Consider the initial value problem $$ \partial_t u = \Delta u$$ $$ u(0,x) = 0$$ for the heat equation in $\mathbb R^n$, where $u: [0,T] \times \mathbb R^n \to \mathbb R$ is a smooth solution up to the time $T$. Suppose there exists a large constant $N>0$ such that $$ |u(t,x)| \le N\exp(N|x|^2t^{-1}) $$ for any $x\in \mathbb R^n$ and $t\in (0,T]$.

Can we prove that $u(t,x)$ must be identically $0$?

Notice that if $|u(t,x)| \le N\exp(N|x|^2)$, then the question is true by Tychonoff's theorem.


Have you tried the classical Tychonoff's example, $$u(t,x) = \sum_{k = 0}^\infty \frac{g^{(k)}(t) x^{2k}}{(2k)!},$$ with $g(t) = e^{-1/t^\alpha}$ and $\alpha > 1$?

As discussed, for example, in these lecture notes, in this case $$|g^{(k)}(t)| \leqslant \frac{k!}{(\theta t)^k} e^{-1/(2 t^\alpha)} \tag{$\star$}$$ for some $\theta > 0$ (edit: see below for a proof), and hence $$ |u(t, x)| \leqslant \sum_{k = 0}^\infty \frac{k! x^{2k}}{(\theta t)^k(2k)!} e^{-1 / (2 t^\alpha)} \leqslant \sum_{k = 0}^\infty \frac{x^{2k}}{(2 \theta t)^k k!} e^{-1 / (2 t^\alpha)} = e^{x^2/(2 \theta t)-1/(2t^\alpha)} \leqslant e^{x^2 / (2 \theta t)} . $$ If I am not mistaken, this (counter-)example implies a negative question to your answer.

Edit: For completeness, here is the proof of ($\star$). This is, of course, completely standard; I reworked it only to convince myself that there is no error in my answer.

The function $g(t) = e^{-1/t^\alpha}$ is holomorphic in the right complex half-plane. By Cauchy's formula (applied to the circle $\Gamma$ centred at $t$ with radius $\theta t$), we have $$ g^{(k)}(t) = \frac{k!}{2\pi i} \int_\Gamma \frac{g(z)}{(z - t)^{k + 1}} dz = \frac{k!}{2\pi} \int_0^{2\pi} \frac{g(t + \theta t e^{i s})}{(\theta t e^{i s})^k} ds . $$ Therefore, $$ |g^{(k)}(t)| \leqslant \frac{k!}{2\pi (\theta t)^k} \int_0^{2\pi} |g(t + \theta t e^{i s})| ds . $$ Choose $\theta > 0$ small enough, so that the image of the disk $D(1, \theta)$ under $z \mapsto z^{-\alpha}$ is contained in the half-plane $\Re z > \tfrac{1}{2}$. Since $|g(z)| = e^{-\Re z^{-\alpha}}$, we have $$|g(t + \theta t e^{i s})| = e^{-t^{-\alpha} \Re (1 + \theta e^{i s})^{-\alpha}} \leqslant e^{-t^{-\alpha} / 2} .$$ It follows that $|g^{(k)}(t)| \leqslant \frac{k!}{(\theta t)^k} e^{-1 / (2 t^\alpha)}$, as desired.

  • $\begingroup$ Kwasnicki For the Tychonoff argument to work, you need $\alpha >1$. In fact, up to geometrical terms $g^{(k)}$ is bounded above by $k^{k(1+\frac{1}{\alpha})}$ and you have in the series a denominator as $k^{2k}$, so you need $1+\frac{1}{\alpha}<2$. A reference is doi.org/10.1017/S001309150000095X $\endgroup$ – Bazin Apr 5 '20 at 11:26
  • $\begingroup$ @Bazin: Thanks for spotting the typo, and for the reference! Regarding the estimate of $g^{(k)}$, unfortunately I do not have time now to check the bound given in the lecture notes (which is left as a "good exercise"). $\endgroup$ – Mateusz Kwaśnicki Apr 5 '20 at 19:46
  • $\begingroup$ @Kwasnicki the good exercise is to prove that your function $g$ belongs to the Gevrey class $G^{1+\frac{1}{\alpha}}$ (this is sharp). Checking the Faa di Bruno theorem is one way to get this, but complex analysis with Cauchy theorem would give you the same answer. $\endgroup$ – Bazin Apr 9 '20 at 12:23
  • $\begingroup$ @Bazin: Thanks for the comment. I am not sure if I understand correctly, though: you seem to refer to uniform bounds for $g^{(k)}$, while here one needs an estimate that can deteriorate as $t \to 0^+$: a uniform bound for $t^k g^{(k)}(t)$. I believe both estimates follow from Faà di Bruno's formula, but, as said above, I did not attempt to check it. If you did, or if you can improve the answer in any other way, please feel free to edit it. $\endgroup$ – Mateusz Kwaśnicki Apr 9 '20 at 21:15
  • $\begingroup$ @Kwasnicki you do not need anything else than $\vert g^{(k)}\vert\le C^{k+1} k^{k(1+\frac{1}{\alpha })}$ with $\alpha >1$ for the Tychonoff example to work: you have applying the heat operator a telescoping series and the convergence follows from the estimate $\vert g^{(k)}\vert\frac{x^{2k}}{(2k)!}\lesssim R^{2k}k^{k(-1+\frac{1}{\alpha})}$. $\endgroup$ – Bazin Apr 10 '20 at 21:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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