Reconstruction methods for inverse scattering problems with phaseless data

This paper investigates phaseless inverse scattering problems for the Schrödinger equation by developing and validating three distinct reconstruction methods based on the inverse Born series framework for far-field total field, total field, and far-field scattered field data.

The Gist

Imagine you are trying to figure out what a hidden object looks like inside a dark room. You can't see the object directly, but you can shine a flashlight at it and watch how the light bounces off. In physics, this is called an inverse scattering problem. Usually, to perfectly reconstruct the object, you need to know two things about the light that bounces back: how bright it is (intensity) and its "timing" or wave pattern (phase).

However, in many real-world situations, our detectors are like cameras that can only see brightness. They are "phase-blind." They tell us how strong the signal is, but they lose the timing information. This makes the puzzle much harder, like trying to solve a jigsaw puzzle where half the pieces are missing their shapes.

This paper by Schotland and Yu is about developing new, clever ways to solve this "phase-blind" puzzle using a mathematical tool called the Inverse Born Series (IBS). Think of the IBS as a step-by-step recipe that starts with a rough guess and keeps refining it until the picture of the hidden object becomes clear.

Here is how they tackle three different versions of this problem:

1. The "Total Light" Puzzle (Phaseless Total Field)

The Scenario: You measure the total brightness of the light at a specific spot. This includes both the original flashlight beam and the light bouncing off the object mixed together.
The Challenge: Because the light waves mix, the brightness you measure is a complicated sum. It's like trying to guess the ingredients of a soup just by tasting the final flavor, but you don't know the ratio of salt to pepper.
The Solution: The authors extended their "recipe" (IBS) to work with just brightness.

• The Analogy: Imagine you are trying to hear a specific instrument in an orchestra, but you only have a microphone that measures total volume. The authors found a way to use the symmetry of the room. If you swap the position of the musician (the source) and the microphone (the observer), you get a second piece of the puzzle. By comparing these two swapped scenarios, they can mathematically "unmix" the signal to figure out the shape of the object, specifically for far-away measurements.

2. The "Bounced Light" Puzzle (Phaseless Scattered Field)

The Scenario: You only measure the light that actually bounced off the object (the scattered field), ignoring the original beam.
The Challenge: Knowing only the brightness of the bounce isn't enough to know the object's shape; it's like knowing how loud a drum hit is, but not knowing if it was a soft tap or a hard slam.
The Solution: They used a trick called polarization.

• The Analogy: Imagine you are trying to guess the shape of a hidden object by throwing balls at it. If you throw just one ball, you can't tell much. But if you throw four different types of balls (some straight, some spinning left, some spinning right, some bouncing back), the way they bounce off reveals the object's shape.
• In their math, they "throw" waves with different mathematical "spins" (using values like 1, -1, i, -i). By measuring the brightness for all four types and combining them, they can mathematically reconstruct the missing "timing" (phase) information. Once they have the phase, they can use their standard recipe to find the object.

3. Making the Recipe Faster (Efficiency)

The Challenge: The mathematical recipe (IBS) involves doing a lot of complex calculations. If you want to make the picture very detailed, the number of calculations can explode, taking forever to run on a computer.
The Solution: The authors found a way to organize the calculations so they don't have to start from scratch every time.

• The Analogy: Imagine you are baking a giant cake that requires layering ingredients. A slow baker makes a new batch of batter for every single layer. The authors' method is like a smart baker who keeps the batter from the previous layer and just adds a little more to it for the next one. This turns a slow, repetitive task into a fast, efficient one, making the computer run much quicker.

What Did They Find?

They tested these methods with computer simulations (digital experiments) using two types of hidden objects: simple circles and complex "clouds" of material.

• Low Contrast (Faint Objects): When the hidden object is weak (doesn't scatter much light), all their methods worked very well. The pictures they reconstructed were sharp and accurate, almost as good as if they had the full "phase" information.
• High Contrast (Strong Objects): When the object is very strong (scatters a lot of light), the math gets unstable. The "recipe" starts to break down, and the pictures become blurry or fail to form. This is a known limit of their method, not a failure of the idea.
• Comparison:
  • Having the full "phase" information is always the best (like having the full jigsaw puzzle).
  • Among the "phase-blind" methods, measuring the scattered light (Method 2) worked better than measuring the total light (Method 1). This is because the scattered light method allowed them to recover more of the missing information without throwing away data.

Summary

In short, this paper provides a toolkit for seeing hidden objects when you can only measure light intensity, not the wave's timing. They showed that by using clever mathematical tricks—like swapping source and detector positions or using multiple "spinning" waves—you can recover the missing information and reconstruct the object, provided the object isn't too "loud" or strong. They also made the math run faster so these techniques could be used in real-world computing.

Technical Summary

Technical Summary: Reconstruction Methods for Inverse Scattering Problems with Phaseless Data

Problem Statement
This work addresses the inverse scattering problem for the Schrödinger equation in $\mathbb{R}^d$, specifically focusing on the recovery of a compactly supported potential $V(x)$ from phaseless measurements. In many practical applications, only the intensity (modulus) of the field is accessible, while the phase information is lost. The authors investigate three specific variants of this problem:

1. Phaseless measurements of the total field ($|u|$).
2. Far-field phaseless measurements of the total field ($|u|$).
3. Far-field phaseless measurements of the scattered field ($|u_s|$).

