Source-linked AI summary
Generalized Jacobi Functions and Their Applications to Fractional Differential Equations
Sheng Chen, Jie Shen, Li-Lian Wang
TL;DR
FDE computation is complicated by nonlocal fractional derivatives, singular kernels and boundary behavior, and mismatched solution–data regularity. The paper introduces GJFs and analyzes GJF-Petrov-Galerkin methods with fractional-derivative approximation theory. For the considered model FDEs, the methods yield sparse systems and data-smoothness-controlled convergence, including exponential convergence for analytic data despite singular solutions.
Problem
FDEs involve nonlocal operators, singular kernels, boundary-singular solutions, and solution–data regularity mismatches that limit standard numerical approaches.
Method
The paper develops two-parameter GJFs linked to fractional calculus and uses them in Petrov-Galerkin spectral methods for prototypical FDEs.
Results
The methods produce sparse matrices and error estimates whose convergence rates depend only on data smoothness; analytic data yield exponential convergence despite singular solutions.
Takeaways & Limitations
For the simple FDEs considered, spectral methods can achieve computational complexity and accuracy comparable to methods for usual PDEs.
Takeaways & Limitations
The paper does not rigorously justify improved L2 error estimates in the fractional case because usual Sobolev regularity is insufficient for the duality argument.
Abstract
from arXiv · showhide
In this paper, we consider spectral approximation of fractional differential equations (FDEs). A main ingredient of our approach is to define a new class of generalized Jacobi functions (GJFs), which is intrinsically related to fractional calculus, and can serve as natural basis functions for properly designed spectral methods for FDEs. We establish spectral approximation results for these GJFs in weighted Sobolev spaces involving fractional derivatives. We construct efficient GJF-Petrov-Galerkin methods for a class of prototypical fractional initial value problems (FIVPs) and fractional boundary value problems (FBVPs) of general order, and show that with an appropriate choice of the parameters in GJFs, the resulted linear systems can be sparse and well-conditioned. Moreover, we derive error estimates with convergence rate only depending on the smoothness of data, so truly spectral accuracy can be attained if the data are smooth enough. The idea and results presented in this paper will be useful to deal with more general FDEs associated with Riemann-Liouville or Caputo fractional derivatives.
1. Introduction
FDEs model anomalous and complex-system dynamics, but their nonlocal operators, singular kernels, boundary singularities, and solution–data regularity mismatch challenge standard numerical methods. The paper develops GJF-based spectral methods to address these difficulties.
- FDEs describe anomalous diffusion, non-exponential relaxation, and complex phenomena across science, engineering, bioengineering, and economics.
- Fractional derivatives are non-local and involve singular kernels or weights, while FDE solutions are usually singular near boundaries.
- Existing finite difference and finite element methods rely on local operations and are poorly suited to singular kernels and weights.
- Nonlocal fractional derivatives typically produce full dense matrices, making coefficient calculation and inversion expensive.
- FDE solutions and data may belong to different Sobolev spaces, so smooth data can still yield limited solution regularity in usual Sobolev spaces.
- The paper introduces GJFs and develops efficient spectral methods intended to address these difficulties for prototypical FDEs.
2. Preliminaries
The preliminaries define fractional operators and review Jacobi polynomials with real parameters, together with hypergeometric and Bateman formulas used in the analysis. They also establish fractional-derivative identities connecting GJFs to polynomial structures.
- Bateman fractional integral formula: The Bateman fractional integral formula is introduced as an important tool for subsequent algorithm development and analysis.
- Fractional integrals and derivatives: The paper reviews Riemann–Liouville and Caputo fractional integrals and derivatives on (−1,1), noting that the formulas extend to general intervals.
- Jacobi polynomials with real parameters: Jacobi polynomials extend beyond classical parameters α, β > −1 to real parameters, remaining polynomials in x, although general-case orthogonality does not persist.
- GJF properties: The generalized basis functions are related to generalized Jacobi functions and to Jacobi poly-fractonomials for s ∈ (0,1).
3. Generalized Jacobi functions
The paper defines generalized Jacobi functions with parameter ranges and fractional-calculus properties tailored to fractional differential equations. These functions support orthogonality, fractional Sturm–Liouville formulations, and connections to Jacobi poly-fractonomials.
- Definition and novelty: GJFs are defined through parameterized Jacobi-polynomial forms across four regions determined by whether α and β are above or below −1.The four regions are ℵ1 through ℵ4, covering the combinations of α, β ≤ −1 and α, β > −1.
- Definition and novelty: The modified GJFs extend earlier definitions to additional parameter ranges, opening new applicability for fractional differential equations.The paper specifically modifies the classical Jacobi-polynomial construction for −1 < α, β < 1 and other ranges.
- Fractional-calculus properties: Suitable fractional derivatives of GJFs produce polynomials, enabling orthogonality-based approximation analysis in fractional Sobolev spaces.The derivative formulas and resulting orthogonality relations are central to studying GJF approximability.
- Fractional-calculus properties: GJFs are eigenfunctions of fractional Sturm–Liouville-type equations, extending the standard Sturm–Liouville framework to fractional derivatives.For fixed s, the associated eigenvalues grow as O(n2s), recovering O(n2) as s → 1.
- Fractional-calculus properties: The fractional Sturm–Liouville operators are not self-adjoint in general, although the singular operators are self-adjoint when s ∈ (0, 1).This qualifies the spectral structure available for the operators used with GJFs.
- Relation to Jacobi poly-fractonomials: GJFs are related to Jacobi poly-fractonomials, but their different constructions yield different admissible parameter ranges.The paper notes that the GJF range relaxes the condition on α relative to the cited JPF construction.
4. Approximation by GJFs
GJF series achieve spectral convergence in weighted Sobolev spaces involving fractional derivatives. Their parameters can encode boundary singularities and conditions while supporting efficient approximation.
- GJFs are analyzed in weighted spaces involving fractional derivatives, yielding optimal-order projection estimates.The approximation results apply when the function has the corresponding fractional-derivative regularity.
- Choosing GJF parameters can match one-sided or two-sided boundary singularities and impose the associated homogeneous boundary conditions.For example, β can be selected to match singularities in fractional initial value problems, while β = −[α] supports two-sided conditions.
- For functions with non-integer boundary singularities, selecting the matching fractional parameter β yields exponential convergence O(e^−cN).The comparison considers functions whose singular factor is represented by the GJF basis.
- For smooth functions, choosing a non-integer β can instead produce only limited convergence rates.The stated limitation arises despite smoothness, because the selected parameter does not match the function’s regularity structure.
5. Applications to fractional differential equations
The paper applies GJFs in Petrov-Galerkin spectral methods for prototypical fractional differential equations. The methods achieve convergence governed by data regularity and can produce sparse or diagonal linear systems.
- The proposed Petrov-Galerkin methods target prototypical fractional differential equations, including problems whose solutions are singular despite regular data.The applications include fractional initial value and boundary value problems of general order.
- The convergence rate depends only on the regularity of the data, regardless of singular behavior in the underlying solution.Thus, smooth input data can yield truly spectral accuracy.
- Suitable GJF parameters make the resulting linear systems usually sparse and sometimes diagonal.The paper reports numerical results intended to validate this structural and approximation analysis.
- The prototypical problems are presented as a basis for studying more complicated fractional differential equations.
5.1. Fractional initial value problems (FIVPs).
For fractional initial value problems, the paper constructs a GJF-based Petrov-Galerkin scheme whose error estimates support spectral accuracy when the source term is sufficiently regular. Extensions cover more general variable-coefficient forms.
- Fractional initial value problems: The FIVP scheme uses GJFs in a Petrov-Galerkin formulation with polynomial test functions and an orthogonal projection of the source term.The numerical solution is obtained by inserting the GJF expansion into the discrete formulation.
- Fractional initial value problems: Theorem 5.1 provides an error estimate for the discrete FIVP solution under weighted Sobolev regularity of the source term.The constant is independent of u, N, and m.
- Generalized FIVPs: The approach extends to more general FIVPs with continuous coefficient functions, while constant coefficients retain sparse linear systems.For general coefficients, a preconditioned iterative algorithm is suggested.
