- #1

Spathi

- 73

- 7

- TL;DR Summary
- We carry out a calculation with a quantum chemical method, for example DFT, and then use the calculation results as a sample for training the molecular mechanics functional;

Experts in physics, physical chemistry and organic chemistry are invited to the thread. I work in the field of quantum chemistry, and have plans of implementing molecular mechanics calculations in my program Chemcraft. Some people say that molecular mechanics can work rather well in some cases, e.g. for finding conformations of organic molecules. And I have a feeling that I can implement something new.

The general idea is as follows. We carry out a calculation with a quantum chemical method, for example DFT, and then use the calculation results as a sample for training the molecular mechanics functional; we adjust the parameters of our MM functional so that we get the best agreement with the results of the DFT calculation. And then we use the resulting custom MM functional to perform calculations for a more complex problem for which the original DFT method is too expensive.

As an example: firstly we calculate the vibrational spectrum with DFT (the second derivatives of energy with respect to the coordinates of atoms), then we use the force field and other parameters to fit the MM functional, and then we use the resulting MM functional to calculate the anharmonic force field (the third derivatives with respect to energy), and this allows you to predict overtones in the vibrational spectrum, or for example more accurate vibrational entropy.

So, for such an approach, it is necessary to implement the MM model in such a way that the MM calculation in it is rather not “good”, but “non-empirical” (“ab initio”). In other words, this MM model should be based on some universal, fundamental principles; then, in the general case, fitting the parameters of the MM functional will work well. What do I mean by fundamental principles? For example, steric repulsion: unbound atoms at distances near the van der Waals radius usually repel each other. An example of another universal principle is the Lennard-Jones potential. How versatile is its formula?

I will write more specifically about my ideas and questions later.

The general idea is as follows. We carry out a calculation with a quantum chemical method, for example DFT, and then use the calculation results as a sample for training the molecular mechanics functional; we adjust the parameters of our MM functional so that we get the best agreement with the results of the DFT calculation. And then we use the resulting custom MM functional to perform calculations for a more complex problem for which the original DFT method is too expensive.

As an example: firstly we calculate the vibrational spectrum with DFT (the second derivatives of energy with respect to the coordinates of atoms), then we use the force field and other parameters to fit the MM functional, and then we use the resulting MM functional to calculate the anharmonic force field (the third derivatives with respect to energy), and this allows you to predict overtones in the vibrational spectrum, or for example more accurate vibrational entropy.

So, for such an approach, it is necessary to implement the MM model in such a way that the MM calculation in it is rather not “good”, but “non-empirical” (“ab initio”). In other words, this MM model should be based on some universal, fundamental principles; then, in the general case, fitting the parameters of the MM functional will work well. What do I mean by fundamental principles? For example, steric repulsion: unbound atoms at distances near the van der Waals radius usually repel each other. An example of another universal principle is the Lennard-Jones potential. How versatile is its formula?

I will write more specifically about my ideas and questions later.

Last edited by a moderator: