Cesar Augusto Vargas-García

SHSLib: A Lean 4 Formalization of Stochastic Hybrid Systems

← Back to Research

Project

SHSLib is a formally verified Lean 4 library formalizing Stochastic Hybrid Systems (SHS) theory — mathematical models describing discrete-mode, continuous-state processes that combine deterministic differential flow with random jumps.


What Are Stochastic Hybrid Systems

Many real systems — a cell that grows continuously but divides in sudden jumps, a gene switching between “on” and “off” transcriptional states — can’t be described by a single continuous differential equation or a single discrete-state Markov chain. They need both at once.

SHSLib formalizes the foundational theory of these systems as laid out by:

An SHS is defined by a 6-tuple $(n, Q, m, f, \phi, \lambda)$: state dimension $n$, discrete mode set $Q$, transition count $m$, continuous drift $f(q,x,t)$, reset maps $\phi_\ell(q,x,t)$, and jump intensities $\lambda_\ell(q,x,t)$.


The Extended Generator

At the core of the library is the extended generator $\mathcal{L}$ — an operator capturing how a system’s expected state evolves through both deterministic drift and stochastic jumps:

\[(\mathcal{L}\psi)(q,x,t) = \nabla_x \psi \cdot f + \frac{\partial \psi}{\partial t} + \sum_{\ell=1}^{m} \left[\psi(\phi_\ell) - \psi\right] \lambda_\ell\]

From this, SHSLib formally proves Dynkin’s Formula:

\[\mathbb{E}[\psi(\mathbf{z}(t))] = \psi(z_0) + \mathbb{E}\left[\int_{t_0}^t (\mathcal{L}\psi)(\mathbf{z}(s))\, ds\right]\]

which connects the generator to the expected evolution of any observable of the system — both a differential and an integral formulation are verified.


Library Structure


Verified Results

As a worked application, the library formalizes cell size regulation for cells with linear growth rate $\mu$ and division rate $k$:

\[\frac{dM_1}{dt} = \mu - \frac{k}{2}M_1 \quad \Rightarrow \quad \mathbb{E}[X^*] = \frac{2\mu}{k}\]
Result Formula Status
Variance $\text{Var}[X] = \frac{4}{3}(\mu/k)^2$ Verified
CV $1/\sqrt{3} \approx 0.577$ Verified
Skewness $6\sqrt{3}/7 \approx 1.485$ Verified
Characteristic function $C(\xi) = \prod_{j=0}^{\infty} (1-i(\mu/k)2^{-j}\xi)^{-1}$ Verified
Density series $\rho^*(x) = \sum_{n=0}^{\infty} c_n e^{-(k/\mu)2^n x}$ Verified
Upwind scheme stability CFL condition Verified
MLE for division rate $\hat{k} = N/T$ maximizes likelihood Verified

Installation

lake build

Requires Lean 4 and Mathlib (version pinned in lakefile.toml).


Applications

SHSLib’s cell size theory formalizes the mathematical framework behind our own experimental and computational work on single-cell size homeostasis:

It also forms the preliminary work for a proposed two-year extension into gene expression dynamics and theoretical systems agriculture — see the SHS Formalization Grant Proposal.



← Back to Research View on GitHub Extension Proposal