Correlated Quantum Dephasometry

• Correlated quantum dephasometry is the estimation of structured dephasing noise using techniques like Ramsey interferometry and single-photon spectroscopy.
• It employs inversion procedures and spectral analysis to reconstruct temporal kernels, spatial covariances, and collective-mode variances from quantum observables.
• Experimental implementations span waveguide-QED systems to superconducting circuits, providing scalable approaches for noise estimation in multiqubit sensor networks.

Correlated quantum dephasometry is the estimation and characterization of correlated dephasing noise from quantum observables. In the single-emitter, waveguide-QED setting, it is a spectroscopic framework in which the complex single-photon transmittance and reflectance encode the same noise information as time-domain Ramsey interferometry; in multiqubit settings, it encompasses the reconstruction of spatial and temporal cross-correlations, cross-spectra, and collective-mode covariances from Ramsey signals, parity oscillations, continuous-control measurements, or correlated channel observables (Ramos et al., 2018). The subject therefore spans at least three closely related regimes: temporally correlated phase noise acting on one quantum emitter, spatially correlated dephasing acting on multiple sensors, and memory effects across successive uses of a quantum channel.

1. Conceptual scope and definitions

A central conceptual distinction is between two views of memory in quantum noise. In one view, “the quantum channel has memory when there exist correlations between successive uses of the channels on a sequence of quantum systems,” so that one speaks of correlated quantum channels. In the other view, memory effects arise from correlations generated during the quantum evolution itself. These two notions are related but not identical, and the literature on correlated dephasometry uses both (Awasthi et al., 2019).

In the single-emitter formulation, dephasing is modeled by a stochastic modulation of the transition frequency,

$$\omega_0 \to \omega_0 + \xi(t), \qquad \langle \xi(t)\rangle = 0,$$

with stationary two-point correlation

$C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$

and spectral density

$$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$

Stationarity allows both Markovian and non-Markovian correlated processes, and the formalism covers both Gaussian and non-Gaussian noise (Ramos et al., 2018).

In many-body sensor-network language, the weak correlated-noise channel is written as

$$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$

where $V$ is the covariance of the stochastic fluctuations and $h_i$ is the local generator, such as $h_i=\sigma_i^z$ for qubit dephasing. The estimation target may be a global amplitude, a correlation coefficient, a correlation length, or a collective-mode variance (Brady et al., 2024).

A complementary collective-dephasing model considers $N$ non-interacting spin-$1/2$ particles in a spatially uniform magnetic field of fluctuating intensity,

$$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$

Here the entire dephasing dynamics is governed by the characteristic function of the accumulated phase, and decoherence-free subspaces arise when the initial state commutes with $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$0 (Carnio et al., 2016).

These formulations share the same operational core: correlated phase noise is not described only by a single $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$1-like parameter, but by a structure of temporal kernels, spatial covariances, or inter-use memory parameters that must be inferred from observables sensitive to coherent phase accumulation.

2. Waveguide-QED and the single-photon scattering formulation

The waveguide-QED formulation gives a particularly explicit realization of correlated quantum dephasometry. The system is a two-level emitter with states $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$2, $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$3 and transition frequency $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$4, coupled to a one-dimensional waveguide with rates $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$5 into right/left channels and to unguided modes with rate $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$6. The total radiative decay is

$C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$7

and the analysis is performed in the rotating-wave approximation and weak-probe linear response (Ramos et al., 2018).

For stationary $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$8, the driven qubit response $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$9 obeys

$$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$0

with stationary solution

$$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$1

where

$$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$2

The averaged elastic amplitudes are then

$$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$3

The same function $$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$4 appears in Ramsey interferometry,

$$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$5

so that the transmittance is a Laplace/Fourier transform of the Ramsey envelope: $$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$6 This identity is the central result of the framework: single-photon spectroscopy and time-resolved Ramsey interferometry contain the same correlated-dephasing information, but spectroscopy does not require direct access to the emitter control (Ramos et al., 2018).

In this picture, correlated dephasing acts as a random classical field that imprints a stochastic phase on the qubit during absorption and re-emission. After averaging over noise realizations, the elastic amplitude is reduced and distorted, and the entire noise dependence is captured by the Ramsey envelope.

