Simultaneous Numerical Determination of Two Time-dependent Coefficients in Second Order Parabolic Equation With Nonlocal Initial and Boundary Conditions

Mohammed A.J.Al-Shatrah; Mohammed Sabah Hussein

doi:10.12688/f1000research.173252.1

Home Browse Simultaneous Numerical Determination of Two Time-dependent Coefficients...

ALL Metrics

-

Views

-

Downloads

Get PDF

Get XML

Export

▬

✚

Research Article

Simultaneous Numerical Determination of Two Time-dependent Coefficients in Second Order Parabolic Equation With Nonlocal Initial and Boundary Conditions

[version 1; peer review: awaiting peer review]

Mohammed A.J.Al-Shatrah¹, Mohammed Sabah Hussein ²

PUBLISHED 10 Feb 2026

Author details Author details

¹ Ministry of Education, Directorate of Education Thi Qar, Thi Qar, 64001, Iraq
² University of Baghdad Al-Jaderyia Campus College of Science, Baghdad, Baghdad Governorate, Iraq

Mohammed A.J.Al-Shatrah
Roles: Data Curation, Formal Analysis, Investigation, Methodology, Software, Visualization, Writing – Original Draft Preparation

Mohammed Sabah Hussein
Roles: Conceptualization, Formal Analysis, Resources, Software, Supervision, Validation, Writing – Review & Editing

OPEN PEER REVIEW

REVIEWER STATUS AWAITING PEER REVIEW

This article is included in the Fallujah Multidisciplinary Science and Innovation gateway.

Abstract

Background

This study establishes a mathematically consistent and computational framework for the simultaneous identification of two time-dependent coefficients in a one-dimensional second-order parabolic partial differential equation. The considered problem is governed by nonlocal initial, boundary, and integral overdetermination conditions.

Methods

The direct problem is solved using the Crank-Nicolson finite difference method (FDM), which ensures unconditional stability and second-order accuracy in both spatial and temporal discretizations. The corresponding inverse problem is reformulated as a nonlinear regularized least-squares optimization problem and efficiently solved used the MATLAB subroutine lsqnonlin from the optimization Toolbox. Due to the intrinsic, ill-posedness of the inverse formulation, small input data errors lead to big output errors. Then, Tikhonov regularization, is employed to enhance numerical stability and robustness.

Results

Extensive numerical experiments are carried out under exact and noisy data to evaluate the numerical accuracy and convergence behavior of the method. The results confirm that the regularization technique effectively damps numerical oscillations, minimizes reconstruction error, and ensures reliable recovery of the unknown coefficients. Sensitivity analysis further reveals the essential role of the regularization parameter in controlling the trade-off between stability and accuracy.

Conclusions

The proposed approach provides an accurate and computationally efficient tool for IP in heat transfer, diffusion processes, and related applied sciences.

Keywords

Finite difference method, Crank-Nicolson, Tikhonov regularization, Inverse problem, Coefficient problem, Ill-posed problem, Nonlocal conditions, Parabolic equation.

Corresponding author: Mohammed Sabah Hussein

Competing interests: No competing interests were disclosed.

Grant information: The author(s) declared that no grants were involved in supporting this work.

Copyright: © 2026 A.J.Al-Shatrah M and Sabah Hussein M. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

How to cite: A.J.Al-Shatrah M and Sabah Hussein M. Simultaneous Numerical Determination of Two Time-dependent Coefficients in Second Order Parabolic Equation With Nonlocal Initial and Boundary Conditions [version 1; peer review: awaiting peer review]. F1000Research 2026, 15:228 (https://doi.org/10.12688/f1000research.173252.1) First published: 10 Feb 2026, 15:228 (https://doi.org/10.12688/f1000research.173252.1) Latest published: 10 Feb 2026, 15:228 (https://doi.org/10.12688/f1000research.173252.1)

1. Introduction

Inverse problems (IPs), arise naturally in many scientific and engineering disciplines as they are typically modeled by differential equations. In the context of ordinary differential equations, the direct problem refers to obtaining solutions for a given system, whereas IP focuses on reconstructing the governing system from observed characteristics. Historically, the role of IPs was recognized in celestial mechanics.¹ In practical applications, IPs are commonly encountered when one seeks to determine unknown causes from their observed outcomes, in contrast to direct problems where effects are predicted from known causes. Compared with direct problems, IPs, are often more challenging to solve due to their ill-posedness, solutions may fail to exist, may not be unique, or may not change continuously with the input data.² Such problems appear in almost every scientific and technological domain, particularly in models derived from social and physical systems. Since most of these models are expressed through differential and integral equations, their analysis requires not only solving the equations but also interpreting the system’s behavior under varying conditions, provided that sufficient information is available to fully describe the model. IPs, associated with such equations arise in a wide range of scientific and engineering applications, as well as in the modeling of social processes. Many physical phenomena are described by mathematical models in the form of initial and boundary value problems for partial differential equations (PDEs). Frequently, these models involve differential and integral equations.³ In boundary type IPs, the goal is to recover unknown boundary conditions, which usually leads to classical ill-posedness because the absence of continuous dependence of the solution on the input data. Numerical solution (NS) of direct mathematical physics problems is presently a well-studied matter. In solving multi-dimensional boundary value problems, difference methods and the finite element method are widely used. Partial differential equations give a basis for applied mathematical models, and their solution can be found in mathematical physics equations and some relations, including additional relations, boundary, and initial conditions, among other elements.⁴ The following reconstruction problems serve as examples of IP applications in daily life, tomography, initial condition estimation of transient problems in conductive heat transfer, detection of non-metallic materials beneath the surface using reflected radiation methods, intensity, position estimation of illuminated radiation from a biological source using experimental radiation measures, and thermal source intensity estimation with functional dependence in space.^5–9 On the other side, there are three different kinds of differential equations, Second-order equations are the most crucial for applications, Elliptic, Parabolic, and hyperbolic equations are examples of these equations. Yuldashev and other authors studied elliptic type integrate differential equations in^10,11 while hyperbolic and parabolic were studied in^12–14 and^15–17 respectively. The alternating direction explicit method used for reconstructing solutions is also efficient and unconditionally stable, in¹⁸ is restructured into a nonlinear regularized least-square optimization problem, and is effectively resolved using the MATLAB subroutine lsqnonlin from the optimization toolbox. The Crank-Nicolson (C-N) FDM together with the TR was used in^19–21 the effectiveness of the computational method was shown together with the proof of the solution’s existence and uniqueness. To find a stable and accurate approximate solution of finite differences. In^22–26 the study shows that for the IP with the time-dependent source coefficient in heat equations, there’s a smooth solution pair as extended as a single data point at an observation point and by providing an explicit formula for the time-dependent source coefficient, additional information at the observation point improves our comprehension of the behavior of the problem. This paper’s IP classical solution for a second order parabolic equation has been shown to be exist and unique by Elvin Azizbayov.²⁷ As a result, the main objective of the current work is a numerical realization of such a problem. In this paper, the article is divided into five sections, Section 2, includes a mathematical formulation of the IP under study. Section 3, a numerical method for solving the forward problem using C-N FDM. In Section 4, the IPs mathematical solution is shown with an initial guess, while, in Section 5, discuss the numerical problem and explain the obtained results.

2. Mathematical formulation

As a mathematical model, we consider $D_{T}$ be a rectangular region that is defined by $D_{T} : {(v, τ) : 0 \leq v \leq 1, 0 \leq τ \leq T}$ and $(T > 0),$ be a fixed number, the one-dimensional IP of determining of unknown functions ${w (v, τ), a (τ), b (τ)}$ for the upcoming parabolic equation,

(1)

f (v, τ) = c (τ) w_{τ} (v, τ) - w_{vv} (v, τ) - a (τ) w (v, τ) - b (τ) g (v, τ), (v, τ) \in D_{T},

with, the nonlocal initial conditions (ICs)

(2)

w (v, 0) + δw (v, T) + \int_{0}^{T} \hat{Ƥ} (τ) w (v, τ) dτ = ζ (v), v \in [0, 1],

with, periodic boundary condition,

(3)

w (0, τ) = γw (1, τ), τ \in [0, T],

and, nonlocal integral condition

(4)

\int_{0}^{1} w (v, τ) dv = 0, τ \in [0, T],

and the overdetermination conditions

(5)

w (v_{i}, τ) = z_{i} (τ) (i = 1, 2), τ \in [0, T],

where,

γ, δ \geq 0, v_{i} \in (0, 1), (i = 1, 2); v_{1} \neq v_{2}

, are fixed numbers,

c (τ) > 0, g (v, τ), f (v, τ), \hat{Ƥ} (τ) \geq 0, ζ (v), z_{i} (τ) (i = 1, 2)

are given functions,

w (v, τ), a (τ), b (τ)

are the sought functions (

v

is space component and

τ

is time component).

2.1 Definition 1

Consider the triplet ${w (v, τ), a (τ), b (τ)}$ represents a classical solution, to the IP $(1) - (5),$ if the functions $w (v, τ), a (τ),$ and $b (τ)$ satisfies the following conditions:

1) The function $w (v, τ)$ and its derivatives $w_{τ} (v, τ), w_{v} (v, τ), w_{vv} (v, τ)$ are continuous in the domain $D_{T} .$
2) The functions $a (τ),$ and $b (τ)$ are continuous on $[0, T] .$
3) Equation (1) and conditions $(2) - (5)$ are satisfied in the classical (usual) sense.

2.2 Existence of the inverse problem classical solution

2.2.1 Theorem 1[28]: Let the assumptions $(μ 1) - (μ 4)$ and the condition

((E (T) + 2) H (T) + G (T) + Q (T)) (E (T) + 2) < 1,

where,

μ 1) ζ (v) \in C^{2} [0, 1], ζ^{(3)} (v) \in L_{2} (0, 1), ζ (0) = γζ (1), ζ^{'} (0) = ζ^{'} (1), ζ^{''} (0) = {γζ}^{''} (1);

where,

\hat{a} = \frac{1 - γ}{1 + γ}

and

\hat{b} = \frac{γ}{1 + γ}

;

γ \neq \pm 1 .

μ 2) f (v, τ) \in C_{v, τ}^{2, 0} (D_{T}), f_{vvv} (v, τ) \in L_{2} (D_{T}), f (0, τ) = γf (1, τ), f_{v} (0, τ) = f_{v} (1, τ), f_{vv} (0, τ) = γ f_{vv} (1, τ), (γ \neq \pm 1), 0 \leq τ \leq T .

μ 3) g (v, τ) \in C_{v, τ}^{2, 0} (D_{T}), g_{vvv} (v, τ) \in L_{2} (D_{T}), g (0, τ) = γg (1, τ), g_{v} (0, τ) = g_{v} (1, τ), g_{vv} (0, τ) = γ g_{vv} (1, τ), (γ \neq \pm 1), 0 \leq τ \leq T .

μ 4) δ \geq 0, c (τ) \in C [0, T], c (τ) > 0, 0 \leq \hat{Ƥ} (τ) \in C [0, T], z_{i} (τ) \in C^{1} [0, T],

z (τ) = z_{1} (τ) g (v_{2}, τ) - z_{2} (τ) g (v_{1}, τ) \neq 0, (i = 1, 2), 0 \leq τ \leq T .

where,

(6)

E_{1} (T) = 2 {(1 + δ)}^{- 1} {‖ ζ (v) ‖}_{L_{2} (0, 1)} + 2 \sqrt{T} (1 + δ {(1 + δ)}^{- 1}) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} {‖ f (v, τ) ‖}_{L_{2} (D_{T})}),

(7)

E_{2} (T) = 2 \sqrt{10} {‖ ζ^{(3)} (v) ‖}_{L_{2} (0, 1)} + 2 \sqrt{10 T} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} {‖ f_{vvv} (v, τ) ‖}_{L_{2} (D_{T})},

(8)

\begin{matrix} E_{3} (T) = 4 \sqrt{3} {‖ ζ^{(3)} (v) (1 - \hat{b} - \hat{a} v) - 3 \hat{a} ζ^{(2)} (v) ‖}_{L_{2} (0, 1)} + 4 \sqrt{3 T} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} \\ {‖ f_{vvv} (v, τ) (1 - \hat{b} - \hat{a} v) - 3 \hat{a} f_{vv} (v, τ) ‖}_{L_{2} (D_{T})} + 12 \sqrt{2} | \hat{a} | (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} {‖ ζ^{(3)} (v) ‖}_{L_{2} (0, 1)} \\ + 12 \sqrt{3} | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]}^{2} \sqrt{{‖ c (τ) ‖}_{C [0, T]}} T \sqrt{T} {(1 + δ)}^{2} {‖ f_{vvv} (v, τ) ‖}_{L_{2} (D_{T})}, \end{matrix}

(9)

H_{1} (T) = (1 + δ {(1 + δ)}^{- 1}) T {‖ \frac{1}{c (τ)} ‖}_{C [0, T]},

(10)

H_{2} (T) = \sqrt{6} T (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]},

(11)

H_{3} (T) = \sqrt{6} T (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} + 6 \sqrt{2} | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]}^{2} \sqrt{{‖ c (τ) ‖}_{C [0, T]}} T^{2} {(1 + δ)}^{2},

(12)

G_{1} (T) = {(1 + δ)}^{- 1} T {‖ \hat{Ƥ} (τ) ‖}_{C [0, T]},

(13)

G_{2} (T) = \sqrt{6} T {‖ \hat{Ƥ} (τ) ‖}_{C [0, T]},

(14)

G_{3} (T) = \sqrt{3} T {‖ \hat{Ƥ} (τ) ‖}_{C [0, T]} [\sqrt{2} + 4 | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} (1 + δ)],

(15)

Q_{1} (T) = (1 + δ {(1 + δ)}^{- 1}) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} \sqrt{T} {‖ b (τ) ‖}_{C [0, T]} {‖ g (v, τ) ‖}_{L_{2} (D_{T})},

(16)

Q_{2} (T) = 4 \sqrt{3 T} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} {‖ g_{vvv} (v, τ) ‖}_{L_{2} (D_{T})},

(17)

\begin{matrix} Q_{3} (T) = 6 \sqrt{2 T} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} ({‖ g_{vvv} (v, τ) (1 - \hat{b} - \hat{a} v) - 3 \hat{a} g_{vv} (v, τ) ‖}_{L_{2} (D_{T})} \\ + | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} \sqrt{{‖ c (τ) ‖}_{C [0, T]}} T (1 + δ) {‖ g_{vvv} (v, τ) ‖}_{L_{2} (D_{T})}), \end{matrix}

where,

E_{4} (T) = E_{1} (T) + E_{2} (T) + E_{3} (T), H_{4} (T) = H_{1} (T) + H_{2} (T) + H_{3} (T),

G_{4} (T) = G_{1} (T) + G_{2} (T) + G_{3} (T), and, Q_{4} (T) = Q_{1} (T) + Q_{2} (T) + Q_{3} (T),

(18)

E_{5} (T) = {‖ {[z (τ)]}^{- 1} ‖}_{C [0, T]} {{‖ z_{1} (τ) (c (τ) {z_{2}}^{'} (τ) - f (v_{2}, τ)) - z_{2} (τ) (c (τ) {z_{1}}^{'} (τ) - f (v_{1}, τ)) ‖}_{C [0, T]} + {(\sum_{k = 1}^{\infty} {λ_{k}}^{- 2})}^{\frac{1}{2}} [2 \sqrt{2} [(1 + 4 \sqrt{3} | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} (1 + δ)) {‖ ζ^{(3)} (v) ‖}_{L_{2} (0, 1)} + {‖ ζ^{(3)} (v) (1 - \hat{b} - \hat{a} v) - 3 \hat{a} ζ^{(2)} (v) ‖}_{L_{2} (0, 1)}] + 2 \sqrt{2 T} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} [(1 + 6 \sqrt{2} | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} \sqrt{{‖ c (τ) ‖}_{C [0, T]}} T (1 + δ)) {‖ f_{vvv} (v, τ) ‖}_{L_{2} (D_{T})}) + {‖ f_{vvv} (v, τ) (1 - \hat{b} - \hat{a} v) - 3 \hat{a} f_{vv} (v, τ) ‖}_{L_{2} (D_{T})}]] {‖ | g (v_{2}, τ) | + | g (v_{1}, τ) | ‖}_{C [0, T]} {‖ \hat{a} v + \hat{b} ‖}_{C [0, 1]}},

(19)

E_{6} (T) = {‖ {[z (τ)]}^{- 1} ‖}_{C [0, T]} {{‖ g (v_{2}, τ) (c (τ) {z_{1}}^{'} (τ) - f (v_{1}, τ)) - g (v_{1}, τ) (c (τ) {z_{2}}^{'} (τ) - f (v_{2}, τ)) ‖}_{C [0, T]} + {(\sum_{k = 1}^{\infty} {λ_{k}}^{- 2})}^{\frac{1}{2}} [2 \sqrt{2} [(1 + 4 \sqrt{3} | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} (1 + δ)) {‖ ζ^{(3)} (v) ‖}_{L_{2} (0, 1)} + {‖ ζ^{(3)} (v) (1 - \hat{b} - \hat{a} v) - 3 \hat{a} ζ^{(2)} (v) ‖}_{L_{2} (0, 1)}] + 2 \sqrt{2 T} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} [(1 + 6 \sqrt{2} | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} \sqrt{{‖ c (τ) ‖}_{C [0, T]}} T (1 + δ)) {‖ f_{vvv} (v, τ) ‖}_{L_{2} (D_{T})}) + {‖ f_{vvv} (v, τ) (1 - \hat{b} - \hat{a} v) - 3 \hat{a} f_{vv} (v, τ) ‖}_{L_{2} (D_{T})}]] х {‖ | z_{1} (τ) | + | z_{2} (τ) | ‖}_{C [0, T]} {‖ \hat{a} v + \hat{b} ‖}_{C [0, 1]}},

