Skip to Content
You are currently on the new version of our website. Access the old version .
  • Article
  • Open Access

27 January 2026

14 Pages

An Expanded Mixed Finite Element Method for Fractional Dispersion Equations with Variable Coefficient

,
and
1
School of Statistics and Mathematics, Shandong University of Finance and Economics, Jinan 250014, China
2
School of Mathematics and Statistics, Shandong Normal University, Jinan 250014, China
3
College of Mathematics and Systems Science, Shandong University of Science and Technology, Qingdao 266590, China
*
Author to whom correspondence should be addressed.

Abstract

In this article, we propose an expanded mixed finite element method for variable-coefficient fractional dispersion equations (FDEs). By introducing two intermediate variables, p = D u and σ = − I θ β p , the FDEs are reformulated into a mixed system involving only lower-order derivatives. Based on this, we construct an expanded mixed variational framework and prove the weak coercivity in the sense of the LBB condition over appropriately chosen Sobolev spaces, thereby ensuring the well-posedness of the formulation. Then, we develop an expanded mixed finite element scheme and prove that the unique expanded finite element solution possesses optimal approximation accuracy to the fractional flux σ , the gradient p and the unknown u. Finally, numerical experiments are conducted to verify the efficiency and accuracy of the proposed method.

1. Introduction

Consider fractional diffusion equations of order 2 − β with variable coefficients
− D k ( x ) I θ β D u + d ( x ) u = g ( x ) , x ∈ Ω : = ( 0 , 1 ) , u ( 0 ) = u ( 1 ) = 0 ,
which have been widely used to model those anomalous diffusive processes in applied sciences. Here, k ( x ) ∈ C 1 ( Ω ) is the diffusive coefficient depending on x with positive upper and lower bounds, i.e., there exist two positive constants satisfying k min ≤ k ( x )   ≤ k max , x ∈ Ω ; the nonnegative function d ( x ) ∈ L ∞ ( Ω ) refers to the chemical reaction term and g ( x ) ∈ L 2 ( Ω ) is the source or sink term; D u : = d u d x stands for the spatial derivative; I θ β : = θ I x β 0 + ( 1 − θ ) I 1 β x defines the fractional integral operator of order β for β ∈ ( 0 , 1 ) and some θ ∈ [ 0 , 1 ] , where the left-sided fractional integral operator I x β 0 and the right-sided counterpart I 1 β x are given by
I x β 0 u = 1 Γ ( β ) ∫ 0 x ( x − t ) β − 1 u ( t ) d t ,       I 1 β x u = 1 Γ ( β ) ∫ x 1 ( t − x ) β − 1 u ( t ) d t ,
respectively [1].
As noted in [2], the presence of variable coefficients may lead to a loss of coercivity in the Galerkin formulations of fractional differential equations, thereby causing difficulties in establishing the solvability of the associated Galerkin equations. Consequently, standard Galerkin-type numerical methods, such as finite element, mixed finite element, and finite volume element, which have been successfully applied to their constant-coefficient counterparts [1,3,4,5,6], cannot be directly extended to the variable-coefficient scenario. To address the loss of coercivity, two main strategies have been proposed. The first and more elementary approach is to impose additional relations on the coefficient k ( x ) and its derivative k ′ ( x ) to ensure coercivity, and following this, Liu [7], Hao [8] and Zheng et al. [8,9,10] proved coercivity using the Gårding’s inequality; the second strategy is to relax the requirement of coercivity to a weaker form over the test and trial function spaces, in accordance with the Ladyzhenskaya–Babu s ˘ ka–Brezzi (LBB) condition [11]. Within this framework, Wang and Yang [2,12] developed a discontinuous Petrov–Galerkin method with rigorously proven weak coercivity, while Chen et al. proposed the H 1 -Galerkin mixed element method [13] and least-squares mixed finite methods [14,15], all of which achieve weak coercivity through a space-decomposition technique.
From an engineering application perspective, the information of the fractional flux − I θ β D u and the gradient D u is of the same importance as the unknown u . Therefore, it is necessary to design a numerical method to approximate the flux, gradient and the unknown simultaneously. The main purpose of this article is to develop an expanded mixed finite element method so that the three variables of (1) are accurately approximated at the same time. This is realized in two systematic stages: (1) Develop an expanded mixed formulation for steady-state FDEs (1) by introducing σ : = − I θ β D u and the gradient p : = D u as intermediate variables, then establish the weak coercivity of this formulation, verified via the inf-sup (LBB) condition, over appropriately constructed Sobolev spaces, thus ensuring its well-posedness. Building on this theoretical foundation, we propose the expanded mixed finite element method and prove that the unique expanded finite element solution possesses an optimal approximation accuracy for fractional flux σ , gradient p, and unknown u. (2) Conduct several numerical experiments to validate the efficiency of the proposed expanded mixed finite element method.
The remainder of this paper is organized as follows: In Section 2, we construct an expanded mixed formulation for steady-state FDEs. Under suitably chosen Sobolev spaces, we establish both the weak coercivity of the bilinear form A ( · , · ) and the inf-sup condition of the bilinear form B ( · , · ) , thus guaranteeing the well-posedness of the mixed formulation. In Section 3, we develop the corresponding expanded mixed finite method and perform a rigorous convergence analysis. In Section 4, we present several numerical experiments to verify the theoretically proven convergence results.

2. Expanded Mixed Variational Formulation for FDEs

In this section, we derive an expanded mixed variational formulation for steady-state FDEs (Equation (2)). By establishing the weak coercivity of the bilinear form A ( · , · ) and the inf-sup condition of the bilinear form B ( · , · ) over appropriately chosen Sobolev spaces, we prove the well-posedness of the formulation. Furthermore, we rigorously establish the equivalence between this expanded variational formulation and the original steady-state FDE.

