Generation of Correlated Time Series for X-ray Astronomy Applications

J\"orn Wilms; Michael A. Nowak; Seth Rowan Larner

Generation of Correlated Time Series for X-ray Astronomy Applications

T0 review · 0 major / 7 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A new algorithm generates pairs of synthetic X-ray light curves with any desired power spectra, coherence, and phase-lag profiles.

desk verdict Correct, well-scoped, and genuinely useful algorithm for generating correlated light curves; the Gaussian/linear limitation is real but fully disclosed. read the letter →

arxiv 2608.02584 v1 pith:QPC7Z5UR submitted 2026-08-03 astro-ph.HE astro-ph.IM

Seth Rowan Larner , Michael A. Nowak , Jörn Wilms This is my paper · ORCID

classification astro-ph.HEastro-ph.IM

keywords syntheticX-raylightcurvespowerspectracoherencephaselagscross-spectrumspectraltimingtimeseriessimulationestimatorcovariance

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper presents an algorithm for generating pairs of X-ray light curves that have any desired power spectra, coherence function, and phase-lag profile. It generalizes the standard Fourier-domain method for synthesizing light curves by splitting the dependent series into a coherent part—a complex linear transfer function applied to the reference series—and an independent Gaussian part, with a normalization that preserves the target power spectrum. The paper also derives closed-form expressions for the variance of coherence and phase-lag estimators and the full covariance with power and cross-spectral quantities, using the delta method. Monte Carlo simulations confirm the formulas, and a two-Lorentzian demonstration reproduces input power spectra, coherence, and lags across 0.01–100 Hz. The motivation is to provide realistic synthetic data for testing joint power-spectrum and cross-spectrum fitting techniques, which have recently revealed hidden variability components.

What carries the argument

The central object is the complex transfer function R(ν) = sqrt(PY γ²/PX) e^{iφ(ν)}: it encodes the target coherence and phase lag by modulating and phase-shifting the reference Fourier coefficients. Together with the normalization K(ν) = sqrt((PY − PX |R|²)/2), which scales an independent Gaussian component, it decomposes the dependent series into coherent and incoherent parts. The derivation also uses the covariance of quadratic forms of Gaussian variables, feeding a delta-method computation of estimator variances and the full 6×6 covariance matrix.

What would settle it

Run a Monte Carlo with n=2 segments and true coherence 0.5; the delta-method variance formula Var(γ̂²)≈2γ²(1−γ²)²/n predicts far less scatter than the actual non-Gaussian distribution of γ̂² will show, and the phase-lag estimator will become substantially biased. If the analytic variances still match the simulation, the large-n assumption is stronger than the paper states.

Watch

Extended reading notes

Core claim

The central claim is that for any user-chosen reference and dependent power spectra PX(ν), PY(ν), coherence γ²(ν), and phase lag φ(ν), the construction Y = K(H + iJ) + R X, with R = (PY γ²/PX)^(1/2) e^{iφ} and K = sqrt((PY − PX |R|²)/2), produces a pair of time series whose expected power spectrum, coherence, and phase lag match the targets. The derivation decomposes the dependent series into a fully coherent component (the reference passed through a complex transfer function R) and an incoherent component (independent Gaussian noise scaled by K), so the four target functions can be specified independently. A second claim is that the variance of the estimated coherence is approximately 2γ²(1

Load-bearing premise

The method's validity rests on the assumption that all correlation between the two bands is linear and second-order, with independent Gaussian Fourier coefficients; real accreting systems violate this with log-normal, multiplicative variability.