(20)

H_{5} (T) = 2 T {‖ {[z (τ)]}^{- 1} ‖}_{C [0, T]} {(\sum_{k = 1}^{\infty} {λ_{k}}^{- 2})}^{\frac{1}{2}} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} (1 + 3 \sqrt{2} | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} \sqrt{{‖ c (τ) ‖}_{C [0, T]}} T (1 + δ)) {‖ | g (v_{2}, τ) | + | g (v_{1}, τ) | ‖}_{C [0, T]} {‖ \hat{a} v + \hat{b} ‖}_{C [0, 1]},

(21)

H_{6} (T) = 2 T {‖ {[z (τ)]}^{- 1} ‖}_{C [0, T]} {(\sum_{k = 1}^{\infty} {λ_{k}}^{- 2})}^{\frac{1}{2}} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} (1 + 3 \sqrt{2} | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} \sqrt{{‖ c (τ) ‖}_{C [0, T]}} T (1 + δ)) {‖ | z_{1} (τ) | + | z_{2} (τ) | ‖}_{C [0, T]} {‖ \hat{a} v + \hat{b} ‖}_{C [0, 1]},

(22)

G_{5} (T) = 2 T {‖ {[z (τ)]}^{- 1} ‖}_{C [0, T]} {(\sum_{k = 1}^{\infty} {λ_{k}}^{- 2})}^{\frac{1}{2}} {‖ \hat{Ƥ} (τ) ‖}_{C [0, T]} (1 + 2 \sqrt{3} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]}) {‖ | g (v_{2}, τ) | + | g (v_{1}, τ) | ‖}_{C [0, T]} {‖ \hat{a} v + \hat{b} ‖}_{C [0, 1]},

(23)

G_{6} (T) = 2 T {‖ {[z (τ)]}^{- 1} ‖}_{C [0, T]} {(\sum_{k = 1}^{\infty} {λ_{k}}^{- 2})}^{\frac{1}{2}} {‖ \hat{Ƥ} (τ) ‖}_{C [0, T]} (1 + 2 \sqrt{3} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]}) {‖ | z_{1} (τ) | + | z_{2} (τ) | ‖}_{C [0, T]} {‖ \hat{a} v + \hat{b} ‖}_{C [0, 1]},

(24)

Q_{5} (T) = 2 \sqrt{2 T} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} {‖ {[z (τ)]}^{- 1} ‖}_{C [0, T]} {(\sum_{k = 1}^{\infty} {λ_{k}}^{- 2})}^{\frac{1}{2}} [(1 + 6 \sqrt{2} | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} \sqrt{{‖ c (τ) ‖}_{C [0, T]}} T (1 + δ)) {‖ g_{vvv} (v, τ) ‖}_{L_{2} (D_{T})} + {‖ g_{vvv} (v, τ) (1 - \hat{b} - \hat{a} x) - 3 \hat{a} g_{vv} (v, τ) ‖}_{L_{2} (D_{T})} {‖ | g (v_{2}, τ) | + | g (v_{1}, τ) | ‖}_{C [0, T]} {‖ \hat{a} v + \hat{b} ‖}_{C [0, 1]},

(25)

Q_{6} (T) = 2 \sqrt{2 T} (1 + δ) {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} {‖ {[z (τ)]}^{- 1} ‖}_{C [0, T]} {(\sum_{k = 1}^{\infty} {λ_{k}}^{- 2})}^{\frac{1}{2}} [(1 + 6 \sqrt{2} | \hat{a} | {‖ \frac{1}{c (τ)} ‖}_{C [0, T]} \sqrt{{‖ c (τ) ‖}_{C [0, T]}} T (1 + δ)) {‖ g_{vvv} (v, τ) ‖}_{L_{2} (D_{T})} + {‖ g_{vvv} (v, τ) (1 - \hat{b} - \hat{a} x) - 3 \hat{a} g_{vv} (v, τ) ‖}_{L_{2} (D_{T})} {‖ | z_{1} (τ) | + | z_{2} (τ) | ‖}_{C [0, T]} {‖ \hat{a} v + \hat{b} ‖}_{C [0, 1]},

where,

E (T) = E_{4} (T) + E_{5} (T) + E_{6} (T), H (T) = H_{4} (T) + H_{5} (T) + H_{6} (T), G (T) = G_{4} (T) + G_{5} (T) + G_{6} (T), and Q (T) = Q_{4} (T) + Q_{5} (T) + Q_{6} (T),

holds, then problem

(1) - (3)

, has a unique solution in the ball

K = K_{R},

of the space

{X_{k}}^{3}

, where,

λ_{k} = 2 kπ; (k = 0, 1, \dots) .

3. FDM scheme for direct (Forward) problem

The direct problem will be resolved in this section, i.e., when unknown, $a (τ)$ , $c (τ)$ and $b (τ)$ are assumed to be given, in order, to solve this problem, Crank-Nicolson (C-N) FDM scheme had been used for finding the solutions of the nonlocal problem given by Equation (1)–(4) $.$ Where the domain $D_{T}$ had been divided into $M \times N$ mesh with spatial step size $∆ v = \frac{1}{M}$ and the time step size $∆ τ = \frac{T}{N}$ where $M$ and $N$ are given positive integers. The grid points have been given by

(26)

v_{i} = i ∆ v, i = \bar{0, M,}

(27)

τ_{j} = j ∆ τ, j = \bar{0, N,}

These quantities’ discretized form is given as follows,

w (v_{i}, τ_{j}) ≕ w_{i, j}, a (τ_{j}) ≕ a_{j}, b (τ_{j}) ≕ b_{j}, c (τ_{j}) ≕ c_{j}, f (v_{i}, τ_{j}) ≕ f_{i, j}, g (v_{i}, τ_{j}) ≕ g_{i, j}, and ζ (v_{i}) ≕ ζ_{i} for i = \bar{0, M,} j = \bar{0, N .}

Applying the C-N FDM, for the discretizes Equation (1) $,$ which approximated as,

(28)

\frac{w_{i, j + 1} - w_{i, j}}{∆ τ} = \frac{1}{2} ({\tilde{d}}_{j} \frac{w_{i + 1, j} - 2 w_{i, j} + w_{i - 1, j}}{{(∆ v)}^{2}} + {\tilde{a}}_{j} w_{i, j} + {\tilde{b}}_{j} g_{i, j} + {\tilde{d}}_{j} f_{i, j} + {\tilde{d}}_{j + 1} \frac{w_{i + 1, j + 1} - 2 w_{i, j + 1} + w_{i - 1, j + 1}}{{(∆ v)}^{2}} + {\tilde{a}}_{j + 1} w_{i, j + 1} + {\tilde{b}}_{j + 1} g_{i, j + 1} + {\tilde{d}}_{j + 1} f_{i, j + 1}), i = \bar{1, M,} j = \bar{0, N,}

where,

{\tilde{d}}_{j} = \frac{1}{c_{j}}

,

{\tilde{a}}_{j} = \frac{a_{j}}{c_{j}}

,

{\tilde{b}}_{j} = \frac{b_{j}}{c_{j}}

,

{\tilde{d}}_{j + 1} = \frac{1}{c_{j + 1}}

,

{\tilde{a}}_{j + 1} = \frac{a_{j + 1}}{c_{j + 1}}

, and,

{\tilde{b}}_{j + 1} = \frac{b_{j + 1}}{c_{j + 1}}

simplifying the above Equation (28), get the following difference equation,

(29)

- {\hat{A}}_{j + 1} w_{i - 1, j + 1} + (1 + {\hat{E}}_{j + 1}) w_{i, j + 1} - {\hat{A}}_{j + 1} w_{i + 1, j + 1} = {\hat{A}}_{j} w_{i - 1, j} + (1 - {\hat{E}}_{j}) w_{i, j} + {\hat{A}}_{j} w_{i + 1, j} + \frac{∆ τ}{2} \hat{H} + \frac{∆ τ}{2} \hat{D},

where,

{\hat{A}}_{j} = \frac{∆ τ {\tilde{d}}_{j}}{2 {(∆ v)}^{2}}

,

{\hat{A}}_{j + 1} = \frac{∆ τ {\tilde{d}}_{j + 1}}{2 {(∆ v)}^{2}}

,

{\hat{E}}_{j + 1} = (2 {\hat{A}}_{j + 1} - \frac{∆ τ}{2} {\tilde{a}}_{j + 1})

,

{\hat{E}}_{j} = (2 {\hat{A}}_{j} - \frac{∆ τ}{2} {\tilde{a}}_{j})

,

\hat{H} = {\tilde{b}}_{j} g_{i, j} + {\tilde{b}}_{j + 1} g_{i, j + 1}

, and,

\hat{D} = {\tilde{d}}_{j} f_{i, j} + {\tilde{d}}_{j + 1} f_{i, j + 1}

.

