Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly  
We are NATO-free zone

Re^4: Question about analytical/numerical DFT hessian

Marco Lombardo

Dear Gena,

neither do I. Simply I read somewhere that if you want to compute accurate activation energies, you have to sum up the total energy and the zero-point energy. Now I think the two should be obtained using the same level of theory.


On Mon Dec 19 '11 7:39pm, Gena wrote
>Dear Marco,

>I didn't mean that you should optimize equilibrium geometry/TS in HF! You should do it as you did before, by DFT. I only suggest to check frequencies of obtained geometries using analytical RHF hessian to save the time. You should still use the difference of DFT energy between equilibrium geometry and TS - this way you get the DFT activation energies, not HF. Hope it's clear.

>Kind regards,

>On Mon Dec 19 '11 5:34pm, Marco Lombardo wrote
>>Dear Gena,
>>thanks for your kind reply. I agree with you than common sense and chemistry knowledge are helpful in understanding if a structure is a possible TS or a minimum on the PES (especially when you deal with almost known reaction mechanisms). I can also understand the technique of calculating the hessian using HF to confirm your assumptions.
>>Anyway, I thik these "shortcuts" are not useful if you want to calculate the ZPE and obtaining consistent values of activation energies. In other terms, using these methods you can obtain "qualitative" results and not "quantitative" ones (also if I know that DFT is not always reliable for energies...)


>>On Mon Dec 19 '11 4:14pm, Gena wrote
>>>Dear Marco,

>>>I am just a non-experienced user as well. But I have done quite a number of calculations similar to those you mentioned. So I got these questions too, and this is what I can suggest:

>>>1) You can use RHF analytical hessian calculation for the TS/equilibrium geometry you had obtained to save the time, this is way faster. It is of course not 100% honest stationary point confirmation, but in many not specific cases it is good enough for usual organic compounds. From my experience the DFT and HF geometries are pretty close for most cases. Again, if you are studying some tricky compound with weak interactions (cluster, charge transfer complex and so on) this definitely can't be said. On the other hand, DFT is not great for these cases either, in my opinion. I personally use semiempirical hessian which can be calculated in minutes as a starting hessian for TS search by DFT/MP2 in most cases, and it usually works. Then I perform RHF hessian calculation for TS confirmation.

>>>2) Also, in many usual cases it is quite obvious to distinguish TS from equilibrium geometry from chemical considerations. For example, I studied a bunch of similar substrates for Diels-Alder reaction and needed equilibrium geometries and TSs for energy barrier estimation to make some conclusions how structure of different substrates affects this barrier. And the point was formation of new C-C bond. So in final product this bond is roughly 1.5 , while in TS it is ~ 1.9-2.0 . So if during TS search I got 1.56 bond length, it was definitely final product, not TS, and I restarted calculation with other parameters (e.g. more accurate RHF hessian instead of semiempirical one for initial geometry). In opposite, if I got ~5 distance for this C-C new bond, it obviously meant that I had obtained initial reagents. The other thing is geometry look - in most cases you can distinguish products, reagents and TS from chemical considerations again. To conclude, I mean that you don't have to perform hessian calculation all the time if you suspect that you got the wrong geometry, you can understand it from common sense.

>>>3) If you perform a series of calculations of the same reaction for analogous molecules, the TSs most likely will be identical too. So in these cases you probably can calculate hessian only for several cases and then see if the structure of TS is right for other analogues from chemical consideration again, without hessian calculation.

>>>4) The problem of such time savings is that you can't publish such calculations, I guess, because they aren't honest. But from experimental point of view - it might be too expensive to perform all calculations honestly all the time. For example, if I hypothetically got 20 energy barriers from MP2 optimizations, then I have to perform 20 numerical MP2 calculations to do everything right even in obvious situations (like inversion barrier for bowl-shaped polyaromatic compound which is bowl-shaped in equilibrium and flat in TS). Believe me, it would be way longer than DFT. And if I need more information in less time - I have to sacrifice accuracy.

>>>I hope it could help you. I also hope that other users will correct me if the considerations above should never be taking into account for time saving :)

>>>Best regards,

>>>On Mon Dec 19 '11 2:09pm, Marco Lombardo wrote
>>>>Dear Firefly users,
>>>>first of all I'm not a computational chemist and I'm just beginning to learn Firefly. Back to the question, using a Core i7 CPU on Windows 7, I was able to optimize two organic reactants systems composed of 85 atoms (no metals) using B3LYP/6-31G(d) and to find the two correspondig transition states at the same level of theory. Each work takes 4 to 8 hours to complete and I think that these are quite impressive results (using this hardware) in terms of computational times.
>>>>Now, if I want to check I have minima and real TSs, I have also to run a HESSIAN calculation for each structure. Using DFT this kind of calculation (only METHOD=NUMERIC is available) takes more than 24 hours for each structure! (85*3 = 255 single point energies have to be computed and each one needs around 6-8 minutes... this means 25.5-34 hours).
>>>>I read somewhere in the forum that analytical DFT gradient method is planned, but it will (probably) not be present in the (near) future 8 release, since it is not a real priority giving only a rather small gain in computational times.
>>>>So, since frequencies have to be computed in order to confirm stationary points or saddle points on the PES, which is the best strategy to save time? I think there is no much advantage to gain 4-8 hours (with respect to other computational softwares) in the optimization steps only to be slow down in the next ones.
>>>>That said, Firefly is really a nice piece of software!


[ Previous ] [ Next ] [ Index ]           Mon Dec 19 '11 8:07pm
[ Reply ] [ Edit ] [ Delete ]           This message read 2154 times