High order tracer variance stable transport with low order energy conserving dynamics for the thermal shallow water equations
David Lee, Kieran Ricardo, Tamara Tambyah · 2025
Read High order tracer variance stable transport with low order energy conserving dynamics for the thermal shallow water equations by David Lee, Kieran Ricardo, Tamara Tambyah (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-20181.txt.
Introduction
Lee, Ricardo, and Tambyah marry high-order tracer transport to low-order fluid dynamics. Their 2025 paper targets the thermal shallow water equations. Weather models need both speed and faithful tracer motion. Low-order dynamics solvers converge fast and stay cheap. But crude transport smears tracers excessively.
A discontinuous Galerkin scheme handles material tracer transport. It provably conserves tracer variance, or damps it with upwinding. The dynamics core stays a low-order mixed finite element method. That core conserves energy by construction. Coupling runs through the dynamics solver's multigrid hierarchy. High-order basis functions sit at the low-order mesh's quadrature points. Standard test cases verified consistency and conservation. Turbulent thermal instability tests showed the payoff. Solutions stayed richer and less diffusive than pure low-order runs. Stability held without excessive tracer smearing. Overall accuracy remains capped by the low-order dynamics. The trick generalizes to other Hamiltonian transport systems.
Catalog record
| CREATOR | David Lee, Kieran Ricardo, Tamara Tambyah |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-20181
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
High order tracer variance stable transport with low order energy conserving dynamics for the thermal shallow water equations David Lee, Kieran Ricardo, Tamara Tambyah Abstract A high order discontinuous Galerkin method for the material transport of thermodynamic tracers is coupled to a low order mixed finite element solver in the context of the thermal shallow water equations. The coupling preserves the energy conserving structure of the low order dynamics solver, while the high order material transport scheme is provably tracer variance conserving, or damping with the inclusion of upwinding. The two methods are coupled via the multigrid hierarchy of the low order dynamics solver, with the basis functions of the high order transport being collocated at the Gauss-Legendre quadrature points with the low order dynamics on the finest scale multigrid mesh. Standard test cases are presented to verify the consistency and conservation properties of the method. While the overall scheme is limited by the formal order of accuracy of the low order dynamics, the use of high order, tracer variance conserving transport is shown to preserve richer turbulent solutions without compromising model stability compared to a purely low order method. Low order numerical schemes are common for implicit atmospheric solvers using finite difference [ 1 , 2 ] , finite volume [ 3 ] and finite element [ 4 ] methods, since the lower condition number of the corresponding Jacobian operator results in faster convergence and matrix assembly compared to high order methods. The approximate Jacobian operators used in these methods typically linearise around the fast acoustic, buoyancy and gravity modes [ 5 , 6 ] , which are not resolved in time, and so there is less emphasis on the accuracy of their representation compared to the slow moving vortical and inertial motions. Conversely transport terms that explicitly resolve these inertial motions are typically based on high order methods for the construction of numerical fluxes using finite volume [ 7 , 8 ] or discontinuous Galerkin [ 9 ] methods, since the upwinding of low order methods as required in order to suppress high frequency artefacts leads to excessive dissipation at low order. In the present work we harmonise these two approaches within a consistent, energy and tracer variance conserving formulation using a mixed finite element method for the low order dynamics and a high order discontinuous Galerkin method for the material transport of thermodynamic tracers. We do this within the context of the thermal shallow water equations, where the buoyancy (the thermodynamic variable) is represented in flux form for the low order dynamics. The low order buoyancy fluxes are computed with respect to the high order buoyancy as derived via discontinuous Galerkin material transport. This same high order representation of the buoyancy is also used in the low order pressure gradient term in the momentum equation, so as to preserve the anti-symmetry of the flux form buoyancy transport and the pressure gradient, and thus satisfy energy conservation [ 10 ] . Meanwhile the high order discontinuous Galerkin representation of the material transport of buoyancy is expressed as a combination of both the weak and strong form advection