PCMSolver Error w/ B97-3c

[corinwagen]

I can't get CPCM(water) to work with B97-3c for my molecule. Here's my input file:

molecule {
  -1 1
  O       -3.79496408 -2.36260414 1.70152926
  C       -2.93101645 -1.66904914 1.14430726
  C       -2.13005829 -0.68789893 1.86141181
  C       -1.14196217 0.05066703 1.29086101
  C       -0.47856119 1.13396394 2.09829283
  C       -0.80787855 -0.19207110 -0.09443095
  N       0.25723669 0.27990621 -0.79382718
  O       1.28867817 0.69574040 0.13535844
  S       2.75258017 0.18914236 -0.18685693
  O       2.81552768 -0.50840706 -1.42180252
  O       3.61564541 1.27844179 0.09948727
  O       2.95557380 -0.82533526 1.00298083
  C       2.03019476 -1.93245912 1.04782605
  C       -1.66037059 -1.05786431 -0.87382233
  C       -1.39498496 -1.19960129 -2.34816003
  C       -2.65299201 -1.76249051 -0.27337471
  H       -2.37740040 -0.54585648 2.90523863
  H       2.38954139 -2.56465364 1.84992540
  H       1.02863634 -1.57685304 1.27002537
  H       2.04236674 -2.46241236 0.09951447
  H       -2.15541649 -1.82673633 -2.80763054
  H       -0.41234738 -1.64107180 -2.50716877
  H       -1.37758315 -0.22482435 -2.83713913
  H       -3.27599978 -2.43724084 -0.84607768
  S       -0.34233722 2.35735393 -1.83532763
  C       1.36095190 2.94818139 -1.81407535
  H       1.71979201 3.09855604 -0.79815280
  H       2.02802634 2.25554347 -2.32132745
  H       1.39841688 3.90181231 -2.33199310
  H       0.52648693 0.85272557 2.39744759
  H       -0.38901252 2.05630112 1.52833438
  H       -1.06736839 1.33094835 2.99083304
}

set {
  pcm true
  pcm_scf_type total
}

pcm = {
   Units = Angstrom
   Medium {
   SolverType = IEFPCM
   Solvent = Water
   }

   Cavity {
   RadiiSet = UFF
   Type = GePol
   Scaling = False
   Area = 0.3
   Mode = Implicit
   }
}

gradient("b97-3c")

And here's what happens when I run the calculation:

$ psi4 input.dat
PCMSolver fatal error.
 In function computeS at line 51 of file /Users/runner/miniforge3/conda-bld/pcmsolver-split_1696645993753/work/src/bi_operators/IBoundaryIntegralOperator.cpp
S matrix is not positive-definite!
Consider changing the average area of the cavity finite elements.
Please report this issue: https://github.com/PCMSolver/pcmsolver/issues

If I delete the PCM sections, the B97-3c gradient runs just fine. The molecule looks normal:

All calculations run on a MacBook Pro using Psi4 1.9.1 from conda-forge.

(I know the message says to report this to PCMSolver but that repository hasn't had any updates in 3+ years...)

[loriab]

I know, it's this issue: https://github.com/PCMSolver/pcmsolver/issues/206 . An additional check was added to pcmsolver that hits larger molecules, and when I updated from v1.2.1 to v1.2.3 when doing the conda-forge transition, that check got turned on. I should do a new binary version commenting out that check so that calcs at least run as before, even if they're slightly less to be trusted.

[corinwagen]

Ah gotcha - makes sense, thanks for quick response

[corinwagen]

FWIW - I'm still running into this issue on a bunch of systems, using a variety of methods (not just B97-3c).

[corinwagen]

If anyone else is having issues here, we've just configured the ddx interface to Psi4 and now solvent is working again on our systems—unfortunately there are no analytical gradients w/ the current interface, but I think this will change in the future (see discussion in #2767). The interface is documented nicely here.