💻 computer science

Quantum Optimization for Electromagnetics: Physics-Informed QAOA for Reconfigurable Intelligent Surfaces

This paper demonstrates that while embedding mutual coupling physics into Quantum Approximate Optimization Algorithm (QAOA) formulations for Reconfigurable Intelligent Surfaces (RIS) improves beamforming precision, sparse, distance-penalized models remain the necessary compromise for achieving feasible execution on current noisy intermediate-scale quantum (NISQ) hardware due to the prohibitive routing overhead of fully dense Hamiltonians.

Original authors: Marco Pasquale, Erik M. Åsgrim, Stefano Markidis, Oscar Quevedo-Teruel

Published 2026-07-20

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Marco Pasquale, Erik M. Åsgrim, Stefano Markidis, Oscar Quevedo-Teruel

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 direct a massive crowd of people to stand in a specific formation to create a giant, invisible spotlight. This is the job of a Reconfigurable Intelligent Surface (RIS). Think of an RIS as a high-tech, digital mirror made of thousands of tiny tiles. Each tile can instantly change how it bounces a radio wave, allowing engineers to steer signals around corners, through walls, or straight to a specific phone without needing a new tower. This technology is the secret sauce being developed for the future of 6G internet, promising to make connections faster and more reliable everywhere.

However, there's a catch. In the real world, these tiles are packed so tightly together that they don't act like independent mirrors. Instead, they "talk" to each other through invisible electromagnetic whispers called mutual coupling. When one tile changes its phase, it slightly disturbs its neighbors, making the whole system behave in a messy, unpredictable way. Solving the puzzle of how to set every single tile to get the perfect signal is a nightmare for traditional computers because the number of possible combinations is astronomically huge. This is where Quantum Computing enters the story. Quantum computers are special machines that can explore many possibilities at once, making them theoretically perfect for solving these massive puzzles. But there's a problem: current quantum computers are still small and noisy, and they struggle with complex, messy math.

This paper explores a fascinating middle ground: Can we teach a quantum computer to solve the RIS puzzle by feeding it a version of the problem that includes just enough "real-world physics" to be accurate, but not so much that the quantum computer chokes? The researchers, working at KTH Royal Institute of Technology, didn't just assume the tiles were perfect; they built four different mathematical models to see how much "physics" a quantum algorithm could handle before it broke.

The Digital Mirror and the Quantum Puzzle

The story begins with a simple question: How do we tell a 5x5 grid of 25 tiny mirror tiles exactly how to tilt their "heads" (their phase) to bounce a radio wave from a starting point to a target destination?

In a perfect, dream-world scenario, you could just calculate the ideal angle for each tile and be done. But in reality, the tiles are neighbors. If you nudge one, the others feel it. This is mutual coupling. If you ignore this, your quantum computer might give you a solution that looks perfect on paper but fails miserably in the real world, like a choir singing perfectly in a vacuum but sounding terrible in a noisy stadium.

The authors set up a simulation to test this. They used a Quantum Approximate Optimization Algorithm (QAOA). Think of QAOA as a very smart, very fast trial-and-error machine. It tries different combinations of tile settings, checks how good the resulting signal is, and slowly learns the best pattern. But to do this, the problem has to be translated into a language the quantum computer understands, called a QUBO (Quadratic Unconstrained Binary Optimization) problem. It's like translating a complex novel into a simple code of zeros and ones.

The Four Models: From "Dream World" to "Real World"

The researchers tested four different ways to translate the physics of the mirror tiles into this code. They wanted to see which translation worked best on their simulated quantum computer.

The "Ideal" Model (Model 1): This model assumes the tiles are perfect strangers. It ignores the fact that they are neighbors and interact. It's the simplest math, but the authors found it physically flawed. It's like telling a choir to ignore the acoustics of the room.
The "Distance Penalty" Model (Model 2): This model adds a simple rule: "The closer you are to a neighbor, the more you affect them." It's a rough approximation, like saying "neighbors are loud," without knowing exactly how loud.
The "Spherical Wave" Model (Model 3): This is a step up. It uses a more realistic formula that accounts for how waves ripple and fade over distance, including the fact that sometimes neighbors cancel each other out and sometimes they boost each other. It's still a simplification, but it's much closer to reality.
The "Full Physics" Model (Model 4): This is the heavy hitter. It tries to include every possible interaction between every single tile, using the full, complex math of electromagnetic waves. It is the most accurate description of reality, but it creates a massive, dense web of connections that is incredibly hard for a quantum computer to navigate.

The Surprising Results

When they ran the simulations, some interesting things happened.

First, the quantum computer (QAOA) was surprisingly good at finding the best answers, even for the most complex models. For the "Full Physics" model (Model 4), which had the most complicated math, the quantum algorithm still managed to find the optimal solution very quickly. This suggests that even if the problem is messy and dense, the quantum approach can handle it in theory.

However, here is the twist: The most accurate model was the hardest to run on real hardware.

The "Full Physics" model (Model 4) required the quantum computer to connect every single "qubit" (the quantum equivalent of a bit) to every other qubit. Imagine trying to hold hands with everyone in a stadium at the same time; it's physically impossible without a lot of extra people helping you pass the message. In quantum terms, this requires a huge amount of "routing" overhead, which current noisy computers can't handle well.

On the other hand, the simpler models (Model 2 and Model 3) only required connections between nearby neighbors. This is like a game of "telephone" where you only talk to the person next to you. This "sparse" structure fits perfectly on current quantum chips.

The Trade-Off: Accuracy vs. Feasibility

The paper's main discovery is a critical trade-off.

If you want the most accurate physics (Model 4), you get the best theoretical beam steering, but the quantum computer needs to be much more powerful than what exists today to run it.
If you want to run it on current technology (NISQ devices), you have to use the "sparse" models (Model 2 or 3). These models ignore some of the distant interactions, but they are "good enough" and can actually be executed on today's machines.

Interestingly, the researchers found that sometimes the "sub-optimal" solution (a solution that wasn't mathematically perfect for the simplified model) actually performed better in the real-world simulation than the "perfect" solution for the simplified model. This highlights a weird mismatch: what looks like the best answer in the quantum code doesn't always translate to the best physical result.

The Bottom Line

The authors conclude that while we can mathematically describe the full, messy physics of these intelligent mirrors, we can't yet run that full description on our current quantum computers. The "Full Physics" model is too heavy.

Instead, the best path forward for the near future is to use sparse, distance-penalized models. These models strike a balance: they include enough physics to be useful (accounting for the fact that neighbors affect each other) but leave out the distant, complex interactions that would overwhelm the hardware.

So, while the dream of a quantum computer perfectly simulating a giant, physics-perfect mirror is still a bit away, this paper shows us a practical stepping stone. By using "physics-informed" but simplified models, we can start optimizing these 6G mirrors today, paving the way for a future where our internet signals can be steered with the precision of a laser, all thanks to a little help from quantum mechanics.

Technical Summary: Quantum Optimization for Electromagnetics

Problem Statement
Reconfigurable Intelligent Surfaces (RIS) are critical for future 6G communication systems, requiring the optimization of discrete phase states across an array to steer electromagnetic waves. This task is a high-dimensional combinatorial optimization problem. Current quantum approaches, particularly those utilizing the Quantum Approximate Optimization Algorithm (QAOA), often rely on simplified models that treat RIS elements as independent radiators. These models neglect physical constraints such as mutual coupling, which arises from sub-wavelength element spacing ( $d \leq \lambda/2$ ). The neglect of these electromagnetic effects leads to discrepancies between optimized configurations and actual engineering performance, degrading real-world beamforming accuracy.

Methodology
The authors propose a framework to embed progressively more physics-informed models of mutual coupling into Quadratic Unconstrained Binary Optimization (QUBO) formulations suitable for QAOA. The study focuses on a $5 \times 5$ RIS grid operating at 30 GHz.

