Implicit solvation geometry optimization fails for higly charged POM

[codecollider]

Hi,

I´m trying to geometry optimize a singly protonated Keggin-Dawson POM structure (P2W18O62H) in the charge state -5 (coords are shown below). The geometry optimization is running fine when using xtb coord --gfn 1 --chrg -5 --opt, but it is not working when using the implicit solvation model, e.g running.

xtb coord --gfn 1 --alpb water --opt --chrg -5

I tried alpb and gbsa with water or methanol and both geometry optimizations do not converge. The first optimization steps are working, but at some point the energy goes crazy. I also tried toluene and in this case it is working fine, maybe because it is less polar (?).

Setting etemp higher also didn´t solve the problem. I´m using GFN 1, because GFN 2 does not work for the POMs (the cage collapses). I there anything I can try beside of what I did?

Thanks for the help and have nice weekend,
Marco

83
energy: -345.346588919820 gnorm: 0.000586692727 xtb: 6.5.1 (579679a)
W -0.05414087995027 6.13859895508601 32.24266257970957
W -0.12265682740682 5.86153053600755 35.51603289244861
W -0.03592852356779 3.17304060652535 33.63937925375328
W 2.93568623968474 2.32137395340652 31.81117629555352
W 2.92428027365558 5.34661761132816 30.37966470381880
W 2.81812138933875 8.30486303850834 32.45016279115233
W 2.74866427299812 8.01716214281666 35.79932114451916
W 2.78229816988649 4.74434051967092 37.33199139285805
W 2.88113590354320 2.00528786214443 35.40943228174974
W 6.52360244561505 2.08864825626183 35.53007516203842
W 6.54501379000732 2.61861508116948 31.98171161350168
W 6.57560099483684 5.46413087450050 30.40761990605730
W 6.46261862685123 8.39225699357474 32.51899865520378
W 6.39954902474657 8.12909016520776 35.92796142201621
W 6.42935551538831 4.83232954854627 37.44129172270882
W 9.36457532027707 6.06413486686656 35.74167344521724
W 9.46999497178439 3.37535041285029 33.86220346612902
W 9.42658065773987 6.33541005841018 32.44312035830787
P 2.85565338124891 5.10412221868781 33.85723633857441
P 6.50084663430476 5.14395470717987 33.93452330099550
O -1.22296613994080 6.83545255189365 31.18571617128805
O 1.20487355141729 5.47462603555795 31.06191605631667
O -0.72678892297167 4.41408056127957 32.44295186335027
O -0.79396888919739 6.50760382619815 33.90685139373433
O 1.13140822814111 7.52803795234835 32.49858288258925
O 1.27076272466977 5.06653762374411 33.81468881763205
O -1.33711430886886 6.37251509094399 36.62659311827197
O 1.06696103251040 7.27030257617617 35.54487818158567
O -0.77822547528528 4.19833424865944 34.99405073284991
O 1.08886309059454 5.00164231346972 36.60891717577421
O -1.18599369822988 1.89654418957246 33.50986563361904
O 1.21818736336920 2.69878952218707 32.35409077221575
O 1.17951711697172 2.49574869711949 34.85168973626003
O 2.67215563101681 0.79031268591524 31.06138206508965
O 4.79884260956113 2.37007796699957 31.58691816476467
O 2.76972940432058 3.48832061384847 30.37777982209272
O 3.23459074334229 1.88749768392493 33.58367067007615
O 3.35620884683202 4.48885763926620 32.54442402073339
O 2.58816425442269 5.73690250846863 28.73506485137034
O 3.17367523200722 6.98992726989430 31.18516717292866
O 4.77211017081738 5.19594866183988 30.24191129076184
O 2.43200136406260 9.69458521076735 31.50688422007194
O 4.64061642442527 8.57529148509737 32.59098130870948
O 2.58509938675820 8.92676971510656 34.18618847192820
O 3.27002071435451 6.56843667455110 33.99150801003643
O 2.32101166029873 9.22681224159158 36.94989215126662
O 4.58094279326772 8.30457330688189 35.78357345691959
O 3.05597955165839 6.50580335157237 36.82816709087960
O 2.37086198393560 4.85159083270440 39.00203047347109
O 2.62931386911554 2.91818190753809 37.00685284727793
O 4.61647911787486 4.56807018488398 37.51793824557321
O 3.29508862418391 4.27624889514562 35.06701675839123
O 2.53847954318640 0.37722115812998 35.86007419827074
O 4.70594214796688 1.97886682730503 35.71942274867686
O 6.92776762311419 0.47257984344113 35.96885262765304
O 8.23081480316275 2.64602438133666 35.01157678698205
O 6.68611091452604 3.00857372104935 37.12080158072131
O 6.25337935375517 2.01649121117523 33.67144257056394
O 6.03719419833980 4.33063249625577 35.13549541730585
O 6.93216034384239 0.95957674509294 31.21007243112032
O 6.81425037551850 3.59929937106011 30.45196531618834
O 8.26901293090502 2.86562347477390 32.50781802243880
O 6.10961513646858 4.43476131851790 32.60665521176269
O 6.98342668564613 5.86236866339056 28.78182500757080
O 8.24513234951111 5.62049559223036 31.20911099299499
O 6.24025855371635 7.08750061388888 31.20224494632710
O 6.84496662656551 9.81544059007181 31.62714448998566
O 8.17968561598137 7.67158302044525 32.64123153635623
O 6.60375735754105 8.97111904246550 34.28477325207822
O 5.99089597262473 6.56708451572115 33.94040080481954
O 6.73807793147476 9.36587889321322 37.07965173968178
O 8.11533690067663 7.41789891978578 35.71772568385139
O 6.12284849919928 6.58693360222220 36.92289721065018
O 6.77047110489783 4.95692360940694 39.12517719672964
O 8.14512883613961 5.14763746339372 36.77247441501781
O 10.51428370993881 6.62532178969173 36.89539418183929
O 10.11188891505777 4.42587051044103 35.24037577436268
O 10.06004174478861 6.72321304665929 34.14759229657180
O 8.07634496483057 5.17413450085425 33.95148239935391
O 10.66392308967842 2.13838916353765 33.76868492693821
O 10.15499734450090 4.63388932926518 32.68233848479431
O 10.62006497031260 7.07396178857442 31.44454712906283
H 6.25848201971418 0.37542236270615 30.86836826518647

[haneug]

I would say it is impressive that GFN1-xTB converges at all for this system, as metal oxides are a typical weakness for GFNn-xTB methods as they were not parameterized for metal oxides. Maybe H0 tuning for W and O could help with convergence? You can generate a DFT gradient as a reference to optimize the pair parameter. See this publication for a nice example of where H0 tuning can help. Alternatively, you can use GFN0-xTB (no SCC so no convergence problems).

[codecollider]

Thanks, I´m looking into this!

[MtoLStoN]

You could try our dd-COSMO implementation with xtb version 6.6.0. For example, the provided POM structure seems to converge to a reasonable geometry using xtb pom.xyz --chrg -5 --gfn 1 --cosmo water --opt.