2nd Midterm: Exam-Style Problems
7 exam-style problems on nonlinear systems, Fourier series and the heat, wave and Laplace equations, worth 5 points each. The second partials of 2019 to 2025 are two problems in 1 to 2 hours; the Exam simulator draws one of each block.
P08 — Two competing species with a competition parameter b: equilibria, Jacobians, the threshold b = 4/3 between coexistence and competitive exclusion
Two species share a habitat. Their populations $x(t)\ge0$, $y(t)\ge0$ (in thousands of individuals, time in years) obey
where $b>0$ is a constant.
(a) Interpret each term of the model. Find all the equilibria as functions of $b$ and the values of $b$ for which there is an equilibrium with both species present ($x>0$, $y>0$).
(b) Compute the Jacobian matrix and classify every equilibrium in the closed first quadrant as $b$ varies. Identify the threshold value of $b$ that separates coexistence from competitive exclusion.
(c) Take $b=1$. Draw (describe) the nullclines, the eigen-directions at each equilibrium and the phase portrait in the first quadrant. What happens in the long run to a population with $x(0)>0$, $y(0)>0$?
(d) Take $b=2$. Answer the same questions. Which species survives?
Setup and canonical formulation
Phase 1. Hypotheses and setup
- Unknowns $x(t)$, $y(t)$; independent variable $t\ge0$; the right-hand side is polynomial, hence $C^1$, so solutions exist and are unique (Picard). Autonomous, nonlinear (quadratic terms), planar.
- The axes are invariant: $x=0\Rightarrow x'=0$, $y=0\Rightarrow y'=0$. By uniqueness no orbit crosses them, so the closed first quadrant is invariant (populations stay non-negative).
- Parameter: $b>0$, the harm done to $y$ by each unit of $x$.
| Coefficient | Meaning |
|---|---|
| $6$, $4$ | intrinsic growth rates of $x$, $y$ (year$^{-1}$) |
| $-2x^2$, $-y^2$ | intraspecific competition; carrying capacities $K_x=3$, $K_y=4$ |
| $-xy$ in $x'$ | harm to $x$ from encounters with $y$ |
| $-bxy$ in $y'$ | harm to $y$ from encounters with $x$ |
Phase 2. Canonical formulation
$\mathbf x'=\mathbf F(\mathbf x)$ with $\mathbf F(x,y)=\big(x(6-2x-y),\ y(4-y-bx)\big)^T$. Near an equilibrium $\mathbf x_0$, $\mathbf u=\mathbf x-\mathbf x_0$ obeys $\mathbf u'\approx D\mathbf F(\mathbf x_0)\,\mathbf u$; when the equilibrium is hyperbolic (no eigenvalue with zero real part) the Hartman-Grobman theorem (Robinson Sec. 32.4.3) says the nonlinear portrait near it is the linear one.
Part (a) · 1 point
Interpretation, equilibria and existence of coexistence
Step 1. Interpretation
Each species grows logistically when alone ($y=0$: $x'=2x(3-x)$, capacity 3; $x=0$: $y'=y(4-y)$, capacity 4). The cross terms $-xy$ and $-bxy$ are both negative: each species reduces the growth of the other, so the model is one of competition (Robinson Sec. 33.1). The larger $b$, the stronger the pressure of $x$ on $y$.
Step 2. Equilibria
$x'=0$ when $x=0$ or $2x+y=6$; $y'=0$ when $y=0$ or $y+bx=4$. The four combinations:
- $x=0$, $y=0$: $(0,0)$.
- $x=0$, $y+bx=4$: $(0,4)$.
- $2x+y=6$, $y=0$: $(3,0)$.
- $2x+y=6$ and $bx+y=4$: subtracting, $(2-b)x=2$, so for $b\neq2$
Step 3. Positivity
$x^*>0\iff b<2$. Then $2-b>0$ and $y^*>0\iff8-6b>0\iff b<\tfrac43$. So the coexistence point is in the open quadrant exactly for $0<b<\tfrac43$. At $b=\tfrac43$: $x^*=\frac{2}{2/3}=3$, $y^*=0$, it coincides with $(3,0)$. For $\tfrac43<b<2$ it lies in $y<0$ (no biological meaning); for $b=2$ the two lines are parallel and there is none.
- Equilibria: $(0,0)$, $(3,0)$, $(0,4)$ and, for $b\neq2$, $\left(\dfrac{2}{2-b},\ \dfrac{8-6b}{2-b}\right)$.
- The coexistence point lies in $x>0$, $y>0$ if and only if $0<b<\tfrac43$; at $b=\tfrac43$ it merges with $(3,0)$.
Part (b) · 2 points
Jacobian and classification as b varies
Step 1. Jacobian
$\dfrac{\partial}{\partial x}\big[6x-2x^2-xy\big]=6-4x-y$, $\dfrac{\partial}{\partial y}\big[6x-2x^2-xy\big]=-x$, $\dfrac{\partial}{\partial x}\big[4y-y^2-bxy\big]=-by$, $\dfrac{\partial}{\partial y}\big[4y-y^2-bxy\big]=4-2y-bx$.
Step 2. Origin
$D\mathbf F(0,0)=\begin{pmatrix}6&0\\0&4\end{pmatrix}$: two positive eigenvalues, unstable node, eigenvectors along the axes. Orbits leave tangent to the slow direction, the $y$ axis ($4<6$).
Step 3. The point (0, 4)
$D\mathbf F(0,4)=\begin{pmatrix}6-4&0\\-4b&4-8\end{pmatrix}=\begin{pmatrix}2&0\\-4b&-4\end{pmatrix}$. Lower triangular: eigenvalues $2$ and $-4$, a saddle for every $b$. Eigenvector of $-4$: $(0,1)$, the $y$ axis is the stable manifold. Eigenvector of $2$: $(D\mathbf F-2I)\mathbf v=\mathbf 0$ gives row 2 $-4bv_1-6v_2=0$, so $\mathbf v=(3,-2b)^T$: the unstable manifold enters the quadrant towards the lower right. Meaning: a few individuals of $x$ introduced into the $y$-only state grow (their per capita rate there is $6-4=2>0$).
Step 4. The point (3, 0)
$D\mathbf F(3,0)=\begin{pmatrix}6-12&-3\\0&4-3b\end{pmatrix}=\begin{pmatrix}-6&-3\\0&4-3b\end{pmatrix}$. Upper triangular: eigenvalues $-6$ (along the $x$ axis) and $4-3b$, which is the per capita growth rate of a few $y$ individuals in the $x$-only state.
- $b<\tfrac43$: $4-3b>0$, saddle ($y$ can invade).
- $b>\tfrac43$: $4-3b<0$, stable node ($y$ cannot invade).
- $b=\tfrac43$: one zero eigenvalue, non-hyperbolic: the linearisation decides nothing there (Robinson Sec. 32.4.3). This is where the coexistence point crosses $(3,0)$.
Step 5. Coexistence point
At $(x^*,y^*)$: $6-4x^*-y^*=(6-2x^*-y^*)-2x^*=-2x^*$ and $4-2y^*-bx^*=(4-y^*-bx^*)-y^*=-y^*$, so
- $\operatorname{tr}=-(2x^*+y^*)=-6$ (because $2x^*+y^*=6$ on the nullcline).
- $\det=2x^*y^*-bx^*y^*=(2-b)x^*y^*=(2-b)\dfrac{2}{2-b}\dfrac{8-6b}{2-b}=\dfrac{2(8-6b)}{2-b}$, positive for $0<b<\tfrac43$.
- Discriminant: $\operatorname{tr}^2-4\det=(-2x^*-y^*)^2-4(2x^*y^*-bx^*y^*)=(2x^*-y^*)^2+4bx^*y^*>0$.
Real eigenvalues, both negative ($\operatorname{tr}<0$, $\det>0$): stable node for every $0<b<\tfrac43$. (The off-diagonal entries have the same sign, which always forbids a focus in a competition model.)
Step 6. Summary and threshold
| $b$ | $(0,0)$ | $(0,4)$ | $(3,0)$ | coexistence |
|---|---|---|---|---|
| $0<b<\tfrac43$ | unstable node | saddle | saddle | stable node in the quadrant |
| $b=\tfrac43$ | unstable node | saddle | non-hyperbolic | merged with $(3,0)$ |
| $b>\tfrac43$ | unstable node | saddle | stable node | outside the quadrant or none |
The threshold is $b_c=\tfrac43$, the value at which $y$ stops being able to invade the $x$-only state: $4-3b_c=0$, i.e. $b_c=r_y/K_x$ (growth rate of $y$ over the carrying capacity of $x$).
- $D\mathbf F=\begin{pmatrix}6-4x-y&-x\\-by&4-2y-bx\end{pmatrix}$.
- $(0,0)$: $\operatorname{diag}(6,4)$, unstable node for all $b$.
- $(0,4)$: $\begin{pmatrix}2&0\\-4b&-4\end{pmatrix}$, eigenvalues $2,-4$: saddle for all $b$.
- $(3,0)$: $\begin{pmatrix}-6&-3\\0&4-3b\end{pmatrix}$: saddle for $b<\tfrac43$, stable node for $b>\tfrac43$, non-hyperbolic at $b=\tfrac43$.
- Coexistence ($0<b<\tfrac43$): $\begin{pmatrix}-2x^*&-x^*\\-by^*&-y^*\end{pmatrix}$, $\operatorname{tr}=-6$, $\det=\dfrac{2(8-6b)}{2-b}>0$, discriminant $>0$: stable node.
- Threshold $b_c=\tfrac43$.
Part (c) · 1 point
Phase portrait and long-time behaviour for b = 1
Step 1. Eigen-data at (2, 2)
$x^*=\frac{2}{1}=2$, $y^*=\frac{2}{1}=2$. $D\mathbf F=\begin{pmatrix}-4&-2\\-2&-2\end{pmatrix}$, $p(\lambda)=(-4-\lambda)(-2-\lambda)-4=\lambda^2+6\lambda+4$, $\lambda=-3\pm\sqrt5$. Row 1 of $(D\mathbf F-\lambda I)\mathbf v=\mathbf 0$: $(-4-\lambda)v_1-2v_2=0$, $v_2=-\tfrac{4+\lambda}{2}v_1$. For $\lambda_s=-3+\sqrt5$: $v_2=-\frac{1+\sqrt5}{2}v_1$, $\mathbf v_s=(2,-1-\sqrt5)$; for $\lambda_f=-3-\sqrt5$: $v_2=\frac{\sqrt5-1}{2}v_1$, $\mathbf v_f=(2,\sqrt5-1)$. Orbits reach $(2,2)$ tangent to $\mathbf v_s$ (slope $-\frac{1+\sqrt5}{2}\approx-1.62$), except the two on the fast line.
Step 2. Saddles
At $(3,0)$, $b=1$: $\begin{pmatrix}-6&-3\\0&1\end{pmatrix}$; for $\lambda=1$, row 1: $-7v_1-3v_2=0$, $\mathbf v=(3,-7)$. Its branch pointing into the quadrant, along $(-3,7)$, leaves $(3,0)$ upwards to the left and runs to $(2,2)$. At $(0,4)$: $\begin{pmatrix}2&0\\-4&-4\end{pmatrix}$, unstable direction $(3,-2)$: its branch in the quadrant leaves $(0,4)$ to the lower right and runs to $(2,2)$. The axes are the stable manifolds of the two saddles.
