⚛️ quantum physics

Quantum-Informed Portfolio Selection: An End-to-End Pipeline Validated on Trapped-Ion Hardware with Real Market Data

This paper presents an end-to-end pipeline using the hybrid qReduMIS algorithm, which leverages QAOA measurements to identify frozen nodes and guide classical reductions, successfully solving large-scale portfolio diversification problems on real market data using Quantinuum's trapped-ion hardware where standalone QAOA fails.

Original authors: Romina Yalovetzky, Martin J. A. Schuetz, Zichang He, Jiayu Shen, Yue Sun, Rudy Raymond, Shauna Sahay, Kishore Perla, Ruben S. Andrist, Grant Salton, Helmut G. Katzgraber, Roger Bongiovanni, Niraj Kuma

Published 2026-07-02

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Romina Yalovetzky, Martin J. A. Schuetz, Zichang He, Jiayu Shen, Yue Sun, Rudy Raymond, Shauna Sahay, Kishore Perla, Ruben S. Andrist, Grant Salton, Helmut G. Katzgraber, Roger Bongiovanni, Niraj Kumar, Rob Otter

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

The Big Picture: Building a "Super-Diverse" Investment Basket

Imagine you are a chef trying to create the ultimate salad. You have a huge basket of 225 different ingredients (stocks). Your goal is to pick as many ingredients as possible to make a big, delicious salad, but with one strict rule: no two ingredients can taste too similar.

If you pick two types of lettuce that taste exactly the same, you aren't really diversifying your salad; you're just doubling down on one flavor. In finance, this is called "correlation." If two stocks move up and down together, they are "correlated." The paper's goal is to find the largest possible group of stocks that are all unique from one another.

The Problem: A Maze Too Big to Solve Alone

Finding this perfect group is like trying to solve a massive maze.

The Old Way (Classical Computers): Traditional computers try to solve this by checking every possible combination. But as the number of ingredients grows, the number of combinations explodes. It's like trying to find a needle in a haystack that keeps getting bigger every second.
The Quantum Way (The New Tool): The researchers tried using a special "Quantum Computer" (specifically, a trapped-ion machine from Quantinuum) to solve this maze. However, just like a human getting tired, the quantum computer struggled. When the maze got too big (like the S&P 100 or Nikkei 225 indices), the quantum computer alone couldn't find the perfect solution. It kept getting lost.

The Solution: The "Hybrid Detective" (qReduMIS)

Instead of asking the quantum computer to solve the whole maze at once, the researchers created a new team-up method called qReduMIS. Think of it as a detective partnership between a Classical Computer (the logical planner) and a Quantum Computer (the intuitive guesser).

Here is how their "End-to-End Pipeline" works, step-by-step:

The Cleanup (Classical Step): First, the Classical Computer looks at the list of 225 stocks and removes the obvious ones that can't be part of the solution. It's like the chef saying, "We definitely don't need both chocolate and pickles," and setting those aside. This shrinks the problem down to a smaller, manageable group (a "kernel").
The Intuitive Guess (Quantum Step): The Classical Computer hands this smaller group to the Quantum Computer. Instead of asking the Quantum Computer, "What is the perfect answer?" (which it might fail at), they ask a different question: "Which ingredients are most likely to be in the perfect salad?"
- The Quantum Computer runs a quick experiment and gives a list of "hot tips." It says, "I'm 90% sure Ingredient #4 belongs in the salad, and I'm 90% sure Ingredient #7 does not."
- In the paper, these "hot tips" are called Frozen Nodes.
The Unblocking (The Magic): The Classical Computer takes these "hot tips" and locks them in place. If the Quantum Computer says "Ingredient #4 is a must," the Classical Computer locks it in and removes its "rivals" (stocks that taste too similar to #4).
Repeat: This process repeats. The Classical Computer shrinks the list, the Quantum Computer gives a few more "hot tips," and the list gets smaller and smaller until the perfect salad is found.

