Unit 6, lesson 49

Laplace Transforms and Initial-Value Problems

How the Laplace transform turns a linear initial-value problem into algebra, for which s it converges, and how steps, impulses and convolution handle switched and sudden inputs.

  • Theory part 97
  • Practice part 98
  • Series and Laplace Methods

A circuit is switched on at t=2t = 2, or a hammer strikes a spring at one instant. Inputs like these jump or act all at once, and the usual guess and check methods struggle with them. The Laplace transform replaces a function of time by a function of a new variable ss, and turns differentiation into multiplication by ss. A linear initial-value problem then becomes an algebra problem, with the initial values already built in.

Key terms

Laplace transform as an improper integral

Definition Let ff be defined for t≥0t \ge 0 and integrable on every interval [0,T][0, T]. Its Laplace transform L{f}\mathcal{L}\{f\} is the function

F(s)=∫0∞e−stf(t) dt=lim⁡T→∞∫0Te−stf(t) dt,\begin{aligned} F(s) &= \int_0^\infty e^{-st} f(t)\, dt \\[0.4em] &= \lim_{T \to \infty} \int_0^T e^{-st} f(t)\, dt, \end{aligned}

defined for those real numbers ss for which this limit exists as a finite number. Only the values of ff for t≥0t \ge 0 matter, which is why it is called the one-sided Laplace transform.

In plain words Multiply f(t)f(t) by the decaying envelope e−ste^{-st} and add up all the area from 00 to ∞\infty. Each value of ss gives one number, and together those numbers form the new function FF.

Example For f(t)=e2tf(t) = e^{2t} and s>2s > 2,

∫0Te−ste2t dt=1−e−(s−2)Ts−2→1s−2 as T→∞,\begin{aligned} \int_0^T e^{-st} e^{2t}\, dt &= \frac{1 - e^{-(s - 2)T}}{s - 2} \\[0.4em] &\to \frac{1}{s - 2} \text{ as } T \to \infty, \end{aligned}

because e−(s−2)T→0e^{-(s - 2)T} \to 0 when s−2>0s - 2 > 0. So L{e2t}=1s−2\mathcal{L}\{e^{2t}\} = \frac{1}{s - 2} for s>2s > 2. The same computation gives L{eat}=1s−a\mathcal{L}\{e^{at}\} = \frac{1}{s - a} for s>as > a, and L{1}=1s\mathcal{L}\{1\} = \frac{1}{s} for s>0s > 0.

Region of convergence

Definition The region of convergence of L{f}\mathcal{L}\{f\} is the set of real ss for which the improper integral converges. A function ff is of exponential order aa if there are constants M>0M > 0 and T0≥0T_0 \ge 0 with

∣f(t)∣≤Meatfor all t≥T0.\lvert f(t) \rvert \le M e^{at} \quad \text{for all } t \ge T_0.

Existence theorem. If ff is piecewise continuous on every interval [0,T][0, T] and of exponential order aa, then the integral converges absolutely for every s>as > a, and F(s)→0F(s) \to 0 as s→∞s \to \infty. Unless it is empty or the whole real line, the region of convergence is a half-line. For absolute convergence this is easy to see, because if the integral of e−s0t∣f(t)∣e^{-s_0 t} \lvert f(t) \rvert is finite, then so is the integral of the smaller function e−st∣f(t)∣e^{-st} \lvert f(t) \rvert for every s>s0s > s_0.

In plain words The envelope e−ste^{-st} has to beat the growth of ff. Past some value of ss it always does for functions that grow no faster than an exponential, and every formula in a table of Laplace transforms comes with that range of ss attached.

Example For f(t)=e2tf(t) = e^{2t} and s≤2s \le 2 the integrand is e(2−s)t≥1e^{(2 - s)t} \ge 1, so the integral diverges. The region of convergence is exactly s>2s > 2. The formula 1s−2\frac{1}{s - 2} still produces numbers for s<2s < 2, for example −1-1 at s=1s = 1, but those numbers are not values of the Laplace transform. At the other extreme, f(t)=et2f(t) = e^{t^2} is of no exponential order and et2−st→∞e^{t^2 - st} \to \infty for every ss, so it has no Laplace transform at all.

Laplace transforms of derivatives

Definition Suppose ff is continuous on [0,∞)[0, \infty) and of exponential order aa, and f′f' is piecewise continuous on every interval [0,T][0, T]. Then for s>as > a

L{f′}(s)=sF(s)−f(0).\mathcal{L}\{f'\}(s) = s F(s) - f(0).

