Unit 6, lesson 33

Laplace Transforms of Derivatives and Initial-Value Problems

How integration by parts gives the Laplace transform of a derivative, how the initial-value terms enter an algebraic equation for the Laplace transform, and how this solves linear initial-value problems.

  • Video 33
  • Initial-Value Problems and Switched Inputs

An initial-value problem picks out one solution from a whole family by its starting values. The Laplace transform replaces each derivative by multiplication by ss, and the starting values have to survive that step, or the algebra would describe the wrong member of the family. This lesson shows how integration by parts keeps them, as known terms beside the unknown, and then solves a second-order problem from start to finish. Throughout, y=y(t)y = y(t) is a function of time t≥0t \ge 0, primes mean derivatives with respect to tt, Y=L{y}Y = \mathcal{L}\{y\}, and ss is a complex number.

Key terms

Integration by parts

Definition Let yy be continuous on [0,T][0, T] and differentiable except at finitely many points, with y′y' piecewise continuous. For every complex ss,

∫0Te−sty′(t) dt=e−sTy(T)−y(0)+s∫0Te−sty(t) dt.\begin{aligned} &\int_0^T e^{-st} y'(t)\,dt \\[0.4em] &\quad = e^{-sT}y(T) - y(0) \\[0.4em] &\qquad + s\int_0^T e^{-st} y(t)\,dt. \end{aligned}

This is integration by parts with u=e−stu = e^{-st} and dv=y′(t) dtdv = y'(t)\,dt, carried out on each piece between the points where y′y' fails to exist. The boundary terms at those interior points cancel because yy is continuous. If instead yy has a jump at a point t0t_0 in (0,T)(0, T), the terms there do not cancel, and the right side gains the extra term −e−st0(y(t0+)−y(t0−))-e^{-st_0}\bigl(y(t_0^+) - y(t_0^-)\bigr).

In plain words Moving the derivative from yy onto the weight e−ste^{-st} costs a boundary term. Differentiating the weight brings out the factor −s-s, and the boundary term keeps the values of yy at the two ends of the interval.

Example The switched signal with y=1y = 1 for 0≤t<10 \le t < 1 and y=0y = 0 for t≥1t \ge 1 has y′=0y' = 0 wherever y′y' exists, so the left side is 00. For T>1T > 1 the formula without a jump term would give

0−1+s∫01e−st dt=−1+(1−e−s)=−e−s,\begin{aligned} &0 - 1 + s\int_0^1 e^{-st}\,dt \\[0.4em] &\quad = -1 + (1 - e^{-s}) \\[0.4em] &\quad = -e^{-s}, \end{aligned}

which is wrong. The jump at t0=1t_0 = 1 has size −1-1, and the extra term +e−s+e^{-s} restores the correct value 00. Continuity of yy is a real hypothesis.

Laplace transform of a derivative

Definition Let yy be continuous on [0,∞)[0, \infty) and of exponential order γ\gamma, so ∣y(t)∣≤Meγt\lvert y(t) \rvert \le Me^{\gamma t} for t≥T0t \ge T_0, and let y′y' be piecewise continuous on [0,∞)[0, \infty). Then L{y′}(s)\mathcal{L}\{y'\}(s) exists for Re⁡s>γ\operatorname{Re} s > \gamma and

L{y′}(s)=sY(s)−y(0).\mathcal{L}\{y'\}(s) = sY(s) - y(0).

If also y′y' is continuous and of exponential order γ\gamma, and y′′y'' is piecewise continuous, then for Re⁡s>γ\operatorname{Re} s > \gamma

L{y′′}(s)=s2Y(s)−sy(0)−y′(0).\begin{aligned} \mathcal{L}\{y''\}(s) &= s^2Y(s) \\[0.4em] &\quad - sy(0) - y'(0). \end{aligned}

The first rule is the integration by parts above as T→∞T \to \infty, because ∣e−sTy(T)∣≤Me−(Re⁡s−γ)T\lvert e^{-sT}y(T) \rvert \le Me^{-(\operatorname{Re} s - \gamma)T} for T≥T0T \ge T_0, and this bound tends to 00. The second applies the first rule to y′y'. In the same way, if y,y′,…,y(n−1)y, y', \ldots, y^{(n-1)} are continuous and of exponential order γ\gamma and y(n)y^{(n)} is piecewise continuous, then L{y(n)}=snY−sn−1y(0)−⋯−y(n−1)(0)\mathcal{L}\{y^{(n)}\} = s^nY - s^{n-1}y(0) - \cdots - y^{(n-1)}(0) for Re⁡s>γ\operatorname{Re} s > \gamma.

