Mantle Convection Numerical Model Experiments: Governing Equations, Benchmarks, and Applications

Mantle convection numerical model experiments use high-performance computational simulations to reproduce the slow, buoyancy-driven circulation of Earth's silicate mantle, delivering quantitative predictions of heat transport, plate velocities, and plume dynamics. By coupling Stokes-flow solvers with energy-transport algorithms and rigorous benchmarking, these experiments bridge geophysical observation and geodynamic theory. This article explains the governing equations, model configurations, diagnostic outputs, verification and validation methods, and practical applications of modern mantle convection modeling.

Technical Overview

Mantle convection is the buoyancy-driven, solid-state creep of Earth's silicate mantle. It operates over geological timescales, transports the planet's internal heat toward the surface, and constitutes the fundamental engine of plate tectonics. Because the mantle cannot be sampled directly beyond a few hundred kilometers depth and flow velocities are only on the order of centimeters per year, numerical simulation serves as the primary experimental tool for studying mantle dynamics. Mantle convection numerical model experiments discretize the governing conservation equations and integrate them forward in time on high-performance computing systems, allowing researchers to test hypotheses that are impossible to examine in a physical laboratory.

Under the Boussinesq approximation and the infinite-Prandtl-number limit appropriate for a fluid with a dynamic viscosity of approximately 10 to the power of 21 pascal-seconds, the mathematical model consists of three coupled conservation laws:

• Conservation of mass, expressed as the incompressibility constraint on the divergence of the velocity field.
• Conservation of momentum, expressed as the Stokes equation, in which inertial forces are negligible because the Prandtl number of the mantle is approximately 10 to the power of 23, so the balance is between pressure gradients, viscous stresses, and thermal buoyancy.
• Conservation of energy, an advection-diffusion equation for temperature that includes conductive heat transport, advective transport by the creeping flow, and volumetric radiogenic heat production.

These equations are closed by constitutive relations describing mantle rheology, which may include Newtonian diffusion creep, non-Newtonian dislocation creep, temperature- and pressure-dependent viscosity, and visco-plastic yielding that localizes deformation in a plate-like manner. The controlling dimensionless parameter is the Rayleigh number, defined as the ratio of buoyancy driving force to viscous and thermal diffusion damping. Convection begins above a critical Rayleigh number of approximately 1708 for basally heated plane layers, whereas the Earth's mantle operates at Rayleigh numbers between roughly one million and one hundred million, producing vigorous, chaotic, time-dependent flow. Consequently, numerical experiments must collect long time series and report statistically averaged diagnostics rather than single deterministic snapshots.

Testing Samples: Numerical Model Configurations

Two-Dimensional Cartesian Box Models

Two-dimensional Cartesian boxes with aspect ratios between 1 and 4 are the standard entry configuration for mantle convection experiments. They support both basal heating and internal radiogenic heating, permit efficient parametric sweeps over Rayleigh number and viscosity contrast, and reproduce the canonical features of thermal convection: cold downwelling sheets, hot rising plumes, and thermal boundary layers at the top and bottom. With viscosity regularizations, viscosity contrasts of up to six orders of magnitude can be resolved, enabling studies of stagnant-lid regimes relevant to Mars and the early Earth.

Three-Dimensional Cartesian and Spherical Shell Models

Three-dimensional box models reveal the true planform of convection, including the spoke-pattern network of sheet-like downwellings connected by linear ridges of hot upwelling. Full spherical shell models reproduce the actual geometry of the mantle, with the ratio of the core-mantle boundary radius to the surface radius set to approximately 0.55 for present-day Earth. These models employ cubed-sphere, icosahedral, or octahedral meshes and routinely involve millions to hundreds of millions of unknowns, requiring petascale computing resources and message-passing parallelization across tens of thousands of processor cores.

Compressible and Thermochemical Models

Advanced configurations extend the basic formulation with the anelastic liquid approximation, which accounts for the adiabatic temperature increase of several hundred kelvin across the mantle, and with equations of state such as the Birch-Murnaghan relation. Phase transitions are represented through effective Clapeyron slopes, most importantly the endothermic transition near 660 kilometers depth, which can impede slab penetration and promote layered convection. Thermochemical models add compositional fields carried by tracer particles, allowing simulation of dense basaltic piles at the core-mantle boundary and the long-term preservation of geochemical reservoirs.

Testing Items: Standard Diagnostic Outputs

Every mantle convection numerical experiment is evaluated against a standardized set of dimensionless and physical diagnostics that characterize the vigor, structure, and heat transport efficiency of the simulated flow:

• Nusselt number at the surface and at the core-mantle boundary, defined as the ratio of total convective heat flux to purely conductive heat flux.
• Root-mean-square velocity of the flow field, typically converted to physical units of centimeters per year for comparison with plate motions.
• Horizontally averaged temperature profiles and bulk mean temperature, used to quantify internal heating effects and adiabatic gradients.
• Thermal boundary layer thicknesses at the top and bottom of the domain, estimated from boundary layer scaling theory.
• Viscous dissipation and full entropy and energy budget closure, which verify that the numerical solution conserves energy to machine-precision tolerances.
• Plume excess temperature, buoyancy flux, and plume spacing statistics for studies of hotspot volcanism.
• Dynamic topography and geoid anomaly predictions for comparison with satellite gravity observations.
• Tracer mixing and residence time statistics for geochemical reservoir evolution studies.

Testing Methods

Discretization and Solver Strategies

Community-standard discretizations include the finite element method with mixed velocity-pressure formulations, finite volume and staggered-grid finite difference schemes, and marker-in-cell or particle-in-cell methods for sharp compositional interfaces. The resulting saddle-point linear systems are solved with geometric or algebraic multigrid preconditioners, Krylov subspace iterations such as GMRES or BiCGStab, or parallel direct sparse solvers. Well-established open-source geodynamic codes, including ASPECT, CitcomS, StagYY, TERRA, and Underworld, implement these algorithms and are routinely cross-compared in community benchmarking initiatives. Adaptive mesh refinement concentrates resolution at thermal boundary layers and plume conduits, improving accuracy at controlled computational cost.

Time Integration and Advection Schemes

Most experiments use operator splitting, treating the Stokes solve implicitly at each time step while advancing temperature explicitly with total-variation-diminishing or flux-limited advection schemes under a diffusive Courant condition. Because the mantle diffusion timescale is on the order of ten billion years while plate-scale advection occurs over tens of millions of years, simulations spanning billions of years of Earth history typically require ten thousand to several million time steps, depending on resolution and regime.

Benchmarking and Verification

Verification establishes that the equations are solved correctly and relies on an internationally recognized set of community benchmarks:

• The Blankenbach benchmark for steady, basally heated two-dimensional convection at Rayleigh numbers from 10 to the fourth power to 10 to the seventh power, with Nusselt number and root-mean-square velocity agreement expected within one percent.
• The van Keken benchmark for time-dependent internally heated convection, which tests advective accuracy through average dissipation profiles.
• The King benchmark suite for temperature-dependent viscosity with contrasts up to 10 to the fifth power.
• The Tosi benchmark for visco-plastic rheology, which validates the plate-like localization behavior central to modern plate generation models.
• The Crameri benchmark for two-dimensional cylindrical geometry, which tests curvilinear staggered-grid formulations.

In addition, systematic grid convergence studies, Richardson extrapolation, and grid convergence indices computed in accordance with ASME V&V 20 practices quantify discretization error, while the method of manufactured solutions verifies solver correctness. Where projects require independent assessment, verification documentation can be reviewed by accredited third-party laboratories operating under ISO/IEC 17025-aligned quality frameworks, and simulation workflows are commonly maintained within ISO 9001-consistent quality management systems.

