lowdanie

From my understanding, accurate simulations at the electron level (post Hartree Fock / DFT) are currently limited to 100 atoms (on a gpu cluster this can take hours or days). Maybe this can be pushed to 1000 atoms with aggressive optimization techniques like FMM.

So at this level of simulation it is currently only possible to simulate one medium size molecule or the interaction of a few small ones.

To simulate larger systems, it is necessary to work at a (semi-)classical level of abstraction that approximates quantum mechanics. For example using molecular dynamics to essentially simulate a fluid with a ball and springs model. In this case, electron level simulation can still be useful for deriving heuristics (conceptually, the spring tension).

I completely agree that it’s interesting to investigate how far the electron level simulation can be pushed.

Theoretically yes, but the method that is currently implemented (Hartree Fock) is notoriously inaccurate for molecular interactions. For example it does not predict the Van Der Waals force between water molecules.

I’m planning to add support for an alternative method called density functional theory which gives better results for molecular interaction.

Thanks!

Thanks!

The short answer is yes, but either memory (if the electron integrals are cached) or runtime (if they are not cached) currently scales like O(n^4) where n is the number of atoms.

In cached mode, it can currently jit compile the graph for molecules of around 10 atoms in ~5 minutes on one T4 gpu. Once the graph is compiled, the actual geometry optimization only takes a few seconds.

I’m working on optimizations that improve the scaling behavior (such as density fitting) with the goal of achieving similar or even better performance for molecules with ~50 atoms.

Since the gradients are computed with jax, the library can be used to differentiate with respect to all inputs including nuclear charge and basis set parameters (exponents and contraction coefficients). I agree that computing gradients for the nuclear charges could be interesting in the context of molecular design.

But for the colab demo I thought that sticking to nuclear positions (i.e atomic forces) would be easier to visualize.

Thanks!