Constrained Transport (CT) Methods

• Constrained Transport (CT) is a class of numerical methods that preserve discrete structural constraints, such as maintaining a divergence-free magnetic field in MHD simulations.
• CT algorithms employ staggered meshes, high-order spatial reconstructions, and upwind compositions to ensure physical accuracy and computational stability in complex flow conditions.
• CT techniques offer robust performance in adaptive mesh refinement and high-order schemes, making them essential for credible simulations in astrophysics, plasma physics, and optimization.

Constrained Transport (CT) is a class of numerical methods designed to preserve discrete structural constraints in continuous transport processes, most notably the divergence-free condition for vector fields or precise marginal/throughput constraints in generalized transport. CT is foundational in computational magnetohydrodynamics (MHD), numerical relativity, and modern optimal transport formulations, providing the structural guarantees necessary for both physical fidelity and computational stability.

1. Mathematical Principles and Invariant Preservation

Classical constrained transport emerged in the context of staggered discretizations of the induction equation in ideal MHD, ensuring that the discrete divergence of the magnetic field, typically written as

$\nabla\cdot{\bf B}=0$

remains exactly (up to round-off) preserved throughout the simulation. In this framework, the magnetic field components are stored on cell faces—rather than at cell centers—and updated via discrete curls of edge-centered electric fields, mirroring Stokes' theorem at the discrete level (Mocz et al., 2014, Olivares et al., 2019, Felker et al., 2017, Cheong et al., 2021).

The critical property is that, when initial data satisfies the constraint, all subsequent time updates at the field’s respective staggered locations maintain this property exactly: $$(\nabla\cdot{\bf B})^{n+1} = (\nabla\cdot{\bf B})^{n} = 0$$ This is a consequence of updating face-centered fluxes as the circulation of electric fields around cell edges; each edge contribution cancels across adjacent faces, making the approach robust against the accumulation of numerical monopoles (Mocz et al., 2014, Miniati et al., 2011).

Extensions of CT appear in generalized mass transport problems—as exemplified in "Monge-Kantorovich Optimal Transport Through Constrictions and Flow-rate Constraints"—where discrete constraints are imposed on throughput across spatial or temporal constrictions via marginal densities or flow-rate bounds, and the admissible class of transport plans is characterized by existence, convexity, and structural theorems originating from measure theory and optimal transport (Dong et al., 2022).

2. Algorithms and Data Structures

Constrained transport methods distinguish themselves from divergence-cleaning or projection approaches by a focus on variable collocation and discrete update design:

• Staggered-mesh CT: Face-centered storage of magnetic fluxes, edge-centered (or face-corner) electric fields, and algorithmic updates via explicit line integrals over face boundaries (Felker et al., 2017, Mocz et al., 2014, Olivares et al., 2019).
• Unstaggered/potential-based CT: Cell-centered representation of the vector potential $\mathbf{A}$, with the magnetic field reconstructed from its discrete curl, guaranteeing $\nabla\cdot{\bf B}=0$ by construction (Mocz et al., 2016, Helzel et al., 2012, Christlieb et al., 2014).
• Prolongation and restriction operators: On adaptive mesh refinement (AMR), divergence-preserving restriction (fine-to-coarse) and prolongation (coarse-to-fine) operators guarantee that multilevel meshes do not introduce constraint violations at refinement boundaries (Olivares et al., 2019, Olivares et al., 2018, Miniati et al., 2011).

In upwind-constrained transport (UCT), edge-centered electric fields required for the update are assembled through a multidimensional composition formula combining Riemann solver results from face-adjacent reconstructed values. This technique maintains consistency with the base fluid solver and enables extensions to high order in both space (e.g., via WENO, MP5, or PPM reconstructions) and time (SSP-Runge-Kutta integrators) (Mignone et al., 2020, Felker et al., 2017).

