6.2 The Laplace Transform Method
The Laplace transform turns a linear initial value problem into algebra: existence and the basic table, both shifting theorems, derivatives, integrals and multiplication by $t$, unit steps, impulses and periodic inputs, partial fractions, convolution, systems and transfer functions — with proofs.
A spring–mass system is struck by a hammer at $t=2$; a circuit is switched on at $t=0$ and off at $t=1$; a drug is injected every eight hours. The equations are linear with constant coefficients, but the forcing terms are switched on and off, or act in an instant. Variation of parameters (§6.1) can handle such problems one interval at a time, with careful matching of the solution at every switch. The Laplace transform does all of it in one stroke. It replaces a function $f(t)$ on $[0,\infty)$ by a function $F(s)$, converts differentiation into multiplication by $s$ (with the initial values appearing automatically), converts a time delay into a factor $e^{-as}$, and converts the convolution integral that expresses a system's response into an ordinary product. Solving the IVP becomes: transform, solve a linear algebraic equation for $Y(s)$, split it into partial fractions, and read the answer off a table.
GATE MA asks for inverse transforms of rational functions (with and without $e^{-as}$ factors), for the solution of an IVP at a particular time, for the values of definite integrals that are Laplace transforms in disguise, and — in MSQs — for exact knowledge of which functions have transforms, what the shifting theorems and the convolution theorem do and do not say, and when the initial and final value theorems may be used. The same transform reappears in §10.3 for partial differential equations on a half-line. This section proves every rule it uses.
The Transform and Its Existence
Proof. On $[0,T_0]$, $f$ is piecewise continuous, hence bounded, say $|f|\le B$. For $s\gt\max(\alpha,0)$,
$$\int_0^\infty e^{-st}|f(t)|\,dt\le B\int_0^{T_0}e^{-st}\,dt+K\int_{T_0}^\infty e^{-(s-\alpha)t}\,dt .$$The right side is at most $\dfrac Bs+\dfrac{K}{s-\alpha}$; so the integral converges absolutely (comparison test) and $|F(s)|\le\frac Bs+\frac K{s-\alpha}\to0$. If $\alpha\lt0$ the same estimate for $s\in(\alpha,0]$ uses $\int_0^{T_0}e^{-st}\le T_0e^{|s|T_0}$. ∎
The conditions are sufficient, not necessary: $t^{-1/2}$ is not piecewise continuous at $0$ (it is unbounded), yet it has a transform (Example 1). The smallest $\sigma$ such that the integral converges for all $s\gt\sigma$ is the abscissa of convergence; Interactive 1 shows the integral settling for $s$ to the right of it and running away to its left.
Proof. (a) is linearity of the integral. (b) (sketch) Put $h=f-g$, so $H(s)=0$ for $s\gt s_0$, and fix $s_1\gt\max(s_0,\alpha,0)$. For $n=0,1,2,\dots$, $H(s_1+n+1)=0$. Substituting $x=e^{-t}$ turns this into $\int_0^1x^n\varphi(x)\,dx=0$ with $\varphi(x)=x^{s_1}h(-\log x)$, a bounded function on $(0,1]$ that is continuous except at finitely many points in each $[\delta,1]$ and tends to $0$ as $x\to0^+$. So $\int_0^1P\varphi=0$ for every polynomial $P$. By the Weierstrass approximation theorem (§4.2), applied after approximating $\varphi$ in the mean by continuous functions, $\int_0^1\varphi^2=0$; hence $\varphi=0$, i.e. $h=0$, at every point of continuity. ∎
So we may read the table below in both directions. Changing $f$ at finitely many points does not change $F$ — this is why the value of a step function exactly at the jump never matters.
