Coherence Ratio: Definitions & Applications

• Coherence ratio is a context-dependent metric that normalizes and compares coherence-bearing quantities against reference benchmarks across diverse measurement models.
• It is employed in fields such as quantum device readout, AGN timing, fMRI mapping, spectroscopy, and quantum resource theory, each using unique formulations.
• Its various forms—ranging from time-based, SNR-weighted, to fidelity-normalized scores—offer actionable insights into performance differences and methodological nuances.

Coherence ratio is a context-dependent term used for quantitative comparisons of coherence-bearing quantities rather than a single standardized invariant. In the cited literature it denotes, among other things, a ratio of coherence times between device topologies, a readout-weighted product of signal-to-noise ratio and coherence time, a normalized robustness measure in resource theory, a ratio of coherent to diffuse or incoherent components in imaging and acoustics, and a ratio of coherent to incoherent correlation strengths or durations in time-series analysis (Das et al., 17 Jan 2026, Fan et al., 2024, Yao et al., 2019, Wang, 2021). In all of these usages, the underlying purpose is comparative: the ratio is introduced to normalize, rank, or operationalize coherence relative to a reference quantity such as lifetime, background noise, incoherent mixing, or a maximally coherent benchmark.

1. Domain-specific definitions

The literature does not assign a single universal mathematical meaning to coherence ratio. Instead, the same phrase is used for several structurally different quantities, each tied to a measurement model or resource-theoretic framework.

Domain	Ratio form	Operational role
Topological quantum-device readout	$CR_{X/Y}=T_2(X)/T_2(Y)$; $CW(X)=\mathrm{SNR}(X)\times T_2(X)$	Compare topologies and readout-weighted coherence
AGN X-ray timing	$\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$	Noise-corrected intrinsic coherence between bands
FF-OCT	$a=A_s/A_r=\sqrt{I_s/I_r}$	Balance coherent sample and reference fields
Phonon spectroscopy	$$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$	Compare temporal coherence time to lifetime
Resting-state fMRI TCM	$R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$, $R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$	Quantify coherence/anti-coherence balance
Quantum resource theory	$\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$; $R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$	Normalize coherence or quantify amplification

These definitions are not interchangeable. Some are ratios of times, some are normalized overlaps, some are coherence functions in the spectral sense, and some are detector-balance variables that control fringe visibility or SNR. That distinction matters because the same numerical increase can mean longer dephasing time in one field, stronger cross-band linear correlation in another, or merely better amplitude matching at a detector in a third (Epitropakis et al., 2017, 2206.12172, Stratton et al., 3 Dec 2025).

2. Topological quantum-device usage

In the 2026 study of Möbius, loop, and trefoil Majorana architectures, the authors state that their paper did not define “coherence ratio” explicitly and then introduce two quantities tied directly to what was measured: the time-based ratio

$CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$

and the readout-oriented metric

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$0

These are extracted from Lorentzian fits to power–frequency spectra obtained from quantum-capacitance measurements. The spectral model is written as

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$1

or equivalently

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$2

with $CW(X)=\mathrm{SNR}(X)\times T_2(X)$3, and the linewidth–coherence relation is

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$4

The parity-readout peak SNR is defined by

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$5

where $CW(X)=\mathrm{SNR}(X)\times T_2(X)$6 is the fitted Lorentzian amplitude and $CW(X)=\mathrm{SNR}(X)\times T_2(X)$7 is the rms noise of the measured spectrum in the fit window, obtained from the residuals or baseline fluctuations after subtracting the fit. Preprocessing removes the mean of $CW(X)=\mathrm{SNR}(X)\times T_2(X)$8, applies a Hann window, computes the one-sided DFT, and defines $CW(X)=\mathrm{SNR}(X)\times T_2(X)$9 before Lorentzian peak fitting (Das et al., 17 Jan 2026).

At $\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$0 and $\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$1, the reported values from Table 1 are: Loop, $\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$2, $\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$3, $\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$4; Möbius, $\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$5, $\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$6, $\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$7; Trefoil, $\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$8, $\gamma^2(f)=|G_{xy}(f)|^2/[G_{xx}(f)G_{yy}(f)]$9, $a=A_s/A_r=\sqrt{I_s/I_r}$0. Consequently,

$a=A_s/A_r=\sqrt{I_s/I_r}$1

The readout-weighted values are

$a=A_s/A_r=\sqrt{I_s/I_r}$2

with ratios

$a=A_s/A_r=\sqrt{I_s/I_r}$3

The linewidth-normalized alternative

$a=A_s/A_r=\sqrt{I_s/I_r}$4

gives $a=A_s/A_r=\sqrt{I_s/I_r}$5, $a=A_s/A_r=\sqrt{I_s/I_r}$6, and $a=A_s/A_r=\sqrt{I_s/I_r}$7, with the same ratios as $a=A_s/A_r=\sqrt{I_s/I_r}$8 in this regime.

