System Level Numerical Analysis of a Monte Carlo Simulation of the E. Coli Chemotaxis

Over the past few years it has been demonstrated that “coarse timesteppers” establish a link between traditional numerical analysis and microscopic/ stochastic simulation. The underlying assumption of the associated lift-run-restrict-estimate procedure is that macroscopic models exist and close in terms of a few governing moments of microscopically evolving distributions, but they are unavailable in closed form. This leads to a system identification based computational approach that sidesteps the necessity of deriving explicit closures. Two-level codes are constructed; the outer code performs macroscopic, continuum level numerical tasks, while the inner code estimates -through appropriately initialized bursts of microscopic simulation- the quantities required for continuum numerics. Such quantities include residuals, time derivatives, and the action of coarse slow Jacobians. We demonstrate how these coarse timesteppers can be applied to perform equation-free computations of a kinetic Monte Carlo simulation of E. coli chemotaxis. Coarse-grained contraction mappings, system level stability analysis as well as acceleration of the direct simulation, are enabled through this computational multiscale enabling technology.

💡 Research Summary

The paper introduces a systematic “equation‑free” computational framework that bridges microscopic stochastic simulations with macroscopic numerical analysis through the use of coarse timesteppers. The central premise is that, although a closed-form macroscopic model may exist for a complex system, it is often unavailable or intractable. Instead of deriving explicit closures, the authors employ a lift‑run‑restrict‑estimate (LRRE) procedure: macroscopic variables are “lifted” to generate consistent microscopic initial conditions; short bursts of kinetic Monte Carlo (KMC) simulations are then “run”; the resulting microscopic data are “restricted” back to a set of low‑order moments (e.g., cell density profile, mean velocity); finally, these moments are used to “estimate” the quantities required for continuum‑level algorithms such as residuals, time derivatives, and Jacobian‑vector products.

Two levels of code are constructed. The inner code is a conventional KMC engine that simulates individual E. coli cells moving up a chemoattractant gradient, tracking both spatial positions and internal biochemical states (receptor occupancy, methylation level). The outer code treats the coarse‑grained quantities as if they obeyed deterministic differential equations and applies standard numerical tools: Newton‑Krylov solvers for fixed‑point (steady‑state) computation, power‑method based eigenvalue estimation for stability analysis, and projective integration for accelerating long‑time dynamics. Crucially, the Jacobian is not formed analytically; it is approximated by perturbing macroscopic variables, rerunning short KMC bursts, and measuring the resulting change in the restricted moments, thus providing an efficient estimate of Jacobian‑vector products.

Applying this methodology to the chemotaxis problem yields several notable outcomes. First, the coarse‑time‑stepper dramatically reduces computational cost: a direct KMC simulation that would require thousands of time units can be replaced by a sequence of short bursts and projective extrapolations, achieving an order‑of‑magnitude speed‑up while preserving statistical accuracy. Second, the Newton‑Krylov iteration converges rapidly to a steady‑state cell density distribution, and the eigenvalue analysis confirms that this fixed point is linearly stable (all eigenvalues have negative real parts). Third, the comparison between the full microscopic distribution and the low‑order moment reconstruction shows excellent agreement, validating the assumption that the system’s dynamics evolve on a slow manifold parameterized by a few moments.

Beyond the specific chemotaxis example, the authors argue that the coarse‑timestepping approach is broadly applicable to any multiscale system where microscopic simulators are available but macroscopic equations are not. By systematically extracting the necessary macroscopic information from ensembles of short microscopic runs, one can perform system‑level tasks—bifurcation analysis, optimal control, uncertainty quantification—without ever writing down the governing PDEs. The paper thus demonstrates a practical pathway for integrating stochastic, particle‑based models with the powerful toolbox of continuum numerical analysis, opening new avenues for efficient, high‑level investigation of complex biological, chemical, and physical systems.