| $f(t)$ | $F(s)$ | for |
|---|---|---|
| $1$ | $\dfrac1s$ | $s\gt0$ |
| $t^n$ ($n\in\mathbb{N}$) | $\dfrac{n!}{s^{n+1}}$ | $s\gt0$ |
| $t^\nu$ ($\nu\gt-1$) | $\dfrac{\Gamma(\nu+1)}{s^{\nu+1}}$ | $s\gt0$ |
| $e^{at}$ | $\dfrac1{s-a}$ | $s\gt a$ |
| $\cos bt$, $\sin bt$ | $\dfrac{s}{s^2+b^2}$, $\dfrac{b}{s^2+b^2}$ | $s\gt0$ |
| $\cosh bt$, $\sinh bt$ | $\dfrac{s}{s^2-b^2}$, $\dfrac{b}{s^2-b^2}$ | $s\gt|b|$ |
Proof. $\int_0^Te^{-(s-a)t}dt=\dfrac{1-e^{-(s-a)T}}{s-a}\to\dfrac1{s-a}$ for $s\gt a$; $a=0$ gives $\mathcal{L}\{1\}$. The same computation holds for complex $a$ when $s\gt\operatorname{Re}a$, since $|e^{-(s-a)T}|=e^{-(s-\operatorname{Re}a)T}$. With $a=ib$, $\dfrac1{s-ib}=\dfrac{s+ib}{s^2+b^2}$, and comparing real and imaginary parts of $\mathcal{L}\{\cos bt+i\sin bt\}$ gives the cosine and sine. $\cosh$ and $\sinh$ are half-sums and half-differences of $e^{\pm bt}$. For $t^\nu$, substitute $u=st$: $\int_0^\infty e^{-st}t^\nu dt$ becomes $s^{-\nu-1}\int_0^\infty e^{-u}u^\nu du=\Gamma(\nu+1)s^{-\nu-1}$, the integral converging at $0$ because $\nu\gt-1$; and $\Gamma(n+1)=n!$. ∎
Shifting, Derivatives and Integrals
Proof. $\int_0^\infty e^{-st}e^{at}f(t)\,dt$ is $\int_0^\infty e^{-(s-a)t}f(t)\,dt=F(s-a)$, which converges when $s-a\gt\alpha$. ∎ So $\mathcal{L}\{e^{at}t^n\}=\dfrac{n!}{(s-a)^{n+1}}$, $\mathcal{L}\{e^{at}\cos bt\}=\dfrac{s-a}{(s-a)^2+b^2}$ and $\mathcal{L}\{e^{at}\sin bt\}=\dfrac{b}{(s-a)^2+b^2}$.
Let $f$ be continuous on $[0,\infty)$ and of exponential order $\alpha$, and let $f'$ be piecewise continuous on $[0,\infty)$. Then $\mathcal{L}\{f'\}$ exists for $s\gt\alpha$ and
$$\mathcal{L}\{f'\}(s)=sF(s)-f(0).$$More generally, if $f,f',\dots,f^{(n-1)}$ are continuous and of exponential order $\alpha$ and $f^{(n)}$ is piecewise continuous, then $\mathcal{L}\{f^{(n)}\}=s^nF-s^{n-1}f(0)-s^{n-2}f'(0)-\dots-f^{(n-1)}(0)$; in particular $\mathcal{L}\{f''\}=s^2F-sf(0)-f'(0)$.
Proof. Fix $T$ and let $0=t_0\lt t_1\lt\dots\lt t_k=T$ include the points where $f'$ is discontinuous. On each $[t_{i-1},t_i]$ integrate by parts:
$$\int_{t_{i-1}}^{t_i}e^{-st}f'(t)\,dt=\Big[e^{-st}f(t)\Big]_{t_{i-1}}^{t_i}+s\int_{t_{i-1}}^{t_i}e^{-st}f(t)\,dt .$$Summing over $i$, the boundary terms telescope because $f$ is continuous, leaving $e^{-sT}f(T)-f(0)+s\int_0^Te^{-st}f\,dt$. As $T\to\infty$, $|e^{-sT}f(T)|\le Ke^{-(s-\alpha)T}\to0$ and the last integral tends to $F(s)$. So $\int_0^Te^{-st}f'$ converges to $sF(s)-f(0)$. The higher formula follows by induction, applying the case $n=1$ to $f^{(n-1)}$. ∎
If $f$ has a jump at $t=a\gt0$, the telescoping leaves the extra term $-e^{-as}\left(f(a^+)-f(a^-)\right)$. In IVPs this is exactly how an impulse enters: it makes $y'$ jump.
Proof. $g(t)=\int_0^tf$ is continuous, $g(0)=0$, and $g'=f$ except at the finitely many jumps of $f$ in each $[0,T]$ (which do not affect the integration by parts above). It is of exponential order $\max(\alpha,0)+\varepsilon$ for every $\varepsilon\gt0$: $|g(t)|\le C+\int_{T_0}^tKe^{\alpha\tau}d\tau$. Theorem 6.19 gives $F=\mathcal{L}\{g'\}=sG-g(0)=sG$. ∎
Let $f$ be piecewise continuous of exponential order $\alpha$.
- $F$ is infinitely differentiable on $(\alpha,\infty)$ and $\mathcal{L}\{t^nf(t)\}=(-1)^nF^{(n)}(s)$.
