Parallel Execution¶

mcising offers three execution modes for temperature scans. The right choice depends on your use case.

The three modes¶

Cooldown (default)¶

Temperatures are processed sequentially from high to low. The spin configuration carries forward — each temperature starts from the previous one's final state.

from mcising import Simulation, SimulationConfig, LatticeConfig, ExecutionMode

config = SimulationConfig(
    lattice=LatticeConfig(size=32),
    temperatures=(3.0, 2.5, 2.269, 2.0, 1.5),
    n_sweeps=1000,
    mode=ExecutionMode.COOLDOWN,  # this is the default
)

results = Simulation(config).run()

Best for: avoiding metastable states at low temperature. Single-threaded.

Independent¶

Each temperature runs from random initialization on a separate CPU core. No communication between temperatures.

config = SimulationConfig(
    lattice=LatticeConfig(size=32),
    temperatures=(3.0, 2.5, 2.269, 2.0, 1.5),
    n_sweeps=1000,
    mode=ExecutionMode.INDEPENDENT,
)

results = Simulation(config).run()

Best for: fast scans with many temperatures. Uses all CPU cores via Rayon.

Parallel Tempering¶

Like independent mode, but replicas periodically swap spin configurations between adjacent temperatures. High-temperature replicas explore freely and pass good configurations down to low-temperature replicas.

config = SimulationConfig(
    lattice=LatticeConfig(size=32),
    temperatures=(3.0, 2.5, 2.269, 2.0, 1.5),
    n_sweeps=1000,
    mode=ExecutionMode.PARALLEL_TEMPERING,
    swap_interval=1,  # attempt swap every sweep (default)
)

results = Simulation(config).run()

Best for: frustrated systems, spin glasses, any system where Metropolis gets stuck in local minima.

When to use which¶

Mode	Parallelism	Sampling quality	Use case
Cooldown	None (1 core)	Good (warm start)	Standard simulations
Independent	Full (all cores)	OK (cold start per T)	Fast scans, error bars
Parallel Tempering	Full + swap sync	Best (replica exchange)	Frustrated/glassy systems

Speed comparison¶

Independent mode runs one temperature per Rayon thread; parallel tempering does the same and adds the replica-exchange synchronisation. The cooldown mode is the single-threaded reference.

Mode	Threads	Wall time	Speed-up vs cooldown
Cooldown	1	3.01 s	1.0×
Independent	1	3.03 s	1.0×
Independent	2	1.62 s	1.9×
Independent	4	0.84 s	3.6×
Independent	8	0.57 s	5.3×
Independent	10	0.54 s	5.6×
Parallel tempering	10	1.47 s	2.0×

Metropolis, 128×128 square lattice, 20 temperatures from 3.5 to 1.5, 500 thermalization + 2,000 production sweeps per temperature (measured every 10), Apple M4 (10 cores: 4 performance + 6 efficiency); medians of 3 runs. The independent and parallel-tempering rows each run in a fresh process with RAYON_NUM_THREADS set to the thread count; the cooldown mode is single-threaded by construction.

Independent mode gives the wall-clock speed-up; it grows with the number of performance cores until the temperatures run out. Parallel tempering pays for the swap synchronisation on top, in exchange for far better sampling of the low-temperature replicas — which is what it is for.

RNG seeding¶

Each mode handles reproducibility differently:

Cooldown: single seed, single RNG stream
Independent: seed + temperature_index per replica — deterministic and independent. The index is the temperature's position in the configured scan, and it sticks to its temperature even when a checkpointed run resumes with some temperatures already done — a resumed scan reproduces the uninterrupted run's streams exactly.
Parallel Tempering: per-replica seeds as in independent mode, plus a separate swap-decision RNG

All modes are fully deterministic: same seed = same results.

Checkpointing¶

All three modes work with checkpoint_run. Granularity differs: cooldown saves after each temperature, independent mode saves the whole batch when it returns (resume runs only the missing temperatures), and parallel tempering is all-or-nothing because the replicas form one coupled ensemble. See Saving results.

CLI¶

# Default (cooldown)
mcising run -L 32 -T 3.0 -T 2.269 -T 1.5

# Independent parallel
mcising run -L 32 --mode independent -T 3.0 -T 2.269 -T 1.5

# Parallel tempering
mcising run -L 32 --mode parallel_tempering -T 3.0 -T 2.269 -T 1.5