🔢 mathematics

Numerical identification of the time-dependent coefficient in the heat equation with fractional Laplacian

This paper addresses the inverse problem of identifying a time-dependent source coefficient in a one-dimensional heat equation with a spectral fractional Laplacian by reducing it to a Volterra equation, proposing a stable and convergent numerical scheme based on fractional matrix powers and Crank–Nicolson discretization, and demonstrating robust reconstruction performance even with noisy data.

Original authors: Arshyn Altybay, Gulzat Nalzhupbayeva

Published 2026-09-10

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Arshyn Altybay, Gulzat Nalzhupbayeva

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the physical world, heat does not always spread out in the smooth, predictable way we learn in school. Sometimes, it moves in strange, erratic patterns, jumping across gaps or lingering in ways that defy simple explanation. This phenomenon, known as anomalous diffusion, occurs in everything from the movement of pollutants in groundwater to the way signals travel through complex biological tissues. To describe these behaviors, scientists use a sophisticated mathematical tool called the fractional Laplacian. Think of this tool as a way to measure how a system "feels" its surroundings not just from immediate neighbors, but from a distance, capturing the long-range interactions that standard models miss. While these equations are powerful for describing nature, they often contain hidden variables—unknown factors that drive the system but cannot be seen directly. Scientists must then work backward, using limited observations to deduce what is happening inside. This is the essence of an inverse problem: trying to reconstruct the invisible engine from the visible effects it produces.

A team of researchers has now tackled a particularly tricky version of this puzzle involving a heat equation governed by the spectral fractional Laplacian. In their study, they focused on a scenario where the source of the heat changes over time, but the only data available is a single, weighted average of the temperature across the entire space, rather than a detailed map of every point. This is a common real-world limitation; sensors often provide aggregate data, like the average temperature of a room, rather than a reading for every square inch. The challenge was to figure out the exact time-varying intensity of the heat source using only this sparse, averaged information, all while accounting for the complex, non-local nature of the fractional diffusion.

The researchers approached this by first translating the continuous mathematical problem into a form that a computer could handle. They broke the physical space into a grid of points and replaced the complex fractional operator with a discrete version that could be calculated efficiently. They then paired this spatial grid with a time-stepping method known as the Crank–Nicolson scheme, a technique famous for its stability and accuracy. By combining these tools, they created a step-by-step algorithm that could simultaneously calculate the temperature distribution and the unknown source coefficient. Crucially, they proved that this method is unconditionally stable, meaning it will not produce wild, nonsensical errors even if the time steps are large, and they demonstrated that the results become twice as accurate every time the grid is made finer or the time steps are shortened.

To solve the inverse problem, the team derived a specific formula that allows the unknown source coefficient to be recovered at each moment in time. This formula relies on a condition that ensures the measurement data is sensitive enough to the source; if this condition holds, the source can be uniquely identified. The researchers showed that as long as the grid is fine enough, this condition remains true in the computer simulation, guaranteeing that the solution is unique. They tested their method with two carefully constructed examples where the true answer was already known. In these tests, the algorithm successfully reconstructed both the temperature field and the time-varying source with high precision, matching the theoretical predictions of second-order accuracy.

A significant part of their work addressed the reality of noisy data. In the real world, measurements are never perfect; they contain small random errors. The researchers found that if one tries to calculate the rate of change of these noisy measurements directly, the errors explode, making the result useless. To fix this, they employed a smoothing technique called Savitzky–Golay differentiation, which fits a local curve to the noisy data before calculating the derivative. This approach acted as a filter, taming the noise while preserving the essential shape of the signal. Their experiments showed that even with measurement errors as high as five percent, the method could still recover the source coefficient with reasonable accuracy. The reconstructed source retained its main profile, and the calculated temperature field remained stable, proving that the method is robust enough to handle the imperfections of real-world data.

The study concludes that it is possible to reliably identify a time-dependent heat source from a single, averaged measurement, even when the underlying physics involves complex, long-range interactions. By combining a stable numerical scheme with a smart way of handling noisy data, the researchers provided a complete toolkit for solving this class of inverse problems. Their work does not just offer a theoretical proof but delivers a practical algorithm that has been tested and verified. The results suggest that for systems where only aggregate data is available, it is possible to peel back the layers of uncertainty and reveal the hidden drivers of the process, provided the data is treated with the right mathematical care.

