betaFeedback

WorksMathematics

Uniqueness for an inverse problem of determining orders and coefficients for multi-term time-fractional diffusion equation

Abstract

This paper investigates a uniqueness problem for determining all fractional orders and their corresponding coefficients in multi-term time-fractional diffusion equations with Caputo derivatives in an abstract Hilbert space. The main result establishes that these parameters can be uniquely identified from a single scalar observation of the solution, of the form (u(t), phi), by comparing the solutions of two evolution equations governed by possibly different positive self-adjoint operators A and B. In contrast to existing results that rely on local space-time measurements and, in particular, may require the initial data to satisfy sign conditions and strong spatial regularity assumptions, our approach uses only the minimal assumption that a belongs to H, without imposing any sign or additional smoothness conditions on the initial value. The proof combines spectral representations of the solutions with a recursive asymptotic comparison method based on fundamental identities for the multinomial Mittag-Leffler function. This approach enables the successive identification of all fractional orders and their associated coefficients, even when the two equations involve different operators. Furthermore, the uniqueness argument extends to multi-term time-fractional wave equations with orders in the interval (1,2).

Introduction

Fractional differential equations have become an important tool for modeling anomalous diffusion, viscoelasticity, heat conduction in heterogeneous media, signal processing, and many other phenomena exhibiting memory and hereditary effects. In recent decades, substantial attention has been devoted to the mathematical analysis of fractional evolution equations, including questions of existence, uniqueness, regularity, asymptotic behavior, controllability, and inverse problems. Among various models, multi-term time-fractional diffusion equations have attracted particular interest because they provide a more accurate description of complex physical processes than their single-term counterparts.

In many practical applications, the parameters appearing in fractional differential equations are not known a priori and must be identified from experimental or observational data. Consequently, inverse problems concerning the determination of fractional orders and associated coefficients have become an active area of research. Such problems are of considerable importance because the fractional orders characterize the memory properties of the underlying medium and therefore play a crucial role in the interpretation of physical processes.

Key words and phrases. Uniqueness in determining orders and coefficients, Multinomial Mittag–Leffler function, Multi-term time-fractional diffusion equation.

Let AA be an arbitrary positive self-adjoint operator defined in a separable Hilbert space HH with domain D(A)D(A). Denote the inner product in HH by (⋅,⋅)(\cdot,\cdot) and the norm by ∥⋅∥\|\cdot\|. Suppose further that the inverse operator A−1A^{-1} is compact. Then it is well known that AA possesses a complete orthonormal system of eigenfunctions {vk}\{v_k\} and a countable set of positive eigenvalues {λk}\{\lambda_k\} with a unique accumulation point at +∞+\infty. We assume that the eigenvalues are ordered in a nondecreasing manner, i.e.,

0<λ1≤λ2≤⋯→+∞.0 < \lambda_1 \le\lambda_2 \le\cdots\to+\infty.

The Caputo fractional derivative of order 0<αj<10 < \alpha_j < 1 (j=1,…,nj=1,\ldots,n) for a function h(t)h(t) is defined as (see [1], p. 92):

∂tαjh(t)=1Γ(1−αj)∫0th′(s)(t−s)αj ds,t>0,\partial_t^{\alpha_j} h(t)=\frac{1}{\Gamma(1-\alpha_j)}\int_0^t\frac{h'(s)}{(t-s)^{\alpha_j}}\,ds,\qquad t>0,

provided that the right-hand side exists. Here, Γ(⋅)\Gamma(\cdot) is Euler’s gamma function.

Consider the following multi-term fractional diffusion equation with Caputo derivatives:

{∑j=1nqj∂tαju(t)+Au(t)=0,0<t≤T,u(0)=a,(1)\left\{ \begin{aligned} \sum_{j=1}^{n}q_j\partial_t^{\alpha_j}u(t)+Au(t)&=0,\qquad0<t\le T,\\ u(0)&=a, \end{aligned} \right. \tag*{(1)}

where

0<αn<αn−1<⋯<α1<1,0<\alpha_n<\alpha_{n-1}<\cdots<\alpha_1<1,

qj>0q_j>0 for j=1,…,nj=1,\ldots,n (without loss of generality, q1=1q_1=1) and a∈Ha\in H.

For single-term time-fractional diffusion equations, inverse problems for determining the fractional order have been extensively investigated by many authors. Various uniqueness and reconstruction results have been obtained using additional observations at interior points, boundary measurements, integral conditions, and overdetermination data. We refer the reader to [2, 3, 4, 5, 6, 7, 8, 9] and the references therein for a comprehensive overview of this topic.

Compared with the single-term case, inverse problems for multi-term time-fractional equations are significantly more challenging because several fractional orders interact simultaneously in the solution representation. As a consequence, the available literature on such problems remains relatively limited.

One of the first systematic studies in this direction was carried out by [10], where uniqueness results for determining fractional orders in multi-term time-fractional diffusion equations were established by means of eigenfunction expansions and Laplace transform techniques.

More recently, Jiang and Li [11] investigated an inverse problem for multi-term time-fractional wave equations and established uniqueness in determining the fractional orders. Their approach is based on local space–time observation data and requires relatively strong assumptions on the initial value. Moreover, their uniqueness result is established under the assumption that two solutions corresponding to different fractional orders coincide on a nonempty space–time subdomain ω×(0,T)\omega\times(0,T), where ω⊂Ω\omega\subset\Omega is nonempty. More precisely, they consider the following pair of initial-boundary value problems:

{∑j=1mcj∂tαju(x,t)−Δu(x,t)+p(x)u(x,t)=0,(x,t)∈Ω×(0,T),u(x,0)=a(x),ut(x,0)=0,x∈Ω,u(x,t)=0,(x,t)∈∂Ω×(0,T).\left\{ \begin{aligned} \sum_{j=1}^{m} c_j \partial_t^{\alpha_j}u(x,t)-\Delta u(x,t)+p(x)u(x,t)&=0, &&(x,t)\in\Omega\times(0,T),\\ u(x,0)&=a(x),\quad u_t(x,0)=0, &&x\in\Omega,\\ u(x,t)&=0, &&(x,t)\in\partial\Omega\times(0,T). \end{aligned} \right.

and

{∑j=1mcj∂tβjv(x,t)−Δv(x,t)+q(x)v(x,t)=0,(x,t)∈Ω×(0,T),v(x,0)=a(x),vt(x,0)=0,x∈Ω,v(x,t)=0,(x,t)∈∂Ω×(0,T).\left\{ \begin{aligned} \sum_{j=1}^{m} c_j \partial_t^{\beta_j}v(x,t)-\Delta v(x,t)+q(x)v(x,t)&=0, &&(x,t)\in\Omega\times(0,T),\\ v(x,0)&=a(x),\quad v_t(x,0)=0, &&x\in\Omega,\\ v(x,t)&=0, &&(x,t)\in\partial\Omega\times(0,T). \end{aligned} \right.

where 1<αm<⋯<α1<21<\alpha_m<\cdots<\alpha_1<2, 1<βm<⋯<β1<21<\beta_m<\cdots<\beta_1<2, cj>0c_j>0, (c1=1)(c_1=1), j=1,…,mj=1,\ldots,m. The proof of the uniqueness of the potential, namely p(x)=q(x)p(x)=q(x), can be found in [12]. In contrast, the present work adopts a different observation strategy and a more general operator framework. We consider two multi-term time-fractional evolution equations governed by possibly distinct positive self-adjoint operators AA and BB in an abstract Hilbert space HH.

Moreover, to determine the fractional orders and their corresponding coefficients, we consider the following observation data:

(u(t),ϕ),0<t<T.(u(t),\phi),\qquad0<t<T.

for some prescribed weight function ϕ\phi. In some existing approaches, uniqueness is established using different types of observation data. For example, in [10], the observation is given by pointwise data at a fixed spatial point, whereas in [11], uniqueness is established by requiring two solutions to coincide throughout a nonempty subdomain ω⊂Ω\omega\subset\Omega over the observation interval (0,T)(0,T). Such spatially distributed pointwise measurements are difficult to obtain in practice, since physical sensors have a finite spatial extent rather than acting as ideal point detectors. Our approach requires only the scalar quantity (u(t),ϕ)(u(t),\phi) at each time t∈(0,T)t\in(0,T) and also allows us to establish uniqueness under weaker regularity assumptions on the initial data aa. Finally, the weight function ϕ\phi may be chosen to vanish outside ω\omega. As is readily seen from the proofs, all the main results remain valid with this choice.

For the single-term time-fractional diffusion equation, uniqueness results for determining the fractional order from observation data were obtained in [13] under such a condition.

The proof is based on the spectral representation of solutions and a detailed asymptotic analysis of the multinomial Mittag–Leffler function. A major difficulty arises from the appearance of multiple competing asymptotic terms generated by different fractional orders. To overcome this difficulty, we repeatedly employ a fundamental identity for the multinomial Mittag–Leffler function and develop a recursive asymptotic comparison procedure. This method allows us to isolate dominant contributions successively and recover each fractional order together with its corresponding coefficient in a step-by-step manner.

The remainder of the paper is organized as follows. In § 2, we present preliminary results concerning spectral theory and the multinomial Mittag–Leffler function. § 3 is devoted to the formulation and analysis of the inverse problem, where the uniqueness theorem is proved. Finally, concluding remarks are given in § 4.

Preliminaries

Let τ\tau be an arbitrary real number. We introduce the power of the operator AA, acting in HH according to the rule:

Aτh=∑k=1∞λkτhkvk.A^\tau h = \sum_{k=1}^{\infty} \lambda_k^\tau h_k v_k.

Here and everywhere below, for the vector h∈Hh \in H, the symbol hkh_k denotes the Fourier coefficients of this vector: hk=(h,vk)h_k = (h,v_k). The domain of definition of this operator is determined by the condition Aτh∈HA^\tau h \in H and has the form:

D(Aτ)={h∈H:∑k=1∞λk2τ∣hk∣2<∞}.D(A^\tau)=\left\{h\in H:\sum_{k=1}^{\infty}\lambda_k^{2\tau}|h_k|^2<\infty\right\}.

The solution of problem (1.1) involves the multivariable (multinomial) Mittag–Leffler function. Therefore, we first present the definition of the multinomial Mittag–Leffler function.

The multinomial Mittag–Leffler function is defined as

E(α1,…,αn),α0(z1,…,zn)=∑k=0∞∑k1+⋯+kn=k(kk1,…,kn)∏j=1nzjkjΓ(α0+∑j=1nαjkj).E_{(\alpha_1,\ldots,\alpha_n),\alpha_0}(z_1,\ldots,z_n)=\sum_{k=0}^{\infty}\sum_{k_1+\cdots+k_n=k}\binom{k}{k_1,\ldots,k_n}\frac{\prod_{j=1}^{n}z_j^{k_j}}{\Gamma\left(\alpha_0+\sum_{j=1}^{n}\alpha_j k_j\right)}.

where (kk1,…,kn)\binom{k}{k_1,\ldots,k_n} denotes the multinomial coefficient

(kk1,…,kn)=k!k1!⋯kn!,with k=∑j=1nkj,\binom{k}{k_1,\ldots,k_n}=\frac{k!}{k_1!\cdots k_n!},\qquad\text{with } k=\sum_{j=1}^{n}k_j,

and kjk_j, 1≤j≤n1\leq j\leq n, are non-negative integers. For more details of the multinomial Mittag–Leffler function, see [14], [15].

Lemma 2.1. [16] Let αj>0\alpha_j>0 (j=1,…,nj=1,\ldots,n) and zj∈Cz_j\in\mathbb{C} (j=1,…,nj=1,\ldots,n). Then

1Γ(α0)+∑j=1nzjE(α1,…,αn),α0+αj(z1,…,zn)=E(α1,…,αn),α0(z1,…,zn).\frac{1}{\Gamma(\alpha_0)}+\sum_{j=1}^{n}z_j E_{(\alpha_1,\ldots,\alpha_n),\alpha_0+\alpha_j}(z_1,\ldots,z_n)=E_{(\alpha_1,\ldots,\alpha_n),\alpha_0}(z_1,\ldots,z_n).

For later use, we adopt the abbreviation

Eα′,β(−λktα1,∗)=E(α1,α1−α2,…,α1−αn),β(−λktα1,−q2tα1−α2,…,−qntα1−αn).E_{\alpha',\beta}(-\lambda_k t^{\alpha_1},*)=E_{(\alpha_1,\alpha_1-\alpha_2,\ldots,\alpha_1-\alpha_n),\beta}(-\lambda_k t^{\alpha_1},-q_2t^{\alpha_1-\alpha_2},\ldots,-q_nt^{\alpha_1-\alpha_n}).

By Lemma (2.1), we obtain the following identity

Eα′,1+α1(−λktα1,∗)=1Γ(1+α1)−λktα1Eα′,1+2α1(−λktα1,∗)−∑j=2nqjtα1−αjEα′,1+2α1−αj(−λktα1,∗).(2)E_{\alpha',1+\alpha_1}(-\lambda_k t^{\alpha_1},*)= \frac{1}{\Gamma(1+\alpha_1)} -\lambda_k t^{\alpha_1}E_{\alpha',1+2\alpha_1}(-\lambda_k t^{\alpha_1},*) -\sum_{j=2}^{n}q_jt^{\alpha_1-\alpha_j}E_{\alpha',1+2\alpha_1-\alpha_j}(-\lambda_k t^{\alpha_1},*). \tag*{(2)}

Lemma 2.2. [17] Let β>0\beta> 0 and α1>α2>⋯>αm>0\alpha_1 > \alpha_2 > \cdots> \alpha_m > 0 be given with α1∈(0,2)\alpha_1 \in(0, 2). Assume that α1π/2<μ<min⁡{α1π,π}\alpha_1\pi/2 < \mu< \min\{\alpha_1\pi, \pi\}, μ≤∣arg⁡(z1)∣≤π\mu\leq|\arg(z_1)| \leq\pi, and there exists K>0K > 0 such that −K≤zj<0-K \leq z_j < 0 for j=2,…,mj = 2,\ldots,m. Then there exists a constant C>0C > 0 depending only on μ\mu, KK, αj\alpha_j (j=1,…,mj = 1,\ldots,m) and β\beta such that

∣E(α1,…,αm),β(z1,…,zm)∣≤C1+∣z1∣.\left|E_{(\alpha_1,\ldots,\alpha_m),\beta}(z_1,\ldots,z_m)\right| \leq\frac{C}{1+|z_1|}.

Lemma 2.3. [14] The solution of (1) is represented by

u(t)=∑k=1∞(1−λktα1Eα′,1+α1(−λktα1,∗))akvk.u(t)=\sum_{k=1}^{\infty}\left(1-\lambda_k t^{\alpha_1}E_{\alpha',1+\alpha_1}(-\lambda_k t^{\alpha_1},*)\right)a_kv_k.

where (a,vk)=ak(a,v_k)=a_k, a∈Ha\in H.

Inverse Problem

Let us formulate our main result. For this, we will look at another system

{∑j=1nμj∂tβjω(t)+Bω(t)=0,0<t≤T,ω(0)=a,(3)\begin{cases} \sum_{j=1}^{n}\mu_j\partial_t^{\beta_j}\omega(t)+B\omega(t)=0, & 0<t\leq T,\\ \omega(0)=a, \tag*{(3)} \end{cases}

where 0<βn<βn−1<⋯<β1<10 < \beta_n < \beta_{n-1} < \cdots< \beta_1 < 1, μj>0\mu_j > 0 for j=1,…,nj=1,\ldots,n (without loss of generality, μ1=1\mu_1=1) and BB is a positive definite self-adjoint operator with compact inverse. By the spectral theorem, the operator BB possesses a complete orthonormal system of eigenfunctions {v~k}\{\widetilde{v}_k\} corresponding to positive eigenvalues {λ~k}\{\widetilde{\lambda}_k\}. Then there exists a unique solution [14]:

ω(t)=∑k=1∞(1−λ~ktβ1Eβ′,1+β1(−λ~ktβ1,∗))a~kv~k.\omega(t)=\sum_{k=1}^{\infty}\left(1-\widetilde{\lambda}_k t^{\beta_1}E_{\beta',1+\beta_1}(-\widetilde{\lambda}_k t^{\beta_1},*)\right)\widetilde{a}_k\widetilde{v}_k.

where (a,v~k)=a~k(a,\widetilde{v}_k)=\widetilde{a}_k and a∈Ha\in H.

Theorem 3.1. Let ϕ∈D(A2)∩D(B2)\phi\in D(A^2)\cap D(B^2), and (a,Aϕ)≠0(a,A\phi)\ne0, (a,Bϕ)≠0(a,B\phi)\ne0. If

(u(t),ϕ)=(ω(t),ϕ),0<t<T,(u(t),\phi)=(\omega(t),\phi),\qquad0<t<T,

then

(a,Aϕ)=(a,Bϕ),αj=βj,qj=μj,j=1,…,n.(a,A\phi)=(a,B\phi),\qquad\alpha_j=\beta_j,\qquad q_j=\mu_j,\qquad j=1,\ldots,n.

Proof. Using Lemma 2.3 together with formula (2), we derive the following expression:

(u(t),ϕ)=∑k=1∞ak(ϕ,vk)−∑k=1∞λkak(ϕ,vk)tα1Γ(1+α1)+∑k=1∞λk2ak(ϕ,vk)t2α1Eα′,1+2α1(−λktα1,∗)+∑k=1∞λkak(ϕ,vk)∑j=2nqjt2α1−αjEα′,1+2α1−αj(−λktα1,∗).(4)\begin{aligned} (u(t),\phi)={}&\sum_{k=1}^{\infty}a_k(\phi,v_k)-\sum_{k=1}^{\infty}\frac{\lambda_k a_k(\phi,v_k)t^{\alpha_1}}{\Gamma(1+\alpha_1)}\\ &+\sum_{k=1}^{\infty}\lambda_k^2a_k(\phi,v_k)t^{2\alpha_1}E_{\alpha',1+2\alpha_1}(-\lambda_k t^{\alpha_1},*)\\ &+\sum_{k=1}^{\infty}\lambda_k a_k(\phi,v_k)\sum_{j=2}^{n}q_jt^{2\alpha_1-\alpha_j}E_{\alpha',1+2\alpha_1-\alpha_j}(-\lambda_k t^{\alpha_1},*). \tag*{(4)} \end{aligned}

and

(ω(t),ϕ)=∑k=1∞a~k(ϕ,v~k)−∑k=1∞λ~ka~k(ϕ,v~k)tβ1Γ(1+β1)+∑k=1∞λ~k2a~k(ϕ,v~k)t2β1Eβ′,1+2β1(−λ~ktβ1,∗)+∑k=1∞λ~ka~k(ϕ,v~k)∑j=2nμjt2β1−βjEβ′,1+2β1−βj(−λ~ktβ1,∗).(5)(\omega(t),\phi)=\sum_{k=1}^{\infty}\widetilde{a}_k(\phi,\widetilde{v}_k)-\sum_{k=1}^{\infty}\frac{\widetilde{\lambda}_k\widetilde{a}_k(\phi,\widetilde{v}_k)t^{\beta_1}}{\Gamma(1+\beta_1)}+\sum_{k=1}^{\infty}\widetilde{\lambda}_k^2\widetilde{a}_k(\phi,\widetilde{v}_k)t^{2\beta_1}E_{\beta',1+2\beta_1}(-\widetilde{\lambda}_k t^{\beta_1},*) +\sum_{k=1}^{\infty}\widetilde{\lambda}_k\widetilde{a}_k(\phi,\widetilde{v}_k)\sum_{j=2}^{n}\mu_jt^{2\beta_1-\beta_j}E_{\beta',1+2\beta_1-\beta_j}(-\widetilde{\lambda}_k t^{\beta_1},*). \tag*{(5)}

Since the estimates in (4) and (5) are obtained in the same manner, it suffices to consider only (4) in the subsequent analysis.

We introduce the following notation:

I1(t)=∑k=1∞λk2ak(ϕ,vk)t2α1Eα′,1+2α1(−λktα1,∗).I_1(t)=\sum_{k=1}^{\infty}\lambda_k^2a_k(\phi,v_k)t^{2\alpha_1}E_{\alpha',1+2\alpha_1}(-\lambda_k t^{\alpha_1},*).
I2(t)=∑k=1∞λkak(ϕ,vk)∑j=2nqjt2α1−αjEα′,1+2α1−αj(−λktα1,∗).I_2(t)=\sum_{k=1}^{\infty}\lambda_ka_k(\phi,v_k)\sum_{j=2}^{n}q_jt^{2\alpha_1-\alpha_j}E_{\alpha',1+2\alpha_1-\alpha_j}(-\lambda_k t^{\alpha_1},*).