- If moreover $\lim_{t\to0^+}f(t)/t$ exists, then $\mathcal{L}\left\{\dfrac{f(t)}{t}\right\}=\displaystyle\int_s^\infty F(\sigma)\,d\sigma$ for $s\gt\alpha$.
Proof. (1) Fix $s_1\gt\alpha$ and $\alpha'\in(\alpha,s_1)$. For $s\ge s_1$, $\left|\frac{\partial}{\partial s}e^{-st}f(t)\right|=te^{-st}|f(t)|\le Ce^{-(s_1-\alpha')t}$ for $t\ge T_0$ (because $te^{-(\alpha'-\alpha)t}$ is bounded), an integrable bound independent of $s$. Hence we may differentiate under the integral sign (dominated convergence, §4.3): $F'(s)=-\int_0^\infty te^{-st}f(t)\,dt$. Induction gives the general case. (2) $g=f/t$ is piecewise continuous (it has a limit at $0$) and of exponential order $\alpha$. By (1), $G'=-\mathcal{L}\{tg\}=-F$, and $G(s)\to0$ as $s\to\infty$ by Theorem 6.15. So $G(s)=-\int_\infty^sF=\int_s^\infty F(\sigma)\,d\sigma$. ∎
- Initial value. If $f$ is continuous on $[0,\infty)$, and $f$ and $f'$ satisfy the hypotheses of Theorem 6.15 ($f'$ piecewise continuous, both of exponential order), then $\displaystyle\lim_{s\to\infty}sF(s)=f(0)$.
- Final value. If $f$ is piecewise continuous on $[0,\infty)$ and $L=\lim_{t\to\infty}f(t)$ exists, then $F(s)$ exists for $s\gt0$ and $\displaystyle\lim_{s\to0^+}sF(s)=L$.
Proof. (1) By Theorems 6.19 and 6.15, $sF(s)-f(0)=\mathcal{L}\{f'\}(s)\to0$. (2) $f$ is bounded, say by $B$, hence of exponential order $0$. Since $s\int_0^\infty e^{-st}dt=1$,
$$sF(s)-L=s\int_0^\infty e^{-st}\left(f(t)-L\right)dt .$$Given $\varepsilon\gt0$ choose $T$ with $|f(t)-L|\lt\varepsilon$ for $t\gt T$. The part of the integral over $[0,T]$ is at most $s\cdot T\cdot2B$, and the part over $[T,\infty)$ at most $\varepsilon\, s\int_T^\infty e^{-st}\le\varepsilon$. Letting $s\to0^+$: $\limsup|sF(s)-L|\le\varepsilon$. ∎
The hypothesis of the final value theorem is that $f(t)$ has a limit — not something you can read off $F$ without care. For a rational $F$ the safe test is: all poles of $sF(s)$ lie in $\operatorname{Re}s\lt0$. Then the partial fraction terms of $f$ are $t^ke^{pt}$ with $\operatorname{Re}p\lt0$ plus a constant, and the limit exists. For $F=\frac1{s^2+1}$ (poles $\pm i$) we get $sF\to0$, but $f=\sin t$ has no limit.
Steps, Impulses and Periodic Inputs
Proof. $\int_0^\infty e^{-st}f(t-a)u(t-a)\,dt$ equals $\int_a^\infty e^{-st}f(t-a)\,dt$, and the substitution $\tau=t-a$ gives $\int_0^\infty e^{-s(\tau+a)}f(\tau)\,d\tau=e^{-as}F(s)$. For the second form apply the first to $f(t)=g(t+a)$. ∎ The shifted function must be $f(t-a)$, delayed as a whole — $\mathcal{L}\{t\,u(t-2)\}$ is not $e^{-2s}/s^2$ but $e^{-2s}\mathcal{L}\{t+2\}=e^{-2s}\left(\frac1{s^2}+\frac2s\right)$.
Proof. $f$ is bounded, so $F$ exists for $s\gt0$. Split $[0,\infty)$ into periods and put $t=\tau+nT$ in the $n$-th: $\int_{nT}^{(n+1)T}e^{-st}f(t)\,dt=e^{-nsT}\int_0^Te^{-s\tau}f(\tau)\,d\tau$. Summing the geometric series $\sum_{n\ge0}e^{-nsT}=\frac1{1-e^{-sT}}$ gives the result. ∎
For $a\ge0$ and $\varepsilon\gt0$ let $d_\varepsilon(t)=\frac1\varepsilon\left(u(t-a)-u(t-a-\varepsilon)\right)$, a pulse of height $\frac1\varepsilon$ and area $1$ on $[a,a+\varepsilon)$. For every function $g$ continuous at $a$, $\int_0^\infty g\,d_\varepsilon\to g(a)$ as $\varepsilon\to0^+$, and
$$\mathcal{L}\{d_\varepsilon\}=e^{-as}\,\frac{1-e^{-\varepsilon s}}{\varepsilon s}\to e^{-as}.$$The unit impulse (Dirac delta) $\delta(t-a)$ is the idealised limit of these pulses. It is not a function (a function vanishing off $\{a\}$ has integral $0$); we use it through the rules $\int_0^\infty g(t)\delta(t-a)\,dt=g(a)$ (the sifting property, $g$ continuous) and $\mathcal{L}\{\delta(t-a)\}=e^{-as}$, in particular $\mathcal{L}\{\delta(t)\}=1$.
Physically, $y''+\dots=c\,\delta(t-a)$ describes a blow at time $a$ that changes the momentum by $c$: $y$ stays continuous and $y'$ jumps by $c$ (divided by the leading coefficient). Note that $\mathcal{L}\{\delta\}=1$ does not tend to $0$ — consistent with Theorem 6.15, since $\delta$ is not a function.
Inverse Transforms and Convolution
By Theorem 6.16, $\mathcal{L}^{-1}$ is well defined on transforms of continuous functions and is linear. For the rational functions that IVPs produce, partial fractions reduce everything to the table.
Let $F=P/Q$ with polynomials $P,Q$, $\deg P\lt\deg Q$.
- If $Q$ has only simple zeros $a_1,\dots,a_n$ (real or complex), then $\displaystyle f(t)=\sum_{k=1}^{n}\frac{P(a_k)}{Q'(a_k)}\,e^{a_kt}$.
- A term $\dfrac{A}{(s-a)^m}$ of the partial fraction expansion contributes $A\,\dfrac{t^{m-1}}{(m-1)!}e^{at}$.
- A term $\dfrac{A(s-\alpha)+B\beta}{(s-\alpha)^2+\beta^2}$ ($\beta\gt0$) contributes $e^{\alpha t}\left(A\cos\beta t+B\sin\beta t\right)$.
In every case $f(t)$ is the sum of the residues of $e^{st}F(s)$ at the poles of $F$ (§5.2).
Proof. Partial fractions write $F$ as a sum of terms $\frac{A}{(s-a)^j}$ (over $\mathbb{C}$), and for real $F$ the conjugate terms combine into the form (3). By the table and Theorem 6.18, $\mathcal{L}\left\{\frac{t^{j-1}}{(j-1)!}e^{at}\right\}=\frac{1}{(s-a)^j}$ and $\mathcal{L}\{e^{\alpha t}\cos\beta t\}=\frac{s-\alpha}{(s-\alpha)^2+\beta^2}$, $\mathcal{L}\{e^{\alpha t}\sin\beta t\}=\frac{\beta}{(s-\alpha)^2+\beta^2}$; linearity finishes (2) and (3). In (1) the coefficient of $\frac1{s-a_k}$ is $\lim_{s\to a_k}(s-a_k)\frac{P(s)}{Q(s)}=\frac{P(a_k)}{Q'(a_k)}$, since $\frac{Q(s)}{s-a_k}\to Q'(a_k)$. The residue of $e^{st}\frac{A}{(s-a)^j}$ at $a$ is $A\frac{t^{j-1}}{(j-1)!}e^{at}$, which gives the last sentence. ∎
Proof. For $s\gt\alpha$ both transforms converge absolutely, so the product of the two integrals is an absolutely convergent double integral over the quadrant (Fubini–Tonelli, §1.3):
$$F(s)G(s)=\int_0^\infty\!\!\int_0^\infty e^{-s(\tau+\sigma)}f(\tau)g(\sigma)\,d\sigma\,d\tau .$$In the inner integral substitute $\sigma=t-\tau$ ($t$ from $\tau$ to $\infty$). The region becomes $\{0\le\tau\le t\lt\infty\}$; changing the order of integration (again justified by absolute convergence),
$$F(s)G(s)=\int_0^\infty e^{-st}\left(\int_0^tf(\tau)g(t-\tau)\,d\tau\right)dt=\mathcal{L}\{f*g\}(s).$$Continuity of $f*g$ follows from the continuity of translation for integrable functions. For the order, we may take $\alpha\ge0$ and, enlarging $K$, assume $|f(t)|,|g(t)|\le Ke^{\alpha t}$ for all $t\ge0$; then $|f*g(t)|\le\int_0^tK^2e^{\alpha\tau}e^{\alpha(t-\tau)}d\tau=K^2te^{\alpha t}$, which is of exponential order $\alpha+\varepsilon$. Commutativity: substitute $\tau\mapsto t-\tau$. Associativity follows from $(FG)H=F(GH)$ and uniqueness, since all the convolutions involved are continuous. ∎
Convolution is not ordinary multiplication: $1*1=t$, not $1$, and $\mathcal{L}\{fg\}\ne FG$ in general. Its meaning is superposition: $(h*f)(t)$ adds up the responses $h(t-\tau)$ to the impulses $f(\tau)\,d\tau$ delivered at all earlier times $\tau$. Interactive 2 shows this happening.
