xtb Single Point Hessian for TS structures unsuitable?

[acd81]

Dear All,

I am trying to employ the SPH methodology aimed at obtaining realistic force constants for a TS structure that has been optimised on a decent DFT level before. Specifically, the TS structure describes H-atom transfer between two adjacent centres, which has a sharp deep change in curvature of the PES - thus v(imag) ~ 1000i cm-1. The numerical hessian at the DFT-optimised structure has indeed one v(imag) ~1000 cm-1 when applying the gnf2 level (via keyword hess).

However, when SPH(gfn2) is applied by using xtb-6.3.3 with the rmsd specification $metadyn rmsd=0.08 $end as recommended in the parent paper (J. Chem. Theory Comput. 2021, 17, 1701−1714) the initial DFT-preoptimised structure is not maintained during metadynamics and the crucial element-H distances for the H-transfer process for SPH-reoptimised and DFT-optimised initial TS structures are found distinctly different. This results in the absense on any imaginary frequency in SPH(gfn2) estimated Hessian. Other smaller rmsd values (0.001-0.01) have also been probed, but to no avail.

I was impressed by the correct reproduction of the TS for H2 addition onto an iridium pincer complex (Figure 10 and Table 2 in the parent paper), but the methodology seems to fail for a seemingly simple TS structure (with a huge imaginary frequency) for H-transfer considered here.

Does anybody have encountered similar problems and perhaps would like to share her/his experince and provide some remedy. Any assistance would be highly appreciated. Thanks in advance, Pitti

the command used is as follows: xtb --gfn2 coord --bhess -I XCONTROL

[acd81]

should have included the DFT-optimised TS structure (Tmole style) - apologies

coord.txt

[acd81]

