Continuous dependence of the solution from the initial data - Meaning

Question

I am looking the following:

What do we mean by the sentence "continuous dependence of the solution of the heat equation from the initial data, on the interval $[0,2\pi]$ with Dirichlet boundary data $0$, for $t>0$ ? Does something equivalent hold for negative times?

$$$$

I have done the following:

Let $u_1(x,t)$ and $u_2(x,t)$ $C^2$ solutions of the problem ($0\leq x\leq 2\pi, \ t\geq 0$)

\begin{align*}&u_t=ku_{xx} \\ &u(0,t)=a_1(t) \\ &u(L,t)=b_1(t) \\ &u(x,0)=f_1(x) \end{align*} and \begin{align*}&u_t=ku_{xx} \\ &u(0,t)=a_2(t) \\ &u(L,t)=b_2(t) \\ &u(x,0)=f_2(x) \end{align*} where $L=2\pi$.

If for some $\epsilon>0$ we have $|f_1(x)-f_2(x)|<\epsilon \ \forall x$, $0\leq x\leq L$ and $|a_1(x)-a_2(x)|<\epsilon$ and $|b_1(x)-b_2(x)|<\epsilon$ $ \ \forall t$, $0\leq t\leq T$ then $|u_1(x,t)-u_2(x,T)|<\epsilon \ \forall x,t$ with $0\leq x\leq 2\pi, \ 0\leq t\leq T$.

That means that little errors at the initial data don't produce big errors at the solution.

Is everything correct so far?

Answer 1

You made already a good start.

continuous dependence of the solution of the heat equation from the initial data, on the interval $[0,2\pi]$ with Dirichlet boundary data $0$, for $t>0$

The initial data refers only to $f$, and Dirichlet boundary data $0$ means that we only need to consider the case $a(t)=b(t)=0$.

Then, as Jeff said in the comments, continuous dependence of the solution from the initial data refers to the following:

For all $\epsilon>0$ there exists an $\delta>0$ such that $$ |f_1(x)-f_2(x)|<\delta \;\forall x, 0\leq x\leq L \quad\Rightarrow\quad |u_1(x,t)-u_2(x,T)|<\epsilon \;\forall x,t, 0\leq x\leq 2\pi, \ 0\leq t\leq T $$ is true. The difference to your version is that you had the same value for $\epsilon$ and $\delta$.

However, in the case of the heat equation with $t>0$ it is possible to show the above statement by choosing $\delta=\epsilon$. (This would then be very similar to the statement in the second-to-last paragraph in your question).

So, if you are able to show that $$ |f_1(x)-f_2(x)|<\epsilon \;\forall x, 0\leq x\leq L \quad\Rightarrow\quad |u_1(x,t)-u_2(x,T)|<\epsilon \;\forall x,t, 0\leq x\leq 2\pi, \ 0\leq t\leq T $$ is true for all $\epsilon>0$, then you have shown that the solution depends continuously on the initial data.

Does something equivalent hold for negative times?

No, for negative $t$ the solution of the heat equation does not depend continuously on the initial data.

This can be shown with a counterexample. We consider the case $T=-1$ and that $t\in [T,0]$. We also set $k=1$. As the initial data we choose $f_1(x)=1/n \sin(nx)$ for some $n\in\mathbb N$. Then it can be shown that the solution $u$ is given by $$ u_1(x,t) = 1/n e^{-n^2 t} \sin (nx). $$ We also choose $f_2(x)=0$ with has the solution $u_2(x,t)=0$.

Then it can be shown that $$ \max_{0\leq x\leq L} | f_1(x)-f_2(x) | = 1/n \qquad\text{and}\qquad \max_{0\leq x\leq L} | u_1(x,T)-f_2(x,T) | = 1/n e^{-n^2 T} = 1/n e^{n^2} $$ So for big $n$, a small error of $1/n$ in the initial data can cause a very big error at the solution at time $T=-1$.