GLOBAL 

Lone keyword: T

Usage: ALLTOALL_SGL YES

All-to-all communication (FFT) should use single precision [ Edit on GitHub ]

how to distribute the processors on the 2d grid needed by BLACS (and thus SCALAPACK) [ Edit on GitHub ]

Lone keyword: T

Usage: BLACS_REPEATABLE

Use a topology for BLACS collectives that is guaranteed to be repeatable on homogeneous architectures [ Edit on GitHub ]

At the end of the run write a callgraph to file, which contains detailed timing informations. This callgraph can be viewed e.g. with the open-source program kcachegrind. [ Edit on GitHub ]

Usage: CALLGRAPH_FILE_NAME {filename}

Name of the callgraph file, which is written at the end of the run. If not specified the project name will be used as filename. [ Edit on GitHub ]

Usage: DLAF_NEIGVEC_MIN 512

Minimum number of eigenvectors for the use of the eigensolver from the DLA-Future library. The eigensolver from the ScaLAPACK library is used as fallback for all smaller cases [ Edit on GitHub ]

Lone keyword: T

Usage: ECHO_ALL_HOSTS NO

Echo a list of hostname and pid for all MPI processes. [ Edit on GitHub ]

Lone keyword: T

Usage: ECHO_INPUT NO

If the input should be echoed to the output with all the defaults made explicit [ Edit on GitHub ]

Specifies the kernel to be used when ELPA is in use [ Edit on GitHub ]

Usage: ELPA_NEIGVEC_MIN 32

Minimum number of eigenvectors for the use of the eigensolver from the ELPA library. The eigensolver from the ScaLAPACK library is used as fallback for all smaller cases [ Edit on GitHub ]

Lone keyword: T

Usage: ELPA_QR

For ELPA, enable a blocked QR step when reducing the input matrix to banded form in preparation for the actual diagonalization step. See implementation paper for more details. Requires ELPA version 201505 or newer, automatically deactivated otherwise. If true, QR is activated only when the the size of the diagonalized matrix is suitable. Print key PRINT_ELPA is useful in determining which matrices are suitable for QR. Might accelerate the diagonalization of suitable matrices. [ Edit on GitHub ]

Lone keyword: T

Usage: ELPA_QR

For ELPA, disable block size limitations when used together with ELPA_QR. Keyword relevant only with ELPA versions 201605 or newer. Use keyword with caution, as it might result in wrong eigenvalues with some matrix orders/block sizes when the number of MPI processes is varied. If the print key PRINT_ELPA is active the validity of the eigenvalues is checked against values calculated without ELPA QR. [ Edit on GitHub ]

Lone keyword: T

Usage: ENABLE_MPI_IO FALSE

Enable MPI parallelization for all supported I/O routines Currently, only cube file writer/reader routines use MPI I/O. Disabling this flag might speed up calculations dominated by I/O. [ Edit on GitHub ]

Usage: EPS_CHECK_DIAG 1.0E-14

Check that the orthonormality of the eigenvectors after a diagonalization fulfills the specified numerical accuracy. A negative threshold value disables the check. [ Edit on GitHub ]

Lone keyword: T

Usage: EXTENDED_FFT_LENGTHS

Use fft library specific values for the allows number of points in FFTs. The default is to use the internal FFT lengths. For external fft libraries this may create an error at the external library level, because the length provided by cp2k is not supported by the external library. In this case switch on this keyword to obtain, with certain fft libraries, lengths matching the external fft library lengths, or larger allowed grids, or grids that more precisely match a given cutoff. IMPORTANT NOTE: in this case, the actual grids used in CP2K depends on the FFT library. A change of FFT library must therefore be considered equivalent to a change of basis, which implies a change of total energy. [ Edit on GitHub ]

References: Frigo2005

FFTW can have improved performance if it is allowed to plan with explicit measurements which strategy is best for a given FFT. While a plan based on measurements is generally faster, differences in machine load will lead to different plans for the same input file, and thus numerics for the FFTs will be slightly different from run to run. PATIENT planning is recommended for long ab initio MD runs. [ Edit on GitHub ]