Technical Summary: Numerical Identification of a Time-Dependent Coefficient in a Heat Equation with the Spectral Fractional Laplacian

Problem Formulation
The paper addresses an inverse coefficient problem for a one-dimensional heat equation governed by the spectral Dirichlet fractional Laplacian, denoted as $(-\Delta_D)^s$ with $0 < s < 1$ . The governing equation is:
$u_t(t, x) + (-\Delta_D)^s u(t, x) = r(t)f(t, x), \quad (t, x) \in (0, T] \times (0, l)$
subject to homogeneous Dirichlet boundary conditions and a prescribed initial state $u(0, x) = \phi(x)$ . The objective is to simultaneously determine the unknown time-dependent source coefficient $r(t)$ and the state $u(t, x)$ . The available data is not the full spatial state but a weighted integral measurement (overdetermination condition):
$\int_0^l u(t, x)\omega(x) \, dx = w(t), \quad 0 \le t \le T,$
where $\omega(x)$ is a known weight function and $w(t)$ is the measured response. The authors distinguish their approach from models using the integral (restricted) fractional Laplacian by employing the spectral definition, which relies on the eigenpairs of the classical Dirichlet Laplacian and preserves the geometry and boundary conditions of the underlying classical problem.

Methodology
The authors develop a complete numerical treatment for this inverse problem through the following steps:

Continuous Formulation: The inverse problem is reduced to a Volterra integral equation of the second kind for the unknown coefficient $r(t)$ . By differentiating the integral measurement condition and utilizing the semigroup representation of the solution, a scalar reconstruction formula is derived. This derivation relies on a nondegeneracy condition, specifically that the inner product $(f(t), \omega)_{L^2}$ remains bounded away from zero.
Spatial Discretization: The spectral fractional operator is discretized using the fractional matrix power $A_h = L_h^s$ , where $L_h$ is the standard second-order Dirichlet finite difference Laplacian. This operator is defined via the discrete sine transform, allowing efficient evaluation of $A_h v$ without explicitly forming the dense matrix.
Time Discretization: A Crank–Nicolson scheme is applied to the semi-discrete system. The scheme is formulated to handle the coupled state and coefficient reconstruction simultaneously.
Reconstruction Algorithm: At each time step, the algorithm performs two linear solves with the same system matrix (derived from the Crank–Nicolson step) to propagate the state and compute the source response. The coefficient $r(t)$ is then recovered using a closed scalar formula involving the discrete inner product of the computed state and the weight function.
Handling Noisy Data: For cases involving noisy integral measurements, the authors employ an adaptive Savitzky–Golay regularized differentiation procedure. This is necessary because direct numerical differentiation of noisy data amplifies high-frequency errors. The smoothing window is selected based on a discrepancy criterion to balance noise reduction and signal fidelity.

Key Contributions and Theoretical Results

Stability and Convergence: The paper proves that the Crank–Nicolson scheme for the direct problem is unconditionally stable. Under a second-order spatial consistency assumption (specifically for finite sine expansions), the scheme achieves an $O(\tau^2 + h^2)$ convergence rate in both $L^2$ and energy norms.
Identifiability: A discrete identifiability condition is established, showing that the denominator in the reconstruction formula remains non-zero provided the continuous nondegeneracy condition holds and the mesh is sufficiently fine.
Conditional Convergence: The authors establish the conditional convergence of the coupled state-coefficient reconstruction. The error estimate for the recovered coefficient depends on the discretization errors ( $\tau^2 + h^2$ ) and the error $\eta$ in the approximation of the derivative of the measurement data ( $w'(t)$ ). The paper explicitly notes that it does not derive a convergence rate for $\eta$ with respect to the raw noise level, but rather treats $\eta$ as a given parameter for the error bound.

Numerical Results
The method is validated using two manufactured examples where the exact solutions are finite sums of Dirichlet eigenfunctions.

Convergence: Numerical experiments confirm the theoretical second-order convergence rates in both time and space for the state and the coefficient in the absence of noise.
Robustness to Noise: Experiments with 1%, 3%, and 5% relative noise in the integral measurements demonstrate that the adaptive regularized differentiation effectively stabilizes the reconstruction. While the error in the recovered coefficient increases with noise levels, the method successfully recovers the main profile of the coefficient. The numerical state is shown to be less sensitive to perturbations than the coefficient, consistent with the algorithmic structure where the state is updated via a stable evolution step.

Significance
The paper claims to provide a complete numerical framework for identifying time-dependent sources in fractional heat equations with spectral fractional Laplacians using integral overdetermination. Its significance lies in the rigorous derivation of a stable, second-order accurate algorithm that includes a closed-form reconstruction formula and a proof of conditional convergence. The work bridges the gap between the theoretical formulation of such inverse problems and their practical numerical solution, specifically addressing the challenges posed by the nonlocal nature of the spectral fractional operator and the instability introduced by noisy derivative data. The authors position their work as a specific contribution to the growing body of literature on inverse problems for fractional diffusion equations, distinguishing their approach by the use of the spectral definition and the associated matrix-power discretization.

More like this