Validation Against Observational Constraints

Validation establishes that the governing physics is adequate for the question posed, and draws on independent geophysical observations:

• Seismic tomography images, compared against predicted temperature anomaly patterns from wave-speed-to-temperature conversions.
• Plate motion reconstructions and space-geodetic velocity fields, compared with predicted surface velocities.
• The observed global surface heat flow and the inferred Urey ratio between radiogenic heat production and total heat loss.
• Glacial isostatic adjustment constraints on mantle viscosity stratification.
• Hotspot track geometry and petrologically inferred plume potential temperatures of 1500 to 1600 degrees Celsius against an ambient mantle value near 1350 degrees Celsius.

Uncertainty Quantification

Modern experiments quantify parameter uncertainty through Latin hypercube sampling, Monte Carlo ensembles, Gaussian-process emulator-based sensitivity analysis, and Bayesian calibration of rheological and thermodynamic parameters, transforming single deterministic runs into probabilistic predictions with confidence intervals.

Applications

Mantle convection numerical experiments directly support research and decision-making across the geosciences:

• Quantifying the driving forces of plate tectonics and testing self-consistent plate generation through visco-plastic rheology.
• Modeling mantle plume initiation at the core-mantle boundary to explain hotspot chains, large igneous provinces, and flood basalt events.
• Simulating subduction dynamics, slab stagnation in the transition zone, and whole-mantle versus layered convection scenarios.
• Reconstructing supercontinent cycles and mantle thermal evolution over gigayear timescales for paleoclimate and tectonic studies.
• Predicting dynamic topography and long-term sea-level change relevant to basin analysis and sedimentary records.
• Assessing geochemical reservoir preservation and mantle homogenization for isotope geochemistry interpretations.
• Simulating planetary interiors, including stagnant-lid convection on Mars, episodic resurfacing scenarios on Venus, and tidal heating in icy moons.
• Constraining deep temperature fields for geothermal resource assessment and nuclear waste repository thermal modeling.

FAQ

What is the Rayleigh number and why is it decisive for these experiments?

The Rayleigh number measures the ratio of thermal buoyancy forces to viscous resistance and thermal diffusion. Convection onsets only above a critical value of roughly 1708 in a basally heated plane layer, and because the Earth's mantle operates at Rayleigh numbers between one million and one hundred million, numerical experiments must capture chaotic, strongly time-dependent flow and report time-averaged statistics.

Why is the Prandtl number treated as infinite?

The mantle viscosity of approximately 10 to the power of 21 pascal-seconds yields a Prandtl number near 10 to the power of 23, meaning inertial forces are utterly negligible compared with viscous forces. The momentum equation therefore reduces to the quasi-static Stokes equation, which is solved at each time step independently of the flow history.

Which numerical codes are commonly used for mantle convection experiments?

Established community codes include ASPECT with adaptive finite elements, CitcomS for spherical shell finite element modeling, StagYY with staggered finite differences, TERRA for multigrid spherical simulations, and Underworld for particle-in-cell methods. These open-source codes are repeatedly cross-verified in published community benchmark studies, ensuring results are reproducible across independent implementations.

How can simulations lasting billions of years be trusted?

Trust is built through a three-layer protocol: verification against community benchmarks to within one percent, quantified discretization error using grid convergence indices consistent with ASME V&V 20 methodology, and validation against independent observations including seismic tomography, heat flow, plate velocities, and glacial rebound. Uncertainty quantification then converts remaining parameter ambiguity into explicit confidence bounds.

When is a compressible formulation required instead of the Boussinesq approximation?

The Boussinesq approximation is adequate for many two-dimensional and moderate Rayleigh number studies, but whole-mantle realistic models require the anelastic liquid approximation to capture the adiabatic temperature gradient of several hundred kelvin, compressional heating, and depth-dependent thermodynamic properties that materially affect plume temperatures and lower mantle viscosity.

How computationally expensive is a typical experiment?

Two-dimensional benchmark-style studies can run in hours on a workstation, whereas fully three-dimensional spherical simulations with thermochemical tracers may consume hundreds of thousands to millions of core-hours on high-performance computing clusters. Adaptive mesh refinement and scalable multigrid solvers are the principal strategies for keeping state-of-the-art experiments within practical budgets.