Chapter 10 · Partial Differential Equations

10.2 Classification and Separation of Variables

Second-order linear equations in two variables: the discriminant $\delta=B^2-4AC$ and the hyperbolic, parabolic and elliptic types, characteristics and canonical forms; then separation of variables and Fourier series for Laplace's equation in a rectangle and a disc, and for the heat and wave equations on an interval.

The three equations of classical mathematical physics — Laplace's equation $u_{xx}+u_{yy}=0$ for steady temperatures and electrostatic potentials, the heat equation $u_t=k\,u_{xx}$ for diffusion, and the wave equation $u_{tt}=c^2u_{xx}$ for vibrations — look almost the same on paper, yet their solutions behave in completely different ways. Heat smooths every corner instantly and forgets its past; waves carry corners along unchanged at a finite speed and never forget; harmonic functions have no time at all and are decided entirely by their boundary values. Which behaviour a given second-order equation shows is decided by one number computed from its top-order coefficients, the discriminant $\delta=B^2-4AC$, exactly as the discriminant of a quadratic form decides whether a conic is a hyperbola, a parabola or an ellipse. Choosing the characteristic curves as coordinates then brings every equation to one of three canonical forms.

The second half of the section solves the three model equations on simple domains by separation of variables: look for solutions that are products of functions of one variable, find which products fit the boundary conditions (an eigenvalue problem), and superpose them with Fourier coefficients chosen to match the remaining data. In GATE MA this produces some of the most reliable marks of the paper: MCQs and MSQs on the type of an equation in a region, its characteristics and canonical form, and the general solution of a constant-coefficient equation; NATs asking for the value of a separated solution of the Laplace, heat or wave equation at a point, or for a Fourier coefficient. Proofs are given for the central results; the longer ones are in the CSIR course linked in the 📚 Go deeper box.

Classification of Second-Order Equations

📘 Definition 10.6 — Second-order linear and semilinear equations; principal part

Let $\Omega\subseteq\mathbb{R}^2$ be open. A linear second-order equation in two variables is

$$A\,u_{xx}+B\,u_{xy}+C\,u_{yy}+D\,u_x+E\,u_y+F\,u=G ,$$

where $A,\dots,G$ are functions of $(x,y)$ on $\Omega$ and $A,B,C$ do not vanish simultaneously; it is homogeneous if $G\equiv0$. It is semilinear if it has the form $A\,u_{xx}+B\,u_{xy}+C\,u_{yy}=\Phi(x,y,u,u_x,u_y)$ with $A,B,C$ depending on $(x,y)$ only. The terms $A\,u_{xx}+B\,u_{xy}+C\,u_{yy}$ form the principal part. Note that $B$ is the whole coefficient of $u_{xy}$.

📘 Definition 10.7 — Discriminant; hyperbolic, parabolic and elliptic equations

The discriminant of the equation at a point $(x_0,y_0)$ is $\delta=B^2-4AC$, evaluated there. The equation is

  • hyperbolic at $(x_0,y_0)$ if $\delta\gt0$ — model: the wave equation $u_{tt}-c^2u_{xx}=0$;
  • parabolic at $(x_0,y_0)$ if $\delta=0$ — model: the heat equation $u_t-k\,u_{xx}=0$;
  • elliptic at $(x_0,y_0)$ if $\delta\lt0$ — model: Laplace's equation $u_{xx}+u_{yy}=0$.

It is of a given type on a set if it is of that type at every point of the set.

Two remarks prevent most mistakes. First, some books write the principal part as $Au_{xx}+2Bu_{xy}+Cu_{yy}$ and then test the sign of $B^2-AC$; with that convention $u_{xx}+2u_{xy}+u_{yy}$ has "$B=1$", with ours $B=2$, and both tests call it parabolic. Second, only the principal part matters: the heat equation $u_t=k\,u_{xx}$, written in the variables $(x,t)$, has $A=k$, $B=C=0$, so $\delta=0$ — it is parabolic even though it is "first order in $t$". With the symmetric matrix $M=\begin{pmatrix}A&B/2\\B/2&C\end{pmatrix}$ we have $\delta=-4\det M$: hyperbolic means $M$ indefinite, elliptic means $M$ definite, parabolic means $M$ singular but nonzero.

📐 Theorem 10.9 — Change of variables; invariance of the type

Let $\xi=\xi(x,y)$, $\eta=\eta(x,y)$ be $C^2$ near $(x_0,y_0)$ with Jacobian $J=\xi_x\eta_y-\xi_y\eta_x\ne0$, and write $u(x,y)=w(\xi,\eta)$. Then the principal part becomes $\bar A\,w_{\xi\xi}+\bar B\,w_{\xi\eta}+\bar C\,w_{\eta\eta}$ with

$$\bar A=A\xi_x^2+B\xi_x\xi_y+C\xi_y^2,\qquad\bar C=A\eta_x^2+B\eta_x\eta_y+C\eta_y^2,$$ $$\bar B=2A\xi_x\eta_x+B(\xi_x\eta_y+\xi_y\eta_x)+2C\xi_y\eta_y,$$

(the first-order terms change, the coefficient of $u$ does not), and

$$\bar B^2-4\bar A\bar C=J^2\,\big(B^2-4AC\big).$$

In particular the type at a point does not depend on the coordinates used.

Proof. By the chain rule $u_x=w_\xi\xi_x+w_\eta\eta_x$ and $u_y=w_\xi\xi_y+w_\eta\eta_y$; differentiating again,

$$u_{xx}=w_{\xi\xi}\xi_x^2+2w_{\xi\eta}\xi_x\eta_x+w_{\eta\eta}\eta_x^2+(\text{first order}),$$ $$u_{xy}=w_{\xi\xi}\xi_x\xi_y+w_{\xi\eta}(\xi_x\eta_y+\xi_y\eta_x)+w_{\eta\eta}\eta_x\eta_y+(\dots),$$ $$u_{yy}=w_{\xi\xi}\xi_y^2+2w_{\xi\eta}\xi_y\eta_y+w_{\eta\eta}\eta_y^2+(\text{first order}).$$

Multiplying by $A,B,C$ and adding gives the three formulas. In matrix language they say $\bar M=PMP^{\mathsf T}$ with $P=\begin{pmatrix}\xi_x&\xi_y\\\eta_x&\eta_y\end{pmatrix}$: indeed $\bar A=\mathbf pM\mathbf p^{\mathsf T}$, $\bar C=\mathbf qM\mathbf q^{\mathsf T}$ and $\tfrac12\bar B=\mathbf pM\mathbf q^{\mathsf T}$ for the rows $\mathbf p,\mathbf q$ of $P$. Taking determinants, $\det\bar M=(\det P)^2\det M=J^2\det M$, and multiplying by $-4$ gives the identity. Since $J^2\gt0$, $\delta$ and $\bar\delta$ have the same sign. ∎

Three consequences are used constantly. (i) Multiplying the equation by a nonvanishing function $h$ multiplies $\delta$ by $h^2\gt0$: the type does not change, even if $h\lt0$. (ii) Lower-order terms never affect the type: $u_{xx}-u_{yy}+100u_x+u^5=0$ is hyperbolic. (iii) If $A,B,C$ are continuous, the sets where the equation is hyperbolic or elliptic are open, while the parabolic set is closed and can be a curve, a single point, a region, or empty. (In $n$ variables the type is read from the signs of the eigenvalues of the coefficient matrix $(a_{ij})$: all of one sign — elliptic; one of the opposite sign — hyperbolic; one zero, the rest of one sign — parabolic. So $\Delta u=u_t$ in $\mathbb{R}^3\times\mathbb{R}$ is parabolic and $u_{tt}=c^2\Delta u$ is hyperbolic.)

✏️ Example 1 — Regions of each type
Find where each equation is hyperbolic, parabolic and elliptic: (a) $u_{xx}+2x\,u_{xy}+y\,u_{yy}=0$; (b) $x\,u_{xx}-4u_{xy}+y\,u_{yy}=0$; (c) $3u_{xx}+2u_{xy}+u_{yy}+u_x-u=0$; (d) the heat equation $u_t=k\,u_{xx}$ ($k\gt0$) and the wave equation $u_{tt}=c^2u_{xx}$ in the $(x,t)$-plane.
(a) $A=1$, $B=2x$, $C=y$: $\delta=4x^2-4y=4(x^2-y)$. Hyperbolic below the parabola $y=x^2$, parabolic on it, elliptic above it ($A=1$, so the equation is of second order everywhere).
(b) $\delta=16-4xy=4(4-xy)$. Hyperbolic where $xy\lt4$ (a region containing both axes), parabolic on the two branches of the hyperbola $xy=4$, elliptic in the two regions $xy\gt4$ (in the first and third quadrants, beyond the branches).
(c) $\delta=4-12=-8\lt0$: elliptic everywhere. The terms $u_x-u$ play no role.
(d) Heat: $k\,u_{xx}-u_t=0$ has $A=k$, $B=C=0$, $\delta=0$: parabolic. Wave: $c^2u_{xx}-u_{tt}=0$ has $A=c^2$, $B=0$, $C=-1$: $\delta=4c^2\gt0$, hyperbolic.
Check Interactive 1 below colours the plane by the sign of $\delta$ for (a) and shows the characteristic directions.

Characteristics and Canonical Forms

📘 Definition 10.8 — Characteristic directions and characteristic curves

At a point $(x,y)$ a nonzero vector $(dx,dy)$ is a characteristic direction of the principal part if

$$A\,dy^2-B\,dx\,dy+C\,dx^2=0 .$$

A curve is a characteristic curve if its tangent is a characteristic direction at each of its points. For a graph $y=y(x)$ this is the characteristic equation $A\left(\dfrac{dy}{dx}\right)^2-B\,\dfrac{dy}{dx}+C=0$, with roots $\dfrac{dy}{dx}=\lambda_\pm=\dfrac{B\pm\sqrt{\delta}}{2A}$ when $A\ne0$.

📐 Theorem 10.10 — Characteristics as level curves; how many there are
  1. A level curve $\varphi(x,y)=c$ of a $C^1$ function with $\nabla\varphi\ne0$ is a characteristic if and only if $A\varphi_x^2+B\varphi_x\varphi_y+C\varphi_y^2=0$ along it.
  2. At a hyperbolic point there are exactly two characteristic directions, at a parabolic point exactly one, at an elliptic point none.