2.1. The Mathematical Model

Let
b ( x ) = − d k ( x ) d x 1 k ( x ) .
Then, the FDEs (Equation (1)) can be rewritten as
− D I θ β D u + b ( x ) I θ β D u + c ( x ) u = f ( x ) , x ∈ Ω ,
u ( 0 ) = u ( 1 ) = 0 ,
where c ( x ) = d ( x ) k − 1 ( x ) ,   f ( x ) = g ( x ) k − 1 ( x ) .
Remark 1. 
If k ( x ) is a constant, then b ( x ) = 0 , and the Equation (2) (or Equation (1)) reduces to a standard constant coefficient fractional diffusion equation. In addition, if β = 0 , the Equation (2) simplifies to the classical second-order advection-diffusion equation.
Noting that Equation (2a) is a combination of a constant coefficient FDE and a fractional low-order term, this motivates us to study steady-state FDEs (Equation (2)) rather than Equation (1).
For μ > 0 , let H μ ( Ω ) be the fractional-order Sobolev space with semi-norms and norms denoted by | · | μ and ∥ · ∥ μ , respectively. Let H − μ ( Ω ) be the negative Sobolev space, which is the dual space of H 0 μ ( Ω ) , and its norm is denoted by ∥ · ∥ − μ . Let J L − μ ( Ω ) and J R − μ ( Ω ) denote the closures of C 0 ∞ ( Ω ) with respect to the norms ∥ v ∥ J L − μ ( Ω ) : = ∥ 0 I x μ v ∥ and ∥ v ∥ J R − μ ( Ω ) : = ∥ x I 1 μ v ∥ , respectively.
The equivalence theories for fractional spaces are presented below [16].
Lemma 1 
([16]). Let μ > 0 . Then the fractional order Sobolev spaces J L − μ ( Ω ) , J R − μ ( Ω ) , and H − μ ( Ω ) are equal with equivalent norms.
Lemma 2 
([16], Lemma 2.6). For μ > 0 ,
( I x μ 0 v , I 1 μ x v ) = cos ( π μ ) ∥ I x μ − ∞ v ˜ ∥ 2 = cos ( π μ ) ∥ I + ∞ μ x v ˜ ∥ 2 ,
where v ˜ is the zero extension of v to R .

