PPPM long-range electrostatics fails with “ZeroDivisionError” during initialization

[rahatun-cep]

Hi, I’m having trouble using PPPM in HOOMD 5.3.
Short-range Ewald works fine, but when I switch to PPPM I get the following warnings and crash:
RuntimeWarning: invalid value encountered in double_scalars (q2)
ZeroDivisionError: float division by zero in hoomd/md/long_range/pppm.py inside _diffpr()

The error happens during the PPPM setup phase, before the simulation begins.

What I think is happening:
It seems that one of the PPPM parameters (kappa, rcut, or the mesh dimensions) is being computed as zero or NaN, so the reciprocal-space grid cannot be constructed.

My questions:
Is this a known issue in HOOMD 5.3?
Are there recommended parameter ranges for PPPM in small boxes or with my charge distribution?
How can I diagnose which PPPM parameter is becoming zero?

[joaander]

Use a debugger and set a breakpoint in _diffpr. Use the debugger to inspect the value of each variable used in the expression that results tin the divide by zero:

hoomd-blue/hoomd/md/long_range/pppm.py

Lines 289 to 302 in f5d00ec

	def _diffpr(hx, hy, hz, xprd, yprd, zprd, N, order, kappa, q2, rcut):
	"""Part of the algorithm that computes the estimated error of the method."""
	lprx = _rms(hx, xprd, N, order, kappa, q2)
	lpry = _rms(hy, yprd, N, order, kappa, q2)
	lprz = _rms(hz, zprd, N, order, kappa, q2)
	kspace_prec = math.sqrt(lprx * lprx + lpry * lpry + lprz * lprz) / math.sqrt(3.0)
	real_prec = (
	2.0
	* q2
	* math.exp(-kappa * kappa * rcut * rcut)
	/ math.sqrt(N * rcut * xprd * yprd * zprd)
	)
	value = kspace_prec - real_prec
	return value

[rahatun-cep]

Thank you very much for your reply. I was able to resolve the previous issue by running the initial conformation-generation script separately (It was my mistake). Now I’m trying to add PPPM to the constant-pH MD/MC code, and I’m running into a new problem. I’m getting the same degree of association whether PPPM is enabled or commented out. It seems that PPPM is not being properly included in the energy calculation. When I comment out PPPM, the calculation of the degree of association is much faster, but when PPPM is included, the computation takes longer—yet the results are identical in the end. I’m not sure what mistake I might be making. Any guidance would be greatly appreciated.