[CP2K-user] [CP2K:20206] Re: Reality check: time needed for FES convergence with WT Metadynamics

[David M. Sherman]

Well, I answered my own question: for a system like this, it took about 220 
ps for each walker (2.2 ns total) to get log K values converged to +/-0.5 
units. The results are pretty interesting and show where both DFT and 
dielectric continuum models fail (and succeed).  Paper in geochemical 
literature is forthcoming....

On Friday, May 3, 2024 at 1:02:43 PM UTC+1 David M. Sherman wrote:

> So, I'm looking at metal-chloride complexation in NPT_I aqueous solutions 
> (55 H2O in box). I'm using the PBE functional with Grimme 
> dispersion. Because I'm doing NPT, I use a plane-wave cutoff to 400/40; the 
> barostat performs very nicely with average pressures fluctuating within a 
> few % of my target (simulations run from 1 to 3000 bars with temperatures 
> ranging from 50 to 350C).  My colvars are M-Cl CN and M-O CN with SCALE 
> small enough to nicely resolve the FES mimina.  I'm using well-tempered 
> metadynamics with a bias factor (T+DeltaT/T) of 10.  I'm adding hills every 
> 150 steps (0.5 fs stepsize).  I'm using 10 walkers.  From the differences 
> in FES minima, I calculate logK values for the various M+nCl = MCln 
> reactions. Qualitatively, the results are pretty good: even after only 50ps 
> per walker, most of the the logK values are within 1-2 units of experiment 
> at 50C, 1 bar and show a nice systematic (and physically correct) change 
> with pressure and temperature.  However, even after each walker has run for 
>  150 ps (i.e., 1.5 ns for each P,T run), I still have the logK values 
> drifting over  +/-  1-2 log units every 20ps or so; this corresponds to 
> fluctuations in the differences between  FES minima of 10-20 kJ/mole. 
>  After 150ps per walker, the added hill height is still about 0.00002 AU 
> (0.05 k/mole).  Given that I'm using WT Metadynamics,  I know that, 
> eventually,   the added hills will be 0, and the FES will become constant, 
> but why is this taking so long?    I know I'm slowing things down by also 
> including the oxygen (water) CN as a collective variable.   Indeed, it 
> seems like the significant hills  that are being added after 150 ps are 
> just for non-important hydration complexes; still, the FES mimima for the 
> MCl, MCl2, MCl3 complexes  haven't really converged either (the free 
> surfaces are clearly on their way to convergence; however the energy 
> differences between the free energy minima are still too noisy).  I know 
> that if I have poorly defined colvars, I will never get FES convergence, 
> but these colvars are well-defined relative to the system and, indeed, give 
> very clear minima in the FES with a barrier height between the minima of 
> around 10-30 kJ/mole. So, my question is: should we expect a highly 
> converged (+/- 5kJ/mole="chemical accuracy") FES in a system like this to 
> require more than 1-2 nanoseconds (100-200 ps/walker with 10 walkers)?  Is 
> everything here normal? Alternatively, Is my bias factor too high?  Should 
> I increase NT-HILLS (am I "hill surfing"?). Thanks!

-- 
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 on the web visit https://groups.google.com/d/msgid/cp2k/603250a2-4a7d-4b30-a039-e1c66609228en%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20240512/e7986850/attachment.htm>