Similarity solutions

Also known as invariant solutions. Although I think that the name similarity solutions refers mainly to one particular group: the scaling group.

Given a scalar PDE in J1(R2,R) (wlog)

F(x,t,u,ux,ut)=0

Suppose we have a one-parameter subgroup of symmetries of the PDE (see symmetry group of a DE system)

x¯=X(x,t,u;ε)t¯=T(x,t,u;ε)u¯=U(x,t,u;ε)

That is, the prolongation of it leaves invariant the hypersurface given by F=0. Then it transforms solutions into solutions.
The infinitesimal generator X of the group, which does satisfy X(F)=0 mod F=0, can be transformed into y with a coordinate change

(x,t,u)(y,s,w)

such that X(y)=1,X(s)=0,X(w)=0 . In this coordinates the PDE is

G(s,w,wy,ws)=0

but since G doesn't depend on y we can assume wy=0, so we obtain the ODE

G(s,w,ws)=0

This change of variables is called similarity variables, and the solutions of the original PDE obtained are called similarity solutions. See @Stephani page 172.

Particular case: travelling wave solution.

Keep an eye: A problem arises if the variable y needs to be the dependent variable. See @Stephani page 172. I think that what is happening here is that you have to distinguish the cases where the flow of the symmetry lies inside solutions u=u(x,t) or transform one solution into a different one. I think that this is explained in @blumanlibro page 303, where they call them invariant solutions. Even more, I think that invariant solutions have to do with characteristic lines (see transport equation).