Canonical: https://knowlpedia.com/partial-differential-equations/heat-smoothing-estimate/

# Heat smoothing estimate

Gaussian convolution gains spatial derivatives and improves integrability with explicit powers of time.

For \(1\le p\le q\le\infty\), a spatial multi-index \(\alpha\), and \(t>0\), the [heat semigroup](https://knowlpedia.com/partial-differential-equations/heat-semigroup/) satisfies
\[
\|\partial_x^\alpha T_tf\|_q
\le C(\nu t)^{-|\alpha|/2-\frac n2(1/p-1/q)}\|f\|_p.
\]
The constant depends on \(n,p,q,\alpha\), and is independent of \(\nu,t,f\).

## Kernel estimate

Every Gaussian derivative is a polynomial times a Gaussian. Scaling gives
\[
\|\partial^\alpha G_\nu(t)\|_r
=C_{\alpha,n,r}(\nu t)^{-|\alpha|/2-\frac n2(1-1/r)}.
\]
Choose \(r\) with \(1+1/q=1/p+1/r\) and apply Young's convolution inequality. The negative powers as \(t\downarrow0\) express the cost of recovering derivatives from rough initial data.
