PaperlayerView exact revision on arXiv

Source-linked AI summary

Computational Methods in the Fractional Calculus of Variations and Optimal Control

Shakoor Pooseh

arXiv:1312.4064v1math.OCcs.NAmath.CAmath.NA

TL;DR

Fractional variational and optimal control problems use arbitrary-order operators, but exact analytical solutions are generally unavailable. The thesis studies series and generalized finite-difference approximations alongside direct and indirect methods. Its conclusion reports improved approximation results in specified derivative evaluations, while noting difficulties in some applications.

  • Problem

    Fractional variational and optimal control problems involve arbitrary-order operators, while analytical solution methods are unavailable for many such problems.

    Source: Introduction · p. 23, Conclusion and future work · p. 169

  • Method

    The thesis develops series and generalized finite-difference approximations and applies direct and indirect approaches to fractional variational problems.

    Source: Introduction · pp. 24, 26, §7.3 · p. 118

  • Results

    The modified approximation provides better results than (5.13), and the thesis reports better results for approximation (5.4) at least when evaluating fractional derivatives.

    Source: Conclusion and future work · p. 171

  • Takeaways & Limitations

    The thesis demonstrates approximation-based treatment of fractional derivatives, integrals, and differential equations through examples.

    Source: Conclusion and future work · p. 171

    For one first-order ODE, the approximation stays close to the exact solution only briefly before diverging, and the authors cannot increase its order without extra information.

    Source: §6.1.4 · p. 95

Abstract

from arXiv · show

The fractional calculus of variations and fractional optimal control are generalizations of the corresponding classical theories, that allow problem modeling and formulations with arbitrary order derivatives and integrals. Because of the lack of analytic methods to solve such fractional problems, numerical techniques are developed. Here, we mainly investigate the approximation of fractional operators by means of series of integer-order derivatives and generalized finite differences. We give upper bounds for the error of proposed approximations and study their efficiency. Direct and indirect methods in solving fractional variational problems are studied in detail. Furthermore, optimality conditions are discussed for different types of unconstrained and constrained variational problems and for fractional optimal control problems. The introduced numerical methods are employed to solve some illustrative examples.

Introduction

The introduction frames fractional variational and control problems as classical optimization problems extended to arbitrary-order derivatives and integrals, for which numerical methods are needed. It presents operator approximations and direct and indirect solution approaches, then extends the framework to constrained problems and fractional optimal control.

Introduction

  • Direct methods for fractional variational problems have received less attention than indirect methods, motivating the thesis’s Euler-like direct approach.Earlier work includes finite-difference, generalized Leitmann, and Ritz methods.Source: Introduction · p. 23
  • The proposed generalized expansion approximates the left Riemann–Liouville derivative by finite sums of integer-order derivatives.Its approximation order is N, and the thesis compares its advantages and disadvantages with another expansion.Source: Introduction · p. 24
  • The thesis investigates these derivative approximations for closed-form functions, discrete tabular data, and fractional differential equations.It also extends related approximation ideas to fractional integrals, deriving decomposition formulas for left and right integrals of Cn functions.Source: Introduction · pp. 24–25
  • The indirect approach substitutes fractional terms with integer-order approximations, reducing a fractional variational problem to an integer-order constrained problem.The resulting classical problem can then be solved using available methods.Source: Introduction · p. 26
  • For fractional optimal control with free final time, the thesis derives necessary optimality conditions but notes that solving them is impeded by limited techniques for fractional differential equations.It also describes approximation-based conversion to a classical optimal control problem solvable by standard computational techniques.Source: Introduction · p. 27
  • The thesis extends variational problems to Lagrangians involving an antiderivative dependent on the unknown function, a fractional integral, and a Caputo derivative.It also considers transversality conditions, integral and holonomic constraints, higher-order Caputo derivatives, Hamiltonian formalism, sufficient conditions, and numerical examples.Source: Introduction · p. 27

The calculus of variations and optimal

This section reviews the classical calculus of variations and optimal control as background for the thesis’s fractional-problem investigations. It introduces variational formulations, necessary conditions, and direct and indirect numerical approaches.