Editorial extensions

If this is right

Joint fits of power spectra and cross-spectra, such as multi-Lorentzian modeling, require the off-diagonal covariance terms derived here; a diagonal likelihood is inadequate.
The variance formulas give a priori segment-count requirements: to achieve a target phase-lag uncertainty at a given coherence, observers can compute how many segments to average.
Simulated pairs with known input coherence and lags provide a ground truth to validate lag-recovery pipelines and reverberation-mapping analysis codes.
The method generates two bands with different power-spectral shapes while preserving arbitrary cross-spectral properties, making it suitable for tests of propagating-fluctuation and two-component accretion models.
The spectral-leakage treatment (Appendix D) predicts how segmenting red spectra biases recovered coherence and lags, and offers mitigation by averaging neighboring frequency bins instead of segments.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Because the construction is linear and Gaussian, its 'arbitrary' targets apply only to second-order, linear correlation; real accreting black holes show log-normal and nonlinear variability, so a non-Gaussian extension (e.g., iterative amplitude rescaling) would be needed to make simulated light curves mimic observed flux distributions.
The transfer function R is a mathematical artifact, not a physical response; the paper itself notes it can imply causality violations. A natural next step would be to benchmark whether existing cross-spectral fitting codes recover known input parameters when Poisson noise and dead-time effects are added.
The phase-lag variance diverges as γ²→0, so low-coherence frequency bins carry enormous statistical weight; the delta-method covariance could be turned into a practical approximate likelihood that downweights these bins appropriately.
One could extend the two-band construction to multiple energy bands by treating the transfer functions between reference and each dependent band jointly, but the paper does not derive the resulting multi-band covariance.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Bottom line: this paper gives the X-ray timing community the missing primitive for generating correlated light curves with specified coherence and phase-lag profiles. The construction in Sec. 3 is simple and correct: Y = R X + K noise with R=(P_Y gamma^2/P_X)^{1/2} e^{i phi} and K chosen to preserve P_Y. I re-derived the cross-spectrum and power expectations; they come out as advertised. The delta-method variance and covariance formulas match the Monte Carlo, and Appendix B is internally consistent. The bias check against Nuttall-Carter is a nice touch because it validates that the simulator reproduces not just input spectra but estimator statistics.

What's genuinely new: Timmer-Koenig gives one series; Emmanoulopoulos et al. give non-Gaussianity but no inter-band correlation. Here you get arbitrary P_X, P_Y, gamma^2, and phi in one construction, plus the full 6x6 covariance needed for joint likelihood fits. The paper is explicitly honest about limits: Gaussianity, noiselessness, linearity, and the fact that R is a mathematical transfer function that can violate causality for positive lags. Appendix D on spectral leakage from segmentation is more thorough than most papers would bother with, and the leakage prediction matches their Fig. D1. Code and derivation notebooks are public, which is real evidence.

Soft spots, in proportion: the main limitation is that the algorithm lives entirely in second-order Gaussian statistics. Real accreting sources are log-normal and nonlinear, so the "arbitrary" in the title is arbitrary only within that class. That is disclosed and is not a hidden flaw. Appendix C on detector mixing is more of a review/reminder than a derivation, and the two-source example is degenerate, but it makes the point. The leakage effects in Appendix D are derived for unity coherence and constant phase; the paper acknowledges but doesn't develop the general case. Minor. The Monte Carlo section tests individual Fourier coefficients rather than full light curves with segmentation, but Section 5's ensemble validation covers the full pipeline.

I do not see a load-bearing error. The central claim is that the construction reproduces target second-order spectra in expectation, and it does. The paper deserves a serious referee; the main things a referee would ask for are a bit more discussion of when the Gaussian approximation fails for the covariance and maybe a direct test against a full simulated light curve with Poisson noise, neither of which changes the verdict.

Referee Report

0 major / 7 minor

Summary. The paper presents an algorithm for generating a pair of synthetic time series with user-specified power spectra P_X(ν), P_Y(ν), coherence γ²(ν), and phase-lag profile φ(ν). The dependent series is built as Y = K(H + iJ) + R X, with R fixed by the target coherence and phase and K fixed by the target PSD (Eqs. 10–16). The authors also derive delta-method approximations for the variance of the coherence and phase-lag estimators, construct a 6×6 covariance matrix for the base and derived estimators, and validate the construction with Monte Carlo experiments. Appendices treat detector/discretization effects and spectral leakage from segmenting.

Significance. If the construction is correct, the paper gives a simple, widely applicable tool for validating cross-spectral fitting methods in X-ray timing. The central derivation is transparent and self-contained, and the variance/covariance formulas are practically useful for likelihood construction. Strengths include the explicit closed-form normalization, the Monte Carlo checks against the external Nuttall–Carter bias formula, the detailed leakage analysis in Appendix D, and the public availability of the code and Mathematica derivation notebooks. The contribution is incremental relative to standard bivariate Gaussian spectral simulation, but it is clearly presented and should be useful to the community.

minor comments (7)