Proof. (1) The tangent to a level curve is perpendicular to $\nabla\varphi$, so it is parallel to $(dx,dy)=(\varphi_y,-\varphi_x)$; substituting,

$$A\,dy^2-B\,dx\,dy+C\,dx^2=A\varphi_x^2+B\varphi_x\varphi_y+C\varphi_y^2 .$$

(2) If $A\ne0$, $dx=0$ would force $A\,dy^2=0$, so we may take $dx=1$ and solve the quadratic $A\lambda^2-B\lambda+C=0$, whose discriminant is $\delta$. If $A=0$ the form is $dx\,(C\,dx-B\,dy)$, giving the two directions $dx=0$ and $C\,dx=B\,dy$ when $B\ne0$ ($\delta=B^2\gt0$) and the single direction $dx=0$ when $B=0$ ($\delta=0$, $C\ne0$). ∎

Comparing (1) with Theorem 10.9: $\bar A=0$ exactly when the curves $\xi=\text{const}$ are characteristics, and $\bar C=0$ exactly when the curves $\eta=\text{const}$ are. Choosing characteristics as coordinate lines therefore kills coefficients. Beware the minus sign in the characteristic equation: $u_{xx}+u_{xy}-6u_{yy}=0$ gives $\lambda^2-\lambda-6=0$, so $\lambda=3$ or $-2$, not $-3$ or $2$.

📐 Theorem 10.11 — Canonical forms

Let $A,B,C$ be $C^2$ near $(x_0,y_0)$, not all zero there, and consider $A\,u_{xx}+B\,u_{xy}+C\,u_{yy}=\Phi(x,y,u,u_x,u_y)$. Near $(x_0,y_0)$ there are new coordinates in which the equation becomes:

  1. hyperbolic ($\delta\gt0$ at the point): $u_{\xi\eta}=\Phi_1(\xi,\eta,u,u_\xi,u_\eta)$, where $\xi=\text{const}$ and $\eta=\text{const}$ are the two families of characteristics; with $\alpha=\xi+\eta$, $\beta=\xi-\eta$ also $u_{\alpha\alpha}-u_{\beta\beta}=\Phi_2$;
  2. parabolic ($\delta\equiv0$ near the point): $u_{\eta\eta}=\Phi_1$, where $\xi=\text{const}$ is the family of characteristics and $\eta$ is any function with $J\ne0$;
  3. elliptic ($\delta\lt0$ at the point, $A,B,C$ real-analytic): $u_{\alpha\alpha}+u_{\beta\beta}=\Phi_1$, where $\alpha+i\beta=\varphi$ is a complex solution of $A\varphi_x^2+B\varphi_x\varphi_y+C\varphi_y^2=0$.

Proof. After interchanging $x$ and $y$ if necessary (or rotating by $45^\circ$ if $A=C=0$ at the point) we may assume $A\ne0$ near the point. (1) $\delta\gt0$ nearby, so $\lambda_+\ne\lambda_-$ are $C^2$ functions. Let $\xi$ and $\eta$ be first integrals of $y'=\lambda_+$ and $y'=\lambda_-$ with $\xi_y,\eta_y\ne0$ (for example, the value at $x=x_0$ of the solution through $(x,y)$). Then $\xi_x=-\lambda_+\xi_y$, so $\bar A=\xi_y^2\big(A\lambda_+^2-B\lambda_++C\big)=0$, and likewise $\bar C=0$. Also $J=\xi_x\eta_y-\xi_y\eta_x=(\lambda_--\lambda_+)\xi_y\eta_y\ne0$, so by Theorem 10.9 $\bar B^2=J^2\delta\gt0$ and we may divide by $\bar B$. For the second form, $\partial_\xi=\partial_\alpha+\partial_\beta$ and $\partial_\eta=\partial_\alpha-\partial_\beta$ give $u_{\xi\eta}=u_{\alpha\alpha}-u_{\beta\beta}$. (2) Now there is one family, $y'=\lambda=\frac{B}{2A}$; take $\xi$ a first integral as before, so $\bar A=0$. Theorem 10.9 gives $\bar B^2=4\bar A\bar C+J^2\delta=0$, so $\bar B=0$ as well. Then $\bar C\ne0$ (otherwise $\bar M=0$, impossible since $\bar M=PMP^{\mathsf T}$ with $M\ne0$ and $P$ invertible), and we divide by $\bar C$. (3) Write $\varphi=\alpha+i\beta$ and separate real and imaginary parts of $A\varphi_x^2+B\varphi_x\varphi_y+C\varphi_y^2=0$: the real part is $\bar A-\bar C=0$ and the imaginary part is $\bar B=0$ (with $\xi=\alpha$, $\eta=\beta$ in Theorem 10.9; $J\ne0$ because the ratio $\varphi_x/\varphi_y$ is a non-real root of $A\mu^2+B\mu+C=0$). Then $\bar\delta=-4\bar A^2\lt0$ forces $\bar A=\bar C\ne0$, and we divide. Analyticity is used only to solve the complex first-order equation (Cauchy–Kovalevskaya). ∎

🧭 Problem-solving strategy — reducing to canonical form
  1. Compute $\delta=B^2-4AC$ and decide the type on the region of interest.
  2. Solve $A\lambda^2-B\lambda+C=0$ for $\lambda=\frac{dy}{dx}$ and integrate $\frac{dy}{dx}=\lambda$ to get the characteristics in the form $\varphi(x,y)=\text{const}$.
  3. Hyperbolic: $\xi,\eta=$ the two families. Parabolic: $\xi=$ the family, $\eta=x$ (or $y$) — check $J\ne0$. Elliptic: $\alpha,\beta=$ real and imaginary parts of the complex family.
  4. Compute $u_x,u_y,u_{xx},u_{xy},u_{yy}$ by the chain rule (or Theorem 10.9 for the top-order part), substitute, divide by the surviving top coefficient, and express everything in the new variables.

When $A,B,C$ are constants and there are no lower-order terms, the canonical form can be integrated at once.

📐 Theorem 10.12 — General solution for constant coefficients

Let $A\ne0,B,C$ be constants and let $m_1,m_2$ be the roots of $Am^2-Bm+C=0$. Every $C^2$ solution of $A\,u_{xx}+B\,u_{xy}+C\,u_{yy}=0$ on $\mathbb{R}^2$ (or on any convex open set) is

  • $u=f(y-m_1x)+g(y-m_2x)$ if $\delta\gt0$ (real $m_1\ne m_2$);
  • $u=f(y-mx)+x\,g(y-mx)$ if $\delta=0$ (double root $m$),

with $f,g$ arbitrary $C^2$ functions. If $\delta\lt0$ the substitution $\alpha=y-\mathrm{Re}\,m\,x$, $\beta=\mathrm{Im}\,m\,x$ turns the equation into a multiple of Laplace's equation.

Proof. Since $m_1+m_2=B/A$ and $m_1m_2=C/A$, the operator factors:

$$A\partial_x^2+B\partial_x\partial_y+C\partial_y^2$$ $$=A(\partial_x+m_1\partial_y)(\partial_x+m_2\partial_y).$$

Hyperbolic. Put $\xi=y-m_1x$, $\eta=y-m_2x$ (Jacobian $m_2-m_1\ne0$). Then $\partial_x=-m_1\partial_\xi-m_2\partial_\eta$ and $\partial_y=\partial_\xi+\partial_\eta$, so $\partial_x+m_1\partial_y=(m_1-m_2)\partial_\eta$ and $\partial_x+m_2\partial_y=(m_2-m_1)\partial_\xi$. The equation becomes $-A(m_1-m_2)^2w_{\xi\eta}=0$, i.e. $w_{\xi\eta}=0$. On a convex set, integrating in $\eta$ shows $w_\xi$ depends on $\xi$ only, and integrating in $\xi$ gives $w=f(\xi)+g(\eta)$. Parabolic. Put $\xi=y-mx$, $\eta=x$; then $\partial_x+m\partial_y=\partial_\eta$ and the equation is $Aw_{\eta\eta}=0$, so $w=f(\xi)+\eta\,g(\xi)$. Conversely both formulas are solutions, since $(\partial_x+m\partial_y)f(y-mx)=0$. ∎

✏️ Example 2 — A general solution with data, and a variable-coefficient reduction
(a) Find the general solution of $u_{xx}+u_{xy}-6u_{yy}=0$ and the solution with $u(x,0)=0$, $u_y(x,0)=x$. (b) Reduce $u_{xx}-x^2u_{yy}=0$ to canonical form in $x\gt0$.
(a) Roots $\delta=1+24=25\gt0$. $m^2-m-6=(m-3)(m+2)$, so the characteristics are $y-3x=c$ and $y+2x=c$, and $u=f(y-3x)+g(y+2x)$.
(a) Data $f(-3x)+g(2x)=0$ and $f'(-3x)+g'(2x)=x$. Differentiating the first: $-3f'(-3x)+2g'(2x)=0$, so $f'(-3x)=\tfrac23g'(2x)$ and $\tfrac53g'(2x)=x$, i.e. $g'(w)=\tfrac{3w}{10}$ and $g(w)=\tfrac{3w^2}{20}$ (constants cancel). Then $f(-3x)=-\tfrac{3x^2}{5}$, so $f(s)=-\tfrac{s^2}{15}$.
(a) Answer $$u=-\frac{(y-3x)^2}{15}+\frac{3(y+2x)^2}{20}=xy+\frac{y^2}{12}.$$ Check: $u_{xx}+u_{xy}-6u_{yy}=0+1-6\cdot\tfrac16=0$, $u(x,0)=0$, $u_y(x,0)=x$. ✓
(b) Characteristics $\delta=4x^2\gt0$; $\lambda^2-x^2=0$ gives $\frac{dy}{dx}=\pm x$, so $\xi=y+\tfrac{x^2}2$, $\eta=y-\tfrac{x^2}2$ ($J=2x\ne0$).
(b) Substitute $u_x=x(w_\xi-w_\eta)$, $u_{xx}=(w_\xi-w_\eta)+x^2(w_{\xi\xi}-2w_{\xi\eta}+w_{\eta\eta})$ and $u_{yy}=w_{\xi\xi}+2w_{\xi\eta}+w_{\eta\eta}$. So $u_{xx}-x^2u_{yy}=(w_\xi-w_\eta)-4x^2w_{\xi\eta}$.
(b) Answer With $x^2=\xi-\eta$: $$w_{\xi\eta}=\frac{w_\xi-w_\eta}{4(\xi-\eta)} .$$
✏️ Example 3 — A parabolic and an elliptic reduction
(a) Reduce $x^2u_{xx}+2xy\,u_{xy}+y^2u_{yy}=0$ ($x\gt0$) to canonical form and find its general solution. (b) Reduce $u_{xx}+x^2u_{yy}=0$ ($x\gt0$) to canonical form.
(a) Type $\delta=4x^2y^2-4x^2y^2=0$: parabolic. $x^2\lambda^2-2xy\lambda+y^2=(x\lambda-y)^2=0$ gives $\lambda=\frac yx$ and the characteristics $\xi=\frac yx$ (rays from the origin). Take $\eta=x$: $J=\xi_x\eta_y-\xi_y\eta_x=-\frac1x\ne0$.
(a) Reduce By Theorem 10.9, $\bar A=\bar B=0$ and $\bar C=A\cdot1^2=x^2$. The first-order terms cancel as well (the coefficient of $w_\xi$ is $x^2\xi_{xx}+2xy\,\xi_{xy}+y^2\xi_{yy}=\frac{2y}{x}-\frac{2y}{x}+0=0$), so the equation is $x^2w_{\eta\eta}=0$, i.e. $$w_{\eta\eta}=0 .$$
(a) Solve $w=f(\xi)+\eta\,g(\xi)$: $$u=f\!\left(\tfrac yx\right)+x\,g\!\left(\tfrac yx\right),$$ the sums of homogeneous functions of degrees $0$ and $1$. (Indeed the operator is $L(L-1)$ with $L=x\partial_x+y\partial_y$, Euler's operator of §1.1.)
(b) Complex characteristics $\delta=-4x^2\lt0$. $\lambda^2+x^2=0$ gives $\frac{dy}{dx}=\pm ix$, i.e. $y\mp i\frac{x^2}2=c$. Take $\alpha=y$, $\beta=\frac{x^2}2$.
(b) Substitute $u_x=x\,w_\beta$, $u_{xx}=w_\beta+x^2w_{\beta\beta}$, $u_{yy}=w_{\alpha\alpha}$. So $u_{xx}+x^2u_{yy}=x^2(w_{\alpha\alpha}+w_{\beta\beta})+w_\beta$, and with $x^2=2\beta$: $$w_{\alpha\alpha}+w_{\beta\beta}=-\frac{w_\beta}{2\beta}.$$

Separation of Variables and Fourier Series

For the heat equation $u_t=k\,u_{xx}$ on a rod $0\lt x\lt L$ with ends held at temperature $0$, try $u=X(x)T(t)$. Substituting, $XT'=kX''T$, and dividing by $kXT$,

$$\frac{T'(t)}{k\,T(t)}=\frac{X''(x)}{X(x)} .$$

A function of $t$ alone equals a function of $x$ alone only if both are a constant, which we call $-\lambda$. So $X''+\lambda X=0$ with $X(0)=X(L)=0$ (from the boundary conditions), and $T'=-k\lambda T$. Only special values of $\lambda$ allow a nonzero $X$: the boundary conditions turn separation of variables into an eigenvalue problem, a Sturm–Liouville problem of the kind met in §6.3.

📐 Theorem 10.13 — The three basic eigenvalue problems

For $X''+\lambda X=0$:

  1. Dirichlet, $X(0)=X(L)=0$: eigenvalues $\lambda_n=\left(\frac{n\pi}{L}\right)^2$, $n=1,2,\dots$, eigenfunctions $\sin\frac{n\pi x}{L}$.
  2. Neumann, $X'(0)=X'(L)=0$: eigenvalues $\lambda_n=\left(\frac{n\pi}{L}\right)^2$, $n=0,1,2,\dots$, eigenfunctions $\cos\frac{n\pi x}{L}$ (including the constant $1$ for $\lambda_0=0$).
  3. Periodic on $[-\pi,\pi]$, $X(-\pi)=X(\pi)$, $X'(-\pi)=X'(\pi)$: eigenvalues $n^2$, $n=0,1,2,\dots$, with eigenfunctions $1$ and, for $n\ge1$, the two-dimensional space spanned by $\cos nx$ and $\sin nx$.

In each case there are no other eigenvalues (in particular none negative), and eigenfunctions belonging to different eigenvalues are orthogonal in $L^2$.

Proof of (1). If $\lambda=-\mu^2\lt0$, then $X=c_1\cosh\mu x+c_2\sinh\mu x$; $X(0)=0$ gives $c_1=0$ and $X(L)=c_2\sinh\mu L=0$ gives $c_2=0$. If $\lambda=0$, $X=c_1+c_2x$ and both constants vanish. If $\lambda=\mu^2\gt0$, $X=c_1\cos\mu x+c_2\sin\mu x$; $c_1=0$, and a nonzero solution needs $\sin\mu L=0$, i.e. $\mu=\frac{n\pi}L$. (2) is the same with derivatives; now $\lambda=0$ gives the constant solution. (3) $\lambda\lt0$: a combination of $e^{\pm\mu x}$ that is $2\pi$-periodic with periodic derivative must vanish; $\lambda=\mu^2\gt0$: $\cos\mu x$, $\sin\mu x$ satisfy the conditions iff $\sin\mu\pi=0$, i.e. $\mu=n$. Orthogonality: $\int_0^L\sin\frac{m\pi x}L\sin\frac{n\pi x}L\,dx=0$ for $m\ne n$ by the product-to-sum formula (or by the Sturm–Liouville argument of §6.3). ∎

📘 Definition 10.9 — Fourier series; sine and cosine series

For $f$ integrable on $[-L,L]$, its Fourier series is

$$\frac{a_0}{2}+\sum_{n=1}^{\infty}\left(a_n\cos\frac{n\pi x}{L}+b_n\sin\frac{n\pi x}{L}\right),$$ $$a_n=\frac1L\int_{-L}^{L}f(x)\cos\frac{n\pi x}{L}\,dx,\qquad b_n=\frac1L\int_{-L}^{L}f(x)\sin\frac{n\pi x}{L}\,dx .$$

For $f$ on $[0,L]$, the Fourier sine series $\sum b_n\sin\frac{n\pi x}{L}$ with $b_n=\frac2L\int_0^Lf(x)\sin\frac{n\pi x}{L}\,dx$ is the Fourier series of the odd extension of $f$; the cosine series $\frac{a_0}2+\sum a_n\cos\frac{n\pi x}{L}$ with $a_n=\frac2L\int_0^Lf(x)\cos\frac{n\pi x}{L}\,dx$ is that of the even extension.

📐 Theorem 10.14 — Convergence of Fourier series (Dirichlet); Parseval

Let $f$ be piecewise $C^1$ on $[-L,L]$ and extended $2L$-periodically. Then its Fourier series converges at every $x$ to $\tfrac12\big(f(x^+)+f(x^-)\big)$. If moreover the periodic extension is continuous, the convergence is absolute and uniform on $\mathbb{R}$. For every square-integrable $f$,

$$\frac1L\int_{-L}^{L}f^2\,dx=\frac{a_0^2}{2}+\sum_{n=1}^{\infty}\left(a_n^2+b_n^2\right)\qquad(\text{Parseval}).$$

(Proof: Dirichlet's kernel and the Riemann–Lebesgue lemma for pointwise convergence; for continuous piecewise $C^1$ extensions $\sum(|a_n|+|b_n|)\lt\infty$ by Cauchy–Schwarz and Bessel applied to $f'$, then the M-test of §4.2; Parseval is the completeness of the trigonometric system in $L^2$, §8.3. Full proofs in the Go deeper links.) For a sine series on $[0,L]$ the odd periodic extension is continuous exactly when $f$ is continuous with $f(0)=f(L)=0$; otherwise the series still converges, but not uniformly, and overshoots near the jumps by about $9\%$ of the jump (the Gibbs phenomenon, visible in Interactive 2).

Laplace's Equation in a Rectangle and a Disc

📘 Definition 10.10 — Dirichlet and Neumann problems
Let $\Omega$ be a bounded domain with boundary $\partial\Omega$ and write $\Delta u=u_{xx}+u_{yy}$. The Dirichlet problem asks for $u\in C^2(\Omega)\cap C(\bar\Omega)$ with $\Delta u=0$ in $\Omega$ and $u=f$ on $\partial\Omega$. The Neumann problem prescribes instead the outward normal derivative $\dfrac{\partial u}{\partial n}=g$ on $\partial\Omega$ (with $u\in C^1(\bar\Omega)$). Solutions of $\Delta u=0$ are the harmonic functions of §5.1.
📐 Theorem 10.15 — Maximum principle, uniqueness and the mean value property
  1. Weak maximum principle. If $u\in C^2(\Omega)\cap C(\bar\Omega)$ is harmonic on the bounded domain $\Omega$, then $\max_{\bar\Omega}u=\max_{\partial\Omega}u$ and $\min_{\bar\Omega}u=\min_{\partial\Omega}u$.
  2. Uniqueness and stability. The Dirichlet problem has at most one solution, and two solutions with data $f_1,f_2$ satisfy $\max_{\bar\Omega}|u_1-u_2|\le\max_{\partial\Omega}|f_1-f_2|$.
  3. Mean value property; strong maximum principle. For every closed disc of radius $r$ about $(x_0,y_0)$ in $\Omega$, $u(x_0,y_0)=\dfrac{1}{2\pi}\displaystyle\int_0^{2\pi}u(x_0+r\cos\theta,\,y_0+r\sin\theta)\,d\theta$. Consequently a harmonic function that attains its maximum (or minimum) at an interior point of a connected $\Omega$ is constant.

Proof. (1) Let $R$ be such that $x^2+y^2\le R^2$ on $\bar\Omega$, let $\varepsilon\gt0$, and put $v=u+\varepsilon(x^2+y^2)$, so $\Delta v=4\varepsilon\gt0$. If $v$ had its maximum over the compact set $\bar\Omega$ at an interior point, there $v_{xx}\le0$ and $v_{yy}\le0$ (second-derivative test in each variable), so $\Delta v\le0$ — a contradiction. Hence

$$\max_{\bar\Omega}u\le\max_{\bar\Omega}v=\max_{\partial\Omega}v\le\max_{\partial\Omega}u+\varepsilon R^2;$$

let $\varepsilon\to0$. The minimum follows by applying this to $-u$. (2) $u_1-u_2$ is harmonic with boundary values $f_1-f_2$; apply (1) to it and to its negative. (3) The mean value formula follows from Cauchy's integral formula (§5.1) applied to an analytic function with real part $u$ on the disc, or from the divergence theorem. If $u(x_0,y_0)=M=\max u$, the formula forces $u=M$ on every circle about $(x_0,y_0)$ in $\Omega$, so the set $\{u=M\}$ is open; it is also closed, hence everything. ∎

📐 Theorem 10.16 — Dirichlet problem in a rectangle

Let $R=(0,a)\times(0,b)$ and let $f$ be continuous and piecewise $C^1$ on $[0,a]$ with $f(0)=f(a)=0$. The solution of $\Delta u=0$ in $R$ with $u=0$ on the sides $x=0$, $x=a$, $y=0$ and $u(x,b)=f(x)$ is

$$u(x,y)=\sum_{n=1}^{\infty}b_n\,\frac{\sinh(n\pi y/a)}{\sinh(n\pi b/a)}\,\sin\frac{n\pi x}{a},$$ $$b_n=\frac2a\int_0^af(x)\sin\frac{n\pi x}{a}\,dx .$$

Data on the other sides are handled the same way (exchange the roles of the variables, or replace $y$ by $b-y$), and general data by adding the four solutions.