3. Applications in Magnetohydrodynamics and Related Fields

CT is the standard for ideal and resistive MHD across astrophysics, plasma physics, and relativistic flows. Key features include:

• Preservation of $\nabla\cdot{\bf B}=0$: Vital for correct Lorentz force computation and to prevent the onset of numerical instabilities or nonphysical solutions (Mocz et al., 2014, Olivares et al., 2019).
• Compatibility with complex meshes: Methods are implemented for unstructured Voronoi tessellations and both static and moving adaptive meshes, as in AREPO and the Black Hole Accretion Code (BHAC), supporting automatic adaptivity and reduced advection errors (Mocz et al., 2016, Mocz et al., 2014, Olivares et al., 2019, Olivares et al., 2018).
• General-relativistic MHD (GRMHD) extensions: Staggered-mesh CT is employed in general-relativistic frameworks, maintaining the discrete divergence-free property in arbitrary stationary spacetimes and under AMR (Cheong et al., 2021, Olivares et al., 2019).
• Divergence cleaning and elliptic projection: For schemes where discretization or mesh adaption can introduce small residual divergence, additional elliptic cleaning steps—solving a Poisson equation for correction—are often used in conjunction with CT (Cheong et al., 2021), although in many cases standard CT alone suffices.

CT’s impact extends to recent developments in:

• High-order DG and WENO schemes: CT maps or vector-potential approaches enable globally divergence-free, entropy-stable high-order schemes on both structured and logically-mapped grids, often combined with characteristic limiters for shock robustness (Rossmanith, 2013, Helzel et al., 2012, Christlieb et al., 2014).
• Positivity-preserving integration: Recent advances combine CT with provably positivity-preserving limiters in high-Mach and high-magnetization regimes, yielding robust schemes for extreme plasma conditions (Pang et al., 2024).

4. Extension to Generalized Optimal Transport with Constraints

Beyond vector field divergence-free preservation, CT is generalized to constrain transport plans in multi-marginal Monge–Kantorovich frameworks. In such settings:

• Discrete flow through constrictions: Constrictions (or “toll stations”) impose pointwise upper bounds on the crossing-time marginals, leading to complex, convex-constrained optimal coupling problems (Dong et al., 2022).
• Convex optimization and entropic regularization: Large-scale CT-constrained transport is solved via discretization to finite-dimensional linear programs (with throughput constraints), and acceleration by regularized Sinkhorn or ADMM-type algorithms (Dong et al., 2022).
• Uniqueness and existence: For certain constraint topologies (e.g., 1D with monotone cost kernels), uniqueness of minimizers and quasi-monotonicity of optimal plans are established (Dong et al., 2022).

A plausible implication is that CT’s formalism encompasses both classical vector field constraints (magnetic, incompressible) and more abstract mass/flow control in modern applied optimization, machine learning, and resource scheduling.

5. Comparison with Divergence Cleaning and Structural Stability

CT is contrasted sharply with divergence-cleaning methods, such as the Generalized Lagrange Multiplier (GLM) approach:

• Magnitude of divergence error: CT maintains $\nabla\cdot{\bf B}$ at round-off (e.g., $L^\infty \sim 10^{-13}$–$10^{-16}$), while GLM and Powell-type cleaning yield divergence errors of $\sim 10^{-3}$–1, which can grow and corrupt solution quality, particularly across shocks or in highly turbulent regimes (Mocz et al., 2014, Olivares et al., 2019, Cheong et al., 2021, Olivares et al., 2018).
• Long-term and shock robustness: CT ensures correct jump conditions, avoids spurious mean-field growth, and preserves structural invariants independent of mesh velocity or geometric complexity (Mocz et al., 2016, Mocz et al., 2014, Felker et al., 2017).
• Impact on physical observables: In astrophysical simulations (e.g., galaxy formation, black hole accretion, relativistic jets), CT yields significantly improved fidelity in both local features and global invariants relative to divergence-cleaning approaches (Mocz et al., 2016, Olivares et al., 2019, Olivares et al., 2018).
• Effect in multi-scale AMR environments: Only CT with correct divergence-preserving prolongation/restriction and edge-correction avoids creation of monopoles at mesh refinement interfaces (Olivares et al., 2019, Olivares et al., 2018, Miniati et al., 2011).