[Eq. (14)] The coherence expression is written with E[XY*] rather than E[X*Y]. Since Eq. (3) defines C = E[X*Y], this is notationally inconsistent and, although harmless for the modulus, it conflicts with the phase convention used in Eq. (15). Please use a single convention throughout.
[Section 3.3] The algorithm should state explicitly, before the inverse transform step, that only N/2−1 positive-frequency coefficients are drawn independently and that the remaining coefficients are filled by Hermitian symmetry, with a specification of the DC and Nyquist terms. This is discussed in Appendix D, but the main algorithm as written could produce a complex time series.
[Section 5.1 / Abstract] The text says averaged spectra are computed using 16 s segments (~16 segments per realization), yet the abstract and Fig. 1 claim coverage from 0.01 Hz. A 16 s segment has a fundamental frequency of 0.0625 Hz. Please clarify whether additional longer segments were used, or correct the claimed frequency range.
[Eq. (28)] State the domain of validity of the variance formulas. The phase-lag variance diverges as γ²→0, where the phase is undefined, and the delta-method gradients are not well behaved at γ²=1. The paper acknowledges large-n/Gaussian limitations, but a short boundary caveat would be helpful.
[Section 4.3 / Eq. (B4)] The text says the coherence estimator's correlations with all four base quantities are proportional to γ²(1−γ²), but the covariances with Ĉ_r and Ĉ_i scale as sqrt(γ²(1−γ²)). The wording should be adjusted to match the formulas.
[Section 5.2] The Monte Carlo variance tests draw n=50 independent single-frequency realizations. For segmented red-noise light curves, contiguous segments are not strictly independent, so the effective n in Eq. (28) may differ. Please state that n is the number of independent segments and note the practical caveat for strongly red spectra.
[Sections 1 and 6] The method would benefit from a brief contextual comparison with existing approaches for simulating multivariate time series with specified cross-spectra (e.g., spectral-matrix factorization or Cholesky-type constructions) and with earlier X-ray-specific correlated-light-curve simulators. This would clarify the paper's specific contribution beyond the closed-form normalization and variance analysis.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: target observables are inputs, and the transfer-function normalization is solved from them; Monte Carlo checks are benchmarked against external results.

full rationale

The central derivation is self-contained. Equations (13) and (16) solve K and R algebraically from the user-chosen targets PX, PY, gamma^2, phi: imposing E[YY*]=PY fixes K, and matching |E[XY*]|^2/(PX PY)=gamma^2 with arg E[X*Y]=phi fixes R. The Section 5 'agreement' is therefore the algorithm reproducing its own inputs, which is exactly what a simulator should do, not a fitted parameter disguised as a prediction. The variance/covariance claims are delta-method consequences of the same explicit Gaussian construction, and their Monte Carlo verification is benchmarked against the external Nuttall & Carter (1976) bias formula and Bendat & Piersol (2010), not against values fitted from the simulation. Self-citations (Vaughan & Nowak 1997; Nowak et al. 1999a,b) are contextual astrophysical background and carry no load in the derivation. The paper explicitly discloses the limitations (Gaussianity, second-order linear correlations only, leakage biases, large-n delta-method approximation), so those constraints are stated assumptions rather than hidden circular dependencies. No circular step can be quoted.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No quantities are fitted to data; the target PSDs, coherence, and lags are user inputs. The transfer function R is explicitly a non-physical mathematical construction, not a new entity. The load-bearing postulates are the linear-Gaussian second-order construction and standard asymptotic statistics.

assumptions (4)

domain assumption The dependent series decomposes as y = y_u + (r * x), an uncorrelated component plus a linearly filtered copy of the reference (equation 6).
This captures all inter-band correlation through a single complex transfer function and excludes higher-order or Volterra correlations, which the paper notes in footnote 2 and in the Conclusions.
domain assumption Fourier coefficients are drawn as independent Gaussians following Timmer and Koenig (1995), so the simulated processes are stationary Gaussian.
This gives exact second-order statistics but cannot reproduce log-normal flux distributions; the paper acknowledges this limitation in Section 6.
standard math Delta-method asymptotics (Wolter 2007) with finite-averaging covariance matrices describe estimator variances at large n.
Used for equation (28) and Appendix B; the paper states these are large-n Gaussian approximations and the Monte Carlo results show deviations from Gaussianity.
standard math The Nuttall and Carter (1976) bias formula and Bendat and Piersol (2010) variance results are correct external benchmarks.
These are used to validate the simulated coherence bias and variance; they are assumed without re-derivation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Generation of Correlated Time Series for X-ray Astronomy Applications." pith.science (2026). https://pith.science/paper/QPC7Z5UR