Solving Initial Value Problems; Systems and Transfer Functions
Let $f$ be piecewise continuous of exponential order. The solution of $ay''+by'+cy=f$, $y(0)=y_0$, $y'(0)=y_1$ is of exponential order, and
$$Y(s)=H(s)F(s)+H(s)\left[(as+b)y_0+ay_1\right].$$Hence $y=h*f+y_{\text{free}}$ (Duhamel's principle): the response to the input with zero initial data, plus the response to the initial data with zero input. If every pole has negative real part, then $h$ decays exponentially, every solution with bounded $f$ is bounded, and the response to the unit step $f=u(t)$ with zero data tends to $H(0)=\frac1c$.
Proof. Let $w=h*f$. At points where $f$ is continuous, differentiating under the integral sign and using $h(0)=0$, $h'(0)=\frac1a$:
$$w'=\int_0^th'(t-\tau)f(\tau)\,d\tau,\qquad w''=\int_0^th''(t-\tau)f(\tau)\,d\tau+\frac{f(t)}a .$$Since $ah''+bh'+ch=0$, this gives $aw''+bw'+cw=f(t)$, with $w(0)=w'(0)=0$. Adding the homogeneous solution with the given initial data gives a solution, which by uniqueness (§6.1, existence and uniqueness for linear equations) is the solution. Its pieces are of exponential order (Theorem 6.26 for $w$; sums of $t^ke^{rt}$ for the rest), and $y$, $y'$ are continuous with $y''$ piecewise continuous, so Theorem 6.19 applies: transforming the equation gives $a(s^2Y-sy_0-y_1)+b(sY-y_0)+cY=F$, which is the formula. If all poles satisfy $\operatorname{Re}p\le-\gamma\lt0$, the partial fraction terms of $h$ are $O(t\,e^{-\gamma t})$, so $\int_0^\infty|h|\lt\infty$ and $|h*f|\le\sup|f|\int_0^\infty|h|$; the free response decays too. The step response is $\int_0^th\to\int_0^\infty h=H(0)$. ∎
- Write the forcing with unit steps and impulses; check it is of exponential order.
- Transform the equation using $\mathcal{L}\{y'\}=sY-y(0)$, $\mathcal{L}\{y''\}=s^2Y-sy(0)-y'(0)$; for a system, transform every equation.
- Solve the algebraic equation (or linear system, by Cramer's rule) for $Y(s)$.
- Split $Y$ into partial fractions; keep each $e^{-as}$ factor aside and apply Theorem 6.23 at the end.
- Invert with the table and the shifting theorems; check the initial values (initial value theorem) and, if the system is stable, the final value.
- Not every function has a transform. $e^{t^2}$ grows too fast; $\frac1t$ is not integrable at $0$.
- The existence conditions are sufficient, not necessary. $t^{-1/2}$ is unbounded at $0$ but has transform $\sqrt{\pi/s}$; $2te^{t^2}\cos(e^{t^2})$ is not of exponential order, yet it is the derivative of the bounded function $\sin(e^{t^2})$ and has a transform (Exercise I.7).
- $F(s)\to0$. Hence $1$, $s$, $\frac{s}{s+1}$, $\frac{s^2}{s^2+1}$ are not transforms of piecewise continuous functions of exponential order ($\frac{s}{s+1}=1-\frac1{s+1}$ is the transform of $\delta(t)-e^{-t}$).
- $\mathcal{L}\{fg\}\ne FG$: $\mathcal{L}\{1\cdot1\}=\frac1s\ne\frac1{s^2}$. The product of transforms belongs to the convolution.
- Second shifting needs the delayed function. $\mathcal{L}\{t\,u(t-2)\}=e^{-2s}\left(\frac1{s^2}+\frac2s\right)$, not $\frac{e^{-2s}}{s^2}$ (that is $(t-2)u(t-2)$).
- The derivative rule needs continuity of $f$. For $f=u(t-1)$, $f'=0$ almost everywhere, but $sF-f(0)=e^{-s}\ne0$: the jump contributes $-e^{-s}\cdot1$.
- Final value theorem misuse. $F=\frac{1}{s^2+1}$: $sF\to0$, yet $\sin t$ has no limit. $F=\frac1{s-1}$: $sF\to0$ as $s\to0$, yet $e^t\to\infty$ (and $s\to0$ is outside the region $s\gt1$).
- Uniqueness is up to values at jumps. $u(t)$ and the function equal to $1$ except $f(0)=0$ have the same transform; Lerch's theorem speaks only about points of continuity.
- Convolution is not multiplication. $1*1=t$; $f*1=\int_0^tf\ne f$. No function is an identity for $*$.
- Undamped resonance. $y''+y=\sin t$ gives $Y=\frac{1}{(s^2+1)^2}$ and the unbounded $\frac{\sin t-t\cos t}{2}$: poles on the imaginary axis are not "stable".
- Variable coefficients. Multiplying by $t$ turns $Y$ into $-Y'$: the transformed equation is a differential equation in $s$, not algebra, and it may admit only one solution with a transform (Example 7(b)).
Two Labs
Exercises
- 1 Find $\mathcal{L}\{3t^2-4e^{-2t}+5\sin3t\}$.
- 2 Find $\mathcal{L}\{\cos^22t\}$.
- 3 Find $\mathcal{L}\{t^{3/2}\}$.
- 4 From the definition, find the transform of $f(t)=e^t$ for $0\le t\lt2$, $f(t)=0$ for $t\ge2$. For which $s$ does it exist?
- 5 Show that $t^{10}e^{3t}\sin t$ is of exponential order $\alpha$ for every $\alpha\gt3$, but not for $\alpha=3$.
- 6 Show that neither $\frac1t$ nor $e^{t^2}$ has a Laplace transform, and say which hypothesis of Theorem 6.15 fails in each case.
- 7 Let $g(t)=\sin(e^{t^2})$. Show that $g'(t)=2te^{t^2}\cos(e^{t^2})$ has a Laplace transform for every $s\gt0$, although it is not of exponential order.
- 1 Find $\mathcal{L}\{t^3e^{-t}\}$.
- 2 Find $\mathcal{L}\{e^{2t}\sin3t\}$.
- 3 Find $\mathcal{L}\{t^2\cos t\}$.
- 4 Find $\mathcal{L}\left\{\dfrac{1-\cos t}{t}\right\}$.
- 5 Find $\mathcal{L}\left\{\int_0^t\tau e^{-\tau}d\tau\right\}$ without evaluating the inner integral.
- 6 Evaluate $\int_0^\infty te^{-3t}\cos t\,dt$.
- 7 Evaluate $\int_0^\infty e^{-t}\dfrac{\sin t}{t}\,dt$.
- 8 For $F(s)=\dfrac{2s+3}{s(s+1)(s+3)}$, use the initial and final value theorems to find $f(0)$ and $\lim_{t\to\infty}f(t)$, then confirm by inverting $F$. Why may the final value theorem not be applied to $F(s)=\dfrac1{s^2-1}$?
- 1 Find $\mathcal{L}\{(t-1)^2u(t-1)\}$.
- 2 Find $\mathcal{L}\{t\,u(t-2)\}$.
- 3 Find the transform of $f(t)=\sin t$ for $0\le t\lt\pi$, $f(t)=0$ for $t\ge\pi$ (one arch of a sine).
- 4 Find the transform of the sawtooth wave $f(t)=t$ for $0\le t\lt1$, extended with period $1$.
- 5 Find $\mathcal{L}^{-1}\left\{\dfrac{3s+1}{(s-1)(s^2+1)}\right\}$.
- 6 Find $\mathcal{L}^{-1}\left\{\dfrac{s+3}{(s+1)^3}\right\}$.
- 7 Find $\mathcal{L}^{-1}\left\{\dfrac{se^{-3s}}{s^2+4}\right\}$.
- 8 Find $\mathcal{L}^{-1}\left\{\tan^{-1}\dfrac2s\right\}$.
- 1 Compute $t*e^t$ directly and check it against the convolution theorem.
- 2 Use convolution to find $\mathcal{L}^{-1}\left\{\dfrac{1}{s(s^2+9)}\right\}$.
- 3 Solve $y'+3y=e^{-3t}$, $y(0)=2$.
- 4 Solve $y''-3y'+2y=4$, $y(0)=1$, $y'(0)=0$.
