⚛️ quantum physics

Calibrated Pressure-Observable Born and Hessian Actions for Quantum-Assisted Waveform Inversion

This paper presents a quantum-assisted framework for acoustic full-waveform inversion that constructs a pressure-consistent interface for Born and Hessian actions using Schrödingerised propagation, proving that explicit receiver calibration is essential for second-order convergence and demonstrating the method's efficacy through a nine-qubit implementation and hybrid classical-quantum optimization.

Original authors: Guanyu Li, Jiwei Jia, Yu Wang, Yuping Duan

Published 2026-07-30

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Guanyu Li, Jiwei Jia, Yu Wang, Yuping Duan

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 find a hidden treasure map buried deep underground. You can't see the map, but you can send sound waves into the ground and listen to the echoes. This is the basic idea behind Full-Waveform Inversion (FWI), a technique used by geologists and engineers to "see" inside the Earth, check the health of bridges, or even look inside the human body without cutting them open. The problem is that the math required to turn those echoes into a clear picture is incredibly difficult. It involves solving massive, complex equations that describe how sound waves bounce, twist, and change speed as they hit different materials.

To make this math easier, scientists have been exploring a new trick: using quantum computers. Think of a quantum computer not as a faster calculator, but as a different kind of machine that speaks the native language of waves. By translating the sound-wave equations into a format that quantum computers understand (a process called "Schrödingerisation"), researchers hope to solve these problems much faster than classical computers ever could. However, there's a catch: just because you can simulate the waves on a quantum machine doesn't mean you automatically know how to read the final answer correctly. It's like having a perfect radio that can play a symphony, but if you don't tune the volume knob correctly, you might hear static instead of music.

This paper tackles that specific "volume knob" problem. The authors, Guanyu Li and colleagues, have built a new "interface" that connects the quantum simulation of sound waves to the actual physical measurements needed to update the underground map. Their main discovery is that to get the right answer, you must account for two things happening at once: the wave changing as it travels, and the fact that the "microphone" (the receiver) itself changes its sensitivity depending on the material it's listening to. They proved that if you ignore the microphone's changing sensitivity, your map will be completely wrong. They tested this idea using a tiny, nine-qubit quantum circuit simulation and showed that including this "calibration" step is essential for the math to work, reducing errors significantly and allowing the system to actually improve the model of the underground structure.

The Story of the "Calibrated Microphone"

So, how does this actually work? Let's imagine you are a detective trying to figure out what a room looks like by shouting and listening to the echo. In the world of Full-Waveform Inversion, the "room" is the Earth, and the "shout" is a sound wave.

The Quantum Trick
Usually, calculating how sound moves through the Earth is like trying to solve a giant puzzle where every piece is moving. The authors use a method called Schrödingerisation. Think of this as translating the messy, real-world sound equations into a "quantum language" that a quantum computer can play with. It's like taking a complex dance routine and rewriting the steps so a robot can perform them perfectly. This allows the computer to simulate how the sound waves (the "propagation") move through different layers of rock or tissue.

The Missing Piece: The Calibrated Microphone
Here is where the paper makes its big discovery. When the sound wave hits a receiver (a microphone), the signal it records depends on two things:

The Wave: How the sound traveled and changed along the way.
The Microphone's Sensitivity: How sensitive the microphone is to the specific material it's touching.

The authors realized that in previous attempts to use quantum computers for this, people often only looked at the first part (the wave) and forgot the second part (the microphone's sensitivity). They called this the "receiver-calibration term."

Imagine you are measuring the temperature of a cup of coffee with a thermometer. If the thermometer itself expands when it gets hot, your reading will be off unless you account for that expansion. In this paper, the "thermometer" is the receiver, and the "expansion" is the fact that the receiver's reading changes based on the speed of sound in the material ( $c$ ). The authors showed that the physical pressure ( $p$ ) is actually the product of the wave's energy and the speed of sound ( $p = c\pi$ ). When you try to calculate how to improve your map, you have to take the derivative of this product.

Mathematically, this means you get two terms:

One term for how the wave changed ( $c_0 \delta \pi$ ).
One term for how the speed of sound changed the reading directly ( $\delta c \pi_0$ ).

The paper proves that if you leave out that second term (the calibration), you aren't just making a small mistake; you are calculating the derivative of the wrong thing entirely. It's like trying to measure the speed of a car but forgetting to account for the fact that your speedometer is broken.

The Proof: A Tiny Quantum Test
To prove this, the authors didn't just do math on paper; they built a tiny, simulated quantum circuit.

The Setup: They created a small model with just 9 qubits (the basic units of quantum information). This is like a miniature, simplified version of the Earth.
The Test: They ran the inversion process (trying to find the hidden map) twice.
1. Without Calibration: They ignored the receiver's changing sensitivity. The result? The error was huge (an order-one error, meaning the answer was basically garbage). The "Gauss-Newton" step (the move to fix the map) pointed in the wrong direction, and the line search (the safety check) rejected the move.
2. With Calibration: They included the receiver's sensitivity. The error dropped dramatically. The system successfully reduced the initial model error from about 10.68% down to a range between 0.28% and 0.73% across ten different test runs.

The "Hybrid" Approach
The paper also describes a "hybrid" method. This means the heavy lifting of the quantum simulation is done by the quantum circuit, but the final steps of assembling the puzzle and deciding how big a step to take are done by a classical computer. They used a Variational Quantum Linear Solver (VQLS), which is a specific type of quantum algorithm designed to find the best solution to a linear equation.

In their experiment, they used Bernoulli samples (random coin flips based on the quantum probabilities) to drive the process. They ran 10,000 shots (measurements) for each part of the calculation to get a good average. The results showed that even with this noisy, finite-shot approach, the calibrated method worked. The "uncalibrated" method failed to improve the model, while the "calibrated" method consistently made the model better.

Why This Matters
This paper doesn't claim to have solved the problem of imaging the entire Earth on a quantum computer today. Instead, it provides a crucial "instruction manual" for how to connect the quantum simulation to the real-world physics correctly. It shows that if you want to use quantum computers for this kind of imaging, you must include the receiver calibration term. Without it, the quantum computer is just simulating a ghost story, not the real physics.

The authors verified their results in multiple ways:

They compared their quantum-style math against standard "finite difference" methods (a classic way to check math).
They checked that their "adjoint" (the reverse calculation) matched perfectly.
They confirmed that their "Hessian action" (a complex math step used to refine the solution) was symmetric and correct.

In the end, the paper concludes that connecting Hamiltonian wave propagation (the quantum part) to local FWI updates (the map-making part) requires a consistent derivative of the physical receiver. It's a small but vital piece of the puzzle that ensures the quantum computer isn't just playing with numbers, but is actually helping us see the hidden world beneath our feet.

Technical Summary: Calibrated Pressure-Observable Born and Hessian Actions for Quantum-Assisted Waveform Inversion

Problem Statement
Full-waveform inversion (FWI) is a computational method used to recover subsurface wavespeed models from measured seismic waveforms. It relies on repeated evaluations of forward pressure data and the application of Born, adjoint, and Gauss–Newton actions. While quantum algorithms offer primitives for Hamiltonian simulation and linear differential equations, a critical gap exists: a Hamiltonian propagation primitive does not inherently define the differentiated physical receiver row, its adjoint, or the specific measured quantities required for a local update. Specifically, in constant-density acoustic FWI, physical pressure is defined as $p = c\pi$ (where $c$ is wavespeed and $\pi$ is an energy variable). Standard quantum formulations often treat the observable as fixed, neglecting the fact that the physical pressure depends explicitly on the coefficient $c$ . Consequently, the directional derivative of the pressure data contains two terms: a propagated wavefield sensitivity and a direct receiver-calibration term. Omitting the latter leads to an incorrect observable and significant errors in the inversion update.

Methodology
The authors construct a pressure-consistent operator-and-readout interface for Born, adjoint, and Gauss–Newton actions using Schrödingerisation, a technique that embeds linear equations into Schrödinger-type dynamics by adding an auxiliary variable.