Step 3. Nullclines and regions
$L_x$: $2x+y=6$ (where $x'=0$, vertical arrows), from $(0,6)$ to $(3,0)$. $L_y$: $x+y=4$ (where $y'=0$, horizontal arrows), from $(0,4)$ to $(4,0)$. They cross at $(2,2)$.
- Above both lines: $x'<0$, $y'<0$ (towards the lower left).
- Below both lines: $x'>0$, $y'>0$ (towards the upper right).
- Between them for $x<2$ (e.g. $(1,3.5)$): $x'>0$, $y'<0$.
- Between them for $x>2$ (e.g. $(3,0.5)$): $x'<0$, $y'>0$.
Step 4. Trapping regions and the long run
Take the triangle $R=\{x+y>4,\ 2x+y<6,\ x<2\}$. On its side on $L_x$, $x'=0$ and $y'=y(4-x-y)=y(x-2)<0$ (there $y=6-2x$, so $4-x-y=x-2$): the field points down, into $R$. On its side on $L_y$, $y'=0$ and $x'=x(6-2x-y)=x(2-x)>0$: the field points right, into $R$. So $R$ is positively invariant; inside it $x$ increases and $y$ decreases, both bounded, so they converge, and the limit must be an equilibrium: $(2,2)$. The other triangle is treated the same way. An orbit starting above or below both lines moves monotonically (lower left or upper right) until it either enters one of the triangles or converges directly to $(2,2)$. Hence every orbit with $x(0)>0$, $y(0)>0$ tends to $(2,2)$: the two species coexist at 2000 individuals each.
- Coexistence point $(2,2)$; $D\mathbf F=\begin{pmatrix}-4&-2\\-2&-2\end{pmatrix}$, $\lambda=-3\pm\sqrt5\approx-0.764,\ -5.236$.
- Slow eigenvector $(2,\,-1-\sqrt5)$, fast eigenvector $(2,\,\sqrt5-1)$.
- Saddles: $(3,0)$ with unstable direction $(3,-7)$ (eigenvalue 1); $(0,4)$ with unstable direction $(3,-2)$ (eigenvalue 2).
- Every orbit with $x(0)>0$, $y(0)>0$ tends to $(2,2)$: coexistence.
Part (d) · 1 point
Phase portrait and long-time behaviour for b = 2
Step 1. Equilibria and linearisation
$2x+y=6$ and $2x+y=4$ never meet, so only $(0,0)$, $(3,0)$, $(0,4)$ remain. At $(3,0)$, $4-3b=-2$: eigenvalues $-6$ and $-2$, a stable node; for $\lambda=-2$, row 1 of $(D\mathbf F+2I)\mathbf v=\mathbf 0$ is $-4v_1-3v_2=0$, so $\mathbf v=(3,-4)$, the slow direction to which orbits are tangent on arrival. At $(0,4)$, $\begin{pmatrix}2&0\\-8&-4\end{pmatrix}$: a saddle, stable manifold the $y$ axis, unstable direction from row 2 $-8v_1-6v_2=0$, $\mathbf v=(3,-4)$.
Step 2. Regions
- $2x+y>6$: $x'<0$ and $y'<0$ (since also $2x+y>4$).
- $4<2x+y<6$: $x'>0$, $y'<0$ (e.g. $(1.5,2)$: $x'=1.5\cdot1>0$, $y'=2\cdot(-1)<0$).
- $2x+y<4$: $x'>0$, $y'>0$.
On $2x+y=6$ the field is vertical and points down, on $2x+y=4$ it is horizontal and points right: both point into the strip, which is positively invariant. In the strip $x$ increases and $y$ decreases, both bounded, so the orbit converges to an equilibrium in the closure of the strip with $x>0$: $(3,0)$. Orbits outside the strip move monotonically into it or straight to $(3,0)$.
Step 3. Long run
Every orbit with $x(0)>0$ (and any $y(0)\ge0$) tends to $(3,0)$: species $y$ becomes extinct and $x$ reaches its carrying capacity. Only the orbits on the $y$ axis ($x(0)=0$) go to $(0,4)$. The unstable branch of the saddle $(0,4)$ runs to $(3,0)$, leaving along $(3,-4)$.
- Equilibria $(0,0)$, $(3,0)$, $(0,4)$ only; nullclines $2x+y=6$ and $2x+y=4$, parallel.
- $(3,0)$: $\begin{pmatrix}-6&-3\\0&-2\end{pmatrix}$, stable node, slow eigenvector $(3,-4)$.
- $(0,4)$: $\begin{pmatrix}2&0\\-8&-4\end{pmatrix}$, saddle, unstable eigenvector $(3,-4)$.
- Every orbit with $x(0)>0$ tends to $(3,0)$: $y$ dies out, competitive exclusion.
Interpretation and checks
Phase 4. Asymptotic and physical checks
- Limits: $b\to0^+$: $x^*\to1$, $y^*\to4$; $y$ no longer feels $x$, and $x$ settles at $6-2x-4=0$. $b\to\tfrac43^-$: $(x^*,y^*)\to(3,0)$ and $\det D\mathbf F\to0$, the exchange of stability with $(3,0)$ (a transcritical bifurcation, extension: a bifurcation where two equilibria cross and swap their stability). For every $b>\tfrac43$ the outcome is that of (d).
- Checks: $\mathbf F=\mathbf 0$ at each equilibrium; triangular Jacobians read directly; $\operatorname{tr}D\mathbf F(x^*,y^*)=-6$ for every $b$; at $b=1$, $\lambda_s+\lambda_f=-6$ and $\lambda_s\lambda_f=9-5=4=\det$. Units: each bracket has units of year$^{-1}$ when the coefficients carry year$^{-1}$ and (thousand individuals)$^{-1}$ year$^{-1}$.
- Common mistakes in the exam: (1) keeping the coexistence point for $\tfrac43<b<2$ although $y^*<0$; (2) using the full Jacobian formula at $(x^*,y^*)$ without the simplification $6-2x^*-y^*=0$, which hides that $\operatorname{tr}=-6$; (3) concluding stability at $b=\tfrac43$ from the linearisation (zero eigenvalue); (4) predicting the long-time outcome from local stability alone, without the nullcline and trapping-region argument.
P09 — Conservative double-well oscillator x'' = 4x - x^3: linearisation, first integral, true centres, the figure-eight separatrix and the orbit types by energy
A particle of unit mass moves on a line under the force $4x-x^3$:
(a) Write the equation as a first-order system for $(x,y)$ with $y=x'$, and find the potential $V(x)$ with $x''=-V'(x)$ and $V(0)=0$. Find the equilibria and classify them by linearisation. Which conclusions can be transferred to the nonlinear system?
(b) Prove, by differentiating along solutions, that $E(x,y)=\tfrac12y^2+V(x)$ is constant on every orbit. Use it to show that the equilibria that are linear centres are true (nonlinear) centres, and give the sense of rotation around them.
(c) Find the explicit equation of the separatrix, the level set of $E$ through the saddle, and its extent in $x$ and $y$. Verify that $x(t)=2\sqrt2\,\operatorname{sech}(2t)$ solves the equation with $x(0)=2\sqrt2$, $x'(0)=0$, and explain why this solution never reaches the saddle in finite time.
(d) Determine the ranges of $E$ for which the orbits are: equilibria, closed orbits inside one well, the separatrix, closed orbits around all three equilibria. Give the turning points as functions of $E$ and describe the phase portrait.
Setup and canonical formulation
Phase 1. Hypotheses and setup
- Unknown $x(t)$, $t\in\mathbb R$. Second-order, autonomous, nonlinear (cubic term), no damping. The right-hand side is $C^1$, so solutions exist and are unique (Picard).
- The force does not depend on $x'$: the system is conservative (Robinson Sec. 34.1), and the energy will organise the whole portrait.
- Symmetries: the force is odd, so $V$ is even and the portrait is symmetric under $x\mapsto-x$; replacing $(y,t)$ by $(-y,-t)$ leaves the equation unchanged, so the portrait is symmetric under $y\mapsto-y$ with arrows reversed.
Phase 2. Canonical formulation
Part (a) · 1.5 points
System, potential, equilibria and linearisation
Step 1. Potential
using Barrow's rule. Check: $V'(x)=x^3-4x=-(4x-x^3)$. It is a double well: $V$ has a local maximum $V(0)=0$ and two minima $V(\pm2)=4-8=-4$.
Step 2. Equilibria
$y=0$ and $4x-x^3=x(2-x)(2+x)=0$: $(0,0)$, $(2,0)$, $(-2,0)$, the critical points of $V$.
Step 3. Linearisation
$D\mathbf F=\begin{pmatrix}0&1\\4-3x^2&0\end{pmatrix}$, with $V''(x)=3x^2-4$.
- At $(0,0)$: $\begin{pmatrix}0&1\\4&0\end{pmatrix}$, $\det(D\mathbf F-\lambda I)=\lambda^2-4$, $\lambda=\pm2$. For $\lambda=2$, row 1: $-2v_1+v_2=0$, $\mathbf v=(1,2)$; for $\lambda=-2$: $2v_1+v_2=0$, $\mathbf v=(1,-2)$. Real eigenvalues of opposite sign: a saddle.
- At $(\pm2,0)$: $4-12=-8$, $\begin{pmatrix}0&1\\-8&0\end{pmatrix}$, $\lambda^2+8=0$, $\lambda=\pm2\sqrt2\,i$: a linear centre, with small-oscillation period $2\pi/(2\sqrt2)=\pi/\sqrt2$.
Step 4. What transfers
The saddle is hyperbolic, so by the Hartman-Grobman theorem (Robinson Sec. 32.4.3) the nonlinear system has a saddle at the origin with stable and unstable curves tangent to $(1,-2)$ and $(1,2)$. The centres are not hyperbolic (zero real part): a small nonlinear term could turn them into weak foci, so the linearisation alone decides nothing. Part (b) settles it.
- $V=\dfrac{x^4}{4}-2x^2$, $V(0)=0$, $V(\pm2)=-4$.
- Equilibria $(0,0)$, $(2,0)$, $(-2,0)$.
- $(0,0)$: $\begin{pmatrix}0&1\\4&0\end{pmatrix}$, $\lambda=\pm2$, eigenvectors $(1,2)$ (unstable), $(1,-2)$ (stable): saddle, also for the nonlinear system.
- $(\pm2,0)$: $\begin{pmatrix}0&1\\-8&0\end{pmatrix}$, $\lambda=\pm2\sqrt2\,i$: linear centres, non-hyperbolic, linearisation inconclusive.
Part (b) · 1.5 points
First integral and true centres
Step 1. Conservation
Along a solution, by the chain rule,
Explicitly: $(x^3-4x)y+y(4x-x^3)=0$. So $E$ is a first integral and every orbit lies on one level set $\tfrac12y^2+\tfrac{x^4}{4}-2x^2=E_0$.
Step 2. Shape of E near the centres
$E_x=V'(x)$, $E_y=y$ vanish at $(2,0)$, and the Hessian there is $\begin{pmatrix}V''(2)&0\\0&1\end{pmatrix}=\begin{pmatrix}8&0\\0&1\end{pmatrix}$, positive definite: $E$ has a strict local minimum $E(2,0)=-4$. Near it, $E\approx-4+4(x-2)^2+\tfrac12y^2$, so for $c$ slightly above $-4$ the level set $E=c$ is a small closed oval around $(2,0)$.
Step 3. Orbits on those level sets are periodic
More precisely, for $-4<c<0$ the right-well part of $\{E=c\}$ is $y=\pm\sqrt{2(c-V(x))}$ for $x\in[x_-,x_+]$, the interval where $V\le c$ (found in (d)): one closed curve, symmetric in $y$, meeting $y=0$ only at the turning points. It contains no equilibrium ($V'(x_\pm)\neq0$), so the speed never vanishes on it and the solution goes round it in a finite time and repeats: every nearby orbit is closed. The equilibria $(\pm2,0)$ are therefore true centres of the nonlinear system (Robinson Sec. 34.1), by symmetry also $(-2,0)$.
Step 4. Sense of rotation
At $(2,1)$, above the centre, $(x',y')=(1,\ 4\cdot2-8)=(1,0)$: the motion is to the right on top, so the orbits are travelled clockwise, as for every conservative system written with $y=x'$ ($x$ increases where $y>0$).
- $\dfrac{dE}{dt}=V'(x)\,y+y\,(-V'(x))=0$.
- $E$ has a strict local minimum $-4$ at $(\pm2,0)$ (Hessian $\operatorname{diag}(8,1)$): level sets $E=c$, $-4<c<0$, are closed curves around each minimum, so the orbits are periodic: true centres.
- Rotation clockwise (at $(2,1)$: $x'=1>0$).
Part (c) · 1 point
Separatrix and the homoclinic solution
Step 1. Level set through the saddle
$E(0,0)=0$. Then $\tfrac12y^2+\tfrac{x^4}{4}-2x^2=0$ gives
Real only for $x^2\le8$, i.e. $\lvert x\rvert\le2\sqrt2$, where $V(\pm2\sqrt2)=\frac{64}{4}-16=0$. Near $x=0$, $y\approx\pm x\sqrt4=\pm2x$: the separatrix leaves the origin along the eigenvectors $(1,\pm2)$ found in (a). It is a figure eight: one loop in $x>0$ around $(2,0)$, its mirror image in $x<0$.
Step 2. Height of the loops
$g(x)=\frac{x^2(8-x^2)}{2}=4x^2-\frac{x^4}{2}$, $g'(x)=8x-2x^3=2x(4-x^2)=0$ at $x=\pm2$, with $g(2)=16-8=8$. So $\lvert y\rvert\le2\sqrt2$, reached above and below the centres.
Step 3. Verification of the homoclinic solution
Let $x=2\sqrt2\operatorname{sech}u$ with $u=2t$, $\frac{du}{dt}=2$. By the chain rule
Product rule, with $\frac{d}{dt}\tanh u=2\operatorname{sech}^2u$:
With $\tanh^2u=1-\operatorname{sech}^2u$: $x''=8\sqrt2\operatorname{sech}u-16\sqrt2\operatorname{sech}^3u$. On the other side, $4x-x^3=8\sqrt2\operatorname{sech}u-(2\sqrt2)^3\operatorname{sech}^3u=8\sqrt2\operatorname{sech}u-16\sqrt2\operatorname{sech}^3u$. They agree. Initial data: $\operatorname{sech}0=1$, $\tanh0=0$, so $x(0)=2\sqrt2$ and $x'(0)=0$; $E=0+V(2\sqrt2)=0$, the solution is on the separatrix.
Step 4. Infinite time
$\operatorname{sech}2t>0$ for every real $t$ and $\operatorname{sech}2t\to0$ only as $t\to\pm\infty$: the solution leaves the origin at $t=-\infty$ along $(1,2)$, reaches the turning point $x=2\sqrt2$ at $t=0$ and returns to the origin at $t=+\infty$ along $(1,-2)$. It is a homoclinic orbit. It cannot reach $(0,0)$ in finite time, because $(0,0)$ is itself a solution and uniqueness forbids two solutions to meet. Its speed is largest when $\operatorname{sech}^2u=\tfrac12$, with $\lvert x'\rvert=4\sqrt2\cdot\tfrac{1}{\sqrt2}\cdot\tfrac{1}{\sqrt2}=2\sqrt2$, as in Step 2.
- Separatrix $E=0$: $y^2=\dfrac{x^2(8-x^2)}{2}$, i.e. $y=\pm x\sqrt{\dfrac{8-x^2}{2}}$, for $\lvert x\rvert\le2\sqrt2$: a figure eight through the origin.
- Tangents at the origin $y=\pm2x$ (the saddle eigenvectors); maximum $\lvert y\rvert=2\sqrt2$ at $x=\pm2$.
- $x(t)=2\sqrt2\operatorname{sech}2t$: $x''=8\sqrt2\operatorname{sech}2t-16\sqrt2\operatorname{sech}^32t=4x-x^3$; $x\to0$ only as $t\to\pm\infty$.
Part (d) · 1 point
Energy ranges, turning points and portrait
Step 1. Admissible energies
$E=\tfrac12y^2+V(x)\ge V(x)\ge-4$. So $E<-4$ gives an empty level set, and $E=-4$ forces $y=0$, $V(x)=-4$: the two centres.
Step 2. Turning points
At a turning point $y=0$ and $V(x)=E$: $\frac{x^4}{4}-2x^2-E=0$, i.e. $X^2-8X-4E=0$ with $X=x^2$. The quadratic formula gives
- For $-4<E<0$: $0<2\sqrt{4+E}<4$, so both roots are positive and $x^2\in[\,4-2\sqrt{4+E},\ 4+2\sqrt{4+E}\,]$ gives two disjoint intervals of $x$, one in each well: two closed orbits (one around each centre).
- For $E>0$: $4-2\sqrt{4+E}<0$ is rejected; the only turning points are $x=\pm\sqrt{4+2\sqrt{4+E}}$, and the level set is one closed curve surrounding all three equilibria, crossing $x=0$ where $\tfrac12y^2=E$.
Step 3. Portrait
Inside each loop of the figure eight there is a family of nested clockwise ovals around $(\pm2,0)$, shrinking to the centre as $E\to-4^+$ and growing to the loop as $E\to0^-$ (their period tends to infinity there). Outside the figure eight there are large clockwise closed curves, peanut-shaped (pinched at $x=0$) for $E$ slightly above $0$, and close to the quartic curves $\tfrac12y^2+\tfrac{x^4}{4}=E$ for large $E$. Every orbit except the equilibria and the separatrix is periodic; there are no attractors, as expected without damping.
- $E<-4$: no orbit; $E=-4$: the centres $(\pm2,0)$.
- $-4<E<0$: two closed orbits, one per well, between $x^2=4-2\sqrt{4+E}$ and $x^2=4+2\sqrt{4+E}$ (e.g. $E=-3$: $\sqrt2\le\lvert x\rvert\le\sqrt6$).
- $E=0$: the saddle and the two homoclinic loops.
- $E>0$: one closed orbit around all three equilibria, turning points $x=\pm\sqrt{4+2\sqrt{4+E}}$, crossing $x=0$ with $y=\pm\sqrt{2E}$ (e.g. $E=5$: $x=\pm\sqrt{10}$, $y=\pm\sqrt{10}$).
Interpretation and checks
Phase 4. Asymptotic and physical checks
- Limits: $E\to-4^+$: small ellipses $4(x-2)^2+\tfrac12y^2=E+4$, period $\to\pi/\sqrt2$ (the linear value). $E\to0$: the period grows without bound (the orbit lingers near the saddle). $E\to\infty$: $V\approx x^4/4$ dominates and the amplitude grows like $(4E)^{1/4}$.
- Checks: $dE/dt=0$ identically; the separatrix leaves the origin with slopes $\pm2$, matching the saddle eigenvectors; the homoclinic solution has $E=0$ and maximum speed $2\sqrt2$, matching the height of the loop. Dimensions, if $x$ is in metres and $t$ in seconds: the coefficients 4 and 1 carry s$^{-2}$ and m$^{-2}$s$^{-2}$, and $E$ is an energy per unit mass, m$^2$s$^{-2}$.
- Common mistakes in the exam: (1) stating that the linear centres are centres of the nonlinear system "by Hartman-Grobman": that theorem does not apply to non-hyperbolic points, the first integral is needed; (2) a sign error in $V$ (taking $V'=4x-x^3$), which turns minima into maxima and swaps saddles and centres; (3) forgetting the root $X=4-2\sqrt{4+E}$ and describing one orbit instead of two for $-4<E<0$; (4) drawing the separatrix as crossing itself transversally at angles other than those of the eigenvectors $(1,\pm2)$.
P10 — Full Fourier series on (-pi, pi) of the function equal to 0 on the left half and x^2 on the right half: coefficients, Dirichlet values and two numerical series
Let $f$ be defined on $(-\pi,\pi)$ by
and extended to $\mathbb R$ as a $2\pi$-periodic function.
(a) Describe the graph of the periodic extension on $[-3\pi,3\pi]$: where is it continuous, where does it jump and by how much? Is $f$ piecewise smooth?
(b) Compute the Fourier series $\dfrac{a_0}{2}+\displaystyle\sum_{n\ge1}\big(a_n\cos nx+b_n\sin nx\big)$ of $f$.
(c) Using the Dirichlet theorem, give the sum of the series at $x=0$, $x=\pi$, $x=-\pi$, $x=\pi/2$ and $x=-\pi/2$.
(d) Deduce the sums of $\displaystyle\sum_{n=1}^\infty\frac{1}{n^2}$ and $\displaystyle\sum_{n=1}^\infty\frac{(-1)^{n+1}}{n^2}$, and check that the two results are consistent with each other.
Setup and canonical formulation
Phase 1. Hypotheses and setup
- Unknowns: the coefficients $a_0$, $a_n$, $b_n$ ($n\ge1$). Independent variable $x\in\mathbb R$, period $2\pi$ ($L=\pi$).
- $f$ is bounded, and on each of $(-\pi,0)$ and $(0,\pi)$ it is a polynomial: $f$ and $f'$ are piecewise continuous, so $f$ is piecewise smooth and the Dirichlet theorem applies.
- No symmetry: $f$ is neither even nor odd, so both $a_n$ and $b_n$ are needed. Since $f=0$ on $(-\pi,0)$, every integral reduces to $(0,\pi)$.
Phase 2. Canonical formulation
Euler formulas on $(-\pi,\pi)$:
Dirichlet theorem: for a piecewise smooth $2\pi$-periodic $f$, the series converges at every $x$ to $\tfrac12\big[f(x^+)+f(x^-)\big]$.
Part (a) · 1 point
Periodic extension, continuity and jumps
Step 1. One period
On $(-\pi,0)$ the graph is the segment $y=0$; on $[0,\pi)$ it is the arc of $y=x^2$, rising from $(0,0)$ to height $\pi^2$ as $x\to\pi^-$.
