sparse grid ao along (0,2) axes #114
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
| @@ -160,7 +160,7 @@ | |||
| " \n", | |||
| " for line in output.split(\"\\n\"):\n", | |||
There was a problem hiding this comment.
Did we check if the unit [eV] was correct (or if it should be [meV])? I think it'd be neat to have a cut-off around 42meV (chemical accuracy). If the units are [eV] then we're probably sparsifying a bit too much (150eV error is >1000x worse error than ML models).
There was a problem hiding this comment.
Yes, the unit was correct. The high error we saw was for 32 carbon atoms placed in a line - the error for placing atoms in a line is high regardless of the grid sparsification (c_32 on the graph).
| grid_AO_dm = grid_AO[0] @ density_matrix # (gsize, N) @ (N, N) -> (gsize, N) | ||
| grid_AO_dm = jnp.expand_dims(grid_AO_dm, axis=0) # (1, gsize, N) | ||
| grid_AO_dm = grid_AO[0] @ density_matrix # (gsize, N) @ (N, N) -> (gsize, N) | ||
| grid_AO_dm = jnp.expand_dims(grid_AO_dm, axis=0) # (1, gsize, N) |
There was a problem hiding this comment.
Amazing we can use the same code for this!
| @@ -199,10 +199,18 @@ def init_dft_tensors_cpu(mol, opts, DIIS_iters=9): | |||
| grid_weights = grids.weights # (grid_size,) = (45624,) for C6H6 | |||
| coord_str = 'GTOval_cart_deriv1' if mol.cart else 'GTOval_sph_deriv1' | |||
| grid_AO = mol.eval_gto(coord_str, grids.coords, 4) # (4, grid_size, N) = (4, 45624, 9) for C6H6. | |||
| if opts.ao_threshold is not None: | |||
| if opts.ao_threshold > 0.0: | |||
There was a problem hiding this comment.
Really like the simplicity: we just pre-process remove and only change grid_size so exchange_correlation code remains the same!
blazejba
force-pushed
the
bb/low-hanging-sparse-grid-ao
branch
from
b26931d to
cd2b2d5
Compare
When
ao_thresholdarg is specified, it reduces the grid size, if any slices along this axis are all zeros. The reduced grid_size applies togrid_AO,grid_coords, andgrid_weights.