1.1 Contemporary mathematical formulation

  • The calculus of variations optimizes functionals, and its formulation supplies a framework for problems with boundary conditions, integral constraints, or dynamic constraints.The chapter reviews this classical framework before discussing problems with fractional operators.Source: control · p. 31, §1.1.2 · pp. 33–35

Euler–Lagrange equation

  • For fixed endpoints, an extremizer must satisfy the Euler–Lagrange equation, obtained by requiring the first variation to vanish and applying integration by parts.Free endpoints add transversality conditions; numerical methods are used when the resulting differential equation is difficult to solve in closed form.Source: Euler–Lagrange equation · pp. 36–37

1.1.3 Solution methods

  • Indirect methods derive and solve necessary optimality equations, whereas direct methods optimize a discretized functional or its finite-dimensional approximation.The distinction provides the section’s basic map of numerical solution approaches.Source: Numerical methods · pp. 37–38

Numerical methods

  • Direct methods discretize the problem: Euler’s method uses mesh-point values, while the Ritz method approximates admissible functions with a finite series.Both approaches reduce the original problem to finite-dimensional optimization.Source: Numerical methods · pp. 37–38

1.2.1 Mathematical formulation

  • Optimal control chooses control parameters to optimize a functional subject to system dynamics, which can be incorporated as a constraint in a variational formulation.The section introduces this variational connection as background for later fractional investigations.Source: §1.2 · p. 38, §1.2.1 · pp. 39–40

1.2.3 Pontryagin’s minimum principle

  • With unbounded controls, variational methods yield necessary conditions; bounded controls instead require Pontryagin’s minimum principle, which minimizes the Hamiltonian over admissible controls.The Hamiltonian’s control derivative vanishing is necessary but is not the stated minimum condition itself.Source: §1.2.3 · p. 41

Fractional Calculus

The section introduces fractional calculus and reviews several definitions of fractional derivatives and integrals. It also outlines special functions and relations among operators used in the thesis.

Fractional Calculus

  • Fractional calculus generalizes ordinary integration and differentiation to arbitrary real or complex orders.The section introduces special functions used in defining or manipulating fractional operators and equations.Source: Fractional Calculus · p. 43, §2.1 · p. 43

2.1 Special functions

  • The gamma function extends factorials to real arguments, while the Mittag–Leffler function generalizes the exponential series.These are among the few special functions encountered in the thesis.Source: §2.1 · p. 44

2.2 A historical review

  • Riemann–Liouville, Grünwald–Letnikov, Hadamard, and Caputo operators provide different definitions of fractional derivatives or integrals.Grünwald–Letnikov derivatives generalize finite differences and depend on function values across an unbounded time direction, reflecting their non-locality.Source: §2.2 · pp. 46–48

2.3 The relation between Riemann–Liouville and Caputo derivatives

  • Riemann–Liouville and Caputo derivatives coincide in cases where the relevant initial or terminal derivatives vanish.The section states their relation for order α > 0 and n = [α] + 1.Source: §2.3 · p. 49

2.4 Integration by parts

  • Fractional integration-by-parts formulas are used in proving Euler–Lagrange necessary optimality conditions.The section recalls formulas for Riemann–Liouville and Caputo derivatives.Source: §2.4 · pp. 49–50

Fractional variational problems

Fractional variational and optimal-control problems extend classical optimization to objectives or constraints involving derivatives and integrals of arbitrary orders. The section formulates these problems, gives fractional necessary conditions, and distinguishes direct from indirect solution methods.