Step 2. Continuity at the origin
$f(0^-)=0$, $f(0^+)=0^2=0$: continuous. The one-sided derivatives are $0$ (left) and $2\cdot0=0$ (right): the graph even has no corner at $x=0$; only $f''$ jumps there (from $0$ to $2$).
Step 3. Jumps at the odd multiples of pi
The extension satisfies $f(x+2\pi)=f(x)$. At $x=\pi$: $f(\pi^-)=\lim_{x\to\pi^-}x^2=\pi^2$, while $f(\pi^+)=f(-\pi^+)=0$. So the extension jumps down by $\pi^2$ at $x=\pm\pi,\pm3\pi$. The graph on $[-3\pi,3\pi]$ is three copies of the period: flat on $(-3\pi,-2\pi)$, arc on $(-2\pi,-\pi)$, flat on $(-\pi,0)$, arc on $(0,\pi)$, flat on $(\pi,2\pi)$, arc on $(2\pi,3\pi)$, with a vertical drop from $\pi^2$ to $0$ at each odd multiple of $\pi$.
- On each period: zero on $(-\pi,0)$, the parabola arc $x^2$ from $0$ up to $\pi^2$ on $(0,\pi)$.
- Continuous at $x=2k\pi$ ($f(0^-)=f(0^+)=0$, and even $f'(0^\pm)=0$).
- Jumps at $x=(2k+1)\pi$: $f(\pi^-)=\pi^2$, $f(\pi^+)=0$, jump of size $-\pi^2$.
- Piecewise smooth: yes.
Part (b) · 2 points
Fourier coefficients
Step 1. Mean value
Step 2. Two auxiliary integrals ($n\ge1$)
By parts with $u=x$, $dv=\sin nx\,dx$, $du=dx$, $v=-\frac{\cos nx}{n}$:
By parts with $u=x$, $dv=\cos nx\,dx$, $du=dx$, $v=\frac{\sin nx}{n}$:
Step 3. Cosine coefficients
By parts with $u=x^2$, $dv=\cos nx\,dx$, $du=2x\,dx$, $v=\frac{\sin nx}{n}$, and then Step 2:
Barrow's rule on $[0,\pi]$, with $\sin n\pi=\sin0=0$ and $\cos n\pi=(-1)^n$:
Step 4. Sine coefficients
By parts with $u=x^2$, $dv=\sin nx\,dx$, $du=2x\,dx$, $v=-\frac{\cos nx}{n}$, and then Step 2:
Barrow's rule on $[0,\pi]$:
Dividing by $\pi$:
Step 5. The series
First terms: $\frac{\pi^2}{6}-2\cos x+\big(\pi-\frac4\pi\big)\sin x+\frac12\cos2x-\frac\pi2\sin2x-\frac29\cos3x+\dots$
- $a_0=\dfrac{\pi^2}{3}$ (so the constant term is $\dfrac{\pi^2}{6}$).
- $a_n=\dfrac{2(-1)^n}{n^2}$: $a_1=-2$, $a_2=\tfrac12$, $a_3=-\tfrac29$.
- $b_n=\dfrac{(-1)^{n+1}\pi}{n}+\dfrac{2\big((-1)^n-1\big)}{\pi n^3}$: $b_n=\dfrac{\pi}{n}-\dfrac{4}{\pi n^3}$ for $n$ odd, $b_n=-\dfrac{\pi}{n}$ for $n$ even.
Part (c) · 1 point
Values of the series by the Dirichlet theorem
Step 1. Points of continuity
$x=0$: $f$ is continuous there, sum $f(0)=0$. $x=\pi/2$: inside $(0,\pi)$, sum $(\pi/2)^2=\pi^2/4$. $x=-\pi/2$: inside $(-\pi,0)$, sum $0$.
Step 2. The endpoints are jump points
For the periodic extension, $x=\pi$ and $x=-\pi$ are the same kind of point: $f(\pi^-)=\pi^2$, $f(\pi^+)=f(-\pi^+)=0$. The series converges to
the same value at $x=-\pi$ (each term is $2\pi$-periodic, so the series takes equal values at $\pm\pi$). It is not $f(\pi)$ of either formula: the series does not "choose" a side.
- $x=0$: $0$. $\ x=\pm\pi$: $\dfrac{\pi^2}{2}$. $\ x=\pi/2$: $\dfrac{\pi^2}{4}$. $\ x=-\pi/2$: $0$.
Part (d) · 1 point
Two numerical series
Step 1. Evaluation at x = 0
$\cos0=1$, $\sin0=0$, and by (c) the sum is $0$:
Step 2. Evaluation at x = pi
$\cos n\pi=(-1)^n$, $\sin n\pi=0$, and by (c) the sum is $\pi^2/2$:
Step 3. Consistency check
Subtracting the two series, the odd terms cancel and the even ones double: $\sum\frac1{n^2}-\sum\frac{(-1)^{n+1}}{n^2}=2\sum_{m\ge1}\frac{1}{(2m)^2}=\frac12\sum_{m\ge1}\frac1{m^2}$. With the values found: $\frac{\pi^2}{6}-\frac{\pi^2}{12}=\frac{\pi^2}{12}$ and $\frac12\cdot\frac{\pi^2}{6}=\frac{\pi^2}{12}$. They agree.
- $x=0$: $0=\dfrac{\pi^2}{6}+2\displaystyle\sum\frac{(-1)^n}{n^2}$, so $\displaystyle\sum_{n=1}^\infty\frac{(-1)^{n+1}}{n^2}=\frac{\pi^2}{12}$.
- $x=\pi$: $\dfrac{\pi^2}{2}=\dfrac{\pi^2}{6}+2\displaystyle\sum\frac{1}{n^2}$, so $\displaystyle\sum_{n=1}^\infty\frac{1}{n^2}=\frac{\pi^2}{6}$.
- Consistency: $\dfrac{\pi^2}{6}-\dfrac{\pi^2}{12}=2\displaystyle\sum_{n\text{ even}}\frac1{n^2}=\frac12\cdot\frac{\pi^2}{6}$.
Interpretation and checks
Phase 4. Asymptotic and physical checks
- Limits: $a_n=O(1/n^2)$ because $f$ is continuous at $0$, while $b_n\sim(-1)^{n+1}\pi/n$ decays only like $1/n$: that slow part is produced by the jump of size $\pi^2$ at $x=\pm\pi$ (a jump of size $J$ at $x=\pi$ always gives a contribution $\frac{J}{\pi}\frac{(-1)^{n}}{n}$ with $J=-\pi^2$ here). Near the jump the partial sums overshoot by about 9% of the jump (Gibbs phenomenon, extension: the overshoot of partial sums near a jump, which does not vanish as $N\to\infty$).
- Checks: the antiderivatives of Steps 3 and 4 give back $x^2\cos nx$ and $x^2\sin nx$ when differentiated; partial sums with 4000 terms reproduce $0$ at $x=0$, $\pi^2/2$ at $x=\pi$ and $\pi^2/4$ at $x=\pi/2$ to $10^{-3}$; the mean value $a_0/2=\pi^2/6$ equals $\frac{1}{2\pi}\int_{-\pi}^{\pi}f\,dx$.
- Common mistakes in the exam: (1) dividing by $2\pi$ instead of $\pi$ in $a_n$, or forgetting that the constant term is $a_0/2$; (2) evaluating the series at $x=\pi$ as $f(\pi^-)=\pi^2$ instead of the midpoint $\pi^2/2$, which gives $\sum1/n^2=5\pi^2/12$ (wrong); (3) dropping the term $2((-1)^n-1)/(\pi n^3)$ of $b_n$, which only lives on odd $n$; (4) treating $f$ as even or odd and discarding one family of coefficients.
P11 — Half-range sine and cosine series of f(x) = x^2 on (0, pi): continuity of the extensions, decay of the coefficients, values of the series and three numerical sums
Let $f(x)=x^2$ for $0<x<\pi$.
(a) Compute the Fourier sine series of $f$ on $(0,\pi)$, writing every integration by parts. Give $b_n$ separately for odd and even $n$.
(b) Compute the Fourier cosine series of $f$ on $(0,\pi)$.
(c) Which of the two periodic extensions (odd or even, of period $2\pi$) is continuous on $\mathbb R$? Explain why this decides how fast the coefficients of (a) and (b) decay. Then give, with justification, the value of each series at $x=\pi$, $x=-\pi/2$ and $x=5\pi/2$.
(d) Use the series to obtain
You may use the Leibniz sum $\displaystyle\sum_{k=0}^{\infty}\frac{(-1)^k}{2k+1}=\frac{\pi}{4}$.
Setup and canonical formulation
Phase 1. Hypotheses and setup
- Data: $f(x)=x^2$ on the half-period $(0,L)$ with $L=\pi$. $f$ is smooth on $[0,\pi]$ with $f(0)=0$, $f(\pi)=\pi^2$, $f'(x)=2x$, so $f'(0)=0$, $f'(\pi)=2\pi$.
- Unknowns: the coefficients $b_n$ (sine series) and $a_0,a_n$ (cosine series), and the pointwise sums of both series on all of $\mathbb R$.
- Regularity: $f$ and $f'$ are continuous on $[0,\pi]$, so both extensions are piecewise smooth and the convergence theorem (Haberman Sec. 3.2) applies: each series converges to the extension where it is continuous and to the average of the side limits at a jump.
Phase 2. Canonical formulation
With $L=\pi$ (Haberman Sec. 3.3):
The sine series represents the odd $2\pi$-periodic extension $f_o$ of $f$; the cosine series represents the even $2\pi$-periodic extension $f_e$.
Part (a) · 1.5 points
Compute the Fourier sine series of f on (0, pi)
Step 1. First integration by parts
Take $u=x^2$, $dv=\sin nx\,dx$, so $du=2x\,dx$, $v=-\dfrac{\cos nx}{n}$:
Step 2. Second integration by parts
Take $u=x$, $dv=\cos nx\,dx$, so $du=dx$, $v=\dfrac{\sin nx}{n}$:
Hence the antiderivative
Check: $F'(x)=-\dfrac{2x\cos nx}{n}+x^2\sin nx+\dfrac{2\sin nx}{n^2}+\dfrac{2x\cos nx}{n}-\dfrac{2\sin nx}{n^2}=x^2\sin nx$.
Step 3. Barrow's rule
$F(\pi)=-\dfrac{\pi^2(-1)^n}{n}+0+\dfrac{2(-1)^n}{n^3}$ and $F(0)=0+0+\dfrac{2}{n^3}$, so
Multiplying by $2/\pi$:
Step 4. Odd and even indices
For odd $n$, $(-1)^{n+1}=1$ and $(-1)^n-1=-2$: $b_n=\dfrac{2\pi}{n}-\dfrac{8}{\pi n^3}$. For even $n$, $(-1)^{n+1}=-1$ and $(-1)^n-1=0$: $b_n=-\dfrac{2\pi}{n}$. The leading term is of order $1/n$ for every $n$.
- $\displaystyle b_n=\frac{2\pi(-1)^{n+1}}{n}+\frac{4\big((-1)^n-1\big)}{\pi n^3}$.
