To keep things simple I just initialize the physics for all elements in G4. valgrind has been reporting an invalid read during G4SeltzerBergerModel::ReadData for the last few G4 versions but we never had any issues related to this. I finally tried to track this down to an apparent array index overflow. When reading the parameters for Z=94 in line 179
   ylimit[Z] = v->Value(0.97, emaxlog, idx, idy);
idx = 22
going down the stack into G4Physics2DVector.icc line 114:
 else if(z < v[idx] || z >= v[idx+1]) { 
in my understanding the size of v is 14 (14 is passed down as idxmax), so v[idx] and v[idx+1] is out of bounds