Show that $E(t,x) = \frac{\textbf{1}_{(0,+\infty)}(t)}{t^{1/3}}\textbf{Ai}(\frac{x}{t^{1/3}})$ is the unique tempered fundamental solution with support in $\{ (t,x) \mid t\geq 0 \}$ of the partial differential operator $P(D) = \partial_t + \frac{1}{3}\partial_x^3$.

I am puzzled about an exercise problem from an undergraduate course about distributions and Fourier transfrom.

The exercise concerns about the constant-coefficients 1+1 dimensional differential operator $$P(D) = \partial_t + \frac{1}{3}\partial_x^3.$$ The first part of the exercise is to check that $$E(t,x) = \frac{\textbf{1}_{(0,+\infty)}(t)}{t^{1/3}}\textbf{Ai}(\frac{x}{t^{1/3}})$$ is a fundamental solution of $P(D)$, where $\textbf{Ai}(x) = \mathcal{F}^{-1}_{\xi \rightarrow x}(e^{i\xi^3/3})$ is the Airy function (it turns out that $\textbf{Ai}$ is $C^\infty$). This part is straightforward and I could work out.

But the second part of the exercise asks to show that $u(t,x) = E(t,x)$ is the unique solution in the sense that $$ u \in \mathcal{S}'(\mathbb{R}^2_{t,x}) \\ (\partial_t + \frac{1}{3}\partial_x^3)u = \delta_{(0,0)} \\ \textbf{supp}~u \subset \mathbb{R}_{t \geq 0} \times \mathbb{R}_x$$

I could not work out. I have consulted Hormander's book $$\textit{The Analysis of Linear Partial Differential Operators II}$$ and found some topics very close to it, mostly in Chap. XII "The Cauchy and Mixed Problems", but it seems that there is no theorem I could directly quote.

$\textbf{My Question}$

  1. Is there any general theorem about the uniqueness of tempered fundamental solutions of linear differential operators with constant coefficients? Is there any necessary or sufficient conditions such that $P(D)u = \delta$ has a unique tempered solution $u$, maybe with some requirement of the support $\textbf{supp}~u$ (for instance $\textbf{supp}~u \subset \text{half space}$ like above.)

  2. Does the requirement $\textbf{supp}~u \subset \mathbb{R}_{t \geq 0} \times \mathbb{R}_x$ take the most important role so that actually this exercise is very easy?


1 Answer 1


It suffices to show that if a tempered distribution $u$ supported on $\{t\ge0\}$ satisfies the homogeneous equation $(\partial_t+\frac13\partial_x^3)u=0$, then $u=0$. This can be shown using a duality argument.

More precisely, pick $\phi\in\mathcal S_{t,x}$ and it suffices to show $u(\phi)=0$. By Duhamel's formula,

$$ \hat\Phi(t,\xi)=\int_t^\infty -e^{i(t-s)\xi^3}\hat\phi(s,\xi)ds $$

satisfies $(\partial_t+\frac13\partial_x^3)\Phi=\phi$, $\Phi(t,\cdot)\to0$ in $\mathcal S_x$ as $t\to+\infty$, and $\Phi\in C_t^\infty(\mathcal S_x)$ (this is tedious (but routine) to check).

Now let $\rho\in C^\infty$ be such that $\rho=1$ on $(-1,+\infty)$ and $\rho=0$ on $(-\infty,-2)$. Then $\rho(t)\Phi\in S_{t,x}$ and

$$ u((\partial_t+\frac13\partial_x^3)(\rho(t)\Phi))=-((\partial_t+\frac13\partial_x^3)u)(\rho(t)\Phi)=0. $$

Since $\rho=1$ on $(-1,+\infty)$, we have

$$(\partial_t+\frac13\partial_x^3)(\rho(t)\Phi)=(\partial_t+\frac13\partial_x^3)\Phi=\phi\text{ on }\{t\ge0\}.$$

Since $u$ is supported on $\{t\ge0\}$,

$$ 0=u((\partial_t+\frac13\partial_x^3)(\rho(t)\Phi))=u(\phi). $$

PS: We leave it to the reader (and the OP) to think about its generalizations.

PPS: Similar duality arguments are used in the proof of Holmgren's uniqueness theorem, see Section 3.5 of Fritz John, Partial Differential Equations.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .