FHI-aims CHANGELOG

How to understand this CHANGELOG (and FHI-aims):

Here, we list a rough overview of recent changes in FHI-aims. A complete record is available directly through the FHI-aims GitLab. Every user is welcome and encouraged to gain access to the FHI-aims GitLab. The GitLab repository is also the location where bug fixes, improvements etc. will be made available.

A final general note: Please have a look at the growing Wiki in the FHI-aims GitLab. You’ll find helpful information about compiling and running FHI-aims: https://aims-git.rz-berlin.mpg.de/aims/FHIaims/-/wikis/home

Changes since release version 240920_2

Dynamical Configuration Interaction (Marc Dvorak and Patrick Rinke, found and merged by Dorothea Golze and Qing-Long Liu)

• Inclusion of the dynamical configuration interaction (DCI) method (M. Dvorak and P. Rinke Phys. Rev. B, (2019) 99, 115134) which was developed by Marc Dvorak and Patrick Rinke, but left un-merged for many years. Dorothea Golze and Qing-Long Liu discovered, fixed conflicts and merged.
• The method is a a quantum embedding theory that combines Green’s function methodology with the many-body wave function, and is fully documented in the manual.

Warnings module (Yair Litman)

• Created a warning module to consolidate the output of warning messages and keep a counter that is printed at the end of the program execution
• This module does two things:
  1. adds a header and footer so the warning messages are easily parsed
     
  2. keeps a counter and writes at the end of the run the number of raised warnings

Usage:

Instead of using call localorb_info(info_str) use

Case a: One-line warning message

   call write_warning_message(info_str)

Case b: Many-line warning message

   call write_warning_message(info_str,'start')  
   call write_warning_message(info_str,'continue')  
   ...  
   call write_warning_message(info_str,'continue')  
   call write_warning_message(info_str,'end')

ScaLAPACK version of use_spin_texture, implementation of use_spin_texture_scf, and Spin Texture Code Refactoring (James A. Green)

• Fixed use_spin_texture with ScaLAPACK
• Implemented calculation of the spin texture on the SCF k-grid via the new keyword use_spin_texture_scf (LAPACK and ScaLAPACK). This produces output files spin_texture_scf.dat and spin_texture_scf.out which are the spin textures on the SCF k-grid, and are exactly analogous in format to spin_texture.dat and spin_texture.out created by the use_spin_texture keyword
• Refactored the spin texture code, including: creating a spin_texture module, moving some routines and variables from band_dos_mulliken for better encapsulation, and reducing duplicate and uneccessary code in the module.
• Implemented calculation of spin texture only for the required states in the energy range.
• Added a new regression test GaAs_spin_texture_Gamma to check the ScaLAPACK version of use_spin_texture.
• Bugfix for if use_spin_texture is requested with output band
• Added a stop in read_control if atomic_zora scalar or include_spin_orbit isn’t selected, since use_spin_texture requires these.
• Added a stop if use_spin_texture is requested without output band or output band_mulliken so that it doesn’t silently do nothing. Error message includes suggestion of new use_spin_texture_scf keyword

Updates to interface to LibRPA driver (Min-Ye Zhang)

• Added fold_C sub-option to output librpa to reduce generated file size while retaining precision in LibRPA calculation.
• Unified output format (E27.18) of double precision numbers of plain text data files.
• Refactored data export function from other post-scf calculations. They are now in src/interface_librpa_ext.f90
• Replaced get_pairwise_coeff_3fn with more memory efficient get_pairwise_coeff_3fn_tuned2 to generate 3-center RI-LVL coefficients
• Supported RI coefficients compression through output librpa csthres:<float>
• Fixed full self-energy matrix calculation with complex eigenvectors
• Fixed eigenvalue decomposition for scalapack and hybrid functional in LibRPA driver
• Added documentation section dedicated to using LibRPA driver, including keywords and examples

Enable selection of CODATA definitions (Sebastian Kokott):

• New keyword: codata VERSION
• If set in control.in, FHI-aims will use the corresponding CODATA definition of the physical constants, such as Hartree, and Bohr.
• VERSION is one of the following strings: 2002, 2018, FHIaims_pre2024. The version FHIaims_pre2024 refers to the internal definitions prior enabling this feature in 2024.
• Default: FHIaims_pre2024.