2.2. Mixed Variational Formulation

Introduce variable p = D u and fractional-order flux σ = − I θ β p . Then Equation (2) can be rewritten as mixed formulas involving lower order derivatives
D u = p ,
− I θ β p = σ ,
D σ + b I θ β p + c u = f .
The weak formulation corresponding to Equation (4) is: find ( u , p , σ ) ∈ W × H × V satisfying
( p , ρ ) + ( u , D ρ ) = 0 , ρ ∈ V ,
( σ , q ) + I ( p , q ) = 0 , q ∈ H ,
( D σ , v ) + I ( p , b v ) + ( c u , v ) = ( f , v ) , v ∈ W ,
with I ( p , q ) = ( I θ β p , q ) = θ ( I x β 2 0 p , I 1 β 2 x q ) + ( 1 − θ ) ( I 1 β 2 x p , I x β 2 0 q ) and V = H 1 ( Ω ) , H = H − β 2 ( Ω ) ,   W = L 2 ( Ω ) .
Let U = W × H be the usual product norm ∥ χ ∥ U 2 = ∥ q ∥ H 2 + ∥ v ∥ W 2 for χ = ( v , q ) ∈ U , and define the following bilinear forms A ( · , · ) : U × U → R and B ( · , · ) : U × V → R :
A ( λ , χ ) = I ( p , q ) + I ( p , b v ) + ( c u , v ) , λ = ( u , p ) , χ = ( v , q ) ∈ U ,
B ( χ , ρ ) = ( q , ρ ) + ( v , D ρ ) , χ = ( v , q ) ∈ U , ρ ∈ V .
Then (5) can be rewritten as: find ( λ , σ ) ∈ U × V with λ = ( u , p ) such that
A ( λ , χ ) + B ( χ , σ ) = F ( χ ) , χ = ( v , q ) ∈ U ,
B ( λ , ρ ) = 0 , ρ ∈ V ,
where the form F ( χ ) is defined
F ( χ ) = ( f , v ) , χ = ( v , q ) ∈ U .
Finally, we define the notation
U 0 = { χ ∈ U : B ( χ , ρ ) = 0 , ρ ∈ V } .
Lemma 3 
([17]). Let χ = ( v , q ) . Then χ ∈ U 0 if and only if q = D v with v ∈ H 0 1 − β 2 ( Ω ) .
Lemma 4 
([3]). The mapping χ = ( v , q ) ∈ U 0 ↦ ∥ q ∥ H defines a norm in U 0 equivalent to the norm ∥ χ ∥ U , i.e., there exist positive constants C 1 , C 2 satisfying
C 1 ∥ q ∥ H ≤ ∥ χ ∥ U ≤ C 2 ∥ q ∥ H .
Assumption 1. 
Assume that b ( x ) ,   c ( x ) ∈ L ∞ ( Ω ) , c ( x ) ≥ 0 and
∥ b ∥ L ∞ ( Ω ) 2 ≤ 4 ( 1 − γ ) c m i n cos ( β 2 π ) ( Γ ( 1 + β 2 ) ) 2 ,
for some 0 < γ < 1 , where c m i n = e s s inf { c ( x ) : x ∈ Ω } .
Remark 2. 
If β = 0 , Equation (8) reduces to
∥ b ∥ L ∞ ( Ω ) 2 ≤ 4 ( 1 − γ ) c m i n ,
which is consistent with the classical stability condition for the second-order advection-diffusion Equation [18].
Lemma 5. 
The bilinear forms A ( · , · ) and B ( · , · ) are continuous on U × U and U × V , respectively. Under condition (8), the bilinear form A ( · , · ) is U 0 -elliptic, i.e., there exists a constant α 0 such that
A ( χ , χ ) ≥ α 0 ∥ χ ∥ U , for χ ∈ U 0 .
The bilinear form B ( · , · ) satisfies the inf-sup condition, i.e.,
sup χ ∈ U B ( χ , ρ ) ∥ χ ∥ U ≥ α 1 ∥ ρ ∥ V , for ρ ∈ V .
Proof. 
The continuity of bilinear forms A ( · , · ) follows from the fact that
A ( λ , χ ) = I ( p , q ) + I ( p , b v ) + ( c u , v ) = θ ( I x β 2 0 p , I 1 β 2 x q ) + ( 1 − θ ) ( I 1 β 2 x p , I x β 2 0 q ) + θ ( I x β 2 0 p , I 1 β 2 x ( b v ) ) + ( 1 − θ ) ( I 1 β 2 x p , I 1 β 2 x ( b v ) ) + ( c u , v ) ≤ C ∥ p ∥ − β 2 ∥ q ∥ − β 2 + C ∥ p ∥ − β 2 ∥ v ∥ + C ∥ u ∥ ∥ v ∥ ≤ C { ∥ p ∥ − β 2 2 + ∥ u ∥ 2 } 1 2 { ∥ q ∥ − β 2 2 + ∥ v ∥ 2 } 1 2 = C ∥ λ ∥ U ∥ χ ∥ U , λ = ( u , p ) , χ = ( v , q ) ∈ U .
The continuity of the bilinear forms B ( · , · ) is obvious.
Now we are to prove the coerciveness of bilinear forms A ( · , · ) . For χ = ( v , q ) ∈ U 0 , we have q = D v and v ∈ H 0 1 − β 2 ( Ω ) according to Lemma 3. Thus
A ( χ , χ ) = I ( q , q ) + I ( q , b v ) + ( c v , v ) = θ ( I x β 2 0 q , I 1 β 2 x q ) + ( 1 − θ ) ( I 1 β 2 x q , I x β 2 0 q ) + I ( q , b v ) + ( c v , v ) = ( I x β 2 0 q , I 1 β 2 x q ) + I ( q , b v ) + ( c v , v ) .
By the boundedness of operators I x β 2 0 and I 1 β 2 x , we derive
I ( q , b v ) = θ ( I x β 2 0 q , I 1 β 2 x ( b v ) ) + ( 1 − θ ) ( I 1 β 2 x q , I x β 2 0 ( b v ) ) ≤ θ ∥ I x β 2 0 q ∥ ∥ I 1 β 2 x ( b v ) ∥ + ( 1 − θ ) ∥ I 1 β 2 x q ∥ ∥ I x β 2 0 ( b v ) ∥ ≤ θ Γ ( 1 + β 2 ) ∥ I x β 2 − ∞ q ˜ ∥ ∥ b v ∥ + 1 − θ Γ ( 1 + β 2 ) ∥ I + ∞ β 2 x q ˜ ∥ ∥ b v ∥ ≤ ∥ b ∥ L ∞ ( Ω ) Γ ( 1 + β 2 ) ∥ I x β 2 − ∞ q ˜ ∥ ∥ v ∥ ,
where q ˜ is the zero extension of q to R . By substituting (13) into Equation (12), applying (3) and noting the equivalence of ∥ q ∥ H and ∥ χ ∥ U with χ = ( v , q ) ∈ U 0 by Lemma 4, under Assumption 1, we obtain that for some γ ∈ ( 0 , 1 )
A ( χ , χ ) ≥ cos ( β 2 π ) ∥ I x β 2 − ∞ q ˜ ∥ 2 − ∥ b ∥ L ∞ ( Ω ) Γ ( 1 + β 2 ) ∥ I x β 2 − ∞ q ˜ ∥ ∥ v ∥ + c m i n ∥ v ∥ 2 = ( cos ( β 2 π ) − ∥ b ∥ L ∞ ( Ω ) 2 4 c m i n ( Γ ( 1 + β 2 ) ) 2 ) ∥ I x β 2 − ∞ q ˜ ∥ 2 + ( c m i n ∥ v ∥ − ∥ b ∥ L ∞ ( Ω ) 2 c m i n Γ ( 1 + β 2 ) ∥ I x β 2 − ∞ q ˜ ∥ ) 2 ≥ ( cos ( β 2 π ) − ∥ b ∥ L ∞ ( Ω ) 2 4 c m i n ( Γ ( 1 + β 2 ) ) 2 ) ∥ q ∥ − β / 2 2 ≥ γ cos ( β 2 π ) ∥ q ∥ − β / 2 2 ≥ C 0 ∥ χ ∥ U 2 .
Finally, we show that the bilinear form B ( · , · ) satisfies the inf-sup condition. For any ρ ∈ V , let χ 0 = ( v 0 , q 0 ) ∈ U with v 0 = D ρ , q 0 = ρ . Then
sup χ ∈ U B ( χ , ρ ) ∥ χ ∥ U ≥ B ( χ 0 , ρ ) ∥ χ 0 ∥ U ≥ ( ρ , ρ ) + ( D ρ , D ρ ) ( ∥ ρ ∥ − β 2 2 + ∥ D ρ ∥ 2 ) 1 2 ≥ ( ∥ ρ ∥ 2 + ∥ D ρ ∥ 2 ) 1 2 = ∥ ρ ∥ V .
We finish the proof of the lemma. □
Theorem 1. 
Suppose that condition (8) holds. The problem (7) (or (5)) has a unique solution ( λ , σ ) ∈ U × V .
Proof. 
Suppose that f = 0 . Let χ = λ , ρ = σ in (7), and then substitute (7b) into (7a). We thus obtain
A ( λ , λ ) = 0 .
Applying the coerciveness of A ( · , · ) , we obtain that λ = 0 . Consequently, (7a) becomes
B ( χ , σ ) = 0 , ∀ χ ∈ U ,
which, together with the inf-sup condition (10), implies that σ = 0 . □
Theorem 2. 
If u ∈ H 0 1 − β 2 ( Ω ) solves Equation (2), then ( λ , σ ) ∈ U × V is the solution of (7) with λ = ( u , p ) , p = D u , and σ = − I θ β p . Conversely, ( λ , σ ) ∈ U × V is the solution of (7) with λ = ( u , p ) , then u ∈ H 0 1 − β 2 solves Equation (2).
Proof. 
Assume that u ∈ H 0 1 − β 2 ( Ω ) solves Equation (2). Set λ = ( u , p ) with p = D u and σ = − I θ β p . Then, according to Lemma 3, it follows that λ ∈ U 0 , which leads to (7b). For any χ = ( v , q ) ∈ U
A ( λ , χ ) + B ( χ , σ ) = ( I θ β p , q ) + ( b I θ β p , v ) + ( c u , v ) + ( D σ , v ) + ( σ , q ) = ( D σ + b I θ β p + c u , v ) + ( σ + I θ β p , q ) = ( f , v ) .
Therefore, (7a) holds.
Conversely, suppose that ( λ , σ ) ∈ U × V is the solution of (7) with λ = ( u , p ) . Then (7b) implies that λ ∈ U 0 . By Lemma 3, this leads to p = D u with u ∈ H 0 1 − β 2 ( Ω ) . Therefore, for v ∈ H 0 1 − β 2 ( Ω ) and q = D v , Lemma 3 leads to χ = ( v , q ) ∈ U 0 , which implies, from (7a)
A ( λ , χ ) = F ( χ ) ,
i.e.,
− D I θ β D u + b I θ β D u + c u , v = ( f , v ) , v ∈ W .
This implies that u solves the Equation (2). □

3. Expanded Mixed Finite Element Method

In this section, we develop an expanded mixed finite element method for steady-state FDEs (1) and perform a rigorous convergence analysis, establishing optimal-order error estimates for the fractional flux σ , the gradient p and the primary solution u in their respective norms.
Let ℑ h be a uniform division of Ω by I i = [ x i − 1 , x i ] , i = 1 , 2 , ⋯ , M with x 0 = 0   , x M = 1 ,   h = 1 M . Let V h × H h × W h ⊂ V × H × W be the Raviart–Thomas space [19,20,21,22]
V h = { v h ∈ V ∩ C 0 ( Ω ¯ ) : v h | I i ∈ P k ( I i ) , i = 1 , 2 , ⋯ , M } , k ≥ 1 , H h = { σ h ∈ H : σ h | I i ∈ P k − 1 ( I i ) , i = 1 , 2 , ⋯ , M } , k ≥ 1 , W h = { w h ∈ W : w h | I i ∈ P k − 1 ( I i ) , i = 1 , 2 , ⋯ , M } , k ≥ 1 ,
where P k ( I i ) is the restriction of the set of all polynomials of degree not bigger than k on I i . Then the expanded mixed finite element formulation is to find ( u h , p h , σ h ) ∈ W h × H h × V h such that
( p h , ρ h ) + ( u h , D ρ h ) = 0 , ρ h ∈ V h ,
( σ h , q h ) + I ( p h , q h ) = 0 , q h ∈ H h ,
( D σ h , v h ) + I ( p h , b v h ) + ( c u h , v h ) = ( f , v h ) , v h ∈ W h .
To facilitate the discussion of the solvability of the mixed finite element formulation, we rewrite Equation (16) as follows using the notation of bilinear forms A ( · , · ) , B ( · , · ) : find ( λ h , σ h )   ∈ U h × V h with λ h = ( u h , p h ) such that
A ( λ h , χ h ) + B ( χ h , σ h ) = F ( χ h ) , χ h = ( v h , q h ) ∈ U h ,
B ( λ h , ρ h ) = 0 , ρ h ∈ V h ,
where U h : = W h × H h .
To conduct the convergence analysis, we introduce three projection operators. The first operator, denoted by Π h : V ↦ V h , is defined by
( D ( ρ − Π h ρ ) , w h ) = 0 , ∀ w h ∈ W h .
The other two operators are standard L 2 -projections onto W h (or H h ), denoted by P h W (or P h H ). The properties of projection operator P h W , P h H and Π h are thoroughly discussed in [17]. Here we enumerate them as follows
∥ v − P h W v ∥ − l ≤ C h s + l ∥ v ∥ s , 0 ≤ l , s ≤ k . ∥ v − P h H v ∥ − l ≤ C h s + l ∥ v ∥ s , 0 ≤ l , s ≤ k . ∥ ρ − Π h ρ ∥ θ ≤ C h s − θ ∥ ρ ∥ s , 0 ≤ θ ≤ 1 , 0 ≤ s ≤ k + 1 .
Define the discrete kernel space
U h 0 = { χ h = ( v h , q h ) ∈ U h : B ( χ h , ρ h ) = 0 , ρ h ∈ V h } .
Generally, space U h 0 may not be a subset of U 0 . This implies that the coerciveness of the bilinear form A ( · , · ) on U h 0 may not hold, the inf-sup condition of the bilinear form B ( · , · ) may not be satisfied, and consequently, the solvability of mixed finite element Formulation (17) is not guaranteed.