- Odd $n$: $b_n=\dfrac{2\pi}{n}-\dfrac{8}{\pi n^3}$; even $n$: $b_n=-\dfrac{2\pi}{n}$. For example $b_1=2\pi-\dfrac8\pi$, $b_2=-\pi$.
- $\displaystyle x^2=\sum_{n=1}^{\infty}\Big[\frac{2\pi(-1)^{n+1}}{n}+\frac{4((-1)^n-1)}{\pi n^3}\Big]\sin nx$ for $0<x<\pi$.
Part (b) · 1 point
Compute the Fourier cosine series of f on (0, pi)
Step 1. Mean value
Step 2. Antiderivative
With $u=x^2$, $dv=\cos nx\,dx$ ($du=2x\,dx$, $v=\sin nx/n$), then $u=x$, $dv=\sin nx\,dx$ ($du=dx$, $v=-\cos nx/n$):
Step 3. Barrow's rule
$G(\pi)=0+\dfrac{2\pi(-1)^n}{n^2}-0$, $G(0)=0$. Therefore
- $a_0=\dfrac{2\pi^2}{3}$, so the constant term is $\dfrac{a_0}{2}=\dfrac{\pi^2}{3}$.
- $a_n=\dfrac{4(-1)^n}{n^2}$ for $n\ge1$.
- $\displaystyle x^2=\frac{\pi^2}{3}+4\sum_{n=1}^{\infty}\frac{(-1)^n}{n^2}\cos nx$ for $0\le x\le\pi$.
Part (c) · 1.25 points
Continuity of the extensions, decay of the coefficients and values of the series
Step 1. The odd extension
$f_o(x)=x^2$ on $(0,\pi)$, $f_o(x)=-x^2$ on $(-\pi,0)$, repeated with period $2\pi$. At $x=0$ both side limits are $0$: continuous. At $x=\pi$: $f_o(\pi^-)=\pi^2$, while $f_o(\pi^+)=f_o(-\pi^+)=-(\pi)^2=-\pi^2$. Since $f(\pi)\ne0$, the odd extension has a jump of size $2\pi^2$ at every odd multiple of $\pi$.
Step 2. The even extension
$f_e(x)=x^2$ on $[-\pi,\pi]$, repeated with period $2\pi$. At $x=\pi$: $f_e(\pi^-)=\pi^2=f_e(-\pi^+)=f_e(\pi^+)$: continuous. The even extension of a continuous function is always continuous. Its derivative is $2x$ on $(-\pi,\pi)$: $f_e'(\pi^-)=2\pi$ and $f_e'(\pi^+)=f_e'(-\pi^+)=-2\pi$, a jump; at $x=0$, $f_e'(0^\pm)=0$, no jump.
Step 3. Why continuity controls the decay
In (a), the term of order $1/n$ came from the boundary value $-\dfrac{x^2\cos nx}{n}\Big|_0^{\pi}$, which is nonzero exactly because $f(\pi)\ne0$, i.e. because $f_o$ jumps. In (b) the corresponding boundary term $\dfrac{x^2\sin nx}{n}\Big|_0^{\pi}$ vanishes for any $f$; the first surviving term is $\Big[\dfrac{2x\cos nx}{n^2}\Big]_0^{\pi}=\dfrac{f'(\pi)(-1)^n-f'(0)}{n^2}$, which is nonzero because $f_e'$ jumps at $x=\pi$ ($f'(\pi)=2\pi\neq0$). Thus a jump in the extension forces $1/n$; a continuous extension with a jump in its derivative gives $1/n^2$. This matches Haberman Sec. 3.4: the cosine series of a continuous $f$ can be differentiated term by term (giving the sine series of $f'=2x$), while the sine series of $x^2$ cannot, because $f_o$ is not continuous.
Step 4. Values of the sine series
By the convergence theorem the sine series converges to $\tfrac12\big[f_o(x^+)+f_o(x^-)\big]$:
- $x=\pi$: $\tfrac12(\pi^2-\pi^2)=0$. Indeed every term $b_n\sin n\pi=0$.
- $x=-\pi/2$: $f_o$ continuous there, $f_o(-\pi/2)=-(\pi/2)^2=-\pi^2/4$.
- $x=5\pi/2=\pi/2+2\pi$: by periodicity $f_o(\pi/2)=\pi^2/4$.
Step 5. Values of the cosine series
$f_e$ is continuous, so the cosine series equals $f_e$ everywhere:
- $x=\pi$: $\pi^2$.
- $x=-\pi/2$: $f_e(-\pi/2)=(\pi/2)^2=\pi^2/4$.
- $x=5\pi/2$: $f_e(\pi/2)=\pi^2/4$.
- Odd extension $f_o$: discontinuous at $x=\pm\pi,\pm3\pi,\dots$ (jump from $\pi^2$ to $-\pi^2$), so $b_n=O(1/n)$.
- Even extension $f_e$: continuous everywhere; only $f_e'$ jumps (from $2\pi$ to $-2\pi$ at odd multiples of $\pi$), so $a_n=O(1/n^2)$.
- Sine series: $0$ at $x=\pi$, $-\pi^2/4$ at $x=-\pi/2$, $\pi^2/4$ at $x=5\pi/2$.
- Cosine series: $\pi^2$ at $x=\pi$, $\pi^2/4$ at $x=-\pi/2$, $\pi^2/4$ at $x=5\pi/2$.
Part (d) · 1.25 points
Numerical sums from the series
Step 1. Cosine series at x = pi
$\cos n\pi=(-1)^n$, so $(-1)^n\cos n\pi=1$ and
Step 2. Cosine series at x = 0
$\cos 0=1$ and $f_e(0)=0$:
Step 3. Sine series at x = pi/2
$x=\pi/2$ is a point of continuity of $f_o$, so the series equals $\pi^2/4$. Only odd $n=2k+1$ survive, with $\sin\big((2k+1)\pi/2\big)=(-1)^k$ and $b_{2k+1}=\dfrac{2\pi}{2k+1}-\dfrac{8}{\pi(2k+1)^3}$:
where $S_3$ is the requested sum. Splitting the series is allowed because both pieces converge. Then
- $\displaystyle\sum_{n\ge1}\frac1{n^2}=\frac{\pi^2}{6}$ (cosine series at $x=\pi$).
- $\displaystyle\sum_{n\ge1}\frac{(-1)^{n+1}}{n^2}=\frac{\pi^2}{12}$ (cosine series at $x=0$).
- $\displaystyle\sum_{k\ge0}\frac{(-1)^k}{(2k+1)^3}=\frac{\pi^3}{32}\approx0.968946$ (sine series at $x=\pi/2$).
Interpretation and checks
Phase 4. Asymptotic and physical checks
- Limits: the sine series converges to $0$ at $x=\pi$ although $f(\pi)=\pi^2$; near $x=\pi$ the partial sums overshoot (Gibbs phenomenon, extension: the overshoot of the partial sums near a jump does not vanish as more terms are added, it only moves closer to the jump). The cosine series converges uniformly, since $\sum 4/n^2<\infty$.
- Checks: $S_3=1-0.037037+0.008+\dots$: the first three terms give $0.970963$, already within $0.0021$ of $\pi^3/32$. Differentiating the cosine series term by term gives $2x=\sum\frac{4(-1)^{n+1}}{n}\sin nx$, the sine series of $2x$ (as Haberman Sec. 3.4 allows for a continuous even extension).
- Common mistakes in the exam: (1) using the general $a_n$ formula with $n=0$ (division by zero) instead of computing $a_0$ apart; (2) forgetting the factor $\tfrac12$ in $a_0/2$, which gives the wrong constant $2\pi^2/3$; (3) claiming the sine series equals $\pi^2$ at $x=\pi$; (4) evaluating at $x=-\pi/2$ with $f$ instead of the extension (the sine series gives $-\pi^2/4$, not $+\pi^2/4$).
P12 — Heat equation with a constant source and fixed, different end temperatures: steady state, separation of variables for the deviation, series solution and the time constant of the slowest mode
A thin laterally insulated rod occupies $0\le x\le L$. Its ends are held at the constant temperatures $u(0,t)=T_A$ and $u(L,t)=T_B$. For $t<0$ the rod has been left long enough to reach the equilibrium with no internal heat source. At $t=0$ an electric current is switched on that heats the rod uniformly, so that for $t>0$ the temperature $u(x,t)$ satisfies
with constant diffusivity $k>0$ and constant source $Q>0$ (in K/s).
(a) Write the initial temperature $u(x,0)$. Find the steady state $u_s(x)$ of the problem for $t>0$ and the steady temperature at the midpoint.
(b) Write the problem satisfied by $v=u-u_s$. Separate variables in it, $v=X(x)T(t)$, and discuss the three cases of the sign of the separation constant to find all eigenvalues and eigenfunctions.
(c) Obtain $u(x,t)$ as a series with all coefficients in closed form.
(d) Find $\lim_{t\to\infty}u$, the time constant $\tau_1$ of the slowest mode, and, keeping only that mode, the time $t_{99}$ after which the midpoint has completed $99\%$ of its final rise. Numbers: $L=0.5$ m, $k=1.0\times10^{-4}$ m$^2$/s, $Q=0.02$ K/s, $T_A=20$ °C, $T_B=40$ °C.
Setup and canonical formulation
Phase 1. Hypotheses and setup
- Unknown $u(x,t)$, independent variables $x\in[0,L]$, $t\ge0$. One-dimensional conduction, constant properties, uniform volumetric source.
- Operator: linear, second order, parabolic; non-homogeneous twice: the source $Q$ in the equation and the Dirichlet data $T_A,T_B$ (non-zero). Separation of variables needs homogeneous boundary conditions, so the inhomogeneities are removed first with the steady state.
- Data: $L$ [m], $k$ [m$^2$/s], $Q$ [K/s], $T_A,T_B$ [°C or K].
Phase 2. Canonical formulation
Split $u(x,t)=u_s(x)+v(x,t)$, where $u_s$ absorbs the source and the boundary data and $v$ solves a homogeneous problem with homogeneous Dirichlet conditions (Haberman Sec. 7.2).
Part (a) · 1 point
Initial temperature and steady state
Step 1. Initial profile
Before the source is switched on the rod is in equilibrium, $u_t=0$, so $k u''=0$ and $u=\alpha+\beta x$. The end values give $\alpha=T_A$, $\alpha+\beta L=T_B$, hence $u_0(x)=T_A+(T_B-T_A)x/L$ (Haberman Sec. 1.4).
Step 2. Steady state with the source
$u_s''=-\dfrac{Q}{k}$. Integrating once: $u_s'=-\dfrac{Q}{k}x+C_1$; again: $u_s=-\dfrac{Q}{2k}x^2+C_1x+C_2$.
Step 3. Constants
$u_s(0)=C_2=T_A$. $u_s(L)=-\dfrac{QL^2}{2k}+C_1L+T_A=T_B$, so $C_1=\dfrac{T_B-T_A}{L}+\dfrac{QL}{2k}$. Substituting and grouping $\dfrac{Q}{2k}(Lx-x^2)$:
Check: $k u_s''=k\cdot\dfrac{Q}{2k}\cdot(-2)=-Q$, and $x(L-x)$ vanishes at both ends.
Step 4. Midpoint
$u_s(L/2)=T_A+\dfrac{T_B-T_A}{2}+\dfrac{Q}{2k}\cdot\dfrac{L}{2}\cdot\dfrac{L}{2}=\dfrac{T_A+T_B}{2}+\dfrac{QL^2}{8k}$. Numbers: $\dfrac{0.02\cdot0.25}{8\cdot10^{-4}}=6.25$ K, so $u_s(L/2)=36.25$ °C.
- $u_0(x)=T_A+(T_B-T_A)\dfrac{x}{L}$.
- $u_s(x)=T_A+(T_B-T_A)\dfrac{x}{L}+\dfrac{Q}{2k}\,x(L-x)$.
- $u_s(L/2)=\dfrac{T_A+T_B}{2}+\dfrac{QL^2}{8k}$; with the numbers $30+6.25=36.25$ °C.
Part (b) · 1.5 points
Problem for v and separation of variables
Step 1. Homogeneous problem
$v_t=u_t-0=k u_{xx}+Q-(k u_s''+Q)=k(u-u_s)_{xx}=k v_{xx}$. At the ends $v(0,t)=T_A-T_A=0$ and $v(L,t)=T_B-T_B=0$. Initially $v(x,0)=u_0-u_s=-\dfrac{Q}{2k}x(L-x)$: the linear parts cancel.
Step 2. Separation
$v=X(x)T(t)$ gives $XT'=kX''T$, and dividing by $kXT$:
The left side depends only on $t$ and the middle only on $x$, so both equal a constant. Hence $T'=-\lambda kT$ and $X''+\lambda X=0$, $X(0)=X(L)=0$ (the conditions on $v$ with $T\not\equiv0$).
Step 3. Case lambda < 0
Write $\lambda=-\mu^2$, $\mu>0$: $X=a\cosh\mu x+b\sinh\mu x$. $X(0)=a=0$; $X(L)=b\sinh\mu L=0$ with $\sinh\mu L>0$, so $b=0$. Only the trivial solution.
Step 4. Case lambda = 0
$X''=0$, $X=a+bx$. $X(0)=a=0$, $X(L)=bL=0\Rightarrow b=0$. Trivial.
Step 5. Case lambda > 0
$\lambda=\mu^2$: $X=a\cos\mu x+b\sin\mu x$. $X(0)=a=0$; $X(L)=b\sin\mu L=0$ with $b\ne0$ requires $\mu L=n\pi$. Hence
and $T'=-k\lambda_nT$ integrates to $T_n(t)=e^{-k\lambda_n t}$.
- $v_t=kv_{xx}$, $v(0,t)=v(L,t)=0$, $v(x,0)=u_0-u_s=-\dfrac{Q}{2k}x(L-x)$.
- $\lambda<0$ and $\lambda=0$: only $X\equiv0$.
- $\lambda_n=\Big(\dfrac{n\pi}{L}\Big)^2$, $X_n=\sin\dfrac{n\pi x}{L}$, $T_n=e^{-k n^2\pi^2t/L^2}$, $n=1,2,\dots$
Part (c) · 1.5 points
Series solution
Step 1. Superposition
Each product $X_nT_n$ satisfies $v_t=kv_{xx}$ and the zero end conditions, and so does any convergent combination:
At $t=0$: $\displaystyle -\frac{Q}{2k}x(L-x)=\sum c_n\sin\frac{n\pi x}{L}$, so by orthogonality $c_n=\dfrac2L\displaystyle\int_0^L\Big(-\frac{Q}{2k}\Big)x(L-x)\sin\frac{n\pi x}{L}\,dx$.
Step 2. The integral
Write $\omega=n\pi/L$. With $u=x(L-x)=Lx-x^2$, $dv=\sin\omega x\,dx$ ($du=(L-2x)\,dx$, $v=-\cos\omega x/\omega$):
The bracket is $0$ because $Lx-x^2$ vanishes at $x=0$ and $x=L$. Next, $u=L-2x$, $dv=\cos\omega x\,dx$ ($du=-2\,dx$, $v=\sin\omega x/\omega$):
since $\sin\omega L=\sin n\pi=0$. Hence $\displaystyle\int_0^L x(L-x)\sin\omega x\,dx=\frac{2(1-(-1)^n)}{\omega^3}=\frac{2L^3(1-(-1)^n)}{n^3\pi^3}$ and
Step 3. Coefficients and solution
Multiplying by $-\dfrac{Q}{2k}$: $c_n=-\dfrac{2QL^2(1-(-1)^n)}{k\pi^3n^3}$, which is $-\dfrac{4QL^2}{k\pi^3n^3}$ for odd $n$ and $0$ for even $n$. Adding $u_s$:
The coefficients decay as $1/n^3$ (the odd extension of $x(L-x)$ is continuous with a continuous derivative), so the series converges fast even at $t=0$.
- $\displaystyle\frac2L\int_0^L x(L-x)\sin\frac{n\pi x}{L}\,dx=\frac{4L^2}{n^3\pi^3}\big(1-(-1)^n\big)$: $\dfrac{8L^2}{n^3\pi^3}$ for odd $n$, $0$ for even $n$.
- $c_n=-\dfrac{4QL^2}{k\pi^3n^3}$ for odd $n$, $c_n=0$ for even $n$.
- $\displaystyle u(x,t)=T_A+(T_B-T_A)\frac xL+\frac{Q}{2k}x(L-x)-\frac{4QL^2}{k\pi^3}\sum_{n\ \mathrm{odd}}\frac{1}{n^3}\sin\frac{n\pi x}{L}\,e^{-kn^2\pi^2t/L^2}$.
Part (d) · 1 point
Long-time limit, time constant and the 99 % time
Step 1. Limit
Every term carries $e^{-kn^2\pi^2t/L^2}\to0$ and $\sum 1/n^3$ converges, so $v\to0$ and $u\to u_s$: the rod forgets its initial state.
Step 2. Slowest mode
The $n=1$ term decays as $e^{-t/\tau_1}$ with $\tau_1=\dfrac{1}{k\lambda_1}=\dfrac{L^2}{k\pi^2}=\dfrac{0.25}{10^{-4}\cdot9.8696}=253.3$ s. The next non-zero mode, $n=3$, has $\tau_3=\tau_1/9$.
Step 3. Midpoint rise
The total rise of the midpoint is $u_s(L/2)-u_0(L/2)=\dfrac{QL^2}{8k}=6.25$ K. At $x=L/2$ the deficit is
At $t=0$ the sum is $\pi^3/32$ (bank item P11), giving $\dfrac{4QL^2}{k\pi^3}\cdot\dfrac{\pi^3}{32}=\dfrac{QL^2}{8k}$, as it must. Keeping $m=0$, the fraction still missing is $\dfrac{32}{\pi^3}e^{-t/\tau_1}$.
Step 4. The 99 % time
$\dfrac{32}{\pi^3}e^{-t_{99}/\tau_1}=0.01\Rightarrow t_{99}=\tau_1\ln\dfrac{3200}{\pi^3}=\tau_1\ln(103.2)=4.637\,\tau_1\approx1174.5$ s. At that time the $m=1$ term is smaller than the $m=0$ term by the factor $\tfrac1{27}e^{-8\cdot4.637}<10^{-17}$: the one-mode answer is exact for practical purposes.
- $u(x,t)\to u_s(x)$ as $t\to\infty$.
- $\tau_1=\dfrac{L^2}{k\pi^2}=253.3$ s; the $n=3$ mode decays $9$ times faster.
- Midpoint: $\dfrac{u_s(L/2)-u(L/2,t)}{QL^2/(8k)}\approx\dfrac{32}{\pi^3}e^{-t/\tau_1}$, so $t_{99}=\tau_1\ln\dfrac{3200}{\pi^3}=4.637\,\tau_1\approx1174.5$ s (about 19.6 min).
Interpretation and checks
Phase 4. Asymptotic and physical checks
- Limits: $Q\to0$ gives $u_s=u_0$ and $v\equiv0$: nothing happens, as expected. $t\to\infty$: $u\to u_s$, a parabola above the straight line between $T_A$ and $T_B$; with these numbers its maximum is interior, at $x^*=\dfrac L2+\dfrac{k(T_B-T_A)}{QL}=0.25+0.2=0.45$ m, since $u_s'(x^*)=0$.
- Checks: dimensions: $\dfrac{Q}{k}x^2$ is $\dfrac{\text{K s}^{-1}}{\text{m}^2\text{s}^{-1}}\text{m}^2=$ K; $\tau_1=L^2/k$ is s. Each mode satisfies $v_t=kv_{xx}$ and vanishes at both ends; the series at $t=0$ reproduces $u_0$ (checked numerically in the script).
- Common mistakes in the exam: (1) separating variables directly in $u$, whose boundary conditions are not homogeneous; (2) taking $v(x,0)=u_0$ instead of $u_0-u_s$; (3) skipping the cases $\lambda\le0$; (4) giving the time constant as $L^2/k$ (forgetting $\pi^2$) or using the decay rate of $v^2$.
P13 — Laplace's equation in a rectangle with Neumann conditions on the vertical sides and Dirichlet data given by finite cosine sums: the zero mode, cosh and sinh modes, flux balance and uniqueness
The steady temperature $u(x,y)$ of the plate $0<x<\pi$, $0<y<1$ satisfies Laplace's equation
The vertical sides are insulated and the horizontal sides are held at prescribed temperatures:
(a) Separate variables, $u=X(x)Y(y)$, and solve the eigenvalue problem for $X$, discussing the three signs of the separation constant.
(b) Solve the equations for $Y$, paying special attention to the zero eigenvalue, and write the general solution that satisfies the conditions on the insulated sides.
(c) Impose the conditions on $y=0$ and $y=1$ and obtain $u(x,y)$ in closed form. Give $u$ at the centre $(\pi/2,\ 1/2)$.
(d) Compute $\displaystyle\Phi(y)=\int_0^{\pi}u_y(x,y)\,dx$ and explain why it does not depend on $y$. Is the solution unique? Would it be unique, and would it exist for any data, if the horizontal sides were also insulated?
Setup and canonical formulation
Phase 1. Hypotheses and setup
- Unknown $u(x,y)$ on the rectangle $R=(0,\pi)\times(0,1)$; steady two-dimensional conduction, no sources.
- Operator: Laplacian, linear, second order, elliptic, homogeneous equation. The conditions are homogeneous (Neumann) in $x$ and non-homogeneous (Dirichlet) in $y$, so the eigenvalue problem is posed in $x$.
- Data: the bottom and top temperatures are finite cosine sums, i.e. already combinations of the eigenfunctions found in (a); no integral is needed for the coefficients.
Phase 2. Canonical formulation
with $f(x)=2+\cos2x$ and $g(x)=4+3\cos x$. Sturm-Liouville problem in $x$: $X''+\lambda X=0$, $X'(0)=X'(\pi)=0$.
Part (a) · 1.5 points
Separation and the eigenvalue problem in x
Step 1. Separation
$u=X(x)Y(y)$ gives $X''Y+XY''=0$. Dividing by $XY$: $\dfrac{X''}{X}=-\dfrac{Y''}{Y}$. The left side depends only on $x$, the right only on $y$; both equal a constant $-\lambda$:
The side conditions $u_x(0,y)=X'(0)Y(y)=0$ for all $y$ and $Y\not\equiv0$ give $X'(0)=0$; likewise $X'(\pi)=0$.
Step 2. Case lambda < 0
$\lambda=-m^2$, $m>0$: $X=a\cosh mx+b\sinh mx$, $X'=m(a\sinh mx+b\cosh mx)$. $X'(0)=mb=0\Rightarrow b=0$. $X'(\pi)=ma\sinh m\pi=0$ with $\sinh m\pi>0\Rightarrow a=0$. Trivial.
Step 3. Case lambda = 0
$X''=0$, $X=a+bx$, $X'=b$. Both conditions give $b=0$, and $a$ is free: $X_0=1$ is an eigenfunction. With Neumann conditions the zero eigenvalue does occur (Haberman Sec. 2.4).
Step 4. Case lambda > 0
$\lambda=m^2$: $X=a\cos mx+b\sin mx$, $X'=m(-a\sin mx+b\cos mx)$. $X'(0)=mb=0\Rightarrow b=0$. $X'(\pi)=-ma\sin m\pi=0$ with $a\ne0$ requires $m=n\in\{1,2,\dots\}$. So $\lambda_n=n^2$, $X_n=\cos nx$.
- $\dfrac{X''}{X}=-\dfrac{Y''}{Y}=-\lambda$; $X''+\lambda X=0$, $X'(0)=X'(\pi)=0$; $Y''-\lambda Y=0$.
- $\lambda<0$: only $X\equiv0$.
- $\lambda_0=0$ is an eigenvalue with $X_0=1$.
- $\lambda_n=n^2$, $X_n=\cos nx$, $n=1,2,\dots$
Part (b) · 1 point
The y equations and the general solution
Step 1. Zero mode
$\lambda=0$: $Y''=0$, integrate twice: $Y'=B_0$, $Y=A_0+B_0y$. The product $X_0Y_0=A_0+B_0y$ is the part of the temperature that varies linearly between the plates.
Step 2. Modes n >= 1
$Y''-n^2Y=0$ has characteristic equation $r^2-n^2=0$, $r=\pm n$, so $Y=\alpha e^{ny}+\beta e^{-ny}$. With $\cosh ny=\tfrac12(e^{ny}+e^{-ny})$ and $\sinh ny=\tfrac12(e^{ny}-e^{-ny})$ this is $Y_n=A_n\cosh ny+B_n\sinh ny$ ($A_n=\alpha+\beta$, $B_n=\alpha-\beta$). The $\cosh/\sinh$ form is preferred because $\sinh 0=0$ and $\cosh0=1$ make the condition at $y=0$ immediate.
Step 3. Superposition
Every product satisfies Laplace's equation and the side conditions, and so does the sum:
- $Y_0=A_0+B_0y$ (linear in $y$).
- $Y_n=A_n\cosh ny+B_n\sinh ny$.
- $\displaystyle u(x,y)=A_0+B_0y+\sum_{n=1}^{\infty}\cos nx\,\big(A_n\cosh ny+B_n\sinh ny\big)$.
Part (c) · 1.5 points
Coefficients from the data and the closed form
Step 1. Bottom side
$u(x,0)=A_0+\sum_n A_n\cos nx=2+\cos2x$. By orthogonality of $\{1,\cos nx\}$ on $(0,\pi)$ the coefficients are unique: $A_0=2$, $A_2=1$, $A_n=0$ for $n\ne2$. (Directly: $\frac1\pi\int_0^\pi(2+\cos2x)dx=2$ and $\frac2\pi\int_0^\pi(2+\cos2x)\cos nx\,dx=\delta_{n2}$.)
Step 2. Top side
$u(x,1)=A_0+B_0+\sum_n(A_n\cosh n+B_n\sinh n)\cos nx=4+3\cos x$:
- constant: $2+B_0=4\Rightarrow B_0=2$;
- $n=1$: $0\cdot\cosh1+B_1\sinh1=3\Rightarrow B_1=\dfrac{3}{\sinh1}$;
- $n=2$: $1\cdot\cosh2+B_2\sinh2=0\Rightarrow B_2=-\dfrac{\cosh2}{\sinh2}=-\coth2$;
- $n\ge3$: $B_n\sinh n=0\Rightarrow B_n=0$.
Step 3. Simplify the n = 2 mode
by $\sinh(a-b)=\sinh a\cosh b-\cosh a\sinh b$. It equals $1$ at $y=0$ and $0$ at $y=1$, as required. Therefore
Step 4. Check
$u_{xx}=-\dfrac{3\sinh y}{\sinh1}\cos x-\dfrac{4\sinh(2-2y)}{\sinh2}\cos2x$ and $u_{yy}=\dfrac{3\sinh y}{\sinh1}\cos x+\dfrac{(-2)^2\sinh(2-2y)}{\sinh2}\cos2x$ (chain rule: $\frac{d}{dy}\sinh(2-2y)=-2\cosh(2-2y)$), so $u_{xx}+u_{yy}=0$. $u_x$ contains $\sin x$ and $\sin2x$, both zero at $x=0,\pi$.
Step 5. Centre
$\cos(\pi/2)=0$, $\cos\pi=-1$: $u(\pi/2,1/2)=2+1-\dfrac{\sinh1}{\sinh2}$. With $\sinh2=2\sinh1\cosh1$:
- $A_0=2$, $B_0=2$; $A_1=0$, $B_1=\dfrac{3}{\sinh1}$; $A_2=1$, $B_2=-\coth 2$; all other coefficients $0$.
- $\displaystyle u(x,y)=2+2y+\frac{3\sinh y}{\sinh1}\cos x+\frac{\sinh\big(2(1-y)\big)}{\sinh2}\cos2x$.
- $u(\pi/2,1/2)=3-\dfrac{1}{2\cosh1}\approx2.675973$.
Part (d) · 1 point
Flux balance and uniqueness
Step 1. The integral
$u_y=2+\dfrac{3\cosh y}{\sinh1}\cos x-\dfrac{2\cosh(2-2y)}{\sinh2}\cos2x$. Since $\int_0^\pi\cos x\,dx=[\sin x]_0^\pi=0$ and $\int_0^\pi\cos2x\,dx=\big[\tfrac12\sin2x\big]_0^\pi=0$:
Step 2. Why it is constant
$\Phi'(y)=\displaystyle\int_0^\pi u_{yy}\,dx=-\int_0^\pi u_{xx}\,dx=-\big[u_x\big]_{x=0}^{x=\pi}=-(0-0)=0$. Physically the heat flux $-K\,u_y$ through any horizontal line is the same: there are no sources inside and the sides are insulated. The temperature rises linearly on average, from mean $2$ at the bottom to mean $4$ at the top, so $B_0=2$ is exactly the mean gradient.
Step 3. Uniqueness (Dirichlet on two sides)
Let $u_1,u_2$ be two solutions and $w=u_1-u_2$: $\Delta w=0$, $w_x=0$ on $x=0,\pi$, $w=0$ on $y=0,1$. Green's first identity (extension: $\iint_R\big(w\Delta w+\lvert\nabla w\rvert^2\big)dA=\oint_{\partial R}w\,\dfrac{\partial w}{\partial n}\,ds$) gives $\iint_R\lvert\nabla w\rvert^2dA=0$, because on each side either $w=0$ or $\partial w/\partial n=0$. Hence $\nabla w=0$, $w$ is constant, and $w=0$ on $y=0$ makes it zero. The solution is unique. Consistently, the maximum principle (Haberman Sec. 2.5) bounds $u$ between the extreme boundary values: $1=u(\pi/2,0)\le u\le u(0,1)=7$ (the maximum principle on this mixed problem is checked on a grid in the script).
Step 4. All sides insulated
If $u_y$ were prescribed on $y=0$ and $y=1$ as well, the zero mode would only fix $B_0$ (twice) and leave $A_0$ free: solutions would differ by constants. Moreover, by Step 2 the prescribed fluxes would have to satisfy $\int_0^\pi u_y(x,0)\,dx=\int_0^\pi u_y(x,1)\,dx$ (solvability condition, Haberman Sec. 2.5): no net heat may enter a steady insulated plate. Data violating it give no solution at all.
- $\Phi(y)=2\pi$ for every $y$: only the zero mode carries net flux.
- Constant because, integrating $u_{xx}+u_{yy}=0$ over $(0,\pi)$, $\Phi'(y)=-[u_x]_0^\pi=0$: what enters through one plate leaves through the other (insulated sides).
- Unique: with Dirichlet data on two sides, $w\equiv0$. With all four sides insulated, solutions exist only if the net flux data are compatible, and are unique only up to an additive constant.
Interpretation and checks
Phase 4. Asymptotic and physical checks
- Limits: the $\cos2x$ mode imposed at the bottom decays upward like $\sinh(2(1-y))/\sinh2\approx e^{-2y}$; higher modes $\cos nx$ would decay like $e^{-ny}$, so fine detail of boundary data is felt only near the side where it is imposed. The zero mode does not decay: it is the only part of the data felt across the whole plate.
- Checks: $u(x,0)=2+0+0+\cos2x$ and $u(x,1)=4+3\cos x+0$; $u_x(0,y)=u_x(\pi,y)=0$; the centre value $2.676$ lies between the boundary extremes $1$ and $7$.
- Common mistakes in the exam: (1) discarding $\lambda=0$ as "trivial", which loses the linear part $2+2y$ and makes the constant data impossible to match; (2) using $\sin nx$ eigenfunctions (those belong to Dirichlet sides); (3) writing $Y_n=A_ne^{ny}+B_ne^{-ny}$ and then solving a $2\times2$ system with large numbers instead of using $\sinh(n(1-y))$; (4) claiming uniqueness only up to a constant, which is true only when every side is insulated.
P14 — Vibrating string with fixed ends started from the rest position with a parabolic initial velocity: modes, coefficients from the velocity, d'Alembert check at the midpoint, energy of each mode and periodicity
A string of length $L$, linear density $\rho$ and tension $T_0$ is fixed at $x=0$ and $x=L$. Its small transverse displacement $u(x,t)$ satisfies
At $t=0$ the string lies along its rest position and is struck by a wide hammer that gives it the velocity profile
where $V>0$ is the speed given to the midpoint.
(a) Separate variables, discuss the sign of the separation constant and find the normal modes compatible with the fixed ends and with $u(x,0)=0$.
(b) Obtain $u(x,t)$ as a series, with the coefficients in closed form.
(c) For $0\le t\le L/(2c)$ use d'Alembert's formula to obtain the displacement of the midpoint, $u(L/2,t)$, in closed form. Check it against the series at $t=L/(2c)$ and deduce $\sum_{n\ \mathrm{odd}}n^{-4}$.
(d) The energy of the string is $E(t)=\displaystyle\int_0^L\Big(\frac{\rho}{2}u_t^2+\frac{T_0}{2}u_x^2\Big)dx$. Compute the energy carried by each mode, check that their sum is the energy given by the hammer, and give the fraction carried by the fundamental. Show that the motion is periodic in time and describe the string at $t=L/(2c)$ and $t=L/c$.