GW improvements (Min-Ye Zhang, Jannis Kockläuner, Qing-Long Liu)

• The GW+cumulant expansion (GW+C) method is implemented for molecules within the contour deformation framework as a method to calculate the spectral function of core states (Jannis Kockläuner).
  • New Keywords: cumulant_expansion, use_decoupling, cumulant_eta
• Added EXX band structure output in periodic GW (Min-Ye Zhang)
• A fix to respect state_upper_limit manually set in control.in for G₀W₀ calculation (Min-Ye Zhang).
• Functionality to skip evaluating completed GW bands for GW restart (Min-Ye Zhang):
  • A control.in flag periodic_gw_skip_completed_bands which allows to check if GW band output files exist at present working directory, and skip the re-evaluation of corresponding bands when they are found.
• Stop periodic GW with hybrid functional as a starting point reference (Min-Ye Zhang)
• Added a stop when an inappropriate blacsdim is chosen for wing vector calculation in periodic GW
• Read in x2c spinors or 1c molecular state from Turbomole to the aims GW calculation (Qing-Long Liu).
• Re-written the part of the evscgw0 routine in both 1c and 2c to remove redundant calculation of W0 (Qing-Long Liu).
• Minor speedup for periodic GW self-energy matrix (Min-Ye Zhang)

Bugfixes:

• Fixed the bug with two-component GW using PBE0 (Qing-Long Liu).
  • While using 2c GW the condense_basis_pair uses n_basis_soc_coll. If condense_basis_pair is called before SOC subroutine it will generate incorrect results because n_basis_soc_coll does not have a value. This bug will affect initialize_prodbas, then ovlp_3fn. While using PBE0, SOC part will keep previously generated incorrect ovlp_3fn, and let it go into GW part.
• Fixed small bugs in read in one-component Turbomole mos file and isolate the read in part of the code into a readin_turbomole.f90 file.
• Fixed Hedin shift bug in periodic gw code and added regression test for it.
• Fixed the bug of printing out the self energy for periodic GW.
• Fixed bugs with angularly averaged inverse dielectric matrix at long wavelength limit for periodic GW (Min-Ye Zhang).
• Change n_high_state bug in 2c GW. It’s not necessary to calculate all the virtual states (Qing-Long Liu).

Periodic BSE improvements (Ruiyi Zhou, Min-Ye Zhang, Yu Cao)

• Solves BSE and outputs file BSE_transiton_dipole.dat, which contains BSE eigenvalues and the corresponding transition_dipole.
• Implemented wing term correction for periodic BSE
  • The subroutines to compute wing term correction and averaged inverse dielectric matrix are ported to periodic BSE
  • A specific frequency grid type is added (internal freq_grid_type set to 99) for periodic BSE to ensure exactly zero frequency is used throughout the current implementation, which adapts the static approximation.
• Supports LDA and PBE starting point.
• A smaller range of bands can be read from an energy_qp file, according to state_lower_limit and state_upper_limit. One may not need to calculate all the bands in GW before BSE any more.

Periodic TERS initial implementation (Krystof Bresina)

• Generalized the numerical near field framework for periodic slabs with the intention of using finite near and far field perturbations to calculate zz-polarizabilities on periodic metallic slabs
• This can be understood as an alternative and generalization of the previous approach using DFPT (as described in Litman et al., J. Phys. Chem. Lett. 2023, 14, 30) that does not allow for such calculations.

XDM improvements (Kyle Bryenton)

• Experimental implementation of XDM z-damping
• Experimental Implementation of XCDM
  • It supplements the XDM model’s exchange hole with the same- and opposite-spin correlation holes.
  • This model can be used by calling xcdm in the control.in in the same way that you call xdm.
  • XCDM supports auto-damping with both BJ and Z damping parameters
  • Neither XCDM and Z damping support the cubic fits for any %HF for the B86bPBE0 or PBE0 functionals
• Fixed a bug for scf reinitialization: created a new boolean variable, xdm_init, which is used to force XDM to initialize properly the first time it’s called.
• Modified warnings to avoid line breaks in new warnings.f90 infrastructure.
• Switched usages of sqrt to dsqrt to avoid precision loss

RI functionality improvements (Joseph Abbott and Francisco Delesma)

