Skip to content

Software Verification

This page collects the software-facing benchmark reports used to verify voids.

The intent is to answer questions such as:

  • does voids reproduce known manufactured or benchmark behavior?
  • how closely does an extracted-network voids workflow track an independent reference discretization on the same geometry?
  • which discrepancies are likely numerical bugs, and which are expected model differences?

Verification is narrower than validation. It does not claim universal physical truth; it tests whether the implementation behaves consistently against a software or numerical reference under controlled assumptions.

Current studies:

  • OpenPNM extracted-network cross-check This report verifies that voids agrees with OpenPNM in the machine-precision regime when both solve the same extracted network with the same throat conductances and the same pressure boundary conditions.

  • External reference CNM benchmark This report compares voids against a fixed external reference dataset. It now cleanly separates same-network imported-CNM parity from the larger image-to-network extraction gap.

  • XLB direct-image permeability benchmark This report includes the implemented LBM formulation, the permeability mapping used in voids, the shared pressure-drop coupling used between PNM and XLB, and a steady Stokes-limit interpretation of the same XLB operator.

  • DRP-443 fracture-network verification overview This report benchmarks DRP-443 against paper values reported from an LBM workflow, so it is treated as numerical-reference verification.

  • DRP-10 Estaillades verification overview This report benchmarks DRP-10 against paper values reported from an OpenFOAM workflow, so it is treated as numerical-reference verification.

The current studies answer different questions:

Reference Geometry seen by reference Main question Expected agreement
OpenPNM Same extracted network as voids Are export/import, BC handling, and solver assembly consistent? Machine precision
External reference CNM workflow Saved external CNM plus original voxel image Does voids match the saved single-phase solve on the same CNM, and how large is the remaining extraction gap from the image workflow? Near machine precision on imported CNM; moderate mismatch for image re-extraction
XLB Original voxel image How different is extracted-network PNM from a direct-image voxel-scale reference? Larger morphology-dependent mismatch is expected
DRP-443 paper reference Published LBM simulation outputs Does the current workflow reproduce paper-scale directional permeability trends on fractured-media volumes? Moderate mismatch is expected
DRP-10 paper reference Published OpenFOAM simulation outputs Does the current workflow reproduce paper porosity/permeability on the Estaillades volume? Moderate mismatch is expected