---
title: Fixed-Strain Biot Splitting
url: https://www.emergentmind.com/topics/fixed-strain-biot-splitting-approach
type: topic
---

# Fixed-Strain Biot Splitting

Searching arXiv for recent and foundational papers on fixed-strain Biot splitting and closely related partitioned poromechanics methods.
Fixed-strain Biot splitting is a sequential partitioning strategy for poromechanics in which the flow subproblem is advanced while the strain, or its rate, is held fixed, after which the mechanics subproblem is updated with the newly computed pressure. In classical Biot systems the fixed quantity is typically the volumetric strain $\nabla\cdot u$; in multilayer poroelastic models it may also include bending-related quantities such as $\Delta_\tau w$ for a thin poroelastic plate; and in thermodynamically consistent soft-material poromechanics the same idea appears as an “undrained-like” split in which $\operatorname{div}((1-\phi)u)$ is frozen during the flow/mass update [2507.10538] [2011.13296].

## 1. Concept and terminology

The defining idea of the fixed-strain split is to decouple mechanics and flow by holding strain fixed during the flow solve. In the standard Biot setting this means that the volumetric strain $\varepsilon_v=\nabla\cdot u$ is treated as known in the flow equation, so the coupling term becomes a source term. The updated pressure is then inserted into the mechanics problem, yielding a sequential flow-first or flow/mass-first partitioning of the coupled system [2210.06206].

This basic idea admits several realizations. In the multilayer Stokes–Biot setting, fixed-strain is applied inside the poroelastic structure: in the thick Biot layer, “strain” refers to the volumetric strain $\nabla\cdot\eta$, while in the thin poroelastic plate it refers to the bending-related quantity $\Delta_\tau w$. The flow subproblem uses lagged mechanical terms, specifically $\xi^n$ and $v^n$, so that the mass equations are linear in pressure and filtration velocities; the mechanics subproblem is then solved with the newly computed pressures treated implicitly [2507.10538].

In the soft-material poromechanics formulation, the same mechanism is described as an undrained-like split. There, the method is derived from alternating minimization of a convex quadratic energy, and the mechanics substep contains a div-div stabilization term that controls the volumetric strain increment. The flow/mass substep is then solved with $u$ fixed, which is the fixed-strain step in the precise sense used in Biot-type splitting [2011.13296].

A persistent point of terminological confusion is the distinction between fixed-strain and drained split. The mixed five-field domain decomposition study states explicitly that the drained split, implemented as a mechanics-first scheme with pressure frozen at the old time level, is not the same as the classical fixed-strain split, which is a flow-first scheme obtained by freezing strain during the flow solve. The two are distinct sequential splittings and are not generally equivalent [2010.15353].

| Setting | Fixed quantity in flow step | Subsequent update |
|---|---|---|
| Classical Biot | $\varepsilon_v=\nabla\cdot u$ | Mechanics with new pressure/head |
| Soft-material poromechanics | $\operatorname{div}((1-\phi)u)$ | Mechanics from alternating minimization |
| Multilayer Biot plate | $\nabla\cdot\eta$ and $\Delta_\tau w$ via lagged $\xi^n,v^n$ | Thick-layer and plate mechanics with new $p,q$ |

## 2. Governing formulations

In standard pressure-based Biot poroelasticity, the coupled model combines linear elasticity and Darcy-type flow. A representative form is
$$
-\nabla\cdot\sigma(u)+\alpha\nabla p=f,\qquad
\sigma(u)=2G\,\varepsilon(u)+\lambda\,\mathrm{tr}(\varepsilon(u))I,
$$
together with
$$
c_0\,\partial_t p+\alpha\,\partial_t(\nabla\cdot u)-\nabla\cdot\!\left(\frac{k}{\mu}\nabla p\right)=g.
$$
The head-based formulation replaces pressure by hydraulic head $h$, with $P=\rho g(h-z)$ and Darcy law $q=-K\nabla h$ [2210.06206].

The mixed five-field formulation of the quasi-static Biot system uses displacement $u$, weakly symmetric stress $\sigma$, rotation $\gamma$, pore pressure $p$, and Darcy velocity $z$. In that setting, the constitutive relation is written as $\sigma=\sigma_e-\alpha pI$ with $A\sigma_e=\varepsilon(u)$, and the mass conservation equation appears in the form
$$
c_0 \dot p+\alpha\,\mathrm{tr}(\dot\varepsilon(u))+\nabla\cdot z=g.
$$
This representation is important for domain decomposition because it exposes interface quantities associated with normal stress and normal Darcy flux [2010.15353].

The soft-material poromechanics model is similar to, but not equivalent to, the classical Biot model. It retains solid inertia $\rho_s(1-\phi)\partial_{tt}u$, fluid inertia $\rho_f\partial_t v$, viscous fluid stress $\operatorname{div}(2\mu_f\varepsilon(v))$, and a drag term $\phi^2k^{-1}(v-\partial_t u)$. Its mass conservation law is
$$
N\,\partial_t p+\operatorname{div}(v)+\operatorname{div}((1-\phi)\partial_t u)=0,
$$
with $N=(1-\phi)^2/\kappa_s$ acting as a storage-like parameter. The model is therefore fully dynamic and uses absolute fluid velocity rather than a pure Darcy law in $-\nabla p$ [2011.13296].

The multilayer fluid–poroelastic interaction formulation extends Biot splitting into a three-domain setting. The thick poroelastic layer $\Omega_b$ contains displacement $\eta$, velocity $\xi=\partial_t\eta$, pore pressure $p$, and relative filtration velocity $u_b$. The thin plate $\Omega_p=\Gamma\times[-H/2,H/2]$ contains transverse displacement $w$, velocity $v=\partial_t w$, curvature variable $\Lambda=-\Delta_\tau w$, pore pressure $q$, and normal filtration velocity $u_p$. The free fluid $\Omega_f$ satisfies the time-dependent Stokes equations with velocity $u$ and pressure $\pi$. Coupling is enforced through Beavers–Joseph–Saffman slip, kinematic continuity of normal velocities, displacement continuity on the middle surface, dynamic balance of normal tractions, and pressure continuity across the plate faces [2507.10538].

## 3. Split algorithms and subproblem structure

The classical fixed-strain algorithm in backward Euler time discretization solves flow first with frozen volumetric strain and then solves mechanics with the updated pressure. In the head-based formulation, over a time step $[t^n,t^{n+1}]$, the flow step is
$$
s_{\mathrm{stor}}\frac{h^{n+1,(m)}-h^n}{\Delta t}
-\nabla\cdot(K\nabla h^{n+1,(m)})
=
Q^{n+1}
-\alpha\,\frac{\varepsilon_v^{*,n+1,(m)}-\varepsilon_v^n}{\Delta t},
$$
where $\varepsilon_v^{*,n+1,(m)}=\nabla\cdot u^{*,n+1,(m)}$ is frozen from the previous inner iteration. The mechanics step then solves
$$
\nabla\cdot\left(\sigma(u^{n+1,(m)})-\alpha P^{n+1,(m)}I\right)=f^{n+1},
\qquad
P^{n+1,(m)}=\rho g(h^{n+1,(m)}-z).
$$
The inner splitting loop stops when both the flow and mechanics residuals satisfy the absolute or relative stopping criteria stated in the discrete formulation [2210.06206].

In the soft-material model, fixed-strain is realized as alternating minimization of a convex quadratic energy $J(u,v)$. At each time step, the iteration alternates between a mechanics solve with div-div stabilization and a flow/mass solve with $u$ fixed. The pressure iterate is given by
$$
p^{n,k}=N\left[\Delta t\,g_p^n-\Delta t\,\operatorname{div}v^{n,k}-\operatorname{div}((1-\phi)u^{n,k})\right].
$$
The mechanics substep includes the stabilization
$$
N\big(\operatorname{div}((1-\phi)(u^{n,k}-u^{n,k-1})),\operatorname{div}((1-\phi)\tilde u)\big),
$$
which is the mechanism enforcing control of the volumetric strain increment. The subsequent flow/mass substep solves for $(v^{n,k},p^{n,k})$ with $u^{n,k}$ held fixed [2011.13296].

In the multilayer Stokes–Biot problem, fixed-strain is embedded in a larger three-step partitioned algorithm. Step 1 solves the thick-layer Darcy problem and the plate flow problem using lagged $\xi^n$ and $v^n$:
$$
\kappa^{-1}u_b^{n+1}=-\nabla p^{n+1},\qquad
c_0 d_t p^{n+1}+\nabla\cdot u_b^{n+1}=G_b^{n+1}-\alpha\nabla\cdot\xi^n,
$$
and
$$
(\kappa^p)^{-1}u_p^{n+1}=-\nabla_{n^p}q^{n+1},\qquad
c_0^p d_t q^{n+1}+\nabla_{n^p}u_p^{n+1}=G_p^{n+1}+\alpha^p z\Delta_\tau \tilde v^n.
$$
Step 2 updates thick-layer and plate mechanics with the new $p^{n+1}$ and $q^{n+1}$. Step 3 solves the Stokes problem in $\Omega_f$ with normal traction set by $-q^{n+1}|_{\Gamma_-}$, Beavers–Joseph–Saffman tangential slip, and a penalty term that weakly enforces normal velocity continuity [2507.10538].