Using Lemma 2.2, we obtain

∣I1(t)∣≤Ct2α1∑k=1∞λk2∣ak∣∣(ϕ,vk)∣,|I_1(t)|\le Ct^{2\alpha_1}\sum_{k=1}^{\infty}\lambda_k^2|a_k||(\phi,v_k)|,
∣I2(t)∣≤C∑j=2nqjt2α1−αj∑k=1∞λk∣ak∣∣(ϕ,vk)∣.|I_2(t)|\le C\sum_{j=2}^{n}q_jt^{2\alpha_1-\alpha_j}\sum_{k=1}^{\infty}\lambda_k|a_k||(\phi,v_k)|.

Using the Cauchy–Schwarz inequality and since a∈Ha\in H, ϕ∈D(A2)∩D(B2)\phi\in D(A^2)\cap D(B^2), we have

∣I1(t)∣≤Ct2α1(∑k=1∞∣ak∣2)1/2(∑k=1∞λk4∣(ϕ,vk)∣2)1/2=Ct2α1∥a∥∥A2ϕ∥.|I_1(t)|\le Ct^{2\alpha_1}\left(\sum_{k=1}^{\infty}|a_k|^2\right)^{1/2}\left(\sum_{k=1}^{\infty}\lambda_k^4|(\phi,v_k)|^2\right)^{1/2}=Ct^{2\alpha_1}\|a\|\|A^2\phi\|.

Similarly,

∣I2(t)∣≤C∑j=2nt2α1−αj(∑k=1∞∣ak∣2)1/2(∑k=1∞λk2∣(ϕ,vk)∣2)1/2=C∑j=2nt2α1−αj∥a∥∥Aϕ∥.|I_2(t)|\le C\sum_{j=2}^{n}t^{2\alpha_1-\alpha_j}\left(\sum_{k=1}^{\infty}|a_k|^2\right)^{1/2}\left(\sum_{k=1}^{\infty}\lambda_k^2|(\phi,v_k)|^2\right)^{1/2}=C\sum_{j=2}^{n}t^{2\alpha_1-\alpha_j}\|a\|\|A\phi\|.

Therefore, min⁡2≤j≤n{2α1−αj}=2α1−α2\min_{2\le j\le n}\{2\alpha_1-\alpha_j\}=2\alpha_1-\alpha_2, we get

(u(t),ϕ)=(a,ϕ)−(a,Aϕ)tα1Γ(1+α1)+O(t2α1)+O(t2α1−α2),t→0.(u(t),\phi)=(a,\phi)-\frac{(a,A\phi)t^{\alpha_1}}{\Gamma(1+\alpha_1)}+O(t^{2\alpha_1})+O(t^{2\alpha_1-\alpha_2}),\qquad t\to0.

Given that a∈Ha\in H, ϕ∈D(A2)∩D(B2)\phi\in D(A^2)\cap D(B^2), similarly

(ω(t),ϕ)=(a,ϕ)−(a,Bϕ)tβ1Γ(1+β1)+O(t2β1)+O(t2β1−β2),t→0.(\omega(t),\phi)=(a,\phi)-\frac{(a,B\phi)t^{\beta_1}}{\Gamma(1+\beta_1)}+O(t^{2\beta_1})+O(t^{2\beta_1-\beta_2}),\qquad t\to0.

Since

(u(t),ϕ)=(ω(t),ϕ),(u(t),\phi)=(\omega(t),\phi),

we have,

−(a,Aϕ)tα1Γ(1+α1)+o(tα1)=−(a,Bϕ)tβ1Γ(1+β1)+o(tβ1),t→0.(6)-\frac{(a,A\phi)t^{\alpha_1}}{\Gamma(1+\alpha_1)}+o(t^{\alpha_1})=-\frac{(a,B\phi)t^{\beta_1}}{\Gamma(1+\beta_1)}+o(t^{\beta_1}),\quad t\to0. \tag*{(6)}

We assume that α1<β1\alpha_1<\beta_1 (or α1>β1\alpha_1>\beta_1). Dividing both sides of (6) by tα1t^{\alpha_1} (or tβ1t^{\beta_1}) and passing to the limit as t→0t\to0, we arrive at

(a,Aϕ)Γ(1+α1)=0(or(a,Bϕ)Γ(1+β1)=0).\frac{(a,A\phi)}{\Gamma(1+\alpha_1)}=0\quad\left(\text{or}\quad\frac{(a,B\phi)}{\Gamma(1+\beta_1)}=0\right).

Using the fact that (a,Aϕ)≠0(a,A\phi)\ne0 (or (a,Bϕ)≠0(a,B\phi)\ne0), this is impossible. Hence, the assumption α1<β1\alpha_1<\beta_1 (α1>β1\alpha_1>\beta_1) is false, and consequently

α1=β1.\alpha_1=\beta_1.

It follows from (6) and the equality α1=β1\alpha_1=\beta_1 that (a,Aϕ)=(a,Bϕ)(a,A\phi)=(a,B\phi). In this way, by subtracting the first and second term from both sides of (4),(5) and repeating the above steps and using Lemma 2.1, we rewrite Eα′,1+2α1−α2(−λktα1,∗)E_{\alpha^\prime,1+2\alpha_1-\alpha_2}(-\lambda_k t^{\alpha_1},*) in the following form:

Eα′,1+2α1−α2(−λktα1,∗)=1Γ(1+2α1−α2)−λktα1Eα′,1+3α1−α2(−λktα1,∗)−∑j=2nqjtα1−αjEα′,1+3α1−α2−αj(−λktα1,∗),\begin{aligned} E_{\alpha^\prime,1+2\alpha_1-\alpha_2}(-\lambda_k t^{\alpha_1},*)={}&\frac{1}{\Gamma(1+2\alpha_1-\alpha_2)}-\lambda_k t^{\alpha_1}E_{\alpha^\prime,1+3\alpha_1-\alpha_2}(-\lambda_k t^{\alpha_1},*)\\ &-\sum_{j=2}^{n}q_jt^{\alpha_1-\alpha_j}E_{\alpha^\prime,1+3\alpha_1-\alpha_2-\alpha_j}(-\lambda_k t^{\alpha_1},*), \end{aligned}

and t→0t\to0, we obtain

q2(a,Aϕ)t2α1−α2Γ(1+2α1−α2)+O(t2α1)+O(t3α1−α2)+O(t3α1−2α2)+O(t2α1−αj)=μ2(a,Bϕ)t2β1−β2Γ(1+2β1−β2)+O(t2β1)+O(t3β1−β2)+O(t3β1−2β2)+O(t2β1−βj),(7)\begin{aligned} \frac{q_2(a,A\phi)t^{2\alpha_1-\alpha_2}}{\Gamma(1+2\alpha_1-\alpha_2)} +O(t^{2\alpha_1})+O(t^{3\alpha_1-\alpha_2})+O(t^{3\alpha_1-2\alpha_2})+O(t^{2\alpha_1-\alpha_j})\\ =\frac{\mu_2(a,B\phi)t^{2\beta_1-\beta_2}}{\Gamma(1+2\beta_1-\beta_2)} +O(t^{2\beta_1})+O(t^{3\beta_1-\beta_2})+O(t^{3\beta_1-2\beta_2})+O(t^{2\beta_1-\beta_j}), \tag*{(7)} \end{aligned}

(where j=3,…,nj=3,\ldots,n), or

q2(a,Aϕ)t2α1−α2Γ(1+2α1−α2)+o(t2α1−α2)=μ2(a,Bϕ)t2β1−β2Γ(1+2β1−β2)+o(t2β1−β2).\frac{q_2(a,A\phi)t^{2\alpha_1-\alpha_2}}{\Gamma(1+2\alpha_1-\alpha_2)}+o(t^{2\alpha_1-\alpha_2})=\frac{\mu_2(a,B\phi)t^{2\beta_1-\beta_2}}{\Gamma(1+2\beta_1-\beta_2)}+o(t^{2\beta_1-\beta_2}).

Having established that α1=β1\alpha_1=\beta_1 and (a,Aϕ)=(a,Bϕ)(a,A\phi)=(a,B\phi), we repeat the above procedure. By the same argument, we get

α2=β2andq2=μ2.\alpha_2=\beta_2\quad\text{and}\quad q_2=\mu_2.

To prove that α3=β3\alpha_3=\beta_3, we subtract from (4) and (5) the first two terms together with

q2(a,Aϕ)t2α1−α2Γ(1+2α1−α2),μ2(a,Bϕ)t2β1−β2Γ(1+2β1−β2).(8)\frac{q_2(a,A\phi)t^{2\alpha_1-\alpha_2}}{\Gamma(1+2\alpha_1-\alpha_2)},\qquad \frac{\mu_2(a,B\phi)t^{2\beta_1-\beta_2}}{\Gamma(1+2\beta_1-\beta_2)}. \tag*{(8)}

Because the first and second terms, also (8) terms (α1=β1\alpha_1=\beta_1, α2=β2\alpha_2=\beta_2, q2=μ2q_2=\mu_2, (a,Aϕ)=(a,Bϕ)(a,A\phi)=(a,B\phi)) are equal. Then we obtain

q3(a,Aϕ)t2α1−α3Γ(1+2α1−α3)+O(t2α1)+O(t3α1−α2)+O(t3α1−2α2)+O(t3α1−α3)+O(t3α1−α3−α2)+O(t2α1−αj)=μ3(a,Bϕ)t2β1−β3Γ(1+2β1−β3)+O(t2β1)+O(t3β1−β2)+O(t3β1−2β2)+O(t3β1−β3)+O(t3β1−β3−β2)+O(t2β1−βj).\begin{aligned} \frac{q_3(a,A\phi)t^{2\alpha_1-\alpha_3}}{\Gamma(1+2\alpha_1-\alpha_3)} +O(t^{2\alpha_1})+O(t^{3\alpha_1-\alpha_2})+O(t^{3\alpha_1-2\alpha_2})\\ +O(t^{3\alpha_1-\alpha_3})+O(t^{3\alpha_1-\alpha_3-\alpha_2})+O(t^{2\alpha_1-\alpha_j}) \\ &= \frac{\mu_3(a,B\phi)t^{2\beta_1-\beta_3}}{\Gamma(1+2\beta_1-\beta_3)}+O(t^{2\beta_1})+O(t^{3\beta_1-\beta_2})+O(t^{3\beta_1-2\beta_2}) \\ +O(t^{3\beta_1-\beta_3})+O(t^{3\beta_1-\beta_3-\beta_2})+O(t^{2\beta_1-\beta_j}). \end{aligned}

where j=4,…,nj=4,\ldots,n. In order to rewrite this equality in the form

q3(a,Aϕ)t2α1−α3Γ(1+2α1−α3)+o(t2α1−α3)=μ3(a,Bϕ)t2β1−β3Γ(1+2β1−β3)+o(t2β1−β3),t→0,(9)\frac{q_3(a,A\phi)t^{2\alpha_1-\alpha_3}}{\Gamma(1+2\alpha_1-\alpha_3)}+o(t^{2\alpha_1-\alpha_3})=\frac{\mu_3(a,B\phi)t^{2\beta_1-\beta_3}}{\Gamma(1+2\beta_1-\beta_3)}+o(t^{2\beta_1-\beta_3}),\quad t\to0, \tag*{(9)}

it is required that

3α1−2α2>2α1−α3,3\alpha_1-2\alpha_2>2\alpha_1-\alpha_3,
3β1−2β2>2β1−β3,3\beta_1-2\beta_2>2\beta_1-\beta_3,

which is not always satisfied. However, by applying Lemma 2.1 ss and pp times to the multinomial Mittag–Leffler functions Eα′,1+2α1−α2(−λktα1,∗)E_{\alpha',1+2\alpha_1-\alpha_2}(-\lambda_k t^{\alpha_1},*), Eβ′,1+2β1−β2(−λktβ1,∗)E_{\beta',1+2\beta_1-\beta_2}(-\lambda_k t^{\beta_1},*) appearing in (4) and (5) there exists s,p∈Ns,p\in\mathbb{N} (here and henceforth let N:={1,2,3,…}\mathbb{N}:=\{1,2,3,\ldots\}) such that

2α1−α2+s(α1−α2)>2α1−α3,2\alpha_1-\alpha_2+s(\alpha_1-\alpha_2)>2\alpha_1-\alpha_3,
2β1−β2+p(β1−β2)>2β1−β3,2\beta_1-\beta_2+p(\beta_1-\beta_2)>2\beta_1-\beta_3,

or, more explicitly,

s(α1−α2)>α2−α3,s(\alpha_1-\alpha_2)>\alpha_2-\alpha_3,
p(β1−β2)>β2−β3,p(\beta_1-\beta_2)>\beta_2-\beta_3,

where s=min⁡{δ∈N:δ(α1−α2)>α2−α3}s=\min\{\delta\in\mathbb{N}:\delta(\alpha_1-\alpha_2)>\alpha_2-\alpha_3\}, p=min⁡{δ∈N:δ(β1−β2)>β2−β3}p=\min\{\delta\in\mathbb{N}:\delta(\beta_1-\beta_2)>\beta_2-\beta_3\}. If p>sp>s, for all terms up to ss, we obtain the following equality when t→0t\to0:

(a,Aϕ)(−1)iq2i+1t2α1−α2+i(α1−α2)Γ(1+2α1−α2+i(α1−α2))+o(t2α1−α2+i(α1−α2))=(a,Bϕ)(−1)iμ2i+1t2β1−β2+i(β1−β2)Γ(1+2β1−β2+i(β1−β2))+o(t2β1−β2+i(β1−β2)),(10)\begin{aligned} (a,A\phi)\frac{(-1)^i q_2^{i+1}t^{2\alpha_1-\alpha_2+i(\alpha_1-\alpha_2)}}{\Gamma(1+2\alpha_1-\alpha_2+i(\alpha_1-\alpha_2))}+o(t^{2\alpha_1-\alpha_2+i(\alpha_1-\alpha_2)}) \\ =(a,B\phi)\frac{(-1)^i\mu_2^{i+1}t^{2\beta_1-\beta_2+i(\beta_1-\beta_2)}}{\Gamma(1+2\beta_1-\beta_2+i(\beta_1-\beta_2))}+o(t^{2\beta_1-\beta_2+i(\beta_1-\beta_2)}), \tag*{(10)} \end{aligned}

where i=1,2,…,s−1i=1,2,\ldots,s-1. As shown above, we already know that α1=β1\alpha_1=\beta_1, α2=β2\alpha_2=\beta_2, q2=μ2q_2=\mu_2, (a,Aϕ)=(a,Bϕ)(a,A\phi)=(a,B\phi).