Proof. With $u=X(x)Y(y)$, $\frac{X''}{X}=-\frac{Y''}{Y}=-\lambda$. The homogeneous conditions on $x=0,a$ give the Dirichlet problem of Theorem 10.13(1): $\lambda_n=\left(\frac{n\pi}a\right)^2$, $X_n=\sin\frac{n\pi x}a$. Then $Y''=\lambda_nY$ with $Y(0)=0$ gives $Y_n=\sinh\frac{n\pi y}a$. Superposing and putting $y=b$ requires $\sum c_n\sinh\frac{n\pi b}{a}\sin\frac{n\pi x}a=f(x)$, so $c_n\sinh\frac{n\pi b}a=b_n$, the sine coefficients. For $y\le b-\varepsilon$ the factor $\frac{\sinh(n\pi y/a)}{\sinh(n\pi b/a)}$ is at most a constant times $e^{-n\pi\varepsilon/a}$, so the series and all its term-by-term derivatives converge uniformly there (M-test) and $u$ is harmonic. Under the hypotheses on $f$ the sine series converges uniformly (Theorem 10.14), and so does the series for $u$ on all of $\bar R$ by the maximum principle applied to the partial sums; so $u$ is continuous up to the boundary with the right values. Uniqueness is Theorem 10.15. ∎

📐 Theorem 10.17 — The disc: series, Poisson's formula, Neumann problem

Let $f$ be continuous and piecewise $C^1$ and $2\pi$-periodic, with Fourier coefficients $a_n,b_n$. The solution of the Dirichlet problem in the disc $r\lt\rho$ with $u(\rho,\theta)=f(\theta)$ is

$$u(r,\theta)=\frac{a_0}2+\sum_{n=1}^{\infty}\left(\frac r\rho\right)^{n}\big(a_n\cos n\theta+b_n\sin n\theta\big),$$

equivalently (Poisson's integral formula)

$$u(r,\theta)=\frac1{2\pi}\int_{-\pi}^{\pi}\frac{(\rho^2-r^2)\,f(\varphi)}{\rho^2-2\rho r\cos(\theta-\varphi)+r^2}\,d\varphi .$$

The Neumann problem $u_r(\rho,\theta)=g(\theta)$ is solvable if and only if $\int_{-\pi}^{\pi}g\,d\theta=0$; its solutions are then $u=C+\sum_{n\ge1}\frac{\rho}{n}\left(\frac r\rho\right)^n(\alpha_n\cos n\theta+\beta_n\sin n\theta)$, with $\alpha_n,\beta_n$ the Fourier coefficients of $g$ and $C$ arbitrary.

Proof. In polar coordinates $\Delta u=u_{rr}+\frac1ru_r+\frac1{r^2}u_{\theta\theta}$. With $u=R(r)\Theta(\theta)$, $\frac{r^2R''+rR'}{R}=-\frac{\Theta''}{\Theta}=\lambda$. Since $u$ must be single-valued and smooth, $\Theta$ is $2\pi$-periodic, so by Theorem 10.13(3) $\lambda=n^2$ and $\Theta\in\operatorname{span}\{\cos n\theta,\sin n\theta\}$. The radial equation $r^2R''+rR'-n^2R=0$ is a Cauchy–Euler equation (§6.1) with indicial roots $\pm n$: solutions $r^n,r^{-n}$ for $n\ge1$ and $1,\log r$ for $n=0$. Boundedness at the centre discards $r^{-n}$ and $\log r$. Superposing and matching at $r=\rho$ gives the series. Inserting the integral formulas for $a_n,b_n$ and summing the geometric series $\frac12+\sum_{n\ge1}s^n\cos n\psi=\frac{1-s^2}{2(1-2s\cos\psi+s^2)}$ ($s=\frac r\rho\lt1$) gives Poisson's formula. For the Neumann problem, the divergence theorem gives $\oint_{\partial\Omega}\frac{\partial u}{\partial n}\,ds=\iint_\Omega\Delta u\,dA=0$, which is the necessary condition; differentiating the series term by term at $r=\rho$ shows that it reproduces $g$ exactly when its constant term $\frac{\alpha_0}2$ is $0$. ∎

Two facts follow at once and are favourites in MSQs. At the centre, $u(0,\theta)=\frac{a_0}2=\frac1{2\pi}\int f$: the value at the centre is the mean of the boundary data. And the Poisson kernel is positive, so $f\ge0$, $f\not\equiv0$ gives $u\gt0$ inside. For an annulus $\rho_1\lt r\lt\rho_2$ nothing is discarded: one needs $1$, $\log r$, $r^{\pm n}$.

✏️ Example 4 — A plate with one hot edge
The square plate $0\le x,y\le\pi$ has three edges at temperature $0$ and the edge $y=\pi$ at temperature $1$. (a) Find the steady temperature $u$ as a series. (b) Find $u$ at the centre exactly, and compare with the first term of the series. (c) Find the solution when instead $u(x,0)=\sin x$ and the other three edges are at $0$, and its value at the centre.
(a) $a=b=\pi$ and $b_n=\frac2\pi\int_0^\pi\sin nx\,dx=\frac{2(1-(-1)^n)}{n\pi}$, i.e. $\frac{4}{n\pi}$ for odd $n$ and $0$ for even $n$: $$u=\frac4\pi\sum_{n\ \mathrm{odd}}\frac{\sinh ny}{n\,\sinh n\pi}\sin nx .$$ (Here $f\equiv1$ does not vanish at the corners, so $u$ is discontinuous at the two top corners; the series still gives the unique bounded solution inside.)
(b) Symmetry Let $u_1,\dots,u_4$ be the solutions with the value $1$ on one edge and $0$ on the others. Their sum has boundary value $1$ on every edge, so it is the constant $1$. A rotation of the square by $90^\circ$ about its centre carries each problem into the next and fixes the centre, so the four values there are equal: $$u\left(\tfrac\pi2,\tfrac\pi2\right)=\tfrac14 .$$
(b) First term $\frac4\pi\cdot\frac{\sinh(\pi/2)}{\sinh\pi}\cdot1=\frac{2}{\pi\cosh(\pi/2)}\approx0.2537$, already within $0.004$ of $0.25$: the higher terms are damped by $\frac{\sinh(n\pi/2)}{\sinh n\pi}\approx e^{-n\pi/2}$.
(c) Replacing $y$ by $\pi-y$: $u=\dfrac{\sinh(\pi-y)}{\sinh\pi}\sin x$. At the centre $u=\frac{\sinh(\pi/2)}{\sinh\pi}=\frac{1}{2\cosh(\pi/2)}\approx0.1993$.
✏️ Example 5 — The unit disc: Dirichlet, Neumann, and a prove-or-disprove
(a) Solve $\Delta u=0$ in $r\lt1$ with $u(1,\theta)=\sin^3\theta$; write $u$ in Cartesian form and evaluate $u(0,0)$ and $u\left(0,\tfrac12\right)$. (b) Decide whether the Neumann problem $u_r(1,\theta)=g(\theta)$ is solvable for $g=\cos^2\theta$ and for $g=\cos2\theta$, and solve it when it is. (c) True or false: if $u$ is harmonic in $r\lt1$, continuous on $r\le1$, and $u(1,\theta)=\cos^2\theta$, then $0\le u\le1$ and $u(0,0)=\frac12$.
(a) Fourier $\sin^3\theta=\frac34\sin\theta-\frac14\sin3\theta$, a finite Fourier series, so $u=\frac34r\sin\theta-\frac14r^3\sin3\theta$.
(a) Cartesian $r\sin\theta=y$ and $r^3\sin3\theta=\mathrm{Im}(x+iy)^3=3x^2y-y^3$: $$u=\frac{3y-3x^2y+y^3}{4}.$$ $u(0,0)=0$ (the mean of $\sin^3$), and $u\left(0,\tfrac12\right)=\frac{3/2+1/8}{4}=\frac{13}{32}$.
(b) $\int_{-\pi}^{\pi}\cos^2\theta\,d\theta=\pi\ne0$: no solution (heat flowing in through the whole boundary cannot be in a steady state). For $\cos2\theta$ the integral is $0$: $u=\frac{r^2}{2}\cos2\theta+C=\frac{x^2-y^2}2+C$; indeed $u_r=r\cos2\theta=\cos2\theta$ at $r=1$.
(c) True The data take values in $[0,1]$, so by the maximum principle $0\le u\le1$; by the mean value property $u(0,0)=\frac1{2\pi}\int\cos^2\theta\,d\theta=\frac12$. Explicitly $u=\frac12+\frac{x^2-y^2}{2}$.

The Heat and Wave Equations on an Interval

📐 Theorem 10.18 — The heat equation on a rod

Let $k\gt0$ and let $f$ be piecewise $C^1$ on $[0,L]$. For $u_t=k\,u_{xx}$, $0\lt x\lt L$, $t\gt0$, with $u(x,0)=f(x)$:

  1. ends at temperature $0$ ($u(0,t)=u(L,t)=0$): $$u=\sum_{n=1}^\infty b_n\,e^{-k(n\pi/L)^2t}\sin\frac{n\pi x}{L},$$ with $b_n$ the sine coefficients of $f$;
  2. insulated ends ($u_x(0,t)=u_x(L,t)=0$): $$u=\frac{a_0}2+\sum_{n=1}^\infty a_n\,e^{-k(n\pi/L)^2t}\cos\frac{n\pi x}{L},$$ with $a_n$ the cosine coefficients; $u\to\frac{a_0}{2}=\frac1L\int_0^Lf$, the mean initial temperature;
  3. ends held at $T_1,T_2$: $u=T_1+(T_2-T_1)\frac xL+w$, where $w$ solves (1) with initial data $f(x)-T_1-(T_2-T_1)\frac xL$. The linear function is the steady state.

