Final Exam: Exam-Style Problems

8 final-exam problems of 2.5 points, one per block of the course (ODEs, systems, Fourier and Sturm-Liouville, PDEs), like the four problems of the finals of 2019 and 2020; the Exam simulator draws one of each.

Final exam · Unit 3 · Reduction of order · 2.5 points · 50 min · Standard exam level

P15 — x y'' - (x+2) y' + 2y = 0 with the known solution e^x: a polynomial second solution by reduction of order, the forced equation with x^3 e^x, and initial-value problems at the singular point x = 0

Consider, for $x>0$, the equation

$$x\,y''-(x+2)\,y'+2y=0 .$$

(a) Check that $y_1(x)=e^x$ is a solution. Using $y=v(x)\,e^x$, find a second solution $y_2$, linearly independent of $y_1$, and compute the Wronskian $W(y_1,y_2)$.

(b) Find the general solution of

$$x\,y''-(x+2)\,y'+2y=x^3e^x,\qquad x>0,$$

again with the substitution $y=v\,e^x$.

(c) The solutions of (b) extend smoothly to $x=0$. Show that all of them satisfy $y'(0)=y(0)$. What does this say about the initial-value problems $y(0)=1,\ y'(0)=0$ and $y(0)=1,\ y'(0)=1$? Why does this not contradict Picard's theorem?

Setup and canonical formulation

Phase 1. Hypotheses and setup

  • Unknown $y(x)$, independent variable $x$, domain $x>0$ (the leading coefficient $x$ vanishes at $x=0$).
  • Operator: linear, second order, variable coefficients; (a) homogeneous, (b) non-homogeneous with $g=x^3e^x$.
  • One solution is known, so the order can be reduced: $v'$ satisfies a first-order linear equation.

Phase 2. Canonical formulation

Dividing by $x$ gives the standard form $y''+p(x)y'+q(x)y=f(x)$ with

$$p(x)=-\frac{x+2}{x}=-1-\frac2x,\qquad q(x)=\frac2x,\qquad f(x)=x^2e^x .$$

$p$, $q$, $f$ are continuous on $x>0$, so every initial-value problem posed at $x_0>0$ has a unique solution there; at $x_0=0$ the hypotheses fail.

Part (a) · 1 point

Second solution by reduction of order and the Wronskian

Step 1. Check of y1

$y_1=y_1'=y_1''=e^x$: $xe^x-(x+2)e^x+2e^x=(x-x-2+2)e^x=0$.

Step 2. Substitution

$y=ve^x$, product rule: $y'=e^x(v'+v)$, $y''=e^x(v''+2v'+v)$. Substituting:

$$xe^x(v''+2v'+v)-(x+2)e^x(v'+v)+2ve^x=e^x\big[xv''+(2x-x-2)v'+(x-x-2+2)v\big]=e^x\big[xv''+(x-2)v'\big]$$

So the homogeneous equation becomes $xv''+(x-2)v'=0$.

Step 3. First-order equation for w = v'

$xw'=-(x-2)w$, separable: $\dfrac{dw}{w}=\Big(-1+\dfrac2x\Big)dx$. Integrating, $\ln\lvert w\rvert=-x+2\ln x+c$, so $w=Kx^2e^{-x}$; take $K=1$.

Step 4. Integrate w

By parts twice ($u=x^2$, $dv=e^{-x}dx$, $du=2x\,dx$, $v=-e^{-x}$; then $u=x$, $du=dx$):

$$\int x^2e^{-x}dx=-x^2e^{-x}+2\int xe^{-x}dx=-x^2e^{-x}+2\Big(-xe^{-x}+\int e^{-x}dx\Big)=-(x^2+2x+2)e^{-x}$$

Then $y=ve^x=-(x^2+2x+2)$. Dropping the constant factor $-1$: $y_2=x^2+2x+2$. Check: $y_2'=2x+2$, $y_2''=2$: $2x-(x+2)(2x+2)+2(x^2+2x+2)=2x-(2x^2+6x+4)+2x^2+4x+4=0$.

Step 5. Wronskian

$$W=\begin{vmatrix}e^x&x^2+2x+2\\e^x&2x+2\end{vmatrix}=e^x(2x+2)-e^x(x^2+2x+2)=-x^2e^x$$

It never vanishes on $x>0$, so $\{y_1,y_2\}$ is a fundamental set. Abel's formula agrees: $W=C\exp\Big(-\int p\,dx\Big)=C\exp\Big(x+2\ln x\Big)=Cx^2e^x$, with $C=-1$.

Key results
  • $x v''+(x-2)v'=0$, so $w=v'=x^2e^{-x}$.
  • $v=-(x^2+2x+2)e^{-x}$, hence $y_2=x^2+2x+2$ (a polynomial).
  • $W(e^x,\ x^2+2x+2)=-x^2e^x\neq0$ for $x>0$.

Part (b) · 1 point

General solution of the forced equation

Step 1. Reduced equation

From Step 2 of (a), $L[ve^x]=e^x\big[xv''+(x-2)v'\big]=x^3e^x$, so $xv''+(x-2)v'=x^3$. With $w=v'$ and dividing by $x$:

$$w'+\Big(1-\frac2x\Big)w=x^2$$

Step 2. Integrating factor

$\mu=\exp\displaystyle\int\Big(1-\frac2x\Big)dx=\exp(x-2\ln x)=\frac{e^x}{x^2}$. Check: $\mu'=\dfrac{e^x}{x^2}-\dfrac{2e^x}{x^3}=\mu\Big(1-\dfrac2x\Big)$. Then $(\mu w)'=\mu w'+\mu'w=\mu\big[w'+(1-2/x)w\big]=\dfrac{e^x}{x^2}x^2=e^x$.

Step 3. Integrate twice

$\mu w=e^x+C$, so $w=x^2+Cx^2e^{-x}$. Integrating, with the antiderivative of (a) Step 4:

$$v=\frac{x^3}{3}-C(x^2+2x+2)e^{-x}+D$$

and $y=ve^x$:

$$y=De^x-C(x^2+2x+2)+\frac{x^3e^x}{3}=C_1e^x+C_2(x^2+2x+2)+\frac{x^3e^x}{3}$$

Step 4. Check of the particular solution

$y_p=\tfrac13x^3e^x$: $y_p'=\big(x^2+\tfrac13x^3\big)e^x$, $y_p''=\big(2x+2x^2+\tfrac13x^3\big)e^x$. Then

$$xy_p''-(x+2)y_p'+2y_p=e^x\Big[2x^2+2x^3+\tfrac13x^4-\big(x^3+\tfrac13x^4+2x^2+\tfrac23x^3\big)+\tfrac23x^3\Big]=x^3e^x$$

Variation of constants with $f=x^2e^x$ and $W=-x^2e^x$ gives $y_p-2e^x$, the same up to a homogeneous solution.

Key results
  • $w'+\Big(1-\dfrac2x\Big)w=x^2$, $\mu=\dfrac{e^x}{x^2}$, $(\mu w)'=e^x$.
  • $w=x^2+Cx^2e^{-x}$, $v=\dfrac{x^3}{3}-C(x^2+2x+2)e^{-x}+D$.
  • $y=C_1e^x+C_2(x^2+2x+2)+\dfrac{x^3e^x}{3}$.

Part (c) · 0.5 points

Initial conditions at the singular point x = 0

Step 1. Values at x = 0

$y=C_1e^x+C_2(x^2+2x+2)+\tfrac13x^3e^x$ gives $y(0)=C_1+2C_2$. $y'=C_1e^x+C_2(2x+2)+\big(x^2+\tfrac13x^3\big)e^x$ gives $y'(0)=C_1+2C_2$. So $y'(0)=y(0)$ always. Directly: at $x=0$ the equation reads $0\cdot y''(0)-2y'(0)+2y(0)=0$.

Step 2. The two initial-value problems

$y(0)=1$, $y'(0)=0$ violates $y'(0)=y(0)$: no solution. $y(0)=y'(0)=1$ only gives $C_1+2C_2=1$: a one-parameter family, no uniqueness.

Step 3. Picard

Picard's theorem needs the standard form $y''=f(x)-p(x)y'-q(x)y$ with coefficients continuous near $x_0$. At $x_0=0$, $p$ and $q$ blow up, so the theorem does not apply. Consistently, $W=-x^2e^x$ vanishes at $x=0$: the matrix $\begin{pmatrix}y_1(0)&y_2(0)\\y_1'(0)&y_2'(0)\end{pmatrix}=\begin{pmatrix}1&2\\1&2\end{pmatrix}$ is singular, so the constants cannot be fitted to arbitrary data.

Key results
  • $y(0)=C_1+2C_2=y'(0)$ for every solution.
  • $y(0)=1,\ y'(0)=0$: no solution. $y(0)=y'(0)=1$: infinitely many ($C_1=1-2C_2$, $C_2$ free).
  • No contradiction: $p=-1-2/x$ and $q=2/x$ are not continuous at $x=0$ (equivalently $W(0)=0$).

Interpretation and checks

Phase 4. Asymptotic and physical checks

  • Limits: as $x\to\infty$ the forced response $\tfrac13x^3e^x$ dominates both $e^x$ and the polynomial. As $x\to0^+$ every solution is smooth, but only data on the line $y'(0)=y(0)$ are reachable.
  • Checks: $y_1$, $y_2$, $y_p$ substituted back (Steps (a)4 and (b)4); Abel's formula reproduces the Wronskian; $W\neq0$ on $x>0$.
  • Common mistakes in the exam: (1) forgetting to divide by $x$ before reading $p(x)$ for Abel's formula or variation of constants (with $f=x^2e^x$, not $x^3e^x$); (2) sign slips in $\int x^2e^{-x}dx$; (3) presenting $y_2=-(x^2+2x+2)e^{-x}$, i.e. $v$ instead of $y=ve^x$; (4) claiming uniqueness at $x_0=0$ by quoting Picard without checking continuity of $p$ and $q$.
Final exam · Unit 3 · Constant coefficients, Cauchy-Euler and forcing · 2.5 points · 50 min · High difficulty / trap

P16 — x^2 y'' - x y' + 2y = x sin(ln x): reduction to constant coefficients by x = e^z, a resonant particular solution x ln x cos(ln x), and a boundary-value problem on [1, b] whose existence and uniqueness depend on b

Consider, for $x\ge1$,

$$x^2y''-x\,y'+2y=x\sin(\ln x).$$

(a) With the change of variable $x=e^z$, $Y(z)=y(e^z)$, write the chain rule for $y'$ and $y''$ explicitly and obtain a constant-coefficient equation for $Y$. Find the general solution of the homogeneous equation in terms of $x$.

(b) Find a particular solution. Explain why the natural guess $x\big(A\cos\ln x+B\sin\ln x\big)$ fails.

(c) Consider the boundary-value problem formed by the equation and $y(1)=0$, $y(b)=0$, with $b>1$.

  • For which $b$ does it have exactly one solution? Write it, and simplify it for $b=e^{\pi/2}$.
  • Show that for $b=e^{n\pi}$ ($n=1,2,\dots$) it has no solution.
  • For $b=e^{\pi}$, replace $y(b)=0$ by $y(e^{\pi})=\beta$. For which $\beta$ are there solutions, and how many?

Setup and canonical formulation

Phase 1. Hypotheses and setup

  • Unknown $y(x)$ on $[1,b]$; $x>0$ so $\ln x$ is defined and the leading coefficient $x^2$ does not vanish.
  • Operator: linear, second order, variable coefficients of Cauchy-Euler type ($x^k y^{(k)}$), non-homogeneous with forcing $x\sin(\ln x)$.
  • Two conditions at different points: a boundary-value problem, for which Picard's theorem gives no existence or uniqueness guarantee.

Phase 2. Canonical formulation

Standard form: $y''-\dfrac1x\,y'+\dfrac{2}{x^2}\,y=\dfrac{\sin(\ln x)}{x}$. After $x=e^z$ the problem becomes a constant-coefficient equation on $0\le z\le\ln b$ with $Y(0)=0$, $Y(\ln b)=0$.

Part (a) · 0.75 points

Change of variable and homogeneous solution

Step 1. First derivative

$y(x)=Y(z(x))$ with $z=\ln x$. Chain rule: $y'=\dfrac{dY}{dz}\dfrac{dz}{dx}=\dfrac1x\,Y'(z)$.

Step 2. Second derivative

Product rule on $\dfrac1x\cdot Y'(z(x))$, and chain rule on the second factor, $\dfrac{d}{dx}Y'(z)=Y''(z)\dfrac1x$:

$$y''=-\frac1{x^2}Y'+\frac1x\cdot\frac1xY''=\frac{1}{x^2}\big(Y''-Y'\big)$$

Step 3. Transformed equation

$x^2y''=Y''-Y'$, $\ xy'=Y'$, $\ x=e^z$, $\ \sin(\ln x)=\sin z$:

$$Y''-Y'-Y'+2Y=e^z\sin z\ \Longrightarrow\ Y''-2Y'+2Y=e^z\sin z$$

Step 4. Homogeneous solution

Characteristic polynomial $r^2-2r+2=0$, $r=\dfrac{2\pm\sqrt{4-8}}{2}=1\pm i$. So $Y_h=e^z(C_1\cos z+C_2\sin z)$ and, back with $e^z=x$, $z=\ln x$:

$$y_h=x\big(C_1\cos(\ln x)+C_2\sin(\ln x)\big)$$
Key results
  • $y'=\dfrac1x\dfrac{dY}{dz}$, $\ y''=\dfrac1{x^2}\Big(\dfrac{d^2Y}{dz^2}-\dfrac{dY}{dz}\Big)$.
  • $Y''-2Y'+2Y=e^z\sin z$; characteristic roots $r=1\pm i$.
  • $y_h=x\big(C_1\cos(\ln x)+C_2\sin(\ln x)\big)$.

Part (b) · 0.75 points

Resonant particular solution

Step 1. Why the natural guess fails

$\alpha+i\beta=1+i$ is a simple root of $r^2-2r+2$; $e^z\cos z$ and $e^z\sin z$ solve the homogeneous equation, so any $e^z(A\cos z+B\sin z)$ is mapped to $0$ and cannot produce $e^z\sin z$. The general rule (Robinson Sec. 14.5; Notes.pdf, the second case of the guess) multiplies the guess by $z$ once (simple root).

Step 2. Shift by e^z

Write $Y=e^zw(z)$. Product rule: $Y'=e^z(w'+w)$, $Y''=e^z(w''+2w'+w)$. Then

$$Y''-2Y'+2Y=e^z\big[w''+2w'+w-2w'-2w+2w\big]=e^z\big(w''+w\big)$$

so the equation becomes $w''+w=\sin z$, and the guess is $w=z(A\cos z+B\sin z)$.

Step 3. Coefficients

$w'=A\cos z+B\sin z+z(-A\sin z+B\cos z)$ and

$$w''=2(-A\sin z+B\cos z)-z(A\cos z+B\sin z)$$

Hence $w''+w=-2A\sin z+2B\cos z=\sin z$: $-2A=1$, $2B=0$, so $A=-\tfrac12$, $B=0$.

Step 4. Back to x

$Y_p=-\tfrac12ze^z\cos z$, and with $z=\ln x$, $e^z=x$:

$$y_p=-\frac12\,x\ln x\,\cos(\ln x)$$

the $x^r\ln x$ structure typical of resonance in Euler equations (here $r=1\pm i$). General solution:

$$y=x\big(C_1\cos(\ln x)+C_2\sin(\ln x)\big)-\frac12x\ln x\cos(\ln x)$$
Key results
  • The guess $e^z(A\cos z+B\sin z)$ is a homogeneous solution: substituting gives $0=e^z\sin z$, impossible.
  • Corrected guess $Y_p=ze^z(A\cos z+B\sin z)$; $A=-\tfrac12$, $B=0$.
  • $y_p=-\tfrac12\,x\ln x\cos(\ln x)$; general solution $y=x\big(C_1\cos\ln x+C_2\sin\ln x\big)-\tfrac12x\ln x\cos\ln x$.

Part (c) · 1 point

Boundary-value problem on [1, b]

Step 1. Condition at x = 1

$\ln1=0$, $\cos0=1$, $\sin0=0$: $y(1)=C_1$. So $C_1=0$ and $y=C_2x\sin(\ln x)-\tfrac12x\ln x\cos(\ln x)$.

Step 2. Condition at x = b

$y(b)=b\Big(C_2\sin(\ln b)-\tfrac12\ln b\cos(\ln b)\Big)=0$, and $b>0$, so

$$C_2\sin(\ln b)=\tfrac12\ln b\cos(\ln b)$$

Step 3. Unique case

If $\sin(\ln b)\neq0$, i.e. $b\neq e^{n\pi}$, then $C_2=\tfrac12\ln b\cot(\ln b)$ and

$$y=\frac x2\Big(\ln b\,\cot(\ln b)\sin(\ln x)-\ln x\cos(\ln x)\Big)$$

For $b=e^{\pi/2}$: $\cot(\pi/2)=0$, so $C_2=0$ and the solution is $y_p$ itself, which already vanishes at $x=1$ and at $x=e^{\pi/2}$ (where $\cos(\ln x)=0$).

Step 4. b = e^{n pi}: no solution

$\sin(n\pi)=0$ and $\cos(n\pi)=(-1)^n$, so Step 2 reads $0=\tfrac12\,n\pi(-1)^n$, impossible. Equivalently $y(e^{n\pi})=(-1)^{n+1}\tfrac{n\pi}{2}e^{n\pi}$ whatever $C_2$ is. The reason: $x\sin(\ln x)$ is a non-trivial solution of the homogeneous boundary-value problem (in $z$, $\sin z$ vanishes at $z=0$ and $z=n\pi$, exactly as the eigenfunctions of Haberman Sec. 2.3), and the forcing is not orthogonal to it. Fredholm alternative (extension): the non-homogeneous problem is solvable if and only if the forcing, written for the self-adjoint form of the operator, is orthogonal to every solution of the homogeneous problem. Dividing the equation by $x^3$ gives the self-adjoint form $\Big(\dfrac{y'}{x}\Big)'+\dfrac{2}{x^3}y=\dfrac{\sin(\ln x)}{x^2}$, and with $z=\ln x$, $dz=dx/x$:

$$\int_1^{e^{n\pi}}\frac{\sin(\ln x)}{x^2}\,x\sin(\ln x)\,dx=\int_0^{n\pi}\sin^2z\,dz=\Big[\frac z2-\frac{\sin2z}{4}\Big]_0^{n\pi}=\frac{n\pi}{2}\neq0$$

Step 5. Changed data at b = e^pi

With $y(e^\pi)=\beta$: $e^\pi\big(C_2\cdot0-\tfrac12\pi(-1)\big)=\beta$, i.e. $\beta=\dfrac{\pi e^\pi}{2}\approx36.35$. If $\beta$ has this value, $C_2$ is free: infinitely many solutions $y=y_p+C_2x\sin(\ln x)$. For any other $\beta$, none.

Key results
  • $y(1)=0\Rightarrow C_1=0$. $y(b)=0\Rightarrow C_2\sin(\ln b)=\tfrac12\ln b\cos(\ln b)$.
  • Unique solution iff $b\ne e^{n\pi}$: $\ y=\dfrac x2\Big(\ln b\cot(\ln b)\sin(\ln x)-\ln x\cos(\ln x)\Big)$. For $b=e^{\pi/2}$: $y=y_p=-\tfrac12x\ln x\cos(\ln x)$.
  • $b=e^{n\pi}$: the condition reads $0=(-1)^{n+1}\dfrac{n\pi}{2}e^{n\pi}\neq0$: no solution.
  • $y(e^\pi)=\beta$: solvable iff $\beta=\dfrac{\pi e^{\pi}}{2}\approx36.35$, and then there are infinitely many solutions $y=y_p+C_2x\sin(\ln x)$.

Interpretation and checks

Phase 4. Asymptotic and physical checks

  • Limits: $b\to e^{n\pi}$ from either side: $\cot(\ln b)\to\pm\infty$, so the unique solution blows up, the signature of an approaching eigenvalue. $b\to1^+$: $\ln b\cot(\ln b)\to1$ and $y\to\tfrac x2\big(\sin(\ln x)-\ln x\cos(\ln x)\big)$ restricted to a shrinking interval, which tends to $0$.
  • Checks: $y_p$ substituted back in the original variable: the script verifies $x^2y_p''-xy_p'+2y_p=x\sin(\ln x)$; the boundary values $y(1)=0$, $y(b)=0$ hold identically in $b$; the solution of Step 3 is the case $b=e$ with $C_2=\tfrac12\cot1\approx0.32105$.
  • Common mistakes in the exam: (1) writing $y''=\dfrac1{x^2}Y''$, forgetting the $-Y'$ term produced by the product rule; (2) using the non-resonant guess and getting the contradiction $0=e^z\sin z$ without multiplying by $z$; (3) concluding "unique by Picard" for a boundary-value problem; (4) assuming that a boundary-value problem with a zero-determinant system always has infinitely many solutions: it has none unless the data satisfy the solvability condition.
Final exam · Unit 4 · Nonlinear systems and stability · 2.5 points · 50 min · Standard exam level

P17 — Predators and prey with logistic prey growth: equilibria, Jacobians, the focus-node threshold of the coexistence point and the long-time outcome

After scaling, the densities $x(t)\ge 0$ of a prey and $y(t)\ge 0$ of its predator obey

$$\dot x=x(1-x)-xy,\qquad \dot y=y\,(x-a),$$

where the parameter $a$, with $0<a<1$, is the death rate of the predator measured against its rate of conversion of prey. Without predators the prey grows logistically with carrying capacity $1$.

(a) Find all the equilibria in the closed first quadrant and the Jacobian matrix of the system. Classify the two equilibria that lie on the axes and give their eigen-directions.

(b) Show that the coexistence equilibrium is asymptotically stable for every $a\in(0,1)$, and find the value $a^*$ that separates a stable focus from a stable node. Describe the phase portrait near this point for $a=\tfrac12$ (eigenvalues, sense of rotation) and for $a=\tfrac{9}{11}$ (eigenvalues, eigenvectors, the change of basis $P$ with $P^{-1}JP$, and the slow direction).

(c) Describe the global phase portrait in the open quadrant (nullclines and the direction of the flow) and state the ecological long-time outcome. Hint for the global part: study $V(x,y)=x-a-a\ln\dfrac{x}{a}+y-b-b\ln\dfrac{y}{b}$ with $b=1-a$.

Setup and canonical formulation

Phase 1. Hypotheses and setup

  • Unknowns $x(t)$, $y(t)$, independent variable $t\ge 0$; state space the closed quadrant $x\ge0$, $y\ge0$ (densities).
  • Autonomous, nonlinear (quadratic) planar system $\dot{\mathbf x}=\mathbf f(\mathbf x)$ with $\mathbf f$ polynomial, hence $\mathbf f$ and $D\mathbf f$ are continuous: the existence and uniqueness theorem for systems (Robinson Thm 25.2) gives one solution through each point, and orbits cannot cross.
  • The axes are invariant: if $x=0$ then $\dot x=0$; if $y=0$ then $\dot y=0$. Hence the open quadrant is invariant and densities never become negative.
SymbolMeaningRange
$x$prey density (scaled by the carrying capacity)$x\ge0$
$y$predator density (scaled)$y\ge0$
$a$predator death rate / conversion rate$0<a<1$
$b=1-a$predator density at coexistence$0<b<1$

Phase 2. Canonical formulation

$$\dot{\mathbf x}=\mathbf f(\mathbf x),\qquad \mathbf f(x,y)=\begin{pmatrix}x-x^2-xy\\ xy-ay\end{pmatrix}.$$

Near an equilibrium $\mathbf x^*$, with $\boldsymbol\xi=\mathbf x-\mathbf x^*$, the linearised system is $\dot{\boldsymbol\xi}=J(\mathbf x^*)\,\boldsymbol\xi$ (Robinson Sec. 32.4.2). When no eigenvalue of $J(\mathbf x^*)$ has zero real part, the Hartman-Grobman theorem (Robinson Sec. 32.4.3) says the nonlinear portrait near $\mathbf x^*$ is the linear one, slightly deformed.

Part (a) · 0.8 points

Equilibria, Jacobian and the boundary equilibria

Step 1. Equilibria

$\dot x=0\iff x=0$ or $y=1-x$; $\dot y=0\iff y=0$ or $x=a$. Combine one condition from each equation:

  • $x=0$ and $y=0$: $O=(0,0)$.
  • $x=0$ and $x=a$: impossible, since $a>0$.
  • $y=1-x$ and $y=0$: $x=1$, so $E_1=(1,0)$.
  • $y=1-x$ and $x=a$: $E^*=(a,1-a)$, inside the open quadrant because $0<a<1$.

Step 2. Jacobian

Differentiate each component with respect to each variable:

$$\frac{\partial}{\partial x}(x-x^2-xy)=1-2x-y,\quad \frac{\partial}{\partial y}(x-x^2-xy)=-x,\quad \frac{\partial}{\partial x}(xy-ay)=y,\quad \frac{\partial}{\partial y}(xy-ay)=x-a,$$
$$J(x,y)=\begin{pmatrix}1-2x-y&-x\\ y&x-a\end{pmatrix}.$$

Step 3. The origin

$$J(0,0)=\begin{pmatrix}1&0\\0&-a\end{pmatrix}.$$

The matrix is diagonal, so the eigenvalues are $\lambda_1=1$ with $\mathbf v_1=(1,0)$ and $\lambda_2=-a<0$ with $\mathbf v_2=(0,1)$ (Robinson Example 27.2). Opposite signs: a saddle (Robinson Sec. 28.3), hyperbolic, so the nonlinear point is a saddle too. The $x$-axis is its unstable direction (prey alone grow), the $y$-axis its stable direction (predators alone die out as $y=y_0e^{-at}$, which is exact because the $y$-axis is invariant).

Step 4. The prey-only equilibrium

$$J(1,0)=\begin{pmatrix}1-2-0&-1\\0&1-a\end{pmatrix}=\begin{pmatrix}-1&-1\\0&1-a\end{pmatrix}.$$

It is upper triangular, so its eigenvalues are the diagonal entries (Robinson Example 27.3): $\lambda_1=-1$ and $\lambda_2=1-a>0$. Eigenvectors from $(J-\lambda I)\mathbf v=\mathbf 0$:

  • $\lambda_1=-1$: $\begin{pmatrix}0&-1\\0&2-a\end{pmatrix}\mathbf v=\mathbf 0$. Row 1 gives $-v_2=0$, so $\mathbf v_1=(1,0)$.
  • $\lambda_2=1-a$: $\begin{pmatrix}-(2-a)&-1\\0&0\end{pmatrix}\mathbf v=\mathbf 0$. Row 1 gives $-(2-a)v_1-v_2=0$, so $v_2=(a-2)v_1$ and $\mathbf v_2=(1,a-2)$.

Opposite signs: $E_1$ is a saddle. Its unstable direction $\pm(1,a-2)$ enters the open quadrant along $(-1,\,2-a)$, up and to the left: a small number of predators introduced into a prey population at capacity grows.

Key results
  • Equilibria: $O=(0,0)$, $E_1=(1,0)$, $E^*=(a,\,1-a)$.
  • $J(x,y)=\begin{pmatrix}1-2x-y&-x\\ y&x-a\end{pmatrix}$.
  • $O$: eigenvalues $1$ (along $(1,0)$) and $-a$ (along $(0,1)$), saddle.
  • $E_1$: eigenvalues $-1$ (along $(1,0)$) and $1-a>0$ (along $(1,a-2)$), saddle.

Part (b) · 1 point

Stability of the coexistence point and the focus-node threshold

Step 1. Jacobian at the coexistence point

Substitute $x=a$, $y=1-a$ in $J$: $1-2a-(1-a)=-a$, $-x=-a$, $y=1-a$, $x-a=0$:

$$J^*=J(a,1-a)=\begin{pmatrix}-a&-a\\1-a&0\end{pmatrix}.$$

Step 2. Characteristic polynomial

$$\det(J^*-\lambda I)=\begin{vmatrix}-a-\lambda&-a\\1-a&-\lambda\end{vmatrix}=(-a-\lambda)(-\lambda)-(-a)(1-a)=\lambda^2+a\lambda+a(1-a).$$

So $\operatorname{tr}J^*=-a$, $\det J^*=a(1-a)$ and

$$\lambda=\frac{-a\pm\sqrt{a^2-4a(1-a)}}{2}=\frac{-a\pm\sqrt{a(5a-4)}}{2}.$$

Step 3. Stability for every $a\in(0,1)$

$\det J^*=a(1-a)>0$ excludes a saddle, and $\operatorname{tr}J^*=-a<0$ gives both eigenvalues negative real part: if they are real, their product is positive and their sum negative, so both are negative; if they are complex, the real part is $-a/2<0$. Every eigenvalue has nonzero real part, so the point is hyperbolic and by Hartman-Grobman (Robinson Sec. 32.4.3) $E^*$ is asymptotically stable in the nonlinear system.