Considering the terms from ss to p−1p-1, we obtain the following equality:

(a,Aϕ)q3t2α1−α3Γ(1+2α1−α3)+o(t2α1−α3)=(a,Bϕ)∑i=sp−1(−1)iμ2i+1t2β1−β2+i(β1−β2)Γ(1+2β1−β2+i(β1−β2))+o(t2β1−β2+i(β1−β2)),(11)\begin{aligned} (a,A\phi)\frac{q_3t^{2\alpha_1-\alpha_3}}{\Gamma(1+2\alpha_1-\alpha_3)}+o(t^{2\alpha_1-\alpha_3}) \\ =(a,B\phi)\sum_{i=s}^{p-1}\frac{(-1)^i\mu_2^{i+1}t^{2\beta_1-\beta_2+i(\beta_1-\beta_2)}}{\Gamma(1+2\beta_1-\beta_2+i(\beta_1-\beta_2))}+o(t^{2\beta_1-\beta_2+i(\beta_1-\beta_2)}), \tag*{(11)} \end{aligned}

In view of the fact that α1=β1\alpha_1=\beta_1, α2=β2\alpha_2=\beta_2, and p>sp>s, we have

2β1−β2+i(β1−β2)>2α1−α3,i=s,s+1,…,p−1.2\beta_1-\beta_2+i(\beta_1-\beta_2)>2\alpha_1-\alpha_3,\qquad i=s,s+1,\ldots,p-1.

Therefore, dividing both sides of the (11) by t2α1−α3t^{2\alpha_1-\alpha_3} and letting t→0t\to0, we obtain

q3(a,Aϕ)Γ(1+2α1−α3)=0.\frac{q_3(a,A\phi)}{\Gamma(1+2\alpha_1-\alpha_3)}=0.

Since q3≠0q_3 \ne0 and (a,Aϕ)≠0(a,A\phi) \ne0, the above equality leads to a contradiction. Therefore, the case p>sp>s is impossible. We next consider the case s>ps>p. For all terms up to p−1p-1, we get (10) equality. From pp to s−1s-1, we obtain following equality:

(a,Aϕ)∑i=ps−1(−1)iq1i+1t2α1−α2+i(α1−α2)Γ(1+2α1−α2+i(α1−α2))+o(t2α1−α2+i(α1−α2))=(a,Bϕ)μ3t2β1−β3Γ(1+2β1−β3)+o(t2β1−β3),(12)\begin{aligned} (a,A\phi)\sum_{i=p}^{s-1}\frac{(-1)^i q_1^{i+1}t^{2\alpha_1-\alpha_2+i(\alpha_1-\alpha_2)}}{\Gamma(1+2\alpha_1-\alpha_2+i(\alpha_1-\alpha_2))}+o\left(t^{2\alpha_1-\alpha_2+i(\alpha_1-\alpha_2)}\right) \\ =(a,B\phi)\frac{\mu_3t^{2\beta_1-\beta_3}}{\Gamma(1+2\beta_1-\beta_3)}+o\left(t^{2\beta_1-\beta_3}\right), \tag*{(12)} \end{aligned}

Taking into account that α1=β1\alpha_1=\beta_1, α2=β2\alpha_2=\beta_2, and s>ps>p, we have

2α1−α2+i(α1−α2)>2β1−β3,i=p,p+1,…,s−1.2\alpha_1-\alpha_2+i(\alpha_1-\alpha_2)>2\beta_1-\beta_3,\qquad i=p,p+1,\ldots,s-1.

Therefore, dividing both sides of the (12) by t2β1−β3t^{2\beta_1-\beta_3} and letting t→0t\to0, we obtain

μ3(a,Bϕ)Γ(1+2β1−β3)=0.\frac{\mu_3(a,B\phi)}{\Gamma(1+2\beta_1-\beta_3)}=0.

A contradiction is obtained in view of μ3≠0\mu_3\ne0 and (a,Bϕ)≠0(a,B\phi)\ne0. Therefore, s=ps=p. Due to the fact that s=ps=p, all the equalities in (10) hold up to ss (or pp). As shown above, we already know that

α1=β1,α2=β2,q2=μ2,(a,Aϕ)=(a,Bϕ).\alpha_1=\beta_1,\qquad\alpha_2=\beta_2,\qquad q_2=\mu_2,\qquad(a,A\phi)=(a,B\phi).

None of these terms contains α3\alpha_3 or β3\beta_3. Therefore, if

3α1−2α2>2α1−α3,3\alpha_1-2\alpha_2>2\alpha_1-\alpha_3,

then (9) is valid. On the other hand, if

3α1−2α2<2α1−α3,3\alpha_1-2\alpha_2<2\alpha_1-\alpha_3,

then there exists ss (or pp) ∈N\in\mathbb{N}, such that

2α1−α2+s(α1−α2)>2α1−α3,2\alpha_1-\alpha_2+s(\alpha_1-\alpha_2)>2\alpha_1-\alpha_3,

we may conclude that (9) is valid. By employing the same method as used previously, we derive a contradiction, thereby establishing the result that α3=β3\alpha_3=\beta_3 and q3=μ3q_3=\mu_3.

It may happen that there exists an integer l∈Nl\in\mathbb{N} such that l=s−1=p−1l=s-1=p-1, i.e., the (s−1)(s-1)-th (or equivalently (p−1)(p-1)-th) iteration, for which

2α1−α2+l(α1−α2)=2α1−α3,2β1−β2+l(β1−β2)=2β1−β3,(13)\begin{aligned} 2\alpha_1-\alpha_2+l(\alpha_1-\alpha_2)&=2\alpha_1-\alpha_3,\\ 2\beta_1-\beta_2+l(\beta_1-\beta_2)&=2\beta_1-\beta_3, \tag*{(13)} \end{aligned}

or, more explicitly,

l(α1−α2)=α2−α3,l(β1−β2)=β2−β3\begin{aligned} l(\alpha_1-\alpha_2)&=\alpha_2-\alpha_3,\\ l(\beta_1-\beta_2)&=\beta_2-\beta_3 \end{aligned}

then we obtain the following equality when t→0t \to0:

\begin{aligned} (a,A\phi)t^{2\alpha_{1}-\alpha_{2}+l(\alpha_{1}-\alpha_{2})}\left(&\frac{q_{3}}{\Gamma(1+2\alpha_{1}-\alpha_{3})}+\frac{(-1)^{l}q_{2}^{l+1}}{\Gamma(1+2\alpha_{1}-\alpha_{2}+l(\alpha_{1}-\alpha_{2}))}\right.\\ &\left.+o\left(t^{2\alpha_{1}-\alpha_{2}+l(\alpha_{1}-\alpha_{2})}\right)\right)\\ ={}&(a,B\phi)t^{2\beta_{1}-\beta_{2}+l(\beta_{1}-\beta_{2})}\left(\frac{\mu_{3}}{\Gamma(1+2\beta_{1}-\beta_{3})}+\frac{(-1)^{l}\mu_{2}^{l+1}}{\Gamma(1+2\beta_{1}-\beta_{2}+l(\beta_{1}-\beta_{2}))}\right.\\ &\left.+o\left(t^{2\beta_{1}-\beta_{2}+l(\beta_{1}-\beta_{2})}\right)\right). \tag*{(14)} \end{aligned}

Due to the fact that α1=β1\alpha_{1}=\beta_{1} and α2=β2\alpha_{2}=\beta_{2}, it follows from (13) that

α3=α2−l(α1−α2)=β2−l(β1−β2)=β3.\alpha_{3}=\alpha_{2}-l(\alpha_{1}-\alpha_{2})=\beta_{2}-l(\beta_{1}-\beta_{2})=\beta_{3}.

Moreover, taking into account that α1=β1\alpha_{1}=\beta_{1}, α2=β2\alpha_{2}=\beta_{2}, α3=β3\alpha_{3}=\beta_{3} and (a,Aϕ)=(a,Bϕ)(a,A\phi)=(a,B\phi) we divide both sides of (14) by t2α1−α2+l(α1−α2)t^{2\alpha_{1}-\alpha_{2}+l(\alpha_{1}-\alpha_{2})} (or t2β1−β2+l(β1−β2)t^{2\beta_{1}-\beta_{2}+l(\beta_{1}-\beta_{2})}) and letting t→0t\to0, we obtain

q3−μ3Γ(1+2α1−α3)+(−1)l(q2l+1−μ2l+1)Γ(1+2α1−α2+l(α1−α2))=0.(15)\frac{q_{3}-\mu_{3}}{\Gamma(1+2\alpha_{1}-\alpha_{3})}+\frac{(-1)^{l}(q_{2}^{l+1}-\mu_{2}^{l+1})}{\Gamma(1+2\alpha_{1}-\alpha_{2}+l(\alpha_{1}-\alpha_{2}))}=0. \tag*{(15)}

We already know that q2=μ2q_{2}=\mu_{2}, it follows from (15) that q3=μ3q_{3}=\mu_{3}.

To prove that α4=β4\alpha_{4}=\beta_{4}, we subtract from (4) and (5) the first two terms and (8) together with

q3(a,Aϕ)t2α1−α3Γ(1+2α1−α3),μ3(a,Bϕ)t2β1−β3Γ(1+2β1−β3).(16)\frac{q_{3}(a,A\phi)t^{2\alpha_{1}-\alpha_{3}}}{\Gamma(1+2\alpha_{1}-\alpha_{3})},\qquad\frac{\mu_{3}(a,B\phi)t^{2\beta_{1}-\beta_{3}}}{\Gamma(1+2\beta_{1}-\beta_{3})}. \tag*{(16)}

Then we obtain

q4(a,Aϕ)t2α1−α4Γ(1+2α1−α4)+O(t2α1)+O(t3α1−α2)+O(t3α1−2α2)+O(t2α1−αj)+O(t3α1−α3)+O(t3α1−α3−α2)+O(t3α1−α4)+O(t3α1−α2−α4)=μ4(a,Bϕ)t2β1−β4Γ(1+2β1−β4)+O(t2β1)+O(t3β1−β2)+O(t3β1−2β2)+O(t2β1−βj)+O(t3β1−β3)+O(t3β1−β3−β2)+O(t3β1−β4)+O(t3β1−β2−β4).\begin{aligned} \frac{q_{4}(a,A\phi)t^{2\alpha_{1}-\alpha_{4}}}{\Gamma(1+2\alpha_{1}-\alpha_{4})} &+O(t^{2\alpha_{1}})+O(t^{3\alpha_{1}-\alpha_{2}})+O(t^{3\alpha_{1}-2\alpha_{2}})+O(t^{2\alpha_{1}-\alpha_{j}})\\ &+O(t^{3\alpha_{1}-\alpha_{3}})+O(t^{3\alpha_{1}-\alpha_{3}-\alpha_{2}})+O(t^{3\alpha_{1}-\alpha_{4}})+O(t^{3\alpha_{1}-\alpha_{2}-\alpha_{4}})\\ ={}&\frac{\mu_{4}(a,B\phi)t^{2\beta_{1}-\beta_{4}}}{\Gamma(1+2\beta_{1}-\beta_{4})}+O(t^{2\beta_{1}})+O(t^{3\beta_{1}-\beta_{2}})+O(t^{3\beta_{1}-2\beta_{2}})+O(t^{2\beta_{1}-\beta_{j}})\\ &+O(t^{3\beta_{1}-\beta_{3}})+O(t^{3\beta_{1}-\beta_{3}-\beta_{2}})+O(t^{3\beta_{1}-\beta_{4}})+O(t^{3\beta_{1}-\beta_{2}-\beta_{4}}). \end{aligned}

where j=5,…,nj=5,\ldots,n. In order to rewrite this equality in the form

q4(a,Aϕ)t2α1−α4Γ(1+2α1−α4)+o(t2α1−α4)=μ4(a,Bϕ)t2β1−β4Γ(1+2β1−β4)+o(t2β1−β4),t→0,(17)\frac{q_{4}(a,A\phi)t^{2\alpha_{1}-\alpha_{4}}}{\Gamma(1+2\alpha_{1}-\alpha_{4})}+o(t^{2\alpha_{1}-\alpha_{4}})=\frac{\mu_{4}(a,B\phi)t^{2\beta_{1}-\beta_{4}}}{\Gamma(1+2\beta_{1}-\beta_{4})}+o(t^{2\beta_{1}-\beta_{4}}),\qquad t\to0, \tag*{(17)}

it is required that

3α1−α2−α3>3α1−2α2>2α1−α4,3\alpha_{1}-\alpha_{2}-\alpha_{3}>3\alpha_{1}-2\alpha_{2}>2\alpha_{1}-\alpha_{4},

which is not always satisfied. However, similarly to the above case, there exists m∈Nm\in\mathbb{N} such that

2α1−α2+m(α1−α2)>2α1−α4.2\alpha_{1}-\alpha_{2}+m(\alpha_{1}-\alpha_{2})>2\alpha_{1}-\alpha_{4}.

Nevertheless, when repeating the procedure for mm times, there may exists r∈Nr\in\mathbb{N} such that

2α1−α2+r(α1−α2)>3α1−α3−α2.2\alpha_{1}-\alpha_{2}+r(\alpha_{1}-\alpha_{2})>3\alpha_{1}-\alpha_{3}-\alpha_{2}.