For $t\gt0$ these series converge together with all their derivatives, so $u$ is $C^\infty$ in $x$ and $t$ for $t\gt0$, whatever $f$ is, and $u$ approaches the steady state like $e^{-k\pi^2t/L^2}$.

Proof. Separation gives $X''+\lambda X=0$ with the boundary conditions of Theorem 10.13(1) or (2), and $T'=-k\lambda_nT$, $T=e^{-k\lambda_nt}$. Superposing and putting $t=0$ requires $\sum b_n\sin\frac{n\pi x}L=f$, which fixes the coefficients. For convergence: $|b_n|\le\frac2L\int_0^L|f|=:K$, and for $t\ge t_0\gt0$ the $j$-th derivative in $x$ and $i$-th in $t$ of the $n$-th term is bounded by $K\left(\frac{n\pi}L\right)^{j+2i}k^ie^{-k(n\pi/L)^2t_0}$, the terms of a convergent series. By the M-test (§4.2) we may differentiate term by term; each term solves the equation, so $u$ does. (3) is linearity: the steady state solves the equation and the boundary conditions, and $w=u-$(steady state) has zero boundary values. ∎

📐 Theorem 10.19 — Maximum principle and energy: uniqueness for the heat equation

Let $u$ be continuous on $[0,L]\times[0,T]$, with $u_t,u_{xx}$ continuous for $t\gt0$, and $u_t=k\,u_{xx}$ for $0\lt x\lt L$, $0\lt t\le T$.

  1. Maximum principle. The maximum and minimum of $u$ are attained on the parabolic boundary $\{t=0\}\cup\{x=0\}\cup\{x=L\}$.
  2. Energy. If $u=0$ at both ends (or $u_x=0$ at both ends), and $u_x$ is continuous up to the ends for $t\gt0$, then $E(t)=\int_0^Lu^2\,dx$ is non-increasing: $E'(t)=-2k\int_0^Lu_x^2\,dx\le0$.

Either part shows that the initial–boundary value problems of Theorem 10.18 have at most one solution.

Proof. (1) Let $v=u+\varepsilon x^2$; then $v_t-kv_{xx}=-2k\varepsilon\lt0$. At a maximum point of $v$ on $[0,L]\times[0,T']$ ($T'\lt T$ arbitrary, $\varepsilon$ small) with $0\lt x\lt L$ and $0\lt t\le T'$ we would have $v_{xx}\le0$ and $v_t\ge0$ (only a one-sided derivative is needed at $t=T'$), so $v_t-kv_{xx}\ge0$ — impossible. So $\max v$ is on the parabolic boundary, and $\max u\le\max_{\text{par.bdry}}u+\varepsilon L^2$; let $\varepsilon\to0$, then $T'\to T$. (2) Differentiating under the integral and integrating by parts,

$$E'=2\int_0^Luu_t\,dx=2k\int_0^Luu_{xx}\,dx$$ $$=2k\big[uu_x\big]_0^L-2k\int_0^Lu_x^2\,dx,$$

and the bracket vanishes under either boundary condition. For uniqueness, the difference $w$ of two solutions has zero data, so (1) gives $0\le w\le0$; or (2) gives $0\le E(t)\le E(0)=0$. ∎

✏️ Example 6 — A rod heating up, and an insulated rod
(a) A rod $0\le x\le1$ with $k=1$ is initially at temperature $0$; for $t\gt0$ the end $x=0$ is kept at $0$ and the end $x=1$ at $1$. Find $u(x,t)$ and $u\left(\tfrac12,0.1\right)$. (b) Using only the first nonzero term at $x=\tfrac12$, estimate the time after which the midpoint is within $0.01$ of its final temperature. (c) If instead both ends are insulated and $u(x,0)=x$, find $u$ and its limit. (d) True or false: in (a), $0\le u\le1$ for all $x,t$.
(a) Steady state $v(x)=x$. Then $w=u-x$ has zero end values and $w(x,0)=-x$, whose sine coefficients are $b_n=2\int_0^1(-x)\sin n\pi x\,dx=\frac{2(-1)^n}{n\pi}$.
(a) Solution $$u=x+\sum_{n=1}^\infty\frac{2(-1)^n}{n\pi}e^{-n^2\pi^2t}\sin n\pi x .$$ At $x=\frac12$ only odd $n$ contribute ($\sin\frac{n\pi}2=\pm1$), and summing numerically $u\left(\tfrac12,0.1\right)\approx0.2628$.
(b) At $x=\frac12$ the $n=1$ term is $-\frac2\pi e^{-\pi^2t}$, the $n=2$ term vanishes and the $n=3$ term carries $e^{-9\pi^2t}$. Solve $\frac2\pi e^{-\pi^2t}=0.01$: $t=\frac{\log(200/\pi)}{\pi^2}\approx0.421$ (in the time units of the problem).
(c) Cosine coefficients of $x$ on $[0,1]$: $a_0=1$, $a_n=2\int_0^1x\cos n\pi x\,dx=\frac{2((-1)^n-1)}{n^2\pi^2}$, i.e. $-\frac{4}{n^2\pi^2}$ for odd $n$. So $$u=\frac12-\frac4{\pi^2}\sum_{n\ \mathrm{odd}}\frac{e^{-n^2\pi^2t}}{n^2}\cos n\pi x\ \to\ \frac12 .$$
(d) True The data on the parabolic boundary (initial values $0$, end values $0$ and $1$) lie in $[0,1]$, so by Theorem 10.19(1) $u$ does too. (Strictly, $u$ is discontinuous at the corner $(1,0)$, where the initial value $0$ meets the end value $1$. Replace the initial data near $x=1$ by a continuous ramp from $0$ up to $1$; the principle applies to that problem, and as the ramp shrinks the series show that its solutions converge to $u$ for $t\gt0$.)
📐 Theorem 10.20 — The vibrating string with fixed ends; energy

For $u_{tt}=c^2u_{xx}$, $0\lt x\lt L$, with $u(0,t)=u(L,t)=0$, $u(x,0)=f(x)$, $u_t(x,0)=g(x)$:

$$u=\sum_{n=1}^{\infty}\left(a_n\cos\frac{n\pi ct}{L}+\frac{L\,\beta_n}{n\pi c}\sin\frac{n\pi ct}{L}\right)\sin\frac{n\pi x}{L},$$

where $a_n$ and $\beta_n$ are the sine coefficients of $f$ and $g$. If $f\in C^2[0,L]$, $g\in C^1[0,L]$ and $f(0)=f(L)=f''(0)=f''(L)=g(0)=g(L)=0$, this is a classical solution; it is periodic in $t$ with period $\frac{2L}c$, and the $n$-th normal mode has frequency $\frac{nc}{2L}$. The energy

$$E(t)=\frac12\int_0^L\left(u_t^2+c^2u_x^2\right)dx$$

is constant in time, and consequently the problem has at most one classical solution.

Proof. Separation gives the Dirichlet eigenfunctions in $x$ and $T''+c^2\lambda_nT=0$, so $T=\cos\frac{n\pi ct}L$ and $\sin\frac{n\pi ct}L$; matching $u$ and $u_t$ at $t=0$ fixes the coefficients ($\frac{n\pi c}{L}\cdot\frac{L\beta_n}{n\pi c}=\beta_n$). Under the stated conditions the odd $2L$-periodic extension $F$ of $f$ is $C^2$ and that of $g$, $G$, is $C^1$, and the series sums to

$$\tfrac12\big(F(x-ct)+F(x+ct)\big)+\tfrac1{2c}\int_{x-ct}^{x+ct}G ,$$

which is $C^2$ — this is d'Alembert's formula of §10.3. Each term has period $\frac{2L}c$ in $t$. Energy: using $u_{tt}=c^2u_{xx}$,

$$E'=\int_0^L\left(u_tu_{tt}+c^2u_xu_{xt}\right)dx$$ $$=c^2\int_0^L\left(u_tu_{xx}+u_xu_{xt}\right)dx=c^2\big[u_tu_x\big]_0^L=0,$$

because $u_t=0$ at the fixed ends. The difference of two solutions has $E(0)=0$, hence $E\equiv0$, so $w_t=w_x=0$ and $w$ is constant, equal to its value $0$ at $t=0$. ∎

