Coarse Graining¶

Coarse graining lets the sampler propose local multi-spin updates on predefined subgroups, rather than updating all spins at once. This keeps proposals tractable on larger systems.

Defining subgroups¶

Create a CoarseGraining object with a list of subgroups and (optionally) their selection probabilities. Each subgroup is a list of spin indices that are updated together.

Every spin must belong to at least one subgroup, and subgroups may overlap (a spin can appear in multiple subgroups).
subgroup_probs must sum to 1.
With the default repeated=True, all subgroups must be the same length. For subgroups of differing sizes, pass repeated=False.

from qemcmc.coarse_grain import CoarseGraining

n = 4
cg = CoarseGraining(
    n=n,
    subgroups=[[0, 1], [2, 3], [1, 3]],
    subgroup_probs=[0.3, 0.5, 0.2],
)

At each MCMC step a subgroup is sampled according to subgroup_probs (uniformly if omitted).

Using it in a proposal¶

Pass the object to QeProposal via coarse_graining; the circuit for each subgroup is built automatically.

from qemcmc.sampler import QeProposal

proposal = QeProposal(model, gamma=(0.3, 0.6), time=(1, 10), coarse_graining=cg)