Source-linked AI summary
Numerical analysis of nonlinear subdiffusion equations
Bangti Jin, Buyang Li, Zhi Zhou
TL;DR
Rigorous numerical analysis of nonlinear time-fractional parabolic equations remains limited, especially for combining nonlinear and discretization errors. The paper develops a framework using fractional discrete Grönwall inequalities, discrete maximal regularity, and nonlinear regularity theory, obtaining O(h^2) spatial error up to a logarithmic factor and O(τ^α) temporal error without extra solution regularity or compatibility conditions.
Problem
Rigorous numerical analysis of nonlinear time-fractional diffusion equations has been scarce, despite the practical importance of fractional models for anomalously slow transport.
Method
The paper combines a fractional discrete Grönwall inequality, discrete maximal regularity, and regularity theory for nonlinear equations, with the inequality established for L1 and BDF convolution-quadrature schemes.
Results
The analysis yields pointwise L2(Ω)-norm error estimates of O(h^2) up to a logarithmic factor for spatial discretization and O(τ^α) for temporal discretization.
Takeaways & Limitations
The framework supports sharp numerical convergence rates for nonlinear time-fractional diffusion equations without extra solution regularity or compatibility conditions on the problem data.
Takeaways & Limitations
The main solution theory and error analysis assume a Lipschitz continuous nonlinear source, although the error results can also hold for non-Lipschitz sources when the continuous solution is uniquely bounded.
Abstract
from arXiv · showhide
We present a general framework for the rigorous numerical analysis of time-fractional nonlinear parabolic partial differential equations, with a fractional derivative of order $α\in(0,1)$ in time. The framework relies on three technical tools: a fractional version of the discrete Grönwall-type inequality, discrete maximal regularity, and regularity theory of nonlinear equations. We establish a general criterion for showing the fractional discrete Grönwall inequality, and verify it for the L1 scheme and convolution quadrature generated by BDFs. Further, we provide a complete solution theory, e.g., existence, uniqueness and regularity, for a time-fractional diffusion equation with a Lipschitz nonlinear source term. Together with the known results of discrete maximal regularity, we derive pointwise $L^2(Ω)$ norm error estimates for semidiscrete Galerkin finite element solutions and fully discrete solutions, which are of order $O(h^2)$ (up to a logarithmic factor) and $O(τ^α)$, respectively, without any extra regularity assumption on the solution or compatibility condition on the problem data. The sharpness of the convergence rates is supported by the numerical experiments.
1. Introduction.
The paper develops a rigorous framework for nonlinear time-fractional diffusion equations, addressing a gap in numerical analysis through new discrete tools and solution regularity theory. It derives sharp spatial and temporal error estimates for finite element discretizations.
- Time-fractional parabolic PDEs model anomalously slow transport and arise in applications including fractal thermal diffusion, heterogeneous aquifers, and protein dynamics.
- The framework combines fractional discrete Grönwall inequalities, discrete maximal regularity, and regularity estimates for nonlinear equations.These tools control nonlinear contributions and combine them with linear errors to obtain global estimates.
- A general criterion is established for fractional discrete Grönwall inequalities, and verified for the L1 scheme and BDF-generated convolution quadratures up to order 6.
- The paper provides existence, uniqueness, and regularity theory for a time-fractional diffusion equation with a Lipschitz continuous nonlinear source.
- O(h^2) spatial and O(τ^α) temporal pointwise L2(Ω)-norm error estimates are obtained for semidiscrete and fully discrete solutions, respectively.The spatial estimate holds up to a logarithmic factor, without extra solution regularity or compatibility conditions on the data.
- The convergence rates are reported as sharp up to a logarithmic factor and are supported by numerical experiments.
2. Discrete Gr¨onwall’s inequality for time-fractional diffusion.
The paper develops fractional continuous and discrete Grönwall inequalities for time-stepping schemes, then gives a criterion that verifies the discrete result for L1 and BDF convolution quadrature methods.
- Continuous fractional Grönwall inequality: Fractional Grönwall inequalities are introduced as tools for analyzing nonlinear subdiffusion time-stepping schemes.The continuous inequality is formulated in a Banach space for α ∈ (0, 1) and p > 1/α.
- Discrete framework: The discrete analysis represents approximations through convolution weights and generating functions for fractional derivative schemes.The framework uses discrete Hardy-type inequalities and discrete Mikhlin multiplier results, including UMD-space assumptions.
- Backward Euler CQ: Backward Euler convolution quadrature satisfies a discrete fractional Grönwall inequality for sufficiently small time steps.The associated constants are independent of σ, τ, N, and the sequence, while depending on α, p, κ, and T.
- General criterion: A general criterion establishes discrete fractional Grönwall inequalities for generating functions satisfying specified analytic and multiplier conditions.The criterion is formulated for UMD spaces and connects other time-stepping schemes to backward Euler convolution quadrature.
- Applications: The criterion applies to the L1 scheme and BDF-generated convolution quadrature for orders 1 through 6.The L1 generating function is verified to satisfy the criterion’s required inequalities.
3. Regularity of the solution.
The paper establishes existence, uniqueness, and regularity for nonlinear time-fractional diffusion solutions with Lipschitz nonlinearities, including spatial-domain and time-derivative estimates without compatibility conditions.