3. Measurement protocols and inversion procedures

The waveguide-QED implementation uses a weak coherent tone of frequency $$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$7 and complex amplitude $$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$8, with drive

$$S(\omega)=\int_{-\infty}^{\infty} d\tau\, e^{i\omega\tau} C(\tau).$$9

so as to remain in the single-photon or linear-response limit. Homodyne detection gives the steady-state complex transmission amplitude directly,

$$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$0

and therefore yields both $$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$1 and $$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$2. Photon counting measures the output flux and gives $$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$3, while $$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$4 is reconstructed through the Kramers–Kronig relation

$$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$5

using causality and analyticity of the response (Ramos et al., 2018).

Once $$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$6 is known, one reconstructs $$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$7. For symmetric $$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$8 distributions, $$\Phi_{\bm V}(\rho)\approx \rho+\sum_{i,j=1}^N V_{ij}\left(h_i\rho h_j-\frac{1}{2}\{h_i h_j,\rho\}\right),$$9 is real, and the inverse-Fourier formula gives the Ramsey envelope directly. For stationary Gaussian noise,

$V$0

so the measured $V$1 determines $V$2. The inversion is unique under stationarity and Gaussianity, and finite-bandwidth reconstructions may be regularized with Tikhonov methods or truncated SVD. For non-Gaussian noise, $V$3 still determines $V$4, but higher-order cumulants are not uniquely specified by two-point spectra alone (Ramos et al., 2018).

A distinct continuous-control implementation reconstructs all self- and cross-spectra of a two-qubit dephasing environment. In the spin-locking protocol, the fit parameter vector is

$V$5

and the estimator minimizes a robust Huber loss. In the reported experiments, the global fit per frequency used $V$6 initial states $V$7 observables $V$8 times, totaling $V$9 data points, while requiring only single-qubit control manipulations and state-tomography measurements (Lüpke et al., 2019).

A further single-qubit-only route uses refined parity oscillations. After preparing $h_i$0, idling, and applying per-qubit analysis rotations, the parity signal is

$h_i$1

where $h_i$2 is the sum of anti-diagonal density-matrix elements with a fixed bitstring imbalance $h_i$3. The discrete Fourier transform over $h_i$4 reconstructs the full set of $h_i$5, so correlated dephasing appears as a characteristic change in the oscillation line shape rather than merely an overall amplitude decay (Gulácsi et al., 13 Feb 2025).

4. Canonical noise models and spectral signatures

The single-photon formalism admits explicit formulas for the paradigmatic dephasing models encountered in solid-state environments. For white, Markovian noise,

$h_i$6

and the transmittance is Lorentzian,

$h_i$7

For colored Gaussian Ornstein–Uhlenbeck noise,

$h_i$8

Its white-noise limit yields the Lorentzian above, while the quasi-static limit is non-Lorentzian and Gaussian-like. Telegraph noise,

$h_i$9

with switching rate $h_i=\sigma_i^z$0, produces broadened non-Lorentzian profiles and, for $h_i=\sigma_i^z$1, two resolved dips near $h_i=\sigma_i^z$2 together with an oscillatory Ramsey envelope. Gaussian $h_i=\sigma_i^z$3 noise,

$h_i=\sigma_i^z$4

gives the long-time law

$h_i=\sigma_i^z$5

which produces broad low-frequency tails and pronounced phase dispersion in $h_i=\sigma_i^z$6; non-Gaussian $h_i=\sigma_i^z$7 noise constructed from sparse telegraph fluctuators yields small “bumps” and granularity in both $h_i=\sigma_i^z$8 and $h_i=\sigma_i^z$9 (Ramos et al., 2018).

Correlated photonic dephasing provides an explicit two-environment analogue. For two photons whose polarization couples locally to correlated frequency environments, the off-diagonal elements are governed by

$N$0

For a single-peak bivariate Gaussian with correlation coefficient $N$1,

$N$2

$N$3

The environmental correlation is then obtained directly from coherence ratios,

$N$4

Double-peak spectral distributions introduce cosine factors, isolated divergences in time-local rates, and coherence revivals, so that nonlocality and non-Markovianity become directly visible in the master-equation structure (Raja et al., 2020).

5. Multiqubit structure and metrological limits

Under collective dephasing, correlated noise can generate and protect correlations rather than merely destroy them. For two qubits in Fano form,

$N$5

the correlation rank is

$N$6

Starting from a product state with $N$7, collective dephasing can yield $N$8, thereby generating discord-type correlations without creating entanglement. Werner states remain invariant for any field direction, and more generally any state satisfying

$N$9

lies in a decoherence-free subspace (Carnio et al., 2016).

For weak correlated-noise estimation in sensor networks, the key quantity is the overlap between the classical covariance of the channel and the quantum generator covariance of the probe. For a pure probe $1/2$0,

$1/2$1

and the main result is

$1/2$2

when $1/2$3. Entanglement helps if and only if the noise is spatially correlated and the probe has aligned generator correlations. In the fully correlated rank-1 case, separable probes give shot-noise scaling, whereas GHZ-like probes give Heisenberg scaling for collective dephasing-amplitude estimation (Brady et al., 2024).

The metrological consequences depend strongly on whether the correlated noise commutes with the signal. In the comb-based treatment of temporally correlated dephasing, parallel dephasing yields a standard-scaling upper bound and “no Heisenberg scaling is achievable for any correlation parameter $1/2$4,” although negative temporal correlations are beneficial. By contrast, for perpendicular dephasing, Heisenberg scaling is possible (Kurdzialek et al., 2024). In the commuting correlated-dephasing model with Gaussian AR(1)-type phase fluctuations, later tensor-network studies found that “spin-squeezed offer practically optimal performance in the regime where phase fluctuations are positively correlated, but can be outperformed by tensor-network optimized strategies for negatively correlated fluctuations” (Ghosh et al., 10 Nov 2025).

These results delimit a recurrent misconception. Correlated dephasing does not imply either universal metrological enhancement or universal metrological degradation. The outcome depends on the alignment between the signal generator and the correlated noise, on the sign structure of the correlations, and on whether the estimation target is the signal itself or the noise parameters.

6. Implementations, applications, and scalable variants

In superconducting circuits, correlated dephasometry has been demonstrated with two transmon qubits dispersively coupled to a shared $1/2$5 resonator. The spin-locking spectroscopy protocol swept $1/2$6, used spin-locking times $1/2$7, and reconstructed $1/2$8, $1/2$9, and $$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$0, including the real and imaginary parts of the cross-spectrum. The reconstructed spectra matched the predicted Lorentzians within $$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$1 confidence intervals and captured the strong asymmetry about zero frequency characteristic of non-classical photon shot noise (Lüpke et al., 2019).

For large registers with sparse long-range pairwise correlations, compressed sensing provides a scalable route. The Markovian dephasing generator is

$$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$2

and random GHZ-type Ramsey experiments on subsets convert decay rates into linear measurements of the off-diagonal part of $$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$3. The method “requires as few as $$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$4 measurement settings,” while the robust RIP-based regime uses $$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$5 and has sample complexity

$$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$6

to be compared with conventional methods of sample complexity $$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$7 (Seif et al., 2021).

A recent extension turns correlated dephasing into a symmetry-resolved nanoscale materials probe. Two spin qubits above a two-dimensional material measure a cross-dephasing rate whose orientation dependence is

$$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$8

so that the harmonics isolate $$H(t)=\hbar\,\omega(t)\,J_{\boldsymbol{n}},\qquad J_{\boldsymbol{n}}=\frac{1}{2}\sum_{k=1}^N \boldsymbol{n}\cdot\sigma^{(k)}.$$9-, $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$00-, and $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$01-wave symmetry channels. In 2D superconductors, the method predicts clear fingerprints that discriminate $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$02-, $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$03-, and $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$04-wave gaps; in magnetic systems, it distinguishes $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$05D $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$06-wave antiferromagnets from $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$07-wave altermagnets. For a representative $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$08 crystalline FeSe film at $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$09 and $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$10, the characteristic single-qubit dephasing time from superconducting noise is $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$11, and $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$12 is achievable for $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$13 (Sun et al., 24 Apr 2026).

Operational extensions reach beyond spectroscopy. In two-use correlated dephasing channels with Pauli-$C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$14 memory parameter $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$15, Bell-basis measurements on system–reference pairs produce a detected capacity

$C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$16

and for correlated dephasing this witness equals the exact quantum capacity. The same measured probability vector $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$17 also determines the dephasing probability $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$18 and the inter-use correlation strength $C(\tau)=\langle \xi(t)\xi(t+\tau)\rangle$19, so noise parameter estimation and channel-capacity certification are obtained simultaneously (Macchiavello et al., 2016).

Taken together, these formulations show that correlated quantum dephasometry is not a single protocol but a family of spectroscopic and estimation methods unified by one principle: correlated dephasing leaves structured, invertible signatures in coherent amplitudes, parity harmonics, multiqubit correlators, and channel observables. The precise observable depends on the platform, but the target is always the same object—the correlation structure of phase noise in time, space, or channel use.