• Restructure real-space grid generation on separable resolution of the identity method (Francisco Delesma)
• RI decomposition of electronic densities constructed from custom occupations (Joseph Abbott):
  1. Construction of an electronic density via user-defined occupations, followed by its RI decomposition using the existing routines. The keyword ri_density_from_occs is used for this purpose, along with the user specification of an input file “eigenstate_occs.in” that contains the custom eigenstate occupations. Crucially, these correspond to k-point-symmetrised set of eigenstates, the eigenvalues for which can be printed with the next keyword:
  2. Output of a file containing eigenstate information. This is controlled by keyword output eigenstate_info. This differs from what is offered by keyword output postscf_eigenvalues: 1) it involves no change in the density of the k-grid, and no re-computation of the eigenvalues, and 2) the printed eigenstates correspond to the k-point symmetrised set of eigenstates (if applicable), and 3) it works for both cluster and periodic systems.
  3. Output cube files evaluated via RI coefficients, using keyword output ri_density. This reads the file “ri_restart_coeffs.out” and uses these to evaluate the corresponding electronic density on a cube grid.
• Introduced a new keyword for the RI-fitting functionality: ri_ovlp_write_triu (Joseph Abbott).
  • When set to true only writes the upper triangle of the RI overlap matrix. This reduces the I/O time and storage requirements of a given RI procedure without losing information (due to overlap symmetry).
• Added new keyword ri_skip_scf (Joseph Abbott).
  • When set to true ensures no SCF cycles are performed prior to RI fitting. This is useful for reading in SCF checkpoint files (i.e. through elsi_restart), performing an SCF initialization step to obtained the density, and then directly fitting to the converged scalar field without running any subsequent SCF cycles.
• Added option to turn off writing of RI overlap matrix (Joseph Abbott).
  • This adds the keyword ri_ovlp_no_write, defaulting to .false., that controls the writing of the RI overlap matrix to file. As the overlap matrix is large and its writing costly, in cases where RI decomposition is being performed on multiple scalar fields with a consistent RI basis definition it is useful to be able to turn writing off.

Output of phonon linewidths and lifetimes due to electron-phonon coupling using the Friction module (Connor Box)

• New keyword: friction_output_vibrations.  Uses finite differences to calculate the Hessian, normal modes, phonon frequencies, linewidths and lifetimes
• Friction with finite difference now supports load_balancing and use_local_index
• Some other tweaks to the friction module including memory usage improvements
• Normal modes calculated through finite difference as part of a friction calculation are now output in JMOL format

Features undergoing testing

Finite Nucleus Model implementation (Thanush Patlolla)

• Added the ability to specify a nuclear charge distribution for each species
• AIMS calculates the potential associated with this nucleus and uses it instead of the standard $-\frac{Z}{r}$ potential throughout the code
• control.in species’ defaults parameter: finite_nucleus <model>
• Several models for finite nucleus charge distribution:
  • Harmonic Oscillator/Modified Harmonic Oscillator (HO / MHO)
  • Two-parameter Fermi (2pF)
  • Three-parameter Fermi (3pF)
  • Three-parameter Gaussian (3pG)

LAPACK support for the local dense format (use_local_index + load_balancing) in relativistic Q4C calculations (Wentao Zhang)

A complete new set of algorithms have been introduced to enable LAPACK support for the local dense format (use_local_index + load_balancing) in relativistic q4c calculations, thereby broadening the range of feasible calculations for the relativistic q4c method in FHI-aims, which save the memory usage for q4c large periodic system calculations.

Before this merged, for the keyword (relativistic q4c), the real-space matrix implementation has support for two matrix formats, which may be classified based on whether the real-space matrix is synchronized across all MPI tasks or not (“global” and “local”, respectively) and whether the matrix is stored in a dense or (nearly) Compressed Sparse Row (CSR) format.

In summary, the relativistic q4c now supports these three formats:

1. Global dense format (PM_none), suitable for non-periodic systems, with LAPACK.
2. Global CSR format (PM_index), suitable for both non-periodic and periodic systems, with LAPACK.
3. Local dense format(use_local_index + load_balancing), suitable for periodic systems, with LAPACK.

Additional changes:

