🔢 mathematics

Propagation of singularities and inverse problems for the viscoacoustic wave equation

This paper establishes the uniqueness of recovering both the spatially varying sound speed and all time derivatives at zero of the memory kernel in the viscoacoustic wave equation from exterior source-to-solution data by constructing semiclassical solutions, proving propagation of singularities, and reducing the problem to lens rigidity and geodesic ray transform inverse problems.

Original authors: Giovanni Covi, Maarten de Hoop, Mikko Salo

Published 2026-03-26

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Giovanni Covi, Maarten de Hoop, Mikko Salo

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 standing outside a mysterious, sealed room. Inside this room, there is a complex fluid that behaves like a mix of honey and water. If you drop a stone outside, the sound waves travel through the air, hit the wall, and enter the room. Inside, the waves bounce around, slow down, and change shape because of the "sticky" nature of the fluid (viscosity) and the fact that the fluid "remembers" how it was deformed in the past (memory).

Your goal? To figure out exactly what the room is made of and how the fluid inside behaves, without ever stepping inside. You can only listen to the sounds coming out of the room after you've made noise outside.

This is the core challenge of the paper "Propagation of Singularities and Inverse Problems for the Viscoacoustic Wave Equation" by Covi, de Hoop, and Salo.

Here is a breakdown of their work using simple analogies:

1. The Problem: The "Black Box"

The scientists are studying a mathematical model called the Viscoacoustic Wave Equation.

The "Acoustic" part: This is like normal sound waves (like a voice or a drumbeat).
The "Visco" part: This adds "memory." Imagine a sponge. If you squeeze it, it doesn't just snap back instantly; it slowly returns to shape. The material "remembers" the squeeze. In the math, this is a memory kernel—a function that says, "How the wave behaves now depends on how it was squeezed yesterday, the day before, and so on."

The big question is: Can we figure out the speed of sound inside the room and the exact "memory recipe" of the fluid just by listening to the waves that come back out?

2. The Old Way vs. The New Way

Previous scientists tried to solve this by sending a single, sharp "click" (a delta function) into the room. They hoped to hear the echo and calculate the distance to the walls.

The Problem: In complex environments (like the Earth's crust or a complex fluid), sound waves can cross over each other and create "caustics" (focusing points where the sound gets super loud and messy). When this happens, the simple "click" method breaks down. It's like trying to read a map when the ink has smeared.

The Authors' Solution: The "Laser Beam" Approach
Instead of a messy "click," the authors invent a special type of wave called a Gaussian Beam.

The Analogy: Imagine a flashlight beam. It's not a single point of light; it's a tight, focused beam that travels in a straight line (or follows a curve) without spreading out too much.
The Magic: They create mathematical "beams" that hug a specific path through the room. Even if the room is messy and has "caustics," these beams stay focused.
The Result: They proved that these beams travel through the room just like light in a vacuum, ignoring the "memory" part of the fluid for their path. The memory only affects how loud the beam is, not where it goes.

3. The Detective Work: Two Steps to Uniqueness

Once they have these focused beams, they act like detectives to solve the mystery in two steps:

Step A: Finding the Shape of the Room (Sound Speed)
By tracking where these beams enter and exit the room, they can map out the "travel times."

The Analogy: If you know how long it takes a runner to run from point A to point B, and you know the runner's speed, you can figure out the distance. Here, they reverse it: they measure the time and figure out the "speed of sound" map inside the room.
The Math: They use a famous mathematical tool called the Geodesic Ray Transform. Think of it as a super-advanced X-ray that can reconstruct the shape of an object just by knowing how long it takes light to cross it from every angle.

Step B: Finding the "Memory Recipe" (The Kernel)
Once they know the speed of sound, they look at the amplitude (loudness) of the beams.

