TaoELi

mxl-driver-tls

by @TaoELi in Tools
6
1
# Install this skill:
npx skills add TaoELi/MaxwellLink --skill "mxl-driver-tls"

Install specific skill from multi-skill repository

# Description

This skill should be used when users want to run the built-in TLS (two-level system) driver in socket or embedded mode and need parameter conventions and example commands.

# SKILL.md


name: mxl-driver-tls
description: This skill should be used when users want to run the built-in TLS (two-level system) driver in socket or embedded mode and need parameter conventions and example commands.


TLS driver (two-level system)

Use for lightweight benchmarks

  • Use the built-in TLS driver (tls) for fast regression and simple light–matter prototyping.

Configure embedded mode

  • Instantiate:
  • Molecule(driver="tls", driver_kwargs=dict(omega=..., mu12=..., orientation=..., pe_initial=..., checkpoint=..., restart=...))

Configure socket mode

  • Launch one process per molecule:
  • mxl_driver --model tls [--unix --address <name> | --address <host> --port <port>] --param "omega=..., mu12=..., orientation=..., pe_initial=..."
  • Keep the hub address/port/unixsocket consistent with the EM script.

Key parameters

  • omega (a.u.), mu12 (a.u.), orientation (0|1|2 for x/y/z), pe_initial (0–1), checkpoint, restart.

Notes

  • Prepare a coherent (superposition) initial TLS state by setting 0 < pe_initial < 1; this seeds emission into vacuum-like classical fields.
  • Read outputs from molecule.additional_data_history (keys include time_au, Pe, Pg, dipoles, energies).
  • Use post-processing guidance in skills/mxl-postprocess/SKILL.md.

# Supported AI Coding Agents

This skill is compatible with the SKILL.md standard and works with all major AI coding agents:

Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.