- Existence and uniqueness: For Lipschitz continuous f and suitable initial data, the nonlinear time-fractional diffusion problem has a unique solution.The proof uses a contraction mapping in an exponentially weighted continuous-function space.
- Derivative estimate: The time derivative satisfies ∂_t u(t) ∈ L^2(Ω) and ∥∂_t u(t)∥_L2(Ω) ≤ c t^(α−1).The estimate captures the possible initial-time singularity for fractional evolution.
- Sharpness and assumptions: Without extra compatibility conditions, the stated regularity and mesh-independent estimates are sharp with respect to temporal Hölder continuity.The paper identifies these estimates as important for the subsequent numerical analysis.
- Time regularity: The solution has C^α([0,T]; X) regularity in time.This follows from difference-quotient estimates based on the mapping properties of the solution operators F and E.
- Spatial regularity: The solution belongs to C([0,T]; D), with bounded Au, under the abstract operator formulation.The argument combines the regularity of F and E with estimates for the nonlinear and initial-data terms.
4. Error estimates.
This section derives spatial and temporal error estimates for nonlinear time-fractional diffusion problems using linear estimates, discrete maximal regularity, and a fractional discrete Grönwall inequality. The resulting rates are sharp up to a logarithmic factor and apply to the L1 scheme and backward Euler convolution quadrature.
- Sharpness: The convergence rates are sharp with respect to solution regularity up to a logarithmic factor and are confirmed by numerical experiments.The framework focuses on the L1 scheme and backward Euler CQ, while other time-stepping schemes can be treated similarly.
- Error-analysis framework: The analysis separates the numerical error into spatial and temporal components and uses corresponding linear subdiffusion estimates as building blocks.An intermediate spatially semidiscrete Galerkin problem is introduced before analyzing the fully discrete scheme.
- Spatial discretization error: The semidiscrete Galerkin finite element solution has an L2(Ω)-norm spatial error of order O(h^2), up to the logarithmic factor ℓh.Here ℓh = log(2 + 1/h), and the estimate remains valid under the stated initial-data interpolation condition.
- Temporal discretization error: The fully discrete nonlinear solution satisfies the pointwise temporal error estimate max 1≤n≤N ∥uh(tn) − un h∥L2(Ω) ≤ cτ^α.The estimate holds for both the L1 scheme and backward Euler CQ under a Lipschitz continuous source term.
- Nonlinear stability mechanism: Discrete maximal ℓp-regularity bounds the nonlinear contribution, while the fractional discrete Grönwall inequality combines it with the linear contribution into a global error estimate.The argument applies discrete maximal regularity with X = L2(Ω) and then invokes discrete Grönwall.
- Scope of the nonlinear assumption: If the nonlinear source is not Lipschitz continuous but the problem has a unique bounded solution, the same error theorems remain valid after proving boundedness of the numerical solutions.The paper assumes Lipschitz continuity for simplicity to avoid these technicalities.
5. Proof of Lemma 4.2 for the L1 scheme.
This section proves the fully discrete L1 error estimate for the linear problem by deriving error representations and resolvent bounds, then treating time-independent and vanishing initial-data cases.
- General-data extension: The L1 scheme is extended from previously analyzed homogeneous problems to the general case considered here.The proof addresses general data rather than only the homogeneous problem.
- Case analysis: The proof separately handles time-independent data and the case v(0) = g(0) = 0 using Laplace-transform error representations and Taylor expansions.The zero-data case uses the identity Phg(t) = 1 ∗ Phg′(t).
- Resolvent comparison: A resolvent estimate bounds the difference between continuous and L1 discrete operator resolvents by c|z|^-ατ in the L2(Ω) operator norm.The estimate is used with Laplace-transform and contour-integral arguments.
- Error representations: The proof represents semidiscrete and fully discrete solutions through convolutions involving the solution operator, projected data, and localized Dirac–Delta forcing.These representations are used to compare continuous and discrete solutions at the time nodes.
- Conclusion of the proof: The resulting estimates complete the proof of Lemma 4.2 for the L1 fully discrete solution.The extension of an intermediate estimate is combined with the preceding representations to obtain the desired bound.
6. Numerical experiments.
Numerical experiments use Galerkin finite elements in space with backward Euler convolution quadrature or L1 time discretization, confirming the predicted spatial and temporal convergence rates in both tested cases.
- 6. Numerical experiments.: The experiments use Galerkin finite elements in space and either backward Euler convolution quadrature or the L1 scheme in time.The domain is the unit square, triangulated with regular right triangles of side length h = 1/M.
- 6. Numerical experiments.: Reference continuous and semidiscrete solutions are computed on finer meshes because the exact solution is unavailable.The continuous reference uses τ = 1/1000 and h = 1/1280; the semidiscrete reference uses h = 1/10 and τ = 1/(64 × 10^4).
- 6. Numerical experiments.: O(h^2) spatial and O(τ^α) temporal error rates are observed for both backward Euler convolution quadrature and L1 in case (a).These observations are reported as fully confirming Theorems 4.3 and 4.4.
- 6. Numerical experiments.: O(h^2) spatial and O(τ^α) temporal convergence rates are also observed in case (b), whose nonlinear source is not Lipschitz continuous.The absolute accuracy of L1 and backward Euler convolution quadrature is comparable in both cases.
- 6. Numerical experiments.: In case (b), spatial error increases slightly with α, whereas temporal error decreases with α.These trends are reported alongside the convergence-rate observations.