Fifth-Order Compact Gas-Kinetic Scheme

• The paper introduces a fifth-order compact gas-kinetic scheme that integrates BGK-based evolution and local derivative updates to achieve high-order spatial accuracy.
• It employs a compact Hermite/HWENO reconstruction that blends cell averages with gradient information, ensuring efficient shock capturing.
• The method couples a two-stage fourth-order temporal discretization with adaptive nonlinear stabilization, balancing accuracy and robustness near discontinuities.

Searching arXiv for recent and foundational papers on fifth-order compact gas-kinetic schemes and closely related CGKS work. A fifth-order compact gas-kinetic scheme is a high-order finite-volume or closely related compact discretization for compressible flow in which the numerical flux is generated from a time-dependent gas-kinetic evolution model, while the compactness of the spatial reconstruction is obtained by evolving not only cell-averaged conservative variables but also additional local derivative information derived from the same interface solution. In the literature, the term does not denote a single canonical algorithm. Rather, it refers to a family of methods that share several structural features: a BGK-based interface evolution, compact Hermite-type or HWENO-type reconstruction using cell averages together with gradients or line-averaged derivatives, nonlinear shock-robust reconstruction or blending near discontinuities, and typically a two-stage fourth-order temporal discretization. Explicit fifth-order compact realizations are now documented for structured-grid gas-kinetic schemes used in implicit large-eddy simulation, compact fifth-order CGKS formulations on 3D structured meshes, and adaptive-stencil compact variants that retain fifth-order accuracy in smooth flow while degrading toward lower-order robust behavior near shocks (Zhao et al., 2022, Yang et al., 27 Aug 2025, Mu et al., 19 Aug 2025).

1. Defining characteristics and scope

The defining distinction of a compact gas-kinetic scheme, as repeatedly emphasized in the compact GKS literature, is that the same time-dependent interface gas distribution function provides both the fluxes required for conservation updates and the interface conservative variables needed to update gradients or slope-like quantities. Compactness is therefore not achieved by an implicit compact finite-difference relation, but by enriching the local reconstruction data with interface-evolved information or cell-averaged derivatives, so that high-order reconstruction can be carried out on a short stencil (Zhao et al., 2019, Zhao et al., 2019).

Within that general class, the fifth-order case occupies an intermediate position. Earlier compact GKS papers developed sixth- and eighth-order structured-grid schemes with the same architecture but did not write down an explicit fifth-order formulation (Zhao et al., 2019, Zhao et al., 2019). By contrast, later work gives explicit fifth-order compact constructions. In the structured-grid turbulence context, the fifth-order compact scheme denoted HGKS-C5T5 uses fifth-order compact reconstruction in the face-normal direction together with fifth-order linear tangential reconstruction, so it is not fully compact in all directions (Zhao et al., 2022). A different structured 3D compact method, denoted CGKS-5th, reconstructs a quartic polynomial over a compact stencil containing the target cell, six face-neighboring cells, and twelve edge-neighboring cells, and supplements cell averages with cell-averaged gradients, line-averaged derivatives, and edge-cell directional derivatives (Yang et al., 27 Aug 2025). A more recent adaptive framework presents the fifth-order compact case as ASE-DFF(5,3)-CGKS, where the fifth-order compact polynomial is the base reconstruction and a discontinuity feedback factor controls adaptive degradation near shocks (Mu et al., 19 Aug 2025).

The expression “fifth-order compact gas-kinetic scheme” must therefore be used carefully. Some papers that are highly relevant to compact GKS methodology are not fifth-order at all. Several rotating-frame, sliding-mesh, ALE, unstructured-mesh, GPU, implicit, and p-multigrid compact GKS papers are explicitly third-order in space, although they retain the characteristic compact-GKS ingredients and are important extensions of the framework (Zhang et al., 2022, Zhang et al., 2024, Yang et al., 2023, Ji et al., 2021, Liu et al., 8 Sep 2025). Likewise, the fifth-order high-order gas-kinetic schemes built on WENO or TENO reconstructions are high-order gas-kinetic schemes, but not compact in the strict compact-GKS sense because they rely on wider finite-volume stencils rather than compact mean-plus-derivative closure (Ji et al., 2017, Mu et al., 2023).

2. Kinetic foundation and interface evolution

The common kinetic foundation is the BGK model,

$f_t+\boldsymbol{u}\cdot \nabla f=\frac{g-f}{\tau},$

or its 2D or 3D variants, where $f$ is the gas distribution function, $g$ is the local Maxwellian equilibrium state, and $\tau$ is the collision time (Yang et al., 27 Aug 2025, Zhao et al., 2022, Ji et al., 2017). The macroscopic conservative variables are obtained as moments,

$$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$

and the fluxes as

$$\boldsymbol{F}=\int \boldsymbol{u} f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$

with the usual collision-invariant vector $\boldsymbol{\psi}$ (Yang et al., 27 Aug 2025, Yang et al., 29 Jun 2026).

