New method estimates Hurst exponents using a one-qubit mirror extension

Brani Vidakovic of the Department of Statistics at Texas A&M University has developed a new method for estimating the Hurst exponent of fractional Brownian motion using quantum wavelets. The work introduces a one-qubit mirror extension that reduces the root mean squared error by a factor of 3-10 for Hurst exponents greater than or equal to 0.5. For paths of length 512, the nondecimated db2 estimator, using exact energy shares, performs 6-14% above the root Cramér-Rao bound, and researchers found Hadamard level tests are less accurate than directly measuring leading qubits.

Quantum Wavelets Estimate Hurst Exponents for fBm Paths

An orthonormal wavelet transform can now be implemented as a quantum unitary for wavelets of any order, enabling the preparation of fractional Brownian motion (fBm) paths as amplitude-encoded states with gate complexity scaling polylogarithmically with the number of time steps. This advance allows for the estimation of the Hurst exponent, H, from repeated preparations of a single path, interrogating the same realization with each measurement.

Researchers extended the ε-decimated quantum nondecimated wavelet transform in three key ways, addressing limitations in circular filtering, estimator sampling distributions, and readout comparisons at equal query budgets. The team removed the endpoint jump created by circular filtering in fBm paths, a refinement that improves accuracy in Hurst exponent estimation. They then derived the sampling distribution of the resulting estimator of H under finite query budgets, providing a statistical basis for assessing estimation accuracy.

This extension preserves every wavelet coefficient that straddles a fold, regardless of filter type or circular shift. Circuits built using this method prepare a path, mirror it, apply an optional shift, and then perform the wavelet transform, culminating in a level test. The resulting probability of measuring a specific level directly corresponds to the share of energy present at that level.

One-Qubit Mirror Extension Removes Circular Filtering Jump

This refinement ensures continuity in the extended path. The extension achieves this by adjoining a qubit and applying controlled-X gates, effectively creating a mirrored and continuous representation of the original path. This reflection serves a dual purpose, functioning as both the controlled unitary for a Hadamard test and the oracle for amplitude estimation.

Paraunitary Lattice Compiles Quantum Wavelet Transforms

At a Hurst exponent of 0.9 and utilizing 220 queries, maximum-likelihood amplitude estimation achieved a root mean squared error of 0.0347, while direct measurement yielded 0.0451 and a method using exact shares reached 0.0295. The construction of the transform relies on compiling from the paraunitary lattice of a compactly supported orthonormal two-channel filter bank, where each rotation acts on a channel qubit distinguishing even and odd samples. This technique allows for the consistent treatment of coefficients that straddle the boundaries of the transformed data, preventing energy concentration issues that arise from circular filtering.

Nondecimated Wavelet Shares from Shift-Register Construction

This advance builds upon existing shift-register techniques, creating a coherent family of circularly shifted orthogonal transforms. Tracing out the shift register yields shift-averaged level energies, while expectations of level projectors recover these energies and their log-spectrum reveals the Hurst exponent, H. A key refinement addresses a long-standing issue in circular filtering, removing the endpoint jump that can distort the estimation of H.

The transform itself is compiled from the paraunitary lattice of its filter bank, using single-qubit rotations and controlled permutations of a position register. Proposition 3 details the construction, beginning with the preparation of an L-qubit register in a superposition state and adding its value to the data register modulo N.

Application of the wavelet transform to the data register results in a reduced state where level shares are nondecimated, calculated as “2^(-L)sum_(s = 0)^(2^L-1)p_j(S^sB) = (2^(j-J)sum_(τ ∈ Z_N)⟨ where gj,τ represents the level-j equivalent filter starting at τ. This formulation allows for filter selection based on desired statistical properties, and the resulting level shares are directly linked to the energy distribution within the transformed signal. The method offers a pathway to more accurate and efficient estimation of the Hurst exponent, an essential parameter in characterizing fractional Brownian motion.

Root-N Central Limit Theorem for Log-Energy Contrasts

This theorem, proven for filters possessing at least two vanishing moments, establishes that the estimator’s distribution approaches a standard normal distribution as the number of measurements increases, provided the query budget is deterministic and the success count at each level diverges. This asymptotic normality holds regardless of the path’s overall norm, a significant advantage for single-record quantum inference where the path norm dictates shot-noise variance and weighting schemes.

