Jonathan E. Moussa

Results 40 comments of Jonathan E. Moussa

What range of problem sizes are you working with? Despite being single-core calculations, MOZYME calculations are usually faster than conventional, multi-threaded MOPAC calculations beyond a few hundred atoms. You can...

We do plan to develop our own Python interface to MOPAC, but there are already two Python interfaces available that I am aware of: MOPAC can be used as a...

So far, we have not attempted to reactivate the old GPU support. It was not a priority for the initial release, and our development plans include an eventual overhaul of...

The presently-deactivated GPU support in MOPAC is unlikely to be helpful for large protein models. It only accelerates conventional, cubic-scaling MOPAC calculations [https://doi.org/10.1021/ct3004645], and you become limited by both GPU...

No, I am pointing out that there are two distinct GPU-related developments in MOPAC, both out of the Rocha group at the Federal University of Paraiba in Brazil, but from...

The existing MOZYME solver is based on an approximate sparsity-preserving matrix factorization process (called "pseudodiagonalization"), and the closest standard sparse matrix method would be ILU factorization. The underlying matrix primitive...

You'll need to give me some time, but I should be able to provide both some matrix examples of various sizes and also a minimal, self-contained single-core reference implementation of...

@aizvorski I have prepared a [first version](https://github.com/godotalgorithm/sparse-pseudo-diag) of a self-contained proxy for the MOZYME solver. It still needs work, but it is already a reasonable reference for any attempts at...

@aizvorski It is unlikely that using something like cusparseSpGEMM will produce good performance for 2-by-2 rotation matrices. I do expect that hand-written CUDA code would be needed to accelerate the...

This is a reasonable request, and the possibility of AM1-BCC in MOPAC has come up before in development discussions. The AM1 atomic charge part of the AM1-BCC model is straightforward...