✏️ Example 7 — The plucked string
A string of length $\pi$ with $c=1$ is pulled up to height $1$ at its midpoint and released from rest: $f(x)=\frac{2x}\pi$ on $\left[0,\frac\pi2\right]$, $f(x)=\frac{2(\pi-x)}\pi$ on $\left[\frac\pi2,\pi\right]$, $g=0$. (a) Find the series solution. Which harmonics are missing, and how large is the third relative to the first? (b) Show that the string is exactly flat at $t=\frac\pi2$ and that $u(x,\pi)=-f(x)$. (c) Find the energy.
(a) Coefficients Integrating by parts on the two halves, $a_n=\frac2\pi\int_0^\pi f(x)\sin nx\,dx=\frac{8}{\pi^2n^2}\sin\frac{n\pi}{2}$. So $$u=\frac8{\pi^2}\left(\cos t\sin x-\frac{\cos3t\sin3x}{9}+\frac{\cos5t\sin5x}{25}-\cdots\right).$$
(a) Harmonics All even harmonics are missing (plucking at the midpoint, a node of every even mode, cannot excite them), and the third harmonic has $\frac19$ of the amplitude of the fundamental. The coefficients decay like $\frac1{n^2}$ because $f$ has a corner.
(b) At $t=\frac\pi2$, $\cos\frac{n\pi}2=0$ for every odd $n$, so $u\left(x,\frac\pi2\right)\equiv0$: all the energy is kinetic. At $t=\pi$, $\cos n\pi=-1$ for odd $n$, so $u(x,\pi)=-f(x)$: the string is inverted, and at $t=2\pi$ it is back where it started.
(c) At $t=0$, $u_t=0$ and $u_x=f'=\pm\frac2\pi$, so $E=\frac12\int_0^\pi\frac4{\pi^2}\,dx=\frac2\pi$, and by Theorem 10.20 it keeps this value. (Check with Parseval at $t=0$: $\frac12\cdot\frac\pi2\sum n^2a_n^2=\frac\pi4\cdot\frac{64}{\pi^4}\cdot\frac{\pi^2}8=\frac2\pi$, using $\sum_{n\ \mathrm{odd}}\frac1{n^2}=\frac{\pi^2}8$.)
⚠️ Counterexamples and pitfalls
  1. Which $B$? With $Au_{xx}+Bu_{xy}+Cu_{yy}$ the test is $B^2-4AC$; with $Au_{xx}+2Bu_{xy}+Cu_{yy}$ it is $B^2-AC$. Mixing the conventions turns $u_{xx}+2u_{xy}+2u_{yy}$ (elliptic: $4-8\lt0$) into a "hyperbolic" equation ($4-2\gt0$).
  2. Type is local. $u_{xx}+2xu_{xy}+yu_{yy}=0$ is of all three types (Example 1(a)); $u_{xx}+(x^4+y^2)u_{yy}=0$ is parabolic only at the origin.
  3. Lower-order terms are irrelevant to the type, and so is the sign of the whole equation: $-u_{xx}-u_{yy}+5u=0$ is elliptic.
  4. The sign in the characteristic equation. It is $A\lambda^2-B\lambda+C=0$ for $\lambda=\frac{dy}{dx}$. For $u_{xx}+u_{xy}-6u_{yy}=0$ the characteristics are $y-3x=c$ and $y+2x=c$.
  5. A parabolic reduction needs $J\ne0$. In Example 3(a), $\eta=\frac yx$ again would give $J=0$; any independent choice such as $\eta=x$ works.
  6. A change of variables cannot change the type ($\bar\delta=J^2\delta$), so no substitution turns Laplace's equation into $u_{\xi\eta}=0$.
  7. Forgetting $\lambda=0$. In Neumann problems the constant mode $\frac{a_0}2$ is an eigenfunction; dropping it gives an insulated rod whose temperature tends to $0$ instead of to its mean.
  8. Disc versus annulus. In a disc, boundedness at the centre removes $r^{-n}$ and $\log r$; in an annulus they must be kept. Periodicity in $\theta$ (not a boundary condition) forces $\lambda=n^2$.
  9. Neumann problems need compatibility. $u_r(1,\theta)=\cos^2\theta$ has no solution in the unit disc; when a solution exists it is unique only up to an additive constant.
  10. The maximum principle is special to $\Delta u=0$ (and $\Delta u\ge0$). $u=\sin x\sin y$ solves $\Delta u+2u=0$ on $(0,\pi)^2$, vanishes on the boundary and is positive inside.
  11. Heat backwards is ill-posed. $u=\frac1ne^{-n^2t}\sin nx$ has initial data of size $\frac1n$ but at $t=-1$ has size $\frac{e^{n^2}}{n}$: tiny changes in data cause enormous changes in the past.
  12. Laplace with Cauchy data is ill-posed (Hadamard): $u=\frac{1}{n^2}\sinh nx\sin ny$ has $u(0,y)=0$ and $u_x(0,y)=\frac{\sin ny}{n}\to0$, yet $u(1,y)$ is of size $\frac{e^n}{2n^2}$.
  13. Gibbs. The sine series of $f\equiv1$ on $(0,\pi)$ converges to $1$ inside but not uniformly; its partial sums overshoot to about $1.18$ near the ends.
📚 Go deeper — full proofs in the CSIR NET course

Two Labs

Interactive 1 — Type Map and Characteristic Curves

The plane is shaded by the sign of $\delta=B^2-4AC$: hyperbolic, parabolic, elliptic. The short strokes are the characteristic directions at grid points (two, one, or none). Through the point $P$ the characteristic curves are traced as far as they stay where $\delta\ge0$. Press Play to move $P$ around, or click the picture to place it.

point P—
A, B, C—
δ = B² − 4AC—
type at P—
slopes dy/dx—
Interactive 2 — One Fourier Series, Three Equations

The initial profile $f$ on $[0,\pi]$ (dashed) is expanded in $N$ sine modes $b_n\sin nx$. Each separated solution multiplies the $n$-th mode by its own factor: $e^{-n^2t}$ for the heat equation $u_t=u_{xx}$, $\cos nt$ for the wave equation $u_{tt}=u_{xx}$ (released from rest), and $\frac{\sinh n(\pi-t)}{\sinh n\pi}$ for Laplace's equation in the square, with the data on the bottom edge and $t$ playing the role of $y$. The thick curve is $u$ at the current $t$; the bars show $|b_n\times\text{factor}|$ against $|b_n|$ (grey). Press Play.