- 5 Solve $y''+9y=\cos3t$, $y(0)=y'(0)=0$, and describe the solution.
- 6 Solve $y''+y=\delta(t-2\pi)$, $y(0)=0$, $y'(0)=1$. What does the blow at $t=2\pi$ do?
- 7 Solve $y''+4y=4u(t-1)$, $y(0)=y'(0)=0$.
- 8 Solve the integro-differential equation $y'+2y+\int_0^ty(\tau)\,d\tau=1$, $y(0)=0$.
- 1 Solve $x'=x-2y$, $y'=3x-4y$, $x(0)=1$, $y(0)=0$.
- 2 Show, by the method of Example 7(b), that the solution of $ty''+2y'+ty=0$ with $y(0)=1$ that has a transform is $\dfrac{\sin t}{t}$.
- 3 For $y''+3y'+2y=f$, find the transfer function, the impulse response and the step response (zero initial data).
- 4 For which real $k$ do all poles of $y''+ky'+4y=f$ lie in $\operatorname{Re}s\lt0$? For $k=0$, find an input bounded by $1$ that produces an unbounded response.
- 5 Prove that $f*g=g*f$ directly from Definition 6.13.
- 6 Prove that there is no piecewise continuous $f$ of exponential order with $\mathcal{L}\{f\}=\dfrac{s}{s^2+1}+1$.
- 7 Modelling. An RL circuit with $L=0.5$ H and $R=10$ Ω is driven by a $20$ V source switched on at $t=0$ and off at $t=1$ s: $\tfrac12i'+10i=20\left(u(t)-u(t-1)\right)$, $i(0)=0$. Find $i(t)$ and the current at $t=1.05$ s.
E1. GATE MA · MCQ $\mathcal{L}^{-1}\left\{\dfrac{s+1}{s^2+6s+13}\right\}$ equals (A) $e^{-3t}(\cos2t-\sin2t)$; (B) $e^{-3t}(\cos2t+\sin2t)$; (C) $e^{3t}(\cos2t-\sin2t)$; (D) $e^{-3t}\left(\cos2t-\tfrac12\sin2t\right)$.
E2. GATE MA · MCQ Let $y$ solve $y''+4y=\delta\left(t-\tfrac\pi2\right)$, $y(0)=y'(0)=0$. Then $y\left(\tfrac{3\pi}{4}\right)$ equals (A) $\tfrac12$; (B) $-\tfrac12$; (C) $0$; (D) $1$.
E3. GATE MA · MSQ Which of the following statements is/are true? (One or more options may be correct.) (A) $\mathcal{L}\{t^{-1/2}\}$ exists for every $s\gt0$. (B) If $f$ is piecewise continuous and of exponential order, then $\mathcal{L}\{f\}(s)\to0$ as $s\to\infty$. (C) For all piecewise continuous $f,g$ of exponential order, $\mathcal{L}\{fg\}=\mathcal{L}\{f\}\,\mathcal{L}\{g\}$. (D) There is a piecewise continuous function of exponential order whose transform is $\dfrac{s}{s+1}$.
E4. GATE MA · MSQ CSIR Part C Consider $y''+2y'+5y=f(t)$. Which of the following is/are true? (One or more options may be correct.) (A) The impulse response is $\tfrac12e^{-t}\sin2t$. (B) If $f=u(t)$ and $y(0)=y'(0)=0$, then $y(t)\to\tfrac15$ as $t\to\infty$. (C) If $f=\sin2t$ and $y(0)=y'(0)=0$, then $y$ is unbounded on $[0,\infty)$. (D) If $f\equiv0$, $y(0)=1$ and $y'(0)=0$, then $Y(s)=\dfrac{s+2}{s^2+2s+5}$.
E5. GATE MA · NAT The value of $\displaystyle\int_0^\infty e^{-t}\,\frac{1-\cos2t}{t}\,dt$ is ______ (correct to two decimal places).
E6. GATE MA · NAT Let $y$ solve $y''+4y'+3y=6\,u(t-1)$, $y(0)=y'(0)=0$. The value of $y(2)$ is ______ (correct to two decimal places).
Suggested plan — 3 lectures × 60 min.
- The transform and its rules. Definitions 6.9–6.10, Theorem 6.15 with proof, Lerch (statement and the idea of the proof), the table, Example 1. Project Interactive 1: show $f=1$ with $s$ sliding through $0$, then $e^{0.5t}$ with $s$ sliding through $0.5$ — the abscissa of convergence becomes visible. Theorems 6.18–6.21 with the integration-by-parts proof of 6.19 on the board; Example 2 (students love the integrals).