The mixed five-field domain decomposition formulation gives another algorithmic interpretation. There, the fixed-strain flow substep is obtained by imposing $\dot\varepsilon(u)=0$, equivalently $\Delta\sigma_e=0$, so the flow mass balance decouples from mechanics:
$$
c_0\frac{p_h^{n+1}-p_h^n}{\Delta t}+ \nabla\cdot z_h^{n+1}=g^{n+1}.
$$
Mechanics is then solved with $p^{n+1}$ as data. The resulting flow and elasticity interface problems are symmetric positive definite and are solved by conjugate gradients in the interface formulation described for the split schemes [2010.15353].

## 4. Stability, convergence, and splitting error

The multilayer Stokes–Biot analysis gives a detailed stability theory for fixed-strain in the presence of a thick Biot layer, a thin poroelastic plate, and Stokes flow. The continuous formal energy contains thick-layer kinetic and elastic terms, plate kinetic and bending terms, storage terms $c_0\|p\|^2$ and $c_0^p\|q\|^2$, and the Stokes kinetic term $\rho_f\|u\|^2$. The dissipation contains Darcy dissipation in the thick layer and plate, viscous Stokes dissipation, and the Beavers–Joseph–Saffman contribution. After spatial finite element discretization, the discrete stability proof yields two regimes. The first is a conditional regime in which $\Delta t$ must satisfy restrictions with scalings $h^2$, $(h/H)^4$, $h$, and $h/H$; the second is a CFL-type regime in which, under the parameter constraints $\alpha^2<c_0\lambda_b$ and $(\alpha^p)^2<12c_0^pD$, stability reduces to a condition linear in $h$ and $h/H$ [2507.10538].

The soft-material undrained-like split admits a different convergence analysis based on generalized gradient flows. At each time step the semi-discrete problem is equivalent to minimizing a strictly convex quadratic functional $J(u,v)$, and alternating minimization yields a linearly convergent iteration. If $(u^n,v^n)$ is the unique minimizer and $(u^{n,k},v^{n,k})$ are the iterates, then
$$
|(u^{n,k}-u^n,v^{n,k}-v^n)|^2
\le
\left(1-\frac{1}{1+\gamma}\right)^2
|(u^{n,k-1}-u^n,v^{n,k-1}-v^n)|^2.
$$
The proof requires $N=(1-\phi)^2/\kappa_s>0$, so the quasi-incompressible limit $N\to 0$ is excluded. The reported interpretation is that convergence deteriorates for nearly incompressible and quasi-impermeable media and may be mildly influenced by strong porosity heterogeneities [2011.13296].

In the finite volume–virtual element study of classical Biot poroelasticity, fixed-strain is described as only conditionally stable, in contrast with the monolithic approach, which is reported as unconditionally stable for the linear Biot model considered. The lagging of volumetric strain introduces a splitting error, and the practical consequence is that strong coupling, large $\Delta t$, stiff skeletons, or low storativity may require smaller time steps or more inner iterations. The study therefore treats fixed-strain as a method whose overall effectiveness depends on both its stability window and the cost of reducing the splitting residuals inside each time step [2210.06206].

A related clarification is negative rather than positive: the mixed five-field domain decomposition paper does not analyze fixed-strain stability. It proves unconditional stability for drained split and fixed-stress split, but not for fixed-strain; its discussion of fixed-strain is algorithmic and comparative rather than theorem-based [2010.15353].

## 5. Discretization, validation, and computational performance

The multilayer Stokes–Biot implementation uses piecewise linear $P1$ finite elements for the thick Biot layer and plate variables and Taylor–Hood $P2$–$P1$ elements for the Stokes velocity–pressure pair. The interface treatment is a primal mixed weak formulation with a penalty on $\Gamma_-$ to improve mass conservation, enforcing $u\cdot n^p\approx v+u_p$. The extension operator $\tilde f$ on $\Omega_p$ is computed by solving $\Delta_{n^p}\tilde f=0$ with Dirichlet data on $\Gamma_-$ and Neumann data on $\Gamma_+$, and the plate average $\overline f$ is implemented by Simpson’s rule in $z$. In manufactured-solution tests, the method shows second order in space and first order in time, and the numerical energy matches the exact energy over long times $T=10\,\mathrm{s}$. In the biologically inspired vessel problem, numerical evidence shows convergence to the Stokes–Biot model without plate as $H\to 0$, while the presence of the thin plate regularizes the dynamics by producing fewer and smaller oscillations in interface displacement and fluid pressure [2507.10538].