t—
factors n = 1, 2, 3—
max |u|—
∫ u² dx (Parseval)—

Exercises

I
Classification and characteristics.
  • 1 Classify: (a) $u_{xx}+6u_{xy}+9u_{yy}=0$; (b) $2u_{xx}-u_{xy}-u_{yy}+u_x=0$; (c) $u_{xx}+u_{xy}+u_{yy}=0$; (d) $u_{xy}=u$.
  • 2 Where is $(1+y)u_{xx}+2u_{xy}+(1-y)u_{yy}=0$ hyperbolic, parabolic, elliptic?
  • 3 Where is $x\,u_{xx}+y\,u_{yy}=0$ hyperbolic, parabolic, elliptic? What happens at the origin?
  • 4 Find the characteristics of $u_{xx}-2u_{xy}-3u_{yy}=0$.
  • 5 Find the characteristics of $y^2u_{xx}-x^2u_{yy}=0$ in the first quadrant, and check them with Theorem 10.10(1).
  • 6 Show that the characteristics of the wave equation $u_{tt}=c^2u_{xx}$ are the lines $x\pm ct=\text{const}$ and those of the heat equation $u_t=ku_{xx}$ are the lines $t=\text{const}$.
  • 7 Prove that if $A,B,C$ are continuous, the set where the equation is elliptic is open. Give an equation whose parabolic set is a single point at which the equation is still of second order.
1 (a) $36-36=0$: parabolic. (b) $1+8=9$: hyperbolic. (c) $1-4=-3$: elliptic. (d) $A=C=0$, $B=1$: $\delta=1$, hyperbolic.
2 $\delta=4-4(1+y)(1-y)=4y^2$: hyperbolic for $y\ne0$, parabolic on the $x$-axis (where $A=C=1$), elliptic nowhere.
3 $\delta=-4xy$: hyperbolic in the second and fourth quadrants, elliptic in the first and third, parabolic on the axes minus the origin. At the origin $A=B=C=0$: the equation is not of second order there and has no type.
4 $\lambda^2+2\lambda-3=(\lambda+3)(\lambda-1)=0$: $y-x=c_1$ and $y+3x=c_2$.
5 $y^2\lambda^2-x^2=0$, $\lambda=\pm\frac xy$: $y\,dy=\pm x\,dx$, so $y^2-x^2=c_1$ and $x^2+y^2=c_2$ (hyperbolas and circles). For $\varphi=y^2\mp x^2$: $y^2\varphi_x^2-x^2\varphi_y^2=4x^2y^2-4x^2y^2=0$. ✓
6 In the $(x,t)$-plane the wave equation $c^2u_{xx}-u_{tt}=0$ gives $c^2\lambda^2-1=0$ for $\lambda=\frac{dt}{dx}$: $t\mp\frac xc=$ const, i.e. $x\pm ct=$ const. The heat equation $ku_{xx}-u_t=0$ has $A=k$, $B=C=0$: $k\lambda^2=0$, so $\frac{dt}{dx}=0$ and $t=$ const.
7 $\delta$ is continuous, so $\{\delta\lt0\}=\delta^{-1}((-\infty,0))$ is open. $u_{xx}+(x^4+y^2)u_{yy}=0$: $\delta=-4(x^4+y^2)$ vanishes only at $(0,0)$, where $A=1$.
II
Canonical forms and general solutions.
  • 1 Find the general solution of $u_{xx}+5u_{xy}+6u_{yy}=0$.
  • 2 Find the general solution of $4u_{xx}+4u_{xy}+u_{yy}=0$.
  • 3 Solve $u_{xx}-u_{xy}-2u_{yy}=0$ with $u(x,0)=x^2$, $u_y(x,0)=0$.
  • 4 Reduce $u_{xx}+2u_{xy}+5u_{yy}=0$ to canonical form.
  • 5 Reduce $u_{xx}-2x\,u_{xy}+x^2u_{yy}=0$ to canonical form.
  • 6 Reduce $u_{xx}-y\,u_{yy}=0$ to canonical form in $y\gt0$.
  • 7 Verify that $u=f(x+y)+g(x-y)$ solves $u_{xx}=u_{yy}$, and find the solution with $u(x,0)=\sin x$, $u_y(x,0)=0$.
1 $m^2-5m+6=0$, $m=2,3$: $u=f(y-2x)+g(y-3x)$.
2 $4m^2-4m+1=(2m-1)^2$, $m=\frac12$: $u=f(2y-x)+x\,g(2y-x)$.
3 $m^2+m-2=0$, $m=1,-2$: $u=f(y-x)+g(y+2x)$. The data give $f(-x)+g(2x)=x^2$ and $f'(-x)+g'(2x)=0$; differentiating the first, $-f'(-x)+2g'(2x)=2x$, so $3g'(2x)=2x$, $g(w)=\frac{w^2}6$, $f(s)=\frac{s^2}3$. Hence $u=\frac{(y-x)^2}3+\frac{(y+2x)^2}6=x^2+\frac{y^2}2$. Check: $2-0-2\cdot1=0$. ✓
4 $m^2-2m+5=0$, $m=1\pm2i$: $\alpha=y-x$, $\beta=2x$. Then $u_{xx}=w_{\alpha\alpha}-4w_{\alpha\beta}+4w_{\beta\beta}$, $u_{xy}=-w_{\alpha\alpha}+2w_{\alpha\beta}$, $u_{yy}=w_{\alpha\alpha}$, and the equation becomes $4(w_{\alpha\alpha}+w_{\beta\beta})=0$.
5 $\delta=4x^2-4x^2=0$; $(m+x)^2=0$, $m=-x$: $\xi=y+\frac{x^2}2$, $\eta=x$. Then $u_x=xw_\xi+w_\eta$ and the equation becomes $w_{\eta\eta}+w_\xi=0$, i.e. $w_{\eta\eta}=-w_\xi$ — a backward heat equation in disguise.
6 $m^2=y$, $m=\pm\sqrt y$: $\frac{dy}{\sqrt y}=\pm dx$, so $\xi=x+2\sqrt y$, $\eta=x-2\sqrt y$. One finds $u_{xx}-yu_{yy}=4w_{\xi\eta}+\frac{w_\xi-w_\eta}{2\sqrt y}$, and with $4\sqrt y=\xi-\eta$: $$w_{\xi\eta}=-\frac{w_\xi-w_\eta}{2(\xi-\eta)}.$$
7 $u_{xx}=f''+g''=u_{yy}$. Data: $f(x)+g(x)=\sin x$, $f'(x)-g'(x)=0$, so $f=g=\frac12\sin$ (up to cancelling constants) and $u=\frac12\left(\sin(x+y)+\sin(x-y)\right)=\sin x\cos y$.
III
Laplace's equation.
  • 1 Solve $\Delta u=0$ in $(0,\pi)^2$ with $u=0$ on three sides and $u(x,\pi)=3\sin2x$.
  • 2 Write down the solution in $(0,2)\times(0,1)$ with $u(x,0)=f(x)$ and $u=0$ on the other three sides.
  • 3 Solve $\Delta u=0$ in the unit disc with $u(1,\theta)=1+2\sin\theta-\cos2\theta$; give $u$ in Cartesian form and $u\left(\tfrac12,0\right)$.
  • 4 Solve the Dirichlet problem in the unit disc with $u(1,\theta)=\cos^2\theta$, and find $u(0,0)$.
  • 5 In the annulus $1\lt r\lt2$, find the harmonic function with $u=0$ on $r=1$ and $u=1$ on $r=2$, and its value on $r=\sqrt2$.
  • 6 For which constants $c$ is the Neumann problem $u_r(1,\theta)=\sin\theta+c$ in the unit disc solvable? Solve it.
  • 7 Prove that the Dirichlet problem for $\Delta u=0$ in a bounded domain has at most one solution, and show by an example that the corresponding statement fails for $\Delta u+2u=0$.
1 Only $n=2$ is present: $u=\dfrac{3\sinh2y\,\sin2x}{\sinh2\pi}$.
2 With $a=2$, $b=1$ and $y$ replaced by $1-y$: $$u=\sum_{n\ge1}b_n\frac{\sinh\left(n\pi(1-y)/2\right)}{\sinh(n\pi/2)}\sin\frac{n\pi x}2,$$ $b_n=\int_0^2f(x)\sin\frac{n\pi x}{2}\,dx$.
3 $u=1+2r\sin\theta-r^2\cos2\theta=1+2y-x^2+y^2$. $u\left(\tfrac12,0\right)=1-\tfrac14=\tfrac34$.
4 $\cos^2\theta=\frac12+\frac12\cos2\theta$, so $u=\frac12+\frac{r^2}2\cos2\theta=\frac12+\frac{x^2-y^2}2$, and $u(0,0)=\frac12$.
5 By uniqueness and rotational symmetry $u$ is radial, so $u=\alpha+\beta\log r$; the data give $u=\dfrac{\log r}{\log2}$, which equals $\frac12$ on $r=\sqrt2$.
6 $\int_{-\pi}^{\pi}(\sin\theta+c)\,d\theta=2\pi c=0$ forces $c=0$. Then $u=r\sin\theta+C=y+C$.
7 The difference of two solutions is harmonic with zero boundary values; by the maximum principle its maximum and minimum are $0$. For $\Delta u+2u=0$ on $(0,\pi)^2$, both $0$ and $\sin x\sin y$ solve the problem with zero boundary values.
IV
The heat equation.
  • 1 Solve $u_t=u_{xx}$ on $(0,\pi)$ with $u(0,t)=u(\pi,t)=0$ and $u(x,0)=2\sin x+5\sin3x$.
  • 2 Solve $u_t=ku_{xx}$ on $(0,L)$ with zero end temperatures and $u(x,0)=1$. Which term dominates for large $t$?
  • 3 Solve $u_t=u_{xx}$ on $(0,\pi)$ with insulated ends and $u(x,0)=\cos^2x$. What is $\lim_{t\to\infty}u$?
  • 4 Solve $u_t=u_{xx}$ on $(0,1)$ with $u(0,t)=10$, $u(1,t)=30$ and $u(x,0)=10+20x+\sin\pi x$.
  • 5 Modelling. A steel bar $50$ cm long ($k=0.15$ cm²/s) is at $100\,^\circ$C when both ends are plunged into ice water ($0\,^\circ$C). Using the first term of the series, estimate the time until the centre reaches $50\,^\circ$C. Is the second nonzero term negligible then?
  • 6 Prove that $\int_0^L u(x,t)^2\,dx$ is non-increasing for a solution with zero end temperatures, and deduce uniqueness.
1 $u=2e^{-t}\sin x+5e^{-9t}\sin3x$.
2 $b_n=\frac2L\int_0^L\sin\frac{n\pi x}L\,dx=\frac{2(1-(-1)^n)}{n\pi}$: $$u=\frac4\pi\sum_{n\ \mathrm{odd}}\frac1ne^{-kn^2\pi^2t/L^2}\sin\frac{n\pi x}L.$$ For large $t$ the $n=1$ term dominates: $u\approx\frac4\pi e^{-k\pi^2t/L^2}\sin\frac{\pi x}L$.
3 $\cos^2x=\frac12+\frac12\cos2x$: $u=\frac12+\frac12e^{-4t}\cos2x\to\frac12$, the mean.
4 Steady state $10+20x$; the remainder starts as $\sin\pi x$: $u=10+20x+e^{-\pi^2t}\sin\pi x$.
5 $u(25,t)\approx\frac{400}{\pi}e^{-\mu t}$ with $\mu=\frac{0.15\,\pi^2}{2500}\approx5.92\times10^{-4}$ s⁻¹. Setting this equal to $50$: $t=\frac{\log(8/\pi)}{\mu}\approx1578$ s $\approx26$ min. The $n=3$ term is then $-\frac{400}{3\pi}e^{-9\mu t}\approx-0.01\,^\circ$C: negligible.
6 Theorem 10.19(2): $E'=-2k\int u_x^2\le0$. If $u_1,u_2$ solve the same problem, $w=u_1-u_2$ has zero data, so $0\le E_w(t)\le E_w(0)=0$ and $w\equiv0$ (it is continuous).
V
The wave equation.
  • 1 Solve $u_{tt}=4u_{xx}$ on $(0,\pi)$ with fixed ends, $u(x,0)=\sin3x$ and $u_t(x,0)=\sin x$.
  • 2 Modelling. A string $0.5$ m long under tension $80$ N has linear density $0.002$ kg/m, so $c=\sqrt{T/\rho}$. Find the frequencies of its first three normal modes.
  • 3 A string of length $L$ is plucked to height $h$ at $x=a$ and released from rest. Show that $a_n=\dfrac{2hL^2}{\pi^2n^2a(L-a)}\sin\dfrac{n\pi a}{L}$, and deduce which harmonics are absent when $a=\frac L3$.
  • 4 For a string released from rest ($g=0$), prove that $u\left(x,\frac Lc\right)=-f(L-x)$.
  • 5 Compute the energy $E(t)$ of $u=\sin x\cos t$ on $(0,\pi)$ ($c=1$) and check that it is constant.
  • 6 For the damped equation $u_{tt}+2u_t=u_{xx}$ on $(0,\pi)$ with fixed ends, find the time factors $T_n(t)$ of the separated solutions $T_n(t)\sin nx$.
1 $c=2$: $u=\sin3x\cos6t+\frac12\sin x\sin2t$ (the coefficient of $\sin x\sin2t$ is $\frac{1}{1\cdot2}$).
2 $c=\sqrt{80/0.002}=200$ m/s; $f_n=\frac{nc}{2L}=200n$ Hz: $200$, $400$ and $600$ Hz.
3 $a_n=\frac2L\left(\int_0^a\frac{hx}a\sin\frac{n\pi x}L\,dx+\int_a^L\frac{h(L-x)}{L-a}\sin\frac{n\pi x}L\,dx\right)$; integrating by parts, the boundary terms cancel and the formula follows. For $a=\frac L3$: $a_n=\frac{9h}{\pi^2n^2}\sin\frac{n\pi}3$, which vanishes for $n=3,6,9,\dots$.
4 $u\left(x,\frac Lc\right)=\sum a_n\cos(n\pi)\sin\frac{n\pi x}L=\sum(-1)^na_n\sin\frac{n\pi x}L$, while $f(L-x)=\sum a_n\sin\left(n\pi-\frac{n\pi x}L\right)=-\sum(-1)^na_n\sin\frac{n\pi x}L$.
5 $E=\frac12\int_0^\pi\left(\sin^2x\sin^2t+\cos^2x\cos^2t\right)dx=\frac12\cdot\frac\pi2\left(\sin^2t+\cos^2t\right)=\frac\pi4$.
6 $T''+2T'+n^2T=0$, roots $-1\pm\sqrt{1-n^2}$. $n=1$: $T_1=(A+Bt)e^{-t}$ (critically damped). $n\ge2$: $T_n=e^{-t}\left(A\cos\sqrt{n^2-1}\,t+B\sin\sqrt{n^2-1}\,t\right)$ — every mode decays like $e^{-t}$.
🎯 GATE corner GATE MA · MCQGATE MA · MSQGATE MA · NAT

E1. GATE MA · MCQ The equation $y\,u_{xx}+2x\,u_{xy}+y\,u_{yy}=0$ is hyperbolic exactly in the region (A) $|x|\gt|y|$; (B) $|x|\lt|y|$; (C) $xy\gt0$; (D) $x^2+y^2\gt1$.

Idea $A=y$, $B=2x$, $C=y$, so $\delta=4x^2-4y^2=4(x-y)(x+y)$.
Decide $\delta\gt0\iff x^2\gt y^2\iff|x|\gt|y|$. It is elliptic for $|x|\lt|y|$ and parabolic on the lines $y=\pm x$ minus the origin (at the origin all coefficients vanish).
Answer (A). Option (B) is the elliptic region.

E2. GATE MA · MCQ Let $u$ solve $u_t=2u_{xx}$ for $0\lt x\lt\pi$, $t\gt0$, with $u(0,t)=u(\pi,t)=0$ and $u(x,0)=3\sin x-\sin2x$. Then $u\left(\frac\pi2,1\right)$ equals (A) $3e^{-2}$; (B) $3e^{-1}$; (C) $3e^{-2}-e^{-8}$; (D) $3e^{-4}$.

Idea With $k=2$, the mode $\sin nx$ decays like $e^{-2n^2t}$: $u=3e^{-2t}\sin x-e^{-8t}\sin2x$.
Evaluate At $x=\frac\pi2$, $\sin2x=\sin\pi=0$, so $u=3e^{-2}\approx0.406$.
Options (B) forgets $k=2$; (C) forgets that $\sin\pi=0$; (D) squares $k$, using $e^{-k^2n^2t}$.
Answer (A).

