⚛️ quantum physics

Quantum generative model on bicycle-sharing system and an application

This paper proposes a novel quantum machine learning model that uses quantum time evolution to analyze bicycle-sharing trends and simulate how proactive redistribution of bicycles can optimize overall system rental numbers.

Original authors: Fumio Nemoto, Nobuyuki Koike, Daichi Sato, Yuuta Kawaai, Masayuki Ohzeki

Published 2026-02-10

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Fumio Nemoto, Nobuyuki Koike, Daichi Sato, Yuuta Kawaai, Masayuki Ohzeki

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 "Quantum Bicycle Predictor": Solving the City’s Bike Shortage

Imagine you’re in a rush to get to work. You walk to the nearest bike-sharing station, only to find it completely empty. You walk to the next one—also empty. Meanwhile, a few blocks away, a station is overflowing with bikes that nobody is using. This "musical chairs" problem is a headache for cities, and it’s exactly what this research aims to solve.

Here is a breakdown of how these researchers are using the strange, "spooky" world of quantum physics to fix a very real-world problem.

1. The Problem: The "Ghost" Demand

In a city, bike demand isn't random; it follows a rhythm. In the morning, bikes "flow" from residential neighborhoods to office districts. In the evening, they flow back.

Traditional computer models try to predict this by looking at history, but they often struggle to see the connections between different locations. They might see that Station A is emptying, but they don't always realize that Station B is about to fill up because of it. It’s like trying to predict a wave in the ocean by looking at only one tiny drop of water at a time.

2. The Solution: The Quantum "Dance"

The researchers decided to stop treating bike stations like isolated dots on a map and started treating them like particles in a quantum dance.

In quantum physics, particles can be "entangled," meaning what happens to one instantly affects the other, no matter how far apart they are. The researchers built a Quantum Generative Model. Instead of just calculating numbers, they designed a "quantum circuit" (a mathematical recipe) that mimics how quantum particles evolve over time.

The Analogy: The Synchronized Swimmers
Think of the bike stations as a team of synchronized swimmers in a pool.

Classical models are like watching each swimmer individually and trying to guess where they will be in ten seconds based on their current speed.
This Quantum model looks at the entire choreography. It understands that if Swimmer A moves left, Swimmer B must move right to keep the pattern. By learning the "choreography" of the city (the correlations between stations), the model can predict not just where one bike will be, but how the entire "dance" of the city will unfold.

3. How It Works (The "Secret Sauce")

To make this work, they used a few clever tricks:

Simplification (SAX): They didn't try to track every single bike. Instead, they simplified the data into "states"—basically asking, "Is the number of bikes going up or down?" It’s like describing a person’s mood as just "Happy" or "Sad" rather than measuring every single chemical in their brain. This makes the math much faster.
The Cost Function: They trained the model by giving it a "grade." If the model's prediction didn't match the actual patterns of how bikes move between residential and office areas, it got a bad grade and had to adjust its "dance moves" (parameters) until it got it right.

4. The "What If?" Machine (The Simulation)

The coolest part of this paper isn't just predicting the future; it’s simulating a better one.

Because the model is "generative," it can act like a high-tech flight simulator. The researchers asked: "What if we proactively moved 100 extra bikes to the residential area at 6:00 AM?"

The model ran thousands of "what-if" scenarios (a Monte Carlo simulation) and found:

The Primary Effect: Those 100 extra bikes prevented shortages in the morning, allowing more people to rent them.
The Secondary Effect: Because those bikes were eventually ridden to the office area, they actually helped prevent a shortage in the office area later in the evening!

It’s a ripple effect. By adding bikes in one place, you create a positive wave that travels through the city.

5. Why Does This Matter?

The researchers proved that their quantum model was actually more accurate at capturing these complex city rhythms than some of the most famous classical AI models (like LSTMs), and it did so using way fewer "brain cells" (parameters).

While we don't have massive quantum computers running our city transit yet, this research shows that the logic of quantum physics is a powerful new tool for managing the complex, moving parts of our modern world. It turns "guessing" into "choreography."

