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

Changes since release version 250822

ScaLAPACK Polarization (Elia Stocco)

• The output polarization keyword has been extended to be able to be used in combination with KS_method parallel. (!2103)
• This permits the calculation of the polarization for periodic calculations with up to thousands of atoms.

---

LibRPA Interface (Minye Zhang)

• The LibRPA library implementing low-scaling algorithms for many-body perturbation theory is included in the external_libraries folder. It is capable of performing periodic RPA and GW calculations with either a file based or API based interface (!2108 !2031 !2111)
• librpa_api module provides subroutines which compute necessary input dataset and wraps LibRPA data parsing APIs.
• interface_librpa_api subroutine to perform MBPT calculation with LibRPA as backend. Currently, RPA correlation energy and quasi-particle energies on regular k-grid is supported. Please note that band structure output is currently disabled for the API interface, however it is possible in the file based interface.
• FHI-aims manual section added about running RPA/GW with LibRPA as backend with either the file based or API based interface, as well as including installation instructions. The LibRPA manual is also available online.
• Additional output and matrix export for LibRPA (!2036 !1990 !1975 !1959)
• Additional control tags for file-based interface to LibRPA (Subtags for suppressing output with output librpa: noC, noV, novxc, novexx, nodielfunc.) (!2011)

---

Fully analytical G0W0 Self-Energy For Molecules (Jannis Kockläuner)

• Implements the fully analytical version of the G0W0 self-energy for closed-shell molecules at N^6 cost in the gw_exact module. (!1961)
• Method can be activated with the fully_analytical_gw keyword.
• Regression test for its usage CO.gw_exact added.

---

Constrained NEO Implementation (Songrui Liu)

• Implementation of the cNEO-DFT method for periodic DFT (!1994 !1997)
• The nuclear–electronic orbital (NEO) method incorporates nuclear quantum effects (NQE) in DFT calculations beyond the Born–Oppenheimer approximation. By incorporating additional constraints to the position operator on quantum particles like protons, the cNEO method facilitates NEO calculations with controlled nuclear localization.
• Description of the method available in Songrui Liu, Jianhang Xu, Yosuke Kanai, J. Chem. Phys., 163, (2025), 084110.

---

Global RI (Moritz Leucke)

• Experimental implementation of Global RI with local metric and new integration for 3-center integrals (!1968)
• global RI: all auxiliary basis functions are used to fit the codensities
• local metric: the coulomb potential can be attenuated with various functions to get linear scaling 3-center integrals
• new integration scheme of 3-center 2-electron integrals: The global RI integration is carried out with domain decomposition just like the 1-electron integrals during the SCF cycle
• GPU acceleration of the 3-center integrals in the new algorithm

---

Inclusion of GreenX Library (Francisco Delesma and James Green)

• The GreenX library is now built by default with FHI-aims, and additional CMake infrastructure has been added to facilitate this. (!2027 !2041)
• Only the time-frequency grids component is switched on by default, other components (e.g. AC_COMPONENT, LBASIS_COMPONENT, PAW_COMPONENT) can be switched on by setting these CMake variables to ON in the initial CMake cache. The build of GreenX can be switched off with USE_GREENX OFF.
• Added regression tests greenx_H2O_RPA_minimax and greenx_Si_RPA_minimax for molecular and periodic testcases, respectively.
• Bumped the cmake_minimum_version for the FHI-aims build to 3.15 from 3.13.

---

Frozen Density Embedding (Danjo De Chavez)

• Initial implementation of FDET with new keywords qm_embedding_type, xc_nakp, and xc_naxcp (!1969 !2105)
• FDET embedding potential then extended to libxc GGA functional (!2055)
• RI decomposed embedding densities and potential (!2074). This allows the decomposition of the embedding potential and densities. These objects are then expressed as linear combination of auxiliary basis which allows faster allocation of embedding potential and densities.

---

RT-TDDFT Improvements (Hannah Bertschi)

• Added D3 and TS dispersion corrections for Ehrenfest dynamics (!1952 !1983)
• Added Hirshfeld analysis to non-periodic RT-TDDFT (!1970)

---

Auto-stop if SCF Not Converging (Konstantin Lion)

• If the SCF is detected as likely to not converge, FHI-aims will stop to avoid wasting computational time. (!2016)
• Utilises SCF analysis infrastructure introduced in version 250822.
• Decision logic on whether to stop the SCF starting from 100 iterations (check can be overridden by sc_iter_limit):
  • Decision point automatically set to 100 or never (free single atom, SCF iterations below 1 s)
  • If the predicted steps to convergence are more than 25, we stop the SCF.
  • If the predicted steps are below/equal 25: if not stalled, we continue to try and converge, if stalled, we stop the SCF
  • Restart logic again after 125, 150, and 175 steps
  • Stop at 175 if not converged
• Deafults can be modified with the keywords: scf_analysis_start, scf_analysis_decision_point, scf_max_allowed_steps_at_decision, scf_max_decisions_allowed

---

ELSI Updated to v2.12 (James Green and Andrei Sobolev)

• Upgraded ELSI to v2.12 to benefit from upstream updates (!2015)
• Included into FHI-aims the ELSI CMake find_package(ExternalELPA) tool, which automatically sets useful CMake variables for an externally compiled version of ELPA, such as ELPA_API_VERSION for version specific features, and GPU kernels to allow better GPU support.
• Introduced the elsi_build_dm_method keyword which uses the new infrastructure in ELSI to select a particular method to build the density matrix. By default, ELSI will try to use the optimal version, depending on runtime conditions (e.g. ELPA version, CPU vs GPU)
• Bugfixes for the elsi_restart infrastructure and regression tests to check functionality (!1998)
• Upgrade of ELSI CMake minimal version to 3.5, to be compatible with CMake > 4.0 (!1964)

---

Initial Density Guess Improvements (James Green)

• A new algorithm to compute the occupation numbers of the free atom minimal basis used to construct the initial density matrix has been implemented. (!2077)
• It can be used with the keyword init_atom_occs_version 2 (the old version, still the default for now, is init_atom_occs_version 1).
• Initial testing of this algorithm has proved promising for reducing the number of SCF steps, and leading to a better SCF solution, by providing a better initial density guess.
• The algorithm is only relevant for spin collinear calculations, and spin none calculations where initial_charge values are specified in geometry.in. It will also likely offer most benefit for calculations involving transition metals, lanthanides or actinides.

---

Implementation of ionic_solver_xc keyword (James Green)

• The ionic_solver_xc keyword is analogous to atomic_solver_xc except that it applies to ionic basis functions generated in get_free_ion.f90 from a call to sratom, rather than the atomic/minimal basis functions that are generated in get_free_atoms.f90 (!2048)
• Allows consistent XC choice to be used for basis functions in a calculation, if an alteration from the default is needed, such as in embASI embedding, where QM/QM calculations require consistent basis functions throughout the separate calculations.
• Previously, setting just atomic_solver_xc meant that only the atomic/minimal basis functions would be generated from sratom with that XC, and this could be inconsistent with the XC used to generate the ionic basis functions.

---

Explicit GGA XC Potential Calculation (Andrei Sobolev)

• Enabled explicit calculation of GGA exchange correlation potential in FHI-aims, as opposed to calculating the matrix element of it using the integration by parts, as shown in Eq. 30 of Blum et al, Computer Physics Communications 180 (2009) 2175–2196. (!2065 !2087)
• This improves the lowest eigenvalues for the heavy atoms (Pb, Bi) calculated on light grids (thus, also slightly improves analytical stress on the sparse grids; see issue #444 for details).
• Currently, the explicit XC potential only works for libxc functionals.
• The explicit XC is only calculated in a post-SCF step to not increase the runtime (Hessian of the density required). Tests showed that using the explicit XC potential during the SCF cycle does not change the final state and energy.

---

Friction and Superconductivity Improvements (Connor Box)

• Temperature scan for friction and superconductivity. Allow users to set multiple friction_temperature values for friction and superconductivity (!2018)
• Additional convenience keywords and bugfixes (!2009)
  • Acoustic sum rule and symmetrise options for friction/superconductivity Hessian (friction_apply_asr, friction_symmetrise_hessian)
  • Large tensor printout can be turned off (friction_print_tensors)
  • alpha2f_broadening_width now correctly in eV and default is now 0.01 eV correctly
  • The translation shift fix for the finite difference bug where an atom crosses into a periodic image is now more elaborate. It checks if the shift actually worked and has support for a manual shift (friction_translation_vec).
• Bugfix for assignment of occupied states (!1951) and other friction bugfixes (!2082 !2035)

---

HTML Version of FHI-aims Manual (Konstantin Lion)

• An HTML version of the FHI-aims manual can now be generated from the TeX files, via LaTXML. (!2033 !2067)
• A single-page or multi-page version of the manual can be generated, with instructions on its use available in the FHI-aims repo Wiki page
• The multi-page version is now also available on the FHI-aims website: https://fhi-aims.org/uploads/manual/index.html

---

Improvements to exx_lr_approximation (Sebastian Kokott)

• Use the cut-Coulomb potential for PBE0’ functional by default, to better align with PBE0 and permit better calculations for sparse k-grids. (BREAKING CHANGE !2084)
• To restore the old behaviour, set cut_coulomb_short_range .false.
• Bugfix for exx_lr_approximation for range-separated hybrids (BREAKING CHANGE). The short-range component of the screened Coulomb potential for k to 0 was not treated correctly (k^2/k2 term). This required an evaluation by a Tayler series. !2086

