The new work, published in the journal SoftwareX by Youssouf Abdelhafiz, Abdenour Amokrane, Gerald Debenest, and Serge Bourbigot, does not introduce a new physical model of pyrolysis. Instead, it tackles a problem that plagues many mature scientific codes: the mathematics inside has become the bottleneck. The original Gpyro relied on fully implicit, iterative solvers that were numerically stable but computationally punishing. Its three-dimensional thermal solver was unstable in demanding cases, documentation was incomplete, and systematic verification was largely absent. For researchers trying to couple solid-phase pyrolysis with the Fire Dynamics Simulator (FDS), Gpyro routinely dominated the runtime of otherwise tractable multiphysics simulations.

To understand why the old approach was so expensive, it helps to look at what Gpyro actually computes. Each material is represented as a collection of condensed-phase species, each with its own temperature-dependent heat capacity, density, anisotropic thermal conductivity, emissivity, and absorption coefficient. As chemical reactions proceed, the code tracks the conservation of total mass and of each species, together with an energy balance that accounts for conduction through the anisotropic solid, chemical heat release, in-depth radiation absorption, and gas-solid thermal exchange. At the boundaries, the surface may exchange heat with the environment through convection, external radiative input, and radiative losses following the Stefan-Boltzmann law. Crucially, the computational mesh must deform as the material shrinks or swells, which couples the local mass, density, and cell volume in a numerically awkward way.

In the legacy version, that coupling was handled with a fully implicit iterative scheme. Stability was excellent, but each time step could demand many iterations, and the cost ballooned in the final stages of degradation of non-charring materials, when the cell volume collapses toward zero. The new version replaces this with an explicit local update for mass and species conservation, performed independently in each computational cell. Because explicit schemes can go wrong when cells become vanishingly small, the team added a stabilization mechanism: any cell that falls below a user-defined volume threshold is declared fully consumed and switched off. A correction procedure catches any predicted negative species masses, scaling destruction rates downward and adjusting production rates so that the local mass balance stays physically sensible, with automatic time-step reduction as a last resort.

The thermal solver received an equally consequential overhaul. The revised code employs an Alternating Direction Implicit (ADI) scheme, a classical strategy that the authors stress is not new in itself; the contribution lies in implementing it efficiently inside Gpyro’s unusual environment of strongly anisotropic meshes and evolving reactive solids. At each time step, the solver sweeps through the grid one direction at a time, in the order Z, then X, then Y, treating the heat fluxes along the active direction implicitly and the others explicitly. Each sweep reduces to a tridiagonal linear system that is solved rapidly with the Thomas algorithm. Because degradation in typical three-dimensional pyrolysis problems propagates mainly along the vertical Z-axis, where users naturally refine the mesh most finely, locking the implicit direction to Z pays off handsomely. In one-dimensional configurations, the solver remains fully implicit and therefore unconditionally stable.

Beyond the new algorithms, the entire codebase was refactored for modularity, readability, and maintainability, and parallelized with OpenMP so it can exploit multicore processors on ordinary workstations. The software is distributed under the GNU GPL 3.0 license on GitHub as Gpyro V2.0, complete with a newly written verification guide, updated user and technical documentation, and a suite of verification and validation cases. Three executables ship with the package: the standalone pyrolysis engine, an inverse modeling tool called Gpyro_propest that estimates material properties from experimental data, and a coupled version under active development for integration with FDS. The inverse tool, previously feasible only on computing clusters, is among the biggest beneficiaries of the speed gains, because parameter estimation requires running the forward model hundreds or thousands of times.

Verifying that all this speed did not corrupt the physics was a central concern, and the team leaned on a benchmark designed for exactly this purpose: a fictitious but physically coherent charring material undergoing two first-order reactions, heated on one face by a constant 50 kilowatts per square meter in a cone-calorimeter-style configuration. Comparisons of the mass loss rate over time show excellent agreement between both versions of Gpyro and the established codes ThermaKin and FDS. Dedicated convergence studies, varying grid spacing from 1 millimeter down to 0.0001 millimeters and time steps from 10 seconds down to half a millisecond, confirm that the new solvers converge reliably toward the same solution as the discretization is refined.

The performance numbers are striking. For the reference one-dimensional case at the minimal converged configuration, the new code is roughly 39 times faster than the legacy version; at a resolution of one micron in the degradation direction, the speed-up reaches 119 times. The gains grow with dimensionality: a two-dimensional case that took the old solver just over nine minutes now runs in 36 seconds, while a three-dimensional case that consumed 13.5 hours finishes in 25 minutes. A validation simulation of anaerobic PMMA pyrolysis, replicating the NIST Gasification Apparatus experiment from the MaCFP benchmark at the same 50 kilowatts per square meter heat flux, reproduced the measured mass loss rate and turned in a 140-fold speed-up. Notably, PMMA is a non-charring material, precisely the regime where the legacy solver struggled most as cells thinned toward zero.

To demonstrate consistency across dimensions, the researchers extended the reference case laterally to a 10 by 10 centimeter domain with adiabatic side boundaries, which should in theory reproduce the one-dimensional answer exactly. It did, in one, two, and three dimensions alike. Enabling convective losses on the lateral faces introduced the expected deviation, and the code visualized the degradation front sweeping through the solid in three dimensions, resolving the spatial distribution of the reaction rate rather than merely a global mass-loss curve. In every comparison, the new and legacy versions produced identical physical results; the only meaningful difference was how long the user had to wait.

The practical consequences extend well faster-than-real-time benchmark bragging rights. Coupled Gpyro-FDS simulations, which treat the feedback between a burning solid and the surrounding gas phase, previously required nearly 24 hours for a reference three-dimensional case; the updated coupling completes it in about one hour, removing Gpyro as the computational choke point in multiphysics fire modeling. The authors say a finalized coupled version will be released to the fire community, alongside an expanding verification and validation database and a revised user guide. The refactored architecture also opens the door to future modules such as mechanical deformation models of degrading solids. For a field where fire safety engineering increasingly depends on trustworthy, accessible simulation, a two-decade-old model reborn as a fast, well-documented, openly licensed tool may prove to be one of the more quietly impactful developments of the year.

Subject of Research: Numerical solver improvements for solid-phase pyrolysis modeling in the Gpyro fire science code

Article Title: Enhancement of the numerical solvers of the pyrolysis code Gpyro

Article References: Abdelhafiz, Y., Amokrane, A., Debenest, G., & Bourbigot, S. (2026). Enhancement of the numerical solvers of the pyrolysis code Gpyro. SoftwareX, 36, Article 103058. https://doi.org/10.1016/j.softx.2026.103058

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103058

Keywords: Gpyro, pyrolysis, fire modeling, numerical solvers, ADI scheme, OpenMP, FDS, cone calorimeter, PMMA, verification and validation, computational performance, open-source software