Fractional variational problems

3.2 A general formulation

  • The general formulation allows vector-valued functions and fractional derivatives, with boundary conditions optional and constraints incorporable through Lagrange multipliers.Source: §3.2 · pp. 52–53

3.3 Fractional Euler–Lagrange equations

  • For a functional depending on left and right Riemann–Liouville derivatives, an extremum requires the corresponding fractional Euler–Lagrange equation under the stated boundary and regularity conditions.Source: §3.3 · pp. 53–54

3.4 Solution methods

  • Direct methods discretize the problem or approximate functions with basis expansions, while indirect methods apply optimality conditions to obtain differential equations.Source: §3.4 · p. 55
  • The thesis generalizes both direct and indirect solution-method classes to fractional problems.Source: §3.4 · p. 56

A short survey on the numerical methods for solving

The survey reviews numerical approaches for fractional variational and optimal-control problems, grouping them into direct approximations and methods that transform problems for classical tools. It describes basis-function, finite-element, and operator-approximation approaches.

fractional variational problems

  • Other surveyed schemes discretize fractional derivatives with Grünwald–Letnikov approximations or use a formula based on an expansion of the fractional derivative.Source: fractional variational problems · pp. 57, 60, 61
  • Ritz methods approximate unknown functions with shifted Legendre polynomials and reduce the problem to a finite system of equations.The cited approach uses orthogonality and closed-form fractional derivatives to obtain 2m + 2 equations in 2m + 2 unknowns.Source: fractional variational problems · p. 58
  • Multiwavelet approaches approximate the functions and solve the resulting algebraic system at collocation points.The survey also describes variants using different basis functions or problem formulations.Source: fractional variational problems · pp. 58–59
  • A finite-element approach divides the time interval into equally spaced subintervals and approximates the unknown function with shape functions such as splines.Source: fractional variational problems · p. 59
  • Fractional optimal-control problems can be converted to integer-order problems using Oustaloup’s approximation, allowing classical optimal-control tools such as RIOTS-95 to be used.Source: fractional variational problems · pp. 60–61

Approximating fractional derivatives

The section develops finite numerical approximations of fractional derivatives using integer-order derivatives and function moments, and illustrates their use in examples, derivative evaluation, tabular-data settings, and fractional differential equations. It also analyzes truncation errors, giving bounds and guidance for choosing the approximation order.

Approximating fractional derivatives

  • The section introduces infinite-sum expansions of Riemann–Liouville, Caputo, and Hadamard derivatives using integer-order derivatives, then truncates them for numerical use.The expansions can transform fractional problems into classical ones, and the derivation focuses mainly on left derivatives while giving right-derivative expansions when needed.Source: Approximating fractional derivatives · p. 65

5.1.2 Approximation using moments of a function

  • The moment-based expansion requires a corrected formulation because an earlier coefficient series diverges, and its finite approximation must retain B(α, N), which cannot be ignored as α approaches one even for large N.The corrected formulation and proof are given in Theorem 25 and Remark 26; the series defining A(α, i) is convergent for i in {1, …, n−1}.Source: §5.1.2 · pp. 68, 69, 72

5.1.3 Numerical evaluation of fractional derivatives

  • For the tested analytic functions, (5.4) converges faster than (5.10), while including the first derivative improves accuracy and higher-order derivatives can further improve the approximation.For x(t) = t4, the (5.4) approximation with N = 4 is exact because derivatives of order five and above vanish; (5.4) fails for non-analytic functions.Source: §5.1.3 · pp. 74–75, Figure 5.2 · p. 74, Figure 5.3 · p. 75

5.1.4 Fractional derivatives of tabular data

  • For tabular data, (5.10) uses sampled function values, finite differences for the first derivative, and moments computed by integration or an auxiliary system; N can be increased until successive approximations differ by less than a chosen tolerance.The approach is proposed for cases where the function is not available in closed form.Source: §5.1.4 · pp. 76–77

5.1.5 Applications to fractional differential equations

  • The moment-based approximation (5.10) gives a better solution than (5.4) for the illustrated fractional initial-value problem, and retaining B(α, N) improves accuracy over setting it to zero.The (5.4) expansion is limited by the available initial condition, while (5.10) avoids requiring extra initial data for its auxiliary variables.Source: §5.1.5 · pp. 77–79, Figure 5.7 · p. 79

