Segregated Finite-Volume Solvers

• Segregated finite-volume solvers are numerical methods that decouple conservation laws into sequential updates for individual field variables.
• They employ a predictor-corrector cycle with lagged value strategies and inner iterations to ensure mass, momentum, and energy conservation.
• Recent advances integrate multiphysics coupling and reduced-order modeling, enhancing performance and applicability in CFD, FSI, and solid mechanics.

Segregated finite-volume solvers are numerical algorithms that solve the coupled systems arising from conservation laws—such as those of fluid mechanics, transport, or solid mechanics—by partitioning the governing equations into subsets and advancing each in sequence. Unlike fully coupled or monolithic schemes, segregated approaches solve for each primary variable (such as velocity components, pressure, energy, or species concentrations) separately within each iteration, using previously computed or lagged values for the remaining variables. This paradigm is foundational to many of the most widely used computational fluid dynamics (CFD) and computational solid mechanics software packages, particularly those based on the finite-volume method (FVM). Recent developments have adapted segregated strategies to include multiphysics coupling, advanced turbulence modeling, multiphase interfaces, large-deformation solids, and efficient reduced-order modeling, while retaining the mesh conservation, robustness, and algorithmic flexibility intrinsic to the segregated approach.

1. Mathematical and Algorithmic Structure

The segregated finite-volume paradigm partitions the global nonlinear discrete system arising from FVM spatial discretization into a set of algebraic subsystems, each predominantly associated with a single field variable or operator block. In nearly all cases, these subsystems are constructed by integrating the governing conservation laws (mass, momentum, energy, constituent transport) over polyhedral control volumes, yielding cell-based unknowns and face-based fluxes amenable to Gauss’s theorem.

Standard Algorithmic Cycle

• Predictor step: A subsystem—typically the momentum block—is assembled and solved for a tentative field value (e.g., provisional velocity $\mathbf{u}^*$) using lagged or old values of the coupled fields (e.g., pressure $p^{n-1}$, density $\rho^{n-1}$).
• Corrector step: The next subsystem—often a Poisson-like equation for pressure correction—is constructed such that mass conservation or divergence-free constraints are enforced. This step typically involves correction terms dependending on provisional field increments.
• Field update: Once correction quantities (e.g., $p'$, $\mathbf{u}'$) are obtained, the primary fields are updated to new iterate values via under-relaxation or linear combination.
• Inner and outer loops: Additional inner correctors (PISO/PIMPLE logic) may be applied to accelerate convergence of nonlinear and coupled dynamics, with under-relaxation factors finely tuned for stability and efficiency (Ngan et al., 2023, Zancanaro et al., 2022, Dritselis et al., 2022).
• Algorithm closure: This cycle is applied iteratively, sequentially refining each variable based on the most up-to-date solution and terminating upon achieving prescribed residual or incremental tolerances.

This structure is observed in schemes ranging from pressure-based solvers for turbulent compressible flows (Zancanaro et al., 2022), moving-mesh fluid-structure interaction (Ngan et al., 2023), multiphase VOF-based flows (Dritselis et al., 2022), to nonlinear solid mechanics (Cardiff et al., 24 Feb 2025).

2. Discretization: Governing Equations and Segregation Principles

Finite-volume segregated solvers enforce the conservation of quantities at the level of each control volume. The segregated viewpoint is especially congruent with conservative integral formulations:

• Fluid mechanics (Navier–Stokes/FVM): Each conserved field (mass, momentum, energy) is updated in sequence using cell-averaged or cell-centered schemes, with spatial fluxes discretized using upwind, central, or limited interpolations to achieve conservative and stable updates (Zancanaro et al., 2022, Ngan et al., 2023).
• Explicit treatment of operator blocks: Convective, diffusive, and source terms may each be updated separately, often by linearization or lagging. At each step, the impact of previously-updated (lagged) field values on the current update is critical to maintain coupling without tightly-coupled algebraic solvers.
• Coupling strategy: For multiphysics or interface problems, interface conditions (e.g., kinematic and dynamic continuity for FSI, interfacial tension for VOF methods) are imposed at the level of segregated updates by incorporating the latest available fluid or solid variables at boundary or interfacial cells (Ngan et al., 2023, Dritselis et al., 2022).
• Conservation properties and challenges: In applications such as tumor growth or free-boundary flows, naive segregated flux computation can violate geometric and totality conservation laws. Segregated-flux constructions—enforcing additivity, V-consistency, and cubic-preservation—are essential for properties such as incompressibility and geometric conservation (Zeng et al., 2018).