At a cell interface, the method uses the BGK integral solution. In a representative 3D form,

$$f(\boldsymbol{x}_{G},t,\boldsymbol{u},\boldsymbol{\xi})=\frac{1}{\tau}\int_0^t g(\boldsymbol{x}',t',\boldsymbol{u}, \boldsymbol{\xi})e^{-(t-t')/\tau}\mathrm{d}t'+e^{-t/\tau}f_0(-\boldsymbol{u}t,\boldsymbol{\xi}),$$

so the interface state is a relaxation-driven combination of transported non-equilibrium initial data $f_0$ and equilibrium evolution $g$ (Yang et al., 29 Jun 2026, Yang et al., 27 Aug 2025). The compact GKS literature repeatedly treats this as the essential physical mechanism behind the method’s ability to combine low dissipation in smooth flow with robust behavior near discontinuities (Zhao et al., 2019, Zhao et al., 2019).

For practical computations, the integral solution is expanded into an explicit time-dependent interface distribution involving equilibrium and non-equilibrium contributions. In CGKS-5th, the second-order explicit interface distribution includes an equilibrium interface state $f$0, reconstructed left and right equilibrium states $f$1 and $f$2, derivative coefficients $f$3, $f$4, $f$5, $f$6, and the Heaviside transport splitting $f$7 (Yang et al., 27 Aug 2025). The fifth-order wall-modeled CGKS framework summarizes the same structure and makes explicit that the interface solution yields not only fluxes but also time derivatives of the flux and interface conservative variables, which are central to compact reconstruction (Yang et al., 29 Jun 2026).

A recurrent point in the literature is that this interface evolution is more informative than a first-order Riemann solver. The family-of-high-order-GKS paper states that beyond the first-order Riemann solver, a high-order gas evolution model seems necessary for the development of high-order schemes, and exploits this property to construct fifth-order time-accurate gas-kinetic schemes, although those schemes remain non-compact in space (Ji et al., 2017). In compact GKS, the same observation becomes the basis for a compact spatial update mechanism.

3. Compactness through evolved means and derivatives

The compactness mechanism is the central differentiator of the method family. Standard wide-stencil finite-volume WENO methods reconstruct high-order interface data from cell averages alone and therefore require geometrically large stencils, especially on unstructured meshes (Zhao et al., 2020). In compact GKS, the interface gas evolution supplies time-accurate interface conservative variables, and those are used to update cell-averaged gradients or line-averaged derivatives, so that the next-step reconstruction uses both averages and derivative moments on a short stencil (Zhao et al., 2019, Yang et al., 27 Aug 2025).

In the earlier compact higher-order framework, the cell-averaged slope in 1D is updated by

$f$8

or equivalently by a Gauss-theorem interpretation of the cell-averaged derivative (Zhao et al., 2019, Zhao et al., 2019). This idea generalizes in 2D and 3D through Gauss or Gauss–Green formulas for cell-averaged gradients. The literature stresses that there is no separate governing equation for these slopes, unlike certain HWENO or DG variants; rather, the slope update comes from the same physical interface solution used for fluxes (Zhao et al., 2019).

In CGKS-5th on structured meshes, the reconstruction data for a target cell $f$9 are partitioned into three groups: $g$0

$g$1

$g$2

and these are used in a compact fifth-order reconstruction over the cell stencil

$g$3

(Yang et al., 27 Aug 2025). Here the phrase “compact” refers not to minimal topological width in an absolute sense, but to the use of nearby cells together with derivative information produced by the interface evolution, rather than expansion to distant wide stencils.

A related but simpler compact fifth-order mechanism appears in HGKS-C5T5. There, the normal-direction reconstruction uses immediate-neighbor cell averages $g$4 and cell-averaged derivatives $g$5, while the derivatives themselves are recovered through Newton–Leibniz from the evolved interface values,

$g$6

The paper explicitly contrasts this with the non-compact fifth-order normal reconstruction that would use $g$7 (Zhao et al., 2022).

This suggests a general interpretation of compactness in gas-kinetic schemes: the spatial order is supported not by pulling information from ever larger geometric neighborhoods, but by augmenting the local representation with moments generated naturally by the kinetic evolution. That interpretation is explicit in the structured compact GKS framework and also in unstructured-mesh compact GKS discussions, even when the formal order is third rather than fifth (Zhao et al., 2020, Yang et al., 2023).

4. Fifth-order spatial reconstruction

The canonical high-order reconstruction in CGKS-5th is a quartic polynomial,

$g$8

with zero-mean basis functions $g$9 written in scaled local coordinates about the cell centroid (Yang et al., 29 Jun 2026, Yang et al., 27 Aug 2025). The coefficients are obtained by a constrained least-squares system that enforces cell-average consistency on the compact fifth-order cell stencil, while matching gradients in face-neighbor cells, directional derivatives in edge-neighbor cells, and a line-averaged derivative inside the target cell: $\tau$0

$\tau$1

$\tau$2

$\tau$3

The scaling parameters $\tau$4 are introduced to improve conditioning (Yang et al., 27 Aug 2025).

A different explicit fifth-order compact formula appears in ASE-DFF(5,3)-CGKS. In 1D, the fifth-order compact reconstruction uses a three-cell big stencil and the cell-averaged derivatives on those cells. The defining constraints are

$\tau$5

$\tau$6

and the resulting interface formulas at $\tau$7 include

$\tau$8

together with explicit first- and second-derivative formulas needed for the gas-kinetic flux (Mu et al., 19 Aug 2025).

In HGKS-C5T5, the fifth-order compactness is only in the normal direction. The paper states that the normal reconstruction is fifth-order HWENO, while the tangential directions still use fifth-order linear reconstruction (Zhao et al., 2022). This hybrid structure is technically important because it shows that “fifth-order compact gas-kinetic scheme” may refer to a directionally compact construction rather than a fully compact multidimensional stencil.

A common misconception is that fifth-order compact GKS must always be fully compact in every coordinate direction. The literature does not support that view. Structured turbulence-oriented work explicitly combines compact normal reconstruction with non-compact tangential reconstruction (Zhao et al., 2022), whereas later 3D CGKS-5th work presents a more globally compact 3D reconstruction using face and edge neighbors with derivative data (Yang et al., 27 Aug 2025).

5. Nonlinear stabilization, shock capturing, and temporal coupling

Compact gas-kinetic schemes rely on nonlinear stabilization because high-order compact reconstructions are oscillatory near shocks if left purely linear. The nonlinear machinery differs across the literature.

In HGKS-C5T5, the normal-direction compact reconstruction is fifth-order HWENO. The paper explains that for smooth flows the linear HWENO weights are used, and that the equilibrium face-averaged derivative is obtained by a linear fourth-order polynomial. At the same time, the scheme still uses fifth-order linear reconstruction in the tangential directions (Zhao et al., 2022). This construction is then embedded in a two-stage fourth-order temporal discretization.

In CGKS-5th, the nonlinear mechanism is GENO-based. The high-order quartic compact reconstruction $\tau$9 is blended with six second-order linear sub-stencil polynomials $$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$0, each reconstructed on a face-based two-cell stencil,

$$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$1

with nonlinear weights

$$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$2

where $$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$3 and $$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$4 (Yang et al., 27 Aug 2025). The paper states that in smooth regions the method approaches the high-order linear compact reconstruction, whereas near discontinuities it is dominated by a second-order ENO-like reconstruction.

ASE-DFF(5,3)-CGKS introduces a different robustification mechanism: an adaptive stencil extension with a discontinuity feedback factor. The discontinuity strength at a Gaussian point uses pressure and Mach-number jumps,

$$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$5

and a cumulative stencil measure yields a feedback factor $$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$6 that compresses the polynomial slopes and higher derivatives (Mu et al., 19 Aug 2025). The paper explicitly states that if $$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$7, the reconstruction reduces to the constant cell average and therefore to first-order behavior; if $$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$8, the full high-order polynomial is retained. This avoids expensive high-order smoothness-indicator calculations and is presented as keeping essentially first-order robustness near discontinuities (Mu et al., 19 Aug 2025).

The time discretization in most fifth-order compact GKS work is not fifth-order in time. Instead, the characteristic choice is two-stage fourth-order temporal discretization. In CGKS-5th the update is

$$\boldsymbol{Q}=\int f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$9

$$\boldsymbol{F}=\int \boldsymbol{u} f\boldsymbol{\psi}\,\mathrm{d}\Xi,$$0

with interface conservative variables evolved through corresponding stage relations (Yang et al., 27 Aug 2025). The same two-stage fourth-order structure appears in structured compact GKS, compact ALE GKS, and the turbulence-oriented compact fifth-order scheme (Zhao et al., 2022, Zhang et al., 2024).

This temporal structure clarifies another frequent ambiguity: a “fifth-order compact gas-kinetic scheme” in current usage usually means fifth-order in space and fourth-order in time, not fully fifth-order in both space and time. The fifth-order-in-time gas-kinetic schemes developed with multi-stage