Step 4. The threshold

The type depends on the sign of $\Delta(a)=a(5a-4)$ (Robinson Ch. 31, summary of phase portraits):

  • $0<a<\tfrac45$: $\Delta<0$, complex pair with real part $-a/2$: stable focus (spiral).
  • $a=\tfrac45$: $\Delta=0$, double eigenvalue $\lambda=-\tfrac25$; $J^*-\lambda I=\begin{pmatrix}-\frac25&-\frac45\\ \frac15&\frac25\end{pmatrix}\neq0$, so there is a single eigen-direction, $\mathbf v=(2,-1)$ from row 1: stable degenerate (improper) node (Robinson Ch. 30). The eigenvalue is still negative, so the point stays hyperbolic and asymptotically stable; only the fine shape (node or slow spiral) may be changed by the nonlinear terms in this borderline case.
  • $\tfrac45<a<1$: $\Delta>0$, two distinct negative eigenvalues: stable node.

So $a^*=\tfrac45$: predators that die fast (large $a$) let the system settle without oscillating.

Step 5. Focus case $a=\tfrac12$

$\lambda^2+\tfrac12\lambda+\tfrac14=0$, so $\lambda=\dfrac{-\frac12\pm\sqrt{\frac14-1}}{2}=-\dfrac14\pm\dfrac{\sqrt3}{4}i$. Real part $-\tfrac14$ (decay rate), imaginary part $\tfrac{\sqrt3}4$ (angular frequency; period of the linear oscillation $2\pi/(\sqrt3/4)=8\pi/\sqrt3$).

Sense of rotation from a test vector (Robinson Ch. 29): at $\boldsymbol\xi=(1,0)$ (more prey than at equilibrium), $\dot{\boldsymbol\xi}=J^*\boldsymbol\xi=(-\tfrac12,\ \tfrac12)$. The second component is positive: the orbit turns from the positive $\xi_1$ axis towards the positive $\xi_2$ axis, i.e. counter-clockwise in the $(x,y)$ plane. Ecologically: an excess of prey makes the predators grow, the predators then eat the prey down, and so on, with oscillations of decreasing amplitude.

Step 6. Node case $a=\tfrac9{11}$

$$J^*=\begin{pmatrix}-\frac9{11}&-\frac9{11}\\ \frac2{11}&0\end{pmatrix},\qquad \lambda^2+\frac9{11}\lambda+\frac9{11}\cdot\frac2{11}=0,\qquad \lambda=\frac{-\frac9{11}\pm\sqrt{\frac{81}{121}-\frac{72}{121}}}{2}=\frac{-\frac9{11}\pm\frac3{11}}{2}.$$

So $\lambda_1=-\tfrac3{11}$ and $\lambda_2=-\tfrac6{11}$ (check: sum $-\tfrac9{11}=\operatorname{tr}$, product $\tfrac{18}{121}=\det$). Eigenvectors from the first row of $(J^*-\lambda I)\mathbf v=\mathbf 0$ (the second row is proportional because $\det(J^*-\lambda I)=0$):

  • $\lambda_1=-\tfrac3{11}$: $\left(-\tfrac9{11}+\tfrac3{11}\right)v_1-\tfrac9{11}v_2=0\Rightarrow -6v_1-9v_2=0\Rightarrow \mathbf v_1=(3,-2)$. Second row check: $\tfrac2{11}\cdot3+\tfrac3{11}\cdot(-2)=0$.
  • $\lambda_2=-\tfrac6{11}$: $\left(-\tfrac9{11}+\tfrac6{11}\right)v_1-\tfrac9{11}v_2=0\Rightarrow -3v_1-9v_2=0\Rightarrow \mathbf v_2=(3,-1)$. Second row check: $\tfrac2{11}\cdot3+\tfrac6{11}\cdot(-1)=0$.

Change of basis (Robinson Sec. 28.2): $P=[\mathbf v_1\mid\mathbf v_2]=\begin{pmatrix}3&3\\-2&-1\end{pmatrix}$, $\det P=3(-1)-3(-2)=3$, and

$$P^{-1}=\frac13\begin{pmatrix}-1&-3\\2&3\end{pmatrix}.$$

$J^*P=[\lambda_1\mathbf v_1\mid\lambda_2\mathbf v_2]=\begin{pmatrix}-\frac9{11}&-\frac{18}{11}\\ \frac6{11}&\frac6{11}\end{pmatrix}$, and entry by entry

$$\begin{aligned} (P^{-1}J^*P)_{11}&=\tfrac13\left[(-1)(-\tfrac9{11})+(-3)\tfrac6{11}\right]=\tfrac13\cdot\left(-\tfrac9{11}\right)=-\tfrac3{11},\\ (P^{-1}J^*P)_{12}&=\tfrac13\left[(-1)(-\tfrac{18}{11})+(-3)\tfrac6{11}\right]=0,\\ (P^{-1}J^*P)_{21}&=\tfrac13\left[2(-\tfrac9{11})+3\cdot\tfrac6{11}\right]=0,\\ (P^{-1}J^*P)_{22}&=\tfrac13\left[2(-\tfrac{18}{11})+3\cdot\tfrac6{11}\right]=-\tfrac6{11}. \end{aligned}$$

So $P^{-1}J^*P=\operatorname{diag}(-\tfrac3{11},-\tfrac6{11})$. In $\mathbf z=P^{-1}\boldsymbol\xi$ coordinates $z_1=c_1e^{-3t/11}$, $z_2=c_2e^{-6t/11}$, and eliminating $t$, $z_2=K z_1^2$: parabola-like orbits. All orbits enter $E^*$ tangent to the slow eigenvector $\mathbf v_1=(3,-2)$ (smaller $\lvert\lambda\rvert$), and far from $E^*$ they run parallel to the fast one $\mathbf v_2=(3,-1)$ (Robinson Sec. 28.3-28.4).

Key results
  • $J(E^*)=\begin{pmatrix}-a&-a\\1-a&0\end{pmatrix}$, $\operatorname{tr}=-a<0$, $\det=a(1-a)>0$: asymptotically stable for all $a\in(0,1)$.
  • Discriminant $a(5a-4)$: stable focus for $0<a<\tfrac45$, stable degenerate node at $a^*=\tfrac45$, stable node for $\tfrac45<a<1$.
  • $a=\tfrac12$: $\lambda=-\tfrac14\pm\tfrac{\sqrt3}{4}i$, counter-clockwise spiral.
  • $a=\tfrac9{11}$: $\lambda_1=-\tfrac3{11}$, $\mathbf v_1=(3,-2)$ (slow), $\lambda_2=-\tfrac6{11}$, $\mathbf v_2=(3,-1)$ (fast); $P^{-1}JP=\operatorname{diag}(-\tfrac3{11},-\tfrac6{11})$.

Part (c) · 0.7 points

Global portrait and long-time outcome

Step 1. Nullclines and direction field

In the open quadrant, $\dot x>0$ below the line $x+y=1$ and $\dot x<0$ above it; $\dot y>0$ to the right of $x=a$ and $\dot y<0$ to the left. The two lines cross at $E^*$ and cut the quadrant into four regions; the flow goes right-and-up (below the line, right of $x=a$), left-and-up (above, right), left-and-down (above, left) and right-and-down (below, left). Going round these four regions in that order is a counter-clockwise turn about $E^*$, consistent with (b). The axes are invariant: on the $x$ axis orbits run to $E_1$, on the $y$ axis to $O$.

Step 2. The function $V$ (extension)

A Lyapunov function (extension: not in the course; a $C^1$ function with a strict minimum at the equilibrium that does not increase along orbits) proves global statements that linearisation cannot. Here $V$ is defined in the open quadrant. Its derivatives are

$$V_x=1-\frac{a}{x}=\frac{x-a}{x},\qquad V_y=1-\frac{b}{y}=\frac{y-b}{y}.$$

By the chain rule, $\dfrac{d}{dt}V(x(t),y(t))=V_x\dot x+V_y\dot y$:

$$\dot V=\frac{x-a}{x}\,x(1-x-y)+\frac{y-b}{y}\,y(x-a)=(x-a)\big[(1-x-y)+(y-b)\big]=(x-a)(1-b-x).$$

With $1-b=a$: $\dot V=(x-a)(a-x)=-(x-a)^2\le0$.

Also $V\ge0$ with equality only at $E^*$: the function $g(s)=s-c-c\ln(s/c)$ has $g'(s)=1-c/s$, negative for $s<c$ and positive for $s>c$, so its minimum is $g(c)=0$; $V$ is the sum of $g(x)$ with $c=a$ and $g(y)$ with $c=b$. $V\to\infty$ at the axes and at infinity, so every level set $V=\text{const}$ is a closed curve around $E^*$ and orbits stay bounded.

Step 3. Long-time outcome

$V$ decreases strictly except where $x=a$. On the line $x=a$ away from $E^*$, $\dot x=a(1-a-y)\neq0$, so the orbit leaves the line at once: no orbit other than $E^*$ stays where $\dot V=0$. Hence (LaSalle's invariance principle, extension) $V(x(t),y(t))\to0$ and every orbit starting in the open quadrant tends to $E^*$. The linearisation of (b) gives the final approach: spirals for $a<\tfrac45$, tangent to the slow eigenvector for $a>\tfrac45$.

Ecological outcome: neither species dies out; prey and predator settle to $x=a$, $y=1-a$. The prey density at coexistence equals the predator's death parameter; a less efficient predator (larger $a$) leaves more prey and fewer predators, and for $a\ge1$ (outside the data) $E^*$ leaves the quadrant and the predator goes extinct.

Key results
  • Nullclines: $\dot x=0$ on $x=0$ and on $x+y=1$; $\dot y=0$ on $y=0$ and on $x=a$.
  • $\dot V=-(x-a)^2\le0$ (extension: Lyapunov function).
  • Every orbit with $x(0)>0$, $y(0)>0$ tends to $E^*=(a,1-a)$: prey and predators coexist; with $a=\tfrac12$ through damped oscillations, with $a=\tfrac9{11}$ without them.

Interpretation and checks

Phase 4. Asymptotic and physical checks

  • Limits: $a\to0^+$: $E^*\to(0,1)$, trace $\to0$, weakly damped oscillations (close to a centre). $a\to1^-$: $E^*\to E_1=(1,0)$ and the eigenvalue $1-a$ of $E_1$ tends to $0$: the two equilibria merge (transcritical bifurcation, extension). For all $a\in(0,1)$, $t\to\infty$ gives $(x,y)\to(a,1-a)$.
  • Checks: $\mathbf f(a,1-a)=\big(a(1-a-(1-a)),\,(1-a)(a-a)\big)=\mathbf 0$; $\operatorname{tr}$ and $\det$ match sums and products of the eigenvalues in both cases; $P^{-1}J^*P$ is diagonal; numerical integration from $(0.2,0.2)$ ends at $E^*$ for $a=\tfrac12$ and $a=\tfrac9{11}$ (script). The model is dimensionless, so no units appear.
  • Common mistakes in the exam: (1) writing the classical Lotka-Volterra centre: the logistic term $-x^2$ adds $-a$ to the trace and makes $E^*$ a sink; (2) concluding "stable" from $\det>0$ alone, without the sign of the trace; (3) taking the fast eigenvector as the direction of approach of a node: orbits arrive tangent to the slow one, $(3,-2)$; (4) using the linearisation to claim global convergence: Hartman-Grobman is only local, the global statement needs $V$.
Final exam · Unit 4 · Linear systems and phase portraits · 2.5 points · 50 min · Standard exam level

P18 — Damped mass-spring as a first-order system: the origin in the trace-determinant plane as the damping grows, and the critically damped motion through a generalised eigenvector

A body of mass $m=1$ kg hangs from a spring of stiffness $k=4$ N/m and moves inside a dashpot of damping coefficient $c\ge0$ (N s/m). Its displacement $x(t)$ (m) from the equilibrium position obeys

$$m\ddot x+c\dot x+kx=0 .$$

(a) Write the equation as a first-order system $\dot{\mathbf X}=A\mathbf X$ for $\mathbf X=(x,\,v)^T$ with $v=\dot x$, for general $m,k>0$ and $c\ge0$. Locate $A$ in the trace-determinant plane and classify the origin as a function of $c$: undamped, under-damped, critically damped and over-damped motion. With $m=1$, $k=4$ give the critical value $c_{cr}$ and the eigenvalues for $c=0$, $c=2$ and $c=5$, and describe each phase portrait in words.

(b) Take the critically damped case $c=c_{cr}$. Find the eigenvalue and eigenvector of $A$, a generalised eigenvector $\mathbf w$ with $(A-\lambda I)\mathbf w=\mathbf v$, the matrix $P=[\mathbf v\mid\mathbf w]$, and check explicitly that $P^{-1}AP$ is the Jordan form. Write the general solution. The body is released from $x(0)=1$ m with velocity $v(0)=v_0$: find $x(t)$, the values of $v_0$ for which the body crosses the equilibrium position, and the crossing time for $v_0=-6$ m/s. For $v_0=0$, find the largest speed reached.

