We are using a previous version of libmesh and would like to reproduce the inverse_map bug found last November. 

I am using Hex8 elements and making sure that has_affine_map is 'true'. I compute Jinv and use it to evaluate dT/dx (i.e., dT/dx = Jinv dT/dxi). These values are in agreement with computations using dphidx from the library. 

Is there anything that I need to consider to reproduce the error ?