An augmented finite-element method for nonlinear poroelasticity matched the convergence predicted by its analysis in a smooth two-dimensional test and was reported as more accurate than a non-augmented version in a Cook’s membrane benchmark. The augmented method was reported as four to 11 times more accurate; on the finest mesh, its deflection was within 0.0002 of the Taylor-Hood result, while the non-augmented error remained around 0.01. These are numerical results from selected test problems, not physical validation.
A formulation built around several fields
The study analyzes the nonlinear Biot equations in two and three dimensions, using poroelastic stress, displacement, pressure and rotation as linked variables. It is a mixed formulation because those fields are treated as a connected set of unknowns. The proposed augmentation adds residual-type terms, weighted by stabilization parameters, to target coercivity of the bilinear form over the stress, displacement and pressure unknowns. In this setting, coercivity is the mathematical control condition the added terms are intended to support.
What the theory actually guarantees
On the theoretical side, the paper uses a Banach fixed-point argument. A fixed point is a solution that remains unchanged by the update used in the argument. The authors report existence and uniqueness for the continuous nonlinear problem under a smallness assumption on the data. At the finite-element level, the nonlinear scheme is likewise reported to have a unique fixed point and a unique discrete solution, with the conclusion depending on the theorem hypotheses.
The conforming finite-element spaces approximate the poroelastic tensor, displacement, pressure and rotation. The selected stress and rotation subspaces satisfy the discrete inf-sup condition, a compatibility requirement used to keep the mixed discretization mathematically stable.
With the stated regularity hypotheses, the paper derives an a priori error estimate whose constant is independent of mesh size. The practical meaning is that the theoretical bound is not loosened merely by changing the mesh, although its rate depends on the regularity and approximation order assumed for the exact fields. The paper also reports a residual-based a posteriori estimator, an error check built from the calculation’s residuals, as reliable: it bounds the combined stress, displacement, pressure and rotation error up to a constant. Under piecewise-polynomial permeability and boundary data, the estimator is reported as efficient up to discretization error and source-data oscillation.
Numerical checks
The numerical studies used FEniCS, Newton-Raphson solves with a residual tolerance of 10 to the power of minus 7, and sparse LU factorization through MUMPS. The smooth two-dimensional convergence test used six uniform mesh-refinement levels.
In the smooth two-dimensional manufactured-solution test, all primary and mixed variables showed the optimal decay predicted by the theoretical error bounds. Estimator effectivity, the reported comparison between estimated and numerical error, remained almost constant across refinements. Because this is a manufactured-solution test rather than a physical test, the result checks convergence behavior rather than validating a physical system.
An adaptive refinement test on a singular domain produced a different kind of evidence. After the initial refinements, adaptive refinement restored approximately quadratic convergence, produced smaller errors with fewer degrees of freedom, and kept effectivity near 2.08. The nonlinear solve required at most three Newton iterations in the reported test.
The three-dimensional parameter study was broadly consistent with the theory in the tested regimes, but not without a caveat. All fields converged optimally except pressure in the low-permeability Set IV, where the behavior was slightly suboptimal. The authors did not expect the method to be robust for low-permeable media, so this case sets a clear limit on how far the reported numerical evidence can be generalized.
The benchmark comparison
The Cook’s membrane benchmark compared three discretizations and monitored vertical deflection at the midpoint of the loaded edge. All three schemes were reported as locking-free. For this single benchmark and this one deflection quantity, the augmented method was reported as four to 11 times more accurate than the non-augmented method. On the finest mesh, its result agreed with the Taylor-Hood deflection within 0.0002, while the non-augmented error stagnated around 0.01.
Evidence with boundaries
The results should be read as conditional numerical evidence. The existence and uniqueness statements depend on small-data and parameter assumptions, the a priori estimate requires regularity of the exact solution, and the efficiency result assumes piecewise-polynomial permeability and boundary data. The manufactured-solution and Cook’s membrane results are problem-specific; they do not establish general computational superiority or real-world predictive performance.
The supplied document is a preprint submitted to Elsevier, identified as arXiv:2609.37618v1 and dated 29 September 2026. Its evidence supports a mathematical and computational scheme under stated conditions, while leaving broader questions about performance outside these test settings unanswered.
Paper data and sources
Original title: An augmented mixed finite element method for the Biot problem with nonlinear permeability
Authors: Felipe Lepe, Gonzalo Rivera, Ricardo Ruiz-Baier, Segundo Villa-Fuentes
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-09-29
DOI: Not available
Original paper · Full text