Setup and canonical formulation

Phase 1. Hypotheses and setup

  • Unknown $x(t)$ (m), independent variable $t\ge0$ (s). Linear, homogeneous, constant coefficients, second order; $m,k>0$, $c\ge0$.
  • Units: $c/m$ and $\sqrt{k/m}$ are s$^{-1}$; $k/m$ is s$^{-2}$. Eigenvalues are rates in s$^{-1}$.
  • The coefficients are constant, so the existence and uniqueness theorem (Robinson Thm 11.1, Thm 25.2) gives a unique solution for every $(x(0),v(0))$, defined for all $t$.
SymbolValueUnit
$m$$1$kg
$k$$4$N/m
$c$$\ge0$, parameterN s/m
$x(0)$$1$m
$v(0)$$v_0$m/s

Phase 2. Canonical formulation

With $x_1=x$, $x_2=v=\dot x$ (Robinson Sec. 25.1): $\dot x_1=x_2$ and, dividing the equation by $m$, $\dot x_2=-\frac km x_1-\frac cm x_2$:

$$\frac{d}{dt}\begin{pmatrix}x\\v\end{pmatrix}=\begin{pmatrix}0&1\\-\frac km&-\frac cm\end{pmatrix}\begin{pmatrix}x\\v\end{pmatrix},\qquad A=\begin{pmatrix}0&1\\-\frac km&-\frac cm\end{pmatrix}.$$

Part (a) · 1 point

Classification of the origin as the damping varies

Step 1. Trace, determinant, characteristic polynomial

$$\det(A-\lambda I)=\begin{vmatrix}-\lambda&1\\-\frac km&-\frac cm-\lambda\end{vmatrix}=(-\lambda)\left(-\frac cm-\lambda\right)-1\cdot\left(-\frac km\right)=\lambda^2+\frac cm\lambda+\frac km .$$

It is the characteristic equation of the scalar ODE divided by $m$, as it must be. Hence $\operatorname{tr}A=-\frac cm\le0$, $\det A=\frac km>0$ and

$$\lambda_\pm=\frac{-\frac cm\pm\sqrt{\frac{c^2}{m^2}-\frac{4k}m}}{2}=\frac{-c\pm\sqrt{c^2-4mk}}{2m}.$$

Step 2. The point $A$ in the trace-determinant plane

$\det A=k/m$ is fixed and positive, so the origin is never a saddle. As $c$ grows from $0$, the point $(\operatorname{tr},\det)=(-c/m,\,k/m)$ moves left along the horizontal line $\det=k/m$, starting on the $\det$ axis and crossing the parabola $\operatorname{tr}^2=4\det$ at $c/m=2\sqrt{k/m}$, i.e. $c_{cr}=2\sqrt{mk}$ (Robinson Ch. 31):

DampingPositionEigenvaluesOrigin
$c=0$ (undamped)on the $\det$ axis$\pm i\sqrt{k/m}$centre (stable, not asymptotically)
$0<c<c_{cr}$ (under-damped)inside the parabola$-\frac c{2m}\pm i\frac{\sqrt{4mk-c^2}}{2m}$stable focus
$c=c_{cr}$ (critical)on the parabola$-\frac c{2m}$, doublestable degenerate node
$c>c_{cr}$ (over-damped)outside the parabolatwo negative realsstable node

With $m=1$, $k=4$: $c_{cr}=2\sqrt{1\cdot4}=4$ N s/m.

Step 3. The three sample values

  • $c=0$: $\lambda^2+4=0$, $\lambda=\pm2i$. Energy $\tfrac12mv^2+\tfrac12kx^2$ is conserved, so the orbits are the ellipses $4x^2+v^2=\text{const}$ (period $\pi$ s). Test vector $\mathbf X=(1,0)$: $A\mathbf X=(0,-4)$ points down, so the motion is clockwise.
  • $c=2$: $\lambda^2+2\lambda+4=0$, $\lambda=\dfrac{-2\pm\sqrt{4-16}}{2}=-1\pm\sqrt3\,i$. Clockwise spirals into the origin (same test vector: $A(1,0)^T=(0,-4)^T$), amplitude decaying as $e^{-t}$.
  • $c=5$: $\lambda^2+5\lambda+4=(\lambda+1)(\lambda+4)=0$, $\lambda_1=-1$, $\lambda_2=-4$. Eigenvectors from row 1 of $(A-\lambda I)\mathbf v=\mathbf 0$, $-\lambda v_1+v_2=0$: $\mathbf v_1=(1,-1)$, $\mathbf v_2=(1,-4)$. Stable node; orbits reach the origin tangent to the slow direction $(1,-1)$ (smaller $\lvert\lambda\rvert$), and far away they are parallel to $(1,-4)$. The body returns to rest without oscillating, and it can cross $x=0$ at most once.
Key results
  • $\lambda^2+\frac cm\lambda+\frac km=0$; discriminant $\dfrac{c^2-4mk}{m^2}$; $c_{cr}=2\sqrt{mk}=4$ N s/m.
  • $c=0$: centre, $\lambda=\pm2i$. $0<c<4$: stable focus ($c=2$: $\lambda=-1\pm\sqrt3\,i$). $c=4$: stable degenerate node, $\lambda=-2$ double. $c>4$: stable node ($c=5$: $\lambda=-1,-4$).
  • Rotation in the $(x,v)$ plane is clockwise in every oscillating case.

Part (b) · 1.5 points

Critically damped motion with a generalised eigenvector

Step 1. Eigenvalue and eigenvector

With $c=4$: $\lambda^2+4\lambda+4=(\lambda+2)^2=0$, so $\lambda=-2$ with algebraic multiplicity 2. Then

$$A+2I=\begin{pmatrix}2&1\\-4&-2\end{pmatrix}.$$

Row 1: $2v_1+v_2=0$; row 2: $-4v_1-2v_2=-2(2v_1+v_2)=0$, the same equation. One free parameter: $\mathbf v=(1,-2)$. Since $A+2I\neq0$, there is no second independent eigenvector.

Step 2. Generalised eigenvector

Solve $(A+2I)\mathbf w=\mathbf v$: row 1 gives $2w_1+w_2=1$, row 2 gives $-4w_1-2w_2=-2$, which is row 1 times $-2$ (consistent). Choose $w_1=0$: $\mathbf w=(0,1)$. Check: $(A+2I)\mathbf w=(2\cdot0+1,\ -4\cdot0-2\cdot1)=(1,-2)=\mathbf v$. Also $(A+2I)^2=\begin{pmatrix}4-4&2-2\\-8+8&-4+4\end{pmatrix}=0$, so $(A+2I)^2\mathbf w=\mathbf 0$ as it must.

Step 3. Change of basis and Jordan form

$$P=[\mathbf v\mid\mathbf w]=\begin{pmatrix}1&0\\-2&1\end{pmatrix},\qquad \det P=1\cdot1-0\cdot(-2)=1,\qquad P^{-1}=\frac11\begin{pmatrix}1&0\\2&1\end{pmatrix}.$$

Columns of $AP$: $A\mathbf v=-2\mathbf v=(-2,4)$ and $A\mathbf w=-2\mathbf w+\mathbf v=(0,-2)+(1,-2)=(1,-4)$; check directly: $A\mathbf w=(0\cdot0+1\cdot1,\ -4\cdot0-4\cdot1)=(1,-4)$. So $AP=\begin{pmatrix}-2&1\\4&-4\end{pmatrix}$ and

$$\begin{aligned} (P^{-1}AP)_{11}&=1\cdot(-2)+0\cdot4=-2, & (P^{-1}AP)_{12}&=1\cdot1+0\cdot(-4)=1,\\ (P^{-1}AP)_{21}&=2\cdot(-2)+1\cdot4=0, & (P^{-1}AP)_{22}&=2\cdot1+1\cdot(-4)=-2, \end{aligned} \qquad P^{-1}AP=\begin{pmatrix}-2&1\\0&-2\end{pmatrix}=J .$$

Step 4. Solution in the canonical coordinates

With $\mathbf z=P^{-1}\mathbf X$: $\dot z_2=-2z_2$ and $\dot z_1=-2z_1+z_2$. The first gives $z_2=c_2e^{-2t}$. The second is linear: multiply by the integrating factor $e^{2t}$, so by the product rule $\frac d{dt}(e^{2t}z_1)=e^{2t}(\dot z_1+2z_1)=e^{2t}z_2=c_2$. Integrate from $0$ to $t$ with Barrow's rule: $e^{2t}z_1(t)-z_1(0)=c_2\,[s]_0^t=c_2t$, so $z_1=(c_1+c_2t)e^{-2t}$ with $c_1=z_1(0)$.

Step 5. General solution

$\mathbf X=P\mathbf z=z_1\mathbf v+z_2\mathbf w$:

$$\mathbf X(t)=c_1e^{-2t}\begin{pmatrix}1\\-2\end{pmatrix}+c_2e^{-2t}\left[t\begin{pmatrix}1\\-2\end{pmatrix}+\begin{pmatrix}0\\1\end{pmatrix}\right], \qquad x(t)=(c_1+c_2t)e^{-2t},\quad v(t)=(-2c_1+c_2-2c_2t)e^{-2t}.$$

Check $v=\dot x$ by the product rule: $\dot x=c_2e^{-2t}+(c_1+c_2t)(-2)e^{-2t}=(c_2-2c_1-2c_2t)e^{-2t}$.

Step 6. Initial conditions

$x(0)=c_1=1$, $v(0)=-2c_1+c_2=v_0\Rightarrow c_2=v_0+2$:

$$x(t)=\big(1+(v_0+2)\,t\big)e^{-2t}.$$

Substitution check: with $x=(1+\beta t)e^{-2t}$, $\beta=v_0+2$, $\dot x=(\beta-2-2\beta t)e^{-2t}$, $\ddot x=(-4\beta+4+4\beta t)e^{-2t}$, and $\ddot x+4\dot x+4x=e^{-2t}\big[(4-4\beta+4\beta t)+(4\beta-8-8\beta t)+(4+4\beta t)\big]=0$.

Step 7. Crossing the equilibrium

$e^{-2t}>0$, so $x(t^*)=0\iff1+(v_0+2)t^*=0\iff t^*=-\dfrac1{v_0+2}$. This is a positive time only if $v_0+2<0$: the body crosses $x=0$ (exactly once) iff $v_0<-2$ m/s, i.e. it is thrown towards the equilibrium fast enough. For $v_0=-6$: $t^*=-\frac1{-4}=0.25$ s. For $v_0\ge-2$ it approaches $x=0$ from above without reaching it.

Step 8. Largest speed for $v_0=0$

$x=(1+2t)e^{-2t}$ and $v=\dot x=(2-2-4t)e^{-2t}=-4te^{-2t}$. Then $\dot v=(-4+8t)e^{-2t}=0$ at $t=\tfrac12$ s, a minimum of $v$ (sign of $\dot v$ changes from $-$ to $+$). $\lvert v\rvert_{\max}=4\cdot\tfrac12e^{-1}=\dfrac2e\approx0.736$ m/s.

Step 9. Phase portrait

Degenerate stable node (Robinson Ch. 30): the only straight-line orbits lie on the eigenline $v=-2x$, with arrows pointing to the origin. Every other orbit is dominated by $c_2te^{-2t}\mathbf v$ both as $t\to+\infty$ (it enters the origin tangent to $v=-2x$) and as $t\to-\infty$ (it comes from far away parallel to the same line). Test vector $(1,0)$: $A(1,0)^T=(0,-4)^T$, so orbits turn clockwise, and those that start above the eigenline with $v_0<-2$ cross the $v$ axis ($x=0$) once before turning back towards the origin.

Key results
  • $A=\begin{pmatrix}0&1\\-4&-4\end{pmatrix}$, $\lambda=-2$ (double), $\mathbf v=(1,-2)$, $\mathbf w=(0,1)$.
  • $P=\begin{pmatrix}1&0\\-2&1\end{pmatrix}$, $P^{-1}=\begin{pmatrix}1&0\\2&1\end{pmatrix}$, $P^{-1}AP=\begin{pmatrix}-2&1\\0&-2\end{pmatrix}$.
  • $x(t)=\big(1+(v_0+2)t\big)e^{-2t}$; crosses $x=0$ once iff $v_0<-2$ m/s, at $t^*=-\dfrac1{v_0+2}$; $v_0=-6$ m/s gives $t^*=0.25$ s.
  • $v_0=0$: $v(t)=-4te^{-2t}$, largest speed $2/e\approx0.736$ m/s at $t=0.5$ s.

Interpretation and checks

Phase 4. Asymptotic and physical checks

  • Limits: $t\to\infty$: $\mathbf X\to\mathbf 0$ for all $c>0$ (all eigenvalues have negative real part). Decay rate of the slowest mode, $\min\lvert\operatorname{Re}\lambda\rvert$: $c/(2m)$ below critical, $2$ s$^{-1}$ at $c=4$, and $\frac{c-\sqrt{c^2-16}}{2}\to0$ as $c\to\infty$ (very strong damping is slow). Critical damping gives the fastest return without oscillation. $c\to0$: focus $\to$ centre.
  • Checks: dimensions: $c_{cr}=2\sqrt{mk}$ has $\sqrt{\text{kg}\cdot\text{N/m}}=\sqrt{\text{kg}^2\text{s}^{-2}}=$ kg/s = N s/m; $t^*=-1/(v_0+2)$: the $2$ is $2$ s$^{-1}\cdot1$ m, so $v_0+2$ is in m/s and $t^*=x(0)/\lvert v_0+2\rvert$ in s. Substitution of $x(t)$ in the ODE and of $\mathbf X$ in $\dot{\mathbf X}=A\mathbf X$ (script); $P^{-1}AP=J$.
  • Common mistakes in the exam: (1) looking for a second eigenvector at $\lambda=-2$ and writing $c_1e^{-2t}\mathbf v+c_2e^{-2t}\mathbf v$, which is one solution twice; (2) writing the second solution as $te^{-2t}\mathbf v$ alone, without $\mathbf w$: it does not satisfy the system; (3) saying a critically damped body can never cross equilibrium: it can, once, when $v_0<-2$ m/s; (4) placing the over-damped case inside the parabola: the parabola $\operatorname{tr}^2=4\det$ separates foci (inside) from nodes (outside).
Final exam · Unit 5 · Sturm-Liouville eigenproblems · 2.5 points · 50 min · Standard exam level

P19 — Sturm-Liouville problem with a Dirichlet end and a Robin end: positivity, the transcendental eigenvalue equation, large eigenvalues, orthogonality and norms

A rod $0\le x\le1$ has its end $x=0$ kept at zero temperature, while its end $x=1$ exchanges heat with surroundings at zero temperature by Newton's law, with a dimensionless transfer coefficient $h>0$. Separating variables in the heat equation leads to the eigenvalue problem

$$X''(x)+\lambda X(x)=0,\quad 0<x<1,\qquad X(0)=0,\qquad X'(1)+h\,X(1)=0 .$$

(a) Prove that the problem has no eigenvalue $\lambda\le0$. For $\lambda=\mu^2>0$ ($\mu>0$) find the eigenfunctions and show that $\mu$ must satisfy $\tan\mu=-\mu/h$.

(b) Show graphically that there are infinitely many eigenvalues $0<\lambda_1<\lambda_2<\dots$, locate $\mu_n$ in an interval of length $\pi/2$, and obtain the asymptotic behaviour of $\mu_n$ and $\lambda_n$ for large $n$. For $h=1$, give $\lambda_1$ and $\lambda_2$ to four significant figures and compare with the asymptotic formula.

(c) Prove, by integrating by parts (Green's identity), that eigenfunctions of different eigenvalues are orthogonal on $[0,1]$. Show that $\displaystyle\int_0^1\sin^2(\mu_nx)\,dx=\frac{\mu_n^2+h^2+h}{2(\mu_n^2+h^2)}$ and write the formula for the coefficients of the expansion $f(x)=\sum_n c_n\sin(\mu_nx)$.

