Simplest potential to minimize molecule

[YuraM]

Dear all,

just out of curiosity, what is the simplest potential to optimize the molecule/evaluate the energy? Molecular mechanics? Really need something very simple and fast to screen a lot of potential complexes.

Is it possible to get the reliable charges and C6 (Lennard-Jones coefficients ) from SP DFT? As a first guess I think only the non-bonded terms from the molecular mechanics would be sufficient.

Any hint/advise is appreciated.

[VTur]

to optimize the molecule/evaluate the energy - это разные вещи. from the molecular mechanics Вы получите приблизительно равновесную геометрию, но стерическую энерргию (т.е. превышение энергии над гипотетической равновесной)
Если нужно в to optimize the molecule/evaluate the energy и то, и другое, тогда LDA.

[Гесс]

Молекулярная динамика (в амбере можно делать "оптимизации", хотя это бредовато), потом DFTB (системы на десятки тысяч атомов), полуэмпирика (в параллельных вариантах (Empire) тысячи атомов, но я незнаю где уже есть импелементированные дисперсионные поправки), HF-3c (ORCA, я попробовал один раз, как то не впечатлило по скорости на размер системы, возможно я неправильно готовил эту кошку), ну а там уже и DFT видно.

[VTur]

LDA быстрее HF. Прочти все функционалы без обмена быстрее HF (есть спецускорители).

Вопрос в изначальной оптимизации. Есть ли знание по исходной геометрии, или нужно проводить конформационный анализ, или отжигать молекулу?

[YuraM]

Is Empire significantly faster than e.g. MOPAC?

Te problem with the molecular mechanics is that you have to know the atom type. There are two issues with that:
a) often it is difficult to obtain the atom type in the automatic fashion based on only Cartesian coordinates (there is BABEL, but I am not sure it can assign the atom type properly in all the cases)

b) in most of the cases there are some atom types which are not define in the force field, e.g. MM3, unless you deal with the simplest organic molecule. Of course there is UFF, but someone told me that this is the least accurate force field.

Another way would be to optimize the molecule at DFT first and try to get the missing parameters, like angle bends, bond stretch, torsions, van-der-Waals and charges. While first three terms are straightforward to obtain, I am not sure regarding the charges and C6 coefficients. Is there a way to extract them from DFT? Because if you extract them from DFT I think these parameters for the similar molecules are much better than coefficients available from general force field.

Ideally I need a method of crude optimization which would cost me 1*10-5 sec per molecule (5-70 atoms). But I suspect that I want too much...

[Гесс]

Is Empire significantly faster than e.g. MOPAC?

Never worked with MOPAC. You can request singlenode (core-parallel) academic version for free http://www.ceposinsilico.de/products/empire.htm Program is in active development.

Te problem with the molecular mechanics is that you have to know the atom type.[...]

Yes, thats one of the reasons why i prefer semiempirics.
I got some unpleasant experiences on getting parameters for new atom in GAFF, i would say its quite challenging and to be sure in a result you really need a lot of work.

Ideally I need a method of crude optimization which would cost me 1*10-5 sec per molecule (5-70 atoms). But I suspect that I want too much...

I hope it was a typo. 10 microseconds per molecule? Do you plan to recalculate all known compounds in few days? From another side 1*10⁵ sec mean 3 hours for 70 atomic molecule. Thats little bit expensive for hybrid DFT even with RI and so on approximations, on LDA i have no experience. For semiempirical optimisation it is not a problem at all - just from yesterday - 180 atoms, optimisation with AM1, RHF, took 227 seconds. Nonbonded interaction can (and will) be quite shitty. Obviously MD can be as long as you want. I still didn't got do you need to search for the "best" conformer or just want to optimise what you gave to a programm to nearest minimum.

[YuraM]

Thank you, nice joke. No, it was not a typo, but rather too optimistic way of thinking.
I think the goal now is not properly defined. I am thinking on the global minimum optimization problem in general in chemistry.
Also, I am thinking on improving my coding abilities. Would it be difficult (or even better say time-consuming) to implement molecular mechanics optimization? Does someone have an experience? Of course there are many existing codes, but if this is not too difficult I would really enjoy doing it myself. To implement the energy is straightforward. There are formulas.
The problem is the energy minimization wrt to the Cartesian coordinates. Of course, the simplest way is to go for numerical first/second derivative - but then I believe the code would be very slow. Therefore analytical first derivative and conjugated gradient method are especially attractive.
Just wonder if the analytical first derivatives for molecular mechanics are published somewhere.

[Ahha]

I think the goal now is not properly defined. I am thinking on the global minimum optimization problem in general in chemistry.
Also, I am thinking on improving my coding abilities.

Well, let me drop by, too. If this above was not another joke, I would recommend you to think towards parallelized algorithms. That's what is a trend nowadays, in my opinion. A single processor's performance has raised but little over last years while number of effective cores in multicore architectures has grown vastly. It doesn't necessarily mean that you'll get tens of microseconds per molecule to optimize, of course, even by molecular mechanics, but this is the road that most likely leads to your success. Or at least I think so.