5.2 Hadamard derivatives

  • Hadamard-derivative expansions are obtained in a similar way to the preceding fractional-derivative expansions [97].Source: §5.2 · p. 79

5.2.1 Approximation by a sum of integer-order derivatives

  • Truncating the infinite expansions at order N yields finite-sum approximations of Hadamard fractional integrals and derivatives in terms of integer-order derivatives.The expansions assume that x admits derivatives of any order.Source: §5.2.1 · p. 80

5.2.2 Approximation using moments of a function

  • A technique based on expanding Riemann–Liouville derivatives yields new left and right Hadamard derivative relations, including approximations using function moments.The stated lemma assumes α ∈ (0, 1) and an absolutely continuous function; the subsequent theorem assumes x is Cn+1.Source: §5.2.2 · p. 81

5.2.3 Examples

  • For α = 0.5, the authors apply approximation (5.24) to derivatives of ln(t) and t^4 and present analytic-versus-numerical results in Figure 5.9.The supplied passages identify the comparison but provide no numerical accuracy values.Source: §5.2.3 · p. 82, Figure 5.9 · p. 83
  • For a Hadamard fractional differential equation with solution x(t) = ln t, the authors compare the analytical solution with the numerical result using N = 2.The example replaces the fractional operator using the expansion with n = 1; the comparison appears in Figure 5.10.Source: §5.2.3 · pp. 82–83, Figure 5.10 · p. 84

5.3 Error analysis

  • Truncation errors for the approximations are bounded, and N should increase for rapidly growing test functions, points farther from a, or higher n.The analysis identifies the second term in (5.27) as the truncation error and gives bounds for left Riemann–Liouville and Hadamard derivative approximations.Source: §5.3 · pp. 84, 86

Approximating fractional integrals

This section develops finite-sum approximations of Riemann–Liouville and Hadamard fractional integrals using integer-order derivative expansions, with error estimates and bounds. It evaluates approximation accuracy, including error behavior as the truncation parameter increases, and illustrates the methods on a fractional integral equation.

6.1.1 Approximation by a sum of integer-order derivatives

  • The new Riemann–Liouville decomposition avoids the large n needed by the earlier derivative series and can achieve appropriate accuracy even for n = 1.The formulas apply to functions of class Cn and support numerical approximations with error estimates.Source: §6.1.1 · p. 87, Approximating fractional integrals · p. 87

6.1.2 Approximation using moments of a function

  • The moments-based Riemann–Liouville decomposition has a general form determined by the test function’s smoothness, and finite approximations depend on derivative order n and sum length N.Although the first n coefficients vanish, retaining them in the truncated sum improves accuracy.Source: §6.1.2 · pp. 89–90

6.1.3 Numerical evaluation of fractional integrals

  • Increasing N improves the Riemann–Liouville approximation, while increasing n reduces error; at matched second-derivative expansions, the new method is more accurate than the prior formula even for small N.The comparisons use x1(t) = t^3 and x2(t) = t^10 on [0, 1].Source: §6.1.3 · pp. 92–93, Figure 6.1 · p. 92, Figure 6.4 · p. 94

6.1.4 Applications to fractional integral equations

  • For a fractional integral equation, the first-order approximation stays close to the exact solution only briefly before diverging, whereas the moments-based expansion is used with increasing N to seek more accurate solutions.The first-order approximation cannot be raised because the problem provides no extra initial or boundary information.Source: §6.1.4 · pp. 95–96, Figure 6.6 · p. 96

6.2.1 Approximation by a sum of integer-order derivatives

  • For Hadamard fractional integrals, prior results provide expansions in integer-order derivatives for functions with derivatives of any order, including a reference for arbitrary α > 0.The passage points to [65, Theorem 3.2] and [35, Theorem 17].Source: §6.2.1 · p. 97

6.2.2 Approximation using moments of a function

  • The moments-based Hadamard expansions cover left and right integrals for x in C^(n+1), and finite sums yield approximations under the stated convergence assumption.The expansions are given for 0 < a < b.Source: §6.2.2 · pp. 99–101

