Energy-Stable Correction Algorithm

• Energy-Stable Correction Algorithm is a numerical scheme that guarantees monotonic energy dissipation by enforcing orthogonality between resolved and unresolved scales.
• It employs dynamic small-scales and generalized-alpha time integration to eliminate spurious energy generation and ensure discrete energy decay.
• The approach enhances stabilized formulations in convection-diffusion problems and improves the reliability of finite element simulations.

An energy-stable correction algorithm is a rigorously constructed scheme or modification in numerical PDE solvers and optimization, specifically designed to guarantee that a discrete or semi-discrete energy functional exhibits monotonic decay (dissipation) in time, faithfully mirroring the physical or mathematical energy dissipation expected from the underlying continuum system. Such corrections are indispensable in stabilized finite element formulations, high-order schemes, and certain optimization flows, where classical methods can introduce spurious or indefinite energy and unphysical behavior.

1. Rationale for Energy-Stable Correction

In stabilized formulations for convection-diffusion and advection-diffusion problems, traditional approaches like SUPG, GLS, or classical variational multiscale (VMS) can produce artificial or even negative energy dissipation due to interactions between resolved and unresolved scales. This manifests as numerical energy wiggles, local energy creation, or non-monotonic global decay even when the exact PDE admits strict monotonic energy evolution. The goal of the energy-stable correction is thus to identify and remove these indefinite contributions by enforcing orthogonality or specific structure in the stabilization operator, ensuring that the discrete energy behaves correctly (Eikelder et al., 2017).

2. Multiscale Formulation and Orthogonal Projection

The foundational decomposition is the VMS (Variational Multiscale) split, where the solution $\phi$ and test function $w$ are decomposed into resolved ($\phi^h$, $w^h$) and unresolved ($\phi'$, $w'$) scales: $$\phi = \phi^h + \phi', \qquad w = w^h + w', \qquad W = W^h \oplus W'$$ Applying an $H^1_0$-orthogonal projector $\mathscr{P}^h$, the unresolved scales are chosen orthogonal in the energy inner product. Inserting this split into the weak form yields dual equations for both scales, which, in their unprojected forms, admit arbitrary sign contributions in the energy evolution. Enforcing

$$\int_\omega \kappa \nabla \phi^h \cdot \nabla \phi'\,dx = 0$$

removes indefinite energy exchange between scales, recovering physical energy behavior.

3. Conversion to SUPG and GLS Under Orthogonality

The imposition of $H_0^1$ orthogonality transforms the two-scale VMS into streamlined forms:

• SUPG-consistent variant: Omitting certain stabilization terms while maintaining orthogonality yields a SUPG-type method where the problematic indefinite energy contribution persists, though less pronounced.
• Galerkin/Least-Squares (GLS) variant: Retaining all necessary terms, the GLS form with dynamic, orthogonal small scales eliminates the indefinite sign contribution and yields strictly dissipative energy evolution.

Both forms require that the large- and small-scales be treated as separate equations and that only appropriately filtered (orthogonal) small-scales contribute to stabilization.

4. Dynamic Small-Scales and Energy-Stable Correction Schemes

The energy-stable correction is further realized by allowing small-scales to evolve dynamically, not statically. Two distinct energy-corrected stabilized schemes are constructed:

• GLSD (Galerkin/Least-Squares with Dynamic Small-Scales):

$$\begin{aligned} & (w^h, \partial_t(\phi^h + \phi'))_\Omega + (w^h, \mathbf{a} \cdot \nabla \phi^h)_\Omega + (\nabla w^h, \kappa \nabla \phi^h)_\Omega - (\mathbf{a} \cdot \nabla w^h - \kappa \Delta w^h, \phi')_\Omega = (w^h,f)_\Omega \ & \partial_t \phi' + \tau^{-1} \phi' = -(\partial_t \phi^h + \mathbf{a}\cdot\nabla\phi^h - \kappa \Delta\phi^h - f) \end{aligned}$$

where $\tau$ is a stabilization parameter.

• DO (Dynamic Orthogonal Formulation):

This approach uses a Lagrange multiplier to enforce $H_0^1$ orthogonality directly, formulating coupled equations for $(\phi^h, \phi', \sigma^h)$ in which orthogonality is never lost.

