Flux CD: Central Differential Flux Solver

• Flux CD is a numerical flux solver that combines central differencing with a carefully scaled upwind-biased dissipation term to reduce spurious pressure oscillations.
• The method is integrated in a hybrid framework, employing primitive-variable updates in smooth regions while switching to conservative shock-capturing where needed.
• CDHD with Flux CD demonstrates a reduction in energy conservation error by up to two orders of magnitude compared to first-order methods, ensuring robust transcritical flow simulations.

Searching arXiv for papers clarifying the term “Flux CD” and related usages. First, searching for the CDHD flux paper associated with “Flux CD”. Searching arXiv for FLUX architecture as another possible interpretation of “Flux”. Searching arXiv for Cd-doped flux-growth literature matching “flux Cd”. “Flux CD” is not a standardized standalone term across arXiv literature, but in a numerical-methods context it most directly corresponds to the Central Differential flux with High-Order Dissipation (CDHD), introduced as a numerical flux treatment for the primitive-variable / pressure-evolution part of a hybrid transcritical-flow solver. CDHD combines a central flux for advection with a minimal, upwind-biased dissipation term, with the stated purpose of suppressing spurious pressure oscillations in transcritical flows while reducing the energy conservation error associated with primitive-variable formulations (Xu et al., 29 Aug 2025).

1. Definition and scope

CDHD denotes Central Differential flux with High-Order Dissipation. It is presented as a numerical flux solver for the primitive-variable discretization used in smooth transcritical regions, rather than as a universal standalone shock-capturing flux. Its target regime is the class of transcritical flows in which strong thermodynamic nonlinearities, especially near the pseudo-boiling/Widom line, generate spurious pressure oscillations and can destabilize conventional conservative schemes (Xu et al., 29 Aug 2025).

The method is explicitly framed against two established numerical tendencies. On one side, conventional conservative schemes can suffer from nonphysical pressure oscillations under real-gas nonlinearities. On the other, primitive-variable approaches, including pressure-evolution formulations and double-flux-type methods, are often more robust with respect to pressure equilibrium but are not strictly energy-conserving, so total-energy error accumulates over time and can distort shock location and shock strength. CDHD is introduced as an attempt to retain the robustness of primitive-variable formulations while reducing both nonphysical oscillations and the associated energy-conservation defect (Xu et al., 29 Aug 2025).

This positioning is important for terminology. The “CD” in the phrase refers to the method’s central differential character, and the flux is not purely central: it is a composite construction whose central convective approximation is augmented by a carefully scaled dissipative stabilization term.

2. Motivation in transcritical-flow simulation

The motivating difficulty is the simulation of transcritical flows under strong real-gas effects. Near the pseudo-boiling/Widom line, thermodynamic properties vary sharply with temperature and pressure. In that regime, purely conservative formulations may exhibit spurious pressure oscillations and loss of robustness. Primitive-variable formulations alleviate those oscillations, but the tradeoff is energy conservation error (Xu et al., 29 Aug 2025).

CDHD is therefore designed for the primitive-variable / pressure-evolution branch of a hybrid algorithm. A plausible implication is that the method is best understood not as an alternative to conservative shock capturing, but as a specialized smooth-region operator for real-fluid thermodynamics. That interpretation is consistent with the paper’s architecture: CDHD is embedded inside a hybrid framework that uses conservative updating where Rankine–Hugoniot fidelity is required and primitive-variable updating where pressure equilibrium and low oscillation are the dominant concerns (Xu et al., 29 Aug 2025).

The paper explicitly states that, relative to a first-order DOTRS-type primitive method, the proposed method reduces the energy conservation error by about two orders of magnitude. That claim defines the numerical significance of CDHD more precisely than the name alone: it is a stabilization mechanism for primitive-variable transport with improved energy behavior, rather than a generic high-order Riemann solver.