6.2.3 Examples

  • For Hadamard examples, N = 3 suffices for good accuracy in one test, while increasing N reduces error; increasing n can raise error for fast-growing functions, so N should also increase.The tests include x4(t) = t^9, for which the sequence x(k,0)(t) = 9^k t^9 grows with k.Source: §6.2.3 · pp. 102–104, Figure 6.7 · p. 103, Figure 6.9 · p. 104

6.3 Error analysis

  • The section estimates truncation errors for finite-sum approximations of both Riemann–Liouville and Hadamard fractional integrals.The estimates address the neglected terms when sums are truncated at order N.Source: §6.3 · p. 105
  • The error formula tends to zero as N increases.Source: §6.3 · p. 106
  • The error for the considered approximation is bounded by an expression.Source: §6.3 · p. 106
  • The error for the right Hadamard integral is bounded.Source: §6.3 · p. 106

Direct methods

This section develops direct discretization methods for fractional variational problems, including Euler-like approaches and a discrete first-variation method using splines. It applies these methods to basic and integral-constrained problems, with the discrete solutions approximating the unknown function at mesh points.

7.1 Finite differences for fractional derivatives

  • Fractional finite-difference approaches include Diethelm’s backward Caputo formula, with order O(h^(2−α)) for 0 < α < 2 and α ≠ 1.Source: §7.1 · p. 108

7.2.1 Euler’s classic direct method

  • The classical Euler direct method replaces admissible curves with grid-point values, finds extrema by solving algebraic equations, and approaches the Euler–Lagrange equation as h tends to zero.Source: §7.2.1 · p. 109

7.2.2 Euler-like direct method

  • The Euler-like method discretizes the fractional derivative with Grünwald–Letnikov approximation and the functional with quadrature, then solves for the interior mesh values; it converges to the fractional Euler–Lagrange equation as h tends to zero.Source: §7.2.2 · pp. 109–110

7.2.3 Examples

  • Linear Euler-like systems yield better solutions as mesh points increase and run quickly when the coefficient matrix is invertible; nonlinear cases differ in runtime and error.Table 7.1 summarizes runtime and error, but the supplied passage does not give its values.Source: §7.2.3 · p. 116
  • Direct-method examples reduce variational problems to algebraic systems: linear systems are fast when their coefficient matrices are invertible, while nonlinear systems have different runtime and error behavior.For the linear examples, increasing the number of mesh points improves the approximations.Source: §7.2.3 · p. 116

7.3 A discrete time method on the first variation

  • The first-variation method uses first-order splines as variations and Grünwald–Letnikov’s discrete derivative to form a system of algebraic equations for the mesh-point values [103].Source: §7.3 · pp. 118–119
  • Solving the discretized first-variation system provides an approximation to the unknown function at the mesh points.The method is also introduced for isoperimetric problems with an integral constraint involving a fractional derivative.Source: §7.3 · pp. 119–120

7.3.2 An isoperimetric fractional variational problem

  • For an integral-constrained problem, the discrete first-variation system is extended with an auxiliary function and a multiplier determined using a discretized constraint.Example 55's resulting algebraic-system solutions for different step sizes are presented.Source: §7.3 · p. 120, §7.3.2 · p. 123

7.3.1 Basic fractional variational problems

Fractional optimal control with free

This section develops and applies fractional optimality conditions and integer-order approximations to fractional optimal control problems, including fixed- and free-final-time cases, endpoint restrictions, a generalized problem with fractional memory, and numerical examples.