Technical Summary: Quantum Generative Model on Bicycle-sharing System and an Application

Journal: Journal of the Physical Society of Japan 95, 024003 (2026)
Authors: Fumio Nemoto, Nobuyuki Koike, Daichi Sato, Yuuta Kawaai, and Masayuki Ohzeki

1. Problem Statement

Bicycle-sharing systems face significant operational challenges due to non-stationary demand. Specifically, intensive commuting patterns lead to localized bicycle shortages in residential areas during morning hours and in office areas during evening hours. Traditional classical models, such as regression, SARIMA, or Graph Convolutional Neural Networks (GCNNs), often struggle to balance computational complexity with the ability to model dynamic, multi-dimensional inter-port correlations (spatial dependencies) effectively.

2. Methodology

The authors propose a novel Quantum Machine Learning (QML) approach using a quantum generative model to capture and simulate these complex dynamics.

Data Discretization (SAX): To bridge the gap between continuous bicycle counts and quantum states, the authors use the Symbolic Aggregate Approximation (SAX) method. They discretize hourly increments into $N=2$ states (e.g., "decreasing/shortage risk" vs. "increasing/return flow").
Quantum Circuit Architecture: The model utilizes a parameterized quantum circuit designed to emulate quantum time evolution.
- Target & Ancilla Qubits: Target qubits represent the states of different port groups, while ancilla qubits enable entanglement to model inter-port correlations.
- Unitary Decomposition: The time-evolution operator is decomposed as $U(\vec{\theta}, t) = V^\dagger(\vec{\theta}_1) W(\vec{\theta}_2 t) V(\vec{\theta}_1)$ . Here, $V$ introduces entanglement (spatial correlation), and $W$ is a diagonal matrix representing the Hamiltonian's eigenspace, allowing for efficient time-scaling.
Correlation-Aware Cost Function: A unique aspect of this work is the composite cost function $C(\vec{\theta})$ $C (θ)$ :
1. KL Divergence Term: Minimizes the difference between the quantum circuit's predicted probability transition matrices and the observed empirical matrices.
2. Correlation Penalty Term: A squared error term that explicitly penalizes discrepancies between the correlation coefficients ( $\rho$ ) of the quantum model and the actual data across different port pairs.
Monte Carlo Simulation: Once trained, the model acts as a generative engine. By repeatedly sampling the circuit, the authors reconstruct full-day trajectories and perform counterfactual simulations.

3. Key Contributions

Integration of Quantum Dynamics with Mobility Data: The study successfully maps discrete, multi-dimensional time-series data onto a quantum evolution framework.
Explicit Correlation Modeling: Unlike standard generative models that may only capture marginal distributions, this model incorporates a specific penalty term to ensure inter-port dependencies (e.g., the flow from residential to office areas) are accurately captured.
Counterfactual Decision Support: The model is not just predictive but prescriptive; it allows operators to simulate "what-if" scenarios, such as the impact of proactively rebalancing bicycles.

4. Results

The model was tested using real-world data from the DATE BIKE system in Sendai City, categorized into "Residential," "Office," and "Others" groups.

Performance vs. Classical Baselines:
- Accuracy: The quantum model achieved a lower Mean Absolute Error (MAE = 4.8) compared to SARIMA (17.1) and LSTM (56.8).
- Correlation Recovery: The model successfully reproduced the qualitative "commuter flow" patterns (e.g., negative correlation between Residential and Office groups during peak hours), which the SARIMA model failed to capture.
- Efficiency: The quantum model used significantly fewer parameters (60) than the LSTM (10,953).
Counterfactual Simulation (The "Pre-adding" Effect):
- The authors simulated adding 100 bicycles to residential ports at 6:00 AM.
- Primary Effect: The simulation showed an increase of 88 additional rentals in the residential area by mitigating shortages.
- Secondary Effect: The arrival of these bicycles in office areas later in the day created an additional 8 rentals by alleviating evening shortages.
- Total System Benefit: A total increase of 96 rentals was estimated.

5. Significance

This research demonstrates that quantum generative models can serve as a powerful, parameter-efficient alternative to deep learning for complex mobility systems. By leveraging the intrinsic probabilistic and entangling nature of quantum mechanics, the model provides a robust tool for urban planners to optimize resource allocation and mitigate service shortages in real-time, paving the way for future applications on actual quantum hardware.