PaperlayerView exact revision on arXiv

Source-linked AI summary

RadonPy: Automated Physical Property Calculation using All-atom Classical Molecular Dynamics Simulations for Polymer Informatics

Yoshihiro Hayashi, Junichiro Shiomi, Junko Morikawa, Ryo Yoshida

arXiv:2203.14090v1cond-mat.mtrl-scicond-mat.softphysics.comp-ph

TL;DR

Open polymer-property databases and computational datasets remain limited, while high-throughput MD workflows are technically challenging. RadonPy automates all-atom classical MD and property calculations, and its calculations were comprehensively verified across more than 1,000 amorphous polymers, with several properties reproducing experimental values.

  • Problem

    Open databases and computational property data for polymers are limited, and high-throughput MD workflows are technically challenging.

    Source: §1 · pp. 3–4

  • Method

    RadonPy automates all-atom classical MD modeling, equilibrium and nonequilibrium simulations, and property calculations from a polymer repeating unit.

    Source: §1 · p. 5

  • Results

    More than 1,000 unique amorphous polymers were comprehensively evaluated, with refractive index, density, and thermal conductivity successfully reproducing experimental values.

    Source: §4 · pp. 29–30

  • Takeaways & Limitations

    RadonPy provides a validated computational-property resource for amorphous polymers, including properties such as refractive index, density, and thermal conductivity.

    Source: §4 · pp. 29–30

    Some calculated–experimental correlations were weak amid large variations, and classical MD calculations showed systematic bias.

    Source: §4 · p. 30

Abstract

from arXiv · show

The rapid growth of data-driven materials research has made it necessary to develop systematically designed, open databases of material properties. However, there are few open databases for polymeric materials compared to other material systems such as inorganic crystals. To this end, we developed RadonPy, the world-first open-source Python library for fully automated all-atom classical molecular dynamics (MD) simulations. For a given polymer repeating unit, the entire process of molecular modeling, equilibrium and nonequilibrium MD calculations, and property calculations can be conducted fully automatically. In this study, 15 different properties, including the thermal conductivity, density, specific heat capacity, thermal expansion coefficients, and refractive index, were calculated for more than 1,000 unique amorphous polymers. The calculated properties were compared and validated systematically with experimental values from PoLyInfo. During the high-throughput data production, eight amorphous polymers with extremely high thermal conductivities, exceeding 0.4 W/mK, were identified, including six polymers with unreported thermal conductivities. These polymers were found to have a high density of hydrogen bonding units or rigid backbones. A decomposition analysis of the heat conduction, which is implemented in RadonPy, revealed the underlying mechanisms that yield a high thermal conductivity of the amorphous polymers: heat transfer via hydrogen bonds and dipole-dipole interactions between the polymer chains with their hydrogen bonding units or via the covalent bonds of the polymer backbone with high rigidity. The creation of massive amounts of computational property data using RadonPy will facilitate the development of polymer informatics, similar to how the emergence of the first-principles computational database for inorganic crystals had significantly advanced materials informatics.

1. INTRODUCTION

Polymer property databases and high-throughput computational data have lagged, while MD workflows are technically challenging to automate. RadonPy automates all-atom classical MD and property calculations, enabling a large amorphous-polymer dataset and the identification of high-thermal-conductivity polymers.

1. INTRODUCTION

  • Open polymer-property databases lag other materials databases, and existing polymer data are sparse across simultaneously recorded properties.PoLyInfo is described as containing approximately 100 properties for more than 18,000 polymers, with few cases recording multiple properties for one polymer.Source: §1 · p. 3
  • Automating polymer MD is technically challenging because workflows must handle conditional parameters, job control, and error handling, while a unified platform is needed to create varied polymeric states such as amorphous and oriented structures.Source: §1 · p. 4
  • RadonPy is an open-source Python library that automates all-atom classical MD modeling, simulations, equilibration checks, restarts, and property calculations from a polymer repeating unit.Its workflow includes equilibrium and nonequilibrium MD.Source: §1 · p. 5
  • More than 1,000 unique amorphous polymers were evaluated for 15 properties, and calculated values were systematically validated against PoLyInfo experiments.The properties include thermal conductivity, density, specific heat capacity, thermal expansion coefficient, and refractive index.Source: §1 · p. 5
  • Eight amorphous polymers had thermal conductivities exceeding 0.4 W/m·K, including six with previously unreported values.These polymers had dense hydrogen-bonding units or rigid, linear backbones.Source: §1 · p. 5
  • Heat-conduction decomposition identified transfer through hydrogen bonds and dipole–dipole interactions between chains, or through covalent bonds in rigid, linear backbones.The analysis is implemented in RadonPy.Source: §1 · p. 5

