Dear Geoff, As promised, here are the molecules we internally used for testing and validation of our QEq implementation. I've attached them as a tar file containing files in COF and MOL2 format. I tried a zip file first but apparently that extension is blacklisted for the mailing list. I hope the .tar file will work, and my apologies if you receive this twice.
The results are divided into three categories: * Metal halides (covering Table II in the Rappé and Goddard paper; we used lambda = 0.5) * Molecules (covering Tables III and (partially) IV) * NaCl clusters (covering Figure 6) We took some care to use the geometries from the references given in the paper. For atoms with a principal quantum number of n = 4 and higher, the analytical expressions we use for the Coulomb integrals become numerically unstable. For those cases we use a rescaling to n = 3 (using equation 17 from Rappé and Goddard). For molecules involving those atoms, the deviations from the values in the literature are thus slightly higher. The only molecules, as far as I know, with charges that really do not match those in the QEq paper of Rappé and Goddard, are lithium hydride (LiH) and silane (SiH4). For SiH4 I suspect that there's an error in the table, and that the hydrogen charges reported for silane actually belong to the molecule directly above that in Table IV, which is ketene, H2C=C=O. For LiH I don't know what the problem is. Any change to the parameters or functional forms for hydrogen will change the hydrogen charges on other molecules so that *they* no longer match, so I could not find a way to reproduce the LiH charges. Kind regards, Paul Becherer Development Scientist Culgi B.V. Galileiweg 8 2333 BD Leiden The Netherlands +31-71-332-2056 -----Original Message----- From: Geoffrey Hutchison <geoff.hutchi...@gmail.com> Sent: Saturday, March 02, 2019 23:14 To: Paul Becherer <paul.beche...@culgi.com> Cc: openbabel-discuss@lists.sourceforge.net Subject: Re: [Open Babel] Contributed code for new file format > Looking at it again now, it seems that the charges from Open Babel's QEq > often have the wrong sign or are simply unreasonably large. I have no idea > what causes that, as I only had a cursory look at the code. But it looks as > if there's more going on than just the additional simplifications to QEq: > even the QTPIE charges do not seem to reproduce those from the QTPIE paper by > Chen and Martinez (e.g. the charges on phenol as reported in the preprint on > arXiv, https://arxiv.org/abs/0807.2068 ). I cannot rule out that I've used > the charge model options in the wrong way, but they seemed quite > straightforward. Ironically, the code for both QEq and QTPIE in Open Babel were contributed by Chen. But I've heard similar things about the signs and magnitudes of the QEq charges. Naturally, considering the source of the code, I didn't question the results at the time, but it's been several years with off-and-on questions about it. > I can send the molecule geometries that we used in validating our QEq against > the Rappé and Goddard article, if that would be of help, or run our QEq > implementation on a set of molecules of your choice and send back the > results. Our coverage of elements is limited to that in the original paper, > which covers H, Li, C, N, O, F, Na, Si, P, S, Cl, K, Br, Rb, I, Cs. If you're willing to send some molecule geometries and charges, that would be very helpful. I probably won't have time to evaluate the code until the summer, but having a validation set would be a helpful contribution. Incidentally, I've been following the work by the Goddard group on their new polarizable QEq model (e.g., PQEq2) and I'm trying to get a test set together for an independent implantation of those. -Geoff
QEQ_results.tar
Description: QEQ_results.tar
_______________________________________________ OpenBabel-discuss mailing list OpenBabel-discuss@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/openbabel-discuss
