Similarity solution (source code)

= Similarity solution

= Self-similar solution
{synonym}

A <similarity solution> has a fixed profile after rescaling its independent variables and, when needed, its amplitude. A typical <similarity ansatz> is
$$
u(x,t)=t^{-\alpha}F(x/t^\beta),\qquad t>0.
$$
Substitution into a <partial differential equation> fixes a scale balance and produces an equation for the <similarity profile> $F$. In one spatial dimension this is often an <ordinary differential equation>; in several spatial dimensions it can remain a <partial differential equation> unless a further symmetry, such as radiality, is imposed.

For $u_t=\Delta u$ on $\mathbb R^n$, balance of time and space <derivatives> gives $\beta=1/2$. Preservation of a finite nonzero spatial integral gives $\alpha=n/2$. Writing $\xi=x/\sqrt t$, the profile equation is
$$
\Delta_\xi F+\tfrac12\xi\cdot\nabla_\xi F+\tfrac n2F=0,
$$
solved by $F(\xi)=C e^{-|\xi|^2/4}$. The <Gaussian heat kernel> is therefore a spreading profile whose total mass stays fixed. A scale-invariant equation need not have a similarity solution compatible with arbitrary prescribed data; the scaling of the data and conserved quantities is part of the construction.