3.1. Solvability of Mixed Finite Element Formulation

In this subsection, we establish the well-posedness of the mixed finite element Formulation (17) by proving the weak coercivity of the bilinear form A ( · , · ) on the discrete space U h 0 and the discrete inf-sup (LBB) condition for the bilinear form B ( · , · ) .
We characterize the discrete space U h 0 using the lemma presented below.
Lemma 6 
([3]). Suppose that ϕ ∈ H and w h ∈ W h are satisfying
( ϕ , ρ h ) + ( w h , D ρ h ) = 0 , ρ h ∈ V h .
Then the following properties hold
w h = P h W ( 0 I x ϕ ) and I x 0 ϕ ∈ H 0 1 − β 2 ( Ω ) , ∥ w h − I x 0 ϕ ∥ ≤ C h | ϕ | − β 2 , ∥ w h ∥ ≤ C ∥ I x 0 ϕ ∥ ≤ C ∥ ϕ ∥ − β 2 .
Lemma 7. 
Under Equation (8), the bilinear form A ( · , · ) is U h 0 -elliptic, i.e., there exists a constant α 0 such that
A ( χ h , χ h ) ≥ α 0 ∥ χ h ∥ U 2 ,   for   χ h ∈ U h 0 .
For sufficiently small h, the bilinear form B ( · , · ) satisfies the inf-sup condition, i.e., there exists a constant α 1 such that
sup χ h ∈ U h B ( χ h , ρ h ) ∥ χ h ∥ U ≥ α 1 ∥ ρ h ∥ V ,   for   ρ h ∈ V h .
Proof. 
For any χ h = ( v h , q h ) ∈ U h 0 , Lemma 6 leads to
∥ v h ∥ = ∥ P h W I x 0 q h ) ∥ ≤ C ∥ I x 0 q h ∥ ≤ C ∥ q h ∥ H .
Then
∥ q h ∥ H ≤ ∥ χ h ∥ U = ( ∥ v h ∥ 2 + ∥ q h ∥ H 2 ) 1 2 ≤ C ∥ q h ∥ H ,
which implies ∥ χ h ∥ U is equivalent to ∥ q h ∥ H , for any χ h = ( v h , q h ) ∈ U h 0 .
For any χ h = ( v h , q h ) ∈ U h 0 , similar to the derivation of Equation (14), we obtain
A ( χ h , χ h ) ≥ ( cos ( β 2 π ) − ∥ b ∥ L ∞ ( Ω ) 2 4 c m i n ( Γ ( 1 + β 2 ) ) 2 ) ∥ q h ∥ − β / 2 2 .
Applying Equation (20), under Equation (8), we obtain
A ( χ h , χ h ) ≥ α 0 ∥ χ h ∥ U 2 .
Next, we will show the inf-sup condition of the bilinear form B ( · , · ) . For any ρ h ∈ V h , let χ h 0 = ( v h 0 , q h 0 ) ∈ U h with v h 0 = D ρ h and q h 0 = P h H ρ h . Then
sup χ h ∈ U h B ( χ h , ρ h ) ∥ χ h ∥ U ≥ B ( χ h 0 , ρ h ) ∥ χ h 0 ∥ U ≥ ( P h H ρ h , ρ h ) + ( D ρ h , D ρ h ) ( ∥ P h H ρ h ∥ H 2 + ∥ D ρ h ∥ 2 ) 1 2 .
Noting that
( P h H ρ h , ρ h ) = ( ρ h , ρ h ) − ( ρ h − P h H ρ h , ρ h ) ≥ ∥ ρ h ∥ 2 − C h ∥ ρ h ∥ V 2 ,
and
∥ P h H ρ h ∥ H ≤ ∥ ρ h ∥ .
Hence, for sufficiently small h with 1 − C h ≥ 1 2 , we obtain
sup χ h ∈ U h B ( χ h , ρ h ) ∥ χ h ∥ U ≥ ( 1 − C h ) ∥ ρ h ∥ V 2 ∥ ρ h ∥ V ≥ α 1 ∥ ρ h ∥ V ,
where α 1 = 1 2 . □
Analogously to the proof of Theorem 1, by applying Lemma 7, we derive the solvability of Equation (17).
Theorem 3. 
Suppose that Equation (8) holds. Then, for sufficiently small h, Equation (17) (or Equation (16)) has a unique solution ( λ h , σ h ) ∈ U h × V h .