The researchers verified these results on a calibration grid, observing a strictly decreasing relationship between the least-squares slope of u(H) and the Hurst exponent itself, alongside a normalized smallest eigenvalue of DΓD⊤ exceeding. The statistical underpinnings of this advancement rest on the negligible contribution of fold terms to the limiting behavior of the mirrored log-energy contrasts, as detailed in Theorem 7.

The team’s analysis, supported by Lemma 8, demonstrates that fine-scale fluctuations remain asymptotically independent of the path norm, a critical property for achieving accurate estimation with deterministic query budgets. For sparse counts the researchers give a working multinomial–lognormal likelihood, exact in its readout model for direct measurement, and check its coverage by simulation.

Direct Measurement Outperforms Hadamard Level Tests

Direct measurement of leading qubits proves more efficient than Hadamard level tests across all query budgets, a surprising result given the established use of Hadamard tests in quantum estimation. Simulations reveal that direct measurement consistently outperforms level tests because each measurement informs every level, maximizing information gain from each preparation of an amplitude-encoded path. This advantage is particularly noticeable when informative shares are rare, a condition frequently encountered in practical applications.

Researchers tested query budgets of 217, 220, and 223, consistently finding direct measurement to be the superior approach. While amplitude estimation exhibits the lowest RMSE at the highest query budget, it fails to return an estimate on 16% of records at H = 0. 9 with the smallest budget, highlighting a conditional advantage.

“Direct measurement is more efficient than level tests at every setting,” the study confirms, emphasizing the practical benefits of this approach for estimating Hurst exponents from limited quantum resources.

Amplitude Estimation Minimizes RMSE at H = 0.9

Maximum-likelihood amplitude estimation (MLAE) achieves a root mean squared error (RMSE) of 0. 0347 at a Hurst exponent (H) of 0. 9 and 220 queries, surpassing the 0. 0451 RMSE obtained through direct measurement using the same records. This improved accuracy is particularly notable given that MLAE delivers this performance on 84% of records, returning an estimate for that proportion of trials.

The comparison accounts for a slight increase in computational cost, with each MLAE query requiring 1. 6-2. 0% more CNOT gates than a direct readout query, maintaining comparable ideal CNOT budgets overall. This indicates that MLAE not only improves upon direct measurement but also approaches the theoretical limit of accuracy achievable from the raw data itself. the ratio of MLAE’s RMSE to direct readout is 0. 77, with a standard error of 0. 04, demonstrating a consistent and statistically significant improvement in estimation precision.

The study highlights that this enhanced performance is most evident when informative shares, the signal contributing to the estimation, are small. While MLAE demonstrates superior accuracy in many scenarios, its success rate is a factor. At H = 0. Direct measurement, in contrast, consistently provides an estimate for every record, though at a slightly lower level of accuracy. 65 to 0. 98, and an RMSE of 0. 052 against 0. 067 at H = 0. 8 and 220 queries.

👉 More information
🗞 Hurst Exponent Estimation with Quantum Wavelets
✍️ Brani Vidakovic (Department of Statistics, Texas A&M University)
🧠 ArXiv: https://arxiv.org/abs/2611.24882v1

Stay current

See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.

Avatar of Rusty Flint

Rusty Flint

Rusty is a quantum science nerd. He's been into academic science all his life, but spent his formative years doing less academic things. Now he turns his attention to write about his passion, the quantum realm. He loves all things Quantum Physics especially. Rusty likes the more esoteric side of Quantum Computing and the Quantum world. Everything from Quantum Entanglement to Quantum Physics. Rusty thinks that we are in the 1950s quantum equivalent of the classical computing world. While other quantum journalists focus on IBM's latest chip or which startup just raised $50 million, Rusty's over here writing 3,000-word deep dives on whether quantum entanglement might explain why you sometimes think about someone right before they text you. (Spoiler: it doesn't, but the exploration is fascinating)

Latest Posts by Rusty Flint: