Math 201 Lecture 18: Convolution

Feb. 17, 2012

Many examples here are taken from the textbook. The rst number in () refers to the problem number in the UA Custom edition, the second number in () refers to the problem number in the 8th edition. 0. Review

Laplace transform of functions with jumps: 1. Represent the function using unit jump functions: g (t) = g1(t) + [ g2(t) g1(t)] u(t t1) + 2. Transforming each g(t) u(t a) as follows: a. Obtain g (t + a). b. Obtain L{ g (t + a)}. c. L{ g (t) u(t a)} = eas L{ g (t + a)}(s). + [ gk(t) gk 1(t)] u(t tk 1). (1)

Inverse Laplace transform of functions involving eas: 1. Write the function to be inverse transformed as eas F (s). 2. Obtain f (t) = L1{F (s)}. 3. Obtain f (t a). 4. The inverse transform is f (t a) u(t a).

Laplace transform of periodic functions: If f has period T and is piecewise continuous on [0, T ], then L{ f }(s) =
T 0

Quiz: L{ f (t)}, f (t) =

est f (t) dt . 1 esT L 1 e 3 s . s3

(2)

1 0<t<2 ; t t>2

(3)

1. Basic Information Equation with general right hand sides and general initial conditions. In practice, we often face the following situation: We need to compute the output of a system for many dierent inputs, sometimes we also need to nd out a general relation between the inputs and the outputs. After mathematical modeling, this situation is translated to the following. Given a dierential operator (corresponding to the system), we need to compute the solutions for many dierent right hand sides/initial values. For example, we may need to consider problems like y + y = g(t); and need to solve it for many dierent g s. We follow our three steps. 1. Transform the equation. Using the initial values and denoting Y (s) = L{ y }, G(s) = L{ g }, we have (s2 + 1) Y = G + s a + b. 1 (5) y (0) = a, y (0) = b (4)

Math 201 Lecture 18: Convolution

2. Solve the transformed equation. We have Y (s) = 1 [G(s) + s a + b]. s2 + 1 (6)

3. Transform back. We need to compute ( s) y = L1{Y } = L1 H (s) G (7)

where H (s) 6

1 , s2 + 1

(s) 6 G(s) + s a + b. and G

From the above we can make the following observations. A) The Laplace transform of the solution is the product of two functions. One of them, H (s), is (s), is only dependent on determined by the dierential operator (the system); The other, G the right hand side and the initial values (the inputs). Thus it is possible to simplify the process of solving equations with many dierent inputs, as H (s) remains the same. B) To be able to take advantage of the above observation, we need to be able to compute the inverse Laplace transforms of product of two functions. Or more specically, if we know that using h and g =g ? , can we obtain L1 H G L1{H } = h, L1 G The answer is yes: = h g L 1 H G , (8) the convolution of h and g .

Transfer function. Denition 1. (Transfer function/Impulse response function) Consider the linear system a y + b y + c y = g, y (0) = y (0) = 0, (9)

Let Y = L{ y }, G = L{ g }. The function H (s) = Y /G is called the transfer function of the linear system. Its inverse Laplace transform, h(t) = L1{H (s)}(t) , is called the impulse response function for the system. Remark 2. We will see later that h(t) solve the equation a y + b y + c y = (t) where (t) is a special function such that L{ } = 1. (11) (10)

It turns out that this (t) cannot be a usual function. It belongs to the so-called generalized functions. Example 3. Find the transfer function and the impulse response function for y + 9 y = g(t). (12)

Solution. Taking Laplace transform we have (recall that when talking about transfer function, we always assume y (0) = y (0) = 0. ) s2 Y + 9 Y = G So the transfer function is H (s) = 1 . s2 + 9 (14) H= Y 1 = . G s2 + 9 (13)

Feb. 17, 2012

The impulse response function is then given by h(t) = L1{H } = Convolution. Denition 4. (Convolution) Let f (t) and g(t) be piecewise continuous on [0, ). The convolution of f (t) and g(t) , denoted f g, is dened by
t

1 sin 3 t. 3

(15)

( f g )(t): =

f (t v ) g (v ) dv.

(16)

Example 5. For example, the convolution of f = t and g = t2 is

t t

t t2 =

(t v ) v 2 dv =

t v 2 v 3 dv =

1 4 1 4 t4 t t = . 3 4 12

(17)

The key property of convolution is the following Theorem 6. (Convolution Theorem) Let f (t) and g(t) be piecewise continuous on [0, ) and of exponential order and set F (s) = L{ f }(s) and G(s) = L{ g }(s). Then or equivalently L{ f g }(s) = F (s) G(s), L1{F (s) G(s)} = ( f g)(t). Example 7. (7.7 13; 7.7 13) Find the Laplace transform of
t

(18) (19)

f (t) =
0

(t v ) e3 v dv.

(20)

Solution. We recognize that in fact Thus f (t) = te3 t. L{ f } = L{t } L{e3 t } = What convolution can do: 1 1 1 = . s2 s 3 s2 (s 3) (21)

1. An alternative method of computing inverse Laplace transforms; 2. Enable us to solve special integral-dierential equations; 3. Obtain formula for solution when the right hand side or initial values are not given. 2. Things to be Careful/Tricky Issues See Common Mistakes for examples. 3. Examples Convolution as alternative method fo inverse Laplace transforms: Example 8. (7.7 5; 7.7 5) Compute L 1 1 s (s2 + 1) (22)

Math 201 Lecture 18: Convolution

Solution. Using Convolution Theorem, we have L 1 1 s (s2 + 1) = L 1

t

1 1 L1 2 s s +1 = 1sin t sin v dv
0

= 1 cos t. Example 9. (7.7 12; 7.7 12) Compute L 1 = L 1 s+1 . (s2 + 1)2

(23)

(24)

Solution. We have L 1 s+1 (s2 + 1)2

s+1 1 s2 + 1 s2 + 1 s+1 1 = L 1 2 L 1 2 s +1 s +1 = [cos t + sin t]sin t

t

=
0 t

cos (t v ) sin v dv +

t 1 1 [sin t sin (t 2 v )] dv + [cos (t 2 v ) cos t] dv 0 2 0 2 t t 1 1 t (sin t cos t) = sin (t 2 v ) dv + cos (t 2 v ) dv 2 0 2 0 1 1 =t t t (sin t cos t) [cos (t 2 v )N v = v =0 sin (t 2 v )N 0 ] 4 2 1 1 = t (sin t cos t) sin t. 2 2

sin (t v ) sin v dv

(25)

Solving integral-dierential equations: Example 10. (7.7 15; 7.7 15 ) Solve

t

y (t) + 3
0

y (v ) sin (t v ) dv = t.

(26)

Solution. We try following the same three steps. 1. Transform the equation. The key is to notice that the integral term is in fact a convolution:
t 0

y (v ) sin (t v ) dv = y (t)sin t.

(27)

Thus taking Laplace transform of both sides, we have Y (s) + 3 Y (s) 2. Solve the transformed equation. We have Y ( s) = 1 + s2 1 1 1 1 3 1 = 2 3 = . 2 2 2 2 (4 + s ) s s (4 + s ) s 4 s2 4 s 2 + 4 (29) 1 1 = . 1 + s2 s2 (28)

3. Transform back. Compute y (t) = L1 1 1 3 1 4 s2 4 s2 + 4 = t 3 sin 2 t. 4 8 (30)

Feb. 17, 2012

Example 11. (7.7 21; 7.7 21 ) Solve

t

y (t) + y (t) Solution. Again,

y (v ) sin (t v ) dv = sin t,

y(0) = 1.

