Non-Stationary Decoherence in Superconducting Qubits: Memory Multi-Fractional Brownian Motion and a Time-Dependent Quantum Brownian Motion Extension

This paper proposes a unified stochastic drift model for superconducting charge qubits based on memory multi-fractional Brownian motion and a time-dependent Caldeira–Leggett environment, which accurately captures non-stationary 1/f noise and long-range correlations to predict coherence times and non-Markovian decay patterns that surpass the limitations of conventional Markovian approaches.

The Gist

Imagine a superconducting qubit (the basic building block of a quantum computer) as a very delicate, spinning top. In a perfect world, this top would spin forever without slowing down. But in the real world, it's sitting in a noisy room. The air currents, vibrations, and temperature changes in that room push the top, causing it to wobble and eventually fall over. This falling-over process is called decoherence, and it's the biggest enemy of quantum computers.

For a long time, scientists thought the "noise" in the room was like white noise—random static that changes instantly and forgets everything immediately. They thought if the top wobbled now, it had no memory of wobbling five seconds ago.

This paper argues that the noise is actually much more complex. It's not just random static; it's a memory. The noise "remembers" what happened in the past, and that memory changes over time.

Here is a breakdown of the paper's main ideas using simple analogies:

1. The "Walking in a Fog" Analogy (The Noise Model)

The authors propose a new way to describe this noise called Memory Multi-Fractional Brownian Motion (mmfBm).

• Old View (Standard Models): Imagine walking in a fog where the wind blows randomly every second. If you stumble today, it has nothing to do with how you walked yesterday. The wind is "stationary" (it doesn't change its nature).
• New View (This Paper): Imagine walking in a fog where the wind is lazy and forgetful, but also drifting.
  • Memory: If the wind pushes you hard today, it's likely to push you hard tomorrow too. The noise has "long-range memory."
  • Drifting (Non-Stationary): The "personality" of the wind changes over time. Sometimes the wind is gentle and predictable; other times it's chaotic and wild. The paper introduces a "Hurst exponent" ($H(t)$), which is like a dial that tells us how "sticky" or "memory-heavy" the noise is at any specific moment. This dial moves up and down as time passes.

2. The "Shifting Gear" Analogy (The Quantum Extension)

The paper doesn't just look at the noise; it connects this "lazy wind" to the actual physics of the quantum computer using a Caldeira–Leggett model.

Think of the quantum computer as a car engine. The noise is the road.

• Classical View: We used to think the road was just bumpy in a fixed way.
• This Paper: The road is made of billions of tiny springs (the environment). The paper shows that if you look at these springs from a distance (high temperature), they act exactly like the "lazy wind" described above. But if you look closely (low temperature), you see the quantum nature of the springs.
• The Bridge: The authors proved that their "lazy wind" math is actually the high-temperature shadow of a complex quantum reality. They built a bridge between the messy, real-world noise and the clean, microscopic laws of physics.

3. The "Stretchy Rubber Band" (The Results)

When the authors simulated how the qubit (the spinning top) behaves with this new "memory wind," they found something surprising:

• Not a Straight Line: In old models, the top's energy decays in a smooth, predictable curve (like a ball rolling down a hill).
• Stretched Exponential: With the new model, the decay is like a rubber band being stretched. It doesn't fall at a constant speed. Sometimes it holds on tight, and sometimes it snaps loose. This "stretched" pattern matches real experiments much better than the old models.
• The "Memory" Effect: Because the noise remembers the past, the qubit doesn't just lose information; it loses it in a way that depends on how long it has been running. The paper found that the qubit can hold its state for surprisingly long times (millions of nanoseconds) if the noise is dominated by this specific type of charge fluctuation.

4. The "Tuning the Radio" Analogy (Experimental Predictions)

The paper suggests that scientists can test this by listening to the "static" on the qubit.

• They propose a method to measure the "Hurst exponent" (the memory dial) by looking at how the qubit's signal fades during specific tests (called Ramsey and Echo experiments).
• If the signal fades in a "stretched" way rather than a straight exponential way, it confirms that the noise has memory and that the "dial" is moving.

5. The "Optimal Speed" (Gate Optimization)

The paper also looks at how fast we should run quantum calculations (gates).

• If you go too slow, the qubit gets tired and falls over (relaxation).
• If you go too fast, the "memory wind" hasn't settled, and the qubit gets confused (dephasing).
• The authors found a "sweet spot" or an optimal speed where the error is lowest. This speed depends on how "sticky" the noise is at that moment.

Summary of What the Paper Claims

• The Problem: Current models assume noise is simple and forgetful, but real qubits experience noise that has a long memory and changes over time.
• The Solution: They created a new mathematical model (mmfBm) that treats noise as a "drifting memory."
• The Proof: They showed mathematically that this model comes from real quantum physics (Caldeira–Leggett) and simulated it on a computer.
• The Result: The simulations show that qubits decay in a "stretched" pattern, not a simple one, and that this model predicts how long qubits can stay coherent much more accurately than before.
• The Limitation: The paper admits that while the math works, simulating this "drifting memory" on a computer is very hard, especially at extremely low temperatures, and the current computer models sometimes struggle to perfectly match the theoretical predictions.

In short, the paper says: "Stop treating quantum noise like random static. It's a living, breathing, remembering force that changes its mind over time, and we need new math to understand it."

