How to sweep one parameter with checkpoint restart¶

Use this when you want to sweep one parameter (e.g. rocking frequency omega_b) across several values, chained via checkpoint restart so each segment after the first starts from an already-developed flow instead of cold. For why this is worth doing and how the restart ramp works, see Checkpoint restart and warm-start chains.

Tip

To sweep multiple parameters at once, or write configs in JSON instead of YAML, use Sweep any parameter combination instead.

1. Write a chain config¶

Copy config/chain_config.yaml and edit:

# ── Fixed params (same for all segments) ────────────────────────────────────
fidelity: 7
geometry: {a: 0.20, b: 0.09, n: 2.0}   # superellipse; n=2 → ellipse
fill_level: 0.4

# ── Timing ───────────────────────────────────────────────────────────────────
n_mix_cycles: 80         # rocking cycles before O₂ injection (segment 0)
n_transition_cycles: 10  # rocking cycles before O₂ re-injection (segments 1+)
t_buffer: 150            # non-dim kLa measurement window

# ── Sweep: one segment per value ─────────────────────────────────────────────
sweep:
  parameter: omega_b
  values: [3.14159, 6.28318]   # 0.5 Hz, 1.0 Hz

# ── Base motion (non-swept params, fixed across all segments) ─────────────────
motion:
  n_harmonics: 1
  theta_max:      [5.0, 0.0, 0.0]
  phi_angular:    [0.0, 0.0, 0.0]   # index 0 must stay 0.0
  omega_h:        6.28
  amplitude_h:    [0.02, 0.0, 0.0]
  phi_horizontal: [0.3, 0.0, 0.0]

videos: false          # set true to use BioReactor-video and render MP4s
submit: true
walltime: "04:00:00"

Supported sweep parameters:

Name	Maps to
`omega_b`	rocking angular frequency (rad/s)
`omega_h`	horizontal translation frequency (rad/s)
`theta_max_0` / `_1` / `_2`	`theta_max[0..2]`
`amplitude_h_0` / `_1` / `_2`	`amplitude_h[0..2]`
`phi_angular_1` / `_2`	`phi_angular[1..2]` (index 0 is always 0)
`phi_horizontal_0` / `_1` / `_2`	`phi_horizontal[0..2]`

2. Smoke-test before submitting long jobs¶

uv run python scripts/chain.py config/chain_config_smoke.yaml

Then check (see Your first sweep for what this looks like in full):

runs/<seg1_id>/logstats.dat starts at t > 0, not t=0
runs/<seg1_id>/results.json has finite kLa values

3. Submit for real¶

python scripts/chain.py config/chain_config.yaml

Prints one line per segment with run IDs and SLURM job IDs. SLURM --dependency=afterok ensures each segment starts only after the previous one succeeds.

4. Collect results¶

Each segment writes independently to runs/<run_id>/results.json — there's no aggregation script for Workflow B specifically; collect manually or use collect_results.py (see Scripts reference).