GPU4PySCF v0.5 release
Bugfixes
- Align the initial guess with PySCF
- Fix CUDA _sync_threads in contract_rho
Improvements
- Compress CDERI in the 'ij' direction
- Save CDERI in CPU memory, and async data transfer between CPU and GPU
- Evaluate spherical AO on grids
- Make use of the sparsity in DFT integrals
- Tensor contraction with cupyx.cutensor
- Improve VV10 calculations
- Multiple streaming for gradient calculations