🔢 mathematics

Convergence of Substructuring Waveform Relaxation Algorithms for Hyperbolic PDEs with Time Delay

This paper analyzes the stability, convergence, and computational efficiency of Dirichlet-Neumann and Neumann-Neumann waveform relaxation algorithms for solving hyperbolic partial differential equations with time delay, utilizing Fourier and Laplace transforms to derive linear convergence estimates, characterize finite-step convergence, and determine optimal parameters for heterogeneous domains.

Original authors: Bankim Chandra Mandal, Deeksha Tomer

Published 2026-07-07

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Bankim Chandra Mandal, Deeksha Tomer

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

Imagine you are trying to solve a giant, complex puzzle, but the pieces are moving, and they have a strange quirk: they react to what happened in the past.

This is the world of the paper you shared. It deals with mathematical equations (called Hyperbolic PDEs with Time Delay) that describe things like sound waves or vibrations where the current state depends on what happened a moment ago. Solving these on a computer is hard because the "memory" of the system makes the calculations heavy and slow.

The authors, Bankim C. Mandal and Deeksha Tomer, are proposing a smarter way to solve these puzzles. Instead of one giant computer trying to do all the work at once, they split the problem into smaller chunks and let different computers (or different parts of one computer) work on them simultaneously. They call this "Waveform Relaxation."

Here is a simple breakdown of their findings using everyday analogies:

1. The Problem: The "Echo" Effect

Think of a wave traveling through a room. In a normal room, the wave moves forward. But in this specific type of problem, the wave has an echo. If you shout now, the room reacts not just to your shout, but also to a shout you made 3 seconds ago. This "time delay" makes the math messy. If you try to calculate the whole room's behavior at once, the computer gets overwhelmed.

2. The Solution: Splitting the Room

The authors suggest cutting the room into two or more smaller rooms (subdomains).

DNWR (Dirichlet-Neumann): Imagine two neighbors, Alice and Bob, sharing a wall.
- Step 1: Alice solves her side of the room, telling Bob, "Here is the exact state of the wall on my side." (This is the Dirichlet part).
- Step 2: Bob takes that information and solves his side, then tells Alice, "Here is the force pushing against the wall on my side." (This is the Neumann part).
- They repeat this back-and-forth until they agree on what the wall is doing.
NNWR (Neumann-Neumann): Imagine a row of neighbors. Everyone solves their own room based on what their neighbors told them, then they all adjust their walls simultaneously to match up.

3. The Big Discovery: "Finite-Step" Convergence

Usually, when neighbors try to agree on a shared wall, they might take forever to get it perfect. They might get 90% right, then 95%, then 99%... and never quite finish.

However, the authors discovered something magical about these specific "echo" equations: They can finish in a finite number of steps.

The Analogy: Imagine the "error" (the part where they disagree) is a ghost. In a normal problem, the ghost lingers. But in this specific setup, every time the neighbors swap information, the ghost gets pushed further and further into the future.
The Result: If you only care about the next 10 seconds (a "time window"), the ghost gets pushed so far into the future that it disappears from your view entirely after just a few rounds of talking. The paper proves that if your time window is small enough, the neighbors will agree perfectly in just 2 or 3 steps, no matter how complex the math is.

4. The "Speed" Factor

The paper also looked at what happens if the "floor" of the rooms is different. Maybe Alice's room has a carpet (slow waves) and Bob's has a wooden floor (fast waves).

They found that if the rooms are sized correctly relative to how fast the waves travel in them, the neighbors can agree instantly (in just 2 steps) if they use the right "volume" setting (a parameter called $\theta$ ). It's like tuning a radio to the exact frequency where the static disappears immediately.

5. Testing the Theory

The authors didn't just write math; they ran computer simulations to prove it works.

They tested different sizes of rooms and different lengths of time.
They compared their method to the "old way" (called Classical Schwarz), which is like neighbors arguing over a shared wall that overlaps slightly.
The Winner: Their new method (DNWR and NNWR) was faster and more efficient. It didn't even need the "overlap" (the shared argument space) to work, making it cleaner and quicker.

Summary

This paper is about a new, highly efficient way for computers to solve complex wave problems that have a "memory" of the past. By splitting the problem into smaller pieces and having them talk to each other in a specific rhythm, the computer can find the perfect answer in just a handful of steps, rather than waiting forever. It's like turning a long, drawn-out negotiation between neighbors into a quick, decisive handshake.

Technical Summary: Convergence of Substructuring Waveform Relaxation Algorithms for Hyperbolic PDEs with Time Delay

Problem Statement
This study addresses the numerical solution of linear hyperbolic partial differential equations (PDEs) featuring constant time delays, specifically modeled by the wave-type equation $u_{tt} = c^2 u_{xx} + \lambda u(x, t-\tau) + h(x, t)$ . Such equations arise in physical and engineering contexts where system dynamics depend on past states, including wave propagation in viscoelastic materials, control systems with delayed feedback, and biological processes. The primary challenge lies in the efficient and stable parallel solution of these equations, particularly when the time delay introduces coupling between past and future states, increasing computational complexity. While Domain Decomposition Methods (DDMs) like Schwarz Waveform Relaxation (SWR) are established for time-dependent problems, their rigorous application and convergence analysis for hyperbolic PDEs with time delays—specifically using non-overlapping substructuring methods—remain underexplored.

