Reduced Circulation Models

Updated 5 June 2026

Reduced circulation models are mathematical approximations that simplify complex vascular and oceanic systems by reducing spatial and parametric complexity.
They employ dimensional reduction, projection-based ROMs, and data-driven surrogates to accurately capture fluxes, pressures, and velocities in network simulations.
Applications span hemodynamics and geophysical flows, offering significant computational speedups with maintained accuracies useful in patient-specific and climate modeling.

A reduced circulation model is any mathematical approximation of a circulation system—such as vascular, cerebrospinal, or oceanic flow—that systematically eliminates spatial, parametric, or dynamical complexity to enable efficient simulation or analysis. Such models may be derived from full three-dimensional partial differential equations (PDEs) via dimensionality reduction (e.g., to 1D or 0D systems) or via reduced-order modeling techniques (e.g., projection-based, data-driven surrogates). The essential objective is to provide computationally tractable frameworks that encode the principal physical mechanisms of circulation, maintain accurate predictions of aggregate or local quantities of interest (fluxes, pressures, velocities), and facilitate scalable modeling, parameter inference, and control in complex biological or geophysical flow networks (Daversin-Catty et al., 2021, Vanderborght et al., 22 Oct 2025, Tezzele et al., 2017, Adjoua et al., 2019, Puelz et al., 2015, San et al., 2012, Siena et al., 20 Oct 2025, Pegolotti et al., 2020).

1. Foundations and Mathematical Formulation

Reduced circulation models originate by systematically averaging or projecting the governing fluid dynamics equations—typically incompressible Navier–Stokes or Stokes systems—over appropriate geometric or dynamic structures.

In physiological contexts, commonly employed reductions include:

One-dimensional (1D) models: Cross-sectionally averaged mass and momentum balances along centerlines; pressure–area constitutive laws for compliance; friction/vessel drag closures via profile-based resistance terms (Puelz et al., 2015, Pfaller et al., 2021, San et al., 2012).
Zero-dimensional (0D, lumped-parameter) models: Circuit analogues (resistor, capacitor, inductor), yielding ODE or DAE systems for pressures and flows at discrete compartments; Windkessel elements for outflow boundary conditions (Pfaller et al., 2021).
Hierarchical (mixed-dimension) couplings: Embedding 1D models in 3D domains for large vessels, or coupling 1D–0D networks for multiscale circulation (Adjoua et al., 2019, Köppl et al., 2018).

Oceanographic and geophysical reduced models exploit:

Box models (e.g., Stommel/Cessi two-box, S-box) for analyzing regime transitions in global meridional overturning circulation, typically yielding ODE systems for temperature and salinity differences (Noory, 12 Feb 2026, Smith et al., 2024).
Reduced-dimension boundary-plane or latitude–depth models, collapsing the interior to dynamically-relevant interfaces (e.g., western–eastern boundaries in the Atlantic) to resolve essential overturning dynamics with strong topological or mixing asymmetries (Vanderborght et al., 22 Oct 2025).

Generalized model structure:

State variables: Cross-sectionally averaged fluxes or pressures (e.g., $q^i(s,t)$ , $p^i(s,t)$ in a network segment indexed by $i$ ) (Daversin-Catty et al., 2021).
Governing PDEs (for 1D axesymmetric flows):

$\frac{\rho}{A^i} \partial_t q^i - \frac{\mu}{A^i} \partial_{ss} q^i + \mu \frac{\alpha^i}{A^i} q^i + \partial_s p^i = 0,\qquad \partial_s q^i = f^i(s,t),$

where $A^i$ is annular area, $\alpha^i$ is the resistance parameter, and $f^i$ encodes wall motion (Daversin-Catty et al., 2021).

Network coupling: Flux conservation and normal stress continuity at junctions, typically enforced via:
- $q^{\mathrm{parent}}(b) = \sum_{\mathrm{daughters}} q^{\mathrm{daughter}}(b)$
- $\sigma^{\mathrm{parent}}(b) = \sigma^{\mathrm{daughter}_j}(b)$ , with $\sigma$ a 1D stress analogue (Daversin-Catty et al., 2021, Puelz et al., 2015).
Constitutive closures: Velocity profile specification (Poiseuille, plug, or Womersley), friction/resistance terms, elastic wall laws, and, when necessary, boundary- or wall-driven source terms.

