History#

2026.10.22 – The heat flux may be sampled every 4 fs

  • The step needed the heat flux every two timesteps. It now needs it at least every 4 fs, which is as fine as the Green-Kubo integral needs: tested on an electrolyte’s heat flux, 4 fs aliases 0.05 % of the conductivity into it, 10 fs about 20 %. Each sample costs about two MD steps, so sampling every 4 fs rather than every 2 makes a production run about a quarter faster.

  • The fix offered by the check of the flowchart now sets 4 fs, and the user guide explains the choice.

2026.10.21 – Bugfix: no thermal conductivity from an MD run over MDI

  • When the forces of the MD came from an external engine over MDI – a machine-learned forcefield, for example – LAMMPS has no per-atom energies or stresses, so the heat flux held only its kinetic term and the step reported meaningless values (72 ± 913 W/m/K for a carbonate electrolyte). The step now says so and skips the analysis.

2026.10.18 – Bugfix: the thermal conductivity of mixtures was too large

  • In a mixture the heat flux includes the energy carried by the interdiffusion of the species, which is not conduction of heat, so κ was too large: for 1 m NaCl in SPC/E water about 1.15 W/m/K instead of 0.81. The step now removes it, using the mass fluxes of the species that the LAMMPS step samples with the heat flux (Bertossa, Grasselli, Ercole and Baroni, Phys. Rev. Lett. 122, 255901 (2019)), by both Green-Kubo and the Helfand moments. The correction for one run or block is too noisy, so it uses the averages over the runs or blocks, with errors by the jackknife. The notes give the uncorrected value too.

  • For a mixture whose MD did not sample the mass fluxes, made with an older LAMMPS step, the step warns that κ is too large.

2026.10.17 – Analyze MD run before the step; one isotropic κ with labelled errors

  • The step analyzes MD run before it in the flowchart – the runs of a Loop around the MD, as independent replicas (“the last MD run of each loop iteration”, the default) – or md.h5md files; it no longer runs its own MD in a sub-flowchart. Older flowcharts that stored its settings for that can still be read; the check before the job runs says what replaces them.

  • The check before the job runs also warns when the analysis is inside the loop holding its MD, when the loop changes the temperature or pressure, and when the thermostat does not suit the analysis, with fixes; the step warns before the MD when the system is small.

  • Errors are the 95% confidence interval, Student’s t for the number of runs or blocks, and the output says so; stored errors are standard errors.

  • The state of the runs – temperature, pressure, volume and density with their errors, and the equilibrium density where an NPT run before them gives the bulk modulus – is reported and stored. The values of each run or block are in step.out.

  • κ is the average of the three diagonal components within each run, with a check that they agree and that the off-diagonal terms are zero; κ and its components are stored as results.

  • Bugfix: in a loop, the step kept the data of earlier iterations.

2026.10.16 – Bugfix: a short run stopped the analysis with an error

  • A run, or block of a run, shorter than about 10 ps stopped the analysis with an IndexError. Each must now be at least 10 ps long, and the step says so.

2026.10.15.2 – Analyze MD run earlier in the flowchart

  • New “MD to analyze” choice: run the MD in the sub-flowchart (as before), the last MD run earlier in the flowchart, the last MD run of each loop iteration, or md.h5md files, including other jobs’. A single run is divided into blocks to estimate the errors.

  • Before the job runs, the step checks that the MD will sample the heat flux every two timesteps, and the job does not start if it does not.

  • The dialog shows only the settings that apply, and hides the sub-flowchart when it is not used.

  • When the run (or block) is too short to fit the Helfand moments, the output now says so; before, those rows were silently missing.

2026.10.15.1 – Reads the H5MD files; needs a periodic system

  • Reads the heat flux, temperature and volume from the md.h5md files that the LAMMPS step now writes, and from the older text files for runs made before.

  • Stops before running any MD if the system is not periodic, since the heat flux is divided by the volume of the cell.

  • The temperature is now averaged over the same samples as the heat flux, the whole run. It was averaged only after the equilibration detected in the run, while the heat flux was used whole. The thermal conductivity changes very little, e.g. by 0.05% for liquid argon.

  • Bugfix: a run whose temperature and volume were missing stopped with an obscure error; the message now says what is missing. The sampling interval was taken to be in fs even when the file gave it in ps.

2026.10.15 – Bugfix: crash with coarse sampling; unreliable fits

  • The Green–Kubo fit failed when the heat flux was sampled less often than every 0.4 ps. It no longer fails.

  • The Green–Kubo fit could accept a second exponential with a negative amplitude, or one too slow to level off within the data, and give a value far from the plateau of the integral. Such fits are now rejected in favour of a single exponential. The error of a two-exponential fit, which could be enormous when the two time constants were nearly equal, now accounts for their correlation.

  • The errors were from the fits to the curves averaged over the runs, which treat their strongly correlated points as independent: the Green-Kubo and Helfand-moment errors were far too small (e.g. +/- 0.0004 W/m/K where the runs differed by 0.03), and the Helfand-derivative errors too large. The error of each value is now the standard error of the values fitted to each run, and the 95% interval is twice that.

  • With a single run the Helfand-moment fits failed and printed 1.00; they now give the slope. With a single run there is no spread between runs to estimate errors from, so none are printed.

  • The slope of the Helfand moments (“Helfand Derivative”) was shifted by one sample, and its errors were in the wrong units. Both are corrected.

  • The analysis now uses the shared seamm_md_analysis library, also used by the Diffusivity step, instead of its own copy. The Helfand moments are computed much faster (by FFT): for a million samples, 0.4 s instead of about 2 minutes.

  • Thermal conductivities from the LAMMPS Heat Flux sub-step need lammps-step 2026.10.15 or later: earlier versions computed the heat flux wrongly.

  • The step no longer needs pandas or statsmodels, and its tests run with uv in CI using the package’s declared dependencies.

2026.3.1 – Internal: switching from deprecated library pkg_resources to importlib

2024.6.3 – Bugfix: handling of options for subflowchart

  • Fixed a bug where the options for the subflowchart were not being parsed correctly.

2024.3.22 – Updated for new scheme for running background tasks.

2024.1.5 – Bugfix: Thermal conductivity

  • If the Helfand moments fit in the thermal conductivity step failed it stopped the entire job. This is fixed, as well as some of the underlying causes for convergence issues.

2023.5.29 – Converged with general approach for trajectory analysis

2023.5.6 – Bugfix

  • Fixed an error handling Nan’s and Inf’s that caused a crash

  • Added the predictions from the derivatives of the Helfand moments to the output to give a better feel for the quality of the results.

2023.5.5 – Improved analysis

  • Considerable improvements to the analysis, results now seem solid

  • Fixed issues with fitting the linear portion of the Helfand moments

  • Added plot of the slope from the Helfand moments, which is similar to the Green-Kubo integral.

  • Cleaned up both the output and graphs.

2023.4.24 – Initial working version

  • Initial tests seem to work but needs more thorough testing.

  • Needs documentation!

2023.4.18 (2023-04-18)

  • Plug-in created using the SEAMM plug-in cookiecutter.