If also f′f' is continuous on [0,∞)[0, \infty) and of exponential order aa, and f′′f'' is piecewise continuous on every [0,T][0, T], then for s>as > a

L{f′′}(s)=s2F(s)−sf(0)−f′(0).\begin{aligned} \mathcal{L}\{f''\}(s) &= s^2 F(s) \\ &\quad - s f(0) - f'(0). \end{aligned}

The proof is integration by parts on [0,T][0, T], where the boundary term e−sTf(T)e^{-sT} f(T) tends to 00 because s>as > a.

In plain words Differentiating in time becomes multiplying by ss, minus a correction that carries the initial value. That is how the initial conditions enter the algebra.

Example Take f(t)=sin⁡3tf(t) = \sin 3t, so F(s)=3s2+9F(s) = \frac{3}{s^2 + 9} for s>0s > 0 and f(0)=0f(0) = 0. The rule predicts

L{3cos⁡3t}=s⋅3s2+9−0=3ss2+9,s>0,\begin{aligned} \mathcal{L}\{3 \cos 3t\} &= s \cdot \frac{3}{s^2 + 9} - 0 \\[0.4em] &= \frac{3s}{s^2 + 9}, \qquad s > 0, \end{aligned}

which agrees with three times the entry L{cos⁡3t}=ss2+9\mathcal{L}\{\cos 3t\} = \frac{s}{s^2 + 9} in the table of Laplace transforms on the reference sheet.

Inverse Laplace transform

Definition Uniqueness theorem. Let ff and gg be piecewise continuous on every [0,T][0, T] and of exponential order, with L{f}(s)=L{g}(s)\mathcal{L}\{f\}(s) = \mathcal{L}\{g\}(s) for all ss greater than some s0s_0. Then f(t)=g(t)f(t) = g(t) at every t≥0t \ge 0 where both are continuous. In particular two continuous functions with the same Laplace transform are equal everywhere on [0,∞)[0, \infty). The inverse Laplace transform L−1{F}\mathcal{L}^{-1}\{F\} is the continuous ff with L{f}=F\mathcal{L}\{f\} = F when there is one. Otherwise it is a piecewise continuous ff, determined except at its points of discontinuity, where the values do not matter. The inverse is linear.

In plain words The function FF determines ff up to values at isolated jumps, so you can read answers backward from a table. In practice you split FF into pieces that appear in the table, usually by partial fractions.

Example Partial fractions give

s+5(s+1)(s+2)=4s+1−3s+2,\frac{s + 5}{(s + 1)(s + 2)} = \frac{4}{s + 1} - \frac{3}{s + 2},

with 4=−1+5−1+24 = \frac{-1 + 5}{-1 + 2} and −3=−2+5−2+1-3 = \frac{-2 + 5}{-2 + 1}. Since L{e−t}=1s+1\mathcal{L}\{e^{-t}\} = \frac{1}{s + 1} and L{e−2t}=1s+2\mathcal{L}\{e^{-2t}\} = \frac{1}{s + 2},

L−1{s+5(s+1)(s+2)}=4e−t−3e−2t.\begin{aligned} &\mathcal{L}^{-1}\left\{\frac{s + 5}{(s + 1)(s + 2)}\right\} \\[0.4em] &\quad = 4e^{-t} - 3e^{-2t}. \end{aligned}

Step functions and the delay rule

Definition For c≥0c \ge 0 the unit step function is

uc(t)={0,0≤t<c,1,t≥c,u_c(t) = \begin{cases} 0, & 0 \le t < c, \\ 1, & t \ge c, \end{cases}

with L{uc}=e−css\mathcal{L}\{u_c\} = \frac{e^{-cs}}{s} for s>0s > 0. Delay rule. If L{f}=F(s)\mathcal{L}\{f\} = F(s) exists for s>as > a and c≥0c \ge 0, then for s>as > a

L{uc(t)f(t−c)}=e−csF(s).\mathcal{L}\{u_c(t) f(t - c)\} = e^{-cs} F(s).

The graph of uc(t)f(t−c)u_c(t) f(t - c) is the graph of ff moved cc units to the right, with the value 00 before t=ct = c.

In plain words A step switches something on at time cc. Delaying a signal by cc multiplies F(s)F(s) by e−cse^{-cs}. The shift must be applied to the whole function, so you rewrite f(t)f(t) in terms of t−ct - c before using the rule.

Example A ramp input rises like tt until t=2t = 2 and then holds at 22. As a single formula,

g(t)=t−u2(t)(t−2),t≥0.g(t) = t - u_2(t)(t - 2), \qquad t \ge 0.

The second term is f(t)=tf(t) = t delayed by 22, so

G(s)=1s2−e−2ss2,s>0.G(s) = \frac{1}{s^2} - \frac{e^{-2s}}{s^2}, \qquad s > 0.

Impulse as a generalized input

Definition For c>0c > 0 and ε>0\varepsilon > 0, the pulse dε(t)=1ε(uc(t)−uc+ε(t))d_\varepsilon(t) = \frac{1}{\varepsilon}\bigl(u_c(t) - u_{c + \varepsilon}(t)\bigr) has height 1ε\frac{1}{\varepsilon} on [c,c+ε)[c, c + \varepsilon) and total area 11. For each s>0s > 0 its Laplace transform satisfies

L{dε}=e−cs 1−e−εsεs→e−cs as ε→0+.\begin{aligned} \mathcal{L}\{d_\varepsilon\} &= e^{-cs} \, \frac{1 - e^{-\varepsilon s}}{\varepsilon s} \\[0.4em] &\to e^{-cs} \text{ as } \varepsilon \to 0^+. \end{aligned}

The unit impulse at cc, written δ(t−c)\delta(t - c), is the generalized input whose integral against each gg continuous on [0,∞)[0, \infty) is the limit of the pulse integrals,

∫0∞δ(t−c) g(t) dt=lim⁡ε→0+∫0∞dε(t) g(t) dt=g(c).\begin{aligned} &\int_0^\infty \delta(t - c)\, g(t)\, dt \\[0.4em] &\quad = \lim_{\varepsilon \to 0^+} \int_0^\infty d_\varepsilon(t)\, g(t)\, dt \\[0.4em] &\quad = g(c). \end{aligned}

The choice g(t)=e−stg(t) = e^{-st} gives L{δ(t−c)}=e−cs\mathcal{L}\{\delta(t - c)\} = e^{-cs}, the limit above. The impulse is not an ordinary function. The condition c>0c > 0 keeps the kick inside the interval of integration, and an impulse at t=0t = 0 needs an extra convention about the lower limit. No piecewise continuous function ff of exponential order has Laplace transform e−cse^{-cs}. Otherwise its running integral h(t)=∫0tf(τ) dτh(t) = \int_0^t f(\tau)\, d\tau would have Laplace transform e−css\frac{e^{-cs}}{s} by the derivative rule, so uniqueness would give h=uch = u_c away from t=ct = c, which is impossible because hh is continuous and ucu_c jumps at cc.

In plain words An impulse is a kick of total size 11 delivered in no time at all. It is the idealization of a very short, very strong push, and in the ss variable it is as easy to handle as a step.

Example With c=1c = 1 and s=1s = 1 the values approach e−1≈0.3679e^{-1} \approx 0.3679 as the pulse narrows.

ε\varepsilonL{dε}(1)\mathcal{L}\{d_\varepsilon\}(1)
110.23250.2325
0.10.10.35010.3501
0.010.010.36600.3660

Convolution

Definition For ff and gg piecewise continuous on every [0,T][0, T], their convolution is

(f∗g)(t)=∫0tf(τ) g(t−τ) dτ(f * g)(t) = \int_0^t f(\tau)\, g(t - \tau)\, d\tau

for t≥0t \ge 0. Convolution theorem. If ff and gg are piecewise continuous on every [0,T][0, T] and of exponential order aa, then f∗gf * g is of exponential order bb for every b>ab > a, and for s>as > a

L{f∗g}(s)=F(s) G(s).\mathcal{L}\{f * g\}(s) = F(s)\, G(s).

In plain words A product of Laplace transforms is the Laplace transform of a convolution, not of the product f(t)g(t)f(t) g(t). Convolution adds up the input received at each earlier time τ\tau weighted by how the system responds after the remaining time t−τt - \tau.

Example Take f(t)=1f(t) = 1 and g(t)=sin⁡tg(t) = \sin t. Then

(1∗sin⁡)(t)=∫0tsin⁡(t−τ) dτ=1−cos⁡t.\begin{aligned} (1 * \sin)(t) &= \int_0^t \sin(t - \tau)\, d\tau \\[0.4em] &= 1 - \cos t. \end{aligned}

The convolution theorem agrees. For s>0s > 0

1s⋅1s2+1=1s−ss2+1=L{1−cos⁡t}.\begin{aligned} \frac{1}{s} \cdot \frac{1}{s^2 + 1} &= \frac{1}{s} - \frac{s}{s^2 + 1} \\[0.4em] &= \mathcal{L}\{1 - \cos t\}. \end{aligned}

Solving linear initial-value problems

Definition Consider ay′′+by′+cy=g(t)a y'' + b y' + c y = g(t) with constants a≠0a \ne 0, bb, cc, initial values y(0)=y0y(0) = y_0 and y′(0)=y1y'(0) = y_1, and gg piecewise continuous on every [0,T][0, T] and of exponential order. Then the solution yy and its derivative y′y' are continuous and of exponential order, so the rules above apply. Taking Laplace transforms gives

(as2+bs+c) Y(s)=G(s)+a(sy0+y1)+by0,\begin{gathered} (a s^2 + b s + c)\, Y(s) = G(s) \\ {} + a(s y_0 + y_1) + b y_0, \end{gathered}

valid for all ss large enough. Solve for YY, apply the inverse Laplace transform, and the result is the solution for t≥0t \ge 0. When gg has jumps, y′′y'' jumps there too and the equation holds except at those points. With an impulse input k δ(t−t0)k\,\delta(t - t_0), where t0>0t_0 > 0, the solution stays continuous and y′y' jumps by ka\frac{k}{a} at t=t0t = t_0.

In plain words Apply L\mathcal{L} to the equation, solve a linear equation for YY, and go back with the inverse. The initial values are used in the first step, so no constants are left to fit at the end.

Example Solve y′′+3y′+2y=2y'' + 3y' + 2y = 2 with y(0)=0y(0) = 0 and y′(0)=0y'(0) = 0. Applying L\mathcal{L} to both sides gives (s2+3s+2)Y=2s(s^2 + 3s + 2) Y = \frac{2}{s} for s>0s > 0, so

Y(s)=2s(s+1)(s+2)=1s−2s+1+1s+2.\begin{aligned} Y(s) &= \frac{2}{s(s + 1)(s + 2)} \\[0.4em] &= \frac{1}{s} - \frac{2}{s + 1} + \frac{1}{s + 2}. \end{aligned}

Inverting term by term,

y(t)=1−2e−t+e−2t,t≥0.y(t) = 1 - 2e^{-t} + e^{-2t}, \qquad t \ge 0.

To check, y(0)=1−2+1=0y(0) = 1 - 2 + 1 = 0 and y′(0)=2−2=0y'(0) = 2 - 2 = 0. Substituting into the equation, the e−te^{-t} terms give −2+6−4=0-2 + 6 - 4 = 0, the e−2te^{-2t} terms give 4−6+2=04 - 6 + 2 = 0, and the constant gives 22.

Check yourself

  1. Solve y′+y=u2(t)y' + y = u_2(t) with y(0)=0y(0) = 0. This is a switched input that turns on at t=2t = 2.
Show the answer

Applying L\mathcal{L} gives (s+1)Y=e−2ss(s + 1) Y = \frac{e^{-2s}}{s} for s>0s > 0, so

Y(s)=e−2s(1s−1s+1).Y(s) = e^{-2s}\left(\frac{1}{s} - \frac{1}{s + 1}\right).

By the delay rule

y(t)=u2(t)(1−e−(t−2)).y(t) = u_2(t)\bigl(1 - e^{-(t - 2)}\bigr).

The solution is 00 until t=2t = 2, then rises toward 11. It is continuous, while y′y' jumps from 00 to 11 at t=2t = 2.

  1. Solve y′′+y=δ(t−π)y'' + y = \delta(t - \pi) with y(0)=0y(0) = 0 and y′(0)=1y'(0) = 1, and describe what the impulse does.
Show the answer

Applying L\mathcal{L} gives (s2+1)Y−1=e−πs(s^2 + 1) Y - 1 = e^{-\pi s} for s>0s > 0, so

Y(s)=1s2+1+e−πss2+1.Y(s) = \frac{1}{s^2 + 1} + \frac{e^{-\pi s}}{s^2 + 1}.

By the delay rule y(t)=sin⁡t+uπ(t)sin⁡(t−π)=sin⁡t−uπ(t)sin⁡ty(t) = \sin t + u_\pi(t) \sin(t - \pi) = \sin t - u_\pi(t) \sin t. So y=sin⁡ty = \sin t for 0≤t<π0 \le t < \pi and y=0y = 0 for t≥πt \ge \pi. Just before t=πt = \pi the velocity is cos⁡π=−1\cos \pi = -1, and the unit impulse raises it by 11 to 00, which stops the oscillation exactly at rest.

  1. Find the Laplace transform of h(t)=u1(t)(t−1)h(t) = u_1(t)(t - 1) and state where it converges.
Show the answer

This is the ramp f(t)=tf(t) = t delayed by 11, and L{t}=1s2\mathcal{L}\{t\} = \frac{1}{s^2} for s>0s > 0. The delay rule gives e−ss2\frac{e^{-s}}{s^2}, valid for s>0s > 0.

How this lesson connects

It also uses Improper integrals, Linear differential equations and initial values and Integration by parts.