2. Model Reduction Methodologies

Multiple frameworks exist for reducing circulation models, each with specific algorithmic pipelines and scopes of applicability:

Dimensional Reduction: Direct formal averaging of 3D PDEs over cross-sections to yield 1D or 0D systems, under specified geometric and physical assumptions (axisymmetry, slow cross-sectional variation, specified velocity profile) (Daversin-Catty et al., 2021, Vanderborght et al., 22 Oct 2025, Puelz et al., 2015). The extended 1D formalism for variable-radius vasculature corrects standard assumptions to retain leading-order effects from stenotic or tapered geometry (Canic et al., 2024).
Model Order Reduction (ROM): Projection-based (e.g., proper orthogonal decomposition (POD) and Galerkin projection), often trained on full-order snapshots (velocity, pressure) and retaining a small number of modes. Inf-sup stabilization via supremizer enrichment is critical to ensure pressure–velocity coupling in incompressible flows (Siena et al., 20 Oct 2025, Tezzele et al., 2017, Pegolotti et al., 2020).
Active Subspace/Parameter Space Reduction: Dimensionality reduction of large parameter spaces via eigendecomposition of the uncentered covariance in the gradient of a quantity of interest (e.g., pressure drop), allowing for parametric compression prior to projection (Tezzele et al., 2017).
Data-Driven and Hybrid ROMs: Surrogate models for dynamical evolution, e.g., combining POD with feedforward or LSTM neural networks to approximate temporal or parametric progression of modal coefficients (Siena et al., 20 Oct 2025, Jr et al., 2021). Data-driven surrogates based on kernel methods (e.g., Gaussian processes) also provide rapid, accurate mapping from parameters (e.g., stenosis severity) to output time-series, enabling real-time state or parameter estimation (Köppl et al., 2018).
Geometric and Modular Decomposition: Domain decomposition into building blocks (e.g., tubes, bifurcations), with local reduced basis and coupling via Lagrange multipliers or Riemann solvers at interfaces, as in modular arterial network models (Pegolotti et al., 2020, Pfaller et al., 2021).
Automated Model Generation: Extraction of networks from imaging data, identification of centerlines and junctions, calibration of element parameters, and fully automated mapping from 3D geometry to coupled 1D–0D networks (Pfaller et al., 2021).

3. Boundary, Coupling, and Closure Conditions

Accurate enforcement of physical interface and boundary conditions remains central to fidelity:

Inflow/Outflow: Prescribed velocity or pressure at inlets; three-element Windkessel (R–C–R) or pure resistance at terminal points (Pfaller et al., 2021, Puelz et al., 2015).
Junctions/Bifurcations: Mass conservation ( $p^i(s,t)$ 0), and continuity of static or total pressure, with further closure by Riemann invariants, total momentum, or by solving small nonlinear systems (Puelz et al., 2015, Canic et al., 2024).
Wall Motion/Compliance:
- Explicitly incorporated as source terms (e.g., wall displacement or velocity entering flux divergence) (Daversin-Catty et al., 2021).
- Implicitly through elastic closure relations (e.g., pressure–area law in 1D models) (San et al., 2012).
Stenosis and Geometry Variability: Extended 1D models retain effects of longitudinal radius gradients and induced local accelerations, necessary for accurate post-stenotic flow and pressure predictions (Canic et al., 2024).
Hybrid and Data-Driven BCs: Lifting function strategies enable the inclusion of non-homogeneous boundary conditions within ROM snapshots, and neural networks can represent physiologically realistic outflow characteristics (Siena et al., 20 Oct 2025).

4. Numerical Implementation and Scalability

Reduced circulation models are implemented with numerical strategies designed for computational efficiency and scalability:

Discretization:
- Finite element (FEM), finite volume (FVM), discontinuous Galerkin (DG), and spectral/hp element frameworks adapted to 1D, 0D, or modular blocks (Puelz et al., 2015, Pegolotti et al., 2020).
- Mixed-element (e.g., continuous P1 for pressure, discontinuous P2 for flux) spaces for networked problems (Daversin-Catty et al., 2021).
Time Integration: Implicit or explicit backward Euler, BDF, or Runge–Kutta schemes, with careful consideration of stiffness and CFL conditions for hyperbolic or parabolic systems (San et al., 2012).
Solver Architecture:
- Saddle-point formulations for mixed systems.
- Use of Schur complements and modular local solves in block-decomposed domains for nearly linear scaling up to thousands of vessels (Pegolotti et al., 2020).
- PETSc and FEniCS platforms for networked finite-element assembly/solving (Daversin-Catty et al., 2021).
Offline/Online Decomposition: For ROMs, computational expenditure is concentrated in the offline (snapshot, basis extraction, pre-factorization) phase; online prediction phases often execute in sub-second timescales for complex networks or over cardiac cycles (Siena et al., 20 Oct 2025, Tezzele et al., 2017, Pegolotti et al., 2020).
Validation and Error Quantification: Systematic benchmarking against full-order 3D PDE simulations, with typical average errors in pressures and flows at vessel outlets at the 1–10% level, and maximal errors in complex network regions (bifurcations, severe stenosis) not exceeding 50% (Daversin-Catty et al., 2021, Pfaller et al., 2021, Canic et al., 2024).

