + a3 x3 + a4 x4 + be the series solution about x = 0 of the initial value problem: y 2xy + y = 0, y (0) = 2, y (0) = 1. Find a0 , a1 , a2 , a3 , a4 . Solution: From y = a0 + a1 x + a2 x2 + a3 x3 + a4 x4 + we imply that y = a1 + 2a2 x + 3a3 x2 + 4a4 x3 + , y = 2a2 + 6a3 x + 12a4 x2 + 20a5 x3 + .

Substitute them into the dierential equation, (2a2 + 6a3 x + 12a4 x2 + 20a5 x3 + ) 2x(a1 + 2a2 x + 3a3 x2 + 4a4 x3 + ) +(a0 + a1 x + a2 x2 + a3 x3 + a4 x4 + ) = 0, i.e., (a0 + 2a2 ) + (a1 + 6a3 )x + (12a4 3a2 )x2 = 0. 24

We have a0 = y (0) = 2, a1 = y (0) = 1, a0 + 2a2 = 0, a2 = 1 a1 + 6a3 = 0, a3 = 1/6. 12a4 3a2 = 0 a4 = 1/4.

Example 36. Consider the DE: y 2xy + y = 0. Note that x0 = 0 is a regular point. So we shall attempt to nd a series solution in the form: y=
n=0

an xn .

1) Find the recursive relation of the coecients in the series solution about x0 = 0. 2) Solve the recursive relation. 3) Find the particular solution with y (0) = 1, y (0) = 2. n Solution: 1) From y = n=0 an x we have y = and y =
n=1 n=2 n=0

nan x

n 1

, xy =

n=1

nan xn ,

n(n 1)an x

n2

(n + 2)(n + 1)an xn .

Substitute all of them into the DE, we imply that an+2 = 2n 1 an . (n + 1)(n + 2)

2) From the recursive relation above, a2n = 3 7 11 (4n 5) a0 , (2n)! 1 5 9 (4n 3) a1 . (2n + 1)! 25

a2n+1 =

We have

1 2 3 7 11 (4n 5) 2n y = a0 1 x x 2! (2 n )! n=2 ) ( 1 5 9 (4n 3) +a1 x + x2n+1 (2 n + 1)! n=1

3) From y (0) = 1 we have a0 = 1; by y (0) = 2, we get a1 = 2. Thus ( ) 1 2 3 7 11 (4n 5) 2n y = 1 x x 2! (2 n )! n=2 ( +2 x +

1 5 9 (4n 3) n=1

) x2n+1

(2n + 1)!

Example 37. Consider the DE: y 2xy + sy = 0. Note that x0 = 0 is a regular point. So we shall attempt to nd a series solution in the form: y=
n=0

an xn .

1) Find the recursive relation of the coecients in the series solution about x0 = 0. 2) Solve the recursive relation. Solution: 1)
y= [(s 2n)an + (n + 1)(n + 2)an+2 ]xn = 0. n=0

2) a2n+2 = (1)n+1 s(s 4)(s 8)...(s 4n) a0 , (2n + 2)! a2n+3 = (1)n+1 (s 2)(s 6)...(s 4n 2) a1 . (2n + 3)!

2.5

Cauchy-Euler Equations

An equation of the form an xn y (n) + an1 xn1 y (n1) + ... + a1 xy + a0 y = 0, is called a Cauchy-Euler equation. Theorem: Second-order Cauchy-Euler equation: a2 x2 y + a1 xy + a0 y = 0, 26

with a2 , a1 and a0 constants. In standard form, the equation is y + The auxiliary equation is a2 r (r 1) + a1 r + a0 = 0, or, a2 r2 + (a1 a2 ) r + a0 = 0. A B y + 2 y = 0. x x

Let r1 and r2 be the two solutions of the auxiliary equation. (i) If r1 = r2 are real, then y1 = |x|r1 and y2 = |x|r2 . (ii) If r1 = r2 (real), then y1 = |x|r1 and y2 = xr1 ln |x|. (iii) If r1 , r2 = i (complex), then y1 = |x| cos( ln |x|) and y2 = |x| sin( ln |x|). A B Proof. Since p(x) = and q (x) = 2 are undened at x = 0, the solution may be x x undened at x = 0. Thus, we assume that x = 0. A Cauchy-Euler equation can be transformed into a constant-coecient equation as follows: For x > 0, let x = et and y (x) = z (t). Then t = ln(x) and, by the chain rule, dy dz dt 1 dz d2 y 1 dz 1 d2 z dt 1 dz 1 d2 z = = , = + = + , dx dt dx x dt dx2 x2 dt x dt2 dx x2 dt x2 dt2 [ 2 ] d z dz dz 2 and the equation x y + Axy + By = 0 becomes + A + Bz = 0, or 2 dt dt dt z + (A 1)z + Bz = 0, which has constant coecients. If z1 (t) and z2 (t) are two independent solutions of z + (A 1)z + Bz = 0, then two independent solutions of x2 y + Axy + By = 0 are given by y1 (x) = z1 (ln x) and y2 (x) = z2 (ln x). Since solutions of a constant-coecient equation are sought in the form z = ert and r y (x) = z (t) with t = ln(x), y (x) = ert = er ln(x) = eln(x ) = xr . Thus, solutions of an Euler equation can be sought directly in the form y = xr .

27

If r1 = r2 are real, then z1 = er1 t and z2 = er2 t y1 = xr1 and y2 = xr2 . If r1 = r2 (real), then z1 = er1 t and z2 = ter1 t y1 = xr1 and y2 = xr1 ln(x). If r1 , r2 = i (complex), then z1 = et cos(t) and z2 = et sin(t) y1 = x cos[ ln(x)] and y2 = x sin[ ln(x)]. For x < 0, let x = et and y (x) = z (t). Then t = ln(x), and the same equation for z (t) results. In either case, t = ln |x|. { Since |x| = x, x > 0 x, x < 0 } , replacing x by |x| gives the solutions for any x = 0. Thus,

If r1 = r2 are real, then y1 = |x|r1 and y2 = |x|r2 . If r1 = r2 (real), then y1 = |x|r1 and y2 = xr1 ln |x|. If r1 , r2 = i (complex), then y1 = |x| cos( ln |x|) and y2 = |x| sin( ln |x|).

Example 38. Solve the following equations: 1. x2 y + 2xy 2y = 0, x > 0. Solution: This is an Euler equation. The auxiliary equation is r2 + r 2 = 0 (r + 2)(r 1) = 0 y = c1 x2 + c2 x = 2. x2 y + 5xy + 4y = 0, x > 0. Solution: This is an Euler equation. The auxiliary equation is r2 + 4r + 4 = 0 (r + 2)2 = 0 y = 3. x2 y + 4xy + 4y = 0, x > 0. Solution: This is an Euler equation. The auxiliary equation is 3 3 9 16 7 2 = i r + 3r + 4 = 0 r = 2 2 2 28 c1 ln(x) + c2 2 . 2 x x c1 + c2 x. x2

[ y=x
3 2

( ) ( )] 7 7 c1 cos ln(x) + c2 sin ln(x) . 2 2

4. 3(x 5)2 y + 6(x 5)y 6y = 0, x = 5. Solution: This is an Euler equation. The auxiliary equation is r2 + r 2 = 0 (r + 2)(r 1) = 0 y = c1 |x 5|2 + c2 |x 5|. 5. x2 y 6y = x3 ln(x), x > 0. Solution: This is an Euler equation. The auxiliary equation is r2 r 6 = 0 (r 3)(r + 2) = 0 y1 = x3 , y2 = x2 , yh = c1 x3 + c2 x2 . x3 ln(x) = x ln(x), x2 x3 x2 ln(x) 1 1 y2 f W [y1 , y2 ] = dx = dx = [ln(x)]2 , = 5, u1 = 2 3 W 5 x 10 3x 2x 1 y1 f 1 1 x4 ln(x) dx = x5 ln(x) + x4 dx = u2 = dx = W 5 25 25 1 5 1 5 x ln(x) + x 25 125 [ ] 1 1 5 1 5 2 2 3 yp = u1 y1 + u2 y2 = [ln(x)] x + x ln(x) + x x = 10 25 125 1 1 1 3 [ln(x)]2 x3 x3 ln(x) + x 10 25 125 1 1 1 3 y = yp + yh = [ln(x)]2 x3 x3 ln(x) + x + c1 x3 + c2 x2 . 10 25 125 f (x) = Theorem: Third-order Cauchy-Euler equation: a3 x3 y + a2 x2 y + a1 xy + a0 y = 0 Let y = xm . Similarly, we have y = mxm1 , y = m (m 1) xm2 , y = m (m 1) (m 2) xm3 Then a3 x3 m (m 1) (m 2) xm3 + a2 x2 m (m 1) xm2 + a1 xmxm1 + a0 xm = 0 xm (a3 m (m 1) (m 2) + a2 m (m 1) + a1 m + a0 ) = 0 Thus we get the auxiliary eqn: a3 m (m 1) (m 2) + a2 m (m 1) + a1 m + a0 = 0. 29

Example 39. Solve the Cauchy-Euler equation (a) x3 y + xy y = 0 (b) x3 y + x2 y 2xy + 2y = 0

1. m3 + (0 3) m2 + (2 0 + 1) m 1 = m3 3m2 + 3m 1 = (m 1)3 = 0, m = 1, 1, 1, y1 = x, y2 = x ln x, y3 = x (ln x)2 yc = c1 x + c2 x ln x + c3 x (ln x)2 2. m3 +(1 3) m2 +(2 1 2) m +2 = m3 2m2 m +2 = (m 1) (m 2) (m + 1) = 0 m = 1, 2, 1, y1 = x, y2 = x2 , y3 = x1 1 yc = c1 x + c2 x2 + c3 x

30

2.6

The Frobenius Method

Frobenius method is important for equations with coecients that have singularities, so that power series method can no longer handle them. Theorem 6. Let x0 = 0 be a regular singular point of the DE y + p(x)y + q (x)y = 0 with xp(x) = Let y=x
r n=0 n=0 n=0 n=0

pn x ,

x q (x) =

qn xn .

cn (r)x =

cn (r)xn+r ,

c0 (r) = 1.

