INTEGRATE OUT · RETAIN INTERACTIONS · FOLLOW THE SCALE
Renormalization, with the Hamiltonian carried through.
The coarse weight is the sum over eliminated spins. Its logarithm includes the constant and every generated interaction. Compare exact decimation, explicit interaction truncation, and a trained full-action approximation.
64 → 32 → 16 → 8 → 4 → 2 → 1 spins; spacing doubles each step.
Loading checked local calculations…
Train the effective action, then check physical errors.
The 3-qubit hybrid predicts all 32 coarse logweights, including their constant. The exact teacher integrates out the two black spins of the genus-two dessin. The varied input family has three tied coupling parameters; its 31 nonconstant output interactions are not 31 independent input parameters.
Executed five-qubit generation, dephasing and sampling code
On these held-out data, the matched-size classical feature readout has lower action MSE, KL and partition error. This experiment establishes neither quantum advantage nor a uniform error guarantee. OOD errors are reported separately. Steps after the learned first action use exact marginalization.
Definitions and reproduction
Full conventions and commands · Exact RG definitions · Actual PennyLane training · Kernel-checked finite RG identities · Complete computed cases and model
The finite dessin Hamiltonian is a physical model chosen on its clean incidence graph. It is not a faithful representation of the infinite gentle algebra or its completed surface order. Finite quantum thermal elimination preserves partition functions and retained equal-time observables; the effective Hamiltonian depends on β and is generally nonlocal. Full real-time dynamics, continuum critical behavior, material-specific predictions, and the GRH arithmetic comparison require additional results.