$coord 3.08697368416802 0.16133515262123 -0.22287118588309 ca -0.21826080938438 -0.43145081025720 3.26007723388471 n -0.25462670928851 0.00843572204312 -3.17270085154633 n -1.27498795788570 1.99253206936190 2.71594812332669 c -3.06934600354442 2.14711065993232 0.81005399602071 c -3.91754172569753 -0.20400679786240 -0.46947618832027 c -2.46687061181998 -1.16597935890000 -2.49476711539567 c -1.36461699309918 -1.70523237882876 2.37915007507800 h 5.82157238583745 3.87446831893388 -0.45632451386381 n 4.96938829938768 6.54076820500094 -0.33941182184646 c 6.92167552407917 7.95562543816785 1.28728181368578 c 7.49614265141746 5.91581985184953 3.27295190370216 c 7.49105349407849 3.36458274489963 1.85948658406644 c 6.56462802395840 1.05423525680233 3.29938249584519 c 6.35905773039874 -2.87980189825293 -0.17691222913666 n 6.61135781098306 -1.26296894871940 1.57944928258688 h 0.03778922270903 -1.16533891335040 5.98104291716238 c 0.45380421489327 -0.15871367836348 -5.85617100541461 c 7.85518605640347 -2.67699238874770 -1.36560485687940 h 6.27104634054616 -5.46308042848838 0.78360313175993 c 7.99294108636729 0.46624209819497 4.68497244180149 h 4.88415377318715 1.58490682010339 4.43101436980176 h 1.31476610215553 0.22485733905613 6.82758127999558 h 0.72111946211944 -2.14009297979098 -6.48020903905871 h -11.01235799928929 -0.80909860082286 4.57474985827969 h -10.67753129320272 -3.85982578378937 1.87150605416598 c -12.40888733595019 -4.78463494326882 2.47065621003485 h -9.24498546970172 -4.86557853107103 -0.05683726657142 c -9.83494525994553 -6.59434765983553 -1.00064933781983 h -7.00450477977925 -3.67393100407755 -0.86879509851790 c -5.51651997203383 -4.63794565710366 -2.87612700276491 c -6.09889031169466 -6.38845435150475 -3.78600390676973 h -3.37858744253560 -3.43651875437164 -3.67272288097804 c -2.27957606596264 -4.28730337573695 -5.17741999368390 h -0.36160411252232 4.19912931733676 3.92064923488273 c -6.59206195585802 3.10622953343025 -2.73439843196822 h -7.09112680157981 1.28982784781309 3.29766158667595 h -2.44717797233414 -0.97957494966415 7.42677388795154 c -2.75090095987486 1.17868255787058 8.93726103941509 c -1.21156572778667 2.53309059917550 9.04201095637845 h -4.93635852467207 1.63775962602065 10.30288538070522 c -5.10348900740834 3.33474108322920 11.44432973610344 h -6.89435064416254 -0.11207286624019 10.19519959202754 c -8.63049655454742 0.19511447034522 11.24656426457020 h -6.60692667261245 -2.26577558678374 8.73877574545691 c -8.13679754215630 -3.63695517905403 8.65493780107483 h -4.41801319306142 -2.75724494987228 7.32924125359225 c -4.47686926130810 -5.24578019993529 5.90484692759246 c -4.64760108578045 -6.74143803654963 7.33746955307947 h -6.24046523001321 -5.28942601009829 4.81088946582397 h -2.31537043557777 -5.98758646284583 4.13451865951829 c -2.70112534631390 -7.91209679131427 3.47262546041877 h -2.37053751338378 -4.84263610911073 2.40645587462300 h 0.23700737894555 -5.95266402229586 5.39062855572635 c 1.50494807364829 -8.22377638964188 5.82456555464069 c 0.64230538431985 -9.97643127071385 5.18509362733948 h 3.82839259561825 -8.28988686811901 7.05219065964254 c 4.77635314802675 -10.08487911521308 7.35674196376448 h 4.93295033993752 -6.05036901876281 7.85763188555148 c 6.75753358330133 -6.06656596116721 8.79618325389790 h 3.68797399183327 -3.77372510856324 7.44506558093499 c 4.55354253103883 -2.01714690063954 8.06531854235670 h 1.34520632380709 -3.69463281145764 6.23896703705977 c -4.06296470848118 4.54531507449234 0.12354705295084 c -9.88748487352446 -1.60930129699457 3.05441378907048 c -1.59795937395219 0.90498226485498 -7.57079411789828 c -2.82205531933842 -0.69181609210071 -9.27184037633885 c -2.18376003354423 -2.63736085272223 -9.45448154562721 h -4.85070803330910 0.14751915753040 -10.72023625080979 c -5.78470374747809 -1.13540038571182 -12.02214564632516 h -5.68405021093886 2.62330386853290 -10.46616456288518 c -7.28072366668926 3.29964052299582 -11.56526091441776 h -4.44830757610254 4.24300157663592 -8.80158841312356 c -5.06974373340872 6.19455657055638 -8.62076368813365 h -2.40665756097019 3.42058114253902 -7.35460075063261 c -0.95858897496368 5.24867136405860 -5.73926216668320 c -0.78477507156337 4.52160849383612 -3.80962679622621 h -1.97548915933091 7.04955040175708 -5.64614292216417 h 1.67183584344474 5.71126084068828 -6.84216067726439 c 1.47760494623785 6.14742944232220 -8.86658289580944 h 2.47682497753965 7.41416991381188 -5.97358296231462 h 3.60087278899806 3.61387296449849 -6.57431449696934 c 6.13771752922418 4.34955491182116 -6.81735418644243 c 6.54712869704878 6.33894242598689 -7.14453914147824 h 8.12724657116311 2.64584803732112 -6.66188702936134 c 10.06298541035028 3.29955724313831 -6.86697064788854 h 7.60421700592952 0.10237125415056 -6.25752463430634 c 9.11948521281739 -1.27928684927859 -6.14660394989553 h 5.10132496990282 -0.66098592657149 -6.04720134698445 c 4.67957309739345 -2.65276572307359 -5.78350745424914 h 3.06763055059820 1.03163851143997 -6.19241569781367 c -3.17423060169269 6.77097168761179 1.37478378146871 c -4.16686623566599 9.15008878734691 0.66460838509394 c -3.47428392792905 10.83484642789696 1.61654418916363 h -5.97916588966148 9.33054783508735 -1.18517177286445 c -6.73429704395157 11.16522015798626 -1.71187123134434 h -6.87081460412053 7.13194074549240 -2.40382689180414 c -6.17483737314576 -1.40431615043033 0.35064661211057 c 6.88862929595305 3.66566967017463 -2.04257484456917 h 5.81815236958954 10.36067791409610 2.43487614444713 c 9.27643776921037 8.58276363415267 -0.27896451556628 c 5.98418738145442 5.89530954870927 4.69667844519151 h 9.28241497053218 6.23620207921102 4.27006913393590 h 9.40854853649909 2.96825397323697 1.15563641766263 h -7.69089257185894 -0.42058135152678 2.33398136583453 c 3.10279636998086 6.63840022537168 0.56596990431846 h 4.78557388155641 7.32427901630002 -2.24782415168916 h -0.61985458285576 8.21288121156019 4.21594619469696 h 4.13694896353771 9.92964441570768 3.56524439048560 h 5.27019009934220 11.71153422185304 0.96151394825222 h 7.19475642253468 11.29544198599456 3.66758864502054 h 10.75019112750154 9.41561473920532 0.91376042333520 h 8.82201691387873 9.94238645888279 -1.77517243980144 h 10.07480688606718 6.89540896488067 -1.17646405758161 h -0.52796760362335 -9.97923845401637 -3.56068630846230 c 1.51429893828903 -8.55006499650452 -3.89636124938306 c 2.63322525860464 -6.84738636869616 -1.94132706693732 c 2.45231255273822 -8.58503011605659 -5.73270297448227 h 5.37815654946076 -7.47885352035504 -1.14538738164649 c 2.61988618853550 -4.89525633505426 -2.66564692674281 h 1.43595847290828 -6.86931753652319 -0.24210737626058 h -1.30958674404374 6.52338548111491 3.27148596023107 c 5.44951904162863 -10.07351211518398 0.13594480006087 c 7.16718893015768 -7.50723143087068 -3.42384679457539 c -8.30586540811605 7.29097691273191 -3.86339739466067 h -5.93464763938382 4.79540991144479 -1.77310368620488 c 8.12062849282527 -6.09879899044537 1.52467407647958 h 1.09345134870450 4.05968150900841 5.36043347261020 h 4.96410664932338 -5.51033909084144 2.40182364011740 h -1.53889513274786 -9.98163257272499 -1.76962788493607 h -1.27040336612665 -11.16945844404465 -5.05905750124074 h 6.73424819810534 -9.06599654630477 -4.71407761172165 h 9.12467732417352 -7.75494407471329 -2.78750918902806 h 7.06871037671708 -5.74254155312886 -4.50114738377804 h 4.22594803301512 -10.10150420209863 1.80444514069996 h 7.37526666820597 -10.53422342608848 0.74720525472397 h 4.80751715869135 -11.55846115322976 -1.15479831322020 h $end

