QCG: Radon gsolv fails with --xtbiff

[alan-arnold]

A series of jobs, identical to the following, run to completion for He to Xe but the Rn instance fails with "error while reading input coordinates" after the Wall scaling factor reaches 1.00 from the default starting value.

crest Rn.xyz --T 24 --qcg h2o.xyz --gsolv --nsolv 10 --alpb h2o --mdlen 5 -xtbiff -keepdir > Rn_crest_5_qcg10_1.out &

| quantum cluster growth: GROW |

Water as solvent recognized, adjusting scaling factor for outer wall pot to 0.70

Constraining solute during Growth Solute: unit ellipsoid axis a,b,c : 0.333 0.333 0.333 Solvent: unit ellipsoid axis a,b,c : 0.386 0.322 0.292

solvent anisotropy : 1.133 solute anisotropy : 1.000 roff inner wall : 0.592 solute max dist : 0.000 solvent max dist : 7.284 inner unit axis : 0.333 0.333 0.333 inner ellipsoid/Bohr : 5.803 5.803 5.803 scaling factor outer ellipsoid: 0.700 outer ellipsoid/Bohr : 6.876 6.876 6.876

Size E De Detot Density Eatom av. R Rlast Volume Opt [Eh] [kcal] [kcal] [u/Å^3] [kcal] [bohr] [bohr] [bohr^3] Wall Potential too small, increasing size by 5 % New scaling factor 0.73 Wall Potential too small, increasing size by 5 % New scaling factor 0.77 Wall Potential too small, increasing size by 5 % New scaling factor 0.81 Wall Potential too small, increasing size by 5 % New scaling factor 0.85 Wall Potential too small, increasing size by 5 % New scaling factor 0.89 Wall Potential too small, increasing size by 5 % New scaling factor 0.94 Wall Potential too small, increasing size by 5 % New scaling factor 0.98 Wall Potential too small, increasing size by 5 % New scaling factor 1.00 <.... output file prematurely stops here ...>

The input structure files are trivial, identical (except for the element) to: Rn.xyz

Strangely, if --xtbiff is NOT used (defaulting to the aISS method), all jobs for He to Rn run to completion:

crest Rn.xyz --T 24 --qcg h2o.xyz --gsolv --nsolv 10 --alpb h2o --mdlen 5 -keepdir > Rn_crest_5_qcg10_2.out &

The versions of crest and xtb that I have been using are the codes from Dec 24 2022 compiled with v2021.6.0 ifort and icc compilers from the 2022.2.0 Intel oneAPI toolkits, running on Ubuntu 22.04 (mentioned in https://github.com/crest-lab/crest/issues/163), and the v1.1 (Sep 27 2017) binary for xtbiff

[alan-arnold]

I've done a little more exploration and it appears that this problem lies in xTB-IFF. The last lines of tmp/tmp_grow/iff.out are

taking internal default parameters S y s t e m 1 solute.lmo Nat: 1 Nlmo: 4 S y s t e m 2 solvent.lmo Nat: 3 Nlmo: 4 charge on system 1 : 0.100000D-07 charge on system 2 : 0.990000D-08 ellipsoid axis lengths : 10.158 10.158 10.158 molecular C6AA D4 /au : 735.7 molecular C6AB D4 /au : 138.0 .# of LP/pi/sigma LMOs : 2 0 0 .# of off-center charges : 4 dipole moment (read/calc): 0.872 0.882 ! NO SOLVENT MODEL USED ! simulation temperature /K: 298.15 LUMO energy 1 (read) : -1.415 HOMO energy 1 (read) : -10.497 LUMO energy 2 (read) : 2.234 HOMO energy 2 (read) : -12.147 cpu time for init 0.49 s wall time for init 0.02 s

initial structure (Rx,Ry,Rz (Bohr) alp,bet,gam (rad): -0.0002 -0.0000 0.0002 0.0000 0.0000 0.0000 R CMA (Angst) : 0.000 R min (Angst) : 0.064 dipole moment : 0.882 intermolecular energies in kcal/mol E Pauli : 31943.040 E disp ATM : 0.000 E disp 2B : -1.388 E disp total : -1.388 E ES atom :188261.289 E ES LMO : 0.000 E ES total :188261.289 E induction : 0.000 E CT : 0.000 E cavity : 0.000 Eint total,gas:220202.941 ** <== Gint total

global optimization in 6D

done in fast mode, i.e, without ATM term but final energies include it. If it should be included use -atm option. ..............................................

initial Rcma /A : 0.00 initial Rmin /A : 0.06 R sytem 1 : NaN R sytem 2 : 0.84 .# of parents : 100 .# of generations : 10 .# final grad opts : 15 R grid step size : 2.50 ang step size /deg : 45.00 .# angular grid size: 512 include input : F pocket search : F stack search : T angular search : T

generation of start ensemble . . # probe R points : 4913

[cplett]

Hi, Thanks for pointing that out. It seems that the xtbiff program is missing parameters for Rn. Unfortunately, this is not easily fixable. However, I'm happy to hear that the aISS works for the example. As it performs usually better than the xtbiff program, I would anyhow recommend using this.