Chapter 4 · Introduction

Chapter 4 — Second-Order Partial Differential Equations

Equations in the second derivatives of an unknown $u(x,y)$: what the sign of $B^2-4AC$ tells you, how to bring an equation to its canonical form, how to solve equations with constant coefficients, and how separation of variables solves the Laplace, heat and wave equations.

In 1747 Jean le Rond d'Alembert wrote down the equation of a vibrating string, $u_{tt}=c^2u_{xx}$, and found its general solution $u=f(x+ct)+g(x-ct)$: every motion of the string is a pair of waves running in opposite directions without changing shape. Sixty years later Joseph Fourier submitted his memoir on the conduction of heat, $u_t=k\,u_{xx}$, and claimed that every initial temperature profile could be written as a sum of sines — each of which simply decays at its own rate. And from Laplace's work on gravitation came $u_{xx}+u_{yy}=0$, the equation of steady states: equilibrium temperatures, electrostatic potentials, the velocity potential of an ideal fluid, the shape of a soap film stretched over a slightly bent wire. Three equations that differ only in signs and in one missing derivative, yet their solutions behave in completely different ways: waves keep their sharp edges and travel at finite speed; heat smooths every corner instantly and is felt everywhere at once; harmonic functions are as smooth as possible and are determined by their boundary values alone.

The reason is hidden in the second-order terms. For $Au_{xx}+Bu_{xy}+Cu_{yy}+\dots=0$ the sign of the discriminant $B^2-4AC$ decides whether the equation is hyperbolic like the wave equation, parabolic like the heat equation or elliptic like Laplace's equation. The type does not change under any smooth change of coordinates, it decides which curves are characteristics — the curves along which information and discontinuities travel — and it decides which problems are sensible: an initial value problem for a wave, an initial-boundary value problem for heat, a boundary value problem for a potential. Section 4.1 develops this classification, reduces every equation to one of three canonical forms, and solves linear equations with constant coefficients by factoring the operator. Sections 4.2 and 4.3 then solve the three model equations by Fourier's method of separation of variables, using the Sturm–Liouville eigenfunctions of Chapter 2, and prove the maximum principles and energy identities that make the solutions unique.

Au_xx + Bu_xy + Cu_yy→B² − 4AC→Hyperbolic · parabolic · elliptic→Characteristics→Canonical forms→F(D, D′)z = f: CF + PI→Separation of variables→Laplace: maximum principle→Heat & wave: Fourier series, d'Alembert
🎯 By the end of this chapter you can…
  • classify a second-order linear equation in two variables at every point, sketch its hyperbolic, parabolic and elliptic regions, and explain why the type is unchanged by a nonsingular change of variables;
  • find the characteristic curves of an equation and reduce it to its canonical form $u_{\xi\eta}=\Phi$, $u_{\eta\eta}=\Phi$ or $u_{\alpha\alpha}+u_{\beta\beta}=\Phi$, and use the canonical form to find general solutions;
  • find the complementary function of a linear equation $F(D,D')z=f$ with constant coefficients, including repeated and non-homogeneous factors, and compute particular integrals for exponential, trigonometric, polynomial and general right-hand sides;
  • use the maximum principle for harmonic functions, solve the Dirichlet problem in a rectangle and in a disc (Poisson's integral formula), and state the compatibility condition of the Neumann problem;
  • solve the heat equation on a rod with Dirichlet or Neumann ends and the vibrating-string problem by separation of variables, find steady states, and prove uniqueness by energy and maximum-principle arguments;
  • use d'Alembert's formula and the domain of dependence, and state the heat kernel solution on the whole line.
Teaser — One Bump, Two Equations: Finite versus Infinite Speed

The same initial bump $u(x,0)=(1-x^2)^2$ on $|x|\le1$ (zero outside) evolves under the wave equation $u_{tt}=u_{xx}$ (starting at rest) and under the heat equation $u_t=u_{xx}$. An observer sits at $x=3$. The wave splits into two half-bumps that travel at speed $1$ and keep their shape — the observer feels nothing until $t=2$. The heat profile spreads and flattens, and the observer's reading is positive at every $t\gt0$, however small. Press Play or drag the time slider.

0.60
time t—
wave: u(3, t)—
heat: u(3, t)—
wave: max u—
heat: max u—

How this is examined

Second-order PDEs are among the most dependable sources of CSIR NET questions. In Part B you may be asked for the region in which an equation with variable coefficients is hyperbolic (the answer is the set where $B^2-4AC\gt0$ — a half-plane, the outside of a circle or parabola, a pair of opposite quadrants); for its characteristic curves or the change of variables that reduces it to canonical form; for the general solution of an equation such as $u_{xx}-5u_{xy}+6u_{yy}=0$, or a particular integral of $F(D,D')z=e^{ax+by}$ in the awkward case $F(a,b)=0$; for the value at an interior point of the solution of a Dirichlet problem in a rectangle or disc; for the steady-state temperature of a rod or the value of a heat or wave solution at a given time; or for the value of a d'Alembert solution, which is pure arithmetic once the formula is known. Part C, where several options may be correct, tests the theory behind the formulas: whether the type can change under a change of variables, whether a solution of Laplace's equation can have an interior maximum, whether a Neumann problem is solvable, whether a heat solution can grow, whether a wave solution vanishes outside a cone, whether a separated series converges. The wrong options are almost always one of the counterexamples collected in the ⚠️ boxes of the three sections. GATE MA asks similar questions, often with a numerical answer.

Sections in this chapter

✅ Before you start

From Chapter 3 you need the idea of a characteristic curve and the solution of first-order linear equations along characteristics (§3.1, especially the theorem that a solution of $a\,p+b\,q=0$ is constant along the curves $\tfrac{dx}{a}=\tfrac{dy}{b}$), and the Cauchy problem and transversality condition (§3.2) — the second-order theory repeats it one derivative higher. From Chapter 2 you need the complementary-function-plus-particular-integral structure and the operator methods $1/f(D)$ for linear ODEs with constant coefficients (§2.1, §2.2), and for Sections 4.2–4.3 the Sturm–Liouville eigenvalue problems $X''+\lambda X=0$ with their orthogonal eigenfunctions and eigenfunction expansions (§2.3). The existence theorem for ODEs (§1.1) is used to produce characteristic coordinates. From the CSIR NET Unit 1 course you need the chain rule for functions of several variables, Jacobians and the inverse function theorem, symmetric $2\times2$ matrices and their definiteness, and uniform convergence of series of functions. If this is your first visit, the welcome page explains how each section is organised.