The central result is that the strict coherence ratio, understood as $a=A_s/A_r=\sqrt{I_s/I_r}$9-ratio, is unity to the quoted precision, while the readout-weighted metrics differ. The paper interprets the identical linewidths as identical dephasing rates $$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$0, consistent with decoherence being dominated by environment or material mechanisms rather than global band topology. By contrast, SNR differences can arise from peak amplitude $$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$1, noise floor $$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$2, coupling to the readout resonator, parasitic capacitances, impedance mismatches, mode-structure differences, windowing effects, or small variations in dissipation. The same paper also notes that geometry-dependent phases in $$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$3 can shift the dominant oscillation frequency $$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$4, indirectly modulating SNR through the transfer function and noise PSD. One factual ambiguity remains: the abstract states that at $$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$5 and $$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$6 the ordering is “Trefoil, Mobius, and Loop,” whereas the detailed values quoted from Table 1 yield $$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$7, corresponding to Trefoil $$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$8 Loop $$R_{\mathbf{k}s}=\tau_{\mathbf{k}s}^c/\tau_{\mathbf{k}s}^p$$9 Möbius (Das et al., 17 Jan 2026).

3. Spectral coherence and correlation-balance formulations

In AGN timing analysis, coherence ratio is treated in the spectral-statistical sense of intrinsic coherence between two stochastic processes. For stationary processes $R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$0 and $R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$1 with cross-spectrum $R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$2 and auto-spectra $R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$3, $R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$4, the population magnitude-squared coherence is

$R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$5

The cited study estimates cross-spectra by splitting light curves with bin size $R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$6 into $R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$7 segments of length $R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$8, computing discrete Fourier transforms on each segment, and averaging the resulting cross-periodograms and periodograms. The intrinsic, noise-corrected estimator follows Vaughan and Nowak: $R_{\mathrm{corr}}=\mathrm{TC}/\mathrm{TAC}$9 with

$R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$0

A key practical result is that the analytic error must be divided by

$R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$1

to approximate the true standard deviation when $R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$2, and intrinsic coherence estimates are approximately Gaussian for $R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$3. Empirically, the intrinsic coherence is modeled as

$R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$4

that is, approximately constant at low frequencies and exponentially decreasing above a break frequency. The low-frequency level is typically $R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$5–$R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$6, while representative $R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$7 values span $R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$8–$R_{\mathrm{time}}=\mathrm{MLP}/\mathrm{MLN}$9. Neither the low-frequency constant intrinsic-coherence value nor the break frequency exhibits a universal scaling with black-hole mass or X-ray Eddington ratio across the full sample (Epitropakis et al., 2017).

Resting-state fMRI temporal coherence mapping uses two explicit ratio forms. After temporal embedding of a voxel time series into transit states and construction of a temporal coherence matrix from Pearson correlations $\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$0, the method defines mean temporal coherence

$\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$1

mean temporal anti-coherence

$\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$2

and the intuitive correlation ratio

$\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$3

A second, recurrence-based time ratio is

$\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$4

where MLP and MLN are normalized mean lengths of positive and negative diagonal segments in thresholded recurrence masks. The paper reports that the ratio formulation was less reproducible and therefore uses the differences

$\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$5

in the current implementation. In healthy resting cortex, anti-coherence tends to dominate, corresponding to $\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$6 or equivalently $\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$7; the coherent/incoherent time ratio $\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$8 had poor reproducibility and is not recommended as a primary outcome (Wang, 2021).

A related acoustic formulation appears in coherence-based spectral enhancement for distant speech recognition, where the ratio between the coherent direct sound field and the diffuse component is estimated in each time-frequency bin. The quantity is the coherent-to-diffuse power ratio (CDR), also called direct-to-diffuse ratio, and under the two-component coherence model

$\overline{C}_{\mathcal R}=(1+C_{\mathcal R})/d$9

it serves as an approximation to the short-time SNR driving a Wiener-type postfilter after MVDR beamforming. This usage is not a normalized coherence function, but it is a literal ratio of coherent to incoherent acoustic power derived from complex spatial coherence estimates (Barfuss et al., 2015).

