Quantum Observers: A NISQ Hardware Demonstration of Chaotic State Prediction Using Quantum Echo-state Networks

This contribution presents a novel design of a Quantum Echo State Network (QESN) that successfully predicts long time series from a chaotic Lorenz system on noisy IBM quantum hardware, thereby demonstrating persistent memory capabilities that exceed the median coherence times of the QPU by more than a factor of 100.

The Gist

The Big Picture: A Quantum Crystal Ball for Chaos

Imagine you are trying to predict the weather. It is a chaotic system where a tiny change today (like the flapping of a butterfly's wing) can lead to a massive storm weeks later. This is the Lorenz System, a famous mathematical model for chaos that the researchers used as a test subject.

Normally, predicting such chaotic systems requires massive classical computers. But this team asked: Can we use a quantum computer for this, even though current quantum computers are noisy and fragile?

Their answer is yes. They built a "Quantum Observer"—a virtual sensor that can observe one part of a chaotic system (like wind speed) and figure out what the other invisible parts (like temperature and pressure) are doing, even on today's imperfect quantum hardware.

The Problem: The "Fragile Glass" of Quantum Computers

Imagine current quantum computers (so-called NISQ devices) as a house of cards made of glass. They are incredibly powerful, but they are also:

1. Noisy: Like trying to hear a whisper at a rock concert.
2. Fragile: The "cards" (qubits) decay very quickly (decoherence). If you try to run a long calculation, the house collapses before you finish.

Previous attempts to use quantum computers for time-series prediction often had to stop every few seconds, reset, and start over because the "house" collapsed. This paper solves the problem by building a structure that can run for a very long time without collapsing.

The Solution: The Quantum Echo-State Network (QESN)

The researchers developed a new design called the Quantum Echo-State Network (QESN). Here is how it works, using an analogy:

1. The "Echo" Room (The Reservoir)

Imagine a large, empty room with strangely shaped walls (the quantum circuit). You shout a sound into the room (input data). Because of the strange walls, the sound bounces around and creates a complex "echo" that mixes the new shout with the echoes of previous shouts.

• In the paper: This is the "Reservoir." It takes in a stream of data and lets it bounce around inside the quantum system. This creates a rich, complex pattern that remembers previous inputs. This is the "memory."

2. The "Sparsity" Trick (Reducing Noise)

Normally, for a quantum computer to work, every qubit must be connected to every other qubit. But that creates too much noise and too many errors.

• The Analogy: Imagine a crowded dance floor where everyone is holding hands. If one person trips, everyone falls.
• The Solution: The researchers decided to let go of most hands. They let only a few people hold hands (this is called Sparsity or "Sparseness").
• The Result: By removing about 50% of the connections, they reduced the probability of errors and let the circuit run faster without losing the ability to remember the past.

3. The "Re-Upload" (Keeping the Beat)

To keep the memory alive, the system doesn't just shout once. It keeps shouting new data into the echo room while the old echoes are still bouncing around.

• The Analogy: It is like a DJ mixing a new track into a song that is still playing. The new track blends with the old one and creates a continuous, evolving sound.
• The Term in the Paper: This is called Data Re-uploading. It allows the quantum computer to process a long stream of data without interruption.

4. The "Reset" (The Magic Trick)

Here comes the cleverest part. In a normal quantum computer, the "magic" disappears when you look at the qubits (measure them), and the calculation stops.

• The Analogy: Imagine a magician performing a trick. If you look at the cards, the trick fails.
• The Solution: The researchers built a system where they only look at half of the qubits (the "readout" qubits) to get the answer, and immediately reset those specific qubits to zero, while the other half (the "memory" qubits) keeps the echo running.
• The Result: They can let the show run for a very long time without the entire system collapsing.

The Record-Breaking Run

The team tested this on a real quantum computer from IBM (the ibm_marrakesh).

• The Challenge: Quantum bits usually last only about 200 microseconds before losing their "quantum nature" (this is called T1/T2 coherence time).
• The Achievement: Their circuit ran for 48,000 microseconds.
• The Metaphor: It is like a runner who can usually only sprint for 2 seconds before collapsing. This team trained their runner so that they could sprint for 100 seconds without interruption. They let the circuit run 100 times longer than the hardware should have held up.

The Results: Predicting the Unpredictable

They fed the system data from the chaotic Lorenz System (only the "X" coordinate). The goal was to predict the "Y" and "Z" coordinates that the system could not see.

• The Result: The Quantum Observer successfully predicted the hidden parts of the chaotic system.
• The Comparison: They compared it to a standard model on a classical computer. The quantum version performed slightly better in simulations and was very competitive on the noisy real hardware. This proves that quantum computers can handle complex tasks with long-term memory, even when they are imperfect.

Summary

This paper shows that we do not need perfect, futuristic quantum computers to do useful work today. Through an intelligent design that:

1. Uses Echoes to remember the past,
2. Cuts unnecessary connections (Sparsity) to reduce errors, and
3. Measures and resets parts of the system while it runs,

...we can build a "Quantum Observer" that can observe chaotic systems and predict their future behavior much longer than anyone thought possible on current hardware. It is proof that quantum machines can already be useful tools for complex predictions, not just in the distant future.

