VASPはオーストリア・ウィーン大学で開発された、擬ポテンシャルと平面波基底を用いた非経験的(ab initio)量子分子動力学(MD)計算プログラムです。結晶や無機固体の周期的境界条件を用いた計算を行います。VASPに実装されている方法は有限温度局所密度近似(LDA, 変分量として自由エネルギーを用いる)に基づき、効果的な行列対角化法とPulay混合を使用してMDステップごとに電子基底状態を精密に評価します。バンド計算、安定構造や遷移状態構造の最適化、拡散パスの検討、反応解析など様々な問題に適用されています。
弊社では、VASPの各バージョンのベンチマークライセンスを所有しております。ご利用ください。
https://www.vasp.at/
We are pleased to announce a new version of VASP with many exciting new features. Please check out their documentation to learn more about what new quantities and methods are available to you now. If you have an active VASP license, you can download the new version in the VASP Portal.
FEATURE
Added output of the imaginary part of the dielectric function from XAS calculations to vaspout.h5 (HDF5), and of the dielectric function with respect to the Fermi level to OUTCAR.
For ELPH_POT_GENERATE = True:
The FFT grid for the primitive cell is now determined automatically if not set via ELPH_POT_FFT_MESH.
VASP generates a CONTCAR_ELPH file that can be used for the subsequent electron-phonon calculation.
The primitive-cell information is also written to vaspout.h5.
IMPROVEMENT
The line search algorithm of the conjugate gradient optimizer (ALGO = A) has been extensively improved:
previously, when performing the line search, it moved incremental along the line. The new version, always starts from the origin for each step during the line search this vastly increases the consistency of the energies.
The line search is now done in a more way and all considered steps are stored into slots to avoid unnecessary redundancy. For acceptance of the final minimum in the line search, it is required that the neighbouring slots have been considered (that first principles energies are known).
The minimum is determined by fitting up to a 4th order polynomial to the data closest to the minimum, and determining the minimum of the polynomial.
If more than 5 data are available, a spline fit is performed through all points (this turns out to be more robust than a 5th order fit).
Last not least: the line search is usually performed using energy evaluations only, as this is faster than gradient calculations. However, if the new gradient is not sufficiently orthogonal to the search direction, one more correction is performed using all yet available data points (again using 4th order polynomial or spline)
The new and improved line search algorithm can be switched on by setting ISEARCH = 1. The legacy line search (ISEARCH = 0) is still the default.
Improvements for BSE:
The dielectric function is written as a scalar (not tensor) for finite q in IBSE = 1, 2, 3.
Hermiticity of the BSE matrix is now enforced for IBSE = 3 to improve stability.
Improved the performance of the Ewald summation used in the truncated Coulomb kernel method for 2D materials by restricting the number of g-vectors that are used in the reciprocal space summation. This truncation of g-vectors is in keeping with the 3D summation, where every value less than
10
−
10
{\displaystyle 10^{−10}} is removed.
BUGFIX
Fixed the handling of “very long full paths of files” with the Intel compiler (these were truncated at 255 characters). Now we allow for 1023 characters and hope this will be enough …
For ISYM = 3, symmetry operation involving spinflips were incorrect under some special fringe circumstances. This bug may have affected hybrid functional calculations for anti-ferromagnetic systems where the spin-up and spin-dn density of states differ from each other at some k-points.
Fixed a bug that prevented building without scaLAPACK support.
Fixed inconsistent ML_FF input/output for non-scaLAPACK build.
The code crashed for LKPROJ = True. This issue is fixed.
SERTA is now correctly listed as “self-energy relaxation-time approximation” in OUTCAR.
The wannier90.UNK files were not correctly written for non-collinear magnetic calculations. This issue is fixed.
If both ISPIN = 2 and LNONCOLLINEAR = True (or LSORBIT = True) were set in the INCAR, the code would stop with an error message (something about the number of entries for MAGMOM being wrong). This issue is fixed: LNONCOLLINEAR = True will now take precedence over ISPIN = 2 (and the latter will default back to ISPIN = 1 internally).
Fixed a problem with the final diagonalization in the occupied subspace before the computation of the forces: this bug sometimes caused errors in the forces when the wave functions were not tightly converged, for ALGO = A or D.
BSE crashed on GPUs with -DCUSOLVERMP and -DCUBLASMP when OMEGAMAX was set. This issue is fixed.
The computation of the dielectric function was not correctly implemented for finite-q in IBSE = 1,3.
BSE crashed for IBSE = 3 with the gamma-only version. This has been fixed.
BSEPREC was overwritten for IBSE = 3. This has been fixed.
The feature to compute the transport function using ELPH_TRANSPORT_NEDOS_PLOT was broken. This has been fixed.
Electron-phonon code crashed for ELPH_MODE = renorm. This issue is fixed.
VASP no longer crashes when running ELPH_POT_GENERATE = True without specifying the FFT grid via ELPH_POT_FFT_MESH.
The LATTICE_CONSTRAINTS were not applied correctly for ISIF = 4, 5. This issue has been fixed.
For ISIF = 4, 5, 6, the header of the XDATCAR file was written twice. This has been fixed.
The truncated Coulomb kernel method for 2D systems only worked correctly if the cell was twice as large as the thickness of the slab. This requirement was imposed by the 2D Ewald summation, where minimum image convention was used in the aperiodic dimension. This requirement has now been removed by setting the distance between two atoms to not follow if the minimum image convention if it is 2D boundary conditions. This issue is fixed.
There was a (small) memory leak in the use of HDF5. This issue has been fixed.
In case the HF exchange was activated via the LTHOMAS tag, VASP wrongly kept IMIX = 4. Now IMIX = 1 is used.
The OpenACC version crashed when compiled with python plugin support (-DPLUGINS,). This issue is fixed and the python plugins are now fully supported in the OpenACC version as well.
For IBRION = 12 or PLUGINS/STRUCTURE = True (i.e. using the python structure plugin) the updated structures were not written to XDATCAR. This issue has been fixed.
The combination of PLUGINS/STRUCTURE = True and ML_MODE = run caused the code to crash. This issue is fixed.
ML_MODE was broken when running on multiple nodes and using shared memory (-Duse_shmem). This issue is fixed.
Fixed broken ML_FF version check in VASPml; was unable to read ML_FFs from VASP 6.5.0 (earlier versions work).
ML_IERR was deprecated in favor of new tag name ML_ESTBLOCK. This fixes “grepping” for ERR in ML_LOGFILE (old tag still works).
The output of the total energy to the OUTCAR was wrong when running ML_MODE = train with ALGO = A. This issue issue has been fixed.
Fixed an incorrectly sized spline grid that sometimes led to wrong force predictions by VASPml.
NVIDIA_Life and Material Sciences Application Performance Guide
平日9:30~17:30 (土曜日、日曜日、祝祭日、年末年始、夏期休暇は、休日とさせていただきます。)