HackYeah

Better descriptions available on Github Readme!!

Inspiration

We started with a simple question: can quantum hardware do more than solve toy problems? Portfolio optimization is a real problem in finance, but many classical approaches rely on manually tuning penalty terms and balancing returns, risk, and constraints by hand. That felt limited to us. We wanted to see whether some of those constraints, especially diversification, could be enforced more naturally by physics itself. That led us to Rydberg atoms. Their interactions make certain combinations physically hard to select together, which felt surprisingly similar to how highly correlated assets should not always appear together in a good portfolio. That connection between finance and physics became the foundation of our project.

What it does

Our project builds a quantum inspired portfolio optimization framework using Rydberg atom dynamics. We start with real financial structure, including expected return, correlation, and risk adjusted signals like Sharpe ratio. Then we encode that structure into a physical system. Asset correlations become distances between atoms, diversification constraints emerge through Rydberg blockade, and asset preference is influenced through local detuning. Instead of only solving the problem as a static optimization equation, we simulate the quantum system and sample bitstrings, where each bitstring represents a candidate portfolio. In that way, the output is not just one abstract answer, but a distribution of feasible portfolios shaped by both finance and physics.

How we built it

We first wrote the portfolio selection problem in QUBO form, then converted it into an Ising style representation so it could connect naturally to quantum hardware. From there, we built the project step by step. We clustered assets based on correlation structure, selected representative assets from those clusters, mapped correlation strength to physical distance between neutral atoms, used the Rydberg interaction term to penalize selecting strongly related assets together, and designed pulse schedules using Rabi amplitude and detuning. We also tested multiple detuning strategies, including uniform, return weighted, and Sharpe weighted versions. Finally, we compared portfolio bitstrings, risk return tradeoffs, and noise robustness across different setups. We also modeled realistic imperfections such as readout bit flip errors and decoherence like mixing to check whether the signal survives beyond an ideal simulation.

Challenges we ran into

One of the hardest parts was not coding, but understanding the mapping itself. At first, it was much easier to think in finance language than in physics language. We had to work through what each object really meant, including what the QUBO matrix was encoding, how Ising variables relate to binary portfolio decisions, what detuning actually biases, and how blockade differs from a hand tuned penalty. Another challenge was deciding how much finance should remain explicit in the model. We did not want this to become just a physics demo, but we also did not want to fall back into a fully classical approach. Finding the right balance between geometry based physical constraints and finance informed signals like Sharpe weighted detuning took a lot of iteration. Noise was another major challenge. Real hardware is imperfect, so it was important to test whether our best portfolio bitstrings still remained meaningful under noisy measurement and decoherence effects.

Accomplishments that we're proud of

We are proud that we built a full pipeline from a real financial optimization problem to a physically meaningful quantum model. Rather than stopping at a classical QUBO formulation, we were able to translate correlation structure into atom geometry, use blockade as a physical diversification mechanism, encode return and Sharpe style preferences through local detuning, generate portfolio candidates through quantum style sampling rather than only deterministic optimization, and show that the strongest portfolio signals remain visible even under noise. What feels most meaningful to us is that the project is not just finance plus quantum in a superficial way. The physics is actually doing something important. It is shaping the feasible set and influencing which portfolios naturally emerge.

What we learned

We learned that quantum optimization is not only about speed. It is also about representation. This project taught us that the way a problem is encoded matters just as much as the solver itself. In classical finance, constraints are often added as penalty terms that must be tuned carefully. In our approach, some of those same constraints can emerge from the hardware structure itself. We also learned that hybrid thinking matters. Our final direction was not to abandon finance metrics, but to encode them differently. Correlation enters through geometry, while return or Sharpe preference enters through detuning. That helped us see that quantum methods do not have to replace financial reasoning. They can implement it in a new way. Most of all, we learned how difficult but rewarding it is to connect ideas across fields. Translating from portfolio theory to QUBO, from QUBO to Ising, and from Ising to Rydberg dynamics forced us to understand each step much more deeply.

What's next for HackYeah

There are several directions we want to explore next. First, we want to scale the framework beyond a small representative subset of assets and test whether the same ideas still hold for larger and more realistic portfolios. Second, we want to improve the way financial signals are encoded. That includes trying richer risk adjusted measures, more dynamic market inputs, and better ways of combining expected return with physical constraints. Third, we want to move closer to real hardware. While simulation helped us understand the system, testing on real neutral atom platforms would make the work much more meaningful. More broadly, we want to keep exploring whether structured financial problems can be solved not just by adding more computation, but by designing physical systems whose natural behavior already reflects the structure of the problem.

Built With

• matplotlib
• numpy
• pandas
• python
• qubo/ising-modeling
• rydberg-atom-simulation-(neutral-atom-quantum-systems)
• scikit-learn
• scipy