Statement

For u0u_0 in the on Rn\mathbb R^n,

u(t,x)=RnGν(t,xy)u0(y)dyu(t,x)=\int_{\mathbb R^n}G_\nu(t,x-y)u_0(y)\,dy

is a smooth solution of tu=νΔu\partial_tu=\nu\Delta u for t>0t>0, with u(t)u0u(t)\to u_0 uniformly as t0t\downarrow0.

Verification

For positive time, derivatives of the Gaussian are integrable and differentiate under the integral. Applying its heat equation gives the equation for uu. To recover the initial value, write the difference as the average of u0(xy)u0(x)u_0(x-y)-u_0(x); uniform continuity handles small yy, and Gaussian concentration handles the complement. Spatial derivatives can be moved onto the Schwartz initial datum, so the same argument gives convergence of every fixed spatial derivative. Uniqueness requires an appropriate growth or norm class and is a separate assertion.

References