Answers are grounded in this paper and cite their sources.
An explicit equilibrium formula is only a starting point when particle counts are restricted to a rare reaction class or read through noisy detectors. This paper gives constructive algorithms for evaluating and controlling conditioned Poisson models, with the numerical precision and output representation included in their resource costs. For any one fixed Presburger conditioning family—an integer-arithmetic description allowing linear constraints, congruences and logical combinations—the forward theorem evaluates log probabilities and constructs approximate conditional samples in polynomial time in the complete binary input, including accuracy digits. Conditioning events may be arbitrarily rare and supports may be infinite. Every possible sampled output satisfies the exact constraint. For classes of a fixed additive congruence, the inverse theorem constructs ordinary positive rational controls within a prescribed linear exponential family. The same controls achieve a requested absolute mean error and a requested relative-entropy error from the canonical information projection. Signed or rank-deficient statistics, unbounded unobserved populations and boundary targets are included. The relative-entropy guarantee is directed from the canonical target law to the controlled full-class law; it does not synthesize arbitrary distributions with the same mean. The proof combines localized factorial-weight approximation and fixed-dimensional lattice summation with local mean witnesses, short projected lifts and quantitative suppression away from a supporting face. An exact marked-count representation extends inference and inverse mean design to missed detections, cross-talk, aggregated reports and Poisson background while preserving the specified detector parameters. Further results identify exact reaction components, compare a finite-reservoir realization with its intended law, and establish obstructions for growing networks and passive experimental acquisition. All preprocessing is charged, but constants and polynomial degrees depend on the fixed structure. The results concern supplied static count models. They do not imply practical running times, fast physical equilibration, experimental acquisition of an unknown model or general prediction of driven dynamics. Established reachability, rounding, summation, exponential-family and product-form machinery is credited separately.
No takes yet. Share an insight, caveat, or question.
Harjas Sandhu (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: