Trinh @ Bath

This is an old revision of the document!


Lecture 33: Uniqueness of solutions I

In this lecture, we'll prove uniqueness for two variants of the heat equation. The different variants you should consider as as follows (all posed on x[0,L]x[0,L]): ut=κuxx,u(x,0)=0,ut=κuxx,u(x,0)=0, with either Dirichlet conditions [D] u=T0(x=0),u(L,t)=T1(x=L),u=T0(x=0),u(L,t)=T1(x=L), or Neumann conditions [N] ux=G0(x=0),ux=G1(x=L),ux=G0(x=0),ux=G1(x=L), or mixed conditions [M] A0u+B0ux=H0(x=0),A1u+B1ux=H1(x=L).A0u+B0ux=H0(x=0),A1u+B1ux=H1(x=L).

Section 19.1: Uniqueness for zero Dirichlet heat equation

Let's show uniqueness of solutions to the heat equation with Dirichlet boundary conditions.

Let u, v be distinct solutions and set w=uv. Then w satisfies wt=κwxx,w(0,t)=T0T0=0,w(L,t)=0,w(x,0)=0,

Essentially this is a heat problem where the system begins at zero heat, has BCs with zero heat everywhere; it is sensible that the only solution is trivial. To prove it, we can manipulate the equation in the following way.

wt=κwxxwwt=κwwxxt12w2=κwwxxtL012w2dx=κL0wwxxdx

Let's call the quantity E(t)=L012w2dx, the energy.

Now use the fact that x(wwx)=w2x+wwxx to re-write the RHS. We then get after integrating once, E(t)=κwwx|L0κL0w2xdx=κL0w2xdx where the second equality occurs because w(0,t)=0=w(L,t) [*]. So in the end, we have the fact that E(t)=κL0w2xdx0, so the energy is always decreasing. But note that E(0)=0 since w(x,0)=0. Finally, note that E(t) is always 0 by its form (the integral of a squared quantity). So the energy is always decreasing, begins from zero, and can never be negative. We have the three statements:

  1. E(t)0 for all time
  2. E(0)=0
  3. E(t)0 for all time

Therefore E(t)0 for all time. Looking at the form of the integrand, you would conclude that w2(x,t)=0 for all x[0,L] and for all t0. So u(x,t)v(x,t) and the solutions must be the same.

Section 19.2: Uniqueness for other BCs of the heat equation

The proof of uniqueness for the Neumann condition [N] is done identically to the above, except in the step marked with [*] we use the fact that wx(0,t)=0=wx(L,t). But you would arrive at the same conclusions as above.