Technical Summary

Technical Summary: Non-Stationary Decoherence in Superconducting Qubits

Problem Statement
Superconducting qubits are highly sensitive to low-frequency environmental fluctuations, particularly $1/f^\beta$ noise arising from dielectric defects, two-level systems (TLS), and quasiparticles. Experimental observations consistently reveal non-exponential decay in Ramsey and spin-echo coherence, along with long-range temporal correlations that cannot be captured by stationary Gaussian models or purely Markovian approaches. Existing models fail to account for two critical features observed in modern devices: (i) the slow temporal drift of the spectral exponent $\beta(t)$, and (ii) long-range correlations extending over many gate cycles. This work addresses the need for a unified framework that describes non-stationary, long-memory decoherence in superconducting charge qubits.

Methodology
The authors propose a unified stochastic drift model (SdM) grounded in Memory Multi-Fractional Brownian Motion (mmfBm) and extend it into a microscopic quantum framework.

1. Classical Sector (mmfBm): The model defines the qubit coordinate evolution driven by a Gaussian process $M(t)$ with a time-dependent Hurst exponent $H(t) \in (1/2, 1)$. The process is constructed via a Riemann–Liouville fractional integral of standard Wiener noise, incorporating an adaptive memory kernel $K(t, s) = (t-s)^{H(t)-1/2}$. This formulation allows the local roughness of the noise to vary slowly over time, generating non-stationary $1/f^\beta$ noise where the spectral exponent is linked to the Hurst parameter by $\beta(t) = 2H(t) - 1$.
2. Quantum Extension: The classical stochastic model is embedded into a microscopic quantum environment using a time-dependent Caldeira–Leggett construction. The environment is modeled as a bosonic bath with a time-dependent spectral density $J(\omega; t) = \eta(t) \omega_c^{1-s(t)} \omega^{s(t)} e^{-\omega/\omega_c}$, where $s(t) = 2H(t) - 1$.
3. Derivations and Simulations:
   • The authors derive the Feynman–Vernon influence functional and the resulting time-dependent Lindblad master equation under Born–Markov and adiabatic approximations.
   • They establish the connection between the quantum bath and the classical mmfBm limit in the high-temperature regime.
   • Numerical simulations are performed using Cholesky decomposition for mmfBm trajectory generation and the QuTiP library for solving the time-dependent master equation.
   • A machine learning approach (CNN) is briefly explored to infer the Hurst exponent from coherence curves, though its limitations are explicitly noted.

Key Contributions

• Unified Framework: The paper establishes a direct theoretical bridge between stochastic noise phenomenology (mmfBm) and microscopic open-system dynamics (time-dependent Caldeira–Leggett bath).
• Time-Dependent Spectral Density: It introduces a spectral density $J(\omega; t)$ that consistently reproduces the experimentally observed time-varying noise exponent $\beta(t) = 2H(t) - 1$.
• Analytical Results:
  • Derivation of the effective time-local Lindblad master equation for non-stationary environments.
  • Proof that the classical mmfBm model emerges as the high-temperature, weak-back-action, adiabatic limit of the quantum bath.
  • Establishment of scaling laws for Ramsey and echo coherence, predicting stretched-exponential decay envelopes of the form $\exp[-A t^{2H(t)}]$ and $\exp[-A t^{2H(t)+2}]$, respectively.
• Numerical Validation: The study provides numerical estimates for relaxation ($T_1$) and dephasing ($T_2$) times under charge-noise dominance, yielding $T_1 \sim 5.00 \times 10^6$ ns and $T_2 \sim 4.18 \times 10^5$ ns.

Results

• Decoherence Dynamics: Simulations demonstrate that decoherence is highly sensitive to the interplay between drift, noise amplitude, and memory effects. Time-dependent Hurst profiles generate more realistic non-stationary behavior compared to constant-$H$ models.
• Non-Markovian Behavior: The system exhibits non-Markovian dephasing with stretched-exponential decay patterns, confirming that simple exponential decay models are insufficient for describing qubit environments dominated by long-memory noise.
• Temperature Crossover: A clear transition is observed between a quantum vacuum-dominated regime (where dephasing is temperature-independent) and a thermal regime (where $\Gamma_\phi \propto T$). The crossover time scale is identified as $t_{cross} \sim \hbar/k_B T$.
• Gate Optimization: The competition between energy relaxation and pure dephasing reveals an optimal gate time that minimizes total error, balancing relaxation decay against rapid dephasing at short times.
• Machine Learning Limitations: The attempt to extract $H(t)$ using a CNN yielded poor generalization ($R^2 = -4.05$), attributed to the use of simplified training data and the inability of global pooling architectures to capture temporal variability. The paper suggests sequence-aware architectures (e.g., LSTMs, Transformers) are required for future work.

Significance and Claims
The paper claims to provide a flexible foundation for understanding non-stationary decoherence, moving beyond the limitations of Markovian approaches. It asserts that the non-exponential decay patterns revealed by the mmfBm framework highlight fundamental constraints on current noise models and offer a guide for designing noise-resilient qubit architectures.

The authors are modest regarding the current state of the work, acknowledging several limitations:

• Finite-time simulations do not fully capture asymptotic low-frequency scaling, leading to discrepancies between fitted and theoretical PSD exponents.
• The locally stationary approximation may break down if the variation of $H(t)$ is too rapid.
• The Born–Markov treatment neglects strong memory effects potentially relevant at millikelvin temperatures.
• The current implementation struggles with numerical stiffness at ultra-low temperatures due to infrared singularities.

The work concludes that while the framework successfully reproduces key physical features (stretched-exponential decay, temperature crossover, non-trivial scaling), future progress requires exact non-Markovian solvers, improved spectral estimation techniques, and more rigorous microscopic derivations connecting mmfBm parameters to specific material defects.