An Adaptive Flux Reconstruction Scheme for Robust Shock Capturing Journal: Journal of Computational Physics
Sai Shruthi Srinivasan, Siva Nadarajah · 2025
Read An Adaptive Flux Reconstruction Scheme for Robust Shock Capturing Journal: Journal of Computational Physics by Sai Shruthi Srinivasan, Siva Nadarajah (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-01564.txt.
Introduction
Sai Shruthi Srinivasan and Siva Nadarajah write on a flaw of high-order flow codes. Near a shock wave, sharp methods such as discontinuous Galerkin sprout wiggles that are not real. Entropy-stable flux reconstruction schemes curb the wiggles and keep provable entropy promises. But they pay for this with lost accuracy and small time steps.
Their fix is a lifting operator that adapts. The operator is the part of the scheme that punishes jumps at cell faces. Switching it as the flow demands keeps the scheme's provable traits, restores accuracy, and allows larger time steps. With a limiter that keeps values positive, the scheme gives sound results through shocks in hyperbolic conservation laws.
Catalog record
| CREATOR | Sai Shruthi Srinivasan, Siva Nadarajah |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-01564
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
An Adaptive Flux Reconstruction Scheme for Robust Shock Capturing Journal: Journal of Computational Physics Sai Shruthi Srinivasan, Siva Nadarajah Abstract 1 Introduction In the case of hyperbolic conservation laws, high-order methods, such as the classical DG method, experience the phenomenon of unwanted high-frequency oscillations in the vicinity of a shock. Shock-capturing methods such as artificial dissipation, solution, flux, or TVD limiting are generally used to eliminate nonphysical oscillations and provide bounds on physical quantities. For entropy-stable schemes, the additional objective would be to retain provable entropy dissipation guarantees of the underlying scheme, i.e. subcell limiting or entropy filtering [ 1 , 2 , 3 , 4 ] . The nonlinearly-stable flux reconstruction (NSFR) semi-discretization given in Eq. 7 with a suitable flux reconstruction scheme has been demonstrated to mitigate spurious oscillations in the presence of shock discontinuities and at CFL values substantially larger than the DG variant of the NSFR scheme whilst retaining the property of entropy stability [ 5 ] . NSFR schemes achieve this by introducing an alternative lifting operator for surface numerical flux penalization, albeit at the expense of accuracy. In this technical note, we present an adaptive approach to the choice of the lifting operator employed, which maintains higher accuracy and allows for larger CFL values while retaining the underlying provable attributes of the scheme. While it cannot eliminate oscillations such as the aforementioned shock-capturing methods, together with a positivity-preserving limiter, the scheme provides for solutions that are essentially oscillation-free. (\bm{M}_{m}+\bm{K}_{m})\frac{d}{dt}\hat{\bm{u}}_{m}(t)^{T}+[\bm{\chi}(\bm{\xi_{v}^{r}})^{T}\bm{\chi}(\bm{\xi_{f}^{r}})^{T}]\left[\left(\tilde{\bm{Q}}-\tilde{\bm{Q}^{T}}\right)\odot\bm{F_{m}^{r}}\right]\bm{1}^{T}+\sum_{f=1}^{N_{f}}\sum_{k=1}^{N_{fp}}\bm{\chi}(\bm{\xi}_{f,k}^{r})^{T}{W}_{f,k}\hat{\bm{n}}^{r}\cdot\bm{f}_{m}^{*,r}=\bm{0}^{T}. (1) Here, \bm{\chi}(\bm{\xi}) are the basis functions written as polynomials in reference space, with \bm{v} and \bm{f} denoting both volume and facet reference points; while \left(\tilde{\bm{Q}}-\tilde{\bm{Q}^{T}}\right) is the general hybridized skew-symmetric stiffness operator involving both volume and surface quadrature evaluations from [ 6 ] and \bm{F_{m}^{r}} is a matrix that contains the reference two-point fluxes. The boundary term is summed over N_{f} faces at N_{fp} surface quadrature points, while \hat{\bm{n}}^{r} is the facet normal in reference space and \bm{f}_{m}^{*,r} is the numerical flux. The NSFR scheme, which serves as a foundation for the novel scheme proposed in this paper, is effectively a filtered DG scheme. The filtering of the DG residual is controlled by the flux reconstruction parameter, c , which is embedded in the modified mass matrix \bm{K_{m}} , \displaystyle(\bm{K}_{m})_{ij} \displaystyle\approx\sum_{s,v,w}c_{(s,v,w)}\int_{{\Omega}_{r}}J_{m}^{\Omega}\partial^{(s,v,w)}\chi_{i}(\bm{\xi}^{r})\partial^{(s,v,w)}\chi_{j}(\bm{\xi}^{r})d{\Omega_{r}}, (2) where \partial^{(s,v,w)} represents a multidimensional index spanning the p -th order broken Sobolev-space [ 7 ] . Setting c=0 allows us to recover a DG scheme, as the lifting operator is unfiltered. On the other hand, setting the c parameter to the largest stable value, c_{+} , recovers an FR scheme that mitigates unwanted oscillations by filtering the lifting operator. The mitigation of the oscillations enhances the robustness of the scheme and allows for a higher CFL. However, the FR scheme that we recover has a greater magnitude of error than its DG counterpart, as demonstrated in [ 8 , 9 , 10 ] . As such, the ideal scheme for shock-dominated problems would employ a DG scheme in smooth regions whilst adapting the lifting operator in regions dominated by strong shocks. This would allow us to retain the DG scheme’s advantage of increased accuracy while also retaining the FR scheme’s advantages of dampened oscillations and timestep efficiency. This is the motivation behind the novel adaptive flux reconstruction (AFR) method. In this technical note, Sec. 2 outlines the preserved properties of the new scheme, Sec. 3 outlines key implementation details, including the shock sensor and timestepping method, and finally Sec. 4 presents a suite of test cases that demonstrate the advantages of the new adaptive scheme over the DG and FR schemes. The implementation and analysis of this novel scheme are summarized in Sec. 5 . 2 Adaptive Flux Reconstruction The adaptive implementation presents a lot of desirable advantages, but it is crucial that the newly proposed scheme still retains the properties of the NSFR scheme that serves as a foundation. Two crucial properties of interest are conservation and nonlinear stability. In the following sections, the two main theorems of the NSFR method presented in [ 11 ] are revisited to show that the AFR scheme conserves these properties. 2.1 Conservation The NSFR scheme possesses the crucial property of global and local conservation. Discretizations that approximate weak solutions to conservation laws must discretely satisfy conservation, and as such, the new scheme must retain this property. Prior to demonstrating that the adaptive flux reconstruction scheme satisfies local conservation, we first revisit several fundamental properties of ESFR schemes. Remark 1 . Following the derivations in Zwanenburg and Nadarajah 2016 , ESFR can be expressed as a filtered DG scheme for general three-dimensional curvilinear coordinates on mixed element types with a modified norm; provided that the matrix \bm{K} satisfy the following definition. Definition 1 . In two-dimensions, the vector correction functions \bm{g}^{f,k}(\bm{\xi}^{r})\in\mathbb{R}^{1\times d} associated with face f , facet cubature node k in the reference element, are defined as the tensor product of the p+1 order one-dimensional correction functions ( \bm{\chi}_{p+1} stores a basis of order p+1 ), with the corresponding p -th order basis functions in the other reference directions, \begin{split}\bm{g}^{f,k}(\bm{\xi}^{r})=\Big[\Big(\bm{\chi}_{p+1}(\xi)\otimes\bm{\chi}({\eta})\Big)\Big(\hat{\bm{g}}^{f,k}_{1}\Big)^{T},\Big(\bm{\chi}({\xi})\otimes\bm{\chi}_{p+1}(\eta)\Big)\Big(\hat{\bm{g}}^{f,k}_{2}\Big)^{T}\Big]\\ =[{g}^{f,k}_{1}(\bm{\xi}^{r}),{g}^{f,k}_{2}(\bm{\xi}^{r})],\end{split} (3) such that \bm{g}^{f,k}(\bm{\xi}_{f_{i},k_{j}}^{r})\cdot\hat{\mathbf{n}}^{r}_{f_{i},k_{j}}=\begin{cases}1\quad\text{if $f=f_{i}$ and $k=k_{j}$}\\ 0\quad\text{otherwise}.\end{cases} (4) Following the derivations in Zwanenburg and Nadarajah [10] on the equivalence of energy stable flux reconstruction and discontinuous Galerkin methods, the modified mass matrix introduced previously is related to the correction function formulation via \nabla^{r}\cdot\bm{g}^{f,k}(\bm{\xi}^{r})=\bm{\chi}(\bm{\xi^{r}})(\bm{M}+\bm{K})^{-1}\bm{\chi}(\bm{\xi}^{r}_{f,k})^{T}W_{f,k}. (5) Hence, NSFR schemes naturally give rise to a modified lifting operator on the surface integral, which can be expressed as (\bm{M}_{m}+\bm{K}_{m})^{-1}\sum_{f=1}^{N_{f}}\sum_{k=1}^{N_{fp}}\bm{\chi}(\bm{\xi}_{f,k}^{r})^{T}W_{f,k}\hat{\bm{n}}^{r}\cdot\bm{f}_{m}^{*,r}. (6) This modified lifting operator will be associated with a conservative scheme provided that the conditions from the following theorem are satisfied. Note that in Eq. 5 , the mass matrix, \bm{M} and FR filter matrix \bm{K} are evaluated on the reference elements as opposed to \bm{M}_{m} and \bm{K}_{m} as in Eq. 2 which includes the element Jacobian. Theorem 1 . The AFR discretization which uses the NSFR discretization in Eq. 7 is locally conserving Proof. Starting with the NSFR discretization, (\bm{M}_{m}+\bm{K}_{m})\frac{d}{dt}\hat{\bm{u}}_{m}(t)^{T}+[\bm{\chi}(\bm{\xi_{v}^{r}})^{T}\bm{\chi}(\bm{\xi_{f}^{r}})^{T}]\left[\left(\tilde{\bm{Q}}-\tilde{\bm{Q}^{T}}\right)\odot\bm{F_{m}^{r}}\right]\bm{1}^{T}+\sum_{f=1}^{N_{f}}\sum_{k=1}^{N_{fp}}\bm{\chi}(\bm{\xi}_{f,k}^{r})^{T}{W}_{f,k}\hat{\bm{n}}^{r}\cdot\bm{f}_{m}^{*,r}=\bm{0}^{T} (7) Given that \bm{\chi}(\bm{\xi}_{v}^{r})\hat{\bm{1}}^{T}=\bm{1}^{T} , we left multiply by \hat{\mathbf{1}} . As outlined in [ 11 ] , \left(\tilde{\bm{Q}}-\tilde{\bm{Q}^{T}}\right)\odot\bm{F_{m}^{r}} is skew-symmetric, its associated term vanishes and we are left with \displaystyle\hat{\mathbf{1}}(\bm{M}_{m}+\bm{K}_{m})\frac{d}{dt}\hat{\bm{u}}_{m}(t)^{T}=-\sum_{f=1}^{N_{f}}\sum_{k=1}^{N_{fp}}{W}_{f,k}\hat{\bm{n}}^{r}\cdot\bm{f}_{m}^{*,r} (8) Using \hat{\bm{1}}\bm{K}_{m}=0 , this is further simplified to, \hat{\mathbf{1}}\bm{M}_{m}\frac{d}{dt}\hat{\bm{u}}_{m}(t)^{T}=-\sum_{f=1}^{N_{f}}\sum_{k=1}^{N_{fp}}{W}_{f,k}\hat{\bm{n}}^{r}\cdot\bm{f}_{m}^{*,r} (9) which recovers local conservation. The adaptive implementation of the flux reconstruction parameter, c , contained within \bm{K}_{m} does not impact local conservation, therefore the scheme is locally conservative. ∎ Remark 2 . Using Eq. 9 , the NSFR scheme is said to be globally conservative after summing across all of the elements and using a telescopic flux. This proof also follows for the AFR scheme. 2.2 Nonlinear Stability Another key property of NSFR is nonlinear stability. Provable nonlinear stability bounds the discrete approximation within a norm to the true solution and ensures that the discretization does not diverge. Provided that positivity of thermodynamic quantities are preserved, the discrete solution remains stable. While stability does not imply convergence, it is still an integral property to conserve in the AFR scheme to ensure that the scheme is stable and satisfies the second law of thermodynamics. As per Cicchino and Nadarajah 2025 , Eq. 7 satisfies the following formulation, \sum_{n=1}^{n_{state}}\hat{\bm{v}}_{n}(\bm{M}_{m}+\bm{K}_{m})\frac{d}{dt}\hat{u}_{m,n}(t)^{T}=\sum_{n=1}^{n_{state}}\sum_{f=1}^{N_{f}}\sum_{k=1}^{N_{fp}}(\bm{\psi}_{n}(\bm{\xi_{fk}^{r}}-v_{n}(\bm{\xi_{fk}^{r}})\bm{f}_{m,n}^{*,r})\cdot\hat{\bm{n}^{r}} (10) With the appropriate boundary