operators so as to provably conserve tracer variance, and hence improve model stability [ 11 , 12 , 13 , 14 ] , since tracer variance constitutes a numerical entropy of the dynamical system. However in contrast the these previous tracer variance conserving formulations, here we express the tracer variance conserving discrete advection operator in material form and not flux form, so as to couple this with the low order flux form dynamics. The integration of the high order discontinuous Galerkin thermodynamic transport into the low order dynamics solver is achieved by constructing the Lagrange polynomial basis functions so as to be orthogonal with respect to the interior Gauss-Legendre quadrature points [ 15 , 16 ] , while the low order mesh is constructed such that the cell centres are coincident with these high order quadrature points, such that there is a 1:1 mapping between the high and low order degrees of freedom. For the low order discretisation, this is done with a hierarchy of coarser meshes as part of a multigrid solver for the implicit low order problem, such that a single high order discontinuous Galerkin element, with four degrees of freedom in each dimension, is coincident with a single low order element two levels down in the multigrid hierarchy from the one on which the dynamics are represented. Multigrid methods are a popular choice for solving the Helmholtz equation that arises from block factorisation of the coupled velocity-pressure system for compressible atmospheric dynamics into a pressure equation [ 3 , 5 ] , since the smoothly varying eigenvalues of the Helmholtz operator are amenable to smoothing via Jacobi iteration across multiple spatial scales. For the mixed velocity-pressure problem Jacobi iteration is less effective owing to the more complex structure of the eigenvalues, and so more sophisticated smoothers are required that account for kernels in the vector field space [ 17 ] , such as decomposition of the vector field into its rotational and divergent components [ 18 ] or hybridised methods [ 19 ] in the context of mixed finite elements, and patch based additive Schwarz schemes in the context of collocated finite volume methods [ 20 ] . Provided that the smoothing operation is local, multigrid methods negate the need to perform global norm and dot products as required of Krylov methods, and so lead to highly scalable algorithms on large parallel computing architectures. The remainder of this article proceeds as follow: In Section 1 we introduce the thermal shallow water equations, which are used as a model of a geophysical system with a distinction between wave-like and inertial processes for the remainder of this article, and their conservation properties. In Section 2 we present the low order dynamics, high order advection variational discretisation with reference to the conservation properties described in Section 1. In Section 3 we present results for standard test cases to verify the method with respect to its consistency and conservation properties, and in Section 4 we discuss the conclusions drawn from this study. 1 Energy and tracer variance conservation for the thermal shallow water equations The rotating thermal shallow water equations may be expressed for the velocity, \boldsymbol{u} , depth h and depth weighted buoyancy, S=sh (with s being the buoyancy) as [ 10 , 21 ] \displaystyle\frac{\partial\boldsymbol{u}}{\partial t}+(\nabla\times\boldsymbol{u}+f)\times\boldsymbol{u}+\frac{1}{2}\nabla(\boldsymbol{u}\cdot\boldsymbol{u}+sh)+\frac{1}{2}s\nabla h \displaystyle=0, (1a) \displaystyle\frac{\partial h}{\partial t}+\nabla\cdot(h\boldsymbol{u}) \displaystyle=0, (1b) \displaystyle\frac{\partial S}{\partial t}+\nabla\cdot(sh\boldsymbol{u}) \displaystyle=0, (1c) where f is the Coriolis term. Alternatively, the buoyancy transport may be formulated as the material transport of s , instead of the flux form transport of S . Expanding ( 1c ) and invoking ( 1b ) we have h\frac{\partial s}{\partial t}+s\frac{\partial h}{\partial t}+h\boldsymbol{u}\cdot\nabla s+s\nabla\cdot(h\boldsymbol{u})=h\Bigg(\frac{\partial s}{\partial t}+\boldsymbol{u}\cdot\nabla s\Bigg)=0, (2) such that \frac{\partial s}{\partial t}+\boldsymbol{u}\cdot\nabla s=0 (3) may instead be used in place of ( 1c ). Using S as the prognostic variable for the buoyancy, the corresponding energy is defined over the domain \Omega\subset\mathbb{R}^{2} as \mathcal{H}=\frac{1}{2}\int h\boldsymbol{u}\cdot\boldsymbol{u}+hS\mathrm{d}\Omega, (4) for which the variational derivatives are given with respect to the prognostic variables, \boldsymbol{u},h,S , as: \displaystyle\frac{\delta\mathcal{H}}{\delta\boldsymbol{u}} \displaystyle=h\boldsymbol{u}:=\boldsymbol{F}, (5a) \displaystyle\frac{\delta\mathcal{H}}{\delta h} \displaystyle=\frac{1}{2}(\boldsymbol{u}\cdot\boldsymbol{u}+S):=\Phi, (5b) \displaystyle\frac{\delta\mathcal{H}}{\delta S} \displaystyle=\frac{h}{2}:=T. (5c) The original system ( 1