3.2. Convergence Analysis

In this subsection, we perform a rigorous convergence analysis for the expanded mixed finite element method, establishing optimal-order error estimates for the unknown solution u, the gradient p, and the fractional flux σ in their respective norms.
Let ξ = P h W u − u h , η = P h H p − p h , ζ = Π h σ − σ h . Subtracting Equation (16) from Equation (5), we obtain the error equation
( p − p h , ρ h ) + ( ξ , D ρ h ) = 0 , ρ h ∈ V h ,
( σ − σ h , q h ) + I ( p − p h , q h ) = 0 , q h ∈ H h ,
( D ζ , v h ) + I ( p − p h , b v h ) + ( c ( u − u h ) , v h ) = 0 , v h ∈ W h .
Lemma 8. 
There exists a constant C > 0 , such that
∥ σ − σ h ∥ V ≤ ∥ σ − Π h σ ∥ V + C { ∥ u − u h ∥ + ∥ p − p h ∥ H } .
Proof. 
Taking v h = D ζ in Equation (21c) and applying Equation (13) of I ( · , · ) , we obtain
∥ D ζ ∥ 2 = − I ( p − p h , b ( D ζ ) ) − ( c ( u − u h ) , D ζ ) ≤ C ∥ p − p h ∥ H ∥ D ζ ∥ + C ∥ u − u h ∥ ∥ D ζ ∥ ,
which implies the estimate of D ζ
∥ D ζ ∥ ≤ C ∥ p − p h ∥ H + C ∥ u − u h ∥ .
Taking q h = ζ in Equation (21b), we obtain
∥ ζ ∥ 2 = ( σ − σ h , ζ ) − ( σ − Π h σ , ζ ) = − I ( p − p h , ζ ) − ( σ − Π h σ , ζ ) ≤ C ∥ p − p h ∥ H ∥ ζ ∥ + ∥ σ − Π h σ ∥ ∥ ζ ∥ ,
which implies the estimate of ζ
∥ ζ ∥ ≤ C ∥ p − p h ∥ H + ∥ σ − Π h σ ∥ .
Combining Equation (22) with Equation (23), we derive
∥ ζ ∥ V ≤ C { ∥ p − p h ∥ H + ∥ u − u h ∥ } + ∥ σ − Π h σ ∥ ,
which, together with the triangle inequality, lead to the result of the lemma. □
Lemma 9. 
There exists a constant C > 0 , such that
∥ u − u h ∥ 2 +   ∥ p − p h ∥ H 2 ≤ C ∥ u − P h W u ∥ 2 + ∥ p − P h H p ∥ H 2 + ∥ σ − Π h σ ∥ V 2 + C ∥ p − P h H p ∥ H ∥ Π h σ − σ h ∥ V .
Proof. 
Letting q h = η ,   ρ h = ζ and v h = ξ in Equation (21) and combining the three formulae of Equation (21), we obtain
A ( P h λ − λ h , P h λ − λ h ) = − A ( λ − P h λ , P h λ − λ h ) + ( σ − Π h σ , η ) − ( p − P h H p , ζ )
where P h : = ( P h W , P h H ) . An application of the coerciveness and continuity of the bilinear form A ( · , · ) on either U 0 or U h 0 , combined with Young’s inequality, yields
α 0 ∥ P h λ − λ h ∥ U 2 ≤ α 0 2 ∥ P h λ − λ h ∥ U 2 + C ∥ λ − P h λ ∥ U 2 + ∥ σ − Π h σ ∥ V 2 + C ∥ p − P h H p ∥ H ∥ ζ ∥ V ,
which leads to
∥ P h λ − λ h ∥ U 2 ≤ C ∥ λ − P h λ ∥ U 2 + C ∥ σ − Π h σ ∥ V 2 + C ∥ p − P h H p ∥ H ∥ ζ ∥ V .
Combining the above estimate and the triangle inequality, we obtain the estimate of the lemma. □
With the help of Lemma 8, Lemma 9 and the properties of the Raviart–Thomas operator, L 2 -projections, we derive the main convergence result.
Theorem 4. 
Suppose that the solution u ∈ H s ( Ω ) , p ∈ H s − 1 ( Ω ) , σ ∈ H s − 1 + β ( Ω ) with s ≥ 2 − β . Then there exists a constant C such that
∥ u − u h ∥ + ∥ p − p h ∥ H + ∥ σ − σ h ∥ V ≤ C h min { s − 2 + β , k } { ∥ u ∥ H s ( Ω ) + ∥ p ∥ H s − 1 ( Ω ) + ∥ σ ∥ H s − 1 + β ( Ω ) } .
Proof. 
Combining Equations (25), (22), and (23) and applying Young’s inequality, we obtain
∥ u − u h ∥ 2 + ∥ p − p h ∥ H 2 ≤ C ∥ u − P h W u ∥ 2 + C ∥ p − P h H p ∥ H 2 + C ∥ σ − Π h σ ∥ V 2 + C ∥ p − P h H p ∥ H ( ∥ u − u h ∥ + ∥ p − p h ∥ H + ∥ σ − Π h σ ∥ V ) ≤ C ∥ u − P h W u ∥ 2 + C ∥ p − P h H p ∥ H 2 + C ∥ σ − Π h σ ∥ V 2 + 1 2 { ∥ u − u h ∥ 2 + ∥ p − p h ∥ H 2 } ,
which leads to
∥ u − u h ∥ 2 + ∥ p − p h ∥ H 2 ≤ C { ∥ u − P h W u ∥ 2 + ∥ p − P h H p ∥ H 2 + ∥ σ − Π h σ ∥ V 2 } .
By Lemma 8, we obtain
∥ σ − σ h ∥ V ≤ C { ∥ u − P h W u ∥ + ∥ p − P h H p ∥ H + ∥ σ − Π h σ ∥ V } .
Applying the properties of the Raviart–Thomas operator and the L 2 projection operators, we complete the proof. □

4. Numerical Experiments

In this section, we conduct several numerical experiments to verify the theoretically established convergence rates for FDEs derived in Section 3.2. The numerical examples employ the lowest-order Raviart–Thomas finite element space R T 0 for the primal variable and the flux, and the space of piecewise constant functions for the gradient variables, denoted by H h .
Example 1. 
Let k ( x ) = 1 x + 1 , d ( x ) = 2 ( x + 1 ) with the right-hand function
g ( x ) = 1 x + 1 − θ ω 2 ( x ) − ( 1 − θ ) ω 2 ( 1 − x ) − 1 ( x + 1 ) 2 − θ ω 1 ( x ) + ( 1 − θ ) ω 1 ( 1 − x ) + 2 ( x + 1 ) x 2 ( 1 − x ) 2 ,
where
ω 1 ( x ) = 24 Γ ( 4 + β ) x 3 + β − 12 Γ ( 3 + β ) x 2 + β + 2 Γ ( 2 + β ) x 1 + β , ω 2 ( x ) = d ω 1 ( x ) d x = 24 Γ ( 3 + β ) x 2 + β − 12 Γ ( 2 + β ) x 1 + β + 2 Γ ( 1 + β ) x β .
The exact solution u , p , σ are
u ( x ) = x 2 ( 1 − x ) 2 , p ( x ) = 4 x 3 − 6 x 2 + 2 x , σ ( x ) = − θ ω 1 ( x ) + ( 1 − θ ) ω 1 ( 1 − x ) .
In this example, the solution is u ∈ H 2.5 − ϵ ( Ω ) , p ∈ H 1.5 − ϵ ( Ω ) and σ ∈ H 1.5 + β − ϵ ( Ω ) , respectively, where ϵ > 0 is sufficiently small. From Table 1, the convergence rates of ∥ u − u h ∥ , ∥ p − p h ∥ H and ∥ σ − σ h ∥ 1 are 1.00 ,   1.40 and 1.00 , which are consistent with projection errors, that is, they are optimal and larger than theoretical errors.
Table 1. Numerical results for Example 1 with θ = 0.6 ,   β = 0.8 .
Example 2. 
Let k ( x ) = 1 x + 1 , d ( x ) = 2 , right-hand function
g ( x ) = 1 x + 1 − θ 1 Γ ( β ) x β − 1 + ( 1 − θ ) 1 Γ ( β ) ( 1 − x ) β − 1 − 1 ( x + 1 ) 2 − θ 1 Γ ( 1 + β ) x β − ( 1 − θ ) 1 Γ ( 1 + β ) ( 1 − x ) β − C κ C * * ( γ + 1 ) Γ ( 1 − β ) + 2 x − C κ x γ + 1 F 1 2 ( β + γ , γ + 1 , γ + 2 , x ) ,
where C κ = F 1 2 ( β + γ , γ + 1 , γ + 2 , 1 ) − 1 , C * * = sin π β sin ( π γ ) + sin ( π ( − β − γ ) ) with γ ∈ ( − 1 , 0 ) satisfying
θ sin ( π ( − β − γ ) ) = ( 1 − θ ) sin ( π γ ) .
Then, the solution u , p , σ are,
u ( x ) = x − C κ x γ + 1 F 1 2 ( β + γ , γ + 1 , γ + 2 , x ) , p ( x ) = 1 − C κ ( γ + 1 ) x γ ( 1 − x ) − β − γ , σ ( x ) = − θ 1 Γ ( 1 + β ) x β − ( 1 − θ ) 1 Γ ( 1 + β ) ( 1 − x ) β − C κ C * * ( γ + 1 ) Γ ( 1 − β ) ,
where
F 1 2 ( a , b ; c ; x ) = Γ ( c ) Γ ( b ) Γ ( c − b ) ∫ 0 1 t b − 1 ( 1 − t ) c − b − 1 ( 1 − x t ) − a d t ,
refers to the Gaussian hypergeometric function.
In this example, we select β = 0.8 and θ = 0.6 . Then, by Equation (28), we have γ = − 0.58 ,   − β − γ = − 0.22 . The appearance of the singularity x γ + 1 F 1 2 ( β + γ , γ + 1 , γ + 2 , x )   = ( γ + 1 ) ∫ 0 x s γ ( 1 − s ) − β − γ d s in the solution u results in a lower regularity. Specifically, u ∈ H min { γ , − β − γ } + 1.5 − ϵ ( Ω ) , p ∈ H min { γ , − β − γ } + 0.5 − ϵ ( Ω ) and σ ∈ H β + 0.5 − ϵ ( Ω ) for a sufficiently small ϵ > 0 . The errors and convergence rates are presented in Table 2. From Table 2, the convergence rates for ∥ u − u h ∥ , ∥ p − p h ∥ H and ∥ σ − σ h ∥ 1 are 0.92 ,   0.32 and 0.30 , respectively. These rates are consistent with the projection errors, which means that they are optimal and larger than the theoretical errors, despite the lower regularity.
Table 2. Numerical results for Example 2 with θ = 0.6 , β = 0.8 .
Example 3. 
Let k ( x ) = 1 x + 1 ,   d ( x ) = 2 , right-hand function g ( x ) = 1 .
For this selection of the right-hand function g ( x ) , the solution u is not available. To compute convergence rates, an approximation solution is employed on a refined mesh (with h = 2 − 13 ) as a reference solution.
In this example, we continue to select β = 0.8 , θ = 0.6 . The lower regularity term x γ + 1 F 1 2 ( β + γ , γ + 1 , γ + 2 , x ) = ( γ + 1 ) ∫ 0 x s γ ( 1 − s ) − β − γ d s might be part of the solution u, since it is a kernel of the fractional operator D ( I θ β D u ) , which results in the regularity of the solution u ∈ H min { γ , − β − γ } + 1.5 − ϵ ( Ω ) , analogous to Example 2. Therefore, the solution is p = D u ∈ H min { γ , − β − γ } + 0.5 − ϵ ( Ω ) . The errors and convergence rates are presented in Table 3. From Table 3, the convergence rates for ∥ u − u h ∥ , ∥ p − p h ∥ H and ∥ σ − σ h ∥ 1 are optimal and exceed the theoretical errors.
Table 3. Numerical results for Example 3 with θ = 0.6 ,   β = 0.8 .

5. Conclusions

In this article, we propose an expanded mixed finite element method for the variable-coefficient space-fractional diffusion equation. We construct an expanded variational formulation and rigorously prove the weak coercivity of the bilinear form A ( · , · ) and the inf-sup condition of the bilinear form B ( · , · ) , which together ensure the existence and uniqueness of the solution to the variational formulation. Based on this well-posedness result, we develop the corresponding expanded mixed finite element scheme and conduct the convergence analysis. Several numerical experiments are carried out to verify the theoretical convergence rates and demonstrate the practical efficiency of the proposed method.
The dense stiffness matrices arising from non-local fractional operators lead to high computational complexity and memory requirements. Therefore, developing efficient fast algorithms and systematically comparing their efficiency should be a priority for future research.

Author Contributions

Methodology, S.Y., H.C. and F.W.; software, S.Y. and F.W.; validation, S.Y. and H.C.; formal analysis, S.Y.; writing—original draft preparation, S.Y.; writing—review and editing, S.Y., H.C. and F.W.; funding acquisition, S.Y., H.C. and F.W. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded in part by the Shandong Provincial Natural Science Foundation under grant No. ZR2022MA068 and No. ZR2022QA061 and the National Natural Science Foundation of China under Grant No. 12171287.