Setup and canonical formulation

Phase 1. Hypotheses and setup

  • Unknowns: the eigenvalue $\lambda\in\mathbb R$ and the eigenfunction $X(x)\not\equiv0$ on $[0,1]$, of class $C^2$.
  • Linear, homogeneous, second-order ODE with constant coefficients; homogeneous boundary conditions: Dirichlet at $x=0$, Robin (third kind) at $x=1$ with $h>0$.
  • Data: $h>0$ (general); $h=1$ in the numerical questions.

Phase 2. Canonical formulation

Sturm-Liouville form (Haberman Sec. 5.3): $\dfrac{d}{dx}\Big(p\dfrac{dX}{dx}\Big)+qX+\lambda\sigma X=0$ with $p=1$, $q=0$, $\sigma=1$, and boundary conditions $\beta_1X(0)+\beta_2X'(0)=0$ with $(\beta_1,\beta_2)=(1,0)$, $\beta_3X(1)+\beta_4X'(1)=0$ with $(\beta_3,\beta_4)=(h,1)$. It is a regular Sturm-Liouville problem ($p,\sigma>0$ on $[0,1]$), so it has real eigenvalues, one eigenfunction per eigenvalue, and orthogonal eigenfunctions with weight $\sigma=1$.

Part (a) · 0.8 points

No eigenvalues $\lambda\le0$ and the eigenvalue equation

Step 1. Case $\lambda=-\mu^2<0$, $\mu>0$

Characteristic equation $r^2-\mu^2=0$, $r=\pm\mu$, so $X=A\cosh(\mu x)+B\sinh(\mu x)$. $X(0)=A=0$. Then $X'=B\mu\cosh(\mu x)$ and

$$X'(1)+hX(1)=B\big(\mu\cosh\mu+h\sinh\mu\big)=0 .$$

For $\mu>0$ and $h>0$ both terms in the bracket are positive, so $B=0$ and $X\equiv0$: no negative eigenvalues.

Step 2. Case $\lambda=0$

$X''=0$, so integrating twice $X=Ax+B$. $X(0)=B=0$; $X'(1)+hX(1)=A+hA=(1+h)A=0$ with $1+h>0$, so $A=0$: $\lambda=0$ is not an eigenvalue.

Step 3. Cross-check with the Rayleigh quotient

Multiplying $X''=-\lambda X$ by $X$ and integrating by parts, $\int_0^1XX''dx=[XX']_0^1-\int_0^1(X')^2dx$, so (Haberman Sec. 5.6)

$$\lambda=\frac{-[XX']_0^1+\int_0^1(X')^2dx}{\int_0^1X^2dx}=\frac{h\,X(1)^2+\int_0^1(X')^2dx}{\int_0^1X^2dx},$$

using $X(0)=0$ and $X'(1)=-hX(1)$. The numerator is $\ge0$, and it is $0$ only if $X'\equiv0$ and $X(1)=0$, i.e. $X\equiv0$. Hence $\lambda>0$, in agreement with Steps 1-2.

Step 4. Case $\lambda=\mu^2>0$

$r^2+\mu^2=0$, $r=\pm i\mu$, so $X=A\cos(\mu x)+B\sin(\mu x)$. $X(0)=A=0$, $X=B\sin(\mu x)$, $X'=B\mu\cos(\mu x)$:

$$X'(1)+hX(1)=B\big(\mu\cos\mu+h\sin\mu\big)=0 .$$

A nontrivial solution needs $B\neq0$, so $\mu\cos\mu+h\sin\mu=0$. If $\cos\mu=0$ this would give $h\sin\mu=\pm h=0$, impossible; so we may divide by $h\cos\mu$:

$$\tan\mu=-\frac{\mu}{h},\qquad X_n(x)=\sin(\mu_nx),\qquad \lambda_n=\mu_n^2 .$$
Key results
  • $\lambda<0$: $\mu\cosh\mu+h\sinh\mu>0$ forces $X\equiv0$. $\lambda=0$: $(1+h)A=0$ forces $X\equiv0$.
  • $\lambda=\mu^2>0$: $X_n=\sin(\mu_nx)$ with $\mu\cos\mu+h\sin\mu=0$, i.e. $\tan\mu=-\mu/h$.

Part (b) · 0.8 points

Infinitely many eigenvalues and their asymptotic behaviour

Step 1. Graphical argument

For $\mu>0$ the line $y=-\mu/h$ is negative and decreasing. The function $\tan\mu$ has branches on $\big((n-\tfrac12)\pi,(n+\tfrac12)\pi\big)$, on each of which it increases continuously from $-\infty$ to $+\infty$; on the first piece $(0,\pi/2)$ it is positive, so there is no root there. On the branch $n\ge1$, $\tan\mu<0$ exactly on $\big((n-\tfrac12)\pi,n\pi\big)$, and the difference $\tan\mu+\mu/h$ goes from $-\infty$ to $n\pi/h>0$ there, increasing strictly (its derivative is $\sec^2\mu+1/h>0$). By the intermediate value theorem it vanishes exactly once:

$$\Big(n-\frac12\Big)\pi<\mu_n<n\pi,\qquad n=1,2,3,\dots$$

There are infinitely many branches, hence infinitely many eigenvalues $\lambda_n=\mu_n^2$, increasing to $+\infty$ (Haberman Sec. 5.3, property 2). Negative roots $-\mu_n$ give the same eigenfunctions up to sign and are not new.

Step 2. Large $n$

Write $p_n=(n-\tfrac12)\pi$ and $\mu_n=p_n+\varepsilon_n$ with $0<\varepsilon_n<\pi/2$. Since $\cos p_n=0$ and $\sin p_n=\pm1$, the addition formulas give $\tan(p_n+\varepsilon)=\dfrac{\sin(p_n+\varepsilon)}{\cos(p_n+\varepsilon)}=\dfrac{\sin p_n\cos\varepsilon}{-\sin p_n\sin\varepsilon}=-\cot\varepsilon$. The eigenvalue equation becomes

$$-\cot\varepsilon_n=-\frac{p_n+\varepsilon_n}{h}\iff\tan\varepsilon_n=\frac{h}{p_n+\varepsilon_n}.$$

The right-hand side tends to $0$, so $\varepsilon_n\to0$ and $\tan\varepsilon_n=\varepsilon_n+O(\varepsilon_n^3)$: $\varepsilon_n=\dfrac h{p_n}+O(p_n^{-3})$. Squaring,

$$\lambda_n=\mu_n^2=p_n^2+2p_n\varepsilon_n+\varepsilon_n^2=\Big(n-\frac12\Big)^2\pi^2+2h+O(n^{-2}).$$

Geometrically: the line becomes steeper relative to the branches, and the roots crowd towards the left asymptotes $(n-\tfrac12)\pi$ (Haberman Sec. 5.9). The leading term $(n-\frac12)^2\pi^2$ is the eigenvalue of the problem with $X(0)=0$, $X'(1)=0$: for large $n$ the eigenfunctions oscillate so fast that the Robin end looks like an insulated end.

Step 3. Numbers for $h=1$

Solving $\mu\cos\mu+\sin\mu=0$ numerically (Newton's method or bisection) inside $(\pi/2,\pi)$ and $(3\pi/2,2\pi)$:

$n$interval$\mu_n$$\lambda_n$asymptotic $\mu_n$asymptotic $\lambda_n$
1$(1.571,\,3.142)$$2.029$$4.116$$\frac\pi2+\frac2\pi=2.207$$\frac{\pi^2}4+2=4.467$
2$(4.712,\,6.283)$$4.913$$24.14$$\frac{3\pi}2+\frac2{3\pi}=4.925$$\frac{9\pi^2}4+2=24.21$

The error of the asymptotic formula drops from about $9\%$ to $0.3\%$ between $n=1$ and $n=2$ for $\lambda_n$.

Key results
  • One root on each branch: $(n-\tfrac12)\pi<\mu_n<n\pi$, $n=1,2,\dots$
  • $\mu_n=(n-\tfrac12)\pi+\dfrac{h}{(n-\frac12)\pi}+O(n^{-3})$, $\ \lambda_n=(n-\tfrac12)^2\pi^2+2h+O(n^{-2})$.
  • $h=1$: $\mu_1=2.029$, $\lambda_1=4.116$; $\mu_2=4.913$, $\lambda_2=24.14$ (asymptotic: $\mu_1\approx2.207$, $\mu_2\approx4.925$).

Part (c) · 0.9 points

Orthogonality and norms

Step 1. Green's identity

For $C^2$ functions $u,v$, the product rule gives $\dfrac{d}{dx}(uv'-vu')=u'v'+uv''-v'u'-vu''=uv''-vu''$. Integrating over $[0,1]$ with Barrow's rule (Haberman Sec. 5.5, Green's formula with $p=1$):

$$\int_0^1(uv''-vu'')\,dx=\big[uv'-vu'\big]_0^1=\big(u(1)v'(1)-v(1)u'(1)\big)-\big(u(0)v'(0)-v(0)u'(0)\big).$$

Step 2. The boundary terms vanish

Take $u=X_m$, $v=X_n$ with $m\neq n$. At $x=0$: $X_m(0)=X_n(0)=0$, so the bracket is $0$. At $x=1$: $X_m'(1)=-hX_m(1)$ and $X_n'(1)=-hX_n(1)$, so

$$X_m(1)X_n'(1)-X_n(1)X_m'(1)=X_m(1)\big(-hX_n(1)\big)-X_n(1)\big(-hX_m(1)\big)=0 .$$

Step 3. Orthogonality

With the ODE, $X_n''=-\lambda_nX_n$ and $X_m''=-\lambda_mX_m$:

$$\int_0^1(X_mX_n''-X_nX_m'')\,dx=(\lambda_m-\lambda_n)\int_0^1X_mX_n\,dx=0 .$$

Since $\lambda_m\neq\lambda_n$ (the $\mu_n$ lie in disjoint intervals), $\displaystyle\int_0^1\sin(\mu_mx)\sin(\mu_nx)\,dx=0$. Note that the $\mu_n$ are not multiples of a common number, so this is not the elementary orthogonality of $\sin(n\pi x)$: it holds because of the boundary condition. (Direct check in the script: the integral $\frac{n\sin m\cos n-m\cos m\sin n}{m^2-n^2}$ vanishes after substituting $\cos\mu=-h\sin\mu/\mu$.)

Step 4. Norm

With $\sin^2s=\frac12(1-\cos2s)$ and the substitution $s=2\mu_nx$, $ds=2\mu_n\,dx$ for the cosine:

$$\int_0^1\sin^2(\mu_nx)\,dx=\Big[\frac x2-\frac{\sin(2\mu_nx)}{4\mu_n}\Big]_0^1=\frac12-\frac{\sin2\mu_n}{4\mu_n}=\frac12-\frac{\sin\mu_n\cos\mu_n}{2\mu_n}.$$

From the eigenvalue equation $\cos\mu_n=-\dfrac{h\sin\mu_n}{\mu_n}$, so $-\dfrac{\sin\mu_n\cos\mu_n}{2\mu_n}=\dfrac{h\sin^2\mu_n}{2\mu_n^2}$. Squaring the same equation, $\mu_n^2\cos^2\mu_n=h^2\sin^2\mu_n$, i.e. $\mu_n^2(1-\sin^2\mu_n)=h^2\sin^2\mu_n$, so $\sin^2\mu_n=\dfrac{\mu_n^2}{\mu_n^2+h^2}$. Therefore

$$\int_0^1\sin^2(\mu_nx)\,dx=\frac12+\frac{h}{2(\mu_n^2+h^2)}=\frac{\mu_n^2+h^2+h}{2(\mu_n^2+h^2)}.$$

For $h=1$, $n=1$: $\dfrac{4.1159+1+1}{2(4.1159+1)}=0.5977$. The norm is slightly above $\tfrac12$ and tends to $\tfrac12$ as $n\to\infty$.

Step 5. Expansion coefficients

If $f(x)=\sum_{n\ge1}c_n\sin(\mu_nx)$, multiply by $\sin(\mu_mx)$, integrate over $[0,1]$ and use Step 3 (only $n=m$ survives; Haberman Sec. 5.3):

$$c_m=\frac{\int_0^1f(x)\sin(\mu_mx)\,dx}{\int_0^1\sin^2(\mu_mx)\,dx}=\frac{2(\mu_m^2+h^2)}{\mu_m^2+h^2+h}\int_0^1f(x)\sin(\mu_mx)\,dx .$$
Key results
  • $\int_0^1(X_mX_n''-X_nX_m'')dx=\big[X_mX_n'-X_nX_m'\big]_0^1=0$, hence $(\lambda_m-\lambda_n)\int_0^1X_mX_n\,dx=0$.
  • $\int_0^1\sin^2(\mu_nx)dx=\dfrac12-\dfrac{\sin2\mu_n}{4\mu_n}=\dfrac{\mu_n^2+h^2+h}{2(\mu_n^2+h^2)}$; for $h=1$, $n=1$: $0.5977$.
  • $c_n=\dfrac{2(\mu_n^2+h^2)}{\mu_n^2+h^2+h}\displaystyle\int_0^1f(x)\sin(\mu_nx)\,dx$.

Interpretation and checks

Phase 4. Asymptotic and physical checks

  • Limits: $h\to0^+$ (insulated end): $\tan\mu=0^-$ pushes $\mu_n\to(n-\frac12)\pi$, the eigenvalues of $X(0)=0$, $X'(1)=0$, and the norm $\to\frac12$. $h\to\infty$ (end held at zero): the line becomes horizontal, $\tan\mu\to0^-$ from the left of $n\pi$, so $\mu_n\to n\pi$, the Dirichlet eigenvalues $n^2\pi^2$, and the norm $\to\frac12$. For every $h>0$, $\lambda_n$ lies between these two Dirichlet-Neumann bounds.
  • Checks: $\sin(\mu_nx)$ satisfies the ODE and both boundary conditions (script); the numerical roots lie in the predicted intervals; $\int_0^1\sin(\mu_1x)\sin(\mu_2x)dx=0$ numerically to 20 digits; the norm formula agrees with direct quadrature. In the heat problem the mode $n$ decays as $e^{-k\lambda_nt}$, so $\lambda_1=4.116$ (for $h=1$) sets the slowest decay rate.
  • Common mistakes in the exam: (1) dropping the case $\lambda<0$ or $\lambda=0$ without proof, or "proving" it with the wrong sign of $h$; (2) writing $\mu_n=n\pi$ or $(n-\tfrac12)\pi$, as for Dirichlet or Neumann ends; (3) placing the roots in $(n\pi,(n+\tfrac12)\pi)$, where $\tan\mu>0$; (4) using $\int_0^1\sin^2=\tfrac12$ for the norm, which is false here because $\sin2\mu_n\neq0$.
Final exam · Unit 5 · Fourier series and convergence · 2.5 points · 50 min · Standard exam level

P20 — Fourier series of the full-wave rectified sine |sin x|: coefficients by product-to-sum formulas, uniform convergence with an error bound, and two numerical series

The output voltage of a full-wave rectifier fed with $\sin x$ is the $2\pi$-periodic function

$$f(x)=\lvert\sin x\rvert,\qquad -\pi<x\le\pi,$$

extended periodically to $\mathbb R$. Use the convention

$$f(x)\sim a_0+\sum_{n=1}^{\infty}\big(a_n\cos nx+b_n\sin nx\big),\qquad a_0=\frac1{2\pi}\int_{-\pi}^{\pi}f\,dx,\quad a_n=\frac1\pi\int_{-\pi}^{\pi}f\cos nx\,dx,\quad b_n=\frac1\pi\int_{-\pi}^{\pi}f\sin nx\,dx .$$

(a) Compute all the Fourier coefficients of $f$, writing out the product-to-sum formula you use. Treat the case $n=1$ separately.