The C-N FDM discretizes Equations (2)–(4) which approximated as,

(30)

w_{i, 0} + δ w_{i, j} + I (v_{i}) = ζ_{i}, i = \bar{1, M,}

where

I (v_{i})

is integral in Equation (2), and we have,

I (v_{i}) = \frac{∆ τ}{2} (\hat{Ƥ} (τ_{0}) w (v_{i}, τ_{0}) + \hat{Ƥ} (τ_{N}) w (v_{i}, τ_{N}) + 2 \sum_{j = 1}^{N - 1} \hat{Ƥ} (τ_{j}) w (v_{i}, τ_{j})),

Finally, the trapezoidal rule and C-N discretizes integral condition ( $4$ ) as;

(31)

w (v_{0}, τ_{j}) + w (v_{M}, τ_{j}) + 2 \sum_{i = 1}^{M - 1} w (v_{i}, τ_{j}) = 0, i = \bar{0, M}

(32)

w_{i, 0} = ζ_{i}, i = \bar{1, M,} j = 0,

(33)

w_{0, j} = γ w_{1, j}, j = \bar{0, N - 1,} i = 0,

From Equation (29) $,$ with the nonlocal initial conditions $w (v, 0) = ζ (v_{i}),$ (assuming $Ƥ (τ) = δ = 0$ ), and using Equations (30)–(33), with each time step $τ_{j}$ , we can be described in matrix form as

(34)

L W_{j + 1} = L 1 W_{j} + \frac{∆ τ}{2} \hat{V}, i = \bar{1, M}, j = \bar{0, N - 1,}

L = {[\begin{matrix} 1 & 0 & 0 & 0 & 0 & \dots & 0 & 0 & 0 & - 2 \\ - {\hat{A}}_{j + 1} & 1 + {\hat{E}}_{j + 1} & - {\hat{A}}_{j + 1} & 0 & 0 & \dots & 0 & 0 & 0 & 0 \\ 0 & - {\hat{A}}_{j + 1} & 1 + {\hat{E}}_{j + 1} & - {\hat{A}}_{j + 1} & 0 & \dots & 0 & 0 & 0 & 0 \\ \dots & \dots & \dots & ⋱ & ⋱ & ⋱ & ⋮ & \dots & ⋮ & ⋮ \\ 0 & 0 & 0 & 0 & 0 & \dots & 0 & - {\hat{A}}_{j + 1} & 1 + {\hat{E}}_{j + 1} & - {\hat{A}}_{j + 1} \\ 1 & 2 & 2 & 2 & 2 & \dots & 2 & 2 & 2 & 1 \end{matrix}]}_{(M) х (M)},

with,

L 1 = [\begin{matrix} 0 \\ {\hat{A}}_{j} w_{1, j} + (1 - {\hat{E}}_{j}) w_{2, j} + {\hat{A}}_{j} w_{3, j} \\ {\hat{A}}_{j} w_{2, j} + (1 - {\hat{E}}_{j}) w_{3, j} + {\hat{A}}_{j} w_{4, j} \\ {\hat{A}}_{j} w_{3, j} + (1 - {\hat{E}}_{j}) w_{4, j} + {\hat{A}}_{j} w_{5, j} \\ ⋮ \\ {\hat{A}}_{j} w_{M - 2, j} + (1 - {\hat{E}}_{j}) w_{M - 1, j} + {\hat{A}}_{j} w_{M, j} \\ {\hat{A}}_{j} w_{M - 1, j} + (1 - {\hat{E}}_{j}) w_{M, j} + {\hat{A}}_{j} w_{M + 1, j} \\ 0 \end{matrix}],

and,

\hat{V} = (\hat{H} + \hat{D})

, where,

\hat{V} = [\begin{matrix} 0 \\ (b_{j} g_{2, j} + b_{j + 1} g_{2, j + 1}) + ({\tilde{d}}_{j} f_{2, j} + {\tilde{d}}_{j + 1} f_{2, j + 1}) \\ (b_{j} g_{3, j} + b_{j + 1} g_{3, j + 1}) + ({\tilde{d}}_{j} f_{3, j} + {\tilde{d}}_{j + 1} f_{3, j + 1}) \\ (b_{j} g_{4, j} + b_{j + 1} g_{4, j + 1}) + ({\tilde{d}}_{j} f_{4, j} + {\tilde{d}}_{j + 1} f_{4, j + 1}) \\ ⋮ \\ (b_{j} g_{M - 1, j} + b_{j + 1} g_{M - 1, j + 1}) + ({\tilde{d}}_{j} f_{M - 1, j} + {\tilde{d}}_{j + 1} f_{M - 1, j + 1}) \\ (b_{j} g_{M, j} + b_{j + 1} g_{M, j + 1}) + ({\tilde{d}}_{j} f_{M, j} + {\tilde{d}}_{j + 1} f_{M, j + 1}) \\ 0 \end{matrix}]

W_{j + 1} = {[w_{1, j + 1}, w_{2, j + 1}, \dots, w_{M + 1, j + 1}, w_{M, j + 1}]}^{T}, W_{j} = {[w_{1, j}, w_{2, j}, \dots, w_{M - 1, j}, w_{M, j}]}^{T}, j = \bar{0, N - 1,}

the Equation (4) desired output data will be approximated using trapezoidal rule as

(35)

0 = \int_{0}^{1} H (v) w (v, τ_{j}) dv = \frac{∆ τ}{2} (w (v_{0}, τ_{j}) + w (v_{M}, τ_{j}) + 2 \sum_{k = 1}^{M - 1} w (v_{k}, τ_{j})) .

3.1 Stability and convergence analysis

The adopted C-N finite difference scheme used to solve the direct problem is unconditionally stable and second order accurate in both spatial and temporal discretization. A brief von Neumann stability analysis shows that the amplification factor satisfies $| G | = 1$ , for all mesh ratios, which guarantees the unconditional stability of the scheme. The local truncation error is of order $ϑ ({(∆ v)}^{2}, {(∆ τ)}^{2})$ confirming that the method provides second order accuracy and reliable convergence toward the analytical solution as the mesh is refined. Hence, the C-N formulation ensures both numerical stability and accuracy for the present problem.

3.2 Examples for direct problem

3.2.1 Example 1: Assume an illustration for the direct problem given by $(1) - (4)$ with $D_{T} : {0 \leq v \leq 1, 0 \leq τ \leq 1},$ and the input data are,

f (v, τ) = \frac{- 1}{6} e^{τ} (1 + τ) v^{3} {(2 v - 1)}^{3} (3 v - 1) + \frac{1}{44} (133 + 1188 e^{3 τ} + 176 π^{2} (3 τ - 1) - τ (6 τ + 1)) Sin (2 πv), z (τ) = \frac{17}{256} e^{τ} (3 τ - 1), c (τ) = 1 + 9 e^{3 τ}, τ \in [0, 1], v \in [0, 1],

and, the analytical solution is,

w (v, τ) = (3 τ - 1) Sin (2 πv), g (v, τ) = e^{τ} {(2 v^{2} - v)}^{3} (3 v - 1), a (τ) = \frac{1}{44} (1 + 2 τ), with, b (τ) = \frac{1}{6} (1 + τ), (v_{1}, v_{2}) = (\frac{1}{4}, \frac{3}{4}), τ \in [0, T], (v, τ) \in D_{T}

We investigate the accuracy of the solution using diverse mesh grid size, $M = N \in {20,40,80,160}$ . In Figure 1 (a)–(d), an excellent agreement between the exact and numerical solutions can be clearly observed, indicating high accuracy. It can be noted that as the number of mesh points increases, the accuracy of the obtained solution improves, which reflects the convergence and stability of the proposed numerical scheme. Moreover, the absolute error graph at each mesh shows that the maximum magnitude of error didn’t exceed $(10^{- 5})$ , as depicted in Figure 1 (a)–(d).

Figure 1. The graphs showing analytical and computational distribution with absolute error for the direct problem $(1) - (4)$ , for example1 $(a 1) - (d 1),$ represent the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

The obtained results clearly indicate that the mesh size has a noticeable influence on the numerical accuracy. As the spatial and temporal mesh are refined (i.e., as $M$ and $N$ increase), the numerical solution becomes smoother and shows a closer agreement with the analytical one. This behavior confirms the expected second-order convergence of the C-N finite difference scheme in both time and space. Beyond a certain refinement level (e.g., $M = N \geq 80$ ), further mesh subdivision produces only negligible changes, implying that the proposed schemeis numerically stable and nearly mesh-independent. Whilst, Figure 2 and Figure 3, show the numerical outcomes for the required information $w (v_{i}, τ), (i = 1, 2),$ evaluated at various mesh sizes such that (s.t.), observed that an excellent agreement is obtained.

Figure 2. The graphs showing the exact and numerical values for, required output $w (v_{1}, τ)$ equation $(5)$ , for example 1 $(a 2) - (d 2),$ with the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

Figure 3. The graphs showing the exact and numerical values for, required output $w (v_{2}, τ)$ equation $(5)$ , for example 1 $(a 3) - (d 3),$ with the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

3.2.2. Example 2: Assume an illustration for the direct problem given by $(1) - (4)$ with $D_{T} : {0 \leq v \leq 1, 0 \leq τ \leq 1},$ and the input data are,

f (v, τ) = \frac{1}{2} e^{τ} ((2 v - 1) (6 + (5 + τ) (v - 1) v) - \frac{2 Sin (πv)}{1 + τ}), z (τ) = \frac{- 3 e^{2 τ}}{32 \sqrt{2}}, c (τ) = 1,

and, the analytical solution is,

w (v, τ) = e^{τ} (1 - v) (v - \frac{1}{2}) v, g (v, τ) = e^{τ} Sin (πv), a (τ) = 6 + τ, with, b (τ) = \frac{1}{(1 + τ)}, (v_{1}, v_{2}) = (\frac{1}{4}, \frac{3}{4}), τ \in [0, T], (v, τ) \in D_{T}

The accuracy of the numerical solution is examined using several mesh grid size, $M = N \in {20,40,80,160}$ . Figure 4 $(a) - (d)$ , clearly demonstrates an excellent agreement between the analytical and numerical solutions, confirming the reliability of the proposed approach. It can be noted that as the number mesh becomes finer, the numerical accuracy improves, indicating the expected convergence and stability of the C-N finite difference scheme. Furthermore, the absolute error plots at different mesh sizes reveal that the maximum error magnitude remains below $(10^{- 8})$ , as illustrated in Figure 4 $(a) - (d)$ .

Figure 4. The graphs showing analytical and computational distribution with absolute error for the direct problem $(1) - (4)$ , for example2 $(a 4) - (d 4),$ represent the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

Similarly, in Example 2, the refinement of the spatial and temporal mesh has a direct impact on the numerical accuracy. As the grid is successively refined, the computed solution aligns more closely with the analytical one, and the absolute error distribution becomes smoother across the domain. This confirms that the proposed C-N finite difference formulation maintains its second-order convergence and numerical stability. It can therefore be concluded that the results are consistent and nearly insensitive to further mesh refinement beyond the tested resolutions, indicating satisfactory mesh-independence of the method. Whilst, Figure 5 and Figure 6, show the numerical outcomes for the required information $w (v_{i}, τ), (i = 1, 2),$ evaluated at various mesh sizes s.t. observed that an excellent agreement is obtained.

Figure 5. The graphs showing the exact and numerical values for, required output $w (v_{1}, τ)$ equation $(5)$ , for example 2 $(a 5) - (d 5),$ with the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

Figure 6. The graphs showing the exact and numerical values for, required output $w (v_{2}, τ)$ equation $(5)$ , for example 2 $(a 6) - (d 6),$ with the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

4. Inverse problem solution for Equations (1)–(5)

For the nonlinear IP $(1) - (5),$ we seek precise and stable identification of $w (v, τ), a (τ)$ and $b (τ)$ , that mean potential term $a (τ)$ , and $b (τ)$ is unknown. The one-dimensional second-order parabolic equation together with $w (v, τ)$ satisfies the problem given by Equation (1)–(5), the problem is reformulated as a nonlinear least-squares minimization task. The resulting minimization problem is solved numerically using the lsqnonlin routine available in MATLAB’s optimization Toolbox, which implements a trust-region reflective algorithm suitable for nonlinear least squares problems with bound constraints. Due to the ill-posedness of such problems, especially in the presence of noise data, Tikhonov Regularization (TR) is employed to stabilize the solution. The regularization parameter $β \geq 0$ , is selected carefully to balance the trade-off between measured and the numerically computed solution. For this purpose, MATLAB routine lsqnonlin is employed, to minimize the Tikhonov Regularized objective functional, the procedure begins with a suitably selected initial guess. The TR functional is derived based on condition $(5)$ , and the functional error is incorporated as follows:

(36)

F (a) = \sum_{i = 1}^{2} {‖ w (v_{i}, τ) - z_{i} (τ) ‖}^{2}_{L^{2} [0, τ]} + β ({‖ a (τ) ‖}^{2}_{L^{2} [0, τ]} + {‖ b (τ) ‖}^{2}_{L^{2} [0, τ]}),

where $β > 0,$ is a regularization parameter. The discretization form of $(29)$ is

(37)

F (\underline{a}) = \sum_{i = 1}^{2} \sum_{j = 1}^{N} {[w (v_{i}, τ_{j}) - z_{i} (τ_{j})]}^{2} + β (\sum_{j = 1}^{N} {a_{j}}^{2} + \sum_{j = 1}^{N} {b_{j}}^{2}),

The unregularized case, i.e., $β = 0,$ produces the regular a nonlinear least-squares functional, which is inherently unstable when dealing with noisy data. The MATLAB routine lsqnonlins is used to minimize $F$ under some physical constraint. The following parameters had been used for the subroutine:

• (Maxlter) maximum number of iterations $= 10^{6} х (N)$ .
• Solution and Objective function tolerance $= 10^{- 10}$
• The lower and upper bounds on the component of the vector $\underline{a}$ are $- 10^{- 2}$ and $10^{2}$ , respectively.

The IP $(1) - (5)$ are resolved via both precise and noisy measurement $(5)$ . By including a random error, the noisy data is numerically simulated:

(38)

z^{ε} (τ_{j}) = z (τ_{j}) + ε_{j} j = \bar{0, N .}

where

ε

, is random Gaussian normal distribution vectors with mean zero and standard deviations

σ

, given by

(39)

σ = ρ х {max}_{τϵ [0, T]} | z (τ) |

where,

ρ

is the percentage of noise. We use the MATLAB function “normrnd”, as:

(40)

\underline{ε} = normrnd (0, σ, N),

to generate the random variables

\underline{ε} = (ε_{j})

and,

j = \bar{0, N .}

5. Results and discussion

To evaluate the precision and stability of the numerical methods, a set of numerical experiments is conducted. These experiments are designed to evaluate the accuracy of the computed results by simulating realistic measurement conditions, where noise is introduced into the input data. To quantitatively evaluate the discrepancy between the exact and the numerically computed solutions, the root mean squares errors (RMSE) are utilized by the following expression

(41)

RMSE (a) = \sqrt{\frac{1}{N} \sum_{i = 1}^{N} {(a_{i}^{exact} - a_{i}^{numerical})}^{2}},

we take,

T = 1

for simplicity.

5.1. Example 3: IP

Consider the IP $(1) - (5)$ with the following input data,

f (v, τ) = (- 2 + 4 π^{2} (3 + τ) - τ (7 + 2 τ)) Sin (2 πv) - τ^{2} (7 + 2 τ) Sin (4 πv), τ \in [0, T],

z (τ) = \frac{- 1}{4} \sqrt{15 + \sqrt{5 (33 - 8 \sqrt{2})}} (3 + τ) (7 + 2 τ), c (τ) = 1, τ \in [0, T],

where

γ = 2

, and, the analytical solution for this IP is provided as follows,

w (v, τ) = (3 + τ) Sin (2 πv), g (v, τ) = (7 + 2 τ) Sin (4 πv), a (τ) = 2 τ + 1, with, b (τ) = τ^{2}, (v_{1}, v_{2}) = (\frac{1}{8}, \frac{2}{5}), τ \in [0, T], (v, τ) \in D_{T}

The time-dependent coefficient $a (τ)$ , and $b (τ)$ , are then reconstructed (in case, $M = N = 80$ , is taken), and consider the case of noise-free in the measurements, see Table 1, this mean $p = 0,$ in $(37)$ and, no regularization applied, Figure 7 shows the objective function $(37)$ without regularization i.e. $β = 0,$ while, in Figure 9, noise without regularization.

Table 1. Numerical information for no regularization $β = 0$ and noise $p = 0$ .

$M = N$	80
RMSE a( $τ$ )	0.0241
RMSE b( $τ$ )	2.1592 $\times 10^{- 4}$
Computational time	3594 Sec
No. of Iterations	16
Obj. Function	1.2221× $10^{- 10}$

Figure 7. The graph showing the objective function for the IP $(1) - (5)$ , in example 3 when the size of mesh $N = M = 80$ .

Figure 8. The graphs $(a 8)$ and $(b 8)$ showing the reconstructed coefficient $a (τ)$ and $b (τ)$ in comparison with exact value for the IP $(1) - (5)$ , in above example when the size of mesh $N = M = 80$ .

Figure 9. The graph showing the objective function when $p = 0.01 %,$ no regularization for the IP $(1) - (5)$ , in example 2, when the size of mesh $N = M = 80$ .

We investigate the time-dependent coefficient $a (τ)$ and $b (τ)$ under both exact and perturbed measurements, as demonstrated in Figures 7 $-$ 12. The associated results are unstable, as illustrated in Figure 9- Figure 10. This reveals that the problem is not properly posed, a noticeable instability appears in the results, which is anticipated due to the ill-posed nature of the investigated IP. The unregularized solutions exhibit noticeable instability even in the noise-free case, demonstrating the sensitivity of the problem to small perturbations in the input data. The TR approach is used by incorporating the penalty term $β ({‖ a (τ) ‖}^{2}_{L^{2} [0, τ]} + {‖ b (τ) ‖}^{2}_{L^{2} [0, τ]})$ into the classical least-squares formulation, as presented in Equation (37). In Figure 11, $p = 0.01 %$ noise strategy is employed to enhance the solution’s robustness with regularization parameters, $β \in {10^{- 3}, \dots, 10^{- 10}}$ , yielding numerically stable regularized reconstructions of $a (τ)$ and $b (τ)$ , as shown in Table 2. These results indicate that smaller values of $β$ are effective when the data is not contaminated.

Figure 10. The graphs $(a 10)$ and $(b 10)$ showing the exact and (unstable) numerical solution for $a (τ)$ and $b (τ)$ when $p = 0.01 %,$ no regularization for the IP $(1) - (5)$ , in example 3, when the size of mesh $N = M = 80$ .

Figure 11. The graph showing the objective function when $p = 0.01 %$ noise with regularization, $β \in {10^{- 3}, \dots, 10^{- 7}}$ , for the IP $(1) - (5)$ , in example 3, when the size of mesh $N = M = 80$ .

Figure 12. The graphs $(a 12)$ and $(b 12)$ showing the exact and (stable) numerical solution for $a (τ)$ and $b (τ)$ when $p = 0.01 %,$ and regularization, $β \in {10^{- 3}, \dots, 10^{- 7}}$ , for the IP $(1) - (5)$ , in example 3, when the size of mesh $N = M = 80$ .

Table 2. Numerical information for various regularization parameters and noise $p = 0.01 %, (N = 80$ ).

$β$	$10^{- 3}$	$10^{- 4}$	$10^{- 5}$	$10^{- 6}$	$10^{- 7}$	$10^{- 8}$	$10^{- 9}$	$10^{- 10}$
RMSE a( $τ$ )	0.3302	0.1220	0.0714	0.0617	0.1189	0.2408	0.2835	0.2865
RMSE b( $τ$ )	0.0552	0.0225	0.0259	0.0344	0.0365	0.0373	0.0377	0.0377
Computational time	84320 Sec	9211 Sec	1453 Sec	1451 Sec	3139 Sec	6277 Sec	5814 Sec	5506 Sec
No. of Iterations	401	43	6	6	14	22	22	22

Next, we will study the proportion of noise contamination with a ratio of $p = 0.01 %$ noise, in $(38)$ via $(39)$ , the associated results are unstable, as illustrated. This reveals that the problem is not properly, posed and small error in the data input cause large errors in the output solution, thus, the problem needs to be regularized. As a consequence, for the situation of $p = 0.01 %$ noise, the best selection for the regularization parameters is $10^{- 6}$ for $a (τ)$ and $10^{- 4}$ for $b (τ)$ as in Table 2, which results in the lowest RMSE values.

6. Conclusions

In this study, the identification of the unknown time-dependent coefficient $a (τ)$ and $b (τ)$ in a one-dimensional parabolic PDE are investigated. The problem is formulated from a numerically evaluated nonlocal integral subjected to initial and boundary conditions. The direct problem is solved using the Crank-Nicolson (C-N) scheme, while the IP is transformed into a nonlinear least squares optimization problem using the lsqnonlin routine in MATLAB. An example test cases are employed in numerical experiments to evaluate the accuracy and stability of the proposed method. According to the results, using TR significantly enhances the solution’s stability. When the root-mean square error (RMSE) numbers are their lowest, the most effective approach is identified. Furthermore, numerical results have been provided to illustrate the accuracy and stability of the numerical method. It found that applying TR method stabilizes the solution. Cases of numerical results exist with noise (with and without) regularization. In the first case, consider when $p = 0.01 %$ noise and without regularization as in Figure 9- Figure 10 in example 3, the results are unstable and highly oscillated, on the contrary. Other case, when $p = 0.01 %$ noise introduced with $β \in {10^{- 3}, \dots, 10^{- 10}},$ regularization, obtain are stable and accurate results for the reconstructed coefficient $a (τ)$ and $b (τ)$ , which are (stable) numerical solutions. As for the remaining cases, we observe that the results are stable and steady when regularization values are $10^{- 6}$ for $a (τ)$ and, $10^{- 4}$ for $b (τ)$ in example 3, as shown in Table 2.

Data availability

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.

References

1. Llibre J, Ramirez R: Inverse Problems in Ordinary Differential Equations and Applications. Progress in Mathematics. 2016; 313. 978-3-319-26337-3, 978-3-319-26339-7 (eBook). Publisher Full Text
2. Lesnic D: Inverse Problems with Applications in Science and Engineering. CRC Press; 2nd ed. 2022. ISBN 978-0-367-00198-8(hbk), ISBN 978-1-032-12538-1(pbk), ISBN 978-0-429-40062-9(ebk). Publisher Full Text
3. Hasanov Hasanoglu A, Romanov VG: Introduction to Inverse Problems for Differential Equations. Switzerland AG: Springer Nature; 2nd ed. 2021. ISBN 978-3-030-79426-2, ISBN 978-3-030-79427-9(eBook). Publisher Full Text
4. Samarskii AA, Vabishchevich PN: Numerical Methods for Solving Inverse Problems of Mathematical Physics. Walter de Gruyter; 2007. 978-3-11-019666-5.
5. Moura Neto FD, Silva Neto AJ: An Introduction to Inverse Problems with Applications. Berlin Heidelberg: Springer-Verlag; 2013. 978-3-642-32556-4. Publisher Full Text
6. Tassiopoulou S, Koukiou G, Anastassopoulos V: Algorithms in Tomography and Related Inverse problems –A Review. Algorithms. 2024; 17: 71. Publisher Full Text
7. Colaco MJ, Orlande HRB, Dulikravich GS: Inverse and Optimization Problems in Heat Transfer. J. Braz. Soc. Mech. Sci. Eng. 2006; XXVIII, 28(1): 1–24.
8. Macdonald CM, Arridge SR, Munro PRT: On the Inverse Problem in Optical Coherence Tomography. Sci. Rep. 2023; 13: 1507. PubMed Abstract | Publisher Full Text | Free Full Text
9. Molnar A, Kiss D, Domozi Z: Radiation Source Localization Using a Model-Based Approach. Sensors. 2023; 23: 5983. PubMed Abstract | Publisher Full Text | Free Full Text
10. Yuldashev TK: Nonlocal Inverse Problem for a Pseudohyperbolic-Pseudoelliptic Type Integro-Differential Equation. Axioms, T.K. 2020; 9(2): 45. Publisher Full Text
11. Qahtan JA, Hussein MS: Dyhoum T.E.: An Approximate Solution for a Second Order Elliptic Inverse Coefficient Problem with Nonlocal Integral. Ibn Al-Haitham Journal for Pure and Applied Sciences. 2024; 37(2): 427–439. Publisher Full Text
12. Qahtan JA, Hussein MS: Numerical Solution to Inverse Coefficient Problem for Hyperbolic Equation under Overspecified Condition of General Integral Type. AIP Conference Proceedings. 2024: 2922–090003. Publisher Full Text
13. Ibraheem QW, Hussein MS: Determination of Time-Dependent Coefficient in Inverse Coefficent Problem of Fractional Wave Equation. AIP Conference Proceedings. 2024; 090004. 2933-2922. Publisher Full Text
14. Gani S, Hussein MS: Numerical Identification of Timewise Dependent Coefficient in Hyperbolic Inverse Problem. Advances in the Theory of Nonlinear Analysis and its Applications. 2024; 7(4): 290. Publisher Full Text
15. Weli SS, Hussein MS: Numerical Determination of Thermal Conductivity in Heat Equation under Nonlocal Boundary Conditions and Integral as Over Specified Condition. Iraqi Journal of Science. 2022; 63(11): 4944–4956. Publisher Full Text
16. Huntul MJ, Hussein MS: Simultaneous Identification of Thermal Conductivity and Heat Source in the Heat Equation. Iraqi Journal of Science. 2021; 62(6): 1968–1978. Publisher Full Text
17. Gani S, Hussein MS, Dyhoum TE: Timewise-dependent Coefficients Identification Problems for Third-Order Pseudo-Parabolic Equations from Nonlocal Extra Conditions. Ibn Al-Haitham Journal for Pure and Applied Sciences. 2025; 38(1): 2521–3407. Publisher Full Text
18. Huntul MJ: Identification of the unknown heat source terms in a 2D parabolic equation. Journal of King Saud University. 2021; 33(4): 101524. Publisher Full Text
19. Huntul MJ, Oussaeif T: Solvability of the Nonlocal Inverse Parabolic Problem and Numerical Results. Comput. Syst. Eng. 2022; 40(3): 1105–1126. Publisher Full Text
20. Hussein MS, Adil Z: Numerical Solution for Two-Sided Stefan Problem. Iraqi Journal of Science. 2020; 61(2): 444–452. Publisher Full Text
21. Qassim M, Hussein MS: Numerical Solution to Recover Time-Dependent Coefficient and Free Boundary from Nonlocal and Stefan Type Overdetermination Conditions in Heat Equation. Iraqi Journal of Science. 2021; 62(3): 950–960. Publisher Full Text
22. Ismailov MI, Ozawa T, Suragan D: Inverse Problems Of Identifying The Time-Dependent Source Coefficient For Subelliptic Heat Equations. Inverse Problems and Imaging. 2023; 19(1): 56–72. Publisher Full Text
23. Hussein MS, Dyhoum TE, Hussein SO, et al.: Identification of Time-Wise Thermal Diffusivity Advection Velocity on The Free-Boundary Inverse Coefficient Problem. Mathematics. 2024; 12(17): 2629. Publisher Full Text
24. Weli SS, Hussein MS: Finding Timewise Diffusion Coefficient From Nonlocal Integral Condition In One-Dimensional Heat Equation’. AIP Conference Proceeding. 2024; 2922: 090006. Publisher Full Text
25. Gani S, Hussein MS: A fourth Order Pseudoparabolic Inverse Problem to Identify the Time Dependent Potential Term from Extra Condition. Iraqi Journal of Science. 2024; 65(8): 4529–4549. Publisher Full Text
26. Ibraheem QW, Hussein MS: Determination of Timewise-Source Coefficient in Time-Fractional Reaction-Diffusion Equation from First Order Heat Moment. Iraqi Journal of Science. 2024; 65(3): 1612–1628. Publisher Full Text
27. Azizbayov E: The Nonlocal Inverse Problem of The Identification of The Lowest Coefficient and The Right-Hand Side in a Second-Order Parabolic Equation with Integral Conditions. Bound. Value Probl. 2019; 2019: 11. Publisher Full Text

Comments on this article Comments (0)

Version 1

VERSION 1 PUBLISHED 10 Feb 2026

Author details Author details

¹ Ministry of Education, Directorate of Education Thi Qar, Thi Qar, 64001, Iraq
² University of Baghdad Al-Jaderyia Campus College of Science, Baghdad, Baghdad Governorate, Iraq

Mohammed A.J.Al-Shatrah
Roles: Data Curation, Formal Analysis, Investigation, Methodology, Software, Visualization, Writing – Original Draft Preparation

Mohammed Sabah Hussein
Roles: Conceptualization, Formal Analysis, Resources, Software, Supervision, Validation, Writing – Review & Editing

Competing interests

No competing interests were disclosed.

Grant information

The author(s) declared that no grants were involved in supporting this work.

Article Versions (1)

version 1

Published: 10 Feb 2026, 15:228

https://doi.org/10.12688/f1000research.173252.1

Copyright

© 2026 A.J.Al-Shatrah M and Sabah Hussein M. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Download

Export To

metrics

	Views	Downloads
F1000Research	-	-
PubMed Central Data from PMC are received and updated monthly.	-	-

Citations

0

SEE MORE DETAILS

CITE

how to cite this article

A.J.Al-Shatrah M and Sabah Hussein M. Simultaneous Numerical Determination of Two Time-dependent Coefficients in Second Order Parabolic Equation With Nonlocal Initial and Boundary Conditions [version 1; peer review: awaiting peer review]. F1000Research 2026, 15:228 (https://doi.org/10.12688/f1000research.173252.1)

NOTE: If applicable, it is important to ensure the information in square brackets after the title is included in all citations of this article.

track

receive updates on this article

Track an article to receive email alerts on any updates to this article.

Open Peer Review

Comments on this article Comments (0)

Version 1

VERSION 1 PUBLISHED 10 Feb 2026

Open Peer Review

Reviewer Status

AWAITING PEER REVIEW

Comments on this article

All Comments(0)

Add a comment

Sign up for content alerts

Browse by related subjects

[1] 1. Llibre J, Ramirez R: Inverse Problems in Ordinary Differential Equations and Applications. Progress in Mathematics. 2016; 313. 978-3-319-26337-3, 978-3-319-26339-7 (eBook). Publisher Full Text

[2] 2. Lesnic D: Inverse Problems with Applications in Science and Engineering. CRC Press; 2nd ed. 2022. ISBN 978-0-367-00198-8(hbk), ISBN 978-1-032-12538-1(pbk), ISBN 978-0-429-40062-9(ebk). Publisher Full Text

[3] 3. Hasanov Hasanoglu A, Romanov VG: Introduction to Inverse Problems for Differential Equations. Switzerland AG: Springer Nature; 2nd ed. 2021. ISBN 978-3-030-79426-2, ISBN 978-3-030-79427-9(eBook). Publisher Full Text

[4] 4. Samarskii AA, Vabishchevich PN: Numerical Methods for Solving Inverse Problems of Mathematical Physics. Walter de Gruyter; 2007. 978-3-11-019666-5.

[5] 5. Moura Neto FD, Silva Neto AJ: An Introduction to Inverse Problems with Applications. Berlin Heidelberg: Springer-Verlag; 2013. 978-3-642-32556-4. Publisher Full Text

[6] 6. Tassiopoulou S, Koukiou G, Anastassopoulos V: Algorithms in Tomography and Related Inverse problems –A Review. Algorithms. 2024; 17: 71. Publisher Full Text

[7] 7. Colaco MJ, Orlande HRB, Dulikravich GS: Inverse and Optimization Problems in Heat Transfer. J. Braz. Soc. Mech. Sci. Eng. 2006; XXVIII, 28(1): 1–24.

[8] 8. Macdonald CM, Arridge SR, Munro PRT: On the Inverse Problem in Optical Coherence Tomography. Sci. Rep. 2023; 13: 1507. PubMed Abstract | Publisher Full Text | Free Full Text

[9] 9. Molnar A, Kiss D, Domozi Z: Radiation Source Localization Using a Model-Based Approach. Sensors. 2023; 23: 5983. PubMed Abstract | Publisher Full Text | Free Full Text

[10] 10. Yuldashev TK: Nonlocal Inverse Problem for a Pseudohyperbolic-Pseudoelliptic Type Integro-Differential Equation. Axioms, T.K. 2020; 9(2): 45. Publisher Full Text

[11] 11. Qahtan JA, Hussein MS: Dyhoum T.E.: An Approximate Solution for a Second Order Elliptic Inverse Coefficient Problem with Nonlocal Integral. Ibn Al-Haitham Journal for Pure and Applied Sciences. 2024; 37(2): 427–439. Publisher Full Text

[12] 12. Qahtan JA, Hussein MS: Numerical Solution to Inverse Coefficient Problem for Hyperbolic Equation under Overspecified Condition of General Integral Type. AIP Conference Proceedings. 2024: 2922–090003. Publisher Full Text

[13] 13. Ibraheem QW, Hussein MS: Determination of Time-Dependent Coefficient in Inverse Coefficent Problem of Fractional Wave Equation. AIP Conference Proceedings. 2024; 090004. 2933-2922. Publisher Full Text

[14] 14. Gani S, Hussein MS: Numerical Identification of Timewise Dependent Coefficient in Hyperbolic Inverse Problem. Advances in the Theory of Nonlinear Analysis and its Applications. 2024; 7(4): 290. Publisher Full Text

[15] 15. Weli SS, Hussein MS: Numerical Determination of Thermal Conductivity in Heat Equation under Nonlocal Boundary Conditions and Integral as Over Specified Condition. Iraqi Journal of Science. 2022; 63(11): 4944–4956. Publisher Full Text

[16] 16. Huntul MJ, Hussein MS: Simultaneous Identification of Thermal Conductivity and Heat Source in the Heat Equation. Iraqi Journal of Science. 2021; 62(6): 1968–1978. Publisher Full Text

[17] 17. Gani S, Hussein MS, Dyhoum TE: Timewise-dependent Coefficients Identification Problems for Third-Order Pseudo-Parabolic Equations from Nonlocal Extra Conditions. Ibn Al-Haitham Journal for Pure and Applied Sciences. 2025; 38(1): 2521–3407. Publisher Full Text

[18] 18. Huntul MJ: Identification of the unknown heat source terms in a 2D parabolic equation. Journal of King Saud University. 2021; 33(4): 101524. Publisher Full Text

[19] 19. Huntul MJ, Oussaeif T: Solvability of the Nonlocal Inverse Parabolic Problem and Numerical Results. Comput. Syst. Eng. 2022; 40(3): 1105–1126. Publisher Full Text

[20] 20. Hussein MS, Adil Z: Numerical Solution for Two-Sided Stefan Problem. Iraqi Journal of Science. 2020; 61(2): 444–452. Publisher Full Text

[21] 21. Qassim M, Hussein MS: Numerical Solution to Recover Time-Dependent Coefficient and Free Boundary from Nonlocal and Stefan Type Overdetermination Conditions in Heat Equation. Iraqi Journal of Science. 2021; 62(3): 950–960. Publisher Full Text

[22] 22. Ismailov MI, Ozawa T, Suragan D: Inverse Problems Of Identifying The Time-Dependent Source Coefficient For Subelliptic Heat Equations. Inverse Problems and Imaging. 2023; 19(1): 56–72. Publisher Full Text

[23] 23. Hussein MS, Dyhoum TE, Hussein SO, et al.: Identification of Time-Wise Thermal Diffusivity Advection Velocity on The Free-Boundary Inverse Coefficient Problem. Mathematics. 2024; 12(17): 2629. Publisher Full Text

[24] 24. Weli SS, Hussein MS: Finding Timewise Diffusion Coefficient From Nonlocal Integral Condition In One-Dimensional Heat Equation’. AIP Conference Proceeding. 2024; 2922: 090006. Publisher Full Text

[25] 25. Gani S, Hussein MS: A fourth Order Pseudoparabolic Inverse Problem to Identify the Time Dependent Potential Term from Extra Condition. Iraqi Journal of Science. 2024; 65(8): 4529–4549. Publisher Full Text

[26] 26. Ibraheem QW, Hussein MS: Determination of Timewise-Source Coefficient in Time-Fractional Reaction-Diffusion Equation from First Order Heat Moment. Iraqi Journal of Science. 2024; 65(3): 1612–1628. Publisher Full Text

[27] 27. Azizbayov E: The Nonlocal Inverse Problem of The Identification of The Lowest Coefficient and The Right-Hand Side in a Second-Order Parabolic Equation with Integral Conditions. Bound. Value Probl. 2019; 2019: 11. Publisher Full Text

Simultaneous Numerical Determination of Two Time-dependent Coefficients in Second Order Parabolic Equation With Nonlocal Initial and Boundary Conditions

Abstract

Background

Methods

Results

Conclusions

Keywords

1. Introduction

2. Mathematical formulation

(1)

(2)

(3)

(4)

(5)

2.1 Definition 1

2.2 Existence of the inverse problem classical solution

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

(14)

(15)

(16)

(17)

(18)

(19)

(20)

(21)

(22)

(23)

(24)

(25)

3. FDM scheme for direct (Forward) problem

(26)

(27)

(28)

(29)

(30)

(31)

(32)

(33)

(34)

(35)

3.1 Stability and convergence analysis

3.2 Examples for direct problem

Figure 1. The graphs showing analytical and computational distribution with absolute error for the direct problem (1)−(4) , for example1 (a1)−(d1), represent the size of mesh N=M={20,40,80,160} , respectively.

Figure 2. The graphs showing the exact and numerical values for, required output w(v1,τ) equation (5) , for example 1 (a2)−(d2), with the size of mesh N=M={20,40,80,160} , respectively.

Figure 3. The graphs showing the exact and numerical values for, required output w(v2,τ) equation (5) , for example 1 (a3)−(d3), with the size of mesh N=M={20,40,80,160} , respectively.

Figure 4. The graphs showing analytical and computational distribution with absolute error for the direct problem (1)−(4) , for example2 (a4)−(d4), represent the size of mesh N=M={20,40,80,160} , respectively.

Figure 5. The graphs showing the exact and numerical values for, required output w(v1,τ) equation (5) , for example 2 (a5)−(d5), with the size of mesh N=M={20,40,80,160} , respectively.

Figure 6. The graphs showing the exact and numerical values for, required output w(v2,τ) equation (5) , for example 2 (a6)−(d6), with the size of mesh N=M={20,40,80,160} , respectively.

4. Inverse problem solution for Equations (1)–(5)

(36)

(37)

(38)

(39)

(40)

5. Results and discussion

(41)

5.1. Example 3: IP

Table 1. Numerical information for no regularization β=0 and noise p=0 .

Figure 7. The graph showing the objective function for the IP (1)−(5) , in example 3 when the size of mesh N=M=80 .

Figure 8. The graphs (a8) and (b8) showing the reconstructed coefficient a(τ) and b(τ) in comparison with exact value for the IP (1)−(5) , in above example when the size of mesh N=M=80 .

Figure 9. The graph showing the objective function when p=0.01%, no regularization for the IP (1)−(5) , in example 2, when the size of mesh N=M=80 .

Figure 10. The graphs (a10) and (b10) showing the exact and (unstable) numerical solution for a(τ) and b(τ) when p=0.01%, no regularization for the IP (1)−(5) , in example 3, when the size of mesh N=M=80 .

Figure 11. The graph showing the objective function when p=0.01% noise with regularization, β∈{10−3,…,10−7} , for the IP (1)−(5) , in example 3, when the size of mesh N=M=80 .

Figure 12. The graphs (a12) and (b12) showing the exact and (stable) numerical solution for a(τ) and b(τ) when p=0.01%, and regularization, β∈{10−3,…,10−7} , for the IP (1)−(5) , in example 3, when the size of mesh N=M=80 .

Table 2. Numerical information for various regularization parameters and noise p=0.01%,(N=80 ).

6. Conclusions

Data availability

References

Comments on this article Comments (0)

Open Peer Review

Comments on this article Comments (0)

Open Peer Review

Reviewer Status

Figure 1. The graphs showing analytical and computational distribution with absolute error for the direct problem $(1) - (4)$ , for example1 $(a 1) - (d 1),$ represent the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

Figure 2. The graphs showing the exact and numerical values for, required output $w (v_{1}, τ)$ equation $(5)$ , for example 1 $(a 2) - (d 2),$ with the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

Figure 3. The graphs showing the exact and numerical values for, required output $w (v_{2}, τ)$ equation $(5)$ , for example 1 $(a 3) - (d 3),$ with the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

Figure 4. The graphs showing analytical and computational distribution with absolute error for the direct problem $(1) - (4)$ , for example2 $(a 4) - (d 4),$ represent the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

Figure 5. The graphs showing the exact and numerical values for, required output $w (v_{1}, τ)$ equation $(5)$ , for example 2 $(a 5) - (d 5),$ with the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

Figure 6. The graphs showing the exact and numerical values for, required output $w (v_{2}, τ)$ equation $(5)$ , for example 2 $(a 6) - (d 6),$ with the size of mesh $N = M = {20, 40, 80, 160}$ , respectively.

Table 1. Numerical information for no regularization $β = 0$ and noise $p = 0$ .

Figure 7. The graph showing the objective function for the IP $(1) - (5)$ , in example 3 when the size of mesh $N = M = 80$ .

Figure 8. The graphs $(a 8)$ and $(b 8)$ showing the reconstructed coefficient $a (τ)$ and $b (τ)$ in comparison with exact value for the IP $(1) - (5)$ , in above example when the size of mesh $N = M = 80$ .

Figure 9. The graph showing the objective function when $p = 0.01 %,$ no regularization for the IP $(1) - (5)$ , in example 2, when the size of mesh $N = M = 80$ .

Figure 10. The graphs $(a 10)$ and $(b 10)$ showing the exact and (unstable) numerical solution for $a (τ)$ and $b (τ)$ when $p = 0.01 %,$ no regularization for the IP $(1) - (5)$ , in example 3, when the size of mesh $N = M = 80$ .

Figure 11. The graph showing the objective function when $p = 0.01 %$ noise with regularization, $β \in {10^{- 3}, \dots, 10^{- 7}}$ , for the IP $(1) - (5)$ , in example 3, when the size of mesh $N = M = 80$ .

Figure 12. The graphs $(a 12)$ and $(b 12)$ showing the exact and (stable) numerical solution for $a (τ)$ and $b (τ)$ when $p = 0.01 %,$ and regularization, $β \in {10^{- 3}, \dots, 10^{- 7}}$ , for the IP $(1) - (5)$ , in example 3, when the size of mesh $N = M = 80$ .

Table 2. Numerical information for various regularization parameters and noise $p = 0.01 %, (N = 80$ ).