The authors emphasize that phaseless inverse problems are significantly more challenging than their conventional counterparts due to the sublinear nature of the absolute-value function and associated issues of non-uniqueness.

Methodology
The core framework employed is the Inverse Born Series (IBS), which expresses the solution to the inverse problem as an explicitly computable functional of the measured data. The methodology is adapted for phaseless data through three distinct approaches corresponding to the data types:

• Extension of IBS for Phaseless Total Field:
  The authors derive a Born series expansion for the squared magnitude of the total field, $|u|^2 - |u_0|^2$. This yields a forward series involving multilinear operators $K_m$. The inverse problem is solved by constructing an inverse series. To handle the nonlinearity and ensure computational efficiency, the authors develop a recursive algorithm to evaluate the multilinear operators $K_n$. By utilizing prefix and suffix recursions for intermediate Born terms, the computational complexity of evaluating $K_n$ is reduced from $O(n^2)$ to $O(n)$ convolution operations.

• Fourier-Based Reconstruction for Far-Field Total Field:
  For far-field data, the authors propose a method to recover the Fourier coefficients of the potential, $\hat{V}(k(\hat{x} - \hat{y}))$. This exploits the scattering reciprocity between incident and observation directions. By interchanging the incident direction and the observation point, a linear system is formed that allows the recovery of Fourier coefficients. The authors note that this system can be ill-conditioned; therefore, a filtering strategy is implemented to discard "bad" Fourier samples where the condition number of the $2 \times 2$ system exceeds a threshold. The remaining coefficients are reconstructed using a Non-Uniform Fast Fourier Transform (NUFFT) based least-squares inversion.

• Polarization-Based Approach for Far-Field Scattered Field:
  Since $|\hat{V}|$ alone does not uniquely determine $V$, the authors employ a polarization-based strategy for scattered field data. By using incident fields that are superpositions of plane waves with specific phase shifts ($a \in \{\pm 1, \pm i\}$), the authors utilize the polarization identity to recover the complex scattering amplitude $A_B$. This allows the recovery of phase information, enabling the use of standard IBS reconstruction techniques on the reconstructed complex-valued data.

Key Contributions
The paper makes the following specific contributions:

1. Extension of IBS: The IBS framework is successfully extended to treat phaseless measurements for both total and scattered fields.
2. Fourier Method: A Fourier-based reconstruction method is proposed for far-field phaseless total-field data, leveraging symmetry between incident and observation directions to recover potential Fourier coefficients.
3. Polarization Method: A polarization-based approach is introduced to recover phase information from far-field phaseless scattered-field data, facilitating IBS reconstruction.
4. Convergence Analysis: The radius of convergence and error estimates for the IBS are analyzed for all three problem variants.
5. Algorithmic Efficiency: An efficient recursive implementation is developed for the multilinear operators in the phaseless total-field case, significantly reducing computational cost.
6. Numerical Validation: Comprehensive numerical experiments are conducted using smoothed circular and Gaussian mixture potentials to validate the proposed methods and compare their performance against conventional phase-based reconstruction.

Results
Numerical experiments were conducted in 2D with wavenumber $k=5.0$ using 400 incident waves. The results indicate:

• Low-Contrast Potentials: All proposed methods perform well for low-contrast potentials. Reconstructions from phase data are slightly more accurate than those from phaseless data, but the phaseless methods yield acceptable results.
• High-Contrast Potentials: For high-contrast potentials, the IBS iteration fails to converge in both phase and phaseless settings, consistent with the theoretical analysis of the convergence radius.
• Comparison of Data Types: Reconstructions from phaseless scattered-field data generally outperform those from phaseless total-field data. The authors attribute this to the fact that the total-field approach requires discarding a portion of the Fourier information (due to ill-conditioning in the linear system), whereas the scattered-field approach, via the polarization method, retains more informative data.
• Error Metrics: Relative errors ($\|V - V_{true}\|_2 / \|V_{true}\|_2$) are reported, showing that while phaseless methods incur higher errors than phase-based methods, they remain effective within the low-contrast regime.

Significance and Claims
The authors claim that this work provides a systematic investigation of reconstruction methods for phaseless inverse scattering within the IBS framework. The significance lies in:

• Providing explicit reconstruction algorithms for scenarios where phase information is unavailable, a common constraint in practical applications.
• Establishing the convergence properties of the IBS for these phaseless variants.
• Demonstrating that while phaseless reconstruction is inherently more difficult (smaller radius of convergence and higher error), it is feasible for low-contrast potentials.
• Showing that the scattered-field setting is superior to the total-field setting for phaseless data recovery due to the availability of more informative data and the avoidance of discarding Fourier samples.

The paper concludes modestly, noting that the methods are validated for low-contrast scenarios and fail for high-contrast ones, consistent with the theoretical limits of the Born series. Future work is suggested to explore phaseless inverse problems for wave equations of elasticity and nonlocal equations in optics.