QUBO Formulation: The RIS beam-steering objective is mapped to a QUBO problem where binary variables $x \in \{-1, +1\}$ represent the phase states of the elements. The cost Hamiltonian $H_C$ is constructed using an Ising interaction matrix $J_{ij}$ .
Four Interaction Models: The paper evaluates four distinct models for constructing the $J_{ij}$ $J_{ij}$ matrix, varying in physical fidelity and quantum hardware feasibility:
- Model 1 (Ideal Dense Phase Coupling): Based purely on idealized phase differences ( $\cos(\Delta\phi_{ij})$ ). It ignores mutual coupling entirely, resulting in a fully dense Hamiltonian.
- Model 2 (Sparse Distance-Penalized Coupling): Adds a generic inverse-distance penalty ( $k/d_{ij}$ ) with a spatial cutoff. It approximates coupling decay but lacks rigorous wave physics. It yields a sparse matrix.
- Model 3 (Sparse Spherical Wave Coupling): Uses the real part of a spherical wave function ( $\alpha \cos(k d_{ij})/d_{ij}$ ) with a spatial cutoff. This incorporates oscillating physical coupling behavior while maintaining sparsity.
- Model 4 (Dense Corrected QUBO Mapping): Derives interactions directly from the far-field objective using the full complex scalar free-space Green's function. This captures both active and reactive near-field power flows, resulting in a fully dense matrix that mathematically pre-compensates for global coupling.
Optimization and Validation: The QAOA circuit is simulated using a custom state-vector simulator (PyTorch) with the Adam optimizer. The optimization minimizes the energy expectation value of the cost Hamiltonian. Crucially, the final performance is validated not just by the QUBO objective, but by simulating the scattered far-field radiation pattern using a fully coupled electromagnetic model to calculate the true angular pointing error ( $\epsilon$ ).

Key Results

Convergence and Approximation Ratio: All four models achieved comparable Approximation Ratios (AR) ranging from 0.79 to 0.90 after 500 optimization steps. Notably, Model 4, despite inducing a highly complex and dense optimization landscape, converged rapidly to high-quality solutions, demonstrating that detailed physical models do not inherently prevent QAOA convergence.
Physical vs. Mathematical Optimality: A significant finding is the mismatch between QUBO optimality and physical accuracy. For Model 1, the most probable bitstring found by QAOA was not the global optimum of the QUBO objective, yet it yielded a lower angular error ( $\epsilon = 2.23^\circ$ ) than the mathematically optimal bitstring ( $\epsilon = 3.66^\circ$ ). This highlights that optimizing the simplified QUBO objective does not always directly translate to improved physical beamforming performance.
Hardware Feasibility Trade-off: Models 1 and 4 require all-to-all qubit connectivity (300 interactions for 25 qubits), incurring prohibitive routing overhead on current Noisy Intermediate-Scale Quantum (NISQ) devices. In contrast, Models 2 and 3 utilize spatial cutoffs to create sparse interaction graphs (approx. 40 interactions), which are naturally compatible with planar hardware topologies.
Performance: While Model 4 offers the most rigorous physical description, the sparse models (2 and 3) offer a necessary compromise. They retain sufficient physical insight to outperform idealized models while remaining executable on near-term hardware without excessive gate overhead.

Significance and Claims
The paper claims to demonstrate that physics-informed quantum optimization is mathematically viable for RIS design. It establishes that:

Feasibility of Dense Models: Even fully dense, physically rigorous Hamiltonians can be effectively encoded and optimized within the QUBO framework, achieving convergence comparable to simpler models.
Necessity of Compromise: While dense models maximize theoretical beamforming precision, sparse, distance-penalized models are a necessary compromise for execution on current NISQ devices due to routing constraints.
Limitations of Current Approaches: The work underscores that "toy" models ignoring mutual coupling are insufficient for accurate engineering, yet fully realistic models face hardware barriers. The proposed framework bridges this gap by offering a spectrum of models that balance physical fidelity with quantum hardware constraints.

The authors conclude that while the approach is promising, substantial steps remain before these methods become practical design tools for large-scale 6G RIS systems, particularly regarding scalability and the handling of noise on real hardware.