4. Imaging, interferometry, and spectroscopy

In adjustable-ratio full-field OCT, the paper does not use the phrase “coherence ratio” explicitly, but it identifies the quantity most directly relevant to fringe visibility as the ratio of coherent field amplitudes at the detector: $R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$0 With polarization-controlled splitting, the detector-side coherent intensity ratio is

$R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$1

so that

$R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$2

The standard visibility is

$R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$3

maximized at $R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$4. This yields the visibility-optimal source-side splitting

$R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$5

When incoherent background reaches the detector,

$R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$6

so the visibility optimum and SNR optimum need not coincide. The paper’s SNR model gives, under the assumption $R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$7 and $R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$8,

$R_j=C_{\mathrm{out}}/C_{\mathrm{in}}$9

with optimum

$CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$0

Experimentally, the reported optimum was $CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$1 for transparent multilayer tape and $CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$2 for fingerprint imaging at $CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$3; for the fingerprint experiment, imaging depth increased from $CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$4 to $CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$5, with SNR gains of $CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$6 in epidermis and $CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$7 in dermis versus a previous fixed-ratio approach. The paper explicitly distinguishes this detector-balance notion from coherence length (Fan et al., 2024).

In phonon spectroscopy, the coherence ratio is a direct ratio of timescales: $CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$8 Here the coherence time is defined from a Gaussian envelope in the time-domain normal-mode coordinate,

$CR_{X/Y}=\frac{T_2(X)}{T_2(Y)},$9

while the conventional phonon lifetime is obtained from a Lorentzian spectral line,

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$00

The same ratio can therefore be written as

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$01

and, using the Gaussian FWHM,

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$02

The study reports that confined modes exhibit a pronounced wavelike behavior characterized by a higher ratio of coherence time to lifetime, and that the ratio decreases as the confined dimension increases (2206.12172).

A third spectroscopic use arises in heterodyned chiral CARS with strongly prepared molecular coherence. There the measurable ratio is

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$03

where $CW(X)=\mathrm{SNR}(X)\times T_2(X)$04 and $CW(X)=\mathrm{SNR}(X)\times T_2(X)$05 are heterodyned chiral and achiral components. Because the common factors $CW(X)=\mathrm{SNR}(X)\times T_2(X)$06, $CW(X)=\mathrm{SNR}(X)\times T_2(X)$07, and $CW(X)=\mathrm{SNR}(X)\times T_2(X)$08 cancel, the ratio directly measures the chiral-to-achiral parameter ratio $CW(X)=\mathrm{SNR}(X)\times T_2(X)$09. The same paper attributes signal enhancement relative to spontaneous ROA to coherent $CW(X)=\mathrm{SNR}(X)\times T_2(X)$10 scaling and reports that the enhanced chiral signal due to strong molecular coherence is up to four orders of magnitude higher than that of spontaneous Raman optical activity (Begzjav et al., 2018).

5. Quantum-information meanings

In quantum coherence resource theory, one explicit use of coherence ratio is the normalized robustness of coherence

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$11

This quantity maps incoherent states to $CW(X)=\mathrm{SNR}(X)\times T_2(X)$12 and maximally coherent states to $CW(X)=\mathrm{SNR}(X)\times T_2(X)$13, and is described as the ratio of a state’s robustness to that of a maximally coherent state. The same work defines the quantum coherence fraction

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$14

proves the bound

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$15

and shows exact equality for all qubit and qutrit states. It also gives the GIO–fidelity formulation

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$16

For $CW(X)=\mathrm{SNR}(X)\times T_2(X)$17 random states per dimension, the largest observed relative gaps between $CW(X)=\mathrm{SNR}(X)\times T_2(X)$18 and $CW(X)=\mathrm{SNR}(X)\times T_2(X)$19 were approximately $CW(X)=\mathrm{SNR}(X)\times T_2(X)$20, $CW(X)=\mathrm{SNR}(X)\times T_2(X)$21, $CW(X)=\mathrm{SNR}(X)\times T_2(X)$22, and $CW(X)=\mathrm{SNR}(X)\times T_2(X)$23 for $CW(X)=\mathrm{SNR}(X)\times T_2(X)$24, respectively (Yao et al., 2019).

A distinct paper, “Coherence Fraction,” explicitly states that it introduces coherence fraction rather than coherence ratio. Its central quantity is

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$25

with $CW(X)=\mathrm{SNR}(X)\times T_2(X)$26. For qubits,

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$27

The paper treats $CW(X)=\mathrm{SNR}(X)\times T_2(X)$28 as a fidelity-type, basis-dependent proximity measure to the set of maximally coherent states, not as a ratio in the normalization sense (Karmakar et al., 2019).

