When do cross-diffusion systems have an entropy structure?

Necessary and sufficient conditions for the existence of an entropy structure for certain classes of cross-diffusion systems with diffusion matrix $A(u)$ are derived, based on results from matrix factorization. The entropy structure is important in the analysis for such equations since $A(u)$ is typically neither symmetric nor positive definite. In particular, the normal ellipticity of $A(u)$ for all $u$ and the symmetry of the Onsager matrix implies its positive definiteness and hence an entropy structure. If $A$ is constant or nearly constant in a certain sense, the existence of an entropy structure is equivalent to the normal ellipticity of $A$. Several applications and examples are presented, including the $n$-species population model of Shigesada, Kawasaki, and Teramoto, a volume-filling model, and a fluid mixture model with partial pressure gradients. Furthermore, the normal elipticity of these models is investigated and some extensions are discussed.