CoChat

Cp2k: An electronic structure and molecular dynamics software package-quickstep: Efficient and accurate electronic structure calculations

Open access

The DOI matches its Crossref record; 444 of the 445 references it cites were found in a registry or an index.

Abstract

CP2K is an open source electronic structure and molecular dynamics software package to perform atomistic simulations of solid-state, liquid, molecular, and biological systems. It is especially aimed at massively parallel and linear-scaling electronic structure methods and state-of-the-art ab initio molecular dynamics simulations. Excellent performance for electronic structure calculations is achieved using novel algorithms implemented for modern high-performance computing systems. This review revisits the main capabilities of CP2K to perform efficient and accurate electronic structure simulations. The emphasis is put on density functional theory and multiple post–Hartree–Fock methods using the Gaussian and plane wave approach and its augmented all-electron extension.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1063/5.0007045

Cites

444 of 444 references verified0 citation errors · 0 not checkable

Statistics

Figures

FIG. 1 Comparison of the timings for the calculation of the RPA correlation energy using the quartic- and the cubic-scaling implementation on a CRAY XC50 machine with 12 cores per node (flat MPI). The system sizes are n × n × n supercells (with n = 1, 2, 3) of a unit cell with 32 water molecules and a density of 1 g/cm 3 . The intersection point where the cubic scaling methods become favorable is at ∼ 90 water molecules. The largest system tested with the cubic scaling RPA contains 864 water molecules (6912 electrons, 49 248 primary basis functions, and 117 504 RI basis functions) and was calculated on 256 compute nodes (3072 cores). The largest tensor of size 117 504 × 49 248 × 49 248 has an occupancy below 0.2%.
FIG. 1 Comparison of the timings for the calculation of the RPA correlation energy using the quartic- and the cubic-scaling implementation on a CRAY XC50 machine with 12 cores per node (flat MPI). The system sizes are n × n × n supercells (with n = 1, 2, 3) of a unit cell with 32 water molecules and a density of 1 g/cm 3 . The intersection point where the cubic scaling methods become favorable is at ∼ 90 water molecules. The largest system tested with the cubic scaling RPA contains 864 water molecules (6912 electrons, 49 248 primary basis functions, and 117 504 RI basis functions) and was calculated on 256 compute nodes (3072 cores). The largest tensor of size 117 504 × 49 248 × 49 248 has an occupancy below 0.2%.CC-BY · From the PDF
FIG. 2 Properties predicted by the GW method: Ionization potentials and electron affinities as measured by (a) photoemission and (b) inverse photoemission spectroscopy, respectively. (c) Level alignment of the HOMO and LUMO upon adsorption of a molecule at metallic surfaces, which are accounted for in CP2K by an IC model avoiding the explicit inclusion of the metal in the GW calculation.
FIG. 2 Properties predicted by the GW method: Ionization potentials and electron affinities as measured by (a) photoemission and (b) inverse photoemission spectroscopy, respectively. (c) Level alignment of the HOMO and LUMO upon adsorption of a molecule at metallic surfaces, which are accounted for in CP2K by an IC model avoiding the explicit inclusion of the metal in the GW calculation.CC-BY · From the PDF
Image 3
CC-BY · From the PDF
FIG. 4 Illustration of the DFT self-interaction error using the PBE XC functional for the reaction H + 2 → H + + H. However, employing the fragment constraint states |H + H ⟩ and |HH + ⟩ as the basis, the correct profile can be recovered with CDFT-CI.
FIG. 4 Illustration of the DFT self-interaction error using the PBE XC functional for the reaction H + 2 → H + + H. However, employing the fragment constraint states |H + H ⟩ and |HH + ⟩ as the basis, the correct profile can be recovered with CDFT-CI.CC-BY · From the PDF