Another use is dynamical amplification. In unspeakable coherence concentration, two uncorrelated copies of a coherent state are processed by globally coherence non-increasing unitaries, and the coherence ratio is defined as the input–output amplification factor

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$29

For qubits, with $CW(X)=\mathrm{SNR}(X)\times T_2(X)$30, the optimal single-shot ratio becomes

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$31

When $CW(X)=\mathrm{SNR}(X)\times T_2(X)$32, the paper identifies “bound coherence,” for which $CW(X)=\mathrm{SNR}(X)\times T_2(X)$33 even though the state may still have nonzero coherence. In the multi-copy concatenation protocol, for $CW(X)=\mathrm{SNR}(X)\times T_2(X)$34 and any $CW(X)=\mathrm{SNR}(X)\times T_2(X)$35, there exist initial states such that

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$36

for all $CW(X)=\mathrm{SNR}(X)\times T_2(X)$37, establishing that the input–output coherence ratio can be amplified unboundedly, although the absolute coherence remains bounded by initial purity (Stratton et al., 3 Dec 2025).

A further normalization appears in the exposition of the fidelity-based coherence measure $CW(X)=\mathrm{SNR}(X)\times T_2(X)$38, where a “coherence ratio” is obtained by dividing by the basis-dependent maximal value: $CW(X)=\mathrm{SNR}(X)\times T_2(X)$39 and for qubits

$CW(X)=\mathrm{SNR}(X)\times T_2(X)$40

Here the ratio is a normalized fidelity-based coherence score in $CW(X)=\mathrm{SNR}(X)\times T_2(X)$41, with $CW(X)=\mathrm{SNR}(X)\times T_2(X)$42 for the maximally coherent qubit (Liu et al., 2017).

6. Interpretation, normalization, and recurrent caveats

Several recurring caveats govern how coherence ratio should be interpreted. First, ratios and weighted metrics need not agree. In the topological-device study, the strict time-based ratios $CW(X)=\mathrm{SNR}(X)\times T_2(X)$43 are all $CW(X)=\mathrm{SNR}(X)\times T_2(X)$44 at $CW(X)=\mathrm{SNR}(X)\times T_2(X)$45, $CW(X)=\mathrm{SNR}(X)\times T_2(X)$46, yet the readout-weighted quantities $CW(X)=\mathrm{SNR}(X)\times T_2(X)$47 and $CW(X)=\mathrm{SNR}(X)\times T_2(X)$48 differ and inherit the SNR ordering (Das et al., 17 Jan 2026). Second, the visibility-optimal condition in interferometry, $CW(X)=\mathrm{SNR}(X)\times T_2(X)$49, is not the same as the SNR-optimal condition when incoherent background varies with the tuning parameter; in FF-OCT, the visibility criterion gives $CW(X)=\mathrm{SNR}(X)\times T_2(X)$50, whereas the SNR optimum is $CW(X)=\mathrm{SNR}(X)\times T_2(X)$51 under the stated approximation (Fan et al., 2024). Third, ratio forms may be less stable than difference forms: in temporal coherence mapping, $CW(X)=\mathrm{SNR}(X)\times T_2(X)$52 and $CW(X)=\mathrm{SNR}(X)\times T_2(X)$53 are conceptually straightforward, but the paper adopts $CW(X)=\mathrm{SNR}(X)\times T_2(X)$54 and $CW(X)=\mathrm{SNR}(X)\times T_2(X)$55 because ratios amplify noise when the denominator is small and $CW(X)=\mathrm{SNR}(X)\times T_2(X)$56 showed poor reproducibility (Wang, 2021).

A further source of confusion is terminological overlap with adjacent concepts. The detector-balance ratio in FF-OCT is explicitly distinct from coherence length (Fan et al., 2024). In resource theory, coherence fraction and coherence ratio are not synonyms: one is an overlap with maximally coherent states, the other may be a normalized robustness or a normalized fidelity-based score (Karmakar et al., 2019, Yao et al., 2019). In AGN timing, intrinsic coherence is a noise-corrected frequency-dependent correlation function and not a ratio of coherence times or coherence resources (Epitropakis et al., 2017). The same phrase therefore identifies different mathematical objects depending on whether the relevant coherence is dephasing time, cross-spectral correlation, optical interference balance, phonon wavepacket duration, or resource-theoretic distance from incoherence.

The unifying feature is operational comparison. A coherence ratio always relates a coherence-bearing quantity to a reference scale: another topology, a lifetime, an incoherent background, a maximally coherent benchmark, or an input state before processing. Its technical content is therefore inseparable from the estimation protocol that defines the numerator and denominator.