The effects of errors in assumed conductivities and geometry on numerical solutions to the inverse problem of electrocardiography