6. High-Order and Generalized CT Schemes

Advances toward high-order CT are characterized by:

• High-order spatial reconstruction: WENO, PPM, or piecewise-polynomial reconstructions to faces and edges, enabling formal third- or fourth-order accuracy in both conservation laws and CT updates (Felker et al., 2017, Helzel et al., 2012, Christlieb et al., 2014, Rossmanith, 2013).
• Upwind CT (UCT) composition: General frameworks for composing edge-centered electric fields from face Riemann solutions, consistent with base Godunov solvers, and readily extensible to arbitrary order and to relativistic MHD (Mignone et al., 2020, Felker et al., 2017).
• Implicit/explicit splitting and positivity: Schemes such as PPCT couple energy-conserving implicit CT updates for the magnetic subsystem with provably positivity-preserving FV updates for the fluid subsystem using Strang splitting. The implication is that CT approaches can now address stringent physical constraints (e.g., positivity of density and pressure) in extreme parameter regimes (Pang et al., 2024).
• Vector-potential CT and unstaggered methods: Cell-centered, vector-potential updates (with or without moving mesh) guarantee structural invariance and interface robustly with mesh adaptivity and non-structured grids (Mocz et al., 2016, Helzel et al., 2012, Christlieb et al., 2014).

7. Computational Implementation, Performance, and Benchmarking

CT schemes incur moderate computational overhead (typically $\sim 5$–$$(\nabla\cdot{\bf B})^{n+1} = (\nabla\cdot{\bf B})^{n} = 0$$0 over flux-only schemes due to additional reconstructions, Poisson projections, or edge-face mappings) but are essential for correctness in practical computations. Standardized benchmarks such as Orszag–Tang vortex, magnetic loop advection, relativistic rotors, strong MHD shocks, and astrophysical jets consistently demonstrate:

• Formal order of accuracy (second to fourth), with divergence error at machine precision
• Stability at high Lorentz factors and in strongly magnetized regimes
• Elimination of spurious field amplification or unphysical forces
• Efficient scalability in parallel/MPI settings and compatibility with block AMR (Mocz et al., 2014, Cheong et al., 2021, Olivares et al., 2019, Pang et al., 2024, Felker et al., 2017).

Key sources for the above are:

• Mocz et al., "A Constrained Transport Scheme for MHD on Unstructured Static and Moving Meshes" (Mocz et al., 2014)
• Olivares et al., "The Black Hole Accretion Code: adaptive mesh refinement and constrained transport" (Olivares et al., 2018)
• Porth et al./Ripperda et al., "Constrained transport and adaptive mesh refinement in the Black Hole Accretion Code" (Olivares et al., 2019)
• Del Zanna et al., Mignone et al., "Systematic construction of upwind constrained transport schemes for MHD" (Mignone et al., 2020)
• Rossmanith, "High-Order Discontinuous Galerkin Finite Element Methods with Globally Divergence-Free Constrained Transport for Ideal MHD" (Rossmanith, 2013)
• Christlieb et al., "Positivity-Preserving Finite Difference WENO Schemes with Constrained Transport for Ideal Magnetohydrodynamic Equations" (Christlieb et al., 2014)
• Pang & Wu, "Provably Positivity-Preserving Constrained Transport (PPCT) Second-Order Scheme for Ideal Magnetohydrodynamics" (Pang et al., 2024)
• Mikami et al., "Monge-Kantorovich Optimal Transport Through Constrictions and Flow-rate Constraints" (Dong et al., 2022)

Constrained transport remains the reference approach for structure-preserving simulation of magnetic, incompressible, and constrained flow phenomena spanning computational physics, optimal transport, and applied mathematics.