E3. GATE MA · MSQ CSIR Part C Let $u$ be harmonic in the open unit disc, continuous on the closed disc, with $u(\cos\theta,\sin\theta)=f(\theta)$, $f$ continuous. Which of the following statements is/are true? (One or more options may be correct.) (A) $u(0,0)=\frac1{2\pi}\int_0^{2\pi}f(\theta)\,d\theta$. (B) If $f\ge0$ and $f\not\equiv0$, then $u\gt0$ in the open disc. (C) If $f(\theta)=\cos3\theta$, then $u(x,y)=x^3-3xy^2$. (D) If $f$ is not constant, $|u|$ may attain its maximum over the closed disc at the centre.

(A) True Mean value property (Theorem 10.15(3)), or $u(0)=\frac{a_0}2$ in Theorem 10.17.
(B) True The Poisson kernel is positive, and $f\ge0$ is positive on some arc, so the integral in Poisson's formula is positive.
(C) True $u=r^3\cos3\theta=\mathrm{Re}(x+iy)^3=x^3-3xy^2$, harmonic with the right boundary values.
(D) False If $|u|$ were largest at the centre, then $u$ or $-u$ would attain an interior maximum, so $u$ would be constant (strong maximum principle) and so would $f$.
Answer (A), (B) and (C).

E4. GATE MA · MSQ Consider $u_{xx}-4u_{xy}+4u_{yy}=0$. Which of the following statements is/are true? (One or more options may be correct.) (A) The equation is parabolic everywhere. (B) $u=(y+2x)^3$ is a solution. (C) $u=x^2(y+2x)$ is a solution. (D) $u=e^{y-2x}$ is a solution.

(A) True $\delta=16-16=0$. The characteristic equation $m^2+4m+4=0$ has the double root $m=-2$, so by Theorem 10.12 the solutions are $f(y+2x)+x\,g(y+2x)$.
(B) True It is $f(y+2x)$ with $f(s)=s^3$.
(C) False With $s=y+2x$: $u_{xx}=2s+8x$, $u_{xy}=2x$, $u_{yy}=0$, so the left side is $2s+8x-8x=2(y+2x)\ne0$. Only the factors $1$ and $x$ are allowed in front of functions of $y+2x$.
(D) False $u_{xx}=4u$, $u_{xy}=-2u$, $u_{yy}=u$: the left side is $(4+8+4)u=16u\ne0$. ($y-2x$ is not the characteristic variable.)
Answer (A) and (B).

E5. GATE MA · NAT Let $u$ be harmonic in the unit square $0\lt x,y\lt1$, with $u=0$ on the sides $x=0$, $x=1$, $y=0$ and $u(x,1)=\sin\pi x+\sin2\pi x$. The value of $u\left(\tfrac14,\tfrac12\right)$ is ______ (correct to two decimal places).

Solution Theorem 10.16 with $a=b=1$ and only two modes: $$u=\frac{\sinh\pi y}{\sinh\pi}\sin\pi x+\frac{\sinh2\pi y}{\sinh2\pi}\sin2\pi x .$$
Evaluate At $\left(\frac14,\frac12\right)$: $\frac{\sinh(\pi/2)}{\sinh\pi}\cdot\frac{\sqrt2}{2}+\frac{\sinh\pi}{\sinh2\pi}\cdot1=\frac{\sqrt2}{4\cosh(\pi/2)}+\frac{1}{2\cosh\pi}$.
Answer $\approx0.1409+0.0431=0.1840$. Answer: $0.18$ (accept $0.17$ to $0.19$).

E6. GATE MA · NAT Let $u$ solve $u_{tt}=9u_{xx}$ for $0\lt x\lt\pi$ with $u(0,t)=u(\pi,t)=0$, $u(x,0)=0$ and $u_t(x,0)=\sin x+\sin2x$. The value of $u\left(\frac\pi4,\frac\pi6\right)$ is ______ (correct to three decimal places).

Solution $c=3$; Theorem 10.20 with $a_n=0$, $\beta_1=\beta_2=1$: $u=\frac13\sin3t\sin x+\frac16\sin6t\sin2x$.
Evaluate At $t=\frac\pi6$: $\sin\frac\pi2=1$ and $\sin\pi=0$, so $u=\frac13\sin\frac\pi4=\frac{\sqrt2}{6}$.
Answer $\frac{\sqrt2}6\approx0.2357$. Answer: $0.236$ (accept $0.235$ to $0.237$).
🧑‍🏫 For the teacher

Suggested plan — 3 lectures × 60 min.

  1. Classification and canonical forms. Definitions 10.6–10.7 with the conic analogy; Theorem 10.9 via $\bar M=PMP^{\mathsf T}$; Example 1. Project Interactive 1 on the first equation and ask where the strokes cross, merge and disappear. Definition 10.8, Theorems 10.10–10.12, the strategy box, Examples 2 and 3. GATE corner E1 and E4 live.
  2. Separation of variables and Laplace's equation. Derive the eigenvalue problem from the heat equation; Theorem 10.13 (do the Dirichlet case on the board), Definition 10.9, Theorem 10.14 (statement). Theorem 10.15 with the $\varepsilon(x^2+y^2)$ proof, Theorems 10.16–10.17, Examples 4 and 5. The symmetry argument of Example 4(b) is a good "aha" moment.
  3. Heat and wave on an interval. Theorems 10.18–10.20 with the energy proofs, Examples 6 and 7. Project Interactive 2 with the square pulse: heat (corners vanish at once, high modes die first), wave (corners survive and travel), Laplace (in between). Finish with E2, E5 and E6.

Discussion questions. (1) Why can a substitution never turn Laplace's equation into the wave equation, while it can turn $u_{xx}-u_{yy}=0$ into $u_{\xi\eta}=0$? (2) In Interactive 2, why does the wave keep the corner of the plucked string forever while the heat equation rounds it off at once — and what does that say about how fast $b_n$ must decay for a smooth function? (3) An insulated rod tends to its mean temperature; a rod with ends at $0$ tends to $0$. Which eigenvalue makes the difference?

Misconceptions to watch for. Using $B^2-AC$ with the full coefficient $B$; calling the heat equation first order and therefore "not second order"; the sign error in $A\lambda^2-B\lambda+C=0$; choosing two dependent coordinates in the parabolic case; forgetting the constant mode in Neumann problems; keeping $r^{-n}$ in a disc or dropping it in an annulus; applying the maximum principle to $\Delta u+cu=0$; expecting the heat series to converge at $t=0$ as fast as for $t\gt0$.

Suggested homework. I(1, 2, 5, 7), II(2, 3, 5, 6), III(1, 3, 5, 6, 7), IV(2, 3, 4, 5), V(1, 3, 4, 6); GATE corner E3, E4, E5.

🔬 Research–Teaching NexusResearch-tutored · optional, about 60 minutes

Research question. A guitarist plucking near the bridge gets a bright, thin sound; plucking near the middle of the string gives a round, mellow one. Can the separated solution of the wave equation explain this, and predict which harmonics a given plucking point removes?

Why it matters. The timbre of a note is its distribution of energy over the normal modes $\sin\frac{n\pi x}L$. Instrument makers, from sitar and guitar builders to piano designers (who place the hammer near $\frac18$ of the string length), choose excitation points by exactly this reasoning; the same mode analysis is used to tune bridges and turbine blades so that dangerous modes are not excited.

What to do.

  1. Use Exercise V3: for a string plucked at $x=a$, $a_n\propto\frac{1}{n^2}\sin\frac{n\pi a}L$. Show that the energy in mode $n$ is proportional to $n^2a_n^2\propto\frac{\sin^2(n\pi a/L)}{n^2}$.
  2. Compute the fraction of the total energy carried by the fundamental for $a=\frac L2$ and for $a=\frac L5$. (Hint: $\sum_{n\ge1}\frac{\sin^2 n\theta}{n^2}=\frac{\theta(\pi-\theta)}{2}$ for $0\le\theta\le\pi$.)
  3. In Interactive 2 choose "wave" and the two plucked profiles; compare the bar charts and the motion. Which harmonics are missing in each case?
  4. If you have a guitar (or a phone spectrum-analyser app), pluck one string at its midpoint and at one fifth of its length and compare the spectra with your predictions.

Tools: pencil and paper, Interactive 2, optionally a spreadsheet and a spectrum-analyser app.

Hand in: the energy formula, the two percentages with working, screenshots or sketches of the two spectra, and a paragraph answering the research question.

Going further. A real string is stiff, and its equation gains a term $-\kappa u_{xxxx}$. Separate variables again and show that the overtones are no longer exact multiples of the fundamental ("inharmonicity"), which is why piano tuners stretch the octaves.

Instructor note: the energy of mode $n$ is $\frac{L}{4}\left(\frac{n\pi c}{L}\right)^2a_n^2$, proportional to $n^2a_n^2\propto\frac{\sin^2(n\pi a/L)}{n^2}$. With $\theta=\frac{\pi a}L$ the total is $\frac{\theta(\pi-\theta)}2$. For $a=\frac L2$: total $\frac{\pi^2}8$, fundamental $1$, share $\frac8{\pi^2}\approx81.1\%$, even harmonics absent. For $a=\frac L5$: total $0.08\pi^2\approx0.7896$, fundamental $\sin^2\frac\pi5\approx0.3455$, share $\approx43.8\%$; harmonics $5,10,15,\dots$ are absent and much more energy sits in the upper harmonics — the "bright" sound. Stiff string: $\omega_n^2=c^2\left(\frac{n\pi}L\right)^2+\kappa\left(\frac{n\pi}L\right)^4$, so $\omega_n/\omega_1$ exceeds $n$.

🧪 Self-check quiz

0/5 correct

Q1. The equation $u_{xx}+4u_{xy}+3u_{yy}+u_x=0$ is

Q2. The eigenvalues of $X''+\lambda X=0$, $X'(0)=X'(\pi)=0$ are

Q3. For $u_t=u_{xx}$ on $(0,\pi)$ with zero end temperatures, the mode $\sin3x$ decays like

Q4. $u$ is harmonic in the unit disc, continuous up to the boundary, with $u(1,\theta)=2+\cos\theta$. Then $u(0,0)$ is

Q5. A string of length $1$ with $c=2$ and fixed ends, released from rest, returns to its initial shape for the first time (for a general initial shape) after time