How to reduce numerical diffusion problem when simulating micromixing

Hello Comsol friends,

I have a short question - my simulations of micromixing exhibit rather fast mixin compared to the measured results, especially at high Peclet numbers (~ order of Pe~ 10 000).

I've red in the manual that this can be reduced by making smaller mesh and using higher order of discretization. I've reduced the mesh to about as much as my computor can hendle it.

Has anybody of you come across this situation? What would be your advice how to fix this, if reducing the mesh size is rather difficult?

Thanks in advance,
best regards,
Izidor