- Switches, impulses, inversion. Definition 6.11, Theorems 6.23–6.24, Definition 6.12 (draw the pulses $d_\varepsilon$ shrinking), Example 3. Theorem 6.25 and Example 4 as a drill: give each row of the class one part, three minutes each. Theorem 6.22 and the "final value trap".
- Convolution and IVPs. Theorem 6.26 with the Fubini proof, Example 5; Theorem 6.27 and the strategy box; Examples 6 and 7. Project Interactive 2: first $h=e^{-t}$ with the step (the response climbs to $H(0)=1$), then $h=\sin t$ with $\sin t$ (resonance builds up visibly), then the damped system with the pulse. Finish with GATE corner E4 deciding every option.
Discussion questions. (1) Why does the derivative rule need $f$ to be continuous, and what extra term appears at a jump? Relate it to the jump in $y'$ caused by an impulse. (2) The convolution $1*1=t$. What does this say about the response of the system $y'=f$ to a constant input? (3) Example 7(b) produced only one solution of a second-order equation. Where did the other one go?
Misconceptions to watch for. Writing $\mathcal{L}\{fg\}=FG$; applying the second shifting theorem to $f(t)u(t-a)$ instead of $f(t-a)u(t-a)$; forgetting the initial values in $\mathcal{L}\{y''\}$; completing the square but not adjusting the numerator; using the final value theorem when there are poles on the imaginary axis or in the right half-plane; treating $\delta$ as an ordinary function with $\mathcal{L}\{\delta\}\to0$.
Suggested homework. I(3, 4, 7), II(3, 4, 6, 8), III(2, 3, 4, 5, 8), IV(4, 5, 6, 8), V(1, 3, 4, 7); GATE corner E1, E3, E6.
Research question. A patient receives the same intravenous dose of an antibiotic every $T$ hours. How high and how low does the drug concentration in the blood settle, and how should the dose and the interval be chosen to keep it inside a safe and effective window?
Why it matters. Hospital dosing schedules for antibiotics, anti-epileptics and anticoagulants are designed exactly this way. In the simplest (one-compartment) model the concentration $C(t)$ decays at a rate proportional to itself, $C'=-kC$, and each injection of dose $D$ into a volume $V$ raises $C$ instantly by $D/V$: an impulse. Too low a trough lets bacteria recover; too high a peak can damage the kidneys or the ear.
What to do.
- Write the model as $C'+kC=\dfrac DV\displaystyle\sum_{n\ge0}\delta(t-nT)$, $C(0^-)=0$, and show that $\mathcal{C}(s)=\dfrac{D/V}{(s+k)\left(1-e^{-sT}\right)}$.
- Invert term by term to get $C(t)=\dfrac DV\sum_{n\ge0}e^{-k(t-nT)}u(t-nT)$, and show that the peak just after the $N$-th dose is $\dfrac DV\cdot\dfrac{1-e^{-NkT}}{1-e^{-kT}}$.
- Find the steady-state peak and trough, and the accumulation factor $R=\dfrac{1}{1-e^{-kT}}$. Take a drug with half-life $6$ h ($k=\frac{\log2}6$ per hour), $D/V=10$ mg/L and $T=8$ h: compute the steady peak and trough.
- Choose $T$ (and then $D/V$) so that at steady state the trough is at least $5$ mg/L and the peak at most $20$ mg/L. Is there a range of admissible intervals? Plot $C(t)$ for your choice.
Tools: pencil and paper for 1–3; a spreadsheet or Python for the plot in 4.
Hand in: the derivation of $\mathcal{C}(s)$ and $C(t)$, the steady-state formulas, a table for the example drug, your dosing choice with a plot, and a paragraph on the trade-off between frequent small doses and rare large ones.
Going further. A two-compartment model (blood and tissue) is a $2\times2$ linear system. Use the Laplace transform to find its transfer function from dose to blood concentration, and compare the impulse response — a sum of two decaying exponentials — with the one-compartment model.
🧪 Self-check quiz
Q1. $\mathcal{L}\{te^{2t}\}$ equals
Q2. $\mathcal{L}^{-1}\left\{\dfrac{e^{-2s}}{s^2}\right\}$ equals
Q3. If $y(0)=1$ and $y'(0)=2$, then $\mathcal{L}\{y''\}$ equals
Q4. Which of these cannot be the Laplace transform of a piecewise continuous function of exponential order?
Q5. The solution of $y''+y=f(t)$, $y(0)=y'(0)=0$ (with $f$ piecewise continuous of exponential order) is