Imagine you're building a house of cards on a moving train. Every time the train lurches, you need two things simultaneously: the cards must not fall over (stability), and the ones already standing must stay exactly where you placed them (balance). Most numerical methods for atmospheric simulation pick one or the other — they can be stable, or they can hold a resting atmosphere perfectly still, but doing both at once in a principled way has been an open problem. This paper solves it for the compressible Euler equations in a specific formulation that climate modelers actually want to use. The committed claim: for the first time, the authors construct discontinuous Galerkin spectral element methods (DGSEM) for the compressible Euler equations in vector-invariant form — with potential temperature as a prognostic variable — that are simultaneously entropy-stable AND well-balanced for isothermal and isentropic hydrostatic backgrounds. Previous work achieved entropy stability or well-balancedness separately, or achieved both only in the conservative-variable formulation. The vector-invariant form matters because it is the formulation used in next-generation atmospheric dynamical cores (think MPAS, the spectral element core in E3SM). The mechanism is flux differencing. The DGSEM framework lets you replace a standard numerical flux with a two-point flux that satisfies discrete analogues of continuous identities — conservation laws, entropy inequalities. The authors show that by splitting the nonconservative terms (which arise because the vector-invariant form is not in divergence form) into symmetric and antisymmetric products, they can design two-point volume fluxes that discretely conserve both thermodynamic entropy and total energy. Then they layer on a well-balancing correction: a source-term discretization that exactly cancels the gravitational flux at hydrostatic equilibrium. The result is a method that, on a mesh at rest, produces zero residual to machine precision — and under perturbation, doesn't blow up because it dissipates entropy in the right direction. On the ladder: the natural comparison is to existing entropy-stable DGSEM methods for the Euler equations in conservation form (Winters et al. 2020, Chan 2019, Ranocha 2018) and to well-balanced DG methods that use conservative variables (Waruszewski et al. 2022). This paper doesn't claim to beat those methods on raw accuracy — the point is structural. It extends the entropy-stability + well-balancing toolkit to a formulation (vector-invariant, potential-temperature-based) that prior methods couldn't handle. The numerical experiments show convergence at the expected polynomial orders, machine-precision well-balancing, and robustness on standard test cases (rising thermal bubble, density current, inertia-gravity waves) up to polynomial degree 7. Integrity is solid for a methods paper. The proofs are mathematical — discrete entropy conservation and well-balancedness are proven analytically, not merely observed empirically. The numerical experiments are standard community benchmarks (rising thermal bubble, density current, Skamarock-Klemp inertia-gravity wave). The reproducibility repository is public on GitHub with all code. The main limitation: all tests are 2D. There is no 3D experiment, no real-data validation, and no performance comparison (wall-clock time, memory) against existing dynamical cores. The milestone trajectory is clear. The method needs to be demonstrated in 3D on a cubed-sphere mesh with realistic atmospheric profiles and compared head-to-head on wall-clock cost against production dynamical cores (FV3, MPAS, the spectral element dynamical core in CAM-SE). The gap is engineering, not theory — the mathematical infrastructure is now in place. A realistic timeline is 2-4 years for a 3D implementation competitive enough to benchmark against production codes. The obvious experiment the authors did not run is a 3D cubed-sphere simulation. The honest read: this is (a) compute and engineering budget — going from 2D tensor-product elements to 3D on the sphere requires substantial implementation work. The theoretical machinery proven here carries over (the authors state this explicitly for curvilinear meshes), but the implementation hasn't been built yet. This is the next paper, and they know it.