The Analogy: The "Finger in the Air" vs. The "Map"

Standalone Quantum (The Map): Imagine trying to drive across a country using a GPS that is broken and only gives you a 1% chance of getting the right turn. If you rely only on that GPS, you will likely get lost.
qReduMIS (The Finger in the Air): Instead of asking the broken GPS for the whole route, you ask it, "Is the next turn left or right?" It might be right 90% of the time. You take that hint, drive a bit, and ask again. By taking small, high-probability steps and using a human (the Classical Computer) to fill in the gaps, you successfully cross the country even though the GPS is imperfect.

What They Actually Found

The researchers tested this on real stock market data from four major indices (DAX, FTSE, S&P 100, and Nikkei 225) using a real quantum computer with 98 "qubits" (quantum bits).

The Failure of the Solo Act: When they let the Quantum Computer try to solve the biggest problems (S&P 100 and Nikkei 225) all by itself, it failed completely. It found the perfect solution 0% of the time.
The Success of the Team: When they used the qReduMIS team-up method:
- For the S&P 100, they found the perfect solution 40% of the time.
- For the Nikkei 225, they found the perfect solution 95% of the time.
- Across all tests, the solutions they found were at least 96% as good as the perfect theoretical solution.

The "Time" Factor

They also measured how long it took to solve these problems. They found that as the problems got bigger, the "Team" method (qReduMIS) got slower much more slowly than the "Solo" method (Standalone QAOA).

Analogy: If the Solo method is a runner who gets exhausted and slows down exponentially as the hill gets steeper, the Team method is a runner with a bicycle. Even on a steep hill, the bicycle keeps them moving at a steady pace. The paper claims the Team method is 3.2 times more efficient in how it scales with problem size.

Summary

This paper doesn't claim to have built a robot that can predict the stock market or make you rich. Instead, it proves a specific technical point: You don't need a perfect quantum computer to solve hard problems.

By using a quantum computer just to give "hints" (identifying which parts of the problem are "frozen" or certain) and letting a classical computer do the heavy lifting of organizing those hints, you can solve massive financial puzzles that a quantum computer alone cannot crack. It turns a "broken" quantum tool into a highly effective partner.

Technical Summary: Quantum-Informed Portfolio Selection via qReduMIS on Trapped-Ion Hardware

Problem Formulation
The paper addresses the Portfolio Selection Problem (PSP), a fundamental task in investment management aimed at constructing a diversified basket of assets to balance risk and return. The authors formalize PSP as a Maximum Independent Set (MIS) problem on an asset correlation graph. In this formulation:

Nodes represent financial assets.
Edges connect pairs of assets if their absolute correlation exceeds a specific threshold $\lambda$ .
The goal is to find the largest subset of nodes (assets) such that no two nodes are connected, ensuring the selected portfolio contains mutually uncorrelated assets.

The authors highlight that financial market graphs differ significantly from standard combinatorial optimization benchmarks. They are dense, non-local, and exhibit heterogeneous degree distributions, unlike the sparse, regular, or geometrically constrained graphs (e.g., unit-disk graphs) typically studied in quantum optimization literature. These structural properties pose challenges for both classical exact algorithms and current quantum approaches, particularly regarding scalability and the suppression of success probability as problem hardness increases.

Methodology: The qReduMIS Pipeline
The core contribution is an end-to-end hybrid quantum-classical pipeline leveraging the qReduMIS algorithm, adapted here for gate-based quantum hardware. Rather than using the quantum processor to directly output the final optimal solution, the pipeline uses quantum measurements to identify "frozen nodes"—vertices with a high marginal probability of belonging (or not belonging) to the optimal MIS.

The pipeline operates recursively through the following steps:

Classical Reduction: The input market graph undergoes exact classical reduction techniques (specifically, removing "exposed corner nodes" or simplicial vertices). This reduces the graph to a smaller "kernel" while maintaining sets of selected and removed nodes.
Quantum Subroutine (QAOA): The reduced kernel is mapped to a Quantum Approximate Optimization Algorithm (QAOA) circuit. The circuit is executed on a trapped-ion quantum processor (Quantinuum Helios).
Frozen Node Identification: Instead of searching for the optimal bitstring directly, the algorithm analyzes the marginal probability distribution of the nodes from the QAOA measurements. Nodes with high marginal probability (in-set) or low marginal probability (out-set) are identified as "frozen."
Recursive Unblocking: A frozen node is selected (using a semi-greedy strategy based on top-ranked marginals), fixed in the solution, and removed from the graph along with its neighbors. This "unblocks" the graph, allowing the classical reduction step to proceed further in the next iteration.
Termination: The process repeats until the kernel is fully reducible or a stopping criterion is met, yielding a final diversified portfolio.

