Agentic transport: mobilities, resistivity and emissivity vs experiment#

PAOFLOW.agentic.transport compares PAOFLOW’s electron-phonon transport with experimental data: the phonon-limited carrier transport of semiconductors (transport/mobility.py of an electron-phonon system folder) and the transport of metals from the transport Eliashberg function (transport/resistivity.py), plus the bridge that feeds the same scattering channels, with impurity and residual terms, to PAOFLOW.transport(scattering_channels=...). The data toolkit (canonical datasets with provenance, units, PDF/raster ingestion, digitising, metrics, publication plots) is the one of thermo-optics, PAOFLOW.agentic.thermo_optics.

The agent playbook, lessons learned, prompts and templates live outside the package, in agentic/transport/.

Modules#

module

role

mobility

MobilityModel: the adapter for transport/<tag>/mobility.dat and carrier_scattering.npz (elphon.carrier_scattering): mobility(T, carrier, kind, interpolate) (SERTA / MRTA drift mobilities, ln μ linear in ln T off the grid), carrier_density (ln n linear in 1/T), conductivity, seebeck, power_law_exponent, bracket, metadata (grids, smearing, doping, gap, scissor mode, electrons provenance, screening, window states), describe(); from_system(folder, output=)

resistivity

ResistivityModel: the adapter for transport/<tag>/resistivity.dat (total and per-channel ρ in µΩ cm, channels named in the header), emissivity.npz (extended-Drude ε(ω, T), hemispherical / directional / total emissivity, Planck coverage) and the eliashberg.npz scalars (λ, λ_tr, ω_p, ω_log, Tc): resistivity(T, channel, rho0_uohm_cm, interpolate, law) (Matthiessen: a residual resistivity is added after the fact), slope, residual_resistivity_ratio, emissivity_total, emissivity_spectrum, dielectric, describe(); from_system

baselines

bloch_gruneisen, fit_bloch_gruneisen (ρ₀, A, θ; a coarse θ scan + bounded refinement, θ or ρ₀ can be fixed), fit_power_law, fit_linear, matthiessen_offset (the residual that reconciles a phonon-limited model with a sample: rms before/after, slope ratio), matthiessen_mobility

channels

metal_channels(eliashberg_npz, rho0_uohm_cm, b_ee_per_ev) (the list transport/resistivity.py builds: TransportSpectralChannel + ResidualResistivity + FermiLiquidRate), carrier_channels(scattering_rates_npz, kind, band_edges, extra) (EnergyResolvedRate aligned to band edges + extra channels), impurity_rate_from_mobility, impurity_channel (a ConstantRate reproducing a mobility)

PAOFLOW.agentic.thermo_optics.paoflow_transport remains as an alias of PAOFLOW.agentic.transport.mobility.

Design choices#

  • Adapters are thin and read-only and return the observable at a condition; interpolation laws are physical (power law for μ, activated for n, linear for ρ above θ_D/5, power law below) and every off-grid value is meant to be labelled.

  • Corrections are explicit arguments. The Hall factor, the residual resistivity and Matthiessen impurity mobilities are quoted next to the uncorrected comparison; the library offers them as diagnostics (matthiessen_offset, matthiessen_mobility), not as fitted models.

  • Baselines are trivial models with named parameters (Bloch-Grüneisen, power laws), so a deviation is always quoted against something.

  • The Boltzmann bridge reuses the run outputs: no re-derivation of the vertex is needed to add impurity or residual scattering to σ, Seebeck and κ_e.

Tests#

tests/unit/agentic/transport/:

  • MobilityModel on a synthetic transport/output (power-law and activated interpolation, exponents, from_system, the alias);

  • ResistivityModel on a synthetic metal output (channel parsing, interpolation laws, residual addition, slope, RRR, emissivity spectra and totals, dielectric function, from_system);

  • Bloch-Grüneisen limits and the recovery of (ρ₀, A, θ) from a synthetic curve; power-law, linear and Matthiessen-offset fits; channel lists from synthetic eliashberg.npz and scattering_rates.npz files, impurity rates.