Add GPU density-fitted CASCI and diagonal-Hessian CASSCF - #869
Open
wtpeter wants to merge 4 commits into
Open
Conversation
sunqm
approved these changes
Sep 1, 2026
| required_memory = ((2 * max_space + work_vectors) * vector_bytes + | ||
| max_space**2 * x0.dtype.itemsize) | ||
| available_memory = get_avail_mem() | ||
| memory_limit = min(max_memory * 1e6, available_memory) |
Collaborator
There was a problem hiding this comment.
max_memory in pyscf controls CPU memory. Since GPU memory is assumed to be dedicated to a single process, we can ignore max_memory when estimating GPU memory usage.
| f'Failed to allocate GPU Davidson subspace of size {max_space}') \ | ||
| from err | ||
|
|
||
| converged = cp.zeros(nroots, dtype=bool) |
Collaborator
There was a problem hiding this comment.
converged can be put on host memory, as a numpy array
| out += (corevhf,) | ||
| timing = getattr(casci, 'timing', None) | ||
| if isinstance(timing, dict): | ||
| timing['h1e_wall'] = (timing.get('h1e_wall', 0.) + |
Collaborator
There was a problem hiding this comment.
wall time is inaccurate in the GPU code.
If you just need to print cost of individual step, it's recommended to use gpu4pyscf.lib.logger.timer functions. If you want to collect the timing data, you need to use the cupy Event object, see also gpu4pyscf.scf.jk._TimingCollector
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR adds initial GPU support for density-fitted CASCI and CASSCF.
The main additions are:
direct_spin1FCI solver with GPU Hamiltonian contraction, Davidson diagonalization, and RDM construction.DF.ao2mo.DFCASCIsupport for restricted HF and KS references.DFCASSCForbital optimizer.The CASSCF orbital step uses a positive diagonal-Hessian model with a common level shift. The final orbital rotation parameters are limited by
max_stepsizewithout increasing the level shift solely to enforce the step limit.Current limitations
MemoryErrorwhen GPU memory is insufficient.This implementation is intended as a lightweight GPU CASSCF fallback rather than a replacement for the more robust second-order CPU optimizer.