FIG. 5 Using a fragment-based CDFT constraint, the system is first divided into two fragments with atomic positions fixed in the same configuration as in the full system. The electron and spin densities of the fragment systems are then saved to cube files and subsequently used as input files for the CDFT calculation, where the constraint target value is calculated from the superimposed fragment densities.
FIG. 5 Using a fragment-based CDFT constraint, the system is first divided into two fragments with atomic positions fixed in the same configuration as in the full system. The electron and spin densities of the fragment systems are then saved to cube files and subsequently used as input files for the CDFT calculation, where the constraint target value is calculated from the superimposed fragment densities.CC-BY · From the PDF
Image 6
CC-BY · From the PDF
FIG. 7 Comparison of the wall time required for the purification using the submatrix method (red circles, with a direct eigendecomposition for the density matrix of the submatrices) and the second order Newton-Schulz sign iteration (blue boxes) for different relative errors in the total energy after one SCF iteration. The corresponding reference energy has been computed by setting ϵfi lter = 10 -16 . All calculations have been performed on a system consisting of 6192 H2O molecules, using KSDFT together with a SZV basis set, utilizing two compute nodes (80 CPU-cores) of the 'Noctua' system at PC 2 .
FIG. 7 Comparison of the wall time required for the purification using the submatrix method (red circles, with a direct eigendecomposition for the density matrix of the submatrices) and the second order Newton-Schulz sign iteration (blue boxes) for different relative errors in the total energy after one SCF iteration. The corresponding reference energy has been computed by setting ϵfi lter = 10 -16 . All calculations have been performed on a system consisting of 6192 H2O molecules, using KSDFT together with a SZV basis set, utilizing two compute nodes (80 CPU-cores) of the 'Noctua' system at PC 2 .CC-BY · From the PDF
FIG. 8 Orthogonal (bottom) and non-orthogonal (top) LMOs on the covalent bond of the adjacent carbon atoms in a carborane molecule C2B10H12 (isosurface value is 0.04 a.u.).
FIG. 8 Orthogonal (bottom) and non-orthogonal (top) LMOs on the covalent bond of the adjacent carbon atoms in a carborane molecule C2B10H12 (isosurface value is 0.04 a.u.).CC-BY · From the PDF
FIG. 9 Accuracy and efficiency of O ( N ) DFT for liquid water based on the twostage CLMO optimization. Calculations were performed using the BLYP functional and a TZV2P basis set for 100 snapshots representing liquid water at constant temperature (300 K) and density (0.9966 g/cm 3 ). (a) Dependence of the average number of neighbors on the localization radius, expressed in units of the elements' van der Waals radii (vdWR). (b) Energy error per molecule relative to the energies of fully delocalized orbitals. For simulations, in which the coordination number of molecules does not change drastically, the mean error represents a constant shift of the potential energy surface and does not affect the quality of simulations. In such cases, the standard deviation (error bars) is better suited to measure the accuracy of the CLMO methods. (c) Wall-time required for the variational minimization of the energy on 256 compute cores. The localization radius is set to 1.6 vdWR for the CLMOs methods. The CLMO methods are compared to the cubically scaling optimization of delocalized orbitals (OT SCF), 199,212 as well as to the O ( N ) optimization of density matrix. 115 Perfect linear- and cubic-scaling lines are shown in gray and cyan, respectively. See Ref. 249 for details.