(b) To which function does the series converge, and is the convergence uniform on $\mathbb R$? Justify both answers, and give an $N$ such that the partial sum $S_N$ (all terms with $n\le N$) differs from $f$ by less than $0.035$ everywhere.

(c) Evaluating the series at suitable points, find

$$S=\sum_{k=1}^{\infty}\frac1{4k^2-1}\qquad\text{and}\qquad T=\sum_{k=1}^{\infty}\frac{(-1)^{k+1}}{4k^2-1}.$$

Check the value of $S$ independently with partial fractions.

Setup and canonical formulation

Phase 1. Hypotheses and setup

  • Data: $f(x)=\lvert\sin x\rvert$ on $(-\pi,\pi]$, period $2L=2\pi$ ($L=\pi$).
  • Regularity: $f$ is continuous on $\mathbb R$ (also across $x=\pm\pi$, where $\lvert\sin\pi\rvert=0=\lvert\sin(-\pi)\rvert$), and piecewise smooth: $f'(x)=\cos x$ on $(0,\pi)$ and $-\cos x$ on $(-\pi,0)$, with corners (jumps of $f'$) at $x=k\pi$.
  • Symmetry: $f(-x)=\lvert-\sin x\rvert=f(x)$, so $f$ is even. In fact $f$ has period $\pi$, which will kill the odd harmonics.

Phase 2. Canonical formulation

Even $f$ (Haberman Sec. 3.3): $b_n=0$ and the series is a cosine series with

$$a_0=\frac1\pi\int_0^{\pi}f\,dx,\qquad a_n=\frac2\pi\int_0^{\pi}f\cos nx\,dx ,$$

and on $[0,\pi]$, $f(x)=\sin x$.

Part (a) · 1.2 points

Fourier coefficients

Step 1. Sine coefficients

$f(x)\sin nx$ is even times odd, i.e. odd, so its integral over the symmetric interval $(-\pi,\pi)$ vanishes: substituting $x=-s$, $dx=-ds$, $\int_{-\pi}^0f(x)\sin nx\,dx=\int_{\pi}^{0}f(s)(-\sin ns)(-ds)=-\int_0^{\pi}f(s)\sin ns\,ds$. Hence $b_n=0$ for all $n$.

Step 2. Mean value

$$a_0=\frac1{2\pi}\int_{-\pi}^{\pi}\lvert\sin x\rvert\,dx=\frac1\pi\int_0^\pi\sin x\,dx=\frac1\pi\big[-\cos x\big]_0^\pi=\frac1\pi\big(-\cos\pi+\cos0\big)=\frac2\pi .$$

Step 3. The case $n=1$

$\sin x\cos x=\tfrac12\sin2x$, so

$$a_1=\frac2\pi\int_0^\pi\frac{\sin2x}{2}\,dx=\frac1\pi\Big[-\frac{\cos2x}{2}\Big]_0^\pi=\frac1\pi\Big(-\frac{\cos2\pi}2+\frac{\cos0}2\Big)=0 .$$

Step 4. The case $n\ge2$

Product-to-sum with $A=x$, $B=nx$: $\sin x\cos nx=\tfrac12\big[\sin(1+n)x+\sin(1-n)x\big]=\tfrac12\big[\sin(n+1)x-\sin(n-1)x\big]$. For an integer $m\ge1$, with the substitution $s=mx$, $ds=m\,dx$:

$$\int_0^\pi\sin mx\,dx=\Big[-\frac{\cos mx}{m}\Big]_0^\pi=\frac{1-\cos m\pi}{m}=\frac{1-(-1)^m}{m}.$$

With $m=n+1$, $1-(-1)^{n+1}=1+(-1)^n$; with $m=n-1\ge1$, $1-(-1)^{n-1}=1+(-1)^n$. Therefore

$$a_n=\frac2\pi\cdot\frac12\Big[\frac{1+(-1)^n}{n+1}-\frac{1+(-1)^n}{n-1}\Big]=\frac{1+(-1)^n}{\pi}\cdot\frac{(n-1)-(n+1)}{n^2-1}=-\frac{2\big(1+(-1)^n\big)}{\pi(n^2-1)} .$$
  • $n$ odd: $1+(-1)^n=0$, so $a_n=0$ (consistent with Step 3 and with the period $\pi$ of $f$).
  • $n=2k$ even: $1+(-1)^n=2$ and $n^2-1=4k^2-1$, so $a_{2k}=-\dfrac{4}{\pi(4k^2-1)}$.

Step 5. The series

$$\lvert\sin x\rvert\sim\frac2\pi-\frac4\pi\sum_{k=1}^\infty\frac{\cos2kx}{4k^2-1}.$$

First terms: $a_2=-\frac4{3\pi}$, $a_4=-\frac4{15\pi}$, $a_6=-\frac4{35\pi}$.

Key results
  • $b_n=0$; $a_0=\dfrac2\pi$; $a_1=0$; $a_n=0$ for odd $n\ge3$; $a_{2k}=-\dfrac{4}{\pi(4k^2-1)}$.
  • $\lvert\sin x\rvert=\dfrac2\pi-\dfrac4\pi\displaystyle\sum_{k=1}^\infty\frac{\cos2kx}{4k^2-1}=\frac2\pi-\frac4\pi\Big(\frac{\cos2x}{3}+\frac{\cos4x}{15}+\frac{\cos6x}{35}+\cdots\Big)$.

Part (b) · 0.6 points

Convergence

Step 1. Pointwise limit

$f$ is piecewise smooth on $[-\pi,\pi]$, so by the Fourier convergence theorem (Haberman Sec. 3.2) its series converges at each $x$ to $\tfrac12\big[f(x^+)+f(x^-)\big]$ of the periodic extension. The periodic extension of $\lvert\sin x\rvert$ is $\lvert\sin x\rvert$ itself on all of $\mathbb R$, which is continuous, so the average equals $f(x)$: the series converges to $\lvert\sin x\rvert$ everywhere, including the corners $x=k\pi$. There is no Gibbs overshoot (Haberman Sec. 3.3: a continuous, piecewise smooth periodic extension gives a continuous limit).

Step 2. Uniform convergence

For all $x$, $\Big\lvert\dfrac{\cos2kx}{4k^2-1}\Big\rvert\le\dfrac{1}{4k^2-1}=M_k$ and $\sum M_k<\infty$ (it behaves like $\sum\frac1{4k^2}$). By the Weierstrass M-test the series converges absolutely and uniformly on $\mathbb R$ (equivalently, Haberman Sec. 3.4: a continuous periodic extension with piecewise smooth derivative has a uniformly convergent Fourier series).

Step 3. Error bound

Partial fractions: $\dfrac{1}{4k^2-1}=\dfrac{1}{(2k-1)(2k+1)}=\dfrac12\Big(\dfrac1{2k-1}-\dfrac1{2k+1}\Big)$ (check: $\frac12\cdot\frac{(2k+1)-(2k-1)}{(2k-1)(2k+1)}=\frac{1}{4k^2-1}$). The tail telescopes:

$$\sum_{k=K+1}^{\infty}\frac{1}{4k^2-1}=\frac12\Big[\Big(\frac1{2K+1}-\frac1{2K+3}\Big)+\Big(\frac1{2K+3}-\frac1{2K+5}\Big)+\cdots\Big]=\frac{1}{2(2K+1)} .$$

The partial sum with all harmonics $n\le N=2K$ keeps the terms $k\le K$, so

$$\lvert f(x)-S_{2K}(x)\rvert\le\frac4\pi\sum_{k>K}\frac1{4k^2-1}=\frac{2}{\pi(2K+1)}\quad\text{for all }x .$$

$\frac{2}{\pi(2K+1)}<0.035\iff2K+1>\frac{2}{0.035\pi}=18.19\iff K\ge9$. $K=9$ ($N=18$) already gives $\frac2{19\pi}=0.0335$; with $K=10$ ($N=20$) the bound is $\frac{2}{21\pi}\approx0.0303$. Any $N\ge18$ works. The bound is attained at $x=0$, where every cosine equals $1$.

Key results
  • The series converges to $\lvert\sin x\rvert$ at every $x\in\mathbb R$ (no jumps), and the convergence is uniform.
  • $\sup_x\lvert f-S_{2N}\rvert\le\dfrac4\pi\displaystyle\sum_{k>N}\frac1{4k^2-1}=\frac{2}{\pi(2N+1)}$; any $N\ge18$ works; $N=18$ (terms up to $\cos18x$) gives $\le\dfrac{2}{19\pi}\approx0.0335<0.035$.

Part (c) · 0.7 points

Two numerical series

Step 1. At $x=0$

By (b) the series equals $f(0)=\lvert\sin0\rvert=0$ there:

$$0=\frac2\pi-\frac4\pi\sum_{k=1}^\infty\frac{1}{4k^2-1}\ \Longrightarrow\ S=\frac{2/\pi}{4/\pi}=\frac12 .$$

Step 2. At $x=\pi/2$

$f(\pi/2)=1$ and $\cos(k\pi)=(-1)^k$:

$$1=\frac2\pi-\frac4\pi\sum_{k=1}^\infty\frac{(-1)^k}{4k^2-1}=\frac2\pi+\frac4\pi\sum_{k=1}^\infty\frac{(-1)^{k+1}}{4k^2-1}=\frac2\pi+\frac4\pi\,T .$$

So $T=\dfrac\pi4\Big(1-\dfrac2\pi\Big)=\dfrac{\pi-2}{4}\approx0.2854$. Check with the first terms: $\frac13-\frac1{15}+\frac1{35}-\frac1{63}=0.2794$, approaching $0.2854$ with alternating error smaller than the next term $\frac1{99}$.

Step 3. Independent check of $S$

From the telescoping of (b), Step 3, with $K=0$: the partial sum is $S_K=\sum_{k=1}^K\frac1{4k^2-1}=\frac12\Big(1-\frac1{2K+1}\Big)=\frac12-\frac{1}{2(2K+1)}$, so $S=\lim_{K\to\infty}S_K=\frac12$, the same value as in Step 1.

Key results
  • $x=0$: $S=\dfrac12$.
  • $x=\pi/2$: $T=\dfrac{\pi-2}{4}\approx0.2854$.
  • Partial sums $S_K=\dfrac12-\dfrac1{2(2K+1)}\to\dfrac12$.

Interpretation and checks

Phase 4. Asymptotic and physical checks

  • Limits: the coefficients decay like $1/n^2$, the rate expected for a continuous function with a jump in its derivative (a function with jumps would give $1/n$); the mean value $a_0=2/\pi\approx0.637$ is the DC output of the rectifier, the classical result for a full-wave rectifier of unit amplitude.
  • Checks: every coefficient by direct integration over $(-\pi,0)$ and $(0,\pi)$ (script, $n\le11$); the series at $x=0.7$ and $x=-2.5$ matches $\lvert\sin x\rvert$; Parseval's identity (extension: $\frac1\pi\int_{-\pi}^{\pi}f^2dx=2a_0^2+\sum a_n^2$ in this convention) gives $\sum\frac{1}{(4k^2-1)^2}=\frac{\pi^2-8}{16}$, also checked. Term-by-term differentiation (Haberman Sec. 3.4) is allowed because $f$ is continuous and periodic: the derived series $\frac8\pi\sum\frac{k\sin2kx}{4k^2-1}$ converges to $f'$, and to $0$ (the average of the jump $\pm1$) at $x=k\pi$.
  • Common mistakes in the exam: (1) using the formula for $a_n$ with $n=1$, which divides by $n-1=0$; (2) integrating $\sin x\cos nx$ over $(-\pi,\pi)$ without the absolute value, which gives $a_n=0$ for all $n$; (3) mixing conventions: with $\frac{a_0}{2}$ in front, $a_0=\frac4\pi$, not $\frac2\pi$; (4) claiming only "convergence in the mean" or a Gibbs overshoot at $x=0$: the extension is continuous, so convergence is uniform.
Final exam · Unit 6 · Laplace equation by separation · 2.5 points · 50 min · Standard exam level

P21 — Laplace's equation in a circular wedge with one edge held at zero and the other insulated: angular eigenfunctions, bounded radial powers and a datum that is a hidden finite sum

A thin plate has the shape of the circular sector $D=\{(r,\theta):\ 0<r<a,\ 0<\theta<\alpha\}$, with $0<\alpha\le2\pi$. Its faces are insulated, so its steady temperature $u(r,\theta)$ solves

$$\Delta u=u_{rr}+\frac1ru_r+\frac1{r^2}u_{\theta\theta}=0\quad\text{in }D,$$
$$u(r,0)=0,\qquad u_\theta(r,\alpha)=0\quad(0<r<a),\qquad \lvert u(r,\theta)\rvert\ \text{bounded as } r\to0^+,\qquad u(a,\theta)=f(\theta)\quad(0<\theta<\alpha).$$

The straight edge $\theta=0$ is held at temperature $0$, the straight edge $\theta=\alpha$ is insulated, and the arc $r=a$ has a prescribed temperature.

(a) Separate variables, $u=R(r)\Theta(\theta)$, and solve the angular eigenvalue problem, discussing the three possible signs of the separation constant. Show that the eigenfunctions are $\Theta_n=\sin\big(\nu_n\theta\big)$ with $\nu_n=\dfrac{(2n-1)\pi}{2\alpha}$, $n=1,2,\dots$

(b) Solve the radial equation, justify which solutions are admissible, and write the general bounded solution together with the formula for its coefficients in terms of $f$.

