Simulating compressible flows with direct numerical simulation places extraordinary demands on the mathematics underneath. The solver must resolve an enormous spread of wavelengths in Fourier space, from the largest eddies down to the finest acoustic ripples, while simultaneously capturing shock waves and contact discontinuities without spurious oscillations. A study published in the journal Advances in Aerodynamics by Jianhua Pan, Luxin Li, Ji Yin, Wei-Gang Zeng and Cheng Jiang addresses both challenges at once. The researchers propose a series of third- and fifth-order hybrid compact least-squares central weighted essentially non-oscillatory schemes, abbreviated CLS-CWENO, applied on structured curvilinear grids within a finite volume framework. In smooth regions of the flow, compact least-squares reconstruction maintains high resolution of multiscale structures; near discontinuities, central weighted essentially non-oscillatory reconstruction takes over, giving the hybrid schemes an oscillation-free shock-capturing capability.
The compact least-squares approach builds on an idea with a long lineage. High-resolution schemes designed to broaden the resolved wavenumber range trace back to the seminal 1993 work of Tam and Webb, who optimized the dispersion properties of a seven-point finite difference scheme by minimizing integrated dispersion error in Fourier space. Separately, Lele introduced compact finite difference schemes with superior spectral properties, relating variable derivatives to variable values through sparse matrices. But compact schemes of that era were formulated for uniform grids, and extending them to non-uniform curvilinear meshes raised difficulties: coordinate-transformed schemes require sufficiently smooth meshes, and some non-uniform formulations lack conservation, risking non-physical solutions when shocks appear. The finite volume method on curvilinear grids is more robust and genuinely conservative for hyperbolic conservation laws, which is the framework the new work adopts.
A central technical contribution lies in how the reconstruction polynomials are determined. In the compact least-squares method, a polynomial is assumed for each control volume, and its unknown coefficients are found by solving local least-squares problems that penalize differences of derivatives between neighboring cells at their shared interfaces. The authors call this variant CLS-ID, for interfacial differences, and contrast it with an earlier formulation, CLS-CD, that instead penalizes differences of averaged derivatives across neighboring control volumes. The free weights in these least-squares problems were optimized to maximize the resolved bandwidth at different levels of dissipation. According to the spectral analysis, the dispersion error of the third-order CLS-ID scheme is significantly smaller than that of the CLS-CD scheme, and even better than a fifth-order upwind scheme in the reported comparisons. Notably, the authors state this is the first time, to their knowledge, that fifth-order CLS schemes have been optimized and applied to curvilinear grids.
The optimization itself blends classical and evolutionary techniques. For the third-order scheme, a straightforward grid search produced optimized weight pairs for target resolved wavenumbers of 1.0, 1.2 and 2.0, with allowable weights shrinking as the target bandwidth grows. For the fifth-order scheme, with four free parameters, a genetic algorithm minimized an objective combining the resolved wavenumber with a dissipation term scaled by a tunable hyper-parameter. The analysis shows that dissipation error approaches zero as the weights tend toward zero, at which point the schemes reduce to earlier spline reconstruction methods, so the weights were constrained to remain positive to retain enough numerical dissipation for stability. An r-index diagnostic confirmed that the optimized third-order weights for targets 1.0 and 1.2 provide adequate dissipation, while the 2.0 target leaves the scheme under-dissipated at high wavenumbers.
High resolution alone is not enough for flows containing shocks, so the team hybridized the compact schemes with CWENO reconstruction through a modified linear system. A shock detector, denoted sigma, scales each cell’s equation: it approaches one in smooth regions, where the system reduces to the pure CLS scheme, and approaches zero near discontinuities, where the gradients are instead determined by the CWENO reconstruction. The detector, proposed in the authors’ previous work, uses five cell-averaged values rather than the three used in a popular earlier detector due to Ren and colleagues. That distinction matters because, as Baeza and co-workers pointed out, three cell averages cannot distinguish a smooth local extremum from a genuine discontinuity. Under stated assumptions, the authors prove a theorem that their detector equals one up to higher-order terms in smooth regions containing first-order critical points, and becomes small only when a discontinuity is present.
The reconstruction linear systems across all control volumes couple into a block tridiagonal matrix, solved efficiently to yield an explicit polynomial per cell. The authors also show that with a particular choice of characteristic lengths the matrix becomes symmetric and the method reduces to the variational finite volume framework, so both formulations are treated collectively as CLS-FV schemes. Boundary conditions, whether Dirichlet or Neumann, enter as weakly constrained terms in the least-squares problem, keeping the matrix block tridiagonal and cheap to solve. Time advancement uses strong stability preserving Runge-Kutta methods, and the two-dimensional Navier-Stokes equations are handled dimension by dimension, with a common shock detector computed from density and pressure and an approximate Roe Riemann solver with H-correction supplying the inviscid fluxes.
The validation campaign spans one and two dimensions, linear and nonlinear equations, inviscid and viscous problems, on uniform and non-uniform curvilinear grids. In accuracy tests on the linear advection equation with smooth initial data, the hybrid schemes achieved errors orders of magnitude smaller than comparable CWENO schemes, with observed convergence rates of about three for the third-order scheme and about five for the fifth-order scheme, confirming the detector’s ability to recognize smooth first-order extrema. Under equal computational time, the hybrid schemes reached smaller errors, and the authors report the overhead in simulation time for two-dimensional cases is acceptable relative to the accuracy gains. In broadband sound wave propagation tests with energy spectra peaking at wavenumbers 4 and 12, the hybrid schemes resolved acoustic turbulent structures better than CWENO, and the proposed detector clearly outperformed the three-point detector in identifying local extrema.
Shock-dominated benchmarks reinforced the picture. In the Lax shock tube problem, density and pressure were captured without oscillations, and the proposed detector resolved discontinuities at least as sharply as the earlier detector, with the two performing very closely overall. In the Shu-Osher problem, where a moving shock interacts with a sinusoidal density disturbance to generate multiscale structures, an instructive result emerged: the third-order coefficients optimized for the widest bandwidth performed worse than those optimized at a narrower target, because insufficient high-wavenumber dissipation allowed oscillations, a direct confirmation that resolution and robustness must be balanced. In a two-dimensional Riemann problem on a 400 by 400 grid, the hybrid schemes captured shock diffraction patterns, Mach stems and Kelvin-Helmholtz roll-ups with high resolution, though reduced dissipation in the fifth-order scheme allowed small numerical disturbances to amplify into asymmetry, a phenomenon the authors note has been observed in other studies as well.
A viscous shock tube case, involving strong shock interaction with boundary layers and vortices at Reynolds number 200, tested the schemes on the full Navier-Stokes equations. On a 300 by 300 grid the boundary layer was not fully resolved, but on a 600 by 600 grid a secondary vortex appeared near the main vortex and the main vortex height moved closer to the reference value of about 0.165 reported by Daru and Tenaud, with density distributions confirming the better performance of the hybrid schemes. The authors conclude that the hybrid CLS-CWENO schemes robustly capture discontinuities while resolving flow structures across a broad bandwidth, making them promising candidates for direct numerical simulation of compressible flows. The study is open access, and the data supporting the findings are available from the corresponding author upon reasonable request.
Subject of Research: Development and validation of hybrid compact least-squares and CWENO numerical schemes for hyperbolic conservation laws on structured curvilinear grids
Article Title: Hybrid compact least-squares and central weighted essentially non-oscillatory schemes for hyperbolic conservation laws on structured curvilinear grids
Article References: Pan, J., Li, L., Yin, J., Zeng, W.-G., & Jiang, C. (2026). Hybrid compact least-squares and central weighted essentially non-oscillatory schemes for hyperbolic conservation laws on structured curvilinear grids. Advances in Aerodynamics, 8(1), Article 19. https://doi.org/10.1186/s42774-025-00246-z
Image Credits: AI Generated
DOI: 10.1186/s42774-025-00246-z
Keywords: numerical methods, compressible flows, shock capturing, compact schemes, CWENO, finite volume method, curvilinear grids, spectral optimization, Navier-Stokes equations, hyperbolic conservation laws, shock detector, direct numerical simulation
News Source: Denise Maddox. (October 29, 2026). New Hybrid Numerical Schemes Capture Shocks and Fine Structures in Compressible Flows. Scienmag.