Hamiltonian Construction: The authors define energy variables $\pi = c^{-1}\partial_t u$ and $q = \nabla u$ to form an auxiliary-space Hamiltonian. While this yields a skew-adjoint block form suitable for simulation, physical pressure is recovered as the coefficient-dependent observable $p = c\pi$ .
Calibrated Derivative: Using the product rule, the directional derivative of the physical pressure is shown to be $D(c\pi)[c_0](\delta c) = c_0 \delta \pi + \delta c \pi_0$ . The first term represents the propagated sensitivity, while the second is the receiver-calibration term. The authors combine Duhamel differentiation with the receiver-row derivative to ensure both contributions are retained in the Born map, its adjoint, and the Gauss–Newton normal action.
Quantum Implementation:
- State Preparation: Source and receiver states are prepared as normalized profiles, with the physical pressure scale tracked classically.
- Propagation: The system uses product-formula propagation (Trotterization) for the Hamiltonian evolution.
- Born Action: The Born map is implemented via a Linear Combination of Unitaries (LCU) block for the derivative Hamiltonian, combined with a separate estimation of the receiver-calibration term.
- Hybrid Inversion: A compiled nine-qubit instance realizes the circuit. The resulting Bernoulli samples drive a four-parameter hybrid inversion loop. A two-qubit Variational Quantum Linear Solver (VQLS) represents the normalized update direction, while the assembly of the normal system, line search, and model refresh remain classical.

Key Contributions

Pressure-Consistent Formulation: The paper establishes that the physical-pressure derivative must include the receiver-calibration term ( $\delta c \pi_0$ ). It proves that omitting this term differentiates a "frozen" observable rather than the physical pressure, leading to order-one errors.
Operator-Readout Interface: The authors provide a complete construction connecting Schrödingerised propagation to local FWI updates, explicitly accounting for coefficient-dependent observables.
Resource Model: A detailed resource model is formulated, accounting for coefficient access, state preparation, input loading, LCU normalization, quadrature, and selected-output measurement.
Verification Framework: The construction is verified through multiple independent methods: finite-difference checks, independently coded tangent/reverse-adjoint recurrences, autodiff evaluations, and explicit-Jacobian comparisons.

Results

Consistency and Convergence: Numerical experiments confirm second-order convergence for smooth periodic refinement. Crucially, omitting the receiver-calibration term results in an order-one Born error and substantially alters the regularized Gauss–Newton direction (reducing correlation to 0.068 and causing the line search to reject the step).
Operator Checks: Across three finite-dimensional scales, the Born finite-difference discrepancy ranges from $9.33 \times 10^{-5}$ to $5.56 \times 10^{-4}$ , while adjoint and Hessian symmetry defects are at the level of machine precision ( $10^{-15}$ to $10^{-16}$ ).
Compiled Circuit Performance: A nine-qubit compiled circuit reproduces dense midpoint Born values to $1.89 \times 10^{-6}$ .
Hybrid Inversion: Ten finite-shot runs using the compiled circuit and VQLS successfully reduced the initial model error from $1.068 \times 10^{-1}$ to a range of $4.54 \times 10^{-4}$ to $7.26 \times 10^{-3}$ .
Readout Efficiency: The paper demonstrates that "selected-output" measurements (sketches) can preserve update correlations near 0.98, avoiding the linear output-size cost of reconstructing a full receiver-time gather.

Significance and Claims
The paper claims to provide the necessary physical-pressure derivative and selected-output measurements to connect Schrödingerised propagation to a local FWI update. It emphasizes that the energy scaling enabling the Hamiltonian construction simultaneously creates the coefficient dependence of pressure, necessitating the specific calibration term. The authors state that their work addresses the "coefficient-dependent physical-pressure derivative" and constructs compatible actions for a classically orchestrated local update.

The significance is framed as a foundational step: connecting Hamiltonian wave propagation to local FWI requires a consistent derivative of the physical receiver and the extraction of update observables at an accounted measurement cost. The authors are modest regarding broader implications, noting that runtime and asymptotic quantum speedup require separate scaling studies and that variable density, hardware-mapped circuits, and augmented PML formulations require further analysis. The work is presented as a finite-dimensional construction verified by discrete-action checks and a compiled small-instance realization, rather than a claim of immediate large-scale quantum advantage.

The Story of the "Calibrated Microphone"

More like this