[awvwgk]

cc @sespic

[sespic]

In xtb, there is currently no possibility to constrain an imaginary frequency (TS optimization). Thus, a general prerequisite is that the DFT TS geometry is also a TS on the xtb potential energy surface. In your example, this seems to be the case, since hess reproduces the correct imag. frequency. SPH calculations apply a RMSD constrain so that the geometry stays as close to the DFT structure as possible during a biased geometry optimization. The difficulty in your example is that the initial gradient is very large and the hydrogen is re-attached to the methyl group already after 3(!) optimization cycles at the GFN2-xTB level of theory. You can check this in the xtbopt.log file that is written. A constrain to prevent this must be extremely high and is not recommended. So in your example, SPH calculations unfortunately don't work for the TS. The good news is that hess gives the correct result and can thus be used here.

[acd81]

Thanks for the swift response. I am well aware that the initial DFT TS geometry rapidly deteriorates during the biased geometry optimisation as far as crucial N-H and C-H distances for a seeminly simple protonylysis are concerned. Even a more strict RMSD constraint (0.001 instead of the recommended 0.08) doesn't work either as the initial DFT TS rapidly relaxes to either reactant or product sides in biased geometry optimisation. I've even tried to fix the crucial four centres of the protonolysis TS during metadynamics, but to no avail.

Is their any other way (other than a more strict RMSD criterion of fixing of crucial centres) to ensure that the DFT TS geometry doesn't deteriorate to such significant degree as in the current example.

Moreover, I have tried SPH for another rather straightforward type of TS structure - olefin insertion into a metal-carbon bond via a classical planar 4-centre TS structure. Here as well, SPH fails as the DFT TS geometry again becomes significntly diffent upon biased geometry optimisation. One wonders as to whether the story in the parant publication (J. Chem. Theory Comput. 2021, 17, 1701−1714) about the TS for H2 addition onto an Ir pincer complex has been spun too optimistically?

Any further insight into these issues are highly appreciated.

[sespic]

The Ir-princer complex was a proof-of-principle example to show that the sign function used in Eq. 13 of the original publication works. No extensive testing on the general performance of SPH on TS geometries was done. Further, the scaling function is an approximation to remove the added bias from the vibrational frequencies. However, for a very large bias (small RMSD enforcements), this approximation is invalid. For this particular example, any kind of geometry optimization (constrained or not) can't be applied and the results from the hess calculation have to be taken instead. I am pretty sure there are many other examples where SPH will fail as well.