[CP2K-user] [CP2K:22018] Sensitivity of SCCS convergence to RHO_MIN / RHO_MAX and Poisson solver choice

[Mridhula]

Hello CP2K community,

I am having difficulty identifying stable and physically reasonable values 
for RHO_MIN and RHO_MAX in the SCCS implicit solvation model, and I would 
greatly appreciate any guidance.

My system is a highly negatively charged molecular system (total charge -6) 
treated under 0D (non-periodic) conditions using the MT Poisson solver, 
with BLYP as an initial functional. For context, the system is a 3GC 
nucleotide with total charge -6; the neutralized version includes six Na⁺ 
counterions. Due to the large net charge, the SCCS response appears to be 
extremely sensitive to the cavity definition.

*What I tried initially*

After reading a Google Groups discussion 
<https://groups.google.com/g/cp2k/c/uCBgj5pbIOo> reporting SCCS parameters 
for OH⁻, where a relatively narrow cavity with higher density thresholds 
was used, I explored RHO_MIN / RHO_MAX values scaled around the OH⁻ 
reference point (RHO_MIN = 0.0024, RHO_MAX = 0.0155), approximately 
preserving their ratio. In practice, this corresponded to scanning values 
in the ranges RHO_MIN ≈ 0.0018-0.0042 and RHO_MAX ≈ 0.0116-0.0271 (a.u.).

However, these choices consistently led to severe convergence issues:

   - the SCCS cycle did not converge even after 2000-3000 iterations, 
   - large oscillations and occasional energy blow-ups were observed, 
   - SCF iterations became extremely slow (up to ~200 s per iteration), 
   even with BLYP. 

To improve stability, I also tried adjusting SCCS-related settings, 
including:

   - decreasing the SCCS MIXING parameter to 0.02, 
   - setting EPS_SCCS = 1.0E-06 so that SCCS iterations begin only after 
   SCF convergence. 

I then hypothesized that, due to the large -6 charge, placing the solvation 
cavity too close to the solute leads to an overly reactive solvent response 
that destabilizes the SCF/SCCS coupling. Based on this, I explored cavity 
definitions at lower electronic densities, shifting the dielectric 
transition further into the low-density tail of the electronic density. 
Specifically, I tested RHO_MIN / RHO_MAX values in the ranges 3.0 × 
10⁻⁴-1.0 × 10⁻³ and 1.0 × 10⁻³-3.0 × 10⁻³ (a.u.), respectively.

While these values improved stability in some cases, I still encountered 
SCCS convergence issues. To check whether this behavior was specific to the 
highly charged system, I also neutralized the system by adding counterions 
(6 Na⁺), but the neutral system exhibited similar convergence problems.

*Additional observation*

I also noticed a strong sensitivity to the Poisson solver. For identical 
SCCS parameters, the number of SCCS iterations and overall stability 
differed significantly when switching between the MT and analytic Poisson 
solvers, with the analytic solver sometimes behaving more robustly for very 
low-density cavity definitions. Orbital transformation also seemed much 
more robust than using standard diagonalisation. 

    

The first image shows the SCCS results obtained with the analytic Poisson 
solver, whereas the second image shows the corresponding results using the 
MT Poisson solver; in both cases, orbital transformation was employed.

The above image corresponds to a simulation using standard diagonalization 
together with the MT Poisson solver. As shown in the screenshots, there is 
significant sensitivity to these solver and SCF choices even when all other 
settings are kept the same. I have also attached the input file 
corresponding to this setup using the analytic Poisson solver.

*Questions*

I would appreciate any advice on the following points:

   1. For highly charged molecular systems, is it generally advisable to 
   shift the SCCS dielectric transition further into the low-density tail, 
   even if this leads to a relatively loose cavity? 
   2. Are there recommended ballpark ranges for RHO_MIN / RHO_MAX for large 
   anionic systems beyond small test cases like OH⁻? 
   3. Are there additional diagnostics or best practices to assess whether 
   a given cavity definition is physically reasonable and numerically stable?

Thank you very much for your time and help.

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/f97e9136-1cf3-4328-99c4-2b384d58ba91n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20260104/a293c062/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: cp2k_implicit.inp
Type: chemical/x-gamess-input
Size: 3195 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20260104/a293c062/attachment-0001.inp>