In plain words Differentiating yy multiplies its Laplace transform by ss and subtracts the starting value. Each further derivative brings one more starting value with it.

Example Take y=cos⁡2ty = \cos 2t, with Y(s)=s/(s2+4)Y(s) = s/(s^2 + 4) for Re⁡s>0\operatorname{Re} s > 0, y(0)=1y(0) = 1 and y′(0)=0y'(0) = 0. The rules predict

L{y′}=s2s2+4−1=−4s2+4\begin{aligned} \mathcal{L}\{y'\} &= \frac{s^2}{s^2 + 4} - 1 \\[0.4em] &= -\frac{4}{s^2 + 4} \end{aligned}

and

L{y′′}=s3s2+4−s=−4ss2+4.\begin{aligned} \mathcal{L}\{y''\} &= \frac{s^3}{s^2 + 4} - s \\[0.4em] &= -\frac{4s}{s^2 + 4}. \end{aligned}

Both agree with the direct calculation, since y′=−2sin⁡2ty' = -2\sin 2t and y′′=−4cos⁡2ty'' = -4\cos 2t, whose Laplace transforms are −2⋅2/(s2+4)-2 \cdot 2/(s^2 + 4) and −4s/(s2+4)-4s/(s^2 + 4).

Initial-value terms

Definition The initial-value terms are the known quantities built from y(0),y′(0),…y(0), y'(0), \ldots that the derivative rules attach to YY. For the expression y′′+py′+qyy'' + py' + qy with constants pp and qq, and yy meeting the hypotheses of the second-derivative rule,

L{y′′+py′+qy}=(s2+ps+q) Y(s)−(sy(0)+y′(0)+py(0)).\begin{aligned} &\mathcal{L}\{y'' + py' + qy\} \\[0.4em] &\quad = (s^2 + ps + q)\,Y(s) \\[0.4em] &\qquad - \bigl(sy(0) + y'(0) + py(0)\bigr). \end{aligned}

The initial data enter the calculation only through these terms. Dropping them gives the Laplace transform of the solution with y(0)=0y(0) = 0 and y′(0)=0y'(0) = 0, which is in general a different function.

In plain words The derivative rules leave behind a few starting values. They are known numbers, so they move to the right side of the equation, where they act like an extra input supplied by the initial state.

Example Consider y′′+3y′+2y=4y'' + 3y' + 2y = 4 with y(0)=1y(0) = 1 and y′(0)=0y'(0) = 0. On a half-plane where all of these Laplace transforms exist, the terms become

L{y′′}=s2Y−s,L{3y′}=3sY−3,L{2y}=2Y,L{4}=4/s.\begin{aligned} \mathcal{L}\{y''\} &= s^2Y - s, \\[0.4em] \mathcal{L}\{3y'\} &= 3sY - 3, \\[0.4em] \mathcal{L}\{2y\} &= 2Y, \\[0.4em] \mathcal{L}\{4\} &= 4/s. \end{aligned}

The initial-value terms are −s-s and −3-3. Writing each term out before collecting anything keeps every sign easy to check.

Algebraic equation for the Laplace transform

Definition Consider y′′+py′+qy=g(t)y'' + py' + qy = g(t) with constants pp and qq, initial values y(0)=y0y(0) = y_0 and y′(0)=y1y'(0) = y_1, and a piecewise continuous forcing gg of exponential order with Laplace transform GG. Applying L\mathcal{L} to both sides gives, on a half-plane where all the Laplace transforms exist,

P(s) Y(s)=G(s)+I(s),P(s)\,Y(s) = G(s) + I(s),

where

P(s)=s2+ps+q,I(s)=sy0+y1+py0.\begin{gathered} P(s) = s^2 + ps + q, \\[0.4em] I(s) = sy_0 + y_1 + py_0. \end{gathered}

The polynomial PP is the characteristic polynomial of the equation, written in the variable ss, and II collects the initial-value terms. For Re⁡s\operatorname{Re} s large enough P(s)≠0P(s) \ne 0, and

Y(s)=G(s)P(s)+I(s)P(s).Y(s) = \frac{G(s)}{P(s)} + \frac{I(s)}{P(s)}.