FIG. 9 Accuracy and efficiency of O ( N ) DFT for liquid water based on the twostage CLMO optimization. Calculations were performed using the BLYP functional and a TZV2P basis set for 100 snapshots representing liquid water at constant temperature (300 K) and density (0.9966 g/cm 3 ). (a) Dependence of the average number of neighbors on the localization radius, expressed in units of the elements' van der Waals radii (vdWR). (b) Energy error per molecule relative to the energies of fully delocalized orbitals. For simulations, in which the coordination number of molecules does not change drastically, the mean error represents a constant shift of the potential energy surface and does not affect the quality of simulations. In such cases, the standard deviation (error bars) is better suited to measure the accuracy of the CLMO methods. (c) Wall-time required for the variational minimization of the energy on 256 compute cores. The localization radius is set to 1.6 vdWR for the CLMOs methods. The CLMO methods are compared to the cubically scaling optimization of delocalized orbitals (OT SCF), 199,212 as well as to the O ( N ) optimization of density matrix. 115 Perfect linear- and cubic-scaling lines are shown in gray and cyan, respectively. See Ref. 249 for details.CC-BY · From the PDF
FIG. 10 Accuracy and efficiency of O ( N ) AIMD based on the two-stage CLMO SCF for liquid water. All simulations were performed using the PBE XC functional and a TZV2P basis set at constant temperature (300 K) and density (0.9966 g/cm 3 ). For the CLMO methods, Rc = 1.6 vdWR. (a) Radial distribution function, (b) kinetic energy distribution (the gray curve shows the theoretical Maxwell-Boltzmann distribution), and (c) IR spectrum were calculated using the fully converged OT method for delocalized orbitals (black line) and CLMO AIMD with ϵ SCF = 10 -2 a.u. (d) Timing benchmarks for liquid water on 256 compute cores. Cyan lines represent perfect cubic-scaling, whereas gray lines represent perfect linear-scaling. (e) Weak scalability benchmarks are performed with ϵ SCF = 10 -2 a.u. Gray dashed lines connect systems simulated on the same number of cores to confirm O ( N ) behavior. See Ref. 250 for details.
FIG. 10 Accuracy and efficiency of O ( N ) AIMD based on the two-stage CLMO SCF for liquid water. All simulations were performed using the PBE XC functional and a TZV2P basis set at constant temperature (300 K) and density (0.9966 g/cm 3 ). For the CLMO methods, Rc = 1.6 vdWR. (a) Radial distribution function, (b) kinetic energy distribution (the gray curve shows the theoretical Maxwell-Boltzmann distribution), and (c) IR spectrum were calculated using the fully converged OT method for delocalized orbitals (black line) and CLMO AIMD with ϵ SCF = 10 -2 a.u. (d) Timing benchmarks for liquid water on 256 compute cores. Cyan lines represent perfect cubic-scaling, whereas gray lines represent perfect linear-scaling. (e) Weak scalability benchmarks are performed with ϵ SCF = 10 -2 a.u. Gray dashed lines connect systems simulated on the same number of cores to confirm O ( N ) behavior. See Ref. 250 for details.CC-BY · From the PDF
FIG. 11 Comparison of the calculated radial distribution functions and IR spectra computed for liquid water based on AIMD simulations at the BLYP + D3/TZV2P level of theory with and without the CT terms. See Ref. 332 for details.
FIG. 11 Comparison of the calculated radial distribution functions and IR spectra computed for liquid water based on AIMD simulations at the BLYP + D3/TZV2P level of theory with and without the CT terms. See Ref. 332 for details.CC-BY · From the PDF
FIG. 12 Simulation of molecules at metallic surfaces using the IC-QM/MM approach. The isosurface of the electrostatic potential at 0.0065 a.u. (red) and -0.0065 a.u. (blue) for a single thymine molecule at Au(111). Left: standard QM/MM approach, where the electrostatic potential of the molecule extends beyond the surface into the metal. Right: IC-QM/MM approach reproducing the correct electrostatics expected for a metallic conductor.
