Source-linked AI summary
Bochner Stability for B-stable DIRK Schemes
Anthony E. Ramirez, Abner J. Salgado
TL;DR
The paper addresses whether U-stability and B-stability coincide for two- and three-stage DIRK schemes and whether this supports convergence under minimal regularity. It proves the equivalence, derives Bochner norm a priori estimates for B-stable methods, and applies the framework to linear evolution problems and gradient flows.
Problem
The paper examines the relationship between U-stability and B-stability for DIRK schemes and the resulting stability and convergence properties for evolution problems.
Method
The paper proves equivalence by relating positive semidefiniteness of the B-stability matrix M to that of the U-stability quadratic form Qs, then uses energy estimates and compactness.
Results
U-stability is equivalent to B-stability for two- and three-stage DIRK schemes, providing Bochner norm a priori estimates for B-stable methods.
Takeaways & Limitations
The framework proves convergence under minimal regularity for linear evolution problems in Gelfand triples and supplies discrete energy-dissipation properties for gradient flows.
Takeaways & Limitations
The discussion is limited to two- and three-stage schemes, and gradient-flow energy analysis faces stringent time-step restrictions when Hessian regularity is assumed.
Abstract
from arXiv · showhide
In Abner J. Salgado and Ignacio Tomas. Diagonally implicit Runge-Kutta schemes: discrete energy-balance laws and compactness properties. J. Number. Math., 31(4):313-341, 2023, the notion of $U$-stability for Diagonally Implicit Runge-Kutta (DIRK) schemes was introduced. Here we establish the equivalence between $U$- and $B$- (algebraic) stability for two- and three-stage DIRK schemes, which then} provides suitable Bochner norm a priori estimates for $B$-stable methods. As applications, we first prove the convergence of $U$-stable DIRK schemes for linear coercive evolution problems under minimal regularity via energy estimates and compactness. Second, we apply these discretizations to gradient flows, which allow us to derive discrete local energy dissipation inequalities and provide counterexamples that demonstrate the limitations of stagewise energy monotonicity.
1. Introduction
The paper connects U-stability with B-stability for two- and three-stage DIRK schemes, then applies the relationship to convergence under minimal regularity and gradient flows.
- Motivation: Many classical stability notions provide only L∞(0, T; H) stability, which is insufficient for convergence of the schemes considered.
- Main contribution: U- and B-stability coincide for the DIRK schemes studied, yielding Bochner norm a priori estimates for B-stable methods.The paper also states that this establishes convergence of U-stable schemes in scenarios where B-stable schemes are known to converge.
- Applications: The paper proves convergence for linear coercive evolution problems in Gelfand triples under minimal regularity using energy estimates and compactness arguments.
- Applications: For gradient flows, the discretizations yield new discrete properties while addressing energy behavior between consecutive stages.The paper investigates energy evolution from one step through its stages to the next step.
- Scope: The analysis is restricted to two- and three-stage schemes because higher stage counts provide little consistency gain.
2. Notation and preliminaries
The preliminaries define Gelfand-triple evolution problems, DIRK discretizations, and U- and B-stability, together with the energy estimates motivating their equivalence.
- Evolution problems: A Gelfand triple consists of a separable Banach state space V continuously embedded in a separable Hilbert pivot space H, with H identified with H′ and embedded in V′.
- Evolution problems: The spatial operator is assumed to satisfy p-coercivity, q-growth, and local solvability conditions.The local solvability condition gives a unique solution v ∈ V for the stated class of equations.
- DIRK schemes: DIRK schemes use a lower-triangular coefficient matrix A with positive diagonal entries and stage updates over a partition of [0, T].The stages approximate the forcing at times tn + ciτn, although these point values need not be meaningful without further approximation choices.
- Stability notions: U-stability requires positive weights bi and positive semidefiniteness of the matrix Q(s), equivalently its bilinear form Qs.This property produces local energy identities, global estimates, and dual norm bounds.
