Is there a good reference for the following problem? Consider any smooth bounded domain $\Omega$ and solve the heat equation

\begin{align} \partial_t u^\kappa &= \kappa \Delta u^\kappa,\\ u^\kappa|_{\partial\Omega}&=0,\\ u^\kappa|_{t=0}&=u_0. \end{align}

What is the rate of convergence as $\kappa\to 0$ of $u^\kappa(t)$ to $u_0$?

  • $\begingroup$ Try to solve this firs when $\Omega$ is a rectangle (I know, it does not have smooth boundary) or $\Omega$ is a disk. In these cases you can solve the equation by separation of variables and you can draw some preliminary conclusions that could guide you. $\endgroup$ May 19, 2018 at 16:58

1 Answer 1


In a nutshell: The rate of convergence depends on the smoothness of $u_0$.

Here are the details.


  • Throughout I assume that $\kappa > 0$ is a real number, i.e. it does not depend on the spatial variable.

  • I'm going to consider the equation in the space $L^2(\Omega)$ (but similar observations also hold on $L^p(\Omega)$ for $p \in [1,\infty)$).

  • Nomenclature: For functions $f: (0,\infty) \to L^2(\Omega)$ and $r: (0,\infty) \to (0,\infty)$ I will write $f \in \mathcal{O}(r)$ if there exists a real number $c \ge 0$ such that $\|f(s)\| \le c r(s)$ for all sufficiently small $s \in (0,\infty)$ (where $\|\cdot\|$ denotes the norm in $L^2(\Omega)$).

  • Let $A$ denote the realisation of the Dirichet Laplace operator on $L^2(\Omega)$, i.e. \begin{align*} D(A) & = \{u \in H^1_0(\Omega): \; \Delta u \in L^2(\Omega)\}, \\ Au & = \Delta u; \end{align*} here $\Delta u$ is to be understood in the sense of distributions. Note that $D(A)$ coincides with $H^2(\Omega) \cap H^1_0(\Omega)$ if $\Omega$ has $C^2$-boundary (by elliptic regularity theory).

  • The operator $\kappa A$ (with domain $D(\kappa A) = D(A)$) generates a $C_0$-semigroup $(e^{t \kappa A})_{t \ge 0}$; each operator $e^{t \kappa A}$ has norm at most $1$. For every $u_0 \in L^2(\Omega)$ the (mild) solution of the heat equation in the question is given by the mapping \begin{align*} [0,\infty) \ni t \mapsto e^{t \kappa A}u_0 \in L^2(\Omega); \end{align*} this is even a classical solution if $u_0 \in D(A)$.

To answer the question we can use the following idea: If we fix a time $t \in (0,\infty)$, then the behaviour of $e^{t\kappa A}$ as $\kappa \to 0$ is the same as the behaviour of $e^{sA}$ as $s \to 0$ (set $s := t\kappa$); thus, the question is essentially "how continuous" the mapping $s \mapsto e^{sA}u_0$ is at $s = 0$.

Now, the answer consists of a negative and a positive result:

Theorem 1 (negative result). There is no rate which works for all $u_0 \in L^2(\Omega)$. More precisely: Fix $t \in (0,\infty)$. For every continuous function $r: (0,\infty) \to (0,\infty)$ which fulfils $r(s) \to 0$ as $s \to 0$ there exists a vector $u_0 \in L^2(\Omega)$ such that the mapping $\kappa \mapsto e^{t\kappa A}u_0 - u_0$ is not contained in $\mathcal{O}(r)$.

Proof. Assume to the contrary that we can find a continuous function $r: (0,\infty) \to (0,\infty)$ which fulfils $r(s) \to 0$ as $s \to 0$ as well as $\big(\kappa \mapsto e^{t\kappa A}u_0 - u_0\big) \in \mathcal{O}(r)$ for each $u_0 \in L^2(\Omega)$. Then it follows from the uniform boundedness principle that the operator family $\big((e^{t \kappa A} - \operatorname{id})/r(\kappa)\big)_{\kappa \in (0,1]}$ is bounded with respect to the operator norm (here, $\operatorname{id}$ denotes the identity operator on $L^2(\Omega)$). Hence, $e^{t\kappa A}$ converges to $\operatorname{id}$ with respect to the operator norm as $\kappa \to 0$. Consequently, $e^{sA}$ converges to $\operatorname{id}$ with respect to operator norm as $s \to 0$. By standard $C_0$-semigroup theory this can only be true if $A$ is a bounded operator, which is not fulfilled. This proves Theorem 1.

Theorem 2 (positive result). Let $u_0 \in D(A)$. Then \begin{align*} \|e^{t\kappa A} u_0 - u_0\| \le t\kappa \|\Delta u_0\| \end{align*} for all $t,\kappa \in (0,\infty)$. Hence, the convergence of $e^{t\kappa A}u_0$ to $u_0$ as $\kappa \to 0$ has a linear rate.

Proof. Fix $t,\kappa \in (0,\infty)$. Since $u_0 \in D(A)$, we have \begin{align*} e^{t\kappa A}u_0 - u_0 = \int_0^{\kappa t} e^{sA}A u_0 \, ds \end{align*} (see for instance [Engel/Nagel: One-Parameter Semigroups for Linear Evolution Equations (2000), Lemma II.1.3(iv)]). The assertion now readily follows by taking the norm on both sides of this equation (use that $Au_0 = \Delta u_0$ and that $\|e^{sA}\| \le 1$ for each $s$).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.