Technical Summary

Technical Summary: Quantum Observers: A NISQ Hardware Demonstration of Chaotic State Prediction via Quantum Echo-State Networks

Problem Statement
While artificial intelligence has demonstrated remarkable capabilities on classical hardware, scalability and efficiency remain constrained by computational limits. Quantum computers offer the potential to overcome these limits; however, integrating quantum computing with neural networks (NNs) on current Noisy Intermediate-Scale Quantum (NISQ) hardware is hindered by decoherence, noise, high error rates, and short coherence times (T1 and T2). Existing approaches for Quantum Reservoir Computing (QRC) often rely on specialized analog devices, require circuit re-initialization or interruption (leading to "Buffer Overflow" errors), or lack persistent memory. Furthermore, transferring classical reservoir concepts—such as fading memory, asymptotic stability (the "echo-state" property), and nonlinearity—into universal gate-based quantum circuits poses a significant challenge due to the unitary evolution of closed systems and the collapse of state measurement.

Methodology
The authors propose a scalable Quantum Echo-State Network (QESN) designed as a "quantum observer" capable of reconstructing latent state variables of a dynamical system from partial measurements. The architecture is specifically developed for universal gate-based NISQ hardware (particularly IBM's superconducting qubits) and avoids the need for circuit resets or interruptions during the execution of long time series.

Key methodological components include:

• Circuit Architecture: The QESN employs a "measure-and-reset" paradigm, where readout qubits are measured and deterministically reset to $|0\rangle$ at each time step, while memory qubits retain their state. This enables continuous evolution over extended periods.
• Input Encoding: Classical data are embedded onto the Bloch sphere using "angle embeddings" (rotations $R_z(\gamma)R_x(\beta)R_z(\alpha)$). A "Context Window" approach maps arbitrary input sizes onto these rotation angles via a fully connected classical layer ($W_{in}$), thereby decoupling reservoir size from input dimension.
• Sparsity and Entanglement: To mitigate gate errors and reduce circuit depth (minimizing SWAP operations), the authors introduce sparsity by randomly deleting a subset of two-qubit entanglement gates. The circuit utilizes an adjustable "Data Re-uploading" block, where input-dependent rotations are repeated ($n_c$ times) between entanglement gates to increase nonlinearity.
• Response Analysis: The authors extend classical time-domain response analysis from control theory to the quantum domain. By analyzing the system response to step, ramp, and sinusoidal inputs, they characterize the memory capacity, nonlinearity, and stability of the QESN. This analysis guides the selection of sparsity levels and the number of re-uploading blocks.
• Training and Optimization: The quantum circuit generates a high-dimensional feature space (probability distributions over the Hilbert space). These features are fed into a classical least-squares regression model with Elastic-Net regularization to predict target values.

Main Contributions

1. Novel QESN Design: A universal gate-based QESN circuit that operates continuously on NISQ hardware without intermediate interruptions or full system resets, utilizing a measure-and-reset scheme for readout qubits.
2. Response Analysis Framework: An extension of classical control theory to quantum circuits for systematically tuning sparsity and nonlinearity (via re-uploading blocks), demonstrating that significant sparsity (up to ~50% gate removal) can be achieved without compromising the echo-state property.
3. Hardware Validation: A comprehensive demonstration of the QESN as a quantum observer on the $ibm\_marrakesh$ QPU, successfully predicting the chaotic dynamics of a Lorenz system.

Results

• Simulation: Simulations on the Aer simulator (using an $ibm\_fez$ noise model) showed that using the full probability distribution as features yielded a lower Root Mean Squared Error (RMSE) than using expectation values alone. The QESN demonstrated superior performance compared to classical ESNs in noise-free simulations with an equal number of reservoir nodes, although the authors note this is a modest gain.
• Hardware Execution: The QESN was executed on the $ibm\_marrakesh$ QPU with 12 qubits to predict the $y(t+1)$ and $z(t+1)$ coordinates of a Lorenz system based solely on the $x(t)$ input.
  • The circuit ran coherently for a duration of approximately 48,000 $\mu$s, which is over 100 times longer than the median coherence times T1 (213.92 $\mu$s) and T2 (119.57 $\mu$s) of the qubits.
  • The experiment successfully processed a time series of 2,000 steps, significantly surpassing previous demonstrations (e.g., 100 steps in [15], 20 steps in [14]).
  • The hardware results achieved a test RMSE of 0.0922 and closely matched expectations from a noisy simulation.
  • The system successfully mitigated "Buffer Overflow" errors by introducing a 24 $\mu$s delay after measurement, enabling longer sequential executions.

Significance and Claims
The work claims to demonstrate the longest time-series prediction algorithm ever executed on gate-based NISQ hardware to date. The authors emphasize that their results prove current NISQ computers are capable of maintaining coherent memory and performing prediction tasks over periods far exceeding the coherence times of individual qubits.

The work positions the QESN as a viable "quantum observer" for dynamical systems and highlights that:

• Sparsity is a critical design parameter for NISQ optimization, reducing error rates while preserving memory and nonlinearity.
• The measure-and-reset paradigm, combined with data re-uploading, enables persistent memory and stable operation without circuit re-initialization.
• Although the current performance gain over classical methods is modest, the exponential growth of the feature space provided by quantum computers promises significant future advantages as hardware improves.

The authors explicitly clarify that they do not claim a definitive "quantum advantage" at this stage, but rather demonstrate the practical feasibility of quantum machine learning approaches on current hardware for complex, chaotic dynamical systems.