3. Specialized Segregated Methods Across Physical Domains

Fluid-Structure Interaction (FSI) and Moving Mesh

Segregated algorithms with Arbitrary Lagrangian–Eulerian (ALE) mesh motion decouple the fluid and structure subproblems using partitioned (loose) coupling. The fluid subproblem is advanced with updated mesh positions, often via a spring-analogy or RBF mesh displacement scheme, while coupling forces are exchanged at the interface and the mesh motion is integrated prior to the next fluid update (Ngan et al., 2023).

Multiphase and Non-Newtonian Flows

In multiphase contexts, e.g., VOF-based solvers for Newtonian or complex rheology fluids, segregated algorithms advance phase fractions, velocities, and pressures in sequence. Interface capturing, artificial compression, and VOF smoothing are integrated via solver-specific substeps, and non-Newtonian constitutive relations are coupled via phase-weighted extra-stress and segregated stress updates. These methods yield robust and stable simulations even in the presence of capillary and viscoplastic/viscoelastic effects (Dritselis et al., 2022).

Solid Mechanics

Segregated finite-volume solvers in solid mechanics break the momentum-balance system into independent updates for each Cartesian displacement component, typically adopting a block Gauss–Seidel approach and under-relaxed corrections. This approach reduces both memory and computational demands while accommodating nonlinear constitutive behavior, though at the cost of increased outer iterations for strongly coupled (e.g., elastoplastic) problems (Cardiff et al., 24 Feb 2025).

Geophysical and Reduced Models

For large-eddy simulation of quasi-geostrophic systems, a three-step segregated approach solves for vorticity, filtered vorticity, and stream function in sequence, each corresponding to an elliptic or transport-type solve, allowing efficient simulation even on coarse meshes (Girfoglio et al., 2022).

4. Reduced-Order, Hyper-Reduction, and Data-Driven Segregated Solvers

Segregated finite-volume solvers form the backbone of several classes of reduced-order models (ROMs) and hyper-reduction strategies, enabling rapid many-query and real-time evaluations:

• POD–Galerkin ROMs: Proper orthogonal decomposition (POD) is used to generate low-dimensional bases for velocity, pressure, and (if relevant) mesh displacement. Each segregated substep (predictor, corrector, etc.) is projected into reduced space, constructing small ($N_r \times N_r$) algebraic systems that preserve the overall logic of the full-order segregated solver (Ngan et al., 2023, Zancanaro et al., 2022).
• Hyper-reduction via DEIM: Discretize-then-project frameworks assemble full-order operators and then project onto a reduced basis at a subset of sparsely-sampled mesh points, using empirical interpolation (EIM, DEIM) to ensure the online cost is independent of the global mesh size. Each segregated block can be hyper-reduced independently, and the method is parallelizable and mesh-scalable (Ngan et al., 11 Jan 2026).
• Hybrid turbulence treatment: For turbulent compressible ROMs, eddy-viscosity fields are approximated in POD space and evaluated in the online phase via neural networks, maintaining independence from specific turbulence closures and accelerating segregated iterations (Zancanaro et al., 2022).

Validation against full-order models consistently demonstrates that reduced and hyper-reduced segregated solvers can match key physical metrics (e.g., forces, limit-cycle amplitudes, phase portraits) at several orders-of-magnitude lower computational cost (Ngan et al., 2023, Zancanaro et al., 2022, Ngan et al., 11 Jan 2026).

5. Performance, Robustness, and Numerical Stability

The segregated finite-volume framework offers several distinct performance and stability characteristics:

• Algorithmic efficiency: By solving only one field at a time and relying heavily on sparse algebraic structures, segregated algorithms offer lower per-iteration cost and scalable parallelization, especially in the context of large or complex multiphysics problems (Dritselis et al., 2022, Cardiff et al., 24 Feb 2025).
• Numerical stability: Under-relaxation of field updates is standard and critical for converging strongly coupled or nonlinear systems. Typical relaxation factors range from 0.3 for pressure to 0.7 for velocities or displacements (Ngan et al., 2023, Zancanaro et al., 2022).
• Dynamic time-stepping: Adaptive control of time step via CFL, capillary, or viscous stability conditions allows large stable steps, especially using fully implicit or SIMPLEC variants (Dritselis et al., 2022, Ngan et al., 2023).
• Conservation fidelity: Advanced flux splitting and correction techniques (e.g., V-consistency, additivity, cubic-preservation) enable machine-precision satisfaction of integral and geometric conservation laws, especially in incompressible or free-boundary settings (Zeng et al., 2018).
• Solver flexibility: Segregated algorithms naturally accommodate switching between explicit and implicit time integration, higher-order spatial discretization, and a range of linear, nonlinear, or Jacobian-free solvers depending on stiffness and problem size (Cardiff et al., 24 Feb 2025, Lucca et al., 2023).

Numerical benchmarks consistently show that properly tuned segregated solvers achieve rapid convergence and high accuracy per iteration, with robustness across wide regimes of Reynolds number, Mach number, non-Newtonian behavior, and multiphase configurations (Dritselis et al., 2022, Zancanaro et al., 2022).

6. Extensions: Implicit, Hybrid, and Advanced Solver Architectures

Fully implicit segregated schemes, as realized in hybrid finite-volume/finite-element (FV/FE) solvers, further relax the stability constraints of explicit/semi-implicit variants. In these approaches, convective and diffusive operators are assembled in fully implicit form over dual FV and primal FE meshes, and momentum and pressure updates are handled via staggered discontinuous/continuous Galerkin elements. Nonlinear inner systems are solved with NV/GMRES or BiCGStab-Krylov algorithms, and energy-stable fluxes (e.g., semi-implicit Ducros flux) enforce discrete kinetic energy stability. These techniques are particularly effective for very high CFL flows, stiff viscous regimes, and realistic patient-specific vascular benchmarks (Lucca et al., 2023).

Additionally, Jacobian-free Newton–Krylov (JFNK) strategies offer Newton-type convergence acceleration for legacy segregated FVM codes in solid mechanics. Here, the action of the full Jacobian is approximated by residual differences within a Krylov subspace framework, requiring minimal code modification and delivering dramatic speedup, especially for large, stiff elastic and hyperelastic calculations (Cardiff et al., 24 Feb 2025).

7. Applications, Validation, and Best-Practice Recommendations

Segregated finite-volume solvers have demonstrated broad applicability: from vortex-induced vibration in FSI at Reynolds numbers near lock-in, to multiphase interaction for surface-tension-dominated droplets, to turbulent transonic flow, under-resolved geophysical LES models, and large-deformation elastodynamics in solid mechanics. Best practice within these methods includes:

• Adopting algorithm-tailored spatial and temporal discretization schemes (e.g., Van Leer, MULES, TVD-RK2, compact stencils),
• Ensuring strict enforcement of conservation via flux partitioning and stabilization,
• Coupling segregated solvers with hyper-reduction or machine learning for rapid ROM deployment,
• Tuning under-relaxation and preconditioning strategies adaptively for scale and stiffness,
• Utilizing implicit and hybrid strategies for stiff or multiscale applications (Ngan et al., 2023, Dritselis et al., 2022, Cardiff et al., 24 Feb 2025, Ngan et al., 11 Jan 2026).

Collectively, segregated finite-volume solvers represent a highly adaptive, efficient, and rigorously validated class of numerical algorithms for PDE-dominated multiphysics applications in research and industry.