3. Primitive-variable formulation

The primitive-variable equations are written in quasi-linear form as

$$\frac{\partial \mathbf{V}}{\partial t} + B(\mathbf{V})\frac{\partial \mathbf{V}}{\partial x}=0,$$

with primitive variables

$\mathbf{V}=[\rho,u,p]^T,$

and Jacobian

$$B(\mathbf{V})= \begin{bmatrix} u & \rho & 0 \ 0 & u & 1/\rho \ 0 & \rho a^2 & u \end{bmatrix},$$

where $a$ is the sound speed obtained from the real-gas equation of state (Xu et al., 29 Aug 2025).

Within this formulation, the spatial operator is split into an accuracy component and a stability component:

$$\mathbf{L}_v = \underbrace{\mathbf{L}_v^{\text{central}}}_{\text{accuracy}} + \underbrace{\mathbf{L}_v^{\text{diss}}}_{\text{stability}}.$$

This decomposition is the core structural idea of CDHD. The central term supplies the main convection approximation; the dissipative term is intentionally small in smooth regions and acts only to suppress oscillations and improve robustness (Xu et al., 29 Aug 2025).

The formulation presumes a real-gas EOS sufficiently smooth for the primitive Jacobian and its eigenstructure to remain well behaved along the dissipation path integral. This assumption is not incidental: the path-integral construction used for the dissipative term depends on that smooth eigenstructure.

4. Central differential term and high-order dissipation

The central part is a finite-difference approximation of the primitive-variable gradient using WENO-5 reconstructed interface values:

$$\left.\partial_x \mathbf{V}\right|_i \approx \frac{\mathbf{V}^{-}_{i+1/2}-\mathbf{V}^{+}_{i-1/2}}{\Delta x},$$

so that

$$\mathbf{L}_v^{\text{central}(i)} = -B(\mathbf{V}_i)\, \frac{\mathbf{V}^{-}_{i+1/2}-\mathbf{V}^{+}_{i-1/2}}{\Delta x}.$$

Although the reconstruction is fifth-order in smooth regions, the paper emphasizes that the two-point midpoint differencing makes the central gradient operator formally second-order overall (Xu et al., 29 Aug 2025).

The dissipation term is constructed from a path-integral form of the Dumbser–Osher–Toro Riemann solver (DOTRS):

$$\mathbf{H}_{i+1/2}^{\pm} \approx \left(\int_0^1 B^{\pm}(\Psi(s))\,ds\right)\left(\mathbf{V}_R-\mathbf{V}_L\right), \qquad \Psi(s)=\mathbf{V}_L+s(\mathbf{V}_R-\mathbf{V}_L),$$

with

$B^{\pm}=\frac12(B\pm |B|).$

The dissipative contribution is then

$$\mathbf{L}_v^{\mathrm{diss}(i)} = -\frac{1}{\Delta x}\left(\mathbf{H}_{i+1/2}^{-}+\mathbf{H}_{i-1/2}^{+}\right).$$

In practice, the path integral is evaluated with a 3-point Gauss–Legendre quadrature (Xu et al., 29 Aug 2025).

The intended balance is explicit: the dissipation is upwind-biased but minimal. It is designed to be strong enough to damp the troublesome transcritical oscillations, yet small enough not to dominate the scheme in smooth regions. This is the defining feature that distinguishes CDHD from a purely central primitive update and from a more strongly upwind primitive solver.

5. Hybrid role and shock treatment

CDHD is not used in isolation. The paper embeds it in a hybrid framework in which a Primitive Variable Riemann Solver (PVRS) acts as a shock sensor. Where conservative shock capturing is needed, the method switches to a standard conservative finite-volume update with WENO-5 reconstruction and a Roe-type real-gas flux. In smooth transcritical regions, it uses the CDHD primitive-variable update (Xu et al., 29 Aug 2025).

This partition of responsibilities is central to the method’s interpretation. CDHD is intended for smooth but strongly nonlinear real-fluid regions where pressure equilibrium and oscillation control matter most. The conservative branch is retained to ensure that shocks satisfy the Rankine–Hugoniot conditions and are propagated with the correct speed and strength. This suggests a two-regime solver architecture: primitive-variable smooth-region transport plus conservative discontinuity resolution.

A common misconception would be to treat CDHD as a general shock-capturing flux. The paper does not support that reading. It states instead that the method is designed for the primitive-variable portion of the hybrid method and is switched away from near discontinuities (Xu et al., 29 Aug 2025).

6. Accuracy, numerical behavior, and limitations

The formal accuracy discussion is tied directly to the scaling of the dissipative term. For smooth flows, WENO-5 interface reconstruction gives

$\mathbf{V}=[\rho,u,p]^T,$0

so the jump satisfies

$\mathbf{V}=[\rho,u,p]^T,$1

Because the DOTRS fluctuation is proportional to this jump,

$\mathbf{V}=[\rho,u,p]^T,$2

which is asymptotically smaller than the leading $\mathbf{V}=[\rho,u,p]^T,$3 truncation error of the central derivative itself. The dissipation therefore stabilizes the computation without changing the formal order dictated by the central approximation (Xu et al., 29 Aug 2025).

The reported numerical behavior is correspondingly structured. In a 1-D transcritical advection test, CDHD with WENO-5 preserves density, temperature, velocity, and pressure profiles over many convection periods, while a first-order DOTRS-based method smears sharp thermodynamic gradients. The relative energy conservation error stays below about 0.04% at the final time, whereas the first-order DOTRS scheme grows to roughly 10%. The paper also reports a diagnostic counterexample: if the dissipative term is removed and only the central differential flux is kept, the solution becomes unstable and develops growing spurious oscillations, especially in density and temperature near steep gradients. In 1-D shock-tube tests and in 2-D droplet advection, the hybrid method remains robust, captures shocks appropriately, and preserves smooth transcritical structures with very little numerical diffusion (Xu et al., 29 Aug 2025).

Several limitations are stated explicitly. The primitive update is only second-order in smooth regions because of the chosen central differencing structure. If higher order is desired, the central derivative would need to be replaced by a higher-order central operator or a different split-flux construction. The method is also intended for smooth regions where reconstructed left and right states remain close enough for the dissipation to stay high-order small. Finally, the derivation relies on a sufficiently smooth real-gas EOS and a smooth eigenstructure along the DOTRS path.

7. Other arXiv usages of similar terminology

The string “Flux CD” is lexically ambiguous outside the transcritical-flow context. One unrelated usage appears in the materials paper on Cd-doped $\mathbf{V}=[\rho,u,p]^T,$4 single crystals grown by a high-temperature flux method using a $\mathbf{V}=[\rho,u,p]^T,$5 flux in a 5:3 molar ratio. There, “flux” refers to crystal-growth chemistry rather than numerical fluxes, and the study reports that Cd doping gradually suppresses superconductivity, with zero resistivity at $\mathbf{V}=[\rho,u,p]^T,$6 for $\mathbf{V}=[\rho,u,p]^T,$7, while the superconducting anisotropy remains similar to that of undoped crystals (Nagao et al., 2017).

A second distinct usage appears in the reverse-engineering report on FLUX.1, where “Flux” denotes a rectified-flow, transformer-based latent image generator with dual text encoders, explicit token spatial IDs, and a hybrid of 19 Double-Stream Transformer blocks followed by Single-Stream blocks. In that literature, “Flux” refers to a text-to-image model family rather than to a computational-fluid-dynamics flux function (Greenberg, 13 Jul 2025).

Accordingly, the most precise encyclopedic reading is that Flux CD, when used in numerical PDE discourse, refers to CDHD: a central differential flux augmented by high-order dissipation for the primitive-variable component of a hybrid transcritical-flow solver. In other domains, the same word sequence can denote unrelated concepts involving crystal-growth fluxes or the FLUX generative-model family, and the intended meaning depends entirely on disciplinary context.