The Analogy: Imagine the beam is a runner. The path tells you the distance. But if the runner is wearing heavy boots (viscosity/memory), they will be slower and quieter. By measuring exactly how much quieter the beam gets as it travels, they can calculate the "memory" of the fluid.
The Breakthrough: They proved that by measuring the beam at different points, they can figure out not just the memory, but the entire history of how the memory changes over time (all the time derivatives at zero). It's like being able to guess the entire recipe of a soup just by tasting the very first spoonful and knowing how the flavors evolve.

4. The Real-World Application: The "Extended Maxwell Model"

The paper ends with a practical application. They apply their math to a specific model used in geology and materials science called the Extended Maxwell Model.

The Analogy: This model describes materials (like polymers or the Earth's mantle) as a collection of tiny springs and dashpots (shock absorbers).
The Result: They proved that if you know how the material behaves at the very start (the first few moments), you can uniquely identify the stiffness of every spring and the thickness of every dashpot inside the material. No guessing required.

Summary

In short, this paper is a masterclass in mathematical detective work.

The Challenge: We can't see inside a complex, "sticky" medium.
The Tool: We use special, focused "laser-like" sound waves that ignore the chaos of the environment.
The Discovery: By tracking where these waves go and how loud they get, we can uniquely reconstruct the entire internal structure and memory of the medium.

This is huge for fields like seismology (understanding the Earth's interior to find oil or predict earthquakes) and medical imaging (seeing inside tissues that aren't just simple water). It proves that even with complex "memory" effects, the universe leaves enough clues for us to solve the puzzle.

1. Problem Statement

The paper addresses an inverse problem for the viscoacoustic wave equation, a scalar model describing wave propagation in media with memory (viscoelasticity). The goal is to uniquely determine the physical properties of the medium from exterior measurements.

The Model: The pressure $u$ $u$ satisfies the integro-differential equation:
$P u := -\frac{1}{2} \left( \partial_t^2 u - \nabla \cdot (c(x)\nabla u) + \int_0^t \nabla \cdot (G(x, t-s)\nabla u(x, s)) \, ds \right) = 0$
where:
- $c(x)$ is the spatially varying sound speed (time-independent).
- $G(x, t)$ is the memory kernel (space-time dependent), representing the viscous/anelastic behavior.
The Setup: Waves are generated by sources $f$ supported in the exterior of a bounded domain $\Omega \subset \mathbb{R}^n$ ( $n \geq 2$ ). The data available is the source-to-solution map $S$ , which maps exterior sources to the resulting wave field observed in the exterior region.
The Question: Does the map $S$ uniquely determine the sound speed $c(x)$ and the memory kernel $G(x, t)$ (specifically its time derivatives at $t=0$ ) inside $\Omega$ ?

2. Methodology

The authors employ a microlocal analysis approach, moving beyond classical geometric optics to handle complex geometries where caustics (focal points) may form.

A. Construction of Approximate Solutions (Quasimodes)

To probe the interior of the medium, the authors construct special approximate solutions called Gaussian beam quasimodes.

Form: These are high-frequency solutions concentrated along a fixed null bicharacteristic (a geodesic in the metric $c^{-2}g_0$ ).
Ansatz: The solution is a superposition of wave packets with a complex phase function $\phi$ and amplitudes $a_l, a''_l$ :
$u(x,t) \approx \sum (-ik)^{-l} \int e^{ik\phi} a_l \, d\sigma$
Handling Memory: Unlike standard wave equations, the memory term involves an integral. The authors show that the memory term affects the amplitudes but not the phase (propagation path).
- The principal amplitude satisfies a transport equation.
- The secondary amplitudes (required due to the memory term) satisfy Volterra integral equations.
Result: They prove the existence of these quasimodes with an error of order $O(k^{-N})$ for any $N$ , valid even when the sound speed is not "simple" (i.e., allowing for caustics).

B. Propagation of Singularities

The authors establish a propagation of singularities result for the semiclassical wave front set ( $WF_{scl}$ ).