(31)

1. Transform the equation. We have and

t

L{ y } = s Y y (0) = s Y 1. L y (v ) sin (t v ) dv = Y (s) L{sin t } = Y (s) . 1 + s2

(32) (33)

So the equation is transformed into sY +Y 1 2. Solve the transformed equation. We have Y= 3. Transform back. Compute y (t) = L1{Y } = L 1 = e
2 t
1

1 Y (s) = . 1 + s2 1 + s2

(34)

s . s2 + s + 1

(35)

1/2 s + 1/2 2 2 (s + 1/2)2 + 3 /2 (s + 1/2) + 3 /2 1 3t 3t . sin cos 2 2 3

(36)

Obtain formula of solution for general right hand side: Example 12. (7.7 1; 7.7 1 ) Use the convolution theorem to obtain a formula for the solution to the given initial value problem. y 2 y + y = g (t); y (0) = 1, y (0) = 1. (37) (38) (39)

Solution. Using

L{ y } = s2 Y s y(0) y (0) = s2 Y + s 1, L{ y } = s Y y (0) = s Y + 1, we have the transformed equation (s2 2 s + 1) Y = G(s) s + 3 Transforming back, we have L 1 1 G(s) (s 1)2 1 L 1 s1 2 L 1 (s 1)2 = L 1 = et; = 2 et t. Y (s) = 1 1 2 G(s) + . (s 1)2 (s 1) (s 1)2

(40)

1 g(t) = (et t) g (t); (s 1)2

(41) (42) (43)

So the formula for the solution reads

t

y (t) = 2 t et et + (et t) g(t) = 2 t et et +

et v (t v ) g(v ) dv.

(44)

Math 201 Lecture 18: Convolution

Finally lets do an example that combined both 7.6 and 7.7. Example 13. (7.6 39; 7.6 39) Solve y + 5 y + 6 y = g (t), where 0 0 t<1 g (t) = t 1 < t < 5 . 1 5<t (46) y (0) = 0, y (0) = 2, (45)

Solution. 1. Transform the equation.

a. Transform the LHS. Compute L{ y } = s2 Y s y (0) y (0) = s2 Y 2; So L{ y } = s Y y (0) = s Y . L{ y + 5 y + 6 y } = (s2 + 5 s + 6) Y 2. (47) (48) (49)

b. Transform the RHS. We rst need to represent g using unit step functions. There are two discontinuities at t = 1, 5. We write and determine So Recalling We compute g(t) = 0 + A(t) u(t 1) + B (t) u(t 5) A(t) = t, B (t) = 1 t. (50) (51) (52) (53)

g (t) = t u(t 1) + (1 t) u(t 5). L{ g(t) u(t a)} = eas L{ g(t + a)}(s)

L{t u(t 1) + (1 t) u(t 5)} = es L{t + 1} + e5 s L{t 4} 1 1 1 4 + = e s e 5 s 2 + . s s s2 s Thus the transformed equation reads (s2 + 5 s + 6) Y = es 2. Solve the transformed equation. We have Y (s) = es s+1 4s+1 2 e 5 s 2 2 + . s2 (s2 + 5 s + 6) s (s + 5 s +6) s2 + 5 s + 6 1 1 + s s2 e5 s 1 4 + s2 s + 2.

(54)

(55)

(56)

3. Transform back. We need to use the formula L1{eas F (s)} = f (t a) u(t a). Thus we have to compute the inverse transforms of of the inverse transform, we rst compute L 1 s s2 (s2 + 5 s + 6) and L1
s+1 s2 (s2 + 5 s + 6)

(57) and
4s+1 . s2 (s2 + 5 s +6)

Due to linearity

1 . s2 (s2 + 5 s + 6)

(58)

Feb. 17, 2012

We have L 1 s s2 (s2 + 5 s + 6) 1 s (s2 + 5 s + 6) 1 1 = L 1 s (s + 2) (s + 3) 1 1 1 = L 1 L1 s s+2 s+3 = 1(e2 t e3 t } = L 1

t

=
0

( e 2 v e 3 v ) d v

L 1

1 s2 (s2 + 5 s + 6)

1 1 2 t 1 1 3 t e + e 3 3 2 2 1 1 2 t 1 3 t = e + e . 6 2 3 1 1 1 = L1 2 L1 s s+2 s+3 = t(e2 t e3 t) =

t

(59)

=
0 t

(t v ) (e2 v e3 v ) dv

= = = = = Thus we have which leads to L 1 e s L 1

1 1 ( t v ) d e 2 v + e 3 v 2 3 t 1 1 1 2 v 1 3 v v = t + e N e2 v + e3 v d(t v ) (t v ) e v =0 3 2 3 2 0 t 1 1 1 1 e2 v + e3 v dv t t+ 2 2 3 3 0 1 1 2 t 1 1 3 t 1 t+ e e + 6 4 4 9 9 1 1 1 5 t + e 2 t e 3 t . (60) 6 4 9 36
0

s+1 s2 (s2 + 5 s + 6) =

t 1 2 1 + e 2 t + e 3 t 9 36 6 4

(61)

s+1 s2 (s2 + 5 s + 6)

t 1 1 2(t 1) 2 3(t 1) 1 + e + e u(t 1); 6 4 9 36

(62)

For the other term we have L 1 4s+1 s2 (s2 + 5 s + 6) = = 19 t 7 2 t 11 3 t + e + e 9 36 6 4 (63)

which leads to L 1 e 5 s

4s+1 s2 (s2 + 5 s + 6)

19 t 5 7 2(t 5) 11 3(t 5) u(t 5). + e + e 6 4 9 36

(64)

Finally we quickly compute L 1 2 s2 + 5 s + 6 = 2 L 1 1 1 s+2 s+3 = 2 e2 t 2 e3 t. (65)

Summarizing, we reach y ( t) = t 1 1 2(t 1) 2 3(t 1) 1 + e + e u(t 1) 6 4 9 36 19 t 5 7 2(t 5) 11 3(t 5) u(t 5) + 2 e2 t 2 e3 t. + e + e 6 4 9 36

(66)

Math 201 Lecture 18: Convolution

Remark 14. Note that for the above problem, it is actually more ecient to compute the inverse transforms using partial fractions. For example, to compute L 1 s+1 s2 (s2 + 5 s + 6) (67) (68) (69)

we write which leads to

s+1 As+B C D = + + s2 (s2 + 5 s + 6) s2 s+2 s+3 s + 1 = (A s + B ) (s + 2) (s + 3) + C s2 (s + 3) + D s2 (s + 2).

Setting s to be special values, we get 1 B= ; 6 1 s = 2 1 = 4 C C = ; 4 2 s = 3 2 = 9 D D= ; 9 3 Finally comparing coecients for the highest order term s we get s=0 1=6B 0=A+C +D A= 1 . 36 (70) (71) (72)

(73)

4. Notes and Comments Other properties of convolution. Theorem 15. (Other properties of convolution) Let f (t), g(t) , and h(t) be piecewise continuous on [0, ). Then a ) f g = g f; b ) f ( g + h) = ( f g ) + ( f h) ; c ) ( f g ) h = f ( g h) ;

d ) f 0 = 0.

Proof. One way to prove these properties is to use denition (p.425 in the textbook). Here we use the Convolution Theorem. We use F , G, H to denote the Laplace transforms of f , g, h. a) We have b) We have f g = L1{L{ f g }} = L1{F G } = L1{G F } = g f . f ( g + h) = = = = = c) We have ( f g )h = = = = = = L1{L{( f g)h }} L1{L{ f g } H } L1{(F G) H } L1{F (G H )} f L1{G H } f ( gh). L 1{L{ f ( g + h)}} L 1{F (G + H ) L 1{F G + F H } L 1{F G } + L1{F H } f g + f h. (74)

(75)

(76)

Feb. 17, 2012

d) is obvious.

Proof of convolution theorem. Proof. We use denition.

t

L{ f g }(s) = = =

est
0 0 <v<t 0 v 0

f (t v ) g (v ) dv dt

est f (t v ) g (v ) dv dt est f (t v ) g (v ) dt dv
v

=
0

g (v )

est f (t v ) dt dv.

(77)

To evaluate the inner integral, we set x = t v . Then dt = dx, est = esx esv. Thus
v

est f (t v ) dt = esv

esx f (x) dx = esv F (s).

0

(78)

Substituting into the original integral, we have

L{ f g }(s) =

g (v )
0 v

est f (t v ) dt dv = F (s)

g (v ) esv dv = F (s) G(s).

0

(79)

Thus ends the proof.