5.2. Fractional boundary value problems (FBVPs).
The paper develops GJF-Petrov-Galerkin schemes for fractional boundary value problems, using fractional-polynomial spaces that encode boundary conditions and solution singularities. Appropriate GJF parameters yield sparse systems, while weighted fractional-space analysis establishes well-posedness and error estimates.
- Method: The GJF-Petrov-Galerkin method is formulated through fractional integration and finite-dimensional trial and test spaces tailored to the problem order.The construction includes equivalent Galerkin formulations used for error analysis.
- Implementation: Using GJFs as basis functions makes the linear system matrix diagonal for the considered FBVP scheme.The diagonal structure follows from the fractional derivative relations and Legendre-polynomial orthogonality.
- Higher-order FBVPs: The higher-order odd-order FBVP construction likewise produces a diagonal system through fractional derivative formulas and Jacobi-polynomial orthogonality.The approach extends beyond the order-ν problem with 1 < ν < 2.
- Well-posedness and error analysis: For even-order FBVPs, norm equivalence and the Babuška-Brezzi inf-sup condition establish unique solvability of both continuous and discrete problems.The paper then derives spectral error estimates under weighted regularity assumptions.
5.3. Numerical results.
Numerical experiments test FIVP and FBVP cases with either smooth data or smooth solutions. Observed exponential or algebraic convergence agrees with the theoretical estimates, while the fractional-norm error analysis does not rigorously explain the smaller L2 errors.
- FIVPs: Smooth FIVP source data produce exponentially decaying errors even when the unknown solution is singular at x = 1.The convergence rate depends only on the smoothness of f, including in the fractional norms.
- FIVPs: For a smooth FIVP solution and singular source term, the observed convergence slopes agree with the predicted algebraic rates.The experiment uses u(x) = (1 − x^3)(1 − e^(1−x)) and derives f from the equation.
- FBVPs with integral boundary conditions: Smooth source data produce true spectral convergence for the FBVP with integral boundary conditions, with rates determined by source-term smoothness.The result agrees with the error estimate (5.20).
- FBVPs with integral boundary conditions: For the same FBVP with a smooth exact solution and singular source, the observed rates match the predicted order m − 1 for ν = 1.3 and ν = 1.7.The predicted regularity bounds are m < 5.4 and m < 4.6, respectively.
- FBVPs with homogeneous boundary conditions: The homogeneous-boundary-condition FBVP shows exponential convergence with smooth source data and agreement with the expected algebraic rate for the smooth-solution case.The latter experiment uses ν = 1.4, s = 0.4 and ν = 1.9, s = 0.9.
- Error measurement: L2 errors are smaller than fractional-norm errors in all examples, but the paper cannot rigorously justify this through a fractional duality argument.The authors attribute the difficulty largely to limited regularity in the usual Sobolev norm.
6. Extensions, discussions and concluding remarks
The paper extends key GJF derivative formulas and approximation results from Riemann-Liouville to Caputo derivatives. It concludes that the framework supports efficient spectral methods for the studied prototypical FDEs while remaining an initial step toward more general problems.
- Extension to Caputo derivatives: Caputo derivative analogues of important Riemann-Liouville formulas are established, including formulas for fractional derivatives of GJFs.The corresponding parameter ranges can differ slightly from the Riemann-Liouville case.
- Extension to Caputo derivatives: For suitable GJF parameters, the Caputo formulas recover counterparts of the key Riemann-Liouville derivative identities.This applies when α > 0 or β > 0 under the stated conditions.
- Extension to Caputo derivatives: The Caputo derivative formulas support GJF approximation results in weighted Sobolev spaces and corresponding spectral methods, although details are omitted.The paper states that GJFs retain similar approximability for Caputo FDEs.
- Contributions: The paper’s two main contributions are the generalized GJF framework and Petrov-Galerkin methods for arbitrarily high-order FIVPs and FBVPs.The methods yield sparse matrices and error estimates whose convergence rates depend only on data smoothness.
- Conclusion: For the simple FDEs considered, the results indicate spectral methods can achieve computational complexity and accuracy comparable in kind to methods for usual PDEs.This conclusion is explicitly restricted to the simple model equations studied.
- Scope: The study covers only a class of very simple prototypical FDEs, so its principles are presented as an initial step toward more general equations.The paper identifies broader FDEs as a direction opened by the developed approximation results.