(Malgrange – Ehrenpreiss) Every linear partial differential operator with constant coefficients P(D) has a fundamental solution, i.e., there exists a solution of the equation
P(D) u=\delta
Let us define
K(x)=\int_{ \mathbf{R} ^{n-1}} \int_{\mathfrak{J}\left(\xi_n\right)=\Phi\left(\xi^{\prime}\right)} \exp (2 \pi \imath x \cdot \xi)(P(\xi))^{-1} d \xi_n d \xi^{\prime}
where, as before, \xi=\left(\xi_1, \ldots, \xi_{n-1}, \xi_n\right)=\left(\xi^{\prime}, \xi_n\right) \text {, while } \Phi was defined in Example 12.22.
We shall show that K is well defined, and, moreover, is the desired fundamental solution of P. To that end, put first
P_N(\xi)=P(\xi)\left(1+4 \pi^2 \overset{n}{\underset{j=1}{\sum}} \xi_j^2\right)
where N is a (sufficiently big) natural number. Next, put
K_N(x)=\int_{ \mathbf{R} ^{n-1}} \int_{\mathfrak{J}\left(\xi_n\right)=\Phi\left(\xi^{\prime}\right)} \exp (2 \pi \imath x \cdot \xi)\left(P_N(\xi)\right)^{-1} d \xi_n d \xi^{\prime}
where the function Φ was chosen as in Example 12.22. Since then on the domain of integration it holds \left|P_N(\xi)\right| \geq C \cdot\left(1+|\xi|^2\right)^N, the last integral converges for N > n/2.
Next, let us prove
P_N(D) K_N=\delta
Assume \varphi \in \mathcal {D}( \mathbf{R}^n). Since the adjoint operator to P_N(D) \text { is } P_N(-D), it holds
Note that we used the equality
\varphi(0)=\int_{ \mathbf{R}^n} \mathcal {F} \varphi(\xi) d \xi
which follows using the inverse Fourier transformation.
So we obtain
\delta=P_N(D) K_N=P_N(D)(1-\Delta)^N K_N
Now K=(1-\Delta)^N K_N \text {, hence } is the fundamental solution of the operator P(D).