A local route through a global problem
A proposed ocean-model method matched the study’s exact projection reference on reported diagnostics in selected numerical tests. Called AC/DC, it is designed to replace the global pressure solve with a local calculation down each vertical column and a horizontal relaxation of the remaining residual. The paper says that design avoids an additional global solve and extra global communication beyond existing barotropic machinery, with overhead of about 1.2 times a hydrostatic step, independent of resolution.
The method combines artificial-compressibility relaxation with a direct column solve. In ordinary terms, the vertical part of the pressure calculation is handled by an exact tridiagonal solve, while only the horizontal residual is relaxed. The authors present that division as a way to add non-hydrostatic dynamics beyond a hydrostatic model without adding a global solve.
A second claim concerns what the equations preserve. The formulation reports a density-weighted energy conservation structure, exact conservation of carried tracer content with uniformly bounded representation error, and an exact tracer-variance budget. Mixing diagnostics remain computable within that structure.
The theory also predicts a large reduction in divergence error on grids with different horizontal and vertical spacings. Relative to pure artificial compressibility, the predicted AC/DC error is (Δz/Δx)^2 times the pure-AC error. At a grid ratio of Δx/Δz = 100, the reported example gives an AC/DC error of 10^-4 of the pure-AC error. A separate dispersion analysis reports leading-order agreement in 1/α̃ for both methods and an AC/DC-to-pure-AC error ratio of kx²/kz².
The tests stayed close to the reference
The numerical program compared AC/DC with an exact discrete projection reference on the same mesh and time step in cases labelled E1, E2 and E3. The detailed cases were a lock-exchange gravity current, internal-gravity-wave dispersion and open-ocean deep convection. They were representative numerical tests, not a comprehensive global ocean simulation.
In the lock-exchange test, the two pressure treatments were closer to each other than the two reported meshes were on the front-Froude diagnostic, the study’s measure of the advancing front. The AC/DC–projection differences were 2.2 × 10^-4 and 1.8 × 10^-4 at the two resolutions, while changing the mesh shifted the result by 2.1 × 10^-2. On that measure, mesh choice had the larger reported effect.
Budget checks were similarly tight. Carried tracer content was conserved to 1 × 10^-15 relative on the projection run and to zero on AC/DC, while pseudo-mass agreed with the flux-form balance to 1.2 × 10^-10. The lock-exchange calculation also reported spurious mixing at 12 times the prescribed diffusivity on the coarser mesh and 3.4 times on the finer mesh. Even so, AC/DC and projection agreed on the reported Kexpl diagnostic to 0.41% and 0.03%, and on Knum to 0.65% and 0.09%, across the two meshes.
The internal-wave test checked the analysis more directly. At α̃ = 1, the operator-theoretic AC/DC measurement reproduced the analytical proposition for all four reported modes. The paper states an AC-closure error on the order of 10^-5 at that test value and on the order of 10^-10 at a production parameter of about α̃ = 4 × 10^4. The latter is a stated extrapolation; the direct measurement was made at α̃ = 1.
Deep convection exposed the trade-off
Deep convection exposed the method’s parameter sensitivity. At the highest tested α, the AC/DC run completed, whereas the lower-α sweep developed runaway acoustic instability and NaNs. At t = 0.04, AC/DC and projection both gave a plume depth of 0.8 and minimum T′ of −1.00. AC/DC’s maximum vertical speed was 2,105, versus 1,963 for projection. The depth and minimum-temperature diagnostics agreed, while the reported maximum speeds differed.
Paired deep-convection runs showed that when α differed by a factor of nine, the observed ratios for the governed quantities ranged from 8.94 to 9.01, consistent with the reported 1/α law. The analysis describes that relationship as measured in one configuration and resolution, not proved as a general rule.
The speed question remains open
The cost claim is best read as an estimate. Pass-based accounting put AC/DC’s central overhead at 1.9–2.2 F1, or 1.8–2.6 F1 under a wider counting convention, against a hydrostatic baseline of about 12 F1. In the paper’s tracer-equivalent work unit, the ratio was near 1.2. These figures are scale estimates, not a general timing measurement or a source-based projection cost.
One reported 5,200-step timing run pointed the other way at its tested setting. Pressure treatment took 348.9 ms per step for AC/DC and 247.0 ms for projection, making projection 1.41 times faster. Total time was 428.6 ms per step for AC/DC versus 327.2 ms for projection, a 1.31-times speed advantage for projection. The run used one laptop and one implementation, so it does not establish a universal speed ranking.
The variable-resolution proposal is intended to work across coarse and refined regions with one equation set, without an explicit interface between hydrostatic and non-hydrostatic equations. Its stated limiting behavior is hydrostatic in coarse regions without an explicit switch. This is an argument and scaling estimate, not a comprehensive global demonstration.
The evidence boundary is clear. The preprint limits itself to formulation, theoretical analysis and representative tests; a comprehensive ocean simulation is outside its scope, and the Boussinesq approximation is left for future work. The supplied analysis does not provide a one-to-one performance comparison with a production projection method; its main projection-cost comparison is an estimate based on pass counts and global reductions. The evidence supports a method under test, not production-scale global performance.
The open question is whether AC/DC remains robust in a comprehensive global telescoping simulation and how its cost compares with production projection methods. The preprint leaves both questions outside the reported work.
Paper data and sources
Original title: A Computational Model for Global Ocean Dynamics at all Scales
Authors: Peter Korn
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text