8.1 Necessary optimality conditions

  • The problem treats a one-dimensional state and control system with integer- and fractional-order derivatives, optimizing both the trajectory and the free terminal time T.The setup assumes differentiable state and piecewise-continuous control; the results are described as extendable to multiple states and controls, and T is initially unconstrained.Source: §8.1 · p. 126

8.1.1 Fractional necessary conditions

  • For mixed integer- and fractional-order control systems with free terminal time, Theorem 57 gives necessary conditions and transversality relations for an optimal state-control-time triplet.The chapter also considers constraints at the end time and states that the conditions become sufficient under appropriate assumptions.Source: end-points · p. 125, §8.1.1 · p. 128
  • Transversality conditions vary with whether T and x(T) are fixed or free, and fractional chain rules prevent the standard conversion from free to fixed terminal time.The corollary also treats terminal points constrained to a curve or by inequality bounds.Source: §8.1.1 · pp. 129–130

8.1.2 Approximated integer-order necessary optimality conditions

  • Using a fractional-operator approximation transforms the problem into an integer-order problem whose necessary conditions form a two-point boundary value problem.The derivation uses approximation (5.10) and the relation between Caputo and Riemann–Liouville derivatives up to order K.Source: §8.1.2 · pp. 131–132

8.2 A generalization

  • For a generalized problem with cost integral starting at A>a, necessary conditions account for fractional-derivative memory extending before the interval of interest [A, T].The initial condition is placed at A; the text notes that no general uniqueness theory is known for this setup, though uniqueness can occur.Source: §8.2 · pp. 132–134

8.3 Sufficient optimality conditions

  • Under convexity of L and f in x and u and of φ in x, the conditions are sufficient when λ(t) ≥ 0 or f is linear; strict convexity yields a unique minimizer.The convexity assumptions and multiplier or linearity condition are stated in Theorem 65; strict convexity is noted as sufficient for uniqueness.Source: Theorem 65. Let ( · p. 135, Then ( · pp. 135, 137

8.4 Numerical treatment and examples

  • Numerical methods solve problems using fractional necessary conditions or an integer-order approximation; a fixed-final-time example with known exact solution is used for comparison.For the example, approximations transform the fractional problem into a classical integer-order boundary value problem.Source: §8.4 · p. 137, §8.4.1 · pp. 137–138

8.4.1 Fixed final time

  • For the fixed-final-time example, MATLAB bvp4c solves the two-point boundary-value problem for N = 2 and N = 3, with results shown against the exact solution.The numerical solutions use fractional approximations (5.10) and (5.14).Source: §8.4.1 · pp. 140–141, Figure 8.2 · p. 142

8.4.2 Free final time

  • For the free-final-time example, the exact solution is unknown, and the authors apply both fractional necessary optimality conditions and a direct integer-order transformation.The fractional-conditions approach treats T as a parameter after mapping the problem to [0, 1], with λ(T) = 0; the direct approach uses H|t=T = 0.Source: §8.4.2 · pp. 143–145
  • For α = 0.5 and N = 2, the parametric boundary-value problem is solved with MATLAB bvp4c, and Figure 8.3 shows both procedures’ trajectories.The fractional necessary optimality conditions are plotted with dashed lines and the integer-order problem with dash-dotted lines.Source: §8.4.2 · pp. 144–145, Figure 8.3 · p. 144

Fractional variational problems

This section develops optimality conditions for fractional variational problems and fractional optimal control, including problems with Caputo derivatives, constraints, and higher-order derivatives. It also examines fractional-operator expansion methods, their performance and limitations, and applications including a fractional dengue model.

depending on indefinite integrals

  • The chapter extends results in to Lagrangians with an antiderivative depending on the unknown function, a fractional integral, and a Caputo derivative, and also treats related constrained and higher-order problems.It also gives sufficient optimization conditions under suitable convexity assumptions on the Lagrangian [12].Source: depending on indefinite integrals · p. 147

9.1 The fundamental problem

  • The fundamental problem’s necessary fractional Euler–Lagrange equation involves fractional integrals, fractional derivatives, and indefinite integrals; minimizers need not be differentiable.The examples show a minimizer that is not differentiable at 0 for 0 < α < 1.Source: §9.1 · pp. 149–150

9.2 Natural boundary conditions

  • When the terminal time T is variable, minimizers (x, T) satisfy a fractional equation and transversality conditions.The problem optimizes over both the trajectory and terminal time.Source: §9.2 · p. 152

9.3 Fractional isoperimetric problems

  • For fractional isoperimetric problems, minimizers satisfy a multiplier-based fractional equation, with constants λ0 and λ not both zero covering the general case.The stated sufficient condition for the one-multiplier result is that the minimizer is not an extremal for the integral constraint.Source: §9.3 · pp. 155–156

9.4 Holonomic constraints

  • For holonomic constraints, a minimizing pair satisfies fractional Euler–Lagrange equations using a continuous multiplier function λ(t).The theorem is stated for two functions and is noted to generalize to n variables.Source: §9.4 · pp. 157–158

9.5 Higher order Caputo derivatives

  • With fixed endpoint derivatives, higher-order Caputo minimizers satisfy a fractional necessary optimality equation; without prescribed boundary conditions, they also satisfy natural boundary conditions.The results concern non-integer derivative orders.Source: §9.5 · pp. 159, 161

9.6 Fractional optimal control problems

  • For unconstrained fractional Lagrange problems, an optimal state-control pair admits an adjoint function satisfying a Hamiltonian system and stationary condition.The stated conditions apply to a cost with an indefinite integral, fractional dynamics, and boundary conditions.Source: §9.6 · p. 162

9.7 Sufficient conditions of optimality

  • For convex Lagrangians, solutions of the fractional Euler–Lagrange equation meeting either stated condition are global minimizers, with a corresponding result for natural boundary conditions.The passage states sufficient conditions for both prescribed and natural boundary settings, but does not include the conditions’ details.Source: §9.7 · pp. 163–164

9.8 Examples

The classical theory [94] tells us to solve the system

Conclusion and future work

  • The thesis generalizes direct and indirect variational methods and their connections to optimal control to fractional calculus.Source: Conclusion and future work · p. 169
  • The Euler-like direct method is fast for linear systems but can converge slowly because its Grünwald–Letnikov approximation is first order; nonlinear cases may require many mesh points and long runtimes.For Example 51, grids of 5 and 30 points have errors of the same order, 10^-3; Example 52 has error E = 1.4787 at n = 5, and increasing mesh size greatly increases runtime.Source: Conclusion and future work · p. 170
  • Indirect methods enable classical solvers on approximated fractional problems, while the operator singularity complicates solution; approximation (5.4) performs better for derivative evaluation despite higher-derivative difficulties.The approximation studies also cover fractional integrals, Hadamard operators, and Caputo derivatives.Source: Conclusion and future work · pp. 170–171
  • Increasing N improves approximation (5.10) for fractional differential equations, and the authors report better results than their modification of (5.13).Approximation (5.10) can also be generalized to include higher-order derivatives and used to compute fractional derivatives from tabular data.Source: Conclusion and future work · p. 171
  • The expansions support two solution routes: directly approximate the fractional problem and solve it classically, or apply the fractional Euler–Lagrange equation before approximation.The first route is associated with Chapter ?? and the second with Section 8.4.1.Source: Conclusion and future work · p. 171
  • Direct methods could improve through higher-precision derivative approximations, better quadrature, and more sophisticated algebraic solvers; further work should cover more problem types.Diethelm’s backward finite differences are given as an example of a higher-precision approximation.Source: Conclusion and future work · p. 172
  • Transforming fractional problems into classical ones may extend available methods, but singularities in approximations (5.10) and (5.4) need attention.The authors also propose software tools for classes of fractional variational problems, potentially including fractional differential-equation solvers.Source: Conclusion and future work · p. 172
  • A simple fractional dengue model produced surprisingly good results [104], using non-local derivatives to model memory and a tunable order α that is highly result-sensitive.The classical model is recovered as α approaches one; the passage contrasts the simple classical model’s inadequacy with the fractional approach.Source: Conclusion and future work · pp. 172–173
  • Future extensions include fractionalizing more complex models, assigning different derivative orders to state variables, and applying fractional optimal control to epidemic models.Source: Conclusion and future work · p. 173
Loading 1312.4064v1…