Preprint · Hypersonic computational fluid dynamics

Admissible Coarse-to-Fine Continuation of Reacting Hypersonic Flow on Native Hexahedral Grids

A developed reacting-air solution is transferred to an unchanged, 41,962,496-hexahedron domain through an admissibility-checked continuation method. Seven species, two thermal energies, and shear-stress-transport turbulence are coupled through thermodynamic, acoustic, charged-transport, and nonlinear-update equations.

Abstract

A developed reacting-air solution is transferred to an unchanged, 41,962,496-hexahedron domain through an admissibility-checked continuation method. Seven species, two thermal energies, and shear-stress-transport turbulence are coupled through thermodynamic, acoustic, charged-transport, and nonlinear-update equations. A native trilinear inverse map and density-weighted turbulence interpolation locate 41,806,828 of 42,396,569 receiver nodes. The remaining 589,741 nodes retain an independently accepted fine-grid state. Native wall, thermodynamic, and turbulence recovery admits every receiver node, with maximum thermal and modal energy roundtrip errors of 2.08 × 10⁻¹⁴ and 1.44 × 10⁻¹⁴. A bounded-memory backend preserves phase ordering across 64 ownership partitions, combines cached topology with ordered graphics-processor replay in software binary64 arithmetic, and checks processor-side turbulence corrections using recomputed global and equation-wise double-precision residuals. Matched complete-update measurements yield bitwise-identical state and numerical carry; timing comparisons retain their measured workload conditions. Ten transferred fine-grid updates pass native admissibility and linear-direction checks in 568–631 seconds per update. Thermal, species, residual, and matched-coordinate compression diagnostics quantify the continuation; decreasing energy norms and increasing turbulence norms identify ongoing relaxation. The contribution integrates complete-state transfer, quantified geometric disagreement, native recovery, and reproducible full-grid execution for continued convergence and physical assessment.

Explore PALLC

PALLC demonstration