FIG. 12 Simulation of molecules at metallic surfaces using the IC-QM/MM approach. The isosurface of the electrostatic potential at 0.0065 a.u. (red) and -0.0065 a.u. (blue) for a single thymine molecule at Au(111). Left: standard QM/MM approach, where the electrostatic potential of the molecule extends beyond the surface into the metal. Right: IC-QM/MM approach reproducing the correct electrostatics expected for a metallic conductor.CC-BY · From the PDF
FIG. 13 Schema of the DBCSR library for the matrix-matrix multiplication (see text for description).
FIG. 13 Schema of the DBCSR library for the matrix-matrix multiplication (see text for description).CC-BY · From the PDF
FIG. 14 Comparison of DBCSR dense multiplication of square matrices of size 10 000, dominated by different sub-matrix block sizes. With the JIT framework in place, these blocks are batch-multiplied on a GPU using LIBSMM_ACC , instead of on the CPU using LIBXSMM . Multiplications were run on a heterogeneous Piz Daint CRAY XC50 node containing a 12 core Intel Haswell CPU and on a NVIDIA V100 GPU.
FIG. 14 Comparison of DBCSR dense multiplication of square matrices of size 10 000, dominated by different sub-matrix block sizes. With the JIT framework in place, these blocks are batch-multiplied on a GPU using LIBSMM_ACC , instead of on the CPU using LIBXSMM . Multiplications were run on a heterogeneous Piz Daint CRAY XC50 node containing a 12 core Intel Haswell CPU and on a NVIDIA V100 GPU.CC-BY · From the PDF
FIG. 15 Comparison of potential energies between CP2K + SIRIUS and QE during an AIMD simulation. The total energy (black curve) remains constant over time (it is a conserved quantity in the NPT ensemble), while the potential energy calculated with CP2K + SIRIUS (blue curve) and QE (red dashed curve) have the same time evolution up to a constant shift in energy (see the lower panel).
FIG. 15 Comparison of potential energies between CP2K + SIRIUS and QE during an AIMD simulation. The total energy (black curve) remains constant over time (it is a conserved quantity in the NPT ensemble), while the potential energy calculated with CP2K + SIRIUS (blue curve) and QE (red dashed curve) have the same time evolution up to a constant shift in energy (see the lower panel).CC-BY · From the PDF
Image 16
CC-BY · From the PDF
FIG. 16 Parallel scaling of CP2K calculations up to thousands of CPU-cores: single-point energy calculation using GAPW with exact Hartree-Fock exchange for a periodic 216 atom lithium hydride crystal (red line with circles), single-point energy calculation using RI-MP2 for 64 water molecules with periodic boundary conditions (blue line with triangles), and a single-point energy calculation in linearscaling DFT for 2048 water molecules in the condensed phase using a DZVP MOLOPT basis set (green line with boxes). All calculations have been performed with up to 256 nodes (10 240 CPU-cores) of the Noctua system at PC 2 .
FIG. 16 Parallel scaling of CP2K calculations up to thousands of CPU-cores: single-point energy calculation using GAPW with exact Hartree-Fock exchange for a periodic 216 atom lithium hydride crystal (red line with circles), single-point energy calculation using RI-MP2 for 64 water molecules with periodic boundary conditions (blue line with triangles), and a single-point energy calculation in linearscaling DFT for 2048 water molecules in the condensed phase using a DZVP MOLOPT basis set (green line with boxes). All calculations have been performed with up to 256 nodes (10 240 CPU-cores) of the Noctua system at PC 2 .CC-BY · From the PDF
FIG. 17 Energy deviation and deviation from the idempotency condition of the density matrix as a function of matrix truncation threshold for the same system and calculation as in Fig. 6. For the reference energy, a calculation with ϵfi lter = 10 -7 was used.
FIG. 17 Energy deviation and deviation from the idempotency condition of the density matrix as a function of matrix truncation threshold for the same system and calculation as in Fig. 6. For the reference energy, a calculation with ϵfi lter = 10 -7 was used.CC-BY · From the PDF
Image 19
CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 7

