Hybrid Behaviour of Markov Population Models

We investigate the behaviour of population models written in Stochastic Concurrent Constraint Programming (sCCP), a stochastic extension of Concurrent Constraint Programming. In particular, we focus on models from which we can define a semantics of sCCP both in terms of Continuous Time Markov Chains (CTMC) and in terms of Stochastic Hybrid Systems, in which some populations are approximated continuously, while others are kept discrete. We will prove the correctness of the hybrid semantics from the point of view of the limiting behaviour of a sequence of models for increasing population size. More specifically, we prove that, under suitable regularity conditions, the sequence of CTMC constructed from sCCP programs for increasing population size converges to the hybrid system constructed by means of the hybrid semantics. We investigate in particular what happens for sCCP models in which some transitions are guarded by boolean predicates or in the presence of instantaneous transitions.

💡 Research Summary

The paper investigates stochastic population models expressed in Stochastic Concurrent Constraint Programming (sCCP), a language that extends Concurrent Constraint Programming with probabilistic transitions. The authors focus on two complementary semantics for such models: the classical Continuous‑Time Markov Chain (CTMC) semantics, where every transition is treated as a discrete jump, and a hybrid semantics that treats some populations continuously while keeping others discrete. The central contribution is a rigorous convergence theorem showing that, under appropriate regularity conditions, a sequence of CTMCs derived from sCCP programs with increasing population size converges to the hybrid system defined by the hybrid semantics.

The work begins by motivating the need for hybrid approximations. In large‑scale biological or epidemiological systems, a full CTMC quickly becomes intractable because the state space grows combinatorially with the number of individuals. Classical fluid or mean‑field limits replace the whole system by deterministic differential equations, but this discards rare events that may be crucial (e.g., the appearance of a mutant strain). The hybrid approach retains a discrete description for small or critical sub‑populations while approximating the bulk with continuous flows, thereby preserving both computational efficiency and stochastic fidelity where it matters.

The authors formalize sCCP syntax: variables represent population counts, guards are Boolean predicates over these variables, and actions specify increments/decrements together with a rate function. In the CTMC semantics each guarded action yields a Poisson jump with intensity given by the rate function evaluated at the current state. For the hybrid semantics the variable set is partitioned into “large” and “small” components. Large variables are scaled by the population size N and, in the limit N→∞, evolve according to an ordinary differential equation (the fluid limit). Small variables remain discrete and follow the same jump rules as in the CTMC, but their rates may depend on the continuous variables.

The main theoretical result is proved in three steps. First, the authors establish a law of large numbers for the scaled large variables, showing almost‑sure convergence to the solution of the fluid ODE under a Lipschitz condition on the rate functions. Second, they demonstrate tightness of the joint process (continuous part, discrete part) and identify any limit point as a solution of a piecewise‑deterministic Markov process (PDMP), which is precisely the hybrid system. Third, they handle complications introduced by guarded transitions and instantaneous transitions. Guarded transitions create switching surfaces in the continuous state space; near these surfaces the rate functions can change abruptly. By assuming that guards are regular (their zero‑level sets are smooth manifolds) and that the vector field is transversal to these manifolds, the authors prove that the hybrid limit respects the intended switching logic. Instantaneous transitions, which fire without delay, are treated by imposing a priority ordering and a “no‑time‑delay” condition that guarantees that the hybrid trajectory does not accumulate Zeno‑type behaviors. Under these additional assumptions the convergence theorem extends to models containing both guarded and instantaneous transitions.

To illustrate the theory, two case studies are presented. The first is an epidemic model where the majority of the population (susceptible, infected, recovered) is treated continuously, while a small class of superspreaders and a death compartment remain discrete. Simulations show that the hybrid model reproduces the full CTMC distribution for outbreak size and timing while reducing computational cost by roughly a factor of thirty for a population of one million. The second case study concerns a biochemical signaling network with a large pool of a protein and a few transcription factors that appear in low copy numbers. The hybrid approach accurately captures the stochastic timing of transcription factor activation (a rare event) and the deterministic evolution of the bulk protein concentration, matching the full stochastic simulation results.

The discussion acknowledges limitations. The convergence proof requires global Lipschitz continuity of rate functions and smoothness of guard surfaces, which may not hold in systems with strong nonlinearities or abrupt environmental changes. The handling of instantaneous transitions relies on an explicit priority scheme that must be supplied by the modeler, adding a layer of design complexity. Future work is suggested on relaxing these regularity conditions, extending the framework to multi‑scale time dynamics, and developing automated tools for guard analysis and hybrid partitioning.

In summary, the paper provides a mathematically sound foundation for hybrid semantics of sCCP population models, demonstrating that hybrid systems can serve as accurate limits of CTMCs for large populations while preserving essential discrete stochastic effects. This bridges the gap between fully stochastic CTMC analysis and deterministic fluid approximations, offering a powerful methodology for researchers in computational biology, epidemiology, and network science who need both scalability and fidelity.