Data Availability Statement

All data generated or analyzed during this study are included in this article.

Conflicts of Interest

The authors declare that they have no conflicts of interest.

References

  1. Ervin, V.J.; Roop, J.P. Variational formulation for the stationary fractional advection dispersion equation. Numer. Methods Partial Differ. Equ. 2005, 22, 558–576. [Google Scholar] [CrossRef] [Scilit]
  2. Wang, H.; Yang, D.P. Wellposedness of variable-coefficient conservative fractional elliptic differential equations. SIAM J. Numer. Anal. 2013, 51, 1088–1107. [Google Scholar] [CrossRef] [Scilit]
  3. Chen, J.; Wang, F.; Chen, H.Z. Probability-conservative simulation for Lévy financial model by a mixed finite element method. Comp. Math. Appl. 2022, 106, 92–105. [Google Scholar] [CrossRef] [Scilit]
  4. Ervin, V.J.; Heuer, N.; Roop, J.P. Regularity of the solution to 1-D fractional order diffusion equations. Math. Comput. 2018, 87, 2273–2294. [Google Scholar] [CrossRef] [Scilit]
  5. Fu, H.; Liu, H.; Wang, H. A finite volume method for two-dimensional Riemann-Liouville space fractional diffusion equation and its efficient implementation. J. Comput. Phys. 2019, 388, 316–334. [Google Scholar] [CrossRef] [Scilit]
  6. Liu, F.; Zhuang, P.; Turner, I.; Burrage, K.; Anh, V. A new fractional finite volume method for solving the fractional diffusion equation. Appl. Math. Model. 2014, 38, 3871–3878. [Google Scholar] [CrossRef] [Scilit]
  7. Hao, Z.; Park, M.; Lin, G.; Cai, Z. Finite element method for two-sided fractional differential equations with variable coefficients: Galerkin approach. J. Sci. Comput. 2019, 79, 700–717. [Google Scholar] [CrossRef] [Scilit]
  8. Liu, H.; Zheng, X.C.; Wang, H.; Fu, H.F. Error estimate of finite element approximation for two-sided space-fractional evolution equation with variable coefficient. J. Sci. Comput. 2022, 90, 15. [Google Scholar] [CrossRef] [Scilit]
  9. Zheng, X.C.; Ervin, V.J.; Wang, H. An indirect finite element method for variable-coefficient space fractional diffusion equations and its optimal-order error estimates. Commun. Appl. Math. Comput. 2020, 2, 147–162. [Google Scholar] [CrossRef] [Scilit]
  10. Zheng, X.C.; Liu, H.; Wang, H.; Fu, H.F. Optimal-order finite element approximations to variable-coefficient two-sided space-fractional advection-reaction-diffusion equations in three space dimensions. Appl. Numer. Math. 2021, 161, 1–12. [Google Scholar] [CrossRef] [Scilit]
  11. Brezzi, F. On the existence, uniqueness and approximation of saddle-point problems arising from lagrangian multipliers. RAIRO Anal. Numer. 1974, 8, 129–151. [Google Scholar] [CrossRef] [Scilit]
  12. Wang, H.; Zhu, S.F. A Petrov-Galerkin finite element method for variable-coefficient fractional diffusion equations. Comput. Methods Appl. Mech. Engrg. 2015, 290, 45–56. [Google Scholar] [CrossRef] [Scilit]
  13. Li, Y.S.; Chen, H.Z.; Wang, H. A mixed-type Galerkin variational formulation and fast algorithms for variable-coefficient fractional diffusion equations. Math. Method Appl. Sci. 2017, 40, 5018–5034. [Google Scholar] [CrossRef] [Scilit]
  14. Yang, S.X.; Chen, H.Z.; Ervin, V.J.; Wang, H. Solvability and approximation of two-side conservative fractional diffusion problems with variable-Coefficient based on least-squares. Appl. Math. Comput. 2021, 406, 126229. [Google Scholar] [CrossRef] [Scilit]
  15. Yang, S.X.; Chen, H.Z.; Wang, H. Least-squared mixed variational formulation based on space decomposition for a kind of variable-coefficient fractional diffusion problems. J. Sci. Comput. 2019, 78, 687–709. [Google Scholar] [CrossRef] [Scilit]
  16. Deng, W.J.; Hesthaven, S. Local discontinuous Galerkin methods for fractional diffusion equations. ESAIM-Math. Model. Numer. Anal. 2013, 47, 1845–1864. [Google Scholar] [CrossRef] [Scilit]
  17. Chen, H.Z.; Wang, H. Numerical simulation for conservative fractional diffusion equations by an expanded mixed formulation. J. Comp. Appl. Math. 2016, 296, 480–498. [Google Scholar] [CrossRef] [Scilit]
  18. Douglas, J.; Roberts, J.E. Global estimates for mixed methods for second order elliptic equations. Math. Comp. 1985, 44, 39–52. [Google Scholar] [CrossRef]
  19. Brezzi, F.; Douglas, J.; Durán, R.; Fortin, M. Mixed finite elements for second order elliptic problems in three variables. Numer. Math. 1987, 51, 237–250. [Google Scholar] [CrossRef] [Scilit]
  20. Brezzi, F.; Douglas, J.; Fortin, M.; Marini, L.D. Two families of mixed finite element for second-order elliptic problems. Numer. Math. 1985, 47, 217–235. [Google Scholar] [CrossRef] [Scilit]
  21. Brezzi, F.; Fortin, M. Mixed and Hybrid Finite Element Methods; Springer: New York, NY, USA, 1991. [Google Scholar]
  22. Raviat, P.A.; Thomas, J.M. A mixed finite element methods for 2nd order elliptic problems. In Mathmatical Aspects of the Finite Element Method; Lecture Notes in Maths; Springer: Berlin, Germany, 1977; Volume 606, pp. 292–315. [Google Scholar]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.