Add Linear Nk OCCRI method + ISDF #155
Conversation
…ks and write kpt functions.
…imized liibraries.
🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
kosm6966
changed the title
|
Is it better to put these codes in pbc/df? |
| const int nthreads = omp_get_max_threads(); | ||
|
|
||
| // Initialize FFTW multithreading support | ||
| if (!fftw_init_threads()) { |
There was a problem hiding this comment.
There are several exit() in this c code. Could you return these errors to the Python side and raise an exception in Python?
|
|
||
| def __init__( | ||
| self, | ||
| mydf, |
There was a problem hiding this comment.
-
This requires a PBC mean-field instance.
mydfoften refers to the density fitting object. It's better to use a different name. -
The OCCRI object does not depend on the mean-field instance. Passing it to the
__init__function looks a little bit misleading. I think it's better to provide a factory function or a class method likeOCCRI.from_mf(mf), which create a OCCRI instance and validate themfthere.
| if kpts is None: | ||
| kpts = numpy.zeros(3, numpy.float64) | ||
| kpts = kpts.round(6) | ||
| self.kmesh = [ |
There was a problem hiding this comment.
This can potentially result in incorrect kmesh for numerical issues, or non-orthogonal lattice. It's recommended to convert kpts to kmesh using this API https://github.com/pyscf/pyscf/blob/fafef00a03b28a9ddbe23cad2d99ce0f1222180c/pyscf/pbc/tools/k2gamma.py#L39
|
|
||
| # Step 1: Get pivot points | ||
| logger.debug(self, "Selecting ISDFX pivot points") | ||
| interpolation.get_pivots(self) |
There was a problem hiding this comment.
The pivots, aovals, and W attributes of self are initialized in the interpolation module. It's recommend to return these variables and explicitly assign them to in this build method.
|
|
||
| return make_natural_orbitals(self.cell, self.kpts, dms) | ||
|
|
||
| def __del__(self): |
There was a problem hiding this comment.
Why do you create an empty __del__?
| @@ -0,0 +1,20 @@ | |||
| # Copyright 2025 The PySCF Developers. All Rights Reserved. | |||
There was a problem hiding this comment.
The __init__.py in the test directory will cause the test code to be included in the wheel package. This is typically not needed and can be removed.
| @@ -0,0 +1,167 @@ | |||
| #!/usr/bin/env python | |||
There was a problem hiding this comment.
The pytest suite is already configured in the package. It can automatically discover all tests. A manual test driver is unnecessary.
##Summary
• Add OCCRI method for efficient exact exchange evaluation in periodic systems with significant algorithmic improvements.
• Performance is similar to but typically faster than existing occri method
• Designed to integrate with isdfx
• Provide memory-efficient implementations with MO blocking following PySCF patterns
• Implement ISDFX (Interpolative Separable Density Fitting eXchange) module with comprehensive testing and examples
• Exchange build is linear with N_k as shown by Example 2
• Default threshold is set and provides < 50 μHa error for all test systems
• Include comprehensive test suites and educational examples demonstrating usage and performance
Features
• OCCRI: Exchange matrix evaluation using occupied orbital resolution of identity
• ISDFX: Tensor hypercontraction-based exchange with interpolative separable density fitting
• Support for RHF, UHF, RKS, UKS and k-point variants (KRHF, KUHF, KUKS, KRKS)
• Memory management with automatic MO blocking for large systems
• Comprehensive utility functions separated from object methods for better code organization
• Natural orbital construction and exchange matrix building utilities
Performance & Accuracy
• ISDFX provides near-FFTDF accuracy (typically <50 μHa energy differences)
• Scaling analysis shows sub-quadratic performance with k-point sampling
• Memory-efficient MO blocking prevents memory overflow for large systems
• Proper handling of asymmetric k-point meshes and complex conjugation
Testing & Examples
• 84 comprehensive tests covering energy accuracy, k-point handling, and edge cases
• Energy comparison tests with 50 μHa tolerance vs FFTDF reference
• Asymmetric k-point tests to catch symmetry and conjugation bugs
• Educational examples: basic usage, AFM initial guess setup, scaling analysis
• Integration with existing run_tests.py framework with --perf flag support
Code Quality
• Follows PySCF coding conventions with comprehensive docstrings
• Separation of pure algorithms from object methods for better maintainability
• Formatted with Black, isort, and flake8
• Extensive error handling and input validation
• Memory cleanup following PySCF patterns
Test plan