v1.9.0

• New Features (PBC systems)
  • Stress tensor calculations using Gaussian density fitting at the Gamma point and with k-point sampling. (#896)
  • Nuclear gradients and stress tensors with Becke grid response. (#887, #906, #921)
  • All-electron core Hamiltonian nuclear and strain derivatives using GDF. (#916)
  • Atom-resolved magnetic moment initial guesses. (#892)
  • Experimental numerical Gamma-point Hessian and phonon modes using phonopy, without non-analytic corrections. (#860)
  • GHF and collinear GKS at the Gamma point and with k-point sampling; GKS currently supports semilocal functionals with Becke grids. (#942, #965)
  • Berry (Zak) phases, hybrid Wannier centers, and electronic and ionic polarization for closed-shell insulators. (#917)
• New Features (molecular systems)
  • MBIS and Hirshfeld atomic charges and multipoles. (#902, #911)
  • Support cuEST v0.2.0, including density-fitted MO integrals and range-separated hybrid energies and gradients. (#859)
  • GPU density-fitted CASCI and diagonal-Hessian CASSCF with a GPU direct-spin1 FCI solver (experimental). (#869)
  • COACH functional, combining VV10 with an ATM-only D4 correction.
  • SCF conversions between restricted, unrestricted and generalized HF/KS references. (#937)
  • SAD initial guesses, including Q-Chem-style guesses and selectable high-spin or spin-averaged atomic densities; high-spin SAD also supports periodic SCF. (#950, #975)
  • Mulliken population analysis, including generalized HF references. (#963)
  • Named solvents in PCM, with solvent-specific static and optical dielectric constants and explicit overrides shared with SMD. (#962)
• Improvements
  • Multigrid v3 with optimized density evaluation, analytical Fourier transform kernels, and reduced memory usage. (#865, #891)
  • Optimize PBC GDF short-range integrals and reduce memory usage in GDF gradients. (#924, #928)
  • Optimize molecular density-fitting gradient and Hessian kernels. (#934)
  • Combine short- and long-range exchange inside density-fitting kernels for energies, gradients, and Hessians. (#859)
  • Unify PBC nuclear gradient and stress tensor evaluation. (#910)
  • Analytical Ewald stress tensor and AFT exchange-divergence corrections. (#913)
  • Optimize nonlocal pseudopotential integrals and stress tensors. (#883, #912)
  • Cache QM/MM nuclear Coulomb energy across SCF cycles. (#914)
  • Upgrade LibXC to 7.1.2 while retaining compatibility with the 7.0.0 fork. (#935)
  • Add a tutorial for hybrid PySCF/GPU4PySCF workflows. (#882)
  • Improve density-fitting task granularity and batch auxiliary groups for integral builds, gradients and Hessians; reduce gradient workspace and improve TD gradient numerical stability. (#967, #970)
  • Batch AO pairs and reciprocal-space vectors to bound memory usage in PBC GDF diffuse integrals.
  • Update four-center integral CUDA kernels and the in-core CCSD integral driver. (#936)
  • Reduce host/device transfers in gradients and Hessians, and batch TDDFT-ris tensor uploads. (#972)
  • Refactor FSSH trajectory storage, checkpoint/restart handling and Wigner sampling. (#899)
  • Align CPHF linear-dependency thresholds with solver tolerances, including Hessian response calculations. (#951)
  • Support custom shared-library search paths. (#957)
  • Support standard CMake CUDA architecture settings. (#952)
• Fixes
  • Multigrid v3 integrals, densities, nuclear gradients and stress tensors for skewed lattices, and integral screening. (#976, #979)
  • Complex-valued PBC second-order SCF and time-reversal symmetry enforcement. (#964, #974)
  • Molecular and periodic GHF spin-block handling used by GKS calculations. (#965)
  • Incorrect ECP-center contributions in TDDFT gradients and non-adiabatic couplings for systems mixing ECP and all-electron atoms. (#971)
  • Incorrect range-separated exchange matrix construction. (#947)
  • Contracted basis dimensions in the PBC J engine. (#945)
  • PBC GDF initialization when switching from Coulomb-only to exchange calculations. (#960)
  • Frozen orbitals in in-core CCSD. (#961)
  • MO dimension handling in the vovo AO-to-MO integral transformation.
  • TDDFT-ris integral buffer sizing under limited GPU memory. (#967)
  • Non-contiguous weighted-density arrays in UKS Hessians. (#958)
  • SCF damping ignored with PCM and SMD solvents. (#949)
  • Respect custom range-separation omega in GPU LibXC functionals and grid-response Hessian terms. (#923, #929)
  • Incorrect screening for mixed range-separated exchange in unrolled K kernels. (#909)
  • GPU OOM in density-fitted TDHF gradients and PBC GDF and RSJK derivatives. (#885, #907)
  • Insufficient workspace for molecular J/K gradient kernels. (#879)
  • Illegal memory access in large PBC RSDF calculations on multiple GPUs. (#862)
  • CUDA indexing, integer overflow, and CUTLASS stream handling issues. (#863, #868)
  • Race condition in DFT grid screening. (#925)
  • Incorrect periodic g-orbital values, second derivatives, and strain derivatives.
  • Incorrect int1e_r2_origi integrals and one-electron integral accuracy. (#857)
  • Incorrect k-point weights in constrained DFT gradients. (#856)
  • Multigrid v2 stress spikes and single-precision collocation kernels. (#855, #874)
  • Inconsistent orthogonal lattice checks in analytical Fourier transforms. (#918)
  • Mesh consistency in PBC geometry optimization, numerical Hessians, and FFTDF/AFTDF SCF scanners. (#889, #932, #933)
  • Stress units and center-of-mass constraints in ASE geometry optimization. (#878, #889)
  • UHF logging for systems with zero beta electrons. (#853)
• API Updates
  • Multigrid v3 replaces multigrid v2 as the default multigrid integrator.
  • Add lr_factor and sr_factor parameters to the 2c2e and 3c2e integral APIs. (#861)
  • Remove deprecated dft_driver, opt_driver, and opt_3c_driver modules. (#858)