Usage: FFTW_WISDOM_FILE_NAME wisdom.dat

The name of the file that contains wisdom (pre-planned FFTs) for use with FFTW3. Using wisdom can significantly speed up the FFTs (see the FFTW homepage for details). Note that wisdom is not transferable between different computer (architectures). Wisdom can be generated using the fftw-wisdom tool that is part of the fftw installation. cp2k/tools/cp2k-wisdom is a script that contains some additional info, and can help to generate a useful default for /etc/fftw/wisdom or particular values for a given simulation. [ Edit on GitHub ]

Usage: FFT_POOL_SCRATCH_LIMIT {INTEGER}

Limits the memory usage of the FFT scratch pool, potentially reducing efficiency a bit [ Edit on GitHub ]

Lone keyword: T

Usage: FLUSH_SHOULD_FLUSH

Flush output regularly, enabling this option might degrade performance significantly on certain machines. [ Edit on GitHub ]

Usage: OUTPUT_FILE_NAME {filename}

Name of the output file. Relevant only if automatically started (through farming for example). If empty uses the project name as basis for it. [ Edit on GitHub ]

Usage: PREFERRED_DGEMM_LIBRARY SPLA

Valid values:

Specifies the DGEMM library to be used. If not available, the BLAS routine is used. This keyword affects some DGEMM calls in the WFC code and turns on their acceleration with SpLA. This keyword affects only local DGEMM calls, not the calls to PDGEMM (see keyword FM%TYPE_OF_MATRIX_MULTIPLICATION). [ Edit on GitHub ]

Usage: PREFERRED_DIAG_LIBRARY ELPA

Valid values:

Specifies the diagonalization library to be used. If not available, the ScaLAPACK library is used [ Edit on GitHub ]

Usage: PREFERRED_FFT_LIBRARY FFTW3

Valid values:

References: Frigo2005

Specifies the FFT library which should be preferred. If it is not available, use FFTW3 if this is linked in, if FFTW3 is not available use FFTSG. Improved performance with FFTW3 can be obtained specifying a proper value for FFTW_PLAN_TYPE. Contrary to earlier CP2K versions, all libraries will result in the same grids, i.e. the subset of grids which all FFT libraries can transform. See EXTENDED_FFT_LENGTHS if larger FFTs or grids that more precisely match a given cutoff are needed, or older results need to be reproduced. FFTW3 is often (close to) optimal, and well tested with CP2K. [ Edit on GitHub ]

Mentions: ⭐ How to Converge the CUTOFF and REL_CUTOFF

How much output is written out. [ Edit on GitHub ]

Which program should be run [ Edit on GitHub ]

Usage: PROJECT_NAME {STRING}

Mentions: ⭐ Molecular Dynamics

Name of the project (used to build the name of the trajectory, and other files generated by the program) [ Edit on GitHub ]

References: Ceriotti2014 , Schonherr2014

Mentions: ⭐ How to Converge the CUTOFF and REL_CUTOFF , ⭐ Electronic band structure from GW , ⭐ GW + Bethe-Salpeter equation , ⭐ Time-Dependent DFT , ⭐ X-Ray Absorption from TDDFT , ⭐ Real-Time Propagation and Ehrenfest MD , ⭐ Monte Carlo

Type of run that you want to perform Geometry optimization, md, montecarlo,… [ Edit on GitHub ]

Lone keyword: T

Usage: SAVE_MEM

Some sections of the input structure are deallocated when not needed, and reallocated only when used. This reduces the required maximum memory [ Edit on GitHub ]

Usage: SEED {INTEGER} .. {INTEGER}

Initial seed for the global (pseudo)random number generator to create a stream of normally Gaussian distributed random numbers. Exactly 1 or 6 positive integer values are expected. A single value is replicated to fill up the full seed array with 6 numbers. [ Edit on GitHub ]

Lone keyword: T

Usage: TRACE

If a debug trace of the execution of the program should be written [ Edit on GitHub ]

Lone keyword: T

Usage: TRACE_MASTER

For parallel TRACEd runs: only the master node writes output. [ Edit on GitHub ]

Usage: TRACE_MAX 100

Limit the total number a given subroutine is printed in the trace. Accounting is not influenced. [ Edit on GitHub ]

Usage: TRACE_ROUTINES {routine_name1} {routine_name2} …

A list of routines to trace. If left empty all routines are traced. Accounting is not influenced. [ Edit on GitHub ]

Aliases: WALLTI

Usage: WALLTIME {real} or {HH:MM:SS}

Maximum execution time for this run. Time in seconds or in HH:MM:SS. [ Edit on GitHub ]