---

Additional New Features

• Implementation of explicit symmetrization of coordinates and lattice vectors for use_symmetric_forces (Sebastian Kokott !2092)
• Bugfixes for constrained relaxation (Konstantin Lion !2098 !2096 !2091)
• Added keyword cumulant_relative_output to output the GW+C spectral function relative to the QP peak (Jannis Kockläuner !2034)
• Added autodamping support for revPBE50 for XDM(BJ) and XDM(Z) dispersion corrections. Called via xc revpbe0 and hybrid_xc_coeff 0.50, and XDM is called as normal. (Kyle Bryenton !2032)
• Enable CMake variables to be accessed within the FHI-aims code via the template routine cmake_build_info.f90.in (Minye Zhang !2029)
• Enabled vdw_correction_hirshfeld for relativistic q4c calculations (Wentao Zhang !1995)
• Fast exact exchange energy for RPA (Moritz Leucke !1984)
• Enable export and import of BLACS distributed arrays with use_local_index / load_balancing in ASI (Gabriel Bramley !1980)
• Implemented fixed_volume option to relax_unit_cell for constant-volume relaxations (Alberto Otero de la Roza !1966 !1979 !1981)
• Add GPU support and real-space grid generation to separable RI using the COSMA library (Francisco Delesma !1963)
• Implementation of PBEsol for the sratom atomic solver (Andrei Sobolev !1958)

---

Additional Feature Updates

• Updates to RPA forces (Muhammad Tahir !2110 !2109 !2106 !2104 !2100 !2089 !2088 !2085 !2081 !2079 !2076 !2073 !2072 !2071 !2069 !2068 !2064 !2063 !2062 !2061 !2059 !2056 !2054 !2053 !2049 !2047 !2043 !2028 !2021 !2014 !2013 !2012 !2006 !2004 !2003 !2002 !2001 !2000 !1996 !1993 !1992 !1989 !1986 !1985 !1977)
• Updates to DFPT (Muhammad Tahir !2107 !2102 !2101 !2099 !2097 !2050 !2023 !2017 !2010 !2008 !2007)
• Refactoring and simplification of the routine evaluate_H_psi_p2 (Sebastian Kokott !2095)
• Increased band structure and band mulliken calculation precision from 5 to 7 decimal places (Xioquan Liu !2083 !2005)
• Change of the output level for XDM moments, volumes and polarizabilities (Kyle Bryenton !2080)
• Removal of automatic MPI alltoall setting when n_tasks>1024 (James Green !2046) Testing by Ville Havu and Oscar Van Vuren showed no performance benefit for this (indeed, a possible performance decrease). The setting can still be switched on manually with use_alltoall .true.
• Better memory handling for projected density of states calculations (James Green !2025)
• Update to RT fields (Chris Shepard !1999)
• Documentation and regression test for calculate_plasma_frequency (Mariia Pogodaeva and James Green !1967)
• Increase flexibility of sc_init_iter flag by allowing multiple restarts within a single run. Seen to be advantageous on test systems, especially involving lanthanides and “cousins”. (Mariana Rossi !1956)
• Update of lightdense and lightdenser in species_defaults/defaults_next with the basis functions being based on defaults_2020/light, rather than defaults_2010/light. (Sebastian Kokott !2044)
• Add regression test for output band + dos_kgrid_factors + include_spin_orbit + use_local_index (GaAs_SOC_Gamma_band_local_index) and + load_balancing (GaAs_SOC_Gamma_band_load_balancing) (James Green !2105)

---

Additional Bugfixes

• Bug fixed in deltascf_projector for periodic calculations (Matthias Kahk !1897)
• Environ bugfix (Jakob Filser !1955)
• Bugfix when output basis is requested while for_aux basis exists (Minye Zhang !1954)
• Bugfix for parallel calculation of exchange energy density (Nicholas Boyer !1953)
• Correction to RI renormalisation (Alan Lewis !1950)
• Correct rotational free energy for diatomics in the hessian diagonalization routine for the post-processing calculation of numerical vibrations (Gabriel Graf and Tommy Lin !1971)
• Bugfix for turbomole_state_to_aims_format.py utility (Qing-Long Liu !1978)
• Bugfix finite nuclear models for Q4C (Thanush Patlolla !1974)
• Whitespace bugfix for ifx 25.3 compiler (Andrei Sobolev !2024)
• Stop for transport calculations with use_local_index as it is not supported (Konstantin Lion !2026)
• Bugfix for PBEsol for spin-polarized calculations (Andrei Sobolev !2042)
• Bugfix for PZLATRA ScaLAPACK complex trace with intel toolchain (James Green !2027)
• Bugfix for task_node_map in MPI SHM routine (James Green !2027)
• Bugfixes for improperly allocated arrays (James Green !2027)
• Bugfix for NEO RT propagation (James Green !2027)
• Bugfix for improperly set variable in RPA (James Green !2027)
• Bugfix for ASI serial import of Hamiltonians (Gabriel Bramley !2060)
• Bugfix for NEO DIIS propagation (James Green !2058)
• Improvements and bugfixes for transport calculations (Xingtao He !2075)
• Bugfix for TREXIO with GNU compilers (James Green !2078)
• Stop for rlsy_symmetry with analytical stress computations for meta GGAs, as it is currently not working (James Green !2112)