In both approaches, when tested with ($\phi^h, \phi'$), energy analysis demonstrates that only negative-definite contributions remain, ensuring unconditional energy decay.

5. Generalized-$\alpha$ Time Integration and Discrete Energy Stability

Time discretization is performed via a generalized-$\alpha$ method that can be tuned to enforce unconditional energy decay for any time-step $\Delta t$. Key parameters $(\alpha_m, \alpha_f, \gamma)$ are chosen so that, in the absence of external sources ($f=0$), the total discrete energy satisfies

$E_{n+1} \leq E_n - (\text{dissipation terms}),$

with all dissipation strictly non-negative. The correct choice of the stabilization timescale $\tau_{\text{eff}}$ (incorporating diffusive, convective, and temporal factors) ensures consistency with static-scale stabilization and formal second-order accuracy in time.

6. Algorithmic Realization

The energy-stable correction procedure is realized by explicit algorithmic workflow at each time step:

Given φ^h_n, φ'_n, ∂_t φ_n 1. Predict φ^h_{n+α_f} ← (1−α_f)φ^h_n + α_f φ^h_{n+1} [initial guess] 2. Linearize and assemble large-scale residual R^h(φ^h_{n+α_f}) including the term −(a·∇w^h−κΔw^h,φ'_{n+α_f}) 3. Solve for φ^h_{n+1}, ∂_tφ_{n+1} from the generalized-α update coupled with the small-scale ODE below 4. Update small scales via ∂_t φ'_{n+1} + τ_eff^{-1} φ'_{n+1} = −[∂_t φ^h + a·∇φ^h − κΔφ^h − f]_{n+α_f} using closed-form generalized-α update 5. Advance to next step

For the DO scheme, a Newton-type inner iteration or static condensation is used to enforce the orthogonality constraint via the Lagrange multiplier.

7. Numerical Verification and Comparative Findings

Numerical tests on the convection-diffusion problem (unit square, sharp block initial conditions, $a = (1,1)$, $\kappa=5\cdot 10^{-4}$, $32\times32$ NURBS mesh) confirm:

• Global energy: Both GLSD and DO schemes strictly exhibit monotonic energy decay matching reference solutions, while SUPG with static small-scales (SUPGS) can show nonphysical energy wiggles and negative dissipation.
• Small-scale dissipation: GLSD/DO have uniformly nonnegative contributions, while SUPGS displays negative small-scale dissipation up to $10^{-3}$.
• Orthogonality enforcement: The problematic term $(\kappa \Delta \phi^h, \phi')$ disappears in GLSD/DO (exact in DO, analytical cancellation in GLSD) but remains indefinite in SUPGS.
• Time integration: Formal second-order accuracy and exact energy decay are achieved with generalized-$\alpha$ parameter choices $(\alpha_f=\alpha_m=\gamma=1/2)$ (Crank-Nicolson).

These results demonstrate that combining dynamic small-scales, $H_0^1$-orthogonality, and energy-stable time-integration leads to stabilized methods where both local and global energy evolution are physically correct and negative dissipation artifacts are eliminated (Eikelder et al., 2017).

8. Broader Implications and Methodological Extensions

Energy-stable correction principles, as exemplified here, have found application in other stabilized high-order schemes, flux reconstruction contexts (Trojak, 2018), adaptive optimization algorithms (Liu et al., 2020, Liu et al., 2022), and advanced time-marching for gradient flows and DG methods. The central methodology—orthogonality enforcement and dynamic scale separation—provides a paradigm for constructing provably stable algorithms in other PDE and optimization frameworks where multiscale interactions risk spurious energy behavior.

9. Summary Table: Core Methods and Properties

Scheme	Key Correction	Energy Decay	Small-Scale Dissipation
GLSD	Dynamic, $H_0^1$-orthogonal scales	Unconditionally monotonic	Nonnegative
DO (Orthogonal)	Dynamic, enforced orthogonality via multiplier	Unconditionally monotonic	Nonnegative
SUPGS (static)	No dynamic small scales	Non-monotonic, can increase	Indefinite, may be negative

The energy-stable correction methodology thus acts as a rigorous foundation for stabilized formulations in convective-diffusive PDEs, enabling both optimal numerical accuracy and strict fidelity to the underlying energy laws.