Key Contributions
The paper makes three primary contributions:

End-to-End Hardware Demonstration: The authors deployed the full pipeline on Quantinuum's 98-qubit trapped-ion Helios system using real market data from four major indices (DAX, FTSE, S&P 100, Nikkei 225) with up to 225 assets. This represents one of the largest gate-based QAOA demonstrations on real-world combinatorial instances to date.
Algorithmic Extension to Gate-Based Devices: The work extends the qReduMIS framework, previously demonstrated on quantum annealing hardware, to gate-based QAOA. This adaptation is critical because market graphs possess non-local connectivity that exceeds the native embedding capabilities of many quantum annealers but is naturally suited to the all-to-all connectivity of trapped-ion processors.
Rigorous Benchmarking: The study provides a systematic comparison between standalone QAOA and qReduMIS (powered by QAOA) across multiple figures of merit, including success probability, approximation ratio, and optimal time-to-solution (optTTS) scaling.

Experimental Results
The experiments utilized QAOA circuits with $p=2$ layers acting on kernels of up to 78 qubits (for the Nikkei 225 instance, where the full graph would require 225 qubits, exceeding the hardware limit).

Success Probability: Standalone QAOA failed to find the optimal solution for the two largest indices (S&P 100 and Nikkei 225), achieving a success probability ( $P_{MIS}$ ) of 0. In contrast, qReduMIS achieved success probabilities of 0.40 and 0.95, respectively, for these indices.
Approximation Ratio: Across all four indices, qReduMIS maintained an average approximation ratio $\ge 0.96$ , whereas standalone QAOA ranged between 0.73 and 0.83.
Scaling Behavior: On a noisy emulator of the Quantinuum H2-1 device, the authors analyzed the scaling of the optimal time-to-solution (optTTS) against problem size. For $p=2$ , the scaling exponent for qReduMIS was found to be 3.2 times smaller than that of standalone QAOA ( $\beta_{qReduMIS} = 0.164$ vs. $\beta_{QAOA} = 0.524$ ).
Frozen Node Signal: Analysis of the DAX 100 kernel showed that even when the probability of sampling the optimal solution directly was low (<10%), approximately 96% of the marginal probability mass was concentrated on ground-truth frozen nodes. This confirms that the quantum hardware provides sufficient signal to guide the classical reduction even without directly outputting the optimal solution.

Significance and Claims
The paper posits that the primary significance of this work lies in demonstrating a fundamentally different use of quantum resources: leveraging quantum measurements to identify structural properties (frozen nodes) that facilitate classical reductions, rather than expecting the quantum processor to solve the entire optimization problem directly.

The authors claim that this "reduce-and-sample" paradigm allows the pipeline to:

Solve instances that exceed the qubit budget of current hardware (e.g., the 225-asset Nikkei 225 on a 98-qubit device) by operating on reduced kernels.
Outperform standalone quantum optimization on instances that can be encoded, significantly improving success rates and solution quality.
Provide a scalable path forward, as the framework benefits from increased circuit depth (demonstrated by improved performance at $p=6$ ) and hardware fidelity improvements.

The authors remain modest regarding the comparison with classical solvers, noting that for the current problem sizes accessible to quantum hardware, classical metaheuristics like Simulated Annealing (SA) solve most instances trivially. Therefore, the primary validation is a quantum-vs-quantum comparison, establishing that the hybrid approach extracts significantly more utility from limited quantum resources than standard quantum optimization alone. The work suggests that as hardware matures to handle larger, harder instances where classical heuristics struggle, the favorable scaling of qReduMIS may translate into a practical advantage.