Matter and Radiation at Extremes, Volume. 9, Issue 1, 015604(2024)

Combining stochastic density functional theory with deep potential molecular dynamics to study warm dense matter

Tao Chen... Qianrui Liu, Yu Liu, Liang Sun and Mohan Chena) |Show fewer author(s)
Author Affiliations
  • HEDPS, CAPT, College of Engineering and School of Physics, Peking University, Beijing 100871, People’s Republic of China
  • show less
    Figures & Tables(12)
    Workflow of the simulation of WDM with the SDFT and DPMD methods. (a) Stochastic orbitals are employed in SDFT to perform molecular dynamics simulations on smaller systems (32 atoms in a bulk B) and initial training data are collected that include atomic positions, energies, forces, and virial tensors. (b) The gathered training data are used to construct a DP model with the temperature-dependent DPMD model. The deep neural network contains both embedding and fitting networks. (c) The new model enables simulations to be performed on large systems (16 384 atoms) and at extremely high temperatures (350 eV). (d) Several physical quantities such as radial distribution functions, static structure factors, dynamic structure factors, and shear viscosities can be calculated, and the data are converged with large systems and long trajectories.
    Root-mean-square error (RMSE) of atomic forces arising from the SDFT calculations for B. The temperature is set to 350 eV and the density to 2.46 g/cm3. The RMSE is evaluated with respect to (a) the number of stochastic orbitals Nsto, (b) the number of k-points in the Brillouin zone Nk, and (c) the product of Nsto and Nk. For each data point, the RMSE is obtained via nine independent SDFT calculations with different sets of stochastic orbitals.
    Forces acting on each B atom as obtained from nine independent SDFT calculations with different stochastic orbitals. The B system has a density of 2.46 g/cm3, and the temperature is 17.23 eV. For each calculation, the force acting on each atom along the γ direction is denoted by Fsto, and their average is Fave. NKS and Nsto are the numbers of KS and stochastic orbitals, respectively. Two sets of Monkhorst–Pack k-points are utilized: a 1 × 1 × 1 k-point mesh (the Γ k-point) and a shifted 2 × 2 × 2 k-point mesh. The RMSE of forces between Fsto and Fave is computed from Eq. (13). The RMSE is obtained via the mentioned nine independent SDFT calculations.
    (a) Comparison of forces acting on each B atom in a 32-atom cell for densities of 2.46 and 12.3 g/cm3 at a temperature of 17.23 eV. (b) Comparison of forces acting on each C atom for densities of 4.17 and 12.46 g/cm3 at a temperature of 21.54 eV. In the SDFT and KSDFT calculations, we denote the forces acting on each atom along the γ direction (γ ∈ x, y, z) by Fsto and FKS, respectively. The RMSE of forces between Fsto and FKS is also shown.
    DOS for the B and C systems as calculated by the SDFT and KSDFT methods. The densities are selected as (a) 2.46 g/cm3 and (b) 12.3 g/cm3 for the B system, and (c) 4.17 g/cm3 and (d) 12.64 g/cm3 for the C system. The Fermi energy is set to zero. We use two sets of k-point sampling in the KSDFT and SDFT calculations. In addition, we adopt two XC functionals, namely, PBE59 and corrKSDT.65
    Radial distribution functions g(r) of B systems with a density of 2.46 g/cm3 at temperatures of (a) 86 eV and (b) 350 eV. The g(r) obtained by Ext-FPMD with the local density approximation (LDA) functional comes from Blanchet et al.19 The SDFT calculations are performed with the PBE59 and corrKSDT65 XC functionals. The number of B atoms is set to 32 in the first-principles molecular dynamics simulations. DPMD denotes the model trained by the traditional DP method50 and DPMD-T the model trained by the TDDP method.51N is the number of B atoms in a cell, which ranges from 32 to 16 384 in the deep-potential-based simulations.
    Static structure factors S(q) of B with a density of 2.46 g/cm3 at temperatures of (a) 86 eV and (b) 350 eV. The SDFT calculations are performed with the PBE59 and corrKSDT65 XC functionals. The number of B atoms is set to 32 in the first-principles molecular dynamics simulations. DPMD denotes the model trained by the traditional DP method50 and DPMD-T the model trained by the TDDP method.51N is the number of B atoms in a cell and ranges from 32 to 16 384 in the deep-potential-based simulations.
    Intermediate scattering functions F(q, t) of B with a density of 2.46 g/cm3 as calculated from DPMD trajectories. Three system sizes (256, 2048, and 16 384 atoms) are adopted in DPMD simulations at two temperatures of (a) 86 eV and (b) 350 eV. Three wave vectors are chosen: q = 0.51, 1.02, and 2.50 Å−1.
    Ion–ion dynamic structure factors S(q, ω) of B with a density of 2.46 g/cm3 as computed from DPMD trajectories. Three system sizes (256, 2048, and 16 384 atoms) are adopted. The wave vectors q are chosen to be (a) 0.51 Å−1, (b) 1.02 Å−1, and (c) 2.50 Å−1.
    Stress autocorrelation functions [Eq. (21)] of warm dense B with a density of 2.46 g/cm3 at temperatures of (a) 86 eV and (b) 350 eV. (c) Shear viscosity of B. DPMD simulations are used, with cells containing 32, 108, 256, 864, 2048, 6912, and 16 384 atoms. Error bars represent standard deviations.
    • Table 1. Comparison of the total energy per atom E, the pressure P, and the degree of ionization α for B and C systems as obtained from the SDFT and traditional KSDFT methods. Four systems are chosen: two B systems at a temperature of 17.23 eV with densities of 2.46 and 12.3 g/cm3, and two C systems with densities of 4.17 and 12.46 g/cm3 at a temperature of 21.54 eV. Δ denotes the percentage difference between the results obtained by SDFT and KSDFT.