The classical Biot parallel study uses a finite volume–virtual element discretization with MPFA-O for flow and low-order VEM for elasticity, solved by a Bi-CGSTAB iteration with an ILU-based preconditioner. The expected advantage of fixed-strain, namely that the flow and mechanics subproblems have simpler matrix structure than the monolithic coupled block, is only partially realized in practice. In Problem A, fixed-strain required 13–14 inner iterations per time step and was slower overall than the monolithic strategy at every tested core count, although its iteration phase scaled slightly better. In Problem B, both monolithic and fixed-strain exhibited superlinear speedups driven by preconditioner behavior, but monolithic total speedup was slightly higher because assembly consumed a smaller fraction of runtime than in the fixed-strain strategy [2210.06206].

The soft-material benchmarks emphasize iteration counts rather than parallel speedup. The undrained-like split performs well in some compressible and moderately permeable regimes, but deteriorates as $\kappa_s$ increases or permeability decreases. In the swelling test, increasing $\kappa_s$ from $10^2$ to $10^5$ increases the average iteration count from $8.6$ to “no convergence at 200 iterations,” and decreasing permeability from $10^{-9}$ to $10^{-12}$ changes the average count from $17.6$ to “no convergence at 500 iterations.” The same study reports that Anderson acceleration, especially with depth $m=5$, dramatically reduces iteration counts, improves robustness, and enables convergence in regimes where plain splits fail [2011.13296].

## 6. Comparison with fixed-stress, drained, and domain decomposition methods

Fixed-strain and fixed-stress are often discussed together because both are sequential decoupling strategies for Biot-type systems, but their stability properties differ sharply in the reported literature. The multilayer Stokes–Biot study states that fixed-strain is conceptually simple and modular, decoupling the problem into standard Darcy flow, elastic solid, and Stokes subproblems that are easily implemented with existing solvers. The same study also notes that in standard poromechanics fixed-stress splitting is known to be unconditionally stable for $\alpha\ge 0.5$ and is often preferred in geomechanics, whereas fixed-strain is conditional unless additional parameter constraints reduce the timestep restriction to a linear CFL condition. It further observes that adapting fixed-stress robustly to the anisotropic Biot plate with restricted flow and fourth-order bending is nontrivial, which is one reason the fixed-strain extension is developed in that setting [2507.10538].

The soft-material study reaches a related but not identical conclusion. There, the undrained-like split is efficient and robust when the solid is sufficiently compressible, that is, when $N=(1-\phi)^2/\kappa_s$ is “comfortably positive,” and when permeability is moderate. For large $\kappa_s$, quasi-incompressible regimes, or very low permeability, the fixed-stress-like diagonally stabilized split is reported as the more robust option. The recommended stabilized choice is $\,\bar p\approx (1-\phi)^2/(K_{dr}\Delta t)$, with $F=0$ and a conservative solid destabilization $S=-(\phi^2k^{-1})/(2\Delta t)$, and the paper emphasizes the role of inf-sup stable fluid-pressure discretizations and Anderson acceleration in difficult regimes [2011.13296].

The distinction from drained split is structural rather than merely semantic. The mixed five-field domain decomposition study states that drained split is a mechanics-first method with pressure frozen at the old time level, while fixed-strain is a flow-first method with $\dot\varepsilon(u)=0$ during the flow solve. In the domain decomposition setting, this fixed-strain construction yields separate flow and mechanics interface problems that are symmetric positive definite and solved with conjugate gradients; their interface operators have condition number $O(h^{-1})$ in the split formulations described. The same study explicitly states that drained split and fixed-strain are not equivalent [2010.15353].

A plausible implication of these results is that fixed-strain occupies a technically specific niche rather than serving as a universal replacement for monolithic or fixed-stress methods. The reported advantages are modularity, interpretable subproblem structure, and straightforward coupling to additional physics such as Stokes flow and thin-plate poroelasticity. The reported limitations are conditional stability, sensitivity to strong coupling and low storage, and, in some computational settings, a loss of overall efficiency once inner splitting iterations and assembly overhead are accounted for [2210.06206].

Source: https://www.emergentmind.com/topics/fixed-strain-biot-splitting-approach