2. METHODS

This section describes RadonPy’s methods for automating polymer-property calculations, from molecular modeling and amorphous-cell construction through equilibration, simulations, and property evaluation. It also covers thermal-conductivity simulations and selection of a PoLyInfo homopolymer dataset.

2.1.1. Conformation search of a repeating unit

  • For a polymer repeating-unit SMILES, up to 1,000 conformations were generated using ETKDG version 2, Butina-clustered by torsion fingerprint deviation, and the four most stable conformations further optimized by DFT.Source: §2.1.1 · pp. 7–8

2.1.2 Calculation of electronic property of a repeating unit

  • RESP charges were calculated by Hartree–Fock single-point calculation, while the dipole moment, HOMO/LUMO energies, and polarizability tensor were calculated using ωB97M-D3BJ.The RESP calculation used 6-31G(d) on the optimized most-stable conformation; polarizability was obtained by a finite-field method.Source: §2.1.2 · p. 8

2.1.3. Generation of polymer chains

  • Chains were grown by a self-avoiding random walk, joining repeat units with randomized new-bond dihedrals and targeting approximately 1,000 atoms per chain.This approximately equalized molecular weights across polymers, so molecular-weight dependence of properties was ignored; chains were generated as atactic polymers.Source: §2.1.3 · p. 9

2.1.4. Assignment of force field parameters

  • RadonPy automatically assigned GAFF2 parameters, whose bond and angle force constants were based on ab initio calculations and tuned to reproduce experimental vibrational properties.Modified fluorocarbon parameters were used for fluorocarbon polymers; missing angle parameters were empirically estimated in the GAFF2 manner.Source: §2.1.4 · p. 10

2.1.5. Generation of a simulation cell

  • Ten randomly arranged, nonoverlapping polymer chains formed an amorphous cell of approximately 10,000 atoms, initially set to 0.05 g/cm3.The cell density was then increased by packing simulation.Source: §2.1.5 · p. 10

2.2.1 Packing simulation

  • Packing increased amorphous-polymer density for subsequent calculations using 1 ns NVT Nosé–Hoover heating from 300 K to 700 K, followed by 1 ns NVT isotropic cell reduction to 0.8 g/cm3 at 700 K.The Lennard–Jones cutoff was set to 3.0 Å to prevent intramolecular self-aggregation.Source: §2.2.1 · p. 10

2.2.2 Equilibration simulation

  • After the 21-step compression/decompression protocol, NpT simulations ran for more than 5 ns at 300 K and 1 atm until equilibrium was achieved, with checks every 5 ns.Source: §2.2.2 · p. 11
  • Equilibration calculations failing to reach equilibrium within 50 ns were discarded, and an amorphous structure was accepted when its nematic order parameter fell below 0.1.Otherwise, the calculation was treated as a failure and removed from the data.Source: §2.2.2 · p. 11

2.2.3 NEMD simulation for thermal conductivity calculation

  • Thermal conductivity was calculated by reverse NEMD in a triplicated equilibrated cell, with velocity exchanges inducing a temperature gradient; runs with gradient-fit R2 below 0.95 were discarded.The simulation used a 2 ps NVT preheat at 300 K followed by 1 ns of NVE dynamics, and RadonPy checked temperature-gradient linearity.Source: §2.2.3 · p. 12
  • A 100 ps thermal-conductivity decomposition resolved energy transport into convection and interaction contributions, with interactions separated into pairwise, bonded, improper, and long-range Coulomb terms.The interaction terms included pairwise, bond, angle, dihedral, improper, and K-space contributions.Source: §2.2.3 · p. 13

2.3. Calculation of physical properties

  • The study calculated physical properties from simulation observables, including fluctuation-based thermodynamic quantities, refractive index, dielectric constant, and thermal conductivity.The refractive index used the Lorentz–Lorenz equation and DFT-computed repeating-unit polarizability; thermal conductivity followed Fourier’s law, and thermal diffusivity used conductivity, density, and heat capacity.Source: §2.3 · pp. 14–16

2.4. Dataset

  • The dataset comprised 1,138 unique PoLyInfo homopolymers selected for experimental coverage, spanning diverse backbones and restricted to neat, additive-free, linear samples measured at 273–323 K with amorphous or unidentified postforming state.Eligible samples contained no additives, fillers, or dopants.Source: §2.4 · pp. 16–17

3. RESULTS AND DISCUSSION

The results assess calculated polymer properties through comparison with PoLyInfo, characterization of chemical-space coverage and property distributions, and analysis of relationships and tradeoffs among properties. They also examine selected high-conductivity polymers through decomposition analysis of bond contributions and intermolecular interactions.

3.2. Validation of the calculated physical properties

  • Calculated density reproduced experimental values (R2 = 0.890), with slight underestimation and a fitted parity-plot slope of 0.805; calculated thermal conductivity had R2 = 0.490 against experimental PoLyInfo values.Source: §3.2 · pp. 18–19
  • Refractive index reproduced the PoLyInfo dataset well (R2 = 0.809), with slight underestimation and fitted parity-plot slope 0.839; calculated CP was overestimated relative to experimental values (R2 = 0.602).The slight refractive-index underestimation was attributed to slight density underestimation.Source: §3.2 · p. 19
  • Linear and volume expansion coefficients correlated weakly with experiments (R2 = 0.178 and 0.217), with substantial variation within polymers.The authors note that larger simulation timescales and cell sizes may be needed for accurate expansion-coefficient calculations.Source: §3.2 · pp. 19–20

3.3. Data distribution

  • Across 1,070 polymers, density ranged from 0.742 to 1.914 g/cm3 (mean 1.133), while refractive index ranged from 1.274 to 1.839 (mean 1.550).Twelve polymers had density >1.75 g/cm3 and contained abundant halogen atoms; nine were identified as high-refractive-index polymers.Source: §3.3 · pp. 21–22
  • Eight reliable polymers had thermal conductivity above 0.4 W/m·K, including polyethylene PI1 at 0.456 W/m·K and poly(vinyl alcohol) PI241 at 0.439 W/m·K.These eight had repeated-calculation SD < 0.05 W/m·K; PI1’s value was consistent with reported 0.39–0.53 W/m·K, while PI241’s exceeded its reported 0.31 W/m·K.Source: §3.3 · p. 21
  • Thermal conductivities were unreported for six of the eight high-conductivity polymers; the identified structures included hydrogen-bond-rich polymers, rigid aromatic polyamides, and aromatic polyimides.The hydrogen-bond-rich examples were PI241 and PI305; PI626 was a rigid, linear aromatic polyamide, and PI687, PI711, PI715, and PI1093 were aromatic polyimides.Source: §3.3 · pp. 21–22
  • Amorphous polymers with refractive indices greater than 1.75 had large π-conjugated backbones, and the distributions suggest polarizability dominates refractive index.The joint distributions also indicate unexplored regions combining lower thermal conductivity with higher refractive index, or higher conductivity with lower refractive index.Source: §3.3 · pp. 22–23
  • The specific heat capacity and thermal conductivity distributions suggest difficulty achieving both high heat capacity and low conductivity; polymers near this Pareto frontier were mainly polystyrenes and polyacrylates.No Pareto frontier was observed in the higher-thermal-conductivity region.Source: §3.3 · p. 22

3.4. Decomposition analysis of thermal conductivity

  • 0.456 W/m·K was calculated for PI1 (polyethylene), whose high thermal conductivity was attributed to bond bending.The decomposition analysis quantified contributions from different interaction components in the eight high-conductivity polymers.Source: §3.4 · p. 27
  • PI241 (polyvinyl alcohol) and PI305 (poly(vinylene) carbonate) had high thermal conductivity largely from nonbonded interactions and heat transfer through hydrogen bonds and dipole–dipole interactions.Their chains contain highly condensed hydroxyl groups, creating strong intermolecular interactions.Source: §3.4 · pp. 27–28
  • PI626 (Kevlar) showed moderately large bond, angle, dihedral, and nonbonded contributions, consistent with its rigid backbone and amide-group interactions.Its backbone supports heat conduction through covalent bonds, while amide groups enable hydrogen-bond and dipole–dipole interactions.Source: §3.4 · p. 28
  • Bond stretching largely contributed to the thermal conductivity of aromatic polyimides PI687, PI711, PI715, and PI1093, with PI687 especially rigid and linear.The results link rigid, linear backbones to enhanced conductivity through bond stretching.Source: §3.4 · p. 28
  • PI1093 had the largest nonbonded contribution among the four identified aromatic polyimides, suggesting hydrogen-bonding units alongside rigid, linear backbones can further increase conductivity.PI1093 is an aromatic polyimide containing an amide group.Source: §3.4 · p. 28
  • Thermal conductivity correlated clearly with bond, angle, dihedral, and nonbonded contributions, but not significantly with convection.The analysis suggests that increasing the bond, angle, dihedral, and nonbonded contributions can help design thermally conductive amorphous polymers.Source: §3.4 · p. 28

4. CONCLUSIONS

RadonPy automates all-atom classical-MD polymer-property calculations and was used to compute properties for more than 1,000 amorphous polymers. Validation was strong for several properties, but expansion coefficients were weakly correlated, and broader validation remains needed.

4. CONCLUSIONS

  • RadonPy is presented as the first open-source Python library to automate polymer-property calculations using all-atom classical MD.Source: §4 · p. 29
  • More than 1,000 unique amorphous polymers were computed in approximately two months, mainly using the supercomputer Fugaku.The calculation used RadonPy for large-scale polymer-property data production.Source: §4 · pp. 29–31
  • The calculation covers thermophysical properties of amorphous homopolymers; other systems have some automated capabilities, but calculation protocols have not been established.Source: §4 · pp. 29–30
  • Refractive index, density, and thermal conductivity quantitatively reproduced experiment; specific heat capacity correlated highly but retained a classical-MD bias from omitted quantum effects.Source: §4 · p. 30
  • Linear and volumetric expansion coefficients showed weak calculated–experimental correlation because of large variations and uncertainties in both calculations and experiments.Source: §4 · p. 30
  • More rigorous validation against experimental values, including properties not discussed here, is needed to determine appropriate calculation conditions and protocols.Source: §4 · p. 30
  • RadonPy was developed to produce a systematically designed polymer-property database, with the authors expecting growing computational data to facilitate polymer informatics.Source: §4 · pp. 30–31

Molecular Dynamics Simulations for Polymer Informatics

The listed authors have affiliations at Japanese universities and research institutes.

Molecular Dynamics Simulations for Polymer Informatics

A. Relationship between thermal conductivity and its standard deviation

Figure S1 presents the relationship between thermal conductivity and its standard deviation.

A. Relationship between thermal conductivity and its standard deviation

  • Figure S1 relates thermal conductivity to its standard deviation; the caption does not state the direction or magnitude of the relationship.Source: Figure S1 · p. 39

B. Chemical structure of high-density polymers

Figure S2 shows chemical structures of polymers classified as high-density, with density reported in g/cm3.

B. Chemical structure of high-density polymers

C. Chemical structure of high-refractive-index polymers

Figure S3 presents chemical structures of polymers with refractive indices greater than 1.75.

C. Chemical structure of high-refractive-index polymers

  • Figure S3 presents chemical structures of polymers with refractive indices greater than 1.75.Source: Figure S3 · p. 41

D. Correlations of mean atomic weights with specific heat capacity and with density

Figures S4 and S5 address correlations of mean atomic weight with specific heat capacity and density, respectively.

D. Correlations of mean atomic weights with specific heat capacity and with density

  • Figure S4 concerns the correlation between mean atomic weight and specific heat capacity.Source: Figure S4 · p. 42
  • Figure S5 concerns the correlation between mean atomic weight and density.Source: Figure S5 · p. 42
Loading 2203.14090v1…