      View table
      View in Article

      Table 1. Comparison of the total energy per atom E, the pressure P, and the degree of ionization α for B and C systems as obtained from the SDFT and traditional KSDFT methods. Four systems are chosen: two B systems at a temperature of 17.23 eV with densities of 2.46 and 12.3 g/cm3, and two C systems with densities of 4.17 and 12.46 g/cm3 at a temperature of 21.54 eV. Δ denotes the percentage difference between the results obtained by SDFT and KSDFT.

      E (eV)P (GPa)α
      BKSDFT−153.683 171849.1120.476 617
      2.46 g/cm3SDFT−153.707 884852.5240.476 659
      17.23 eVΔ0.0161%0.4019%0.0088%
      BKSDFT−66.923 1778730.0100.390 341
      12.3 g/cm3SDFT−66.924 2358730.8290.390 344
      17.23 eVΔ0.0016%0.0094%0.0008%
      CKSDFT−263.586 5432018.3800.505 693
      4.17 g/cm3SDFT−263.595 8102019.7800.505 707
      21.54 eVΔ0.0035%0.0693%0.0028%
      CKSDFT−190.692 8439168.0500.440 747
      12.46 g/cm3SDFT−190.698 5359171.5520.440 757
      21.54 eVΔ0.0030%0.0382%0.0023%
    • Table 2. RMSE of forces acting on B atoms as obtained from nine independent SDFT calculations with two sets of stochastic orbitals (128 and 256). The B system has a density of 2.46 g/cm3, and the temperatures are set to 17.23, 86, 350, and 1000 eV. NKS is the number of KS orbitals, and Nsto is the number of stochastic orbitals. Two sets of Monkhorst–Pack k-points are utilized: a 1 × 1 × 1 k-point mesh (include the Γ k-point) and a shifted 2 × 2 × 2 k-point mesh. The RMSE of forces is computed from Eq. (13).

      View table
      View in Article

      Table 2. RMSE of forces acting on B atoms as obtained from nine independent SDFT calculations with two sets of stochastic orbitals (128 and 256). The B system has a density of 2.46 g/cm3, and the temperatures are set to 17.23, 86, 350, and 1000 eV. NKS is the number of KS orbitals, and Nsto is the number of stochastic orbitals. Two sets of Monkhorst–Pack k-points are utilized: a 1 × 1 × 1 k-point mesh (include the Γ k-point) and a shifted 2 × 2 × 2 k-point mesh. The RMSE of forces is computed from Eq. (13).

      T (eV)NKSNstok-pointsRMSE (eV/Å)
      17.231281281 × 1 × 10.596
      1281282 × 2 × 20.302
      861281281 × 1 × 13.26
      1281282 × 2 × 21.54
      35002561 × 1 × 16.40
      02562 × 2 × 23.26
      100002561 × 1 × 12.68
      02562 × 2 × 21.30
    Tools

    Get Citation

    Copy Citation Text

    Tao Chen, Qianrui Liu, Yu Liu, Liang Sun, Mohan Chen. Combining stochastic density functional theory with deep potential molecular dynamics to study warm dense matter[J]. Matter and Radiation at Extremes, 2024, 9(1): 015604

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category:

    Received: Jun. 16, 2023

    Accepted: Dec. 13, 2023

    Published Online: Mar. 27, 2024

    The Author Email: Chen Mohan (mohanchen@pku.edu.cn)

    DOI:10.1063/5.0163303

    Topics