• Accepted Paper

Computational framework for structure-property landscape mapping in binary porous media via analytical correlation functions and implicit Krylov solvers

Marina V. Karsanina, Nickolay M. Evstigneev, Aleksei M. Cherkasov, Reza Azin, and Kirill M. Gerke

Phys. Rev. E - Accepted 9 November, 2026

DOI: https://doi.org/10.1103/3rwd-qwqc

Abstract

Designing materials with prescribed physical properties remains a central challenge in computational materials science. We present a systematic framework for exploring the parameter space of analytical two-point correlation functions (S2) and mapping the resulting structure-property relationships for binary porous media. As the first and most tractable realization of this framework, the S2 autocorrelation function is parameterized as a three-component superposition of exponential, damped oscillatory, and quadratic decay basis functions, controlled by three scalar parameters – a deliberate choice that keeps the parameter space well-defined and the results interpretable, while already capturing a wide diversity of structural morphologies. For each parameter combination, a three-dimensional binary structure of 5123 voxels is generated via simulated annealing reconstruction, and its full permeability tensor is computed using a GPU-accelerated implicit Krylov subspace Stokes solver. Applying this pipeline across 168 parameter combinations and nine porosity values between 10% and 50%, we produce a library of 1512 characterized structures with 92% of the samples spanning three orders of magnitude, extending to eight when the near-percolation outliers are included. The resulting structure–property maps reveal that, above φ≈0.25, the apparent dependence of permeability on the correlation function parameters is almost entirely an effect of specific surface area: in the dimensionless form Ks2/φ3 the parameter dependence collapses to within a few per cent, while below that porosity it survives and is governed by pore-space connectivity. The implied Kozeny constant is therefore not constant, varying by more than two orders of magnitude over the porosity range studied. We further analyze the realizability of the sampled correlation functions using energy-based, Gaussian random field, and negative-S2 criteria, and find that the Gaussian field criterion is the most conservative while the energy-based criterion best predicts successful reconstruction. This is due to the fact that reconstruction with limited correlation directions compensates for non-realizability by anisotropy. Crucially, the proposed framework does not optimize toward a single target – it maps the accessible structure-property landscape, making it possible to select or further refine any structure for a desired application. The advent of next-generation implicit solvers means this approach is now tractable for permeability and, equally, for any physical property that can be computed efficiently – opening a systematic, data-rich path toward computational material design that scales naturally to richer structural descriptors and more complex material classes.

Export citation

Export citation

Choose format for download:

Download Citation

If the author has provided any supplemental materials with this article they will be available upon publication of the version of record.

Sign In to Your Journals Account

Filter

Filter

Article Lookup

Enter a citation