Key Finding: The memory term $R$ is of lower order (in the semiclassical parameter $k$ ) compared to the wave operator $Q$ . Consequently, the propagation of wave-type singularities (where the time-frequency $\tau \neq 0$ ) is identical to that of the standard wave equation.
Implication: Singularities travel along the same geodesics determined by the sound speed $c(x)$ , regardless of the memory kernel. This allows the recovery of travel times and the scattering relation from the exterior data.
Note: The paper acknowledges "stationary singularities" (where $\tau=0$ ) which do not propagate, but these are not utilized in the inverse problem solution.

C. The Inverse Problem Strategy

The recovery of the coefficients proceeds in two stages:

Recovery of Sound Speed ( $c$ ):
- Using the propagation of singularities, the source-to-solution map determines the scattering relation and travel times (lens data).
- Under specific geometric assumptions (strict convexity or foliation by strictly convex hypersurfaces), the lens data uniquely determines the sound speed $c(x)$ via results on lens rigidity and the geodesic X-ray transform.
Recovery of Memory Kernel ( $G$ ):
- Once $c(x)$ is known, the authors analyze the amplitude of the Gaussian beam solutions along a geodesic.
- The amplitude depends on an integral involving the memory kernel $G$ and its time derivatives at $t=0$ .
- By recursively analyzing the transport and Volterra equations for higher-order terms in the asymptotic expansion, they recover all time derivatives $\partial_t^j G(x, 0)$ .

3. Key Contributions and Results

Theorem 1.1 & 1.2 (Approximate Solutions & Propagation)

Constructed Gaussian beam quasimodes for the viscoacoustic equation with memory.
Proved that the semiclassical wave front set propagates along null bicharacteristics of the principal symbol, unaffected by the memory term. This holds without restrictions on the sound speed (allowing for caustics).

Theorem 1.3 (Main Uniqueness Result)

Statement: If two viscoacoustic media have the same source-to-solution map $S$ $S$ , and their sound speeds agree on the boundary (to infinite order), then:
1. The sound speeds are identical inside the domain: $c_1 = c_2$ .
2. All time derivatives of the memory kernel at $t=0$ are identical: $\partial_t^j G_1(x, 0) = \partial_t^j G_2(x, 0)$ for all $j \in \mathbb{N}$ .
Conditions: Requires geometric assumptions (A1-A3) ensuring the solvability of the lens rigidity problem (e.g., simple manifolds for $n=2$ , foliated by strictly convex hypersurfaces for $n \geq 3$ ).

Theorem 1.4 (Application to Extended Maxwell Model)

Applied the main result to the Extended Maxwell Model, where the memory kernel is a sum of exponentials representing multiple relaxation processes.
Proved that if the time derivatives of the kernel match up to a certain order (determined by the number of relaxation processes), the model parameters (viscous and elastic coefficients) are uniquely determined.

4. Significance and Impact

Overcoming Caustics: Previous works (e.g., Romanov) relied on delta sources and singularity expansions that break down at caustics. This paper uses Gaussian beams and microlocal analysis to handle general geometries where caustics are present, which is crucial for realistic seismic applications.
Decoupling Propagation and Attenuation: The result that the memory term does not alter the path of singularities (only the amplitude) is a profound theoretical insight. It simplifies the inverse problem by separating the recovery of the metric (sound speed) from the recovery of the attenuation (memory).
Generalization: The methods extend to viscoelastic equations and other integro-differential models with convolutional memory terms.
Practical Application: The uniqueness result for the Extended Maxwell Model provides a theoretical foundation for seismic inversion techniques aimed at recovering subsurface viscosity and elasticity parameters, which are critical for understanding Earth's interior and material properties.

5. Conclusion

The paper successfully solves a challenging inverse problem for viscoacoustic media by combining semiclassical analysis, Gaussian beam constructions, and geometric inverse problem theory. It establishes that exterior measurements uniquely determine both the wave speed and the memory characteristics of the medium, even in the presence of complex wave phenomena like caustics.