The first part is the Laplace transform of the response that starts from rest, and the second is that of the free motion from the initial state with no forcing.

In plain words After the Laplace transform, the differential equation becomes one linear equation for the unknown function YY. Solving it means dividing by the characteristic polynomial, and the answer splits into a part caused by the input and a part caused by the starting state.

Example For the problem above, P(s)=s2+3s+2=(s+1)(s+2)P(s) = s^2 + 3s + 2 = (s + 1)(s + 2) and I(s)=s+3I(s) = s + 3. Collecting terms gives P(s) Y=4/s+s+3=(s2+3s+4)/sP(s)\,Y = 4/s + s + 3 = (s^2 + 3s + 4)/s, so

Y(s)=s2+3s+4s(s+1)(s+2).Y(s) = \frac{s^2 + 3s + 4}{s(s + 1)(s + 2)}.

The two parts are

GP=4s(s+1)(s+2),IP=s+3(s+1)(s+2),\begin{aligned} \frac{G}{P} &= \frac{4}{s(s + 1)(s + 2)}, \\[0.4em] \frac{I}{P} &= \frac{s + 3}{(s + 1)(s + 2)}, \end{aligned}

which invert to 2−4e−t+2e−2t2 - 4e^{-t} + 2e^{-2t}, the solution that starts from rest, and 2e−t−e−2t2e^{-t} - e^{-2t}, the free motion with y(0)=1y(0) = 1 and y′(0)=0y'(0) = 0.

Initial-value problems by Laplace transform

Definition Consider y′′+py′+qy=g(t)y'' + py' + qy = g(t) for t≥0t \ge 0, with constants pp and qq, data y(0)=y0y(0) = y_0 and y′(0)=y1y'(0) = y_1, and gg piecewise continuous and of exponential order. Then the solution yy and its derivative y′y' are continuous and of exponential order, so the derivative rules apply. First take the Laplace transform of every term and keep the initial-value terms. Then solve the algebraic equation for YY, split YY by partial fractions and invert each piece. Finally check the result in the differential equation and in both initial conditions. The answer is the solution for t≥0t \ge 0, and when gg has jumps, y′′y'' jumps there too and the equation holds except at those points. Data given at a time t0≠0t_0 \ne 0 need the shift τ=t−t0\tau = t - t_0 first. The method relies on constant coefficients and linearity. A coefficient that depends on tt brings in derivatives with respect to ss, as in L{ty′}=−Y−s dY/ds\mathcal{L}\{ty'\} = -Y - s\,dY/ds, and a nonlinear term has no rule of this kind, since L{y2}\mathcal{L}\{y^2\} is not Y2Y^2 in general. For y=1y = 1, for example, L{y2}=1/s\mathcal{L}\{y^2\} = 1/s while Y2=1/s2Y^2 = 1/s^2.

In plain words Take the Laplace transform, solve for YY, split it into table entries and invert. The initial values are used at the first step, so no constants are left to fit at the end, and a final check catches algebra slips.

Example For the problem above, write Y(s)=A/s+B/(s+1)+C/(s+2)Y(s) = A/s + B/(s + 1) + C/(s + 2). Multiplying by each factor and evaluating at its zero gives A=4/2=2A = 4/2 = 2, B=2/(−1)=−2B = 2/(-1) = -2 and C=2/2=1C = 2/2 = 1, so for t≥0t \ge 0

y(t)=2−2e−t+e−2t.y(t) = 2 - 2e^{-t} + e^{-2t}.

To check it, y(0)=2−2+1=1y(0) = 2 - 2 + 1 = 1. The derivatives are y′=2e−t−2e−2ty' = 2e^{-t} - 2e^{-2t}, so y′(0)=0y'(0) = 0, and y′′=−2e−t+4e−2ty'' = -2e^{-t} + 4e^{-2t}. In y′′+3y′+2yy'' + 3y' + 2y the coefficients of e−te^{-t} add to −2+6−4=0-2 + 6 - 4 = 0, those of e−2te^{-2t} add to 4−6+2=04 - 6 + 2 = 0, and the constant is 2⋅2=42 \cdot 2 = 4. The poles 00, −1-1 and −2-2 show that YY is the Laplace transform of yy for Re⁡s>0\operatorname{Re} s > 0. As t→∞t \to \infty both exponentials decay and y→2y \to 2, the level at which 2y2y balances the constant input 44. Since y′=2e−t(1−e−t)≥0y' = 2e^{-t}(1 - e^{-t}) \ge 0 for t≥0t \ge 0, the solution rises from 11 toward 22 without overshooting.

Check yourself

  1. Solve y′+2y=6y' + 2y = 6 with y(0)=1y(0) = 1 by the Laplace transform.
Show the answer

The derivative rule gives sY−1+2Y=6/ssY - 1 + 2Y = 6/s, so (s+2)Y=1+6/s=(s+6)/s(s + 2)Y = 1 + 6/s = (s + 6)/s and

Y(s)=s+6s(s+2)=3s−2s+2.\begin{aligned} Y(s) &= \frac{s + 6}{s(s + 2)} \\[0.4em] &= \frac{3}{s} - \frac{2}{s + 2}. \end{aligned}

The coefficients come from 6/2=36/2 = 3 at s=0s = 0 and 4/(−2)=−24/(-2) = -2 at s=−2s = -2. So y=3−2e−2ty = 3 - 2e^{-2t} for t≥0t \ge 0. As a check, y(0)=3−2=1y(0) = 3 - 2 = 1, and with y′=4e−2ty' = 4e^{-2t} the sum y′+2y=4e−2t+6−4e−2t=6y' + 2y = 4e^{-2t} + 6 - 4e^{-2t} = 6.

  1. For y′′−y=0y'' - y = 0 with y(0)=1y(0) = 1 and y′(0)=1y'(0) = 1, a classmate uses L{y′′}=s2Y−y(0)−y′(0)\mathcal{L}\{y''\} = s^2Y - y(0) - y'(0) and gets y=et−e−ty = e^{t} - e^{-t}. Find the error and solve the problem correctly.
Show the answer

The correct rule is L{y′′}=s2Y−sy(0)−y′(0)\mathcal{L}\{y''\} = s^2Y - sy(0) - y'(0), so the factor ss on y(0)y(0) was lost. The wrong rule gives (s2−1)Y=2(s^2 - 1)Y = 2, hence Y=1/(s−1)−1/(s+1)Y = 1/(s - 1) - 1/(s + 1) and the function et−e−te^{t} - e^{-t}, which is 00 at t=0t = 0 instead of 11. Checking the initial condition exposes the mistake. With the correct rule, s2Y−s−1−Y=0s^2Y - s - 1 - Y = 0, so

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

and y=ety = e^{t}. It satisfies y′′−y=0y'' - y = 0, y(0)=1y(0) = 1 and y′(0)=1y'(0) = 1.

  1. Solve y′′+4y=8y'' + 4y = 8 with y(0)=0y(0) = 0 and y′(0)=2y'(0) = 2.
Show the answer

The terms give s2Y−2+4Y=8/ss^2Y - 2 + 4Y = 8/s, so

Y(s)=8s(s2+4)+2s2+4.\begin{aligned} Y(s) &= \frac{8}{s(s^2 + 4)} \\[0.4em] &\quad + \frac{2}{s^2 + 4}. \end{aligned}

For the first fraction write 8/(s(s2+4))=A/s+(Bs+C)/(s2+4)8/(s(s^2 + 4)) = A/s + (Bs + C)/(s^2 + 4). At s=0s = 0 the coefficient is A=8/4=2A = 8/4 = 2, and then 8=2(s2+4)+(Bs+C)s8 = 2(s^2 + 4) + (Bs + C)s forces B=−2B = -2 and C=0C = 0. Hence

Y(s)=2s−2ss2+4+2s2+4,\begin{aligned} Y(s) &= \frac{2}{s} - \frac{2s}{s^2 + 4} \\[0.4em] &\quad + \frac{2}{s^2 + 4}, \end{aligned}

and y=2−2cos⁡2t+sin⁡2ty = 2 - 2\cos 2t + \sin 2t for t≥0t \ge 0, and YY is its Laplace transform for Re⁡s>0\operatorname{Re} s > 0. As a check, y(0)=2−2=0y(0) = 2 - 2 = 0, y′=4sin⁡2t+2cos⁡2ty' = 4\sin 2t + 2\cos 2t gives y′(0)=2y'(0) = 2, and y′′=8cos⁡2t−4sin⁡2ty'' = 8\cos 2t - 4\sin 2t, so y′′+4y=8y'' + 4y = 8.

How this lesson connects