Notes
This is a recommended release for all users. Thanks to everyone who contributed directly, reported an issue, or suggested an improvement. There are many quality of life improvements, bug fixes throughout the application, and updates to the associated testing. As previously announced, the legacy CUDA support (QMC_CUDA=1) is removed in this version. For GPU support, users should transition to the offload code which is more capable and fully usable in production on NVIDIA GPUs.
This version is intended for long-term support of v3 of QMCPACK. Development effort is now focused towards v4. Contributions of tests, fixes, and features from users and developers are still welcome to v3 for a potential future release. However, these will not be ported towards v4 by the core QMCPACK developers without prior arrangement. Please discuss options with QMCPACK developers.
- Simplified checkpointing and enabled it in the batched drivers. Users now only need specify checkpoint={-1,0,N} to checkpoint between blocks. #4646
- NERSC Perlmutter build recipe. #4698
- qmc-fit: Now supports parameter fitting with jackknife for e.g. DFT+U, EXX scans #4475 and for equation of states and morse fits #4518
- Improved error checking including NaN checks to protect against potentially unreliable compilers and libraries, #4697, and checks on GPU matrix inversion #4693
- Significant advances in orbital optimization capability, focusing on LCAO wavefunctions. Development is ongoing for multideterminant support and for spline wavefunctions. See e.g. the Be atom orbital optimization test #4626, #4619, reading and writing of orbital rotation parameters #4580, support for disabled/frozen parameters #4581.
- Magnetization Density Estimator for non-collinear wavefunctions #4531
- Pathak-Wagner regularizer for forces #4477
- The legacy CUDA implementation, the version built with QMC_CUDA=1, has been removed from the codebase, #4431, #4632,#4499, #4442.
- For increased performance with current AMD GPU support, new QMC_DISABLE_HIP_HOST_REGISTER option is enabled by default for ROCm/HIP builds. #4674
- Bugfix: J1Spin indexing was wrong #4612
- Bugfix: 1RDM estimator data written to stat.h5 was incorrect #4568
- Introduced ENABLE_PPCONVERT option and skip ppconvert compilation when cross compiling. #4601
- Faster builds compared to v3.16.0 due to code refactoring #4682
- Many refinements throughout the codebase, cleanup, improved testing.
NEXUS
- Nexus: Equilibration detection algorithm is now deterministic #4557
- Nexus: Support for Kagayaki cluster at JAIST #4598
- Nexus: GPU support fix for NERSC/Perlmutter #4699
- Nexus: Use simplices in convex_hull to support newer scipy versions #4671
- Nexus: Add pdos flag for Projwfc #4655
- Nexus: Adding crowds_serialize_walkers tag to dmc input list #4651
- Nexus: Qdens handles batched driver input/output #4645
- Nexus: Fix namelist read for Projwfc input #4644
Known problems
- When offload builds are compiled with CUDA toolkit versions above 11.2 using LLVM, multideterminant tests and functionality will fail, seemingly due to an issue with the toolkit. This is discussed in llvm/llvm-project#54633 . All other functionality appears to work as expected. As a workaround, the CUDA toolkit 11.2 can be used. The actual NVIDIA drivers can be more recent.