5. Application Domains and Model Performance

Hemodynamic Circulation and Cardiovascular Networks

Applications in cerebral perivascular spaces, whole-arterial trees, peripheral stenoses, and cardiac/aortic flows.
Achievable acceleration: Reduced systems often use less than 1% of the degrees of freedom and memory of full 3D models, with per-cycle computational speedups of $p^i(s,t)$ 1– $p^i(s,t)$ 2 (Daversin-Catty et al., 2021, Siena et al., 20 Oct 2025, Pegolotti et al., 2020).
Patient-specific pipeline: Fully automated extraction and simulation from 3D imaging, with mean errors of 1–10% across >70 anatomical vascular models (Pfaller et al., 2021, Sankaran et al., 2019).
Hybrid equation/data ROMs: Combination of projection-based and machine-learned surrogates for outflow boundaries enables robust sub-millisecond patient-specific hemodynamics simulations (Siena et al., 20 Oct 2025).
Real-time clinical use: Physics-driven response-surface approaches yield per-point FFR predictions on sub-second timescales, supported by >1,300-patient validation (Sankaran et al., 2019).

Oceanic and Geophysical Circulation

Reduced box models (Stommel/Cessi): Capture regime transitions (e.g., AMOC tipping points), admit cusp-bifurcation structure, and explain the destruction of bistability by polar-amplification-driven erosion of background temperature gradients (Noory, 12 Feb 2026, Smith et al., 2024).
Boundary-plane and latitude–depth reductions: Encode geostrophic balance, boundary-intensified mixing, and interhemispheric exchange, achieving quantitative agreement with full 3D MITgcm simulations at orders-of-magnitude less cost (Vanderborght et al., 22 Oct 2025).
Proper orthogonal decomposition and stabilized ROMs: Enable accurate simulation of large-scale quasigeostrophic or lake circulation with $p^i(s,t)$ 3– $p^i(s,t)$ 4-fold computational gains (San et al., 2014, Jr et al., 2021).

Multiscale and Surrogate Modeling

Graph-based stochastic models: Mcrophysiologically realistic vascular trees, capillary beds, and compliant wall mechanics over thousands to millions of vessel segments (Adjoua et al., 2019).
Surrogate methods: Kernel-based or neural network surrogates for parameter-to-output mappings yield rapid estimation of stenosis impact or regime-change probability, with errors below clinical or geophysical tolerances (Köppl et al., 2018, Jr et al., 2021).

6. Limitations, Extensions, and Current Research Frontiers

Limitations

Reduced geometry models neglect fully 3D structures, separation, and secondary flow (e.g., post-stenotic recirculation, strongly curved/bifurcating regions) (Canic et al., 2024, Pegolotti et al., 2020).
Parameterization and closure strictness: Accuracy depends on fidelity of boundary data, compliance parameters, and resistance/friction closures; error increases in highly deformed, tortuous, or geometrically complex scenarios (Daversin-Catty et al., 2021, Pegolotti et al., 2020).

Extensions and Future Directions

Porous-wall and multi-physics coupling: Extensions to Darcy/Brinkman regimes, perfusion, and coupled solute/mass transport (Daversin-Catty et al., 2021).
Nonlinear ROMs and multiparametric/autoencoder reductions: Needed for fully capturing parametric variation beyond geometric or time-variation (Siena et al., 20 Oct 2025).
Advanced data assimilation: Bayesian and ensemble Kalman filtering for real-time inference and prediction in oceanic/atmospheric GCM surrogates (Smith et al., 2024).
Automated network extraction and segmentation for patient-specific modeling, including automated detection of stenoses or junctions (Pfaller et al., 2021).
Extension to global-scale inter-basin geophysical models for coupled climate–carbon–circulation dynamics (Vanderborght et al., 22 Oct 2025).

Reduced circulation models thus provide a unifying computational framework across biophysical and geophysical disciplines, enabling rapid, scalable, and increasingly automated simulation, prediction, and inference of complex, networked transport systems. Their development relies on a blend of asymptotic reduction, projection-based compression, data-driven surrogacy, and algorithmic innovation in network assembly and coupling—together opening new avenues for exploration in physiology, clinical practice, and climate science.