Setup and canonical formulation
Phase 1. Hypotheses and setup
- Unknown $u(x,t)$ on $0\le x\le L$, $t\ge0$; small transverse vibrations, uniform string, no damping, no gravity.
- Operator: linear, second order, hyperbolic, homogeneous, with homogeneous Dirichlet conditions $u(0,t)=u(L,t)=0$.
- Data: $L$ [m], $c$ [m/s], $V$ [m/s], $\rho$ [kg/m], $T_0=\rho c^2$ [N]. Two initial conditions are needed (second order in $t$): zero displacement, non-zero velocity.
Phase 2. Canonical formulation
Eigenvalue problem in $x$: $X''+\lambda X=0$, $X(0)=X(L)=0$. In $t$: $T''+\lambda c^2T=0$, with $T(0)=0$ from the zero initial displacement.
Part (a) · 1.25 points
Separation and normal modes
Step 1. Separation
$XT''=c^2X''T$, so $\dfrac{T''}{c^2T}=\dfrac{X''}{X}=-\lambda$ (each side depends on one variable only). Fixed ends: $X(0)T(t)=0$ for all $t$, so $X(0)=0$, and $X(L)=0$.
Step 2. Case lambda < 0
$\lambda=-\mu^2$: $X=a\cosh\mu x+b\sinh\mu x$; $X(0)=a=0$, $X(L)=b\sinh\mu L=0\Rightarrow b=0$.
Step 3. Case lambda = 0
$X=a+bx$; $X(0)=a=0$, $X(L)=bL=0\Rightarrow b=0$.
Step 4. Case lambda > 0
$\lambda=\mu^2$: $X=a\cos\mu x+b\sin\mu x$; $X(0)=a=0$; $b\sin\mu L=0$ with $b\ne0$ gives $\mu=n\pi/L$. Then $T''+\omega_n^2T=0$ with $\omega_n=c\sqrt{\lambda_n}=\dfrac{n\pi c}{L}$, so $T=\alpha\cos\omega_nt+\beta\sin\omega_nt$. The condition $u(x,0)=X(x)T(0)=0$ gives $\alpha=0$. The normal modes are
- $\dfrac{T''}{c^2T}=\dfrac{X''}{X}=-\lambda$.
- $\lambda\le0$: only $X\equiv0$. $\lambda_n=\Big(\dfrac{n\pi}{L}\Big)^2$, $X_n=\sin\dfrac{n\pi x}{L}$.
- With $u(x,0)=0$: $T_n=\sin\omega_nt$, $\omega_n=\dfrac{n\pi c}{L}$.
Part (b) · 1.5 points
Series solution with coefficients from the velocity
Step 1. Superposition and the velocity condition
$u=\sum_{n\ge1}B_n\sin\dfrac{n\pi x}{L}\sin\omega_nt$ satisfies the equation, the fixed ends and $u(x,0)=0$. Differentiating in $t$ (chain rule: $\frac{d}{dt}\sin\omega_nt=\omega_n\cos\omega_nt$):
Step 2. The integral
Let $k=n\pi/L$. First parts, $u=Lx-x^2$, $dv=\sin kx\,dx$, $du=(L-2x)dx$, $v=-\cos kx/k$:
Second parts, $u=L-2x$, $dv=\cos kx\,dx$, $du=-2dx$, $v=\sin kx/k$:
So $\displaystyle\int_0^Lx(L-x)\sin kx\,dx=\frac{2(1-(-1)^n)}{k^3}=\frac{2L^3(1-(-1)^n)}{\pi^3n^3}$.
Step 3. Coefficients
which is $\dfrac{32V}{\pi^3n^3}$ for odd $n$ and $0$ for even $n$ (the even modes have a node at the midpoint and are not excited by a velocity profile symmetric about it). Dividing by $\omega_n=n\pi c/L$:
- $\displaystyle\frac2L\int_0^L g(x)\sin\frac{n\pi x}{L}\,dx=\frac{32V}{\pi^3n^3}$ for odd $n$, $0$ for even $n$.
- $\omega_nB_n=\dfrac{32V}{\pi^3n^3}$, so $B_n=\dfrac{32VL}{c\pi^4n^4}$ for odd $n$.
- $\displaystyle u(x,t)=\frac{32VL}{c\pi^4}\sum_{n\ \mathrm{odd}}\frac{1}{n^4}\sin\frac{n\pi x}{L}\sin\frac{n\pi ct}{L}$.
Part (c) · 1.25 points
d'Alembert check at the midpoint
Step 1. d'Alembert's formula
The general solution of $u_{tt}=c^2u_{xx}$ is $F(x-ct)+H(x+ct)$; with $u(x,0)=0$ and $u_t(x,0)=G(x)$ it becomes $u(x,t)=\dfrac{1}{2c}\displaystyle\int_{x-ct}^{x+ct}G(s)\,ds$ (extension: the course solves the string by separation of variables; d'Alembert's form is used here only as an independent check).
Step 2. The integral at the midpoint
With $s=L/2+\sigma$, $ds=d\sigma$, the limits $s=L/2\mp ct$ become $\sigma=\mp ct$, and $s(L-s)=\Big(\dfrac L2+\sigma\Big)\Big(\dfrac L2-\sigma\Big)=\dfrac{L^2}{4}-\sigma^2$, so $g=V-\dfrac{4V\sigma^2}{L^2}$:
For small $t$, $u\approx Vt$: the midpoint starts with speed $V$, as imposed.
Step 3. Comparison with the series
At $t=L/(2c)$: d'Alembert gives $\dfrac{VL}{2c}-\dfrac{4Vc^2}{3L^2}\cdot\dfrac{L^3}{8c^3}=\dfrac{VL}{2c}-\dfrac{VL}{6c}=\dfrac{VL}{3c}$. In the series, $\sin\dfrac{n\pi}{2}\sin\dfrac{n\pi}{2}=\big((-1)^{(n-1)/2}\big)^2=1$ for odd $n$:
- $u(L/2,t)=Vt-\dfrac{4Vc^2t^3}{3L^2}$ for $0\le t\le\dfrac{L}{2c}$.
- At $t=L/(2c)$: $u=\dfrac{VL}{3c}$, and the series gives $\dfrac{32VL}{c\pi^4}\sum_{n\ \mathrm{odd}}n^{-4}$.
- $\displaystyle\sum_{n\ \mathrm{odd}}\frac1{n^4}=\frac{\pi^4}{96}$.
Part (d) · 1 point
Energy of the modes and periodicity
Step 1. Energy of one mode
For $u_n=B_n\sin k_nx\,\sin\omega_nt$ ($k_n=n\pi/L$): $\partial_tu_n=B_n\omega_n\sin k_nx\cos\omega_nt$, $\partial_xu_n=B_nk_n\cos k_nx\sin\omega_nt$. With $\int_0^L\sin^2k_nx\,dx=\int_0^L\cos^2k_nx\,dx=L/2$ and $T_0=\rho c^2$:
since $c^2k_n^2=\omega_n^2$. With $\omega_nB_n=32V/(\pi^3n^3)$: $E_n=\dfrac{\rho L}{4}\cdot\dfrac{1024V^2}{\pi^6n^6}=\dfrac{256\rho LV^2}{\pi^6n^6}$.
Step 2. Total energy
$E$ is conserved (the ends do no work), so $E=E(0)=\dfrac\rho2\displaystyle\int_0^L\frac{16V^2}{L^4}x^2(L-x)^2dx$. Expanding $x^2(L-x)^2=L^2x^2-2Lx^3+x^4$ and integrating: $\Big[\dfrac{L^2x^3}{3}-\dfrac{Lx^4}{2}+\dfrac{x^5}{5}\Big]_0^L=L^5\Big(\dfrac13-\dfrac12+\dfrac15\Big)=\dfrac{L^5}{30}$. Hence $E=\dfrac{8\rho V^2}{L^4}\cdot\dfrac{L^5}{30}=\dfrac{4\rho LV^2}{15}$. The modes are orthogonal, so $E=\sum E_n$, and with $\sum_{n\ \mathrm{odd}}n^{-6}=\pi^6/960$ the sum indeed gives $\dfrac{256}{960}\rho LV^2=\dfrac{4}{15}\rho LV^2$.
Step 3. Fraction in the fundamental
$\dfrac{E_1}{E}=\dfrac{256/\pi^6}{4/15}=\dfrac{960}{\pi^6}\approx0.99856$: a smooth, symmetric strike puts $99.86\%$ of the energy in the fundamental; the string sounds an almost pure tone of frequency $c/(2L)$.
Step 4. Periodicity
$\omega_n=n\omega_1$, so every term has period $2\pi/\omega_1=2L/c$ and $u(x,t+2L/c)=u(x,t)$. Only odd $n$ appear, and $\sin(\omega_nt+n\pi)=-\sin\omega_nt$ for odd $n$, hence $u(x,t+L/c)=-u(x,t)$. At $t=L/c$: $u=0$ and $u_t=\sum\omega_nB_n\sin k_nx\cos n\pi=-g(x)$: the string crosses its rest position with the velocity reversed. At $t=L/(2c)$: $\cos(n\pi/2)=0$ for odd $n$, so $u_t=0$: the string is momentarily at rest at its maximum displacement (midpoint $VL/(3c)$).
- $E_n=\dfrac{\rho L}{4}\omega_n^2B_n^2=\dfrac{256\,\rho LV^2}{\pi^6n^6}$ ($n$ odd), constant in time.
- $E(0)=\dfrac{4\rho LV^2}{15}$, and $\sum_{n\ \mathrm{odd}}E_n=\dfrac{256\rho LV^2}{\pi^6}\cdot\dfrac{\pi^6}{960}=\dfrac{4\rho LV^2}{15}$.
- Fraction in the fundamental: $\dfrac{960}{\pi^6}\approx0.99856$.
- Period $\dfrac{2L}{c}$; at $t=L/(2c)$ the string is momentarily at rest with maximum displacement; at $t=L/c$ it crosses the rest position with velocity $-g(x)$, and $u(x,t+L/c)=-u(x,t)$.
Interpretation and checks
Phase 4. Asymptotic and physical checks
- Limits: $c\to\infty$ (very stiff string): $u\to0$, the displacement amplitude $VL/(3c)$ vanishes. Small $t$: $u\approx tg(x)$, consistent with the initial velocity. The coefficients decay as $n^{-4}$, faster than for a plucked string ($n^{-2}$), because the displacement is the time integral of a smooth velocity profile.
- Checks: dimensions: $B_n=\dfrac{32VL}{c\pi^4n^4}$ is $\dfrac{(\text{m/s})\,\text{m}}{\text{m/s}}=$ m; $E=\rho LV^2$ is (kg/m)(m)(m$^2$/s$^2$) $=$ J. Two independent routes (series and d'Alembert) give the same midpoint displacement $VL/(3c)$.
- Common mistakes in the exam: (1) keeping $\cos\omega_nt$ terms although the initial displacement is zero; (2) setting $B_n$ equal to the sine coefficients of $g$, forgetting the factor $1/\omega_n$ from differentiating in $t$; (3) using $g$ itself instead of its odd periodic extension in d'Alembert's formula when $ct>L/2$; (4) giving the period as $L/c$: that is only the half-period after which $u$ changes sign.