In this case, by applying Lemma 2.1 to Eα′,1+2α1−α3(−λktα1,∗)E_{\alpha',1+2\alpha_1-\alpha_3}(-\lambda_k t^{\alpha_1},*) exactly hh times, we can find an integer hh satisfying

2α1−α3+h(α1−α2)>2α1−α4.2\alpha_1-\alpha_3+h(\alpha_1-\alpha_2)>2\alpha_1-\alpha_4.

Throughout the construction of mm and hh no terms involving α4\alpha_4 or β4\beta_4 arise, similarly to the situation in (10). So we may conclude that (17) is valid. By employing the same method as used previously, we derive a contradiction, thereby establishing the result that α4=β4\alpha_4=\beta_4, q4=μ4q_4=\mu_4.

The subsequent steps follow exactly the same procedure as that used in the proof of α5=β5\alpha_5=\beta_5 and q5=μ5q_5=\mu_5. Repeating the above argument recursively, we conclude that

αj=βj,qj=μj,j=1,…,n.\alpha_j=\beta_j,\qquad q_j=\mu_j,\qquad j=1,\ldots,n.

Thus the proof of Theorem 3.1 is complete. □\square

Remark 3.2. By reviewing the proof, one can readily verify that the same arguments remain valid for multi-term time-fractional wave equations. More precisely, the results continue to hold for equations (1) and (3) with

1<αn<αn−1<⋯<α1<2,1<βn<βn−1<⋯<β1<2,1<\alpha_n<\alpha_{n-1}<\cdots<\alpha_1<2,\qquad1<\beta_n<\beta_{n-1}<\cdots<\beta_1<2,

supplemented with the initial conditions

u(0)=a,∂tu(0)=0,w(0)=a,∂tω(0)=0.u(0)=a,\qquad\partial_tu(0)=0,\qquad w(0)=a,\qquad\partial_t\omega(0)=0.

Conclusion

In this paper, we investigated an inverse problem for identifying all fractional orders and their corresponding coefficients in a multi-term time-fractional diffusion equation with Caputo derivatives. The main result establishes the uniqueness of recovering all fractional parameters from a single scalar observation of the solution, namely (u(t),ϕ)(u(t),\phi), in an abstract Hilbert space framework. The proof relies on the spectral representation of the solution together with a careful asymptotic analysis of the multinomial Mittag–Leffler function. A key difficulty stems from the interaction of several asymptotic terms generated by different fractional orders. This difficulty is overcome by repeatedly applying an identity for the multinomial Mittag–Leffler function and developing a recursive asymptotic comparison argument, which makes it possible to separate the dominant contributions and identify the unknown parameters successively. Compared with the existing literature, the proposed approach significantly weakens the assumptions required for uniqueness. In particular, the initial value is required only to belong to the underlying Hilbert space, without any sign or additional regularity assumptions. The observation function, however, must satisfy suitable regularity and nondegeneracy conditions. These features substantially broaden the applicability of the obtained uniqueness theorem. Finally, we can see that the same recursive argument extends naturally to multi-term time-fractional wave equations with fractional orders in (1,2)(1,2).

Acknowledgments

The authors are grateful to professors M. Yamamoto and Z. Sobirov for discussions of these results.

Ravshan Ashurov: V. I. Romanovskiy Institute of Mathematics Uzbekistan Academy of Sciences, University Street 9, Tashkent–100174, Uzbekistan; Engineering School, Central Asian University, 111221, Tashkent, Uzbekistan.

Email address: [email protected]

Damir Shamuratov: V. I. Romanovskiy Institute of Mathematics Uzbekistan Academy of Sciences, University Street 9, Tashkent–100174, Uzbekistan

Email address: [email protected]

References

  1. [1]A.A. Kilbas, H.M. Srivastava, J.J. Trujillo, Theory and Applications of Fractional Differential Equations. Elsevier, Amsterdam (2006).
  2. [2]S. Alimov, R. Ashurov, Inverse problem of determining an order of the Caputo time-fractional derivative for a subdiffusion equation, J. Inverse Ill-Posed Probl. 28(5), 651–658 (2020).DOI
  3. [3]S. Alimov, R. Ashurov, Inverse problem of determining an order of the Riemann–Liouville time-fractional derivative, Progr. Fract. Differ. Appl. 8(4), 1–8 (2022).DOI
  4. [4]R. Ashurov, S. Umarov, Determination of the order of fractional derivative for subdiffusion equations, Fract. Calc. Appl. Anal. 23(6), 1647–1662 (2020).DOI
  5. [5]R. Ashurov, S. Umarov, An inverse problem of determining orders of systems of fractional pseudo-differential equations, Fract. Calc. Appl. Anal. 25(1), 109–127 (2022).arxiv.org/abs/2107.08830
  6. [6]Z. Li, Y. Liu, M. Yamamoto, Inverse problems of determining parameters of the fractional partial differential equations, In: Handbook of Fractional Calculus with Applications, Vol. 2, De Gruyter, Berlin, pp. 431–442 (2019).DOI
  7. [7]X. Zheng, J. Cheng, H. Wang, Uniqueness of determining the variable fractional order in variable-order time-fractional diffusion equations, Inverse Problems 35, 1–11 (2019).DOI
  8. [8]S. Tatar, S. Ulusoy, A uniqueness result for an inverse problem in a space-time fractional diffusion equation, Electron. J. Differ. Equ. 257, 1–9 (2013).DOI
  9. [9]A.V. Pskhu, Order determination inverse problems for fractional diffusion equations with Dzhbashyan–Nersesyan derivative, J. Math. Sci. 300(1), 55–65 (2026).DOI
  10. [10]Z. Li, M. Yamamoto, Uniqueness for inverse problems of determining orders of multi-term time fractional derivatives of diffusion equation, Appl. Anal. 94, 570–579 (2015).DOI
  11. [11]X. Jiang, Z. Li, Unique inversion of orders and potential for multi-term time fractional wave equations, Appl. Math. Lett. 160, Article 109319 (2025).DOI
  12. [12]E. Bazhlekova, Application of subordination principle to coefficient inverse problem for multi-term time-fractional wave equation, Fract. Calc. Appl. Anal. 1–15 (2024).
  13. [13]R. Ashurov, M. Yamamoto, Uniqueness for an inverse problem of determining order and temporal factor of the source for time-fractional evolution equations. https://arxiv.org/pdf/2606.31113
  14. [14]Z. Li, Y. Liu, M. Yamamoto, Initial-boundary value problems for multi-term time-fractional diffusion equations with positive constant coefficients, Appl. Math. Comput. 257, 381–397 (2015).DOI
  15. [15]Y. Luchko, Operational method in fractional calculus, Fract. Calc. Appl. Anal. 2(4), 463–488 (1999).
  16. [16]C.S. Sin, J.U. Rim, H.S. Choe, Initial-boundary value problems for multi-term time-fractional wave equations, Fract. Calc. Appl. Anal. 25(5), 1994–2019 (2022).DOI
  17. [17]F. Maes, K. Van Bockstal, Existence and uniqueness of a weak solution to fractional single-phase-lag heat equation, Fract. Calc. Appl. Anal. 26(4), 1663–1690 (2023).DOI

Paper details

vlnc:W_vzktunjt
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents