transport

name	transport
description	Prepare GPUMD thermal-transport workflows for EMD / HNEMD / NEMD / SHC / HNEMDEC. Use when the user needs `compute_hac`, `compute_hnemd`, `compute_hnemdec`, `compute_shc`, thermal-conductivity extraction, or transport-specific sampling and convergence guidance.
compatibility	Requires GPUMD and a potential that is numerically stable in the target state.
catalog-hidden	true
license	GPL-3.0-only
metadata	{"author":"Jhin","version":"0.2.0"}

GPUMD Thermal Transport

Use this subskill for heat-transport calculations in GPUMD. Diffusion, viscosity, and ionic conductivity are related observables but live in gpumd/diffusion.

Quick method picker

Target	Method	Keyword
bulk κ in a single-component crystal / liquid	HNEMD	compute_hnemd
bulk κ via Green-Kubo	EMD	compute_hac
length dependence or interface κ	NEMD	not yet covered
spectral decomposition of κ	spectral heat current	compute_shc
multi-component coupled transport	HNEMDEC	compute_hnemdec

Agent responsibilities

1. Pick the method from the observable and justify the pick explicitly.
2. Require an explicit equilibration stage before production.
3. Enforce linear-response discipline for HNEMD: start at a small driving field and only increase if the result is field-independent within uncertainty.
4. For low-dimensional systems, require an explicit thickness convention before reporting W m^-1 K^-1.
5. Report the number of seeds, production length, and uncertainty next to the κ value.

Workflow

Step 1. Equilibrate

Always start from a short NVT / NPT equilibration to the target state, then switch ensemble for production. Do not keep an aggressive thermostat on during a Green-Kubo production segment.

Step 2. EMD / Green-Kubo via compute_hac

Annotated pattern (see assets/examples/transport/emd/run.in):

potential   nep.txt
velocity    300
time_step   1

# equilibrate
ensemble    npt_scr 300 300 100 0 0 0 100 100 100 1000
dump_thermo 100
run         100000

# production in NVE
ensemble    nve
dump_thermo 100
compute_hac 20 200 2
run         2000000

• compute_hac sample_interval Nc output_interval
  • sample_interval — store heat-current samples every N steps
  • Nc — maximum number of correlation time steps to be calculated
  • output_interval — how often the running integral is written

Inspect hac.out and the running integral of the correlation function, not just the peak. Use multiple seeds when the noise level matters.

Step 3. HNEMD via compute_hnemd

Annotated pattern (see assets/examples/transport/hnemd/run.in):

potential   nep.txt
velocity    300
time_step   1

# equilibrate
ensemble    npt_scr 300 300 100 0 0 0 100 100 100 1000
dump_thermo 100
run         100000

# HNEMD production (thermostatted)
ensemble    nvt_nhc 300 300 100
compute_hnemd 1000 0.00001 0 0
dump_thermo 1000
run         1000000

• compute_hnemd output_interval Fe_x Fe_y Fe_z
  • Fe_x, Fe_y, Fe_z are the driving-field components in 1/Å.
  • Start at ~1e-5 1/Å for stiff crystalline solids and verify field-independence before trusting the result.

Post-process with the bundled helper:

python scripts/average_hnemd_kappa.py kappa.out

Do not blindly discard a fixed fraction of the time series. Instead, inspect the kappa.out running average to identify when it has equilibrated, then use --discard-rows N with N chosen from that inspection.

Step 3.5. NEMD (stub)

NEMD workflows for length-dependent or interface thermal conductivity are supported by GPUMD but not yet covered in this skill. Refer directly to:

• GPUMD documentation: https://gpumd.org/gpumd/input_parameters/index.html
• Tutorial examples: 07_Carbon_thermal_conductivity_NEMD

Step 4. SHC / spectral decomposition

For spectral heat-current decomposition use compute_shc with appropriate group definitions. Groups are set up in model.xyz via group:I:M in the Properties header, or by add_groups preprocessing.

Step 5. HNEMDEC (multi-component coupled transport)

For multi-component systems with cross-coupling between mass and heat flux (molten salts, ionic conductors, mixed liquids), use compute_hnemdec. Do not apply the single-component HNEMD formula to a multi-component system without checking the cross-coupling.

Relevant tutorial anchor: 29_thermal_transport_multicomponent_HNEMDEC.

Step 6. Report

For a publishable transport number, report at least:

• number of independent seeds or replicas
• production length per seed
• cell size
• uncertainty estimate
• whether the result was tested against a different driving field or a second cell size

Common pitfalls

• leaving a strong thermostat on during Green-Kubo production
• treating a short HNEMD trajectory as converged
• reporting W m^-1 K^-1 for a monolayer without a stated thickness convention
• using compute_hnemd in a regime where the flux is no longer linear in the field

Read first

• references/thermal-transport-workflow.md

Read when needed:

• references/core-files-and-ensembles.md
• references/gpumd-keyword-cheatsheet.md
• references/tutorial-map.md

Bundled templates and helpers

• assets/examples/transport/emd/run.in
• assets/examples/transport/hnemd/run.in
• scripts/average_hnemd_kappa.py
• scripts/parse_thermo.py

Expected output

1. an explicit method choice with the reason
2. a transport-specific GPUMD input template
3. linear-response / convergence / reporting caveats that match the method

References

• compute_hac: https://gpumd.org/gpumd/input_parameters/compute_hac.html
• compute_hnemd: https://gpumd.org/gpumd/input_parameters/compute_hnemd.html
• compute_shc: https://gpumd.org/gpumd/input_parameters/compute_shc.html
• compute_hnemdec: https://gpumd.org/gpumd/input_parameters/compute_hnemdec.html
• Fan et al., Phys. Rev. B 99, 064308 (2019)
• Fan et al., Phys. Rev. B 92, 094301 (2015)