Skip to content

Usage icon Usage

This page is your practical starting point for running DYNAMITE: CLI flags, inputs/outputs, and common workflows.

For the algorithmic background and implementation concepts, continue with:

  • Concepts → Algorithm
  • Concepts → Interpolation grids
  • Concepts → Time integration
  • Concepts → EOMs and observables

Entry point: ./RG-Evo. Show all options with -h.

Model and parameters

  • -p, -q (integers): model orders (e.g., spherical mixed p-spin). Use p>=2; common: p=3, q=4.
  • -l, --lambda (float): coupling strength (dimensionless). Typical 0–1; study transitions by sweeping.
  • -T, --T0 (float|inf): initial temperature; use inf for zero-noise quenches.
  • -G, --Gamma (float): final temperature.

Note: At present, the mixed spherical p-spin equations are hardcoded as in the accompanying paper; other models are not yet implemented. Modular equations of motion will be subject of a future release.

Discretization and accuracy

  • -L grid length: selects Grid_data/<L>/ (available: 512/1024/2048). Larger L → higher accuracy and cost.
  • -m max iterations and -t max physical time bound the run. For exploratory runs, start with -m 1e4.
  • -d minimum time step and -e error tolerance control adaptivity (RK54 default with auto-switch to SSPRK104; SERK2 trials optional). Smaller -e → better accuracy.

Execution controls

  • -g, --gpu boolean: enable GPU kernels (default true when available); set false for CPU-only or reproducibility.
  • -A, --async-export boolean: asynchronous I/O to avoid blocking the integrator (default true).
  • -s save outputs (default true) and -o output directory root.
  • -D debug logging; -v print build/version; -I allow resume across incompatible versions (use with care).

Interpolation mode: - -R, --log-response-interp boolean: interpolate QR and dQR in log space (default false). If enabled, the code interpolates f=log(QR) and g=dQR/QR and exponentiates back; it automatically falls back to linear when any stencil QR<=0. QK/dQK remain linear.

Sparsification: - -w, --sparsify-sweeps INT: number of sparsify sweeps per maintenance pass. - -1 (default): auto — CPU: 1 sweep; GPU: 1 sweep normally, 2 sweeps if GPU memory usage > 50%. - 0: disable sparsification. - >0: fixed number of sweeps.

Inputs (grids)

Interpolation weights/indices are loaded from Grid_data/<L>/: - theta.dat, phi1.dat, phi2.dat, int.dat - posA1y.dat, posA2y.dat, posB2y.dat - indsA1y.dat, indsA2y.dat, indsB2y.dat - weightsA1y.dat, weightsA2y.dat, weightsB2y.dat

Choose the largest L that fits memory/time for your study; verify convergence of observables with L.

Generating grids and interpolation metadata

If Grid_data/<L>/ is missing (or you want to regenerate with different settings), use the built-in grid generator.

For the full ./RG-Evo grid flag reference, outputs, and validation workflow, see How-to → Generate new grids.

If you just want the default grids for a first run, see Tutorial → Generate grids.

Outputs and observables

Each run writes a self-contained output directory that contains:

  • the full state (preferably data.h5, otherwise data.bin),
  • a provenance record (params.txt),
  • and lightweight text summaries for quick plotting.

For a practical guide to locating files and doing sanity checks, see Tutorial → Reading outputs.

For the on-disk formats (including the compressed snapshot files), see Concepts → Architecture.

Resume is automatic if a compatible checkpoint is found; see Concepts → Version compatibility.

Typical runs

  • Short aging run (GPU): 
    ./RG-Evo -L 512 -l 0.5 -m 1e4 -D false
  • CPU-only reproducibility: 
    ./RG-Evo --gpu false -L 512 -l 0.5 -m 5e3
  • Parameter sweep in λ around a transition: 
    for lam in 0.4 0.5 0.6; do ./RG-Evo -L 1024 -l "$lam" -m 2e4 -D false; done