(c) Let $\alpha=\pi/2$ (a quarter disc) and $f(\theta)=\sin^3\theta$. Find $u(r,\theta)$ in closed form, check all the conditions, write it in Cartesian coordinates $x=r\cos\theta$, $y=r\sin\theta$, and give $u$ at $(r,\theta)=(a/2,\pi/2)$ and at $(a/2,\pi/6)$.

Setup and canonical formulation

Phase 1. Hypotheses and setup

  • Unknown $u(r,\theta)$, independent variables $r\in(0,a)$, $\theta\in(0,\alpha)$; $u\in C^2(D)$, continuous up to the boundary except possibly at corners.
  • Linear, homogeneous, second-order elliptic PDE (Laplace). Homogeneous conditions on the two straight edges (Dirichlet at $\theta=0$, Neumann at $\theta=\alpha$) and a boundedness condition at the vertex $r=0$; the only nonhomogeneous condition is on the arc $r=a$. This is the structure that allows separation of variables with eigenfunctions in $\theta$.
  • Data: $a>0$ (radius), $\alpha$ (opening angle), $f$ (arc datum). In (c): $\alpha=\pi/2$, $f=\sin^3\theta$; $u$ is a dimensionless temperature.

Phase 2. Canonical formulation

Multiplying the PDE by $r^2$: $r^2u_{rr}+ru_r+u_{\theta\theta}=0$. With $u=R(r)\Theta(\theta)$:

$$\frac{r^2R''+rR'}{R}=-\frac{\Theta''}{\Theta}=\lambda\quad(\text{constant}),$$
$$\Theta''+\lambda\Theta=0,\ \ \Theta(0)=0,\ \ \Theta'(\alpha)=0;\qquad r^2R''+rR'-\lambda R=0,\ \ \lvert R(0^+)\rvert<\infty .$$

The angular problem is a regular Sturm-Liouville problem with $p=\sigma=1$, $q=0$ (Haberman Sec. 5.3).

Part (a) · 1 point

Separation and the angular eigenvalue problem

Step 1. Separation

$u_{rr}=R''\Theta$, $u_r=R'\Theta$, $u_{\theta\theta}=R\Theta''$. Then $r^2R''\Theta+rR'\Theta+R\Theta''=0$; dividing by $R\Theta$ (where it is not zero),

$$\frac{r^2R''+rR'}{R}=-\frac{\Theta''}{\Theta}.$$

The left side depends only on $r$ and the right only on $\theta$, so both equal a constant $\lambda$. The edge conditions become $R(r)\Theta(0)=0$ and $R(r)\Theta'(\alpha)=0$ for all $r$; for a nontrivial $R$, $\Theta(0)=0$ and $\Theta'(\alpha)=0$.

Step 2. Case $\lambda=-\mu^2<0$

$\Theta=A\cosh\mu\theta+B\sinh\mu\theta$. $\Theta(0)=A=0$. $\Theta'(\alpha)=B\mu\cosh\mu\alpha=0$; since $\mu>0$ and $\cosh\mu\alpha\ge1$, $B=0$. Trivial.

Step 3. Case $\lambda=0$

$\Theta''=0$, $\Theta=A\theta+B$. $\Theta(0)=B=0$, $\Theta'(\alpha)=A=0$. Trivial: unlike the full disc, the constant is not allowed (it would violate $u=0$ on $\theta=0$).

Step 4. Case $\lambda=\nu^2>0$

$\Theta=A\cos\nu\theta+B\sin\nu\theta$. $\Theta(0)=A=0$. $\Theta'(\alpha)=B\nu\cos\nu\alpha=0$ with $B\neq0$, $\nu>0$, so $\cos\nu\alpha=0$:

$$\nu\alpha=\frac{(2n-1)\pi}{2}\ \Longrightarrow\ \nu_n=\frac{(2n-1)\pi}{2\alpha},\qquad \Theta_n(\theta)=\sin\Big(\frac{(2n-1)\pi\theta}{2\alpha}\Big),\qquad n=1,2,\dots$$

Each $\Theta_n$ vanishes at $\theta=0$ and has zero slope at $\theta=\alpha$: a quarter wave, three quarter waves, and so on across the opening.

Step 5. Orthogonality and norm

For $m\neq n$ the eigenvalues differ, so $\int_0^\alpha\Theta_m\Theta_n\,d\theta=0$ (Haberman Sec. 5.3; the boundary terms $[\Theta_m\Theta_n'-\Theta_n\Theta_m']_0^\alpha$ vanish because $\Theta(0)=0$ and $\Theta'(\alpha)=0$). For $m=n$, with $\sin^2s=\frac12(1-\cos2s)$:

$$\int_0^\alpha\sin^2\nu_n\theta\,d\theta=\Big[\frac\theta2-\frac{\sin2\nu_n\theta}{4\nu_n}\Big]_0^\alpha=\frac\alpha2-\frac{\sin\big((2n-1)\pi\big)}{4\nu_n}=\frac\alpha2 .$$
Key results
  • $\lambda<0$ and $\lambda=0$: only $\Theta\equiv0$.
  • $\lambda=\nu^2>0$: $\Theta=\sin\nu\theta$ with $\cos\nu\alpha=0$, so $\nu_n=\dfrac{(2n-1)\pi}{2\alpha}$, $\lambda_n=\nu_n^2$.
  • $\displaystyle\int_0^\alpha\sin\nu_m\theta\,\sin\nu_n\theta\,d\theta=\frac\alpha2\delta_{mn}$.

Part (b) · 0.6 points

Radial equation and the general bounded solution

Step 1. Euler equation

With $R=r^p$: $R'=pr^{p-1}$, $R''=p(p-1)r^{p-2}$, so $r^2R''+rR'-\nu^2R=\big[p(p-1)+p-\nu^2\big]r^p=(p^2-\nu^2)r^p$. Hence $p=\pm\nu_n$ and, since $\nu_n\neq0$, the two roots are distinct (Haberman Sec. 2.5):

$$R_n(r)=c_1r^{\nu_n}+c_2r^{-\nu_n}.$$

Step 2. Boundedness at the vertex

$\nu_n>0$, so $r^{-\nu_n}\to\infty$ as $r\to0^+$ while $r^{\nu_n}\to0$. The condition $\lvert u\rvert<\infty$ at $r=0$ forces $c_2=0$. (The $\lambda=0$ branch $c_1+c_2\ln r$ does not appear, because $\lambda=0$ is not an eigenvalue.)

Step 3. Superposition and coefficients

By linearity (Haberman Sec. 2.2), the sum of the product solutions is a solution of the PDE that satisfies the homogeneous conditions. Writing the constants as $A_n/a^{\nu_n}$ for convenience:

$$u(r,\theta)=\sum_{n=1}^\infty A_n\Big(\frac ra\Big)^{\nu_n}\sin\nu_n\theta .$$

At $r=a$: $f(\theta)=\sum_nA_n\sin\nu_n\theta$. Multiplying by $\sin\nu_m\theta$, integrating over $(0,\alpha)$ and using (a), Step 5:

$$A_m=\frac{\int_0^\alpha f(\theta)\sin\nu_m\theta\,d\theta}{\alpha/2}=\frac2\alpha\int_0^\alpha f(\theta)\sin\nu_m\theta\,d\theta .$$
Key results
  • $R=r^p$ gives $p^2-\nu_n^2=0$, $R=c_1r^{\nu_n}+c_2r^{-\nu_n}$; boundedness forces $c_2=0$.
  • $u(r,\theta)=\displaystyle\sum_{n=1}^\infty A_n\Big(\frac ra\Big)^{\nu_n}\sin\nu_n\theta$, $\ A_n=\frac2\alpha\int_0^\alpha f(\theta)\sin\nu_n\theta\,d\theta$.

Part (c) · 0.9 points

Quarter disc with $f=\sin^3\theta$

Step 1. Eigenvalues for the quarter disc

$\nu_n=\dfrac{(2n-1)\pi}{2\cdot\pi/2}=2n-1$: eigenfunctions $\sin\theta$, $\sin3\theta$, $\sin5\theta,\dots$, and radial factors $r$, $r^3$, $r^5,\dots$

Step 2. The datum as a finite sum

$\sin3\theta=\sin(2\theta+\theta)=\sin2\theta\cos\theta+\cos2\theta\sin\theta=2\sin\theta\cos^2\theta+(1-2\sin^2\theta)\sin\theta$. With $\cos^2\theta=1-\sin^2\theta$: $\sin3\theta=2\sin\theta-2\sin^3\theta+\sin\theta-2\sin^3\theta=3\sin\theta-4\sin^3\theta$. Solving for the cube:

$$\sin^3\theta=\frac34\sin\theta-\frac14\sin3\theta .$$

Both terms are eigenfunctions, so by uniqueness of the coefficients (orthogonality) $A_1=\tfrac34$, $A_2=-\tfrac14$ and $A_n=0$ for $n\ge3$. Cross-check of $A_1$ with the formula of (b): $A_1=\frac4\pi\int_0^{\pi/2}\sin^3\theta\sin\theta\,d\theta=\frac4\pi\cdot\frac{3\pi}{16}=\frac34$, using $\int_0^{\pi/2}\sin^4\theta\,d\theta=\frac{3\pi}{16}$ (from $\sin^4=\frac38-\frac12\cos2\theta+\frac18\cos4\theta$).

Step 3. Solution

$$u(r,\theta)=\frac34\,\frac ra\,\sin\theta-\frac14\Big(\frac ra\Big)^3\sin3\theta .$$

Checks: $u(r,0)=0$ because $\sin0=0$; $u_\theta=\frac34\frac ra\cos\theta-\frac34\big(\frac ra\big)^3\cos3\theta$ vanishes at $\theta=\pi/2$ because $\cos\frac\pi2=\cos\frac{3\pi}2=0$; $u$ is bounded (a polynomial in $r$); $u(a,\theta)=\frac34\sin\theta-\frac14\sin3\theta=\sin^3\theta$. Each term $r^\nu\sin\nu\theta$ is harmonic by Step 1 of (b).

Step 4. Cartesian form

$r\sin\theta=y$, and $r^3\sin3\theta=r^3(3\sin\theta\cos^2\theta-\sin^3\theta)=3(r\cos\theta)^2(r\sin\theta)-(r\sin\theta)^3=3x^2y-y^3$ (using $\sin3\theta=3\sin\theta-4\sin^3\theta=3\sin\theta\cos^2\theta-\sin^3\theta$). Therefore

$$u(x,y)=\frac{3y}{4a}-\frac{3x^2y-y^3}{4a^3}.$$

Direct check: $u_{xx}=-\dfrac{6y}{4a^3}$, $u_{yy}=\dfrac{6y}{4a^3}$, so $\Delta u=0$. On the insulated edge $x=0$ ($\theta=\pi/2$), $u_x=-\dfrac{6xy}{4a^3}=0$: the normal derivative vanishes, as required. On $y=0$ ($\theta=0$), $u=0$.

Step 5. Values

  • $(r,\theta)=(a/2,\pi/2)$: $\sin\frac\pi2=1$, $\sin\frac{3\pi}2=-1$, so $u=\frac34\cdot\frac12+\frac14\cdot\frac18=\frac38+\frac1{32}=\frac{13}{32}\approx0.406$.
  • $(r,\theta)=(a/2,\pi/6)$: $\sin\frac\pi6=\frac12$, $\sin\frac\pi2=1$, so $u=\frac34\cdot\frac12\cdot\frac12-\frac14\cdot\frac18=\frac3{16}-\frac1{32}=\frac5{32}\approx0.156$.

Both lie between the minimum $0$ and the maximum $1$ of the boundary values, as the maximum principle for Laplace's equation requires (Haberman Sec. 2.5).

Key results
  • $\nu_n=2n-1$; $\sin^3\theta=\tfrac34\sin\theta-\tfrac14\sin3\theta$; $A_1=\tfrac34$, $A_2=-\tfrac14$, $A_n=0$ for $n\ge3$.
  • $u(r,\theta)=\dfrac34\,\dfrac ra\sin\theta-\dfrac14\Big(\dfrac ra\Big)^3\sin3\theta$; Cartesian: $u=\dfrac{3y}{4a}-\dfrac{3x^2y-y^3}{4a^3}$.
  • $u(a/2,\pi/2)=\dfrac{13}{32}$, $\ u(a/2,\pi/6)=\dfrac5{32}$.

Interpretation and checks

Phase 4. Asymptotic and physical checks

  • Limits: $r\to0$: $u\approx\frac34\frac ra\sin\theta\to0$, the vertex takes the temperature of the cold edge; the leading mode $r^{\nu_1}$ dominates near the vertex. For a general opening, $\nu_1=\pi/(2\alpha)<1$ when $\alpha>\pi/2$: $u$ stays bounded but $\lvert\nabla u\rvert\sim r^{\nu_1-1}\to\infty$ at the vertex (a corner singularity caused by the mixed conditions; with $u=0$ on both edges it would only appear for $\alpha>\pi$). The quarter disc, $\nu_1=1$, is exactly the borderline case with a finite gradient. Along the insulated edge, $u(r,\pi/2)=\frac34s+\frac14s^3$ with $s=r/a$ increases monotonically from $0$ to $1$.
  • Checks: the PDE, the two edge conditions, boundedness and the arc datum are satisfied (script, in polar and Cartesian form); the coefficient formula of (b) reproduces $A_1=\frac34$, $A_2=-\frac14$ and $A_n=0$ for $n=3,4,5$; the norm $\int_0^{\pi/2}\sin^23\theta\,d\theta=\frac\pi4=\frac\alpha2$. Dimensionless problem: $(r/a)^\nu$ is a pure number for any $a$.
  • Common mistakes in the exam: (1) using the full-disc eigenfunctions $\cos n\theta,\sin n\theta$ or the half-disc ones $\sin n\theta$ ($\nu=n$): the insulated edge requires odd quarter-wave multiples; (2) keeping $r^{-\nu}$ or a $\ln r$ term, which blow up at the vertex; (3) computing the coefficients by integrals over $(0,\pi)$ or with norm $\pi/2$ instead of $\alpha/2=\pi/4$; (4) not recognising $\sin^3\theta$ as a finite sum and writing an infinite series with wrong integrals.
Final exam · Unit 6 · Heat and wave equations by separation · 2.5 points · 50 min · Standard exam level

P22 — Heat equation in a rod with insulated ends: conservation of the total heat, the zero eigenvalue, a cosine-power initial profile and the time for the midpoint to settle

A homogeneous metal rod of length $L=0.5$ m, with lateral surface and both ends thermally insulated, has thermal diffusivity $k=1.0\times10^{-4}$ m$^2$/s. Its temperature $u(x,t)$ (in $^\circ$C) obeys

$$u_t=k\,u_{xx},\quad 0<x<L,\ t>0,\qquad u_x(0,t)=u_x(L,t)=0,\qquad u(x,0)=20+40\cos^4\Big(\frac{\pi x}{2L}\Big).$$

(a) Without solving the equation, prove that the total heat $H(t)=\displaystyle\int_0^Lu(x,t)\,dx$ (proportional to the thermal energy, with constant $c\rho A$) does not change in time. Deduce the uniform temperature $u_\infty$ the rod tends to, assuming it tends to a uniform temperature.

(b) Solve the problem by separation of variables. Discuss the three signs of the separation constant, including the zero eigenvalue, and write the initial profile as a finite cosine combination using a power-reduction identity.

(c) Confirm $u_\infty$ from the solution. Find the time constant of the slowest decaying mode and the time $t_{10}$ after which the deviation of the midpoint temperature $u(L/2,t)$ from $u_\infty$ has fallen to one tenth of its initial value. Give the temperatures of both ends at $t_{10}$.

Setup and canonical formulation

Phase 1. Hypotheses and setup

  • Unknown $u(x,t)$ ($^\circ$C), independent variables $x\in[0,L]$ (m) and $t\ge0$ (s); $u$ is $C^2$ in $x$ and $C^1$ in $t$ for $t>0$.
  • Linear, homogeneous, second-order parabolic PDE with constant coefficient $k>0$; homogeneous Neumann (insulated) conditions at both ends (Haberman Sec. 1.3); nonhomogeneous initial condition, smooth and compatible with the end conditions ($f'(0)=f'(L)=0$).
  • No heat sources and no flux through the boundary.
SymbolValueUnit
$L$$0.5$m
$k$$1.0\times10^{-4}$m$^2$/s
$f(x)=u(x,0)$$20+40\cos^4\frac{\pi x}{2L}$$^\circ$C
$f(0)$, $f(L/2)$, $f(L)$$60$, $30$, $20$$^\circ$C

Phase 2. Canonical formulation

Product solutions $u=X(x)T(t)$ lead to the Sturm-Liouville problem $X''+\lambda X=0$, $X'(0)=X'(L)=0$ ($p=\sigma=1$, $q=0$) and to $T'=-\lambda kT$. The initial profile is rewritten as

$$f(x)=A_0+\sum_{n\ge1}A_n\cos\frac{n\pi x}{L}.$$

Part (a) · 0.6 points

Conservation of the total heat and the final temperature

Step 1. Derivative of the total heat

The limits $0$ and $L$ do not depend on $t$ and $u_t$ is continuous, so we may differentiate under the integral (Haberman Sec. 1.2):

$$H'(t)=\frac{d}{dt}\int_0^Lu(x,t)\,dx=\int_0^Lu_t(x,t)\,dx=k\int_0^Lu_{xx}(x,t)\,dx .$$

An antiderivative of $u_{xx}$ with respect to $x$ is $u_x$, so by Barrow's rule

$$H'(t)=k\big[u_x(x,t)\big]_{x=0}^{x=L}=k\big(u_x(L,t)-u_x(0,t)\big)=k(0-0)=0 .$$

Hence $H(t)=H(0)$ for all $t$: insulated ends let no heat in or out (Haberman Sec. 1.4).

Step 2. Initial total heat

With $\cos^4\theta=\frac{3+4\cos2\theta+\cos4\theta}{8}$ (proved in (b), Step 4) and $\theta=\frac{\pi x}{2L}$, $f=35+20\cos\frac{\pi x}{L}+5\cos\frac{2\pi x}{L}$. For $n\ge1$, with $s=\frac{n\pi x}{L}$, $ds=\frac{n\pi}{L}dx$:

$$\int_0^L\cos\frac{n\pi x}L\,dx=\frac{L}{n\pi}\Big[\sin s\Big]_0^{n\pi}=\frac{L}{n\pi}\big(\sin n\pi-\sin0\big)=0 .$$

So $H(0)=\int_0^L35\,dx=35L$.

Step 3. Final temperature

If $u\to u_\infty$ (a constant) as $t\to\infty$, then $H\to u_\infty L$; but $H=35L$ for all $t$, so

$$u_\infty=\frac1L\int_0^Lf(x)\,dx=35\ ^\circ\text{C},$$

the mean of the initial temperature, not the mean $40\ ^\circ$C of the end values $60$ and $20$.

Key results
  • $H'(t)=k\big[u_x(L,t)-u_x(0,t)\big]=0$, so $H(t)=H(0)$.
  • $H(0)=\int_0^Lf\,dx=35L$, so $u_\infty=\dfrac1L\int_0^Lf\,dx=35\ ^\circ$C.

Part (b) · 1.3 points

Separation of variables and the solution

Step 1. Separation

With $u=X(x)T(t)$: $XT'=kX''T$; dividing by $kXT$,

$$\frac{T'}{kT}=\frac{X''}{X}=-\lambda .$$

The end conditions give $X'(0)T(t)=0$ and $X'(L)T(t)=0$, so $X'(0)=X'(L)=0$ for a nontrivial $T$.

Step 2. The three cases for $X''+\lambda X=0$

  • $\lambda=-\mu^2<0$: $X=A\cosh\mu x+B\sinh\mu x$, $X'=\mu(A\sinh\mu x+B\cosh\mu x)$. $X'(0)=\mu B=0\Rightarrow B=0$. $X'(L)=\mu A\sinh\mu L=0$ with $\sinh\mu L>0$, so $A=0$: trivial.
  • $\lambda=0$: $X=A+Bx$, $X'=B$; $X'(0)=B=0$ and $X'(L)=0$ holds for any $A$. Eigenvalue $\lambda_0=0$ with $X_0=1$: the constant temperature is an equilibrium of the insulated rod (Haberman Sec. 2.4).
  • $\lambda=\mu^2>0$: $X=A\cos\mu x+B\sin\mu x$, $X'=\mu(-A\sin\mu x+B\cos\mu x)$. $X'(0)=\mu B=0\Rightarrow B=0$. $X'(L)=-\mu A\sin\mu L=0$ with $A\neq0$, so $\sin\mu L=0$, $\mu=\frac{n\pi}{L}$: $\lambda_n=\big(\frac{n\pi}L\big)^2$, $X_n=\cos\frac{n\pi x}L$, $n=1,2,\dots$

Step 3. Time factors

$T'=-\lambda_nkT$ is separable: for $T\neq0$, $\frac{dT}{T}=-\lambda_nk\,dt$, integrate from $0$ to $t$ with Barrow's rule, $\ln\lvert T(t)\rvert-\ln\lvert T(0)\rvert=-\lambda_nkt$, so $T_n(t)=T_n(0)e^{-\lambda_nkt}$. For $n=0$, $T_0$ is constant. By superposition (Haberman Sec. 2.2):

$$u(x,t)=A_0+\sum_{n=1}^\infty A_n\,e^{-k(n\pi/L)^2t}\cos\frac{n\pi x}{L}.$$

Step 4. Initial profile as a finite cosine sum

With $\cos^2\theta=\frac{1+\cos2\theta}{2}$ applied twice:

$$\cos^4\theta=\Big(\frac{1+\cos2\theta}2\Big)^2=\frac{1+2\cos2\theta+\cos^22\theta}{4}=\frac14\Big(1+2\cos2\theta+\frac{1+\cos4\theta}2\Big)=\frac{3+4\cos2\theta+\cos4\theta}{8}.$$

With $\theta=\frac{\pi x}{2L}$: $2\theta=\frac{\pi x}L$, $4\theta=\frac{2\pi x}L$, and

$$f(x)=20+40\cdot\frac{3+4\cos\frac{\pi x}L+\cos\frac{2\pi x}L}{8}=20+15+20\cos\frac{\pi x}{L}+5\cos\frac{2\pi x}{L}.$$

Check: $f(0)=35+20+5=60$, $f(L/2)=35+0-5=30$, $f(L)=35-20+5=20$, as in the data table.

Step 5. Coefficients and solution

$f$ is already a combination of eigenfunctions, so matching $u(x,0)=f$ term by term (orthogonality of $\cos\frac{n\pi x}L$ on $[0,L]$, Haberman Sec. 2.4: $A_0=\frac1L\int_0^Lf$, $A_n=\frac2L\int_0^Lf\cos\frac{n\pi x}L$) gives $A_0=35$, $A_1=20$, $A_2=5$ and $A_n=0$ for $n\ge3$:

$$u(x,t)=35+20\,e^{-k\pi^2t/L^2}\cos\frac{\pi x}L+5\,e^{-4k\pi^2t/L^2}\cos\frac{2\pi x}L .$$

Verification: $u_t=-\frac{20k\pi^2}{L^2}e^{-k\pi^2t/L^2}\cos\frac{\pi x}L-\frac{20k\pi^2}{L^2}e^{-4k\pi^2t/L^2}\cos\frac{2\pi x}L$ and, by the chain rule, $ku_{xx}=k\Big[-20\frac{\pi^2}{L^2}e^{-k\pi^2t/L^2}\cos\frac{\pi x}L-5\frac{4\pi^2}{L^2}e^{-4k\pi^2t/L^2}\cos\frac{2\pi x}L\Big]$, the same expression. $u_x$ contains only $\sin\frac{\pi x}L$ and $\sin\frac{2\pi x}L$, which vanish at $x=0$ and $x=L$.

Key results
  • $\lambda<0$: trivial; $\lambda_0=0$: $X_0=1$; $\lambda_n=\big(\frac{n\pi}L\big)^2$: $X_n=\cos\frac{n\pi x}L$, $T_n=e^{-k(n\pi/L)^2t}$.
  • $f(x)=35+20\cos\dfrac{\pi x}{L}+5\cos\dfrac{2\pi x}{L}$.
  • $u(x,t)=35+20\,e^{-k\pi^2t/L^2}\cos\dfrac{\pi x}L+5\,e^{-4k\pi^2t/L^2}\cos\dfrac{2\pi x}L$.

Part (c) · 0.6 points

Final state and settling times

Step 1. Long-time limit

Both exponents are negative, so as $t\to\infty$ the two modes vanish and $u\to A_0=35\ ^\circ$C, the value predicted in (a) from conservation alone. The zero eigenvalue is the reason: its mode does not decay, and its amplitude is the mean of $f$.

Step 2. Time constants

Mode $n$ decays as $e^{-t/\tau_n}$ with $\tau_n=\dfrac{L^2}{k n^2\pi^2}$. The slowest is $n=1$:

$$\tau_1=\frac{(0.5)^2}{(1.0\times10^{-4})\pi^2}=\frac{0.25}{9.870\times10^{-4}}\approx253.3\ \text{s},\qquad \tau_2=\frac{\tau_1}4\approx63.3\ \text{s}.$$

Step 3. Midpoint

At $x=L/2$: $\cos\frac\pi2=0$ and $\cos\pi=-1$, so $u(L/2,t)=35-5e^{-4k\pi^2t/L^2}$. The midpoint starts at $30\ ^\circ$C and warms monotonically to $35\ ^\circ$C; its deviation $5e^{-4k\pi^2t/L^2}$ is governed by the faster mode only. One tenth of the initial deviation:

$$e^{-4k\pi^2t_{10}/L^2}=\frac1{10}\iff\frac{4k\pi^2t_{10}}{L^2}=\ln10\iff t_{10}=\frac{L^2\ln10}{4\pi^2k}=\frac{0.25\cdot2.3026}{4\pi^2\cdot10^{-4}}\approx145.8\ \text{s}.$$

Step 4. Ends at $t_{10}$

$t_{10}/\tau_1=\frac{\ln10}{4}$, so $e^{-t_{10}/\tau_1}=10^{-1/4}\approx0.5623$, and the second mode equals $5\cdot\frac1{10}=0.5$. At $x=0$ (both cosines $=1$): $u=35+20(0.5623)+0.5\approx46.75\ ^\circ$C. At $x=L$ ($\cos\pi=-1$, $\cos2\pi=1$): $u=35-11.25+0.5\approx24.25\ ^\circ$C. The ends are still far from equilibrium when the midpoint has almost settled: the midpoint is a node of the slow mode.

Key results
  • $u\to35\ ^\circ$C; slowest time constant $\tau_1=\dfrac{L^2}{k\pi^2}\approx253.3$ s ($\tau_2=\tau_1/4\approx63.3$ s).
  • $u(L/2,t)=35-5e^{-4k\pi^2t/L^2}$; $t_{10}=\dfrac{L^2\ln10}{4\pi^2k}\approx145.8$ s.
  • At $t_{10}$: $u(0)=35+20\cdot10^{-1/4}+0.5\approx46.75\ ^\circ$C, $u(L)\approx24.25\ ^\circ$C.

Interpretation and checks

Phase 4. Asymptotic and physical checks

  • Limits: $t\to0^+$ reproduces $f$ (Step 4 of (b)); $t\to\infty$ gives the mean $35\ ^\circ$C; for $t\gg\tau_2$, $u-35\approx20e^{-t/\tau_1}\cos\frac{\pi x}{L}$, a single half cosine with the hot end at $x=0$. Doubling $L$ multiplies all times by $4$ (diffusive scaling $t\sim L^2/k$).
  • Checks: dimensions: $k\pi^2t/L^2$ has (m$^2$/s)(s)/m$^2$, dimensionless; $\tau_1=L^2/(k\pi^2)$ is in s. $H(t)=35L$ for all $t$ computed directly from the solution (script). The PDE, both Neumann conditions and the initial condition hold (script); the coefficients also follow from the integrals $\frac1L\int f$ and $\frac2L\int f\cos\frac{n\pi x}L$ with the original $\cos^4$ form.
  • Common mistakes in the exam: (1) forgetting the $\lambda=0$ eigenvalue and its constant mode, which gives a rod that cools to $0\ ^\circ$C although no heat can leave; (2) using sines $\sin\frac{n\pi x}L$ (the Dirichlet eigenfunctions); (3) taking the final temperature as the average of the end values, $40\ ^\circ$C, instead of the spatial mean $35\ ^\circ$C; (4) computing $t_{10}$ with the slow rate $k\pi^2/L^2$ at the midpoint, where that mode is absent ($t$ would come out four times too large, $583$ s).