- Stability notions: For U-stable DIRK schemes, the quadrature weights νi equal the method weights bi.The equality is established for both two- and three-stage cases using consistency and the relation A⊤λ = b.
- Stability notions: B-stability requires positive weights bi and positive semidefiniteness of M := BA + A⊤B − bb⊤.For dissipative systems, the quadratic form associated with M yields a discrete contraction estimate analogous to continuous monotonicity.
3. Equivalence between U- and B-stability
For two- and three-stage DIRK schemes, U-stability and B-stability are equivalent. The proof reduces positive semidefiniteness of the B-stability matrix to that of a quadratic form associated with U-stability.
- The equivalence follows by showing that M is positive semidefinite exactly when the quadratic form Q_s is positive semidefinite.
- The proof applies the similarity-transformed matrix M̃ and compares its spectrum with a matrix representing Q_s.
- Two-stage case: U-stability is equivalent to B-stability for two-stage DIRK schemes with invertible A and positive weights.
- Two-stage case: For two stages, Q_2 is rewritten in differences of stage variables, producing a coefficient matrix K whose positive semidefiniteness matches that of M.
- Three-stage case: The three-stage calculation uses the same strategy, with a larger coefficient representation and a change-of-basis matrix.
- Three-stage case: For three-stage DIRK schemes with invertible A, U-stability is likewise equivalent to B-stability.
4. Convergence of U-stable DIRK schemes for linear evolution problems
For linear coercive evolution problems in Gelfand triples, the paper proves convergence of U-stable DIRK schemes under minimal regularity. The argument combines uniform estimates, compactness, consistency, and identification of the limit.
- Setting: The application assumes a linear coercive operator and sets p = q = 2, making M constant and local solvability valid for every γ > 0.
- Time-discrete functions: Stage sequences are converted into piecewise constant and piecewise polynomial time-dependent functions for compactness and limit passage.
- Consistency: For ε = ε(τ) tending to zero with τ, the reconstructed forcing satisfies f̃_τ → f in L2(0, T; V′).
- Compactness: U-stability and consistency provide bounds that permit extraction of subsequences converging weakly or weak-* in the relevant Bochner spaces.
- Limit identification: The reconstructed stage function, piecewise constant approximation, and discrete derivative are identified with a common limit and its time derivative.
- Convergence: Theorem 4.1 concludes that the limit solves the continuous problem and that the whole family converges as τ → 0.
5. Some remarks on DIRK schemes for gradient flows
For gradient flows, U- or B-stable DIRK schemes yield convergence and local energy-dissipation results, while stagewise energy monotonicity generally requires additional restrictions and can fail.
- Convergence: DIRK gradient-flow approximations converge under consistency, B-stability, and a uniform time step.The result concerns the piecewise constant reconstruction in L∞(0,T;H).
- Contractivity: Exact gradient flows are nonexpansive, but arbitrary Runge–Kutta discretizations need not preserve contractivity.A convex potential can be constructed for which a Runge–Kutta scheme is noncontractive for every time step.
- Regularity limitations: For practical gradient-flow energies, twice differentiability and a Hessian-Lipschitz time-step restriction are often unavailable or stringent.The paper therefore explores alternatives to this regularity-dependent route.
- Variational characterization: Each DIRK stage admits a minimizing-movements-like variational characterization and minimizes an associated stage energy.The resulting inequality compares the stage energy at the computed stage with its value at arbitrary competitors.
- Discrete energy dissipation: U-stability provides a local energy-dissipation estimate for each time step of the DIRK scheme.The estimate is formulated in Proposition 5.1 and summarized in Corollary 5.1.
- Stagewise energy monotonicity: Without time-step restrictions, U-stability alone gives little control over stagewise energy monotonicity, and counterexamples demonstrate failure in general.A two-stage, second-order, B-stable scheme is used in a scalar example to exhibit the relevant behavior.