Let r1 and r2 (where r1 r2 if both are real) be two roots of the indicial equation r2 + (p0 1)r + q0 = 0. Case (i): If r1 r2 is not an integer, then the two linearly independent solutions are given by: n+r1 y1 = cn (r1 )x , y2 = cn (r2 )xn+r2 , x > 0.
n=0 n=0

Case (ii): If r1 = r2 , then the two linearly independent solutions are given by: y1 =
n=0

cn (r1 )x

n+r1

y2 = y1 ln(x) +

n=1

bn xn+r1 , x > 0,

where bn = cn (r1 ). Case (iii): If r1 r2 = N is a positive integer, then the two linearly independent solutions are given by: y1 = where
n=0

cn (r1 )xn+r1 ,

y2 = Ay1 ln(x) +

n=0

bn xn+r2 , x > 0,

d [(r r2 )cn (r)]r=r2 . rr2 dr The radius of convergence of every one of the above is at least as great as the distance from x0 = 0 to the nearest other singular point of the equation. A = lim [(r r2 )cN (r)], bn =

31

Example 40. Find the general solution of 2x2 y + (x + 2x2 )y y = 0 for x > 0 near x0 = 0. Solution: Step 1: Determine whether x0 = 0 is an ordinary point or a regular singular point. Write the equation as y + ( We have xp(x) = 1 1 + 1)y 2 y = 0, 2x 2x 1 1 + x, x2 q (x) = . 2 2

So x0 = 0 is a regular singular point. 1 1 , q0 = 2 . The indicial Step 2: Find and solve the indicial equation. Note that p0 = 2 equation is: 1 1 1 r2 + (p0 1)r + q0 = 0. r2 r = 0. r1 = 1, r2 = . 2 2 2 Note that r1 r2 = 1.5, so we have Case (i). Step 3: Find the recursive relation about cn (r). Let y=
n=0 n=0 n+r1 n=0

cn (r)xn+r ,

c0 (r) = 1.

y =

(n + r)cn (r)x

,y =

(n + r)(n + r 1)cn (r)xn+r2 .

Substitute them into (5.1) we have 2x2

n=0 n=0

(n + r)(n + r 1)cn (r)xn+r2 +(x +2x2 )

n=0

(n + r)cn (r)xn+r1
n=0

n=0

cn (r)xn+r = 0,

2(n + r)(n + r 1)cn (r)xn+r +

n=0

(n + r)cn (r)xn+r

2(n + r)cn (r)xn+r+1

n=0

cn (r)xn+r = 0,

[2(n + r)(n + r 1) + (n + r) 1]cn (r)xn+r + 2(n + r)cn (r)xn+r+1 = 0, n=0 n=0

32

 n=0

[(2n + 2r + 1)(n + r 1)]cn (r)x

n=1

n+ r

n=1

2(n 1 + r)cn1 (r)xn+r = 0,

(2r +1)(r 1)]c0 (r)xr +

{[(2n + 2r +1)(n + r 1)]cn (r)+ 2(n 1+ r)cn1 (r)}xn+r = 0,

(2r + 1)(r 1)]c0 (r) = 0,

{[(2n + 2r + 1)(n + r 1)]cn (r) + 2(n 1 + r)cn1 (r)} = 0, n 1.

Since c0 (r) = 1 = 0, the above rst equation results in our indicial equation (2r +1)(r 1) = 0. The second equation gives cn ( r ) = 2 cn1 (r), 2n + 2r + 1 n 1. (10)

Step 4: Find y1 . Take r = r1 = 1, by (10), cn (1) = 2 cn1 (1), n 1. 2n + 3

2 2 c0 (1) = , 5 5 (2)2 2 c1 (1) = , ..., c2 (1) = 7 5(7) c1 (1) = cn (1) = Therefore, y1 = x +

n=1

(2)n , n 1. 5(7)...(2n + 3) ( 1+ ) (2)n xn . 5(7)...(2n + 3)

(2)n xn+1 = xr1 5(7)...(2n + 3)

n=1

Step 5: Find y2 . Take r = r2 = 1 , by (10), 2 1 2 1 1 1 cn1 ( ) = cn1 ( ), n 1. cn ( ) = 2 2n 2 n 2 1 1 1 c1 ( ) = c0 ( ) = 1, 2 1 2 1 1 1 (2)2 c2 ( ) = c1 ( ) = , ..., 2 2 2 2 (1)n 1 , n 1. cn ( ) = 2 n! Therefore, y2 =
(1)n n=0

n!

xn 2 = x 2 ex .
1 1

33

Step 6: The general solution is y (x) = c1 y1 + c2 y2 , where c1 and c2 are constants. Remark. To get solutions for x < 0 or for x = 0 near x0 = 0, replace xr1 , xr2 and ln x in Theorem 6 by |x|r1 , |x|r2 and ln |x|. Remark. If (x x0 )p(x) and (x x0 )2 q (x) are not polynomials but are given by Taylor series, then its possible to obtain approximate solutions. Example 41. Solve Cauchy-Euler equation x2 y + 2xy + The indicial equation is: 3 = 0. 16 1 1 The two solutions are r1 = 4 , r2 = 3 , r1 r2 = 2 . We get a basis y1 = x1/4 , y2 = x3/4 . 4 r2 + r + Example 42. Solve Hyper-geometric dierential equation x(x 1)y + (3x 1)y + y = 0, x > 0. (11) 3 y = 0, 16 x>0

Note that p(x) = (3x 1)/(x 1) and q (x) = x/(x 1). We have b0 = b(0) = 1, c0 = c(0) = 0. Thus the indicial equation is r2 = 0, which has double root r = 0. We have Case (ii). Substitute y = y = y into (11), we have
n=0 n=0 n=0 n=0 = (n + r)(n + r 1)cn (r)xn+r2 n=0

cn (r)xn+r ,

c0 (r) = 1

(n + r)cn (r)xn+r1 ,

(n + r)(n + r 1)cn x

n+r

n=0

(n + r)(n + r 1)cn x
n=0

n+ r 1

n=0

3(n + r)cn xn+r

which implies that

n=0

(n + r)cn x

n+ r 1

cn xn+r = 0,

[(n+r)(n+r1)cn (n+r+1)(n+r)cn+1 +3(n+r)cn (n+r+1)cn+1 +cn ]xn+r r2 c0 xr1 = 0. 34

Simplifying this and using indicial equation r2 = 0, we get

n=0

(n + r + 1)2 (cn cn+1 )xn+r = 0.

Thus cn+1 = cn = = 1. So y1 (x) =

n=0

xn =

1 . 1x

To get another independent solution, note that bn = cn (r1 ) = 0 for all n, y2 = y1 ln(x) + Example 43. (Case III) Solve (x2 x)y xy + y = 0, x > 0. (12)
n=1

bn xn+r =

ln x . 1x

Solution: The indicial equation is r(r 1) = 0, the roots are r1 = 1 and r2 = 0. This is the Case III with N = r1 r2 = 1. n+1 Let y = . Substitute this into (12), n=0 cn x (x x)
2 n=0

(n + 1)(n + 1 1)cn x

n1

n=0

(n + 1)cn x +

n=0

cn xn+1 = 0.

Collecting like terms

[ 2 ] n cn (n + 2)(n + 1)cn+1 xn+1 = 0. n=0

n2 cn (n = 0, 1, ...), (n + 2)(n + 1) c1 = c2 = = 0. We get a solution y1 (x) = c0 x = x. To get a second solution, we let cn+1 = y2 = ky1 ln(x) +
n=0

We derive

dn xn+r2 = kx ln(x) +

n=0

dn xn ,

where k is a constant. Substitute y2 into the dierential equation, k +

n=0

[dn (n 1)2 dn+1 (n + 1)n]xn = 0, d0 = k, d2 = d3 = ... = 0.

Thus y2 = kx ln(x) + k + d1 x. 35

2.7

Bessels Equation

Denition 6. The DE x2 y (x) + xy (x) + (2 x2 2 )y (x) = 0 is called Bessels equation of order with parameter , where 0 and are constants. If = 1, then it is called Bessels equation of order . If = 0, it is Euler equation. If = 0, by substitution z = x and w(z ) = y (x), the equation above can be changed to z 2 w (z ) + zw (z ) + (z 2 2 )w(z ) = 0. Denition 7. J (x) =
k=0

( x )2k+ (1)k k !( + k + 1) 2

is called the Bessel function of the rst kind of order ; For any 0, Y (x) = lim

cos()J (x) J (x) sin()

is called the Bessel function of the second kind of order . Theorem For any x > 0, two linearly independent solutions y1 (x) and y2 (x) of x2 y (x) + xy (x) + (2 x2 2 )y (x) = 0 are y1 (x) = J (x), y2 (x) = {

J (x), if > 0 is not an integer; Y (x), for any 0.

The general solution is y (x) = c1 y1 + c2 y2 , where c1 and c2 are constants. Remark. If > 0 is an integer, then J (x) = (1) J (x).

Example 44. Find two linearly independent solutions, valid for x > 0: (1) x2 y (x) + xy (x) + (4x2 9)y (x) = 0. Solution: Note that 2 = 4 and 2 = 9, = 2 and = 3. Hence y1 (x) = J3 (2x), 36 y2 (x) = Y3 (2x).

(2) x2 y (x) + xy (x) + (4x2 0.25)y (x) = 0. Solution: integer, Note that 2 = 4 and 2 = 0.25, = 2 and = 0.5. Since is not an y1 (x) = J0.5 (2x), (3) xy (x) + y (x) + 0.04xy (x) = 0. Solution: This not Bessel equation. However, by multiplying two sides with x, we get x2 y (x) + xy (x) + 0.04x2 y (x) = 0 which is Bessel equation with 2 = 0.04 and 2 = 0, = 0.2 and = 0. Since is an integer, y1 (x) = J0 (0.2x), y2 (x) = Y0 (0.2x). Remark. If we need solution for x = 0, then ( J (x) = |x| 2 )
k=0

y2 (x) = J0.5 (2x).

( x )2 k (1)k k !( + k + 1) 2

37

3
3.1

III Fourier Series Pre-knowledge

) = (1)n ; 2

1. Trig Identities: sin(n + cos(n ) = (1)n , n is an integer.

1 + cos 2x 1 cos 2x , sin2 x = . 2 2 cos(a b) cos(a + b) cos(a + b) + cos(a b) sin a sin b = , cos a cos b = . 2 2 sin(a + b) + sin(a b) sin a cos b = . 2 cos2 x = 2. A function f (x) is called p periodic if p > 0 is the smallest number such that f (x + p) = f (x) for any x. The number p is called the period. For example, cos kx and sin kx are 2k -periodic. 3. Odd-Even function: If f (x) = f (x) for all x [a, a], then f (x) is odd on [a, a]; If f (x) = f (x) for all x [a, a], then f (x) is even on [a, a]. For example, sin kx is odd, cos kx is even. a If f (x) is odd on [a, a], then a f (x)dx = 0. a a If f (x) is even on [a, a], then a f (x)dx = 2 0 f (x)dx. 4. If m, n are non-negative integers, and m = n, then cos(nx) cos(mx) dx = 0, sin(nx) sin(mx) dx = 0.

Also, even if we drop the restriction m = n, cos(nx) sin(mx) dx = 0.