  1. Grotendorst, 2004, Quickstep: Make the atoms dance, 29
  2. 1969, ESAIM: Math. Modell. Numer. Anal., 3, 35
  3. 2006, Numerical Optimization
  4. B. Leimkuhler, M. Sachs, and G. Stoltz, “Hypocoercivity properties of adaptive Langevin dynamics,” arXiv:1908.09363 (2019). · arXiv · doi:10.1137/19m1291649
  5. Grotendorst, 2000, Modern Methods and Algorithms of Quantum Chemistry, 257
  6. 1964, Sov. Phys. JETP, 20, 1018
  7. 2014, Using Advanced MPI: Modern Features of the Message-Passing Interface

Cited by

See all 4,468

Metadata

Authors
Kühne, T. D., Iannuzzi, M., Del Ben, M., Rybkin, V. V., Seewald, P., Stein, F., Laino, T., Khaliullin, R. Z., Schütt, O., Schiffmann, F., Golze, D., Wilhelm, J., Chulkov, S., Bani-Hashemian, M. H., Weber, V., Borštnik, U., Taillefumier, M., Jakobovits, A. S., Lazzaro, A., Pabst, H., Müller, T., Schade, R., Guidon, M., Andermatt, S., Holmberg, N., Schenter, G. K., Hehn, A., Bussy, A., Belleflamme, F., Tabacchi, G., Glöß, A., Lass, M., Bethune, I., Mundy, C. J., Plessl, C., Watkins, M., VandeVondele, J., Krack, M., & Hutter, J.
Affiliations
  • Paderborn University (DE)
  • University of Zurich (CH)
  • Computational Research Division, Lawrence Berkeley National Laboratory 3 , Berkeley, California 94720, USA
  • IBM Research - Zurich (CH)
  • McGill University (CA)
  • ETH Zurich (CH)
  • Victoria University (AU)
  • Aalto University (FI)
  • University of Regensburg (DE)
  • University of Lincoln (GB)
  • HPE Switzerland GmbH 14 , Basel, Switzerland
  • Intel (United States) (US)
  • Pacific Northwest National Laboratory (US)
  • University of Insubria (IT)
  • BASF (Germany) (DE)
  • Science and Technology Facilities Council (GB)
  • School of Mathematics and Physics, University of Lincoln 10 , Brayford Pool, Lincoln, United Kingdom
  • Paul Scherrer Institute (CH)
Corresponding author
Kühne, Thomas D.
Venue
The Journal of Chemical Physics
Published
2020-05-19
Volume
152
Issue
19
Pages
194103-194103
ISSN
  • 0021-9606 (ISSN-L)
  • 1089-7690
OpenAlex
W3027308092
Cited by
4,468
References
445
Citations by year
2026: 1066 · 2025: 1195 · 2024: 870 · 2023: 561 · 2022: 431 · 2021: 284 · 2020: 58
FWCI
135.74
Type
journal article
Language
en
Open access
Licence
Locations
  • The Journal of Chemical Physics (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Repository for Publications and Research Data (ETH Zurich) (publishedVersion, cc-by)
  • Lincoln Repository (University of Lincoln) (acceptedVersion)
  • University of Regensburg Publication Server (University of Regensburg) (acceptedVersion)
  • Zurich Open Repository and Archive (University of Zurich) (acceptedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion)
  • Victoria University Research Repository (Victoria University) (submittedVersion, cc-by)
  • Aaltodoc (Aalto University) (submittedVersion, cc-by)
  • DORA PSI (Paul Scherrer Institute) (submittedVersion, cc-by)
  • Figshare (submittedVersion, cc-by)
  • IrInSubria (University of Insubria) (submittedVersion, other-oa)
  • OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) (submittedVersion)
  • Zurich Open Repository and Archive (University of Zurich) (submittedVersion, cc-by)
  • arXiv (Cornell University)
  • ETH Zürich Research Collection (cc-by)
  • Zurich Open Repository and Archive (University of Zurich)
  • University of Regensburg Publication Server (University of Regensburg)
Funders
  • Natural Sciences and Engineering Research Council of Canada (RGPIN-2016-0505)
  • H2020 European Research Council (716142)
  • Schweizerischer Nationalfonds Zur Förderung der Wissenschaftlichen Forschung (PZ00P2 174227)
Topics
  • Physical Sciences › Physics and Astronomy › Atomic and Molecular Physics, and Optics › Advanced Chemical Physics Studies
  • Physical Sciences › Chemistry › Physical and Theoretical Chemistry › Advanced Physical and Chemical Molecular Interactions
  • Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
Keywords
CP2K, electronic structure software, molecular dynamics, ab initio molecular dynamics, density functional theory, post-Hartree-Fock methods, Gaussian basis sets, plane-wave basis, linear-scaling methods, massively parallel computing, atomistic simulation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.