• Restart functionality in RT-TDDFT with changing external field (Jianhang Xu, Chris Shepard, Hannah Bertschi, Mariana Rossi)
• Fixed a bug in the check_use_local_index routine. Any calculation with use_local_index should never start with packed_matrix_format /= PM_index. The check was wrongly skipped due to a misplaced if statement. (Konstantin Lion)
• bands_compare.py script, providing an RMSE value for band structure comparison, is added to utilities folder (Uthpala Herath)
• Added a keyword adjust_scf_param that can change s.c.f parameters modified by the adjust_scf keyword. In this first implementation, a user can modify the default values of charge_mix_param and the width of the occupation_type variable as specified by adjust_scf in the low-gap and large-gap cases. (James A. Green)
• Enable numerical stress calculation in i-pi socket (Shuo Zhao)
• K-grid consistency checks (Konstantin Lion):
  • We calculate n_k_points_xyz(i) * length_of_lattice_vector(i) and compare the ratios between non-vacuum directions. If the ratio is larger than 2 (or smaller than 0.5), we stop unless an override is specified. The check can be skipped with override_kgrid_checks .true.
• ASI improvements (Pavel Stishenko):
  • Added ASE parser hook for dielectric tensor to current version of DFPT code
  • Added ASI_is_scf_converged() and ASI_get_k_points() functions to ASI API
  • Added specific type for ASI_basis_func_descr: ASI_basis_func_descr_aims
  • Fixed initDM callback reset after invocation - now it is done separately for each (kpts,spin) pair
  • Fixed l and m quantum numbers export to ASI
• RPA improvements (Muhammad Tahir)
  • molecular LVL rpa energy for very big systems with less memory
  • lapack version of rSE forces
  • update rSE single point energy for 100+ atoms system
  • update rpa force without Coulomb matrix power
  • update rpa+rSE forces without power of Coulomb matrix
  • force_correction for RPA force calculations
• Kubo-Greenwood improvements (Jingkai Quan).
  • Allow manually setting the number of k-points to be calculated in each loop (set_nkp_parallel).
  • Can output conductivity in all directions & average simultaneously (kg_out_put_all .true.)
  • Various bugfixes (James A. Green)
• Basis sets additions (Jack Morgenstein):
  • Added pcJ-0, pcJ-1, pcJ-2, and pcJ-3 basis sets
  • Completed the pcJ-4 basis sets for H to Ar in FHIaims/species_defaults/defaults_next/nmr/pcJ-n/pcJ-4
  • Completed the NAO-J-n basis sets for H to Ar in FHIaims/species_defaults/defaults_next/nmr/NAO-J-n
• D3 Improvements (James A. Green):
  • Bugfix for sign of D3 forces
  • Automatic parameter recognition is now also possible for LibXC functionals.
  • Define a proper public and private interface in dftd3_interface module.
  • Small refactor to the routines in the dftd3_interface module including better usage of module variables.
  • Added s-dftd3 library citation: Journal of Open Source Software 9 (103), 7169.
• Various Bugfixes (James A. Green):
  • Bugfix for COSMO when specified with initial_charge in geometry.in.
  • Add a stop if either generalized_regular_k_grid or k_points_external are used with many_body_dispersion/many_body_dispersion_nl and no k_grid values are specified for MBD
  • Fix for ScaLAPACK band: ensuring mulliken_decomp is zeroed only in the band_mulliken code branch.
  • Allow dos_kgrid_factors with only projected DOS.
  • Adding a stop if dos_kgrid_factors + an external k-grid (generalized_regular_k_grid or k_points_external) is requested.
  • Add final band structure timing for band_mulliken.
  • Fixed a bug: when an external k-grid was specified, the code was stopping erroneously when output realspace_esp or output cube was requested if the header of k_list.in contained the k-point directions (for reference only) of 0 0 0.
  • Bugfix for output cube first_order_density with DFPT using the old interface (dfpt_centralised .false.)
  • Added a stop if the cube output is requested with dfpt_centralised .true. as this is currently not implemented.
  • Bugfix occupation_type cold in FHI-aims and ELSI, and changed how occupation_type is stored in FHI-aims to match ELSI.
  • A check on Gamma point only output cube calculations has been modified to only trigger when cube_type is specified to be eigenstate or eigenstate_imag, as these are the only quantities to depend on the k-points.
  • Added a stop if the sum of initial charges in geometry.in (specified by initial_charge keyword) is not equal to the total charge on the system, specified by the charge keyword in control.in
  • Added an override keyword override_initial_charge_check that disables this stop, e.g. if the initial_charge values are not obvious to set in geometry.in.
  • Various bugfixes for get_vibrations.py
  • Removal of RelWithDebInfo CMAKE_BUILD_TYPE being made a default.
• Documentation:
  • COSMO Implicit Solvent Model (Yi Yao)
  • Updated FHI-aims documentation to include instruction on building FHI-aims with multiaims support (Uthpala Herath)
  • Improve the appendix about phase convention of real spherical harmonics (Min-Ye Zhang)
  • Re-add the keyword DFPT dielectric to the manual, as it got removed at some point (James A. Green)
  • fix doc typo output_rs_matrices (Zekun Lou)
  • Update documentation for vibration scripts (James A. Green)
  • Fixed documentation to reflect implemented support of empty sites for PBC (Pavel Stishenko)
• WIP: Local hybrid potentials implementation - LH_B86bPBE0_erfs, LH_B86bPBE0_erfcor (Nicholas Boyer):
  • Exchange energy density is implemented as part of XDM for local hybrid xc potentials