@misc{pith2026260802584,
  author       = {Pith},
  title        = {Pith review of: Generation of Correlated Time Series for X-ray Astronomy Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QPC7Z5UR}},
  note         = {Machine review of arXiv:2608.02584}
}

read the original abstract

Cross-spectral methods have become essential for studying accretion physics in X-ray binaries and active galactic nuclei, where coherence and phase lag measurements constrain physical models and reveal variability components invisible in power spectra alone. Recent multi-Lorentzian fitting techniques have uncovered new quasi-periodic features through joint analysis of power spectra and cross-spectra, but testing these methods requires synthetic data with realistic statistical properties. We present an algorithm for generating pairs of time series with arbitrary power spectra, coherence functions, and phase lag profiles. The method extends existing methods by decomposing the dependent time series into coherent and incoherent components, where the coherent part is constructed through a complex transfer function applied to a reference series. We derive the transfer function and normalization required to preserve target spectral shapes while achieving specified cross-spectral properties. We obtain approximate analytic expressions for the variance of coherence and phase lag estimators and construct the approximate covariance matrix relating these quantities to the underlying power and cross-spectra. When fitting models jointly to power and cross-spectra, these correlations must be incorporated into the likelihood. We apply the method to a two-Lorentzian model with component-specific phase lags and demonstrate close agreement between input models and generated power spectra, coherence function, and phase lag profile across four decades in frequency.

Figures

Figures reproduced from arXiv: 2608.02584 by the authors.

**Figure 1.** Ensemble validation of the simulation method applied to a realistic two-component model. Panels show (top to bottom): reference band PSD PˆX, dependent band PSD Pˆ Y , coherence γˆ 2 , and phase lag ϕˆ. Black points with error bars represent ensemble means ± standard errors over Nreal = 10 independent realizations. Solid red curves show scaled input models; dashed red curves in the PSD panels indicate individual Lor… view at source ↗

**Figure 2.** Joint distribution of the six estimators: base measurements (PˆX, Pˆ Y , Cˆr, Cˆ i), and derived quantities (γˆ 2 , ϕˆ). Distributions generated from 50 000 Monte Carlo realizations each averaging 50 instances of a single frequency. In the off diagonal panels, orange lines mark true values and ellipses show 1σ and 2σ contours from the full 6×6 covariance matrix (equation B6). Gray points each mark a single Monte Car… view at source ↗

**Figure 3.** Bias E[ˆγ 2 ]–γ 2 versus true coherence for different numbers of averaged segments (n = 4, 8, 16, 32). Points show Monte Carlo results from 50 000 trials at each coherence value ranging from 0 to 0.999; dashed lines show the exact theoretical bias from Nuttall & Carter (1976). The bias peaks at γ 2 = 0, where E[ˆγ 2 ] = 1/n, vanishes at γ 2 = 1, and scales approximately as (1 − γ 2 ) 2/n for intermediate coherence. … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 1 linked inside Pith

[1]

Investigatingafluctuating-accretion model for the spectral-timing properties of accreting black hole systems,MNRAS,367(2), 801–814

Arévalo,P.&Uttley,P.,2006. Investigatingafluctuating-accretion model for the spectral-timing properties of accreting black hole systems,MNRAS,367(2), 801–814. Arnaud, K. A.,

2006
[2]

The mixing matrix for the real part of the cross-spectrum is Λ ˆCr = √PX PY 4   2αcosϕ0β0 0 2αcosϕ0β β0 0 0 0β0 0   .(A1) The mixing matrix for the imaginary part of the cross- spectrum is Λ ˆCi = √PX PY 4   2αsinϕ0 0β 0 2αsinϕ−β0 0−β0 0 β0 0 0   .(A2) RASTI000, 1–15 (2026) Correlated Time Series for X-ray Astronomy11 The mixing matrix for...

2026
[4]

For two functionsf(ˆB)and g( ˆB)of the estimator vector, the covariance is Cov(f( ˆB), g(ˆB))≈(∇f) T · Σ n · ∇g(B1) where gradients are evaluated at the true value,B(Graybill 1983)

to compute the co- variances between the derived quantitiesˆγ2 and ˆϕand the base quantities( ˆCr, ˆCi, ˆPX , ˆPY ). For two functionsf(ˆB)and g( ˆB)of the estimator vector, the covariance is Cov(f( ˆB), g(ˆB))≈(∇f) T · Σ n · ∇g(B1) where gradients are evaluated at the true value,B(Graybill 1983). The gradients of the coherence and phase lag with respect ...

1983
[10]

M., 2007.Introduction to variance estimation, Statis- tics for social and behavioral sciences, Springer, New York, 2nd edn

Wolter, K. M., 2007.Introduction to variance estimation, Statis- tics for social and behavioral sciences, Springer, New York, 2nd edn. Zhang, W., Jahoda, K., Swank, J. H., Morgan, E. H., & Giles, A. B.,

2007
[13]

cosϕ 2n p γ2 Cov( ˆϕ, ˆPX )≈0 Cov( ˆϕ, ˆPY )≈0. (B5) Toleadingorderinthedeltamethod,thephaselagisuncorre- lated with both power spectra, consistent with the geometric intuition that the phase depends only on the argument of the cross-spectrum, not on the magnitude. B3 Full Covariance Matrix Combining these results with the covariance matrixΣfrom equation ...

2026
[14]

(C3) The measured signals are proportional to each other at all times, so their coherence is identically unity

y(t) = (αY1 +α Y2 )δ(t−t 0). (C3) The measured signals are proportional to each other at all times, so their coherence is identically unity. The two sources are intrinsically unrelated, yet the measurement pro- cess makes them indistinguishable from a single perfectly co- herent source. No amount of averaging can separate them. This deterministic example ...

1993
[15]

practical steps

and the aliased harmonics (k̸= 0). If the cross-spectral properties at the aliased frequencies differ from those atν, the aliased sum mixes these together and the resulting coherence reflects a weighted average rather than the value atνalone. However, this effect can be diminished by choosing a high sampling rate (see e.g., the discussion in van der Klis ...

1989
[16]

Fi- nally, we will neglect the DC and Nyquist frequency compo- nents becauseW 0,j′ andW N/2,j′ vanish forj ′ ∈[1, Ns/2−1]

We note that the effects dis- cussed in this Appendix are applicable to general coherence and phase lag profiles, however detailed discussion of the more general case is beyond the scope of this Appendix. Fi- nally, we will neglect the DC and Nyquist frequency compo- nents becauseW 0,j′ andW N/2,j′ vanish forj ′ ∈[1, Ns/2−1]. By using Equations 10 and 11,...

2026

Show all 17 references

[1984]

Applications of these methods to synthetic data could produce synthetic timeseries which better match their input properties

and recommended methods of mitigation (Papadakis & Lawrence 1993; Zhu & Xue 2016; Epitropakis & Papadakis 2016, 2017). Applications of these methods to synthetic data could produce synthetic timeseries which better match their input properties. Complementar- ily, the method de...

1993
[1993]

Percival, D

Improved methods for power spectrum modelling of red noise,Monthly Notices of the Royal Astronomical Society,261(3), 612–624. Percival, D. B. & Walden, A. T., 1993.Spectral Analysis for Phys- ical Applications, Cambridge University Press, Cambridge. Pottschmidt, K., Wilms, J.,...

1993
[2003]

Long term variability of Cygnus X-1. I. X-ray spectral-temporal correlations in the hard state,A&A, 407, 1039–1058. Rencher, A. C., 2008.Linear models in statistics, Wiley- Interscience, Hoboken, N.J, 2nd edn. Schetzen, M., 1980.The Volterra and Wiener Theories of Non- linear ...

2008
[2014]

Long term variability of Cygnus X-1. VI. Energy-resolved X-ray variability 1999-2011,A&A, 565, A1. Harris, C. R., Millman, K. J., van der Walt, S. J., Gommers, R., Virtanen, P., Cournapeau, D., Wieser, E., Taylor, J., Berg, S., Smith, N. J., Kern, R., Picus, M., Hoyer, S., van...

1999
[2016]

Epitropakis, A

Statistical properties of Fourier-based time-lag estimates,Astronomy & Astrophysics, 591, A113, arXiv:1604.01726 [astro-ph]. Epitropakis, A. & Papadakis, I. E.,

arXiv
[2019]

Stingray: RASTI000, 1–15 (2026) 10S. R. Larner et al. A Modern Python Library for Spectral Timing,ApJ,881(1),

2026
[2022]

Bachetti, M., Huppenkothen, D., Stevens, A., Swinbank, J., Mas- troserio, G., Lucchini, M., Lai, E

Using the bispectrum to probe radio X-ray correlations in GRS 1915+105,MNRAS,514(2), 1720–1732. Bachetti, M., Huppenkothen, D., Stevens, A., Swinbank, J., Mas- troserio, G., Lucchini, M., Lai, E. V., Buchner, J., Desai, A., Joshi, G., Pisanu, F., Pisupati, S. G. D., Sharma, S....