Methodology
The authors investigate two non-overlapping substructuring waveform relaxation algorithms: Dirichlet-Neumann Waveform Relaxation (DNWR) and Neumann-Neumann Waveform Relaxation (NNWR). The study focuses on asymmetric domain decompositions and heterogeneous media (where wave speed $c$ varies across subdomains).

The analysis employs two primary mathematical frameworks:

Fourier Transform Analysis: Applied to the error equations to derive linear convergence estimates. This approach transforms the time-dependent PDEs into parameter-dependent Helmholtz problems in the frequency domain, allowing for the derivation of convergence factors based on subdomain geometry.
Laplace Transform Analysis: Utilized to establish finite-step convergence properties. By analyzing the convolution and time-shifting properties of the inverse Laplace transform, the authors demonstrate how error terms propagate and vanish over specific time windows. This method leverages the delay term $\tau$ to show that errors are effectively shifted out of the computational time window after a finite number of iterations.

The algorithms are tested on a linear wave equation with zero initial and boundary conditions for the error equations. The DNWR method involves solving a Dirichlet problem on one subdomain followed by a Neumann problem on the other, with interface updates governed by a relaxation parameter $\theta$ . The NNWR method extends this to multiple subdomains, solving Dirichlet problems in parallel followed by Neumann corrections.

Key Contributions and Results

DNWR Convergence (Asymmetric Homogeneous Case):
- Using Fourier analysis, the paper establishes a linear convergence bound for the DNWR algorithm with a relaxation parameter $\theta = 1/2$ . The convergence rate depends on the ratio of subdomain lengths ( $a$ and $b$ ).
- Laplace transform analysis proves that the method achieves finite-step convergence. Specifically, if the time window $T$ satisfies $T/k \leq 2 \min\{a/c, b/c\}$ , the error becomes identically zero within $k+1$ iterations. This result highlights that the time-delay term does not degrade the finite-step convergence property observed in standard wave equations; rather, the delay induces a time-shift in the error propagation.
DNWR Convergence (Heterogeneous Media):
- For heterogeneous media with wave speeds $c_1$ and $c_2$ in subdomains $\Omega_1$ and $\Omega_2$ , the study derives an optimal relaxation parameter $\theta = 1/(1 + c_1/c_2)$ .
- Under the geometric condition where scaled subdomain sizes are equal ( $|\Omega_1|/c_1 = |\Omega_2|/c_2$ ), the method achieves exact convergence in two iterations regardless of the time window length.
NNWR Convergence (Multi-Subdomain 1D and 2D):
- The paper mathematically establishes finite-step convergence for the NNWR algorithm in multi-subdomain configurations (1D and 2D strips).
- With a fixed relaxation parameter $\theta = 1/4$ , the algorithm converges in at most $k+1$ iterations provided the time window satisfies $T/k \leq 2 h_{min}/c$ , where $h_{min}$ is the minimum subdomain width.
- The 2D analysis utilizes Fourier sine transforms to reduce the problem to a set of 1D problems, confirming that the convergence behavior extends to higher dimensions.
Numerical Validation:
- Numerical experiments using Leapfrog discretization validate the theoretical bounds. Results confirm that the magnitude of the delay parameter $\tau$ does not negatively impact the convergence rate; instead, convergence is governed by the time window size relative to the subdomain dimensions and wave speed.
- Comparisons with the classical Schwarz Waveform Relaxation (SWR) method demonstrate that DNWR and NNWR exhibit superior iteration efficiency and do not require overlapping subdomains to ensure convergence, making them particularly suitable for heterogeneous coefficients.

Significance and Claims
The authors claim that this work provides the first rigorous analysis of DNWR and NNWR methods applied specifically to delay-dependent hyperbolic systems. The study bridges a gap in the literature regarding the scalability and convergence of non-overlapping domain decomposition methods for problems involving time delays.

The significance of the findings lies in the demonstration that these substructuring methods can achieve finite-step convergence (exact solution in a finite number of iterations) for time-delay problems, similar to their performance in non-delayed wave equations. This property is attributed to the time-shifting nature of the delay term, which pushes error contributions outside the computational window after a specific number of iterations. The paper posits that these methods offer a promising framework for parallel computing in delay differential equations, capable of handling heterogeneous structures and complex geometries without the need for subdomain overlap, thereby reducing computational overhead compared to classical overlapping Schwarz methods. The authors suggest that the framework is extensible to more complex settings, including systems with multiple or distributed delays and nonlinear dynamics, though such extensions are identified as directions for future research rather than results presented in this work.