5. If n is a positive integer, then 2 cos (nx) dx =

sin2 (nx) dx = .

b b vdu. udv = uv | a a a b g(b) 7. Integration by substitution: a f (g (x))g (x)dx = g(a) f (u)du. 6. Integration by parts: b 38

3.2

(Full) Fourier series

Fourier series are named in honor of Joseph Fourier (1768-1830), who made important contributions to the study of trigonometric series. Fourier series have many applications such as, solving partial dierential equations, signal precessing, image processing. A function f (x) is piecewise continuous in interval (a, b) if we have a = t0 < t1 < ... < tm = b, such that f(x) is continuous in each interval (ti , ti+1 ) and the limits lim f (x)
xti

and lim f (x) exist for all i = 0, 1, 2, . . . , m. In the following, we assume that both f and + f are piecewise continuous. Denition. Let f (x) be 2L-periodic function. Then f (x) can be written as
a0 { nx nx } f (x) = + an cos( ) + bn sin( ) . 2 L L n=1 xti

(13)

This series is called the (full) Fourier series for f (x). The coecients an (n 0) are called the Fourier cosine coecients, and the coecients bn (n 1) are called the Fourier sine coecients. Remark. The = occurs at every x [L, L] where f is continuous. If we omit the condition where f is continuous at x, then we may write a0 nx nx f (x) + (an cos( ) + bn sin( )). 2 L L n=1

Theorem. The Fourier coecients can be calculated as follows: nx 1 L f (x) cos( an = ) dx n = 0, 1, 2, . . . . L L L nx 1 L bn = f (x) sin( ) dx n = 1, 2, 3, . . . . L L L Proof. The coecient a0 is the simplest to nd: integrating (23) from L to L, } L L L L { nx nx a0 cos( sin( dx + an ) dx + an ) dx f (x) dx = L L L L L L 2 n=1 L a0 = dx L 2 The series on the right vanishes, and we nd that 1 L a0 = f (x) dx. L L 39

(14)

(15)

We do the same thing to compute, say, bm , except that rst we multiply (23) through by ). We get sin( mx L L L mx a0 mx f (x) sin( ) dx = sin( ) dx + L L L L 2 } L L { nx mx nx mx an cos( ) sin( ) dx + bn sin( ) sin( ) dx L L L L L L n=1 L mx mx = bm sin( ) sin( ) dx = bm L. L L L 1 L mx bm = f (x) sin( ) dx m = 1, 2, 3, . . . . L L L Likewise we can get the formula for am . Formulas (14) and (15) allow us to compute the Fourier coecients of f . Remark 1. Even though f is dened only on [L, L], the right-hand side of (23) is 2L-periodic, so we could view f as being dened over the whole line, but 2L-periodic as well. Remark 2. If f is even on [L, L], then f (x) sin( mx ) is odd on [L, L], so bn = 0 for all L mx n 1; and f (x) cos( L ) is even on [L, L], so 2 L nx an = f (x) cos( ) dx n = 0, 1, 2, . . . . L 0 L ) is odd on [L, L], so an = 0 for all Remark 3. If f is odd on [L, L], then f (x) cos( mx L mx n 0; and f (x) sin( L ) is even on [L, L], so 2 L nx bn = f (x) sin( ) dx n = 1, 1, 2, . . . . L 0 L

3.3

2 -periodic function
a0 f (x) = + {an cos nx + bn sin nx} , 2 n=1

If f is 2 -periodic (i.e., L = ), then

where a0 an bn

1 = f (x) dx. 1 f (x) cos(nx) dx = 1 = f (x) sin(nx) dx 40

(16) n = 0, 1, 2, . . . . n = 1, 2, 3, . . . . (17) (18)

Example 45. Let f (x) = x, for all x [, ), and f (x) be 2 -periodic. Compute the Fourier coecients. Notice that cos(nx) is an even function, while f and sin(nx) are odd functions. 1 1 a0 = f (x)dx = xdx = 0 1 1 an = f (x) cos(nx)dx = x cos(nx)dx = 0 1 1 bn = f (x) sin(nx)dx = x sin(nx)dx = ([ ] [ ] ) x cos(nx) 2 2 sin(nx) n+1 2 x sin(nx)dx = + = ( 1) = 0 n n2 n 0 0 Notice that a0 , an are 0 because x and x cos(nx) are odd functions. Hence the Fourier series for f (x) = x is:
n=1

x = a0 + =

(an cos(nx) + bn sin(nx)) = x (, )

2 (1)n+1 sin(nx), n n=1

Example 46. Let f (x) = x2 , x [, ), and f (x) be 2 -periodic. Compute the Fourier coecients. Since f is even (f (x) = f (x) for all x), then bn = 0. 1 2 2 a0 = x dx = 2 , 3 and for n 1 an 1 =

x2 cos(nx) dx

1 = n = Thus for x (, ),

x sin(nx)
n

} 2x sin(nx) dx

(1) 4 . n2 2 (1)n 4 + cos(nx). 2 3 n n=1

x2 =

41

Example 47. Let f (x) =

0, for x [, 0); , 1, for x (0, ).

and let f (x) be 2 -periodic. Find the Fourier series of f (x). ) ( 0 1 1 1dx = 1, = f (x)dx = 0dx + 0 ( 0 ) 1 1 1 = f (x) cos(nx)dx = 0 cos(nx)dx + 1 cos(nx)dx = sin(nx)| 0 = 0, n 0 ( 0 ) 1 1 1 = f (x) sin(nx)dx = 0 sin(nx)dx + 1 sin(nx)dx = cos(nx)| 0 n 0 { 2 , for odd n; 1 n = (1 (1)n ) = n 0, for even n. Hence the Fourier series for f (x) is: 2 1 + sin(nx) = 2 odd n n

a0 an bn

f (x)

1 2 = + sin(2n + 1)x, 2 n=0 (2n + 1)

x (, ).

3.4

Points of discontinuity and convergence

In equation (1), = means that the series on the right converges to the function on the left at each point x. It often happens that the Fourier series of a function f fails to converge to that function, in particular at the points of discontinuity of f. The facts are: The Fourier Theorem: If the function f (x) is piecewise continuously dierentiable then its Fourier series converges for every x to the average value fav (x) = where f (x+) = lim f (t),
tx +

f (x+) + f (x) , 2 f (x) = lim f (t).

tx

(19)

At the points where f (x) is continuous, fav (x) = f (x). 42

All the functions we shall consider in the sequel are piecewise continuously dierentiable, and therefore the Fourier series will represent the function. In order to ensure that the Fourier series of function f (x) converges to that function at every x R, sometimes it is necessary to redene f (x) at the points of discontinuity x, so that fav (x) = f (x). Example 48. Let 1, for x (, 0); f (x) = x, for x (0, ); 0, for x = 0, , .
1 2

Determine the sums to which the series converges at x = 0, , 88, 101 . Solution: The sum=0.5 at 0; at .

3.5

Geometric interpretation of Fourier series

S1 =

In the example above, if we let 1 2 + sin x, 2 1 2 2 S3 = + sin x + sin 3x, 2 3 1 2 2 2 S5 = + sin x + sin 3x + sin 5x, 2 3 5 each partial sum is a continuous function that approximates the discontinuous function f(x) on the interval (, ). The bigger n, the better the approximation.

3.6

0 < x < 1; x, Example 49. Let f (x) be 2-periodic and f (x) = 0, 1 < x 0. 0.5, x = 1, 1. In this case, L = 1. a0 = 0 1 an = bn The full Fourier series is
1

General 2L-periodic function

1 x dx = , 2 x cos(nx) dx =

(1)n 1 , (n )2 0 1 (1)n = x sin(nx) dx = . n 0

] [ (1)n 1 (1)n 1 cos(nx) + sin(nx) . f (x) = + 4 n=1 (n )2 n 43

3.7

Fourier cosine and sine Series on interval (0, L): Even and odd extensions

Half-range Expansions Let f (x) be dene on (0, L). Three special extensions are important: (i) Consider f (x) as an odd function on (L, L), i.e., { f (x), x (0, L); fodd (x) = f (x), x (L, 0). Then f (x) is 2Lperiodic. Thus an = 0 for all n, (23) becomes ( nx ) f (x) = bn sin L n=1 which is called Fourier sine series of f , where ( mx ) 2 L bm = dx f (x) sin L 0 L

(20)

m = 1, 2, 3, . . . .

(ii) Consider f (x) as an even function on (L, L), i.e., { f (x), x (0, L); feven (x) = f (x), x (L, 0). Then f (x) is 2Lperiodic. Thus bn = 0 for all n, (23) becomes ( nx ) a0 f (x) = + an cos 2 L n=1 which is called Fourier cosine series of f , where 2 L mx am = f (x) cos( ) dx L 0 L

(21)

m = 0, 1, 2, 3, . . . .

The cosine and sine series here are known as HALF-RANGE EXPANSIONs. (iii) Consider f (x) as Lperiodic. Then half-period is L/2. 1 L nx an = f (x) cos( ) dx n = 0, 1, 2, . . . . L 0 L/2 L 1 nx bn = f (x) sin( ) dx n = 1, 2, 3, . . . . L/2 0 L/2 ( ) ( ) a0 2nx 2nx f (x) = + [an cos + bn sin ] 2 L L n=1 which is called (full) Fourier series of f . 44

(22)

{ Example 50. Let f (x) = cosine series.

x, 0 x < 1. . Find the Fourier sine series and Fourier 0, 1 x < 2;

Solution: (i) Fourier sine series: for m = 1, 2, 3, . . . , bm 2 L mx 2 2 mx = f (x) sin( ) dx = f (x) sin( ) dx L 0 L 2 0 2 [ ]1 1 2 mx 4 mx mx ) dx = x cos( ) + 2 2 sin( ) = x sin( 2 m 2 m 2 0 0 2 m 4 m = cos( ) + 2 2 sin( ). m 2 m 2 mx mx 2 L 2 2 f (x) cos( f (x) cos( = ) dx = ) dx L 0 L 2 0 2 [ ]1 1 2 mx mx 4 mx = x cos( ) dx = x sin( ) + 2 2 cos( ) 2 m 2 m 2 0 0 2 m 4 m 4 = sin( ) + 2 2 cos( ) 2 2. m 2 m 2 m

(ii) Fourier cosine series:for m = 0, 1, 2, 3, . . . , am

Remark. (i) 2-periodic extension of f : 0 x < 1; x, (x) = f 0, 1 x < 2. (x), for any x. f (x + 2) = f In this case, L = 1. a0 = 0 1 am = bm The full Fourier series is ] [ (1)n 1 (1)n 1 cos(nx) + sin(nx) . f (x) = + 4 n=1 (n )2 n 45
1

1 x dx = , 2 x cos(mx) dx =

(1)m 1 , (m )2 0 1 (1)m = x sin(mx) dx = . m 0

At x = 2, this series converges to (2+) + f (2) f (2) = 0; =f 2 At x = 1, this series converges to (1+) + f (1) f 1+0 = = 0.5 2 2 At x = 1, this series converges to (1+) + f (1) f 1+0 = = 0.5. 2 2 (ii) 4-periodic odd extension of f : 0, x, (x) = f 0, (x), f (x + 4) = f In this case, L = 2. a 0 = 0, am = 0, 2 2 mx 2 m 4 m bm = f (x) sin( ) dx = cos( ) + 2 2 sin( ). 2 0 2 m 2 m 2 The full Fourier series is f (x) =
[ m=1

2 x < 1; 1 x < 1; 1 x < 2. otherwise.

] 2 m 4 m mx cos( ) + 2 2 sin( ) sin( ), m 2 m 2 2

which is the Fourier sine series. At x = 2, this series converges to (2+) + f (2) f (2) = 0; =f 2 At x = 1, this series converges to (1+) + f (1) f 1+0 = = 0.5 2 2 At x = 1, this series converges to (1+) + f (1) 01 f = = 0.5. 2 2 46

(iii) 4-periodic even extension of f : 0, x, f (x) = x, 0, f (x + 4) = f (x), In this case, L = 2. am bm 2 = = 2 = 0

0

2 x < 1; 1 x < 0; 0 x < 1; 1 x < 2. otherwise.

f (x) cos(

mx 2 m 4 m 4 ) dx = sin( ) + 2 2 cos( ) 2 2, 2 m 2 m 2 m

The full Fourier series is ] [ 2 m 4 m 4 mx f (x) = sin( ) + 2 2 cos( ) 2 2 cos( ), m 2 m 2 m 2 m=1 which is the Fourier cosine series. At x = 2, this series converges to (2+) + f (2) f (2) = 0; =f 2 At x = 1, this series converges to (1+) + f (1) f 1+0 = = 0.5 2 2 At x = 1, this series converges to (1+) + f (1) f 0+1 = = 0.5. 2 2

3.8

(Full) Fourier Series on Interval [a, b]

(x) with Let f (x) be dened on (a, b). Then we can extend f (x) to a periodic function f a period b a. Let 2L = b a, then L = b . 2 Denition 8. The Fourier Series of f (x) on (a, b) is f (x) = nx nx a0 + (an cos( ) + bn sin( )) 2 L L n=1

(23)

at every x [a, b] where f is continuous. The coecients an (n 0) and the coecients bn (n 1) are calculated as follows: 1 b nx 1 b nx an = f (x) cos( ) dx, n = 0, 1, 2, . . . ; bn = f (x) sin( ) dx, n = 1, 2, 3, . . . . L a L L a L 47

{ Example 51. Let f (x) =

3, 1 < x < 0; Find its Fourier series. 2x, 0 x < 3;

Solution: L = 2. 1 3 a0 = f (x)dx = 6, 2 1 ( ) 1 3 nx 1 n 3n 4 3n 4 an = f (x) cos dx = 3 sin + 6 sin + cos , 2 1 2 n 2 2 n 2 n ( ) 1 3 nx 1 n 3n 4 3n bn = f (x) sin dx = 3 + 3 cos 6 cos + sin . 2 1 2 n 2 2 n 2

48

4
4.1

IV Partial Dierential Equations What is a Partial Dierential Equation?

A partial dierential equation (PDE) relates the partial derivatives of a function of two or more independent variables together. For example, Laplaces equation for (x, y ), 2 2 + =0 x2 y 2 (24)

arises in many places in mathematics and physics. For simplicity, we will use subscript notation for partial derivatives, so this equation can also be written xx + yy = 0. We say a function is a solution to a PDE if it satisfy the equation and any side conditions given. Mathematicians are often interested in if a solution exists and when it is unique. Example. 1 = x and 2 = x2 y 2 are solutions to Laplaces equation (24). The order of PDE is the highest partial derivative that appears in the equation. So, Laplaces Equation (24) is second-order. We also dene linear PDEs as equations for which the dependent variable (and its derivatives) appear in terms with degree at most one. Anything else is called nonlinear. Linear equations can further be classied as homogeneous for which the dependent variable (and it derivatives) appear in terms with degree exactly one, and nonhomogeneous which may contain terms which only depend on the independent variable. If two solutions, say u1 and u2 , satisfy a linear homogeneous PDE, that any linear combination of them u = c1 u1 + c2 u2 is also a solution. So, for example, since 1 = x2 y 2 2 = x

both satisfy Laplaces equation, xx + yy = 0, so does any linear combination of them = c1 1 + c2 2 = c1 (x2 y 2 ) + c2 x. This property is extremely useful for constructing solutions which satisfy certain initial conditions and boundary conditions.

4.2

Heat Equation or Diusion Equation

When we derived Newtons Law of cooling we made several assumptions most importantly that the temperature does not vary with location. If we account for the variation 49

of temperature with location, we can derive a PDE called the heat equation or, more generally, the diusion equation. Suppose we consider a metal bar, with a uniform crosssectional area, A, whose temperature, u(x, t), is a function of time, t, and the position, x, along the bar (that is we assume the temperature is uniform in every cross-section). Then u(x, t) satises the diusion equation (DE), ut = c2 uxx , (25)

where c2 is a constant known as the thermal diusivity, determined by the geometry and physical properties of the metal bar. Suppose we consider a bar of nite length L, occupying the region 0 < x < L. At the boundaries of the metal bar we can specify a xed temperature, u(0, t) = u0 u(L, t) = u1 , (26)

which are usually referred to as Dirichlet boundary conditions (BCs). We also need to specify the initial temperature distribution, Initial condition (IC) : 4.2.1 u(x, 0) = f (x) 0 x L. (27)

Homogeneous Boundary Conditions: Separation of Variables

Theorem 11. The solution of the heat equation ut = c2 uxx u(0, t) = 0 u(L, t) = 0 u(x, 0) = f (x) 0 < x < L, t > 0 t>0 0 < x < L, P DE BC IC

is given by u(x, t) = where

n=1

bn sin

( nx ) L

e(

cn 2 t L

) ,

2 bn = L

f (x) sin

( nx ) L

dx.

(28)

Proof. The method to solve the wave equation comprises three steps. Step 1: Separation of variables. Let u(x, t) = X (x)T (t) and substitute it into the equation we have XTt = c2 Xxx T 50

and dividing by XT we nd Tt Xxx = = . (29) 2 cT X where is to be determined. Now because Tt /DT is only a function of t and Xxx /X is only a function of x we know that must be independent of x and t respectively, and therefore must be a constant consequently it is known as the separation constant. We can now solve the resulting ODE for T (t) Tt = c2 T T (t) = CeDt . (30)

Step 2: Eigenfunction. We now look for a solution for the X (x) equation that also satises the homogeneous boundary conditions. From the boundary conditions (BC), we know that u(0, t) = X (0)T (t) = 0 u(L, t) = X (L)T (t) = 0 X (0) = 0 X (L) = 0 (31) (32)

So nally we conclude that we are looking for solutions to the Boundary Value Problem for X (x), Xxx X = 0, X (0) = X (L) = 0. (33) Now we consider three cases: 1. If = 0, then X (x) = Ax + B , it follows from X (0) = 0 that B = 0. Thus X (x) = Ax. From X (L) = 0 we have A = 0, so X (x) = 0, trivial solution. 2. If > 0, then X (x) = Ae x + Be x , it follows from X (0) = X (L) = 0 that A + B = 0, Ae L + Be L = 0. Thus A = B = 0 and so X (x) 0, a trivial solution. 3. If < 0, then X (x) = A cos( x) + B sin( x). Applying the boundary condi tions we see that X (0) = 0 implies that A = 0, and that B sin( L) = 0. To have non-trivial solution, B = 0, so sin( L) = 0, = n L ( nx ) , n = 1, 2, 3, ... Xn (x) = Bn sin L where Bn are constants. Note that = ( n )2 . These special values of are called L eigenvalues and the associated functions, Xn (x), are known as eigenfunctions. This implies cn 2 Tn (t) = Cn e( L ) t .

51

Multiplying the solution for Xn (x) and Tn (t) together nally yields a solution for un (x, t) satisfying BCs un (0, t) = un (L, t) = 0 is ( nx ) cn 2 un (x, t) bn sin e( L ) t n = 1, 2, 3 . . . . (34) L Step 3: Superposition. We use superposition of the eigenfunctions to satisfy the initial condition. To consider the IC: u(x, 0) = f (x), we let u(x, t) =
n=1

un (x, t) =

n=1

bn sin

( nx ) L

e(

cn 2 t L

) ,

f (x) = u(x, 0) = bn = 2 L
0 L

n=1

bn sin

( nx ) L dx.

f (x) sin

( nx ) L

Example 52. Solve u 2u = , t x2 subject to the boundary conditions u(0, t) = u(1, t) = 0, and the initial condition u(x, 0) = f (x) = {

x if 0 < x < 1/2, 1 x if 1/2 < x < 1. ) (1 x) sin(nx)dx . (35)

Solution: c = 1, L = 1. Thus ( 1/2 bn = 2 x sin(nx)dx +

0

1/2

Integration by parts yields, b2m = (1)m u(x, t) =

n=1

1 , m2 2
2

b2m+1 = 0, (1)m 1 m2 2 sin(2mx)e(2m)

2t

bn sin

( nx ) L

( cn t L )

m=1

52

4.2.2

Non-homogeneous Boundary Conditions The Dirichlet Problem for the Diffusion Equation (Non-homogeneous Boundary Conditions)

ut = 2 uxx u(0, t) = A u(L, t) = B u(x, 0) = f (x)

0 < x < L, t > 0 t>0 0 x L.

P DE BC IC

By letting u(x, t) = v (x) + w(x, t), we see that w(x, t) satises Homogeneous Boundary Conditions and v (x) satises v (x) = 0, v (0) = A, v (L) = B.

Theorem 12. The solution of the heat equation (25) subject to BCs: u(0, t) = A, u(L, t) = B and IC: u(x, 0) = f (x) is given by u(x, t) = v (x) + w(x, t) with BA v (x) = x + A, L 2 bn = L
0 L n=1

w(x, t) =

bn sin

( nx ) L

e(

n 2 t L

) ,

where bn satises

[f (x) v (x)] sin

( nx ) L

dx.