1915
[2025]

Graybill, F

A hidden quasi-periodic oscillation in Cygnus X-1 revealed by NICER,A&A,696, A237. Graybill, F. A., 1983.Matrices with applications in statistics, Belmont, Calif. : Wadsworth International Group. Grinberg, V., Pottschmidt, K., Böck, M., Schmid, C., Nowak, M. A., Uttley, P., To...

1983
[2026]

Bendat, J

Stingraysoftware/stingray: v2.3.2. Bendat, J. & Piersol, A., 2010.Random Data: Analysis and Mea- surement Procedures, Fourth Edition, Wiley. Dauser, T., Falkner, S., Lorenz, M., Kirsch, C., Peille, P., Cuc- chetti, E., Schmid, C., Brand, T., Oertel, M., Smith, R., & Wilms, J.,

2010

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [1]

Investigatingafluctuating-accretion model for the spectral-timing properties of accreting black hole systems,MNRAS,367(2), 801–814

Arévalo,P.&Uttley,P.,2006. Investigatingafluctuating-accretion model for the spectral-timing properties of accreting black hole systems,MNRAS,367(2), 801–814. Arnaud, K. A.,

2006

[2] [2]

The mixing matrix for the real part of the cross-spectrum is Λ ˆCr = √PX PY 4   2αcosϕ0β0 0 2αcosϕ0β β0 0 0 0β0 0   .(A1) The mixing matrix for the imaginary part of the cross- spectrum is Λ ˆCi = √PX PY 4   2αsinϕ0 0β 0 2αsinϕ−β0 0−β0 0 β0 0 0   .(A2) RASTI000, 1–15 (2026) Correlated Time Series for X-ray Astronomy11 The mixing matrix for...

2026

[3] [4]

For two functionsf(ˆB)and g( ˆB)of the estimator vector, the covariance is Cov(f( ˆB), g(ˆB))≈(∇f) T · Σ n · ∇g(B1) where gradients are evaluated at the true value,B(Graybill 1983)

to compute the co- variances between the derived quantitiesˆγ2 and ˆϕand the base quantities( ˆCr, ˆCi, ˆPX , ˆPY ). For two functionsf(ˆB)and g( ˆB)of the estimator vector, the covariance is Cov(f( ˆB), g(ˆB))≈(∇f) T · Σ n · ∇g(B1) where gradients are evaluated at the true value,B(Graybill 1983). The gradients of the coherence and phase lag with respect ...

1983

[4] [10]

M., 2007.Introduction to variance estimation, Statis- tics for social and behavioral sciences, Springer, New York, 2nd edn

Wolter, K. M., 2007.Introduction to variance estimation, Statis- tics for social and behavioral sciences, Springer, New York, 2nd edn. Zhang, W., Jahoda, K., Swank, J. H., Morgan, E. H., & Giles, A. B.,

2007

[5] [13]

cosϕ 2n p γ2 Cov( ˆϕ, ˆPX )≈0 Cov( ˆϕ, ˆPY )≈0. (B5) Toleadingorderinthedeltamethod,thephaselagisuncorre- lated with both power spectra, consistent with the geometric intuition that the phase depends only on the argument of the cross-spectrum, not on the magnitude. B3 Full Covariance Matrix Combining these results with the covariance matrixΣfrom equation ...

2026

[6] [14]

(C3) The measured signals are proportional to each other at all times, so their coherence is identically unity

y(t) = (αY1 +α Y2 )δ(t−t 0). (C3) The measured signals are proportional to each other at all times, so their coherence is identically unity. The two sources are intrinsically unrelated, yet the measurement pro- cess makes them indistinguishable from a single perfectly co- herent source. No amount of averaging can separate them. This deterministic example ...

1993

[7] [15]

practical steps

and the aliased harmonics (k̸= 0). If the cross-spectral properties at the aliased frequencies differ from those atν, the aliased sum mixes these together and the resulting coherence reflects a weighted average rather than the value atνalone. However, this effect can be diminished by choosing a high sampling rate (see e.g., the discussion in van der Klis ...

1989

[8] [16]

Fi- nally, we will neglect the DC and Nyquist frequency compo- nents becauseW 0,j′ andW N/2,j′ vanish forj ′ ∈[1, Ns/2−1]

We note that the effects dis- cussed in this Appendix are applicable to general coherence and phase lag profiles, however detailed discussion of the more general case is beyond the scope of this Appendix. Fi- nally, we will neglect the DC and Nyquist frequency compo- nents becauseW 0,j′ andW N/2,j′ vanish forj ′ ∈[1, Ns/2−1]. By using Equations 10 and 11,...

2026