8
$\begingroup$

DFT is a computational tool that is used in optimizing and calculating the electronic structure properties of molecules.

Are there any machine learning codes that can do something similar in a shorter time? Is there any software that does this?

$\endgroup$
  • $\begingroup$ I believe there was a NDDO-SEMO method that employs machine learning - you might want to try that out. $\endgroup$ – ANZGC FlyingFalcon May 3 at 0:56
14
$\begingroup$

There is one Python package that I know: AMP (Atomistic Machine-learning Package). It is based on ASE (Atomic Simulation Environment), a homogeneous interface to a lot of computational chemistry packages.

Basically, after you obtain geometry trajectories through ASE (using any computational chemistry package you like), an approximation to the potential energy surface (PES) is machine-learned. It can be used as an ASE interface.

You can theoretically do anything ASE is capable of with this, from saddle point optimization to quantum dynamics. I am not an expert in this field, but they seem to apply ideas from convolutional neural networks to molecular structures for describing the chemical environment surrounding atoms. Their method is described in Khorshidi, A.; Peterson, A. A. Amp: A Modular Approach to Machine Learning in Atomistic Simulations. Computer Physics Communications 2016, 207, 310–324.

| improve this answer | |
$\endgroup$
2
$\begingroup$

If you're interested in metal complexes, there's a free software, Python-based tool called molSimplify. From their site:

Geometry optimization with density functional theory (DFT), a general procedure to obtain the ground state structures of a complex, is computationally demanding in terms of time and can also easily fail. Two main failure modes are 1) the expected geometry cannot maintain stable during the DFT simulation (e.g., ligand dissociation) and 2) the electronic structure of the optimized geometry is bad, which indicates the system of study is out of the domain of applicability of DFT. Either case can only be identified after a simulation completes, leading to a massive waste of the computational resources (and your time!).

To address this challenge, we built machine learning models to classify the simulation outcomes and readily achieved a good performance on the out-of-sample test data.

It's really handy when you have dozens of structures to create. It gives a good initial geometry for further optimizations, avoiding boring work to assemble each one by hand and fiddling with bond angles. For more details, see their paper Duan, Chenru, et al. “Learning from Failure: Predicting Electronic Structure Calculation Outcomes with Machine Learning Models”. Journal of Chemical Theory and Computation, vol. 15, no 4, abril de 2019, p. 2331–45. ACS Publications, doi:10.1021/acs.jctc.9b00057.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.