(36)

Example. Find the solution of the Heat Equation uxx = 9ut , subject to BCs: u(0, t) = 1, u(2, t) = 3 and IC: u(x, 0) = f (x), where { 2x + 1, 0 x < 1, f (x) = . x + 1, 1 x < 2, Solution: Here = 1 , L = 2, A = 1, B = 3. Thus v (x) = x + 1. By (36), for 3 m = 1, 2, 3, . . . , 2 L mx 2 2 mx bm = [f (x) v (x)] sin( ) dx = [f (x) v (x)] sin( ) dx L 0 L 2 0 2 [ ]1 1 2 mx 4 mx mx ) dx = x cos( ) + 2 2 sin( ) = x sin( 2 m 2 m 2 0 0 2 m 4 m = cos( ) + 2 2 sin( ). m 2 m 2 53

u(x, t) = x + 1 +

[ n=1

] ( nx ) n 2 n 4 n 2 cos( ) + 2 2 sin( ) sin e( 6 ) t . n 2 n 2 2

4.3

The Wave Equation

The wave equation has the form: utt = c2 uxx , (37)

where u = u(x, t) can be thought of as the vertical displacement of the vibration of a string. 4.3.1 Dirichlet Problem and Separation of Variables

If we tie the string at both ends we can have the following boundary conditions:

u(0, t) = A(t), u(L, t) = B (t), where A, B are piecewise functions. For example, we can have a sinusoidal function at one end and a Heaviside function at the other. When the boundary values A and B are 0 we obtain the Dirichlet Problem for the wave equation: utt = c2 uxx , u(0, t) = 0, u(L, t) = 0, u(x, 0) = f (x), ut (x, 0) = g (x) 0 < x < L, t > 0 t>0 0<x<L P DE BC IC.

As you have seen in Heat equation, the method of separating the variables is a very convenient way to obtain solutions for PDEs. In the case of the Dirichlet Problem we will quickly review the method. Theorem 13. The solution of the above Dirichlet Problem (wave problem) is given by: u(x, t) = where: 2 an = L
0 [ n=1

] nct nct nx an cos( ) + bn sin( ) sin( ), L L L 2 bn = nc

0 L

f (x) sin

( nx ) L

dx,

g (x) sin

( nx ) L

dx.

54

Proof. The method to solve the wave equation comprises three steps. Step 1: Separation of variables. Let u(x, t) = X (x)T (t) and substitute it into the equation utt = c2 uxx , to obtain: X (x)T (t) = c2 X (x)T (t),
(t) X (x) or cT 2 T (t) = X (x) , thus the equality is one of functions of dierent variables, so both quotients have to be constant. Say T (t) X (x) = = 2 , c2 T (t) X (x) then we can solve each ordinary dierential equation separately. We have the following three cases: 2 , 2 , and = 0. Step 2: Eigenfunctions. (x) = 2 , are: X (x) = Case 1 When the constant is 2 , then the solutions for X X (x)

c1 ex + c2 ex , and the solutions for

T (t) c2 T (t)

= 2 , are T (t) = d1 ect + d2 ect . Then

u(x, t) = (d1 ect + d2 ect )(c1 ex + c2 ex ). Lets take a look at the boundary conditions: u(0, t) = 0, u(L, t) = 0. We imply that c1 = c2 = 0, and so u(x, t) = 0, only trivial solution. Case 2 When the constant is 0, then the equations become X (x) = T (t) = 0, and X (x) = c1 x + c2 , and T (t) = d1 t + d2 . Then u(x, t) = (d1 t + d2 )(c1 x + c2 ). Lets take a look at the boundary conditions: u(0, t) = 0, u(L, t) = 0. We imply that c1 = c2 = 0, and so u(x, t) = 0, only trivial solution. (x) = 2 , are: Case 3 When the constant is 2 , then the solutions for X X (x) X (x) = c1 sin(x) + c2 cos(x), and the solutions for
T (t) c2 T (t)

= 2 , are: T (t) = d1 sin(ct) + d2 cos(ct).

Then u(x, t) = (d1 sin(ct) + d2 cos(ct))(c1 sin(x) + c2 cos(x)). The boundary conditions u(0, t) = u(L, t) = 0 translate into: (d1 sin(ct) + d2 cos(ct))(c1 sin(0) + c2 cos(0)) = 0 (d1 sin(ct) + d2 cos(ct))(c1 sin(L) + c2 cos(L)) = 0, 55 t > 0,

namely: c2 = 0 c1 sin(L) = 0. From the last condition we obtain = n , and L [ n n ] n un (x, t) = d1n sin( ct) + d2n cos( ct) cn sin( x). L L L Step 3: Superposition. We use superposition of the eigenfunctions to satisfy the initial condition. ] [ nct nct nx u(x, t) = un (x, t) = d1n sin( ) + d2n cos( ) cn sin( ) L L L n=1 n=1 =
[ n=1

] nct nct nx an cos( ) + bn sin( ) sin( ). L L L ( nx )

The only conditions left to check are the initial conditions: , L n=1 ( nx ) nc ut (x, 0) = g (x) = . bn sin L L n=1 u(x, 0) = f (x) = an sin From the 1st one we have 2 an = L From the 2nd one we have 2 bn = nc Then u(x, t) =
[ n=1

f (x) sin

( nx ) L

dx.

g (x) sin

( nx ) L

dx.

] nct nx nct ) + bn sin( ) sin( ). an cos( L L L

Example 53. Solve the following equation: 9utt = uxx , u(0, t) = 0, u(4, t) = 0, u(x, 0) = f (x), ut (x, 0) = 0 56 0 < x < 4, t > 0 t>0 0<x<4 DE BC IC,

where f (x) = Solution: bn = 0. Note that

2x, 2x + 4,

0x2 2<x4

2, f (x) = 2,

0x2 2 x 4,

and f (0) = f (4) = 0, thus 2 L nx an = f (x)sin( )dx L 0 L L 2 L nx L 2 L nx = f (x)( ) cos( )|0 + f (x) cos( )dx L n L L n 0 L 2 4 2 4 nx 2 4 nx = ( ) 2 cos( )dx + ( ) (2) cos( )dx 4 n 0 4 4 n 2 4 32 n = sin( ). 2 2 n 2 Now we can write the formal solution to the plucked string equation: 32 1 n nt nx u(x, t) = 2 sin( ) cos( ) sin( ). 2 n=1 n 2 12 4

*String with free end: Example 54. Solve the following equation: utt = 9uxx , u(0, t) = 0, ux (2, t) = 0, 5x u(x, 0) = 2 sin( ), ut (x, 0) = 0 4 0 < x < 2, t > 0 t>0 0<x<2 DE BC IC.

Solution: Let u(x, t) = X (x)T (t) and substitute it into the equation utt = 9uxx , to obtain: X (x) T (t) = = 2 . 9T (t) X (x)
(x) Case 1 When the constant is 2 , then the solutions for X = 2 , are: X (x) = X (x) c1 ex + c2 ex . From u(0, t) = 0, ux (2, t) = 0 we have X (0) = 0, X (2) = 0, which gives

c1 + c2 = 0, c1 x c2 = 0, c1 = c2 = 0, X (x) = 0, u(x, t) = 0. 57

So only trivial solution. Case 2 When the constant is 0, then the equations become X (x) = T (t) = 0, and so X (x) = c1 x + c2 . From X (0) = 0, X (2) = 0, we imply that c1 = c2 = 0, X (x) = 0, u(x, t) = 0. So only trivial solution. Case 3 When the constant is 2 , then the solutions for X (x) = c1 sin(x) + c2 cos(x), and the solutions for
T (t) 9T (t) X (x) X (x)

= 2 , are:

= 2 , are: T (t) = d1 sin(3t) + d2 cos(3t).

Then u(x, t) = (d1 sin(3t) + d2 cos(3t))(c1 sin(x) + c2 cos(x)). X (0) = 0, c2 = 0, X (x) = c1 sin(x). (2n + 1) 4 (2n + 1) 3(2n + 1) 3(2n + 1) X (x) = c1 sin( x), T (t) = d1 sin( t) + d2 cos( t) 4 4 4 3(2n + 1) (2n + 1) 3(2n + 1) t) + Bn cos( t)] sin( x), un (x, t) = [An sin( 4 4 4 ] [ 3(2n + 1) 3(2n + 1) (2n + 1) u(x, t) = An sin( t) + Bn cos( t) sin( x). 4 4 4 n=1 X (2) = 0 c1 cos(2) = 0 = The only conditions left to check are the initial conditions: u(x, 0) =
n=1

Bn sin(

(2n + 1) 5x x) = 2 sin( ), 4 4

] 15 3(2n + 1) (2n + 1) t) + 2 cos( t) sin( x). u(x, t) = An sin( 4 4 4 n=1

B2 = 2, Bn = 0(n = 2),

ut (x, 0) =

n=1

An

3(2n + 1) (2n + 1) sin( x) = 0, An = 0, 4 4 15 (2n + 1) t) sin( x). 4 4 58

u(x, t) = 2 cos(

4.4

Laplace Equation

Now we consider the following Dirichlet Problem for the Laplace equation: P DE : 2 u = BC :
2u x2

2u y 2

uxx + uyy = 0,

(x, y ) R;

u(x, y ) = f (x, y ),

(x, y ) R.

R is a region. We can consider u(x, y ) as the steady-state (time-independent) temperature distribution. 4.4.1 Let R = {(x, y ) : 0 < x < L, 0 < y < M }. Case 1: Consider the boundary-value problem P DE : BC : uxx + uyy = 0, (x, y ) R; Solve the Laplaces equation inside a rectangle

u(x, 0) = u(x, M ) = u(0, y ) = 0, u(L, y ) = f (y ).

We will use the method of separation of variables again. Let u(x, y ) = X (x)Y (y ) and substitute in the equation to obtain: X Y = = constant = 2 X Y (as before, when constant is 0 or negative, no non-trivial solution!). X 2 X = 0 Y + 2 Y = 0 X = c ex + c ex 1 2 Y = c3 sin y + c4 cos y u(x, y ) = (c1 ex + c2 ex )(c3 sin y + c4 cos y ). The boundary conditions u(x, 0) = 0 c4 = 0; u(x, M ) = 0 c3 sin M = 0;

u(0, y ) = 0, c2 = c1 . u(x, y ) = c1 (ex ex )c3 sin y = 2c1 c3 sinh(x) sin y. 59

To have non-trivial solution, sin M = 0, = n , where n is integer. M ( nx ) ( ny ) un (x, y ) = an sinh sin . M M u(x, y ) = By the boundary condition f (y ) = u(L, y ) = we imply that 2 ) ( an = M sinh nL M
n=1 n=1

an sinh

( nx ) M (

sin

( ny ) M

(38)

an sinh
0

nL M

) sin

( ny ) M

f (y ) sin

( ny ) M

dy.

(39)

Case 2: Consider the boundary-value problem P DE : BC : Let z = L x, equation becomes: P DE : BC : uxx + uyy = 0, (x, y ) R;

u(x, 0) = u(x, M ) = u(L, y ) = 0, u(0, y ) = f (y ).

u(x, y ) = w(z, y ). Then x = 0 z = L; x = L z = 0 and the

wzz + wyy = 0,

(z, y ) R;

w(z, 0) = w(z, M ) = w(0, y ) = 0, w(L, y ) = f (y ).

According to Case 1, the solution is: w(z, y ) = with 2 ( ) an = M sinh nL M Therefore, u(x, y ) =
n=1 n=1

an sinh

( nz ) M

sin

( ny ) M

f (y ) sin (

( ny ) M )

dy,

n 1.

(40)

an sinh

n (L x) M

sin

( ny ) M

(41)

Case 3: Consider the boundary-value problem P DE : BC : uxx + uyy = 0, (x, y ) R;

u(x, 0) = u(0, y ) = u(L, y ) = 0, u(x, M ) = f (x). 60

Interchange x and y we will have Case 1. By Case 1, interchange L and M , u(x, y ) = where 2 ( ) an = L sinh nM L
n=1

an sinh
0

( ny ) L
L

sin

( nx ) L

(42)

f (x) sin

( nx ) L

dx.

(43)

Case 4: Consider the boundary-value problem P DE : BC : uxx + uyy = 0, (x, y ) R;

u(x, M ) = u(0, y ) = u(L, y ) = 0, u(x, 0) = f (x). u(x, y ) = w(x, z ). Then y = 0 z = M ; y =

This is similar to Case 2. Let z = M y, M z = 0 and the equation becomes: P DE : BC : We have w(x, z ) = where 2 ( ) an = L sinh nM L Thus u(x, y ) =
n=1

wxx + wzz = 0,

(x, z ) R;

w(x, 0) = w(0, y ) = w(L, y ) = 0, w(L, M ) = f (x).

n=1

an sinh
0

( nz ) L

sin

( nx ) L

f (x) sin n (M y ) L

( nx ) L ) sin

dx.

(44)

( an sinh

( nx ) L

(45)

Case 5: Consider the boundary-value problem P DE : BC : uxx + uyy = 0, (x, y ) R;

u(x, 0) = f1 (x), u(x, M ) = f2 (x), u(0, y ) = f3 (y ), u(L, y ) = f4 (y ).

Then we just need to combine cases 14 together. Theorem 14. The solution of the above Laplace equation is: u(x, y ) = u1 (x, y ) + u2 (x, y ) + u3 (x, y ) + u4 (x, y ),

61

where u1 (x, y ) = u2 (x, y ) = u3 (x, y ) = u4 (x, y ) =

L ( nx ) ( nx ) n (M y ) 2 ( nM ) f1 (x) sin an sinh sin , an = dx, L L L L sinh L 0 n=1 L ( nx ) ( ny ) ( nx ) 2 ( ) bn sinh sin , bn = f2 (x) sin dx, L L L L sinh nM 0 L n=1 M ( ny ) ( ny ) 2 n (L x) ( nL ) sin , cn = f3 (y ) sin dy, cn sinh M M M M sinh 0 M n=1 M ( ny ) ( ny ) ( nx ) 2 ( ) dn sinh sin , dn = f4 (y ) sin dy. M M M M sinh nL 0 M n=1

Example Find the solution of Laplaces equation uxx + uyy = 0 within R = {(x, y ) : 0 < x < 3, 0 < y < 2} with BC : u(x, 0) = 2x + 2, u(x, 2) = 0, u(0, y ) = 2 y, u(3, y ) = 8 4y. Solution: L = 3, M = 2. dn 2 ( ) = 2 sinh 3n 2 2 ( ) 2 sinh 3n 2
0 2 2

(8 4y ) sin (2 y ) sin
0 3

( ny ) 2 2 3

dy =

cn = an = bn

( ny )

16 ( ), n sinh 3n 2

dy =

2 ( ) 3 sinh 2n 3 = 0.

(2x + 2) sin
0

( nx )

4 ( ), n sinh 3n 2 16 cos(n ) + 4 ( ) , n sinh 2n 3

dx =

( nx ) ( ny ) 16 ( 3n ) sinh sin , 2 2 n sinh 2 n=1 ) ( ( ny ) 4 n (3 x) ( ) , u3 (x, y ) = sin sinh 2 2 n sinh 3n 2 n=1 ) ( ( nx ) 16 cos(n ) + 4 n (2 y ) ( 2n ) sinh u1 (x, y ) = , sin 3 3 n sinh 3 n=1 u4 (x, y ) = u2 (x, y ) = 0. Theorem 15. (Polynomial solution) Consider the boundary-value problem P DE : BC : uxx + uyy = 0, (x, y ) R = {(x, y ) : 0 < x < L, 0 < y < M };

u(x, 0) = f1 (x), u(x, M ) = f2 (x), u(0, y ) = f3 (y ), u(L, y ) = f4 (y ). 62

If all fi are continuous and linear on the boundary, then the PDE has a polynomial solution u(x, y ) = ax + by + cxy + d. Example Find a polynomial solution of Laplaces equation within R = {(x, y ) : 0 < x < 3, 0 < y < 2} with BC : u(x, 0) = 2x + 2, u(x, 2) = 0, u(0, y ) = 2 y, u(3, y ) = 8 4y. Solution: Note that all fi are continuous and linear on the boundary. Let u(x, y ) = ax + by + cxy + d. u(x, 0) = 0 a = d = 2. u(0, y ) = 2 y = by + 2 b = 1. u(3, y ) = 8 4y = 6 y + 3cy + 2 c = 1. u(x, y ) = 2x y xy + 2. 4.4.2 Laplace equation inside a circle

In this case, we use polar coordinates: x = r cos , Then we have the boundary value problem: P DE : BC : 1 1 urr + ur + 2 u = 0, r r u(R, ) = f (). r < R; y = r sin .

Theorem 16. Let f be continuous, 2 -periodic, f be piecewise continuous. Then the solution of the PDE above is a0 n + r [an cos(n) + bn sin(n)], u(r, ) = 2 n=1

where 1 an = Rn 1 bn = Rn

f () cos(n)d,
2

n 0, n 1.

f () sin(n)d,
0

63

Proof. Let u(r, ) = F (r)G(). We imply that r2 F (r) F (r) G () +r = = k. F (r) F (r ) G()

Thus we obtain the Euler-Cauchy equation r2 F (r) + rF (r) kF (r) = 0 and the harmonic equation G () + kG() = 0. Since f () is 2 -periodic, G() should be 2 -periodic. Thus k > 0 and k = n2 , n = 1, 2, .... Then Gn () = An cos(n) + Bn sin(n), F (r) = Cn rn + Dn rn . The eigenfunctions are ( ) un (r, ) = Cn rn + Dn rn (An cos(n) + Bn sin(n)) . By superposition, u(r, ) =
( n=0

) Cn rn + Dn rn (An cos(n) + Bn sin(n)) .

The bounded interior solution is (Dn = 0 since u(0, ) is bounded): u(r, ) =

n=0 n=0

Cn rn (An cos(n) + Bn sin(n))

rn (an cos(n) + bn sin(n)) .

The boundary condition implies that u(R, ) = Thus 1 an = Rn Example. Solve P DE : BC : 1 1 urr + ur + 2 u = 0, r < 3; r r u(3, ) = 27 cos(2) 54 sin(3). 64
0 n=0

Rn (an cos(n) + bn sin(n)) = f ().

0

f () cos(n)d,

1 bn = Rn

f () sin(n)d.

Solution: R = 3. The boundary condition implies that u(3, ) = Thus 32 a2 = 27,

n=0

3n (an cos(n) + bn sin(n)) = 27 cos(2) 54 sin(3).

33 b3 = 54, other an and bn are 0, so then { 3, if n = 2; an = 0, if n = 2. { bn = 2, if n = 3; 0, if n = 3.

u(r, ) = 3r2 cos(2) 2r3 sin(3). 4.4.3 Laplace equation outside a circle

Consider the boundary value problem: P DE : BC : 1 1 urr + ur + 2 u = 0, r r u(a, ) = f (). r > a;

Theorem 17. Let f be continuous, 2 -periodic, f be piecewise continuous. Let u(r, ) be bounded in R. Then the solution of the PDE above is a 0 n u(r, ) = + r [an cos(n) + bn sin(n)], 2 n=1

where

an an = an bn =

f () cos(n)d,
2

n 0, n 1.

f () sin(n)d,
0

Example. Find the bounded solution of urr + 1 u + r12 u = 0 outside the circle r = 2, r r subject to the boundary condition u(2, ) = 4 cos2 (2) 5 sin(3). Solution: a = 2. The solution of the PDE above is a 0 n + r [an cos(n) + bn sin(n)]. u(r, ) = 2 n=1

65

Note that 4 cos2 (2) = 2[cos(4) + 1]. Thus u(2, ) = 2 + 2 cos(4) 5 sin(3). Therefore a0 n 2 + 2 cos(4) 5 sin(3) = + 2 [an cos(n) + bn sin(n)], 2 n=1

which implies that a0 = 4, 2 = 24 a4 , an = 0 for n = 0, 4; 5 = 23 b3 , bn = 0 for n = 3. Hence a0 = 4, a4 = 32, an = 0 for n = 0, 4; b3 = 40, bn = 0 for n = 3. Hence u(r, ) = 2 + 32r4 cos(4) 40r3 sin(3). 4.4.4 Laplace equation within an annulus

Consider the boundary value problem: P DE : BC : 1 1 urr + ur + 2 u = 0, a < r < b; r r u(a, ) = f (), u(b, ) = g ().

This is a combination of the above two sections. Theorem 18. Let f and g be continuous, 2 -periodic, f and g be piecewise continuous. Then the solution of the PDE above is a0 + b0 ln r u(r, ) = + [(an rn + bn rn ) cos(n) + (cn rn + dn rn ) sin(n)], 2 n=1

where a0 + b0 ln a = an a + bn a cn a + dn a
n n n

= =

a0 + b0 ln b = a n b + bn b
n n n

= =

cn b + dn b

1 1 1 1 1 1

f ()d,
2

f () cos(n)d,
0 2

n 1, n 1,

f () sin(n)d,
0 2

g ()d,
0 2

g () cos(n)d,
0 2

n 1, n 1.

g () sin(n)d,
0

66

Example. Solve P DE : BC : 1 1 urr + ur + 2 u = 0, 1 < r < 3; r r u(1, ) = 2 cos2 () 1, u(3, ) = sin(3).

Solution: a = 1, b = 3, f () = cos(2), g () = sin(3). Method 1: 1 2 cos(2)d = 0, a0 + b0 ln 1 = 0 { 2 1, if n = 2; 1 an 1n + bn 1n = cos(2) cos(n)d = 0 0, if n = 2. 2 1 cn 1n + dn 1n = cos(2) sin(n)d = 0, n 1, 0 1 2 a0 + b0 ln 3 = sin(3)d = 0, 0 1 2 n n sin(3) cos(n)d = 0, n 1, an 3 + bn 3 = 0 { 1, if n = 3; 1 2 n n cn 3 + dn 3 = sin(3) sin(n)d = 0 0, if n = 3. Method 2: We start from the boundary conditions directly. a0 + b0 ln r u(r, ) = + [(an rn + bn rn ) cos(n) + (cn rn + dn rn ) sin(n)] 2 n=1

a0 u(1, ) = + [(an + bn ) cos(n) + (cn + dn ) sin(n)] = cos(2) 2 n=1 a0 = 0, a2 + b2 = 1, an + bn = 0 (n = 2), cn + dn = 0 (n 1). 2

(46)

Similarly, a0 + b0 ln 3 u(3, ) = + [(an 3n + bn 3n ) cos(n) + (cn 3n + dn 3n ) sin(n)] = sin(3) 2 n=1

cn 3n +dn 3n = 0 (n = 3). (47) 1 81 27 By (46) and (47), a2 = 80 , other an = 0; b2 = 80 , other bn = 0; c3 = 728 , other cn = 0; 27 d3 = 728 , other dn = 0. Thus ) ( ) ( 27 3 27 3 1 2 81 2 cos(2) + r r sin(3). u(r, ) = r + r 80 80 728 728 67

a0 + b0 ln 3 = 0, 2

an 3n +bn 3n = 0 (n = 1),

c3 33 +d3 33 = 1,

V Sturm-Liouville Problems
a0 (x)y + a1 (x)y + a2 (x)y = 0, a0 (x) = 0

Every linear second-order ordinary dierential equation of the form

can be written in divergence form, [p(x)y ] + s(x)y = 0, where

p(x) = e

a1 (x)/a0 (x)dx

s(x) =

a2 (x)p(x) , a0 (x)

p(x) is called an integrating factor of the dierential eqn. Example 55. Lengendres equation can be written as: [(1 x2 )y ] + n(n + 1)y = 0.
Example 56. Bessels equation z 2 yz + zyz + (z 2 2 )y = 0 can be written as: ) ( 2 2 (xy ) + + k x y = 0. x

Solution: Let z = kx. Then

yz = y /k, yz = y /k 2 .

Hence x2 y + xy + (k 2 x2 2 )y = 0. Dividing by x we get the result.

5.1

Sturm-Liouville Problems

In IV we solved several initial and boundary value problems. While solving these equations we used the method separation of variables which reduces the problem to one of the following types of Sturm-Liouville problems Denition 9. Sturm-Liouville Problem is the problem which consists in nding the nonzero solutions on [a, b] of the second-order ordinary dierential equation (Sturm-Liouville equation): [p(x)y ] + [r(x) q (x)]y = 0 (48) 68

with the boundary conditions k1 y (a) + k2 y (a) = 0, (k1 , k2 ) = (0, 0) l1 y (b) + l2 y (b) = 0, (l1 , l2 ) = (0, 0). (49)

The values of for which there exist nonzero solutions (non-trivial solution) are called eigenvalues, and the corresponding nonzero solutions are called eigenfunctions. The function r(x) will be called the weight function. Remark 1. If p, q , r, p are real-valued and continuous on the interval and p(x) and r(x) are positive throughout that interval (or negative throughout that interval), then all the eigenvalues are real. This is important since eigenvalues are often related to frequencies, energies, or other physical quantities that must be real. Weight function: Example 57. Given the following SLP y + a(x)y + b(x)y = 0, Find the weight function. Solution: By using the integrating factor we get

c < x < d.

r(x) = b(x)e Example 58. Given the following SLP y + 2xy + 3y = 0, Find the weight function. Solution:

a(x)dx

c < x < d.

r(x) = 3e

2xdx

= 3e x .

Now we discuss how to get eigenvalues and eigenfunctions. Example 59. Find the eigenvalues and eigenfunctions: y + y = 0, y (0) = 0, y (L) = 0.

69

Solution: r(x) = 1, q (x) = 0, p(x) = 1, a = 0, b = L, k1 = 1, k2 = 0, l1 = 0, l2 = 1. (i) When < 0. Set = 2 . Then we have y 2 y = 0. The solution is y = cex + dex . The two initial conditions y (0) = 0, y (L) = 0 imply that { c + d = 0, ceL deL = 0. We imply that c = d = 0, and so y 0, which is not an eigenfunction. (ii) When = 0. Then y = 0, hence y = cx + d. Combining this with the two initial conditions we also get c = d = 0, and so, y = 0. (iii) When > 0. Set = 2 . Then we have y + 2 y = 0. The solution is y = c cos x + d sin x. The two initial conditions y (0) = 0, y (L) = 0 imply that { c = 0, c sin L + d cos L = 0. ) (k = 0, 1, 2, ...). Take d = 1. Then We imply that c = 0, L = (k + 1 2 y (x) = sin x. Hence the eigenfunctions are ( y (x) = sin and the eigenvalues are = ( ) 2k + 1 x , 2L )2 , k = 0, 1, 2, .... k = 0, 1, 2, ...

2k + 1 2L

Example 60. Find the eigenvalues and eigenfunctions: (xy ) + y = 0, x Solution: Set x = et . Then y = which implies that d (xy ) = dx

y (1) = 0, y (e) = 0.

dy dy = et , dx dt

( ) ( ) d dy t dy xe = = et yt . dt dx dt 70

Note that, when x = 1, we have t = 0; when x = e, t = 1. Thus we obtain

yt + y = 0, y (0) = 0, yt (1) = 0.

This means: r(x) = 1, q (x) = 0, p(x) = 1, a = 0, b = 1, k1 = 1, k2 = 0, l1 = 0, l2 = 1. (i) When < 0. Set = 2 . Then we have
2 y = 0. yt The solution is y = cet + det . The two initial conditions y (0) = 0, yt (1) = 0 imply that { c + d = 0, ce de = 0.

We imply that c = d = 0, and so y 0, which is not an eigenfunction. (ii) When = 0. Then yt = 0, hence y = ct + d. Combining this with the two initial conditions we also get c = d = 0, and so, y = 0. (iii) When > 0. Set = 2 . Then we have
yt + 2 y = 0. The solution is y = c cos t + d sin t. The two initial conditions y (0) = 0, yt (1) = 0 imply that { c = 0, c sin + d cos = 0.

We imply that c = 0, = (k + 1 ) (k = 0, 1, 2, ...). Take d = 1. Then 2 y (t) = sin t. Note that t = ln x. Hence the eigenfunctions are ( ) 2k + 1 y (x) = sin ln x , 2 and the eigenvalues are = ( 2k + 1 2 )2 , k = 0, 1, 2, ....

k = 0, 1, 2, ...

Example 61. Consider the Sturm-Liouville problem y + 2y + y = 0, y (0) + y (0) = 0, y (1) = 0. (a) Find all eigenvalues n and corresponding eigenfunctions yn (x). (b) Determine the weight function. 71

Solution: (a) The indicial (characteristic) eqn is: r2 + 2r + = 0, r = 1 (i) = 1. Then r = 1 is the only root. y (x) = (A + Bx)ex .

1 .

From the boundary conditions, A = 0 and B = 0. Thus, y = 0, = 0 is not an eigenvalue. (ii) < 1. Let = 1 2 , r = 1 , y (x) = Ae(1+)x + Be(1)x . From the boundary conditions, A = 0 and B = 0. Thus, < 1 is not an eigenvalue. (iii) > 1. Now let = 1 + 2 , r = 1 i , y (x) = ex (A cos x + B sin x). We have y (0) = A, y (0) = A + B. By y (0) + y (0) = 0 we imply that B = 0 and y (x) = ex A cos x. Then by y (1) = 0 we have (2n + 1) (2n + 1)2 2 cos = 0 = , n = 1 + . 2 4 ( ) (2n + 1)x x yn (x) = Ae cos . 2 (b) I (x) = e
2dx

= e2x .

Multiply two sides of the equation by I (x) we imply that e2x y + 2e2x y + e2x y = 0 [e2x y ] + e2x y = 0. Thus the weight function is e2x .

Example 62. Solve the Sturm-Liouville problem: Y + 2Y + ( + 1)Y = 0, Y (0) = 0, Y (2) = 0.

Solution: The characteristic eqn is: c2 + 2c + ( + 1) = 0, c = 1 . (i) = 0. Then c = 1. Y (x) = (A + Bx)ex . From the boundary conditions, A = 0 and B = 0. Thus, = 0 is not an eigenvalue. 72

(ii) Now let = 2 , c = 1 , Y (x) = Ae(1+)x + Be(1)x . From the boundary conditions, A = 0 and B = 0. Thus, = 0 is not an eigenvalue. (iii) Now let = 2 , c = 1 i , Y (x) = ex (A cos x + B sin x). From the boundary conditions, A = 0 and sin 2 = 0. Thus = n , and 2 Yn (x) = ex sin nx . 2

5.2

Regular and Singular Sturm-Liouville Problems

Case 1. Regular problem, p(a) = 0, p(b) = 0. We have seen examples. Case 2. Problem with one singular point, p(a) = 0, say. In this case, a single boundary condition at x = b, suces, l1 y (b) + l2 y (b) = 0, (l1 , l2 ) = (0, 0). Then we require that y and y are bounded at x = a. A similar situation holds if p(b) = 0.

Example 63. Solve the Bessel equation on [0, R]: ) ( 2 n 2 [xy ] + + x y = 0 x with the boundary conditions y (0) and y (0) bounded, y ( R ) = 0.

Solution: Consider the standard Bessels equation of order n:

z 2 yz + zyz + (z 2 n2 )y = 0.

Let z = x. Then we get the given equation. One of the solution is Jn (z ) = Jn (x). The general solution is y = c1 Jn (x) + c2 Yn (x). We take c2 = 0 because Yn is unbounded at the origin, so then y = c1 Jn (x). Jn (z ) has innitely many zeros, say mn : 1n < 2n < . To satisfy the condition at x = R, let mn = mn /R. Then Jn (mn R) = Jn (mn ) = 0. By Theorem ??, the sequence of the Bessel functions of the rst kind Jn (1n x), Jn (2n x), Jn (3n x), forms an orthogonal set on the interval 0 x R with respect to the weight r(x) = x.

73

Case 3. Problem with two singular points, p(a) = p(b) = 0. In this case, no boundary conditions are needed.

Example 64. (Orthogonality of Legendre polynomials) Legendres equation is a SturmLiouville equation [r(x)y ] + y = 0. here p(x) = 1, q (x) = 0. Solution: Since r(1) = r(1) = 0, we need no boundary conditions. For n = 0, 1, 2, , = 0, 1 2, 2 3, ,the Legendre polynomials Pn (x) are solutions of the problem. Hence these are the eigenfunctions. From Theorem ??, they are orthogonal, i.e., 1 Pm (x)Pn (x) dx = 0 (m = n).
1

r(x) = 1 x2 , = n(n + 1),

Case 4. Periodic Sturm-Liouville problem: If p(a) = p(b) = 0, then (49) can be replaced by periodic boundary condition y (a) = y (b), y (a) = y (b).

Example 65. Solve X + X = 0 with periodic boundary conditions X (0) = X (2 ) and X (0) = X (2 ). ( )2 Solution: The eigenvalues are n = n with corresponding eigenfunctions Xn (x) = 2 An cos(nx) + Bn sin(nx) for n = 0, 1, 2, . . . . Notice that 0 = 0 is an eigenvalue since it corresponds to X0 (x) = A0 which is not necessarily zero.

5.3

Bessels equation

Example 66. Bessels equation can be written as a Sturm-Liouville equation. Solution: Consider the Bessels equation x2 y (x) + xy (x) + (2 x2 2 )y (x) = 0. Divide by x2 on two sides, 2 1 2 y (x) + y (x) + ( 2 )y (x) = 0. x x

74

 1 Hence, by multiplying integrating factor exp x dx = exp[ln x] = x, ( 2 ) 2 (xy ) + + x y = 0, x here p(x) = x, q (x) = 2 /x, r(x) = x. We now discuss properties of Bessels theory, not only because of their great practical importance but also as a model case for showing how properties of functions can be discovered from their series. Propositions: d [x J (x)] = x J 1 (x), > 0; dx ] d [ x J (x) = x J +1 (x), 0; dx 2 J 1 (x) + J +1 (x) = J (x), 0; x J 1 (x) J +1 (x) = 2J (x), 0; 2 J1/2 (x) = sin x; x 2 J1/2 (x) = cos x. x Proof. Recall that J (x) = x

(50) (51) (52) (53) (54) (55)

(1)m x2m . 2m+ m!(m + + 1) 2 m=0

(56)

Multiplying the two sides by x we get x J (x) = (1)m x2m+2 . 2m+ m!(m + + 1) 2 m=0

We dierentiate this, by using recursive relation of Gamma function,

(1)m (1)m 2(m + ) 2m+2 1 1 x = x x x2m . (x J (x)) = 2m+ m!(m + + 1) 2m+ 1 m!(m + ) 2 2 m=0 m=0

This proves (50). Similarly, multiplying the two sides of (56) by x , then calculate derivative, then replace m by m + 1, we can get (51). By (50) and (51),
= x J 1 , x 1 J (x) + x J x 1 J (x) + x J = x J +1 .

75

From these two equalities we imply (52) and (53). Since ( 1 ) = . We have 2 3 (m + ) = 2(m+1) (2m + 1)(2m 1) 3 1 . 2 Now for = 1 , 2
J1/2 (x) = x

= = = This proves (54). Now by (50),

(1)m x2m = 2m+1/2 m!(m + 1/2 + 1) 2 m=0

2 (1)m x2m+1 x m=0 22m+1 m!(m + 3/2)

2 (1)m x2m+1 x m=0 22m+1 m!2(m+1) (2m + 1)(2m 1) 3 1 2 (1)m x2m+1 x m=0 (2m + 1)!

2 sin x. x

x This proves (55). Example 67. Integrate Example 68. Integrate

1/2

[ ] J1/2 (x) = xJ1/2 (x) =

2 cos x.

x3 J0 (x)dx. I=
1 2

x3 J4 (x) dx.

Solution. By (51), I=
1 2

[ 3 ] x J3 (x) dx = 23 J3 (2) J3 (1).

Now by (52) and (53), J3 (x) = We imply that J3 (x) = 4 J2 (x) J1 (x), x ( J2 (x) = 2 J1 (x) J0 (x). x

) 8 4 1 J1 (x) J0 (x). 2 x x

(57)

From the table we can nd the value I = 0.0038. 76

Example 69. Represent J4 (x) by using J0 (x) and J1 (x). Solution: From J 1 (x) + J +1 (x) = we imply that 2 J1 (x), x 4 J1 (x) + J3 (x) = J2 (x), x 6 J2 (x) + J4 (x) = J3 (x), x 8 J3 (x) + J5 (x) = J4 (x). x J0 (x) + J2 (x) = Hence J2 (x) = 2 J1 (x) J0 (x), x ( ) 8 4 J3 (x) = 1 J1 (x) J0 (x), 2 x x ( ) ( ) 48 8 24 J4 (x) = J1 (x) 1 J0 (x). x3 x x2 2 J (x) x

Theorem 19. The Bessel functions of the rst kind, Jn (x), satisfy the following orthogonality relations: { R 0, if m = k ; (Jn (nm x), Jn (nk x)) := xJn (nm x)Jn (nk x) dx = 2 R J 2 ( R), if m = k . 0 2 n+1 nk where nk =
nk R

and nk is the kth strictly positive zero of Jn (x).

0 4 2 xJ5 (5k x) dx =

Example 70.

R2 2 2 J (nk R) = 8J6 (5k ). 2 n+1

77

6
6.1

VI Fourier Transform
Fourier transform and its inverse

Denition 10. The Fourier transform of f : () := F (f ) := f

f (x)eix dx

(58)

(): and the following inverse Fourier transform of f 1 1 ()eix d. F (f ) := f (x) = f 2

(59)

() can be considered as spectral density, which measures the Remark. The transform f intensity of f (x) in the frequency domain. Remark. The unitary forms of the Fourier transform are: 1 1 ix 1 ) := f (x) = ()eix d. () := f (x)e dx, F (f f F (f ) := f 2 2 Existence: If f is absolutely integrable, i.e., |f (x)|dx < , then F (f ) exists. Simi is absolutely integrable, then F 1 (f ) exists. larly, if f Example 71. Find the Fourier transform of the function { k, if 0 < x < a; f (x) = 0, otherwise, Solution: From (58) by integration, () = F (f ) = f =
0

f (x)eix dx k (eia 1) . i

keix dx =

Example 72. Show that F (u(x a) u(x b)) = Proof: From (58) by integration, F (u(x a) u(x b)) = =
a b

eib eia . i

(u(x a) u(x b))eix dx eib eia . i

eix dx =

78

Example 73. Show that F (e|x| ) = 2 . 1 + 2

Proof: From (58) by integration, |x| F (e ) = e|x| eix dx 0 x ix = e e dx + ex eix dx =

0

1 1 e(1+i)x |0 e(1+i)x | + 0 1 + i 1 + i

1 1 2 = = . 1 + i 1 + i 1 + 2 Remark. lim e(1+i)x = 0 = lim e(1+i)x .

x x

Example 74. Show that F (a ) = eia . Proof: Note that a (f ) = F (a ) =

(x a)f (x)dx = f (a).

(x a)eix dx = eix |x=a = eia .

6.2

Linearity

The Fourier transform is a linear operation, that is, the following proposition holds. Proposition ( Linearity of the Fourier transform) For constants a and b, and two functions g (x) and h(x), we have F (ag + bh) = aF (g ) + bF (h). By using (58) we can easily prove it. (), then for any real number , Theorem 20. If F (f ) := f ( ) { ( )} 1 1 F{f (x)} = f , F = ||f (x). f || Proof. Its easy to get this by substitution y = x. Example 75. For t > 0, nd F (et|x| ).

79

Solution: Let f (x) = e|x| . Note that () = F (e|x| ) = F (f (x)) = f By the above theorem, F (e
t|x|

2 . 1 + 2

( ) 1 ) = F (f (tx)) = f t t

= Example 76. For t > 0, F { 1

2t 1 2 . ( )2 = 2 t1+ t + 2 }
t

= et|x| . { } Example 77. By the denition, F 1 e|| =

2t t 2 + 2

1 . (1+x2 )

6.3

Shifting theorem

(). Then for any real Theorem 21. Suppose that f is absolutely integrable and F (f ) := f number a, The rst Shifting Theorem: { } ( + a), F eiax f (x) = f The second Shifting Theorem: (), F {f (x a)} = eia f { } () = f (x a). F 1 eia f { } { } ( + a) = eiax F 1 f () . F 1 f

Proof. By the denition we will get both of them.

Example 78.

2eia F (e|x+a| ) = . 1 + 2 { } eia 1 1 F = ebi(x+a)|x+a| . 2 1 + ( b) 2 F (eiax|x| ) = 2 . 1 + ( + a)2

80

6.4

Fourier transform and derivatives

Theorem 22. (Fourier transform of the derivative) Let f (x) be continuous on the x-axis, and let f (x) 0 as x . Then F (f (x)) = iF (f (x)), F (f (x)) = F (f (x)).
2

(60) (61)

Proof. By (58), we deduce from integration by parts that F (f (x)) = f (x)eix dx [ ] ix = f (x)e (i) f (x)eix dx = iF (f (x)).

This proves (60). Now by using (60) twice we imply that F (f (x)) = iF (f (x)) = (i)2 F (f (x)) = 2 F (f (x)). This proves (61). Theorem 23. (Derivative of Fourier transform) Suppose that f (x) and xf (x) are abso(). Then f is dierentiable and lutely integrable and F (f ) := f { } () () df d f F{xf (x)} = i , F 1 = ixf (x). d d Proof. Dierentiating the two sides () = f will give the result. Example 79. F (e
tx2

f (x)eix dx

2 ) = e /(4t) , t

{ } 1 2 2 F 1 et = ex /(4t) . 2 t

2 () + 2t(i)f () = 0, Proof. Let f (x) = etx . Then f (x) + 2txf (x) = 0, if

() df () = ce2 /(4t) , = d, f () 2t f 81

(0) = c=f

tx2

dx = . t

Example 80. Find the Fourier transform of the function f (x) = xex by using 2 tx2 ) = e /(4t) . F (e t
2

Solution: By the linearity, ) ( ) ( ) 1 x2 1 ( 2 x2 F xe = F [e ] = F [ex ] 2 2 ( ) 1 i 2 /4 x2 e = (i)F e = 2 2 1 i 2 /4 = e . 2 Example 81. Find { } 2 /16 1 F e . 16 Solution: By the above theorem, { } { ( )} 2 /16 d 2 /16 1 1 F e = F e 16 d 2 { } 2 /16 1 = ixF e 2 = ixe4x .
2

6.5

Convolution

The convolution f g of functions f and g is dened in Part I: f (x) g (x) = (f g )(x) := f (v )g (x v )dv = f (x v )g (v )dv.

Theorem 24. (Convolution theorem) Let f (x) and g (x) be piecewise continuous, bounded, and absolutely integrable on the x-axis. Then F (f g ) = F (f )F (g ), () (f g )(x) = F 1 {f g ()}.

ex , if x 0; Example 82. Let f (x) = g (x) = 0, if x < 0, F {f (x)}, F {g (x)}, F {f (x) g (x)}. e|x| , F {f (x)} = Solution: f (x) g (x) = 1 2 1 . 1+2 82
1 , 1i

ex , if x 0; Find f (x) g (x), 0, if x > 0.

1 , 1+i

F {g (x)} =

F {f (x) g (x)} =