Structure of the data

• VASP training data generation. The training data for the MTPs is generated in these active learning runs.
  • PE
    • split into segments 15 K – 30 K, 30 K - 150 K, 150 K - 300 K, 300 K - 400 K and 400 K - 500 K
  • PT
    • 15-500 K
  •  P3HT
    • 15 K – 300 K, 300 K – 400 K and 400 K – 500 K
• Moment Tensor Potential (MTP) main training runs. Typically 5 training runs with different random initialisations are performed, which are labelled with a, b, c, d and e.
  •  PE MTP_MD
    • a, b, c, d ("best"), e
  • PE MTP_phonon
    • a, b, c, d ("best"), e
  • PT MTP_MD
    • a, b, c ("best"), d, e
  • PT MTP_phonon
    • a, b ("best"), c, d, e
  • P3HT MTP_MD
    • a, b, c, d ("best"), e
  • P3HT MTP_phonon
    • a, b ("best")
• VASP validation runs. VASP active learning runs are performed at 300 K in an NPT ensemble. The resulting data is used to determine RMSD_MD and for the comparisons in Section 2.8 "Energy, Force and Stress in Molecular Dynamics".
  • PE. This run is also used in the section "Energy, Force and Stress in Molecular Dynamics"
  • PT
  • P3HT
• MTP further training runs. These are various other training runs that resulted in the MTPs in Section 2.2. "Impact of the Choice of the Reference Data, the Level of the MTP, and the Number of Considered Atom Types"
  • temperature ranges. see Figure 4a and 4b.
    • 15-100 K, 15-200 K 15-300 K, 15-400 K each for PE and PT
    • 15-500 K is identical to MTP_MD and can be found in folder MTP_main_training_runs
  • Levels
    • Level 18
    • Level 22 is identical to the 15-100 K training run in the folder MTP_further_training_runs/temperature_ranges
    • Level 26 is identical to MTP_phonon and can be found in the folder MTP_main_training_runs
  •  Atom type splitting for P3HT
    • 1 carbon type, i.e. no splitting
    • 3 carbon types
    • 6 carbon types 
• Relaxations
  • Relaxed unit cells
    • Relaxed unit cells of the DFT calculations and of the "best" MTP_MD and "best" MTP_phonon. These are the unit cell that are used to calculate e.g. phonon band structures.
  • DFT Relaxations
    • Unit cell and atomic relaxations of PE, PT and P3HT. For descriptions of the relaxation procedure we refer to the Method and Introduction section of the main paper and Section S1 of the Supplementary Materials
• Elastic constants calculations
  • DFT (main calculation with high energy cutoff)
  • MTP_MD
    • PE, PT, P3HT
      • clamped ion method (as shown in main paper)
        • strain distance 1 % (used in main paper), 0.1 % and 0.01 %. (0.1 % and 0.01 % are a convergence test)
          • as above, MTPs with different random initialisations are labelled a, b, c, d, e.
      • relaxed ion method (as shown in Supplementary Materials)
  • MTP_phonon
    • same folder structure as for MTP_MD
  • DFT with low 400 eV energy cutoff
  • DFT VASP-internal 900 eV energy cutoff
• Phonon calculations
  • DFT
  • MTP_phonon
    • PE, PT, P3HT
      • best (in terms of RMSD_phonon)
      • median (in terms of RMSD_phonon)
  • MTP_MD
    • PE, PT, P3HT
      • best (in terms of RMSD_MD)
  • AIREBO

• Thermal conductivity calculations

  • DFT
    • displacement 0.03 Å (main calculation)
      • RTA, 10×15×160 q-mesh
      • full BTE, 10×15×60 q-mesh
      • full BTE, 4×6×160 q-mesh
    • displacement 0.05 Å (convergence test)
  • MTP_phonon
    • a, b, c, d ("best"), e
      • RTA, 10×15×160 q-mesh
      • full BTE, 10×5×60 q-mesh
      • full BTE, 4×6×160 q-mesh

• Thermal Expansion

  • MTP_MD
    • a, b, c, d ("best"), e
  • AIREBO

• Energy, Force and Stress in Molecular Dynamics

  • MTP_MD
  • AIREBO
  • Validation MD active learning run at 300 K is contained in the folder VASP_validation_runs/PE

A Note on axis relabelling

In order to align the three materials PE, PT and P3HT consistently, such that the polymer chain is always in z-direction, we relabelled the axis for the paper in the following way:

PE

x --> z

y --> y

z --> x

PT

x --> y

y --> x

z --> z

P3HT

x --> z

y --> y

z --> x

Example

The q-mesh for the phono3py calculation of PE is given as 160×15×10 in the job-submission-script where phono3py is called, but in the paper (and above) it is written as 10×15×160. All calculations (including POSCAR files etc.) are consistent with each other in terms of their axis labels. Just for the paper, the axis relabelling was performed.