Thanks for checking it out. That’s strange, I compiled the August 7 release on gcc 9.1.0 using make mpi on a centos 7 x86_64 machine, with fftw3 single precision, though the issue should be coming before any fft calls.
For the magnitude of forces I’m considering in the system, the relative error of 50 should be very small compared to the forces in question. The forces are on the order of mN per particle.