Pith. sign in

REVIEW 1 major objections 4 minor 25 references

Number of Sign Changes: Segment of AR(1)

T0 review · 1 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read This paper derives an exact closed-form formula for the variance of the number of sign changes in four observations of a stationary AR(1) Gaussian process.

desk verdict A careful, narrow computation of the variance of sign changes in AR(1) segments; the n=4 formula is new and checkable, but the paper should state the extension of Cheng's integral to negative arguments. read the letter →

arxiv 1909.02556 v1 pith:C2THH45Q submitted 2019-09-05 math.HO math.STstat.TH

Steven Finch This is my paper · ORCID
classification math.HOmath.STstat.TH MSC 60G1062M1033B30
keywords signchangesAR(1)processorthantprobabilitiesdilogarithmvarianceindependentintervalapproximationzerocrossingsGaussiantimeseries
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 establishes an exact closed-form formula for the variance of the number of sign changes among four consecutive observations of a stationary first-order autoregressive Gaussian series with correlation parameter ρ. The formula is assembled from quadrivariate normal orthant probabilities, each expressed through a dilogarithm integral, and it supplements the known linear mean formula. The resulting variance is symmetric in ρ, attains its maximum 3/4 at ρ=0, and takes the value 0.721407566... at ρ=1/2. Against the independent-interval approximation, the new expression matches exactly for segments of length two and three and comes very close for length four, with the largest discrepancy near ρ≈-0.897.

What carries the argument

The argument turns on a dilogarithm integral I(h,x) = ∫₀ˣ arcsin((1−h²)t/(h²−t²))/√(1−t²) dt, expressed as a sum of complex dilogarithms. Two orthant-probability functions f(a,b) and g(a,b) are defined from this integral for correlation matrices with entries a, b, ab, and a²b. Every sign pattern in a length-four AR(1) segment, after flipping coordinates, has a covariance matrix of one of those two forms, so every p_{e1 e2 e3 e4} reduces to f or g with arguments ±ρ. Summing the resulting probabilities with the appropriate multiplicities yields the variance.

What would settle it

Compute V(S4) at ρ=-0.5 by direct high-precision numerical integration of the four-dimensional Gaussian orthant probabilities p_{e1e2e3e4}, without using the dilogarithm formula, and compare with the paper's expression; any disagreement beyond integration error refutes it. The same check can be done at ρ=-0.897, where the approximation gap is largest, since a mismatch there would localize the failure.

Watch

Extended reading notes

Core claim

The central result is that V(S4) = 4g(ρ,ρ)+2f(ρ,−ρ)+16g(ρ,−ρ)+8f(−ρ,ρ)+18f(−ρ,−ρ) − 9[1/2 − arcsin(ρ)/π]^2, where f and g denote probabilities that four variables with two specific correlation matrices are all positive. These are evaluated in closed form via a dilogarithm integral, so the variance itself is a closed expression in ρ for every |ρ|<1. The paper verifies symmetry, the maximum 3/4 at ρ=0, and numerical agreement at ρ=1/2, and contrasts the formula with the independent-interval approximation: the approximation is exact for n=2 and n=3 variances and nearly exact for n=4, with the biggest gap of about 0.036 near ρ=-0.897 and about 0.002 near ρ=0.763.

Load-bearing premise

The entire variance expression relies on the dilogarithm evaluation of I(h,x) being correct for the parameter ranges needed, including negative argument ρ; the paper cites this evaluation without proving it and even notes a typo in the source, so if the corrected identity fails for some |ρ|<1, the central formula collapses.

Editorial extensions

If this is right

  • For n=4, the exact variance of sign changes is a closed function of ρ, so no simulation or quadrature is needed for this sample size.
  • The variance is symmetric under ρ→−ρ, meaning short sign-change counts cannot by themselves distinguish positive from negative serial correlation.
  • The independent-interval approximation exactly reproduces the variance for n=2 and n=3 and is nearly exact for n=4, supporting its use for short AR(1) segments.
  • The appendix identifies the two orthant probabilities needed for n=5 that do not fit the f/g pattern, explaining why closed-form expressions stop at n=4.
  • The mean formula (n−1) arccos(ρ)/π remains the expectation for all n, consistent with the classical zero-crossing theorem in discrete time.

Reading between the lines

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

  • A similar f/g sign-flip decomposition should extend to other local pattern counts in AR(1) segments, such as peaks or turning points, where the same covariance forms arise, though this is not pursued in the paper.
  • The exact V(S4) formula gives a cheap benchmark for validating numerical routines for quadrivariate normal probabilities, especially at negative ρ where the approximation is weakest.
  • The IIA model's unexplained asymmetry at negative ρ could be tested directly by substituting |ρ| in the recursion; if the fit improves, the model's physical derivation likely assumed positive correlation.
  • Because the two intractable n=5 orthant probabilities depend on a double-parameter integral J(h,k,x), symbolic progress for n≥5 may require evaluating J at h≠k, which the paper leaves open.
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.

Referee Report

1 major / 4 minor

Summary. The manuscript derives exact formulas for the mean and variance of the number of sign changes in a short segment (n ≤ 4) of a stationary Gaussian AR(1) process with correlation ρ. For n = 2 and n = ít3 the paper recovers known results; for n = 4 it presents a closed-form expression for V(S4) in terms of quadrivariate normal orthant probabilities f and g, which are built from Cheng's dilogarithm evaluation of a one-dimensional integral I(h,x). The paper then compares V(S4) with the independent-interval approximation (IIA) of Nyberg et al., finds close agreement for positive ρ and a small discrepancy for negative ρ, and explains that the IIA asymmetry for negative ρ may indicate an assumption ρ > 0 in the model. An appendix discusses the additional integral needed for n = 5 and reports that it resists closed-form evaluation except in a special case.

Significance. If correct, the V(S4) formula is a useful exact benchmark for zero-crossing statistics of AR(1) processes, extending known results beyond n = 3. The derivation is transparent and parameter-free: it uses only standard orthant probability formulas and Cheng's integral, with no fitting and no use of the model to define theory. The paper is commendably explicit about the assumptions behind the IIA comparison and about numerical verification using R's pmvnorm. However, the exactness of V(S4) rests on a domain extension that is not justified in the manuscript, so the central claim is not fully established as written.

major comments (1)
  1. [Section 1, display for I(h,x) and display for V(S4)] The quoted evaluation of I(h,x) is stated for 0 < x < h^2 < 1, but the displayed formula for V(S4) immediately applies it to f(ρ,−ρ), g(ρ,−ρ), and f(−ρ,−ρ), which involve I(ρ,−ρ^3) and I(−ρ,−ρ^3), i.e., negative upper limits. The paper does not justify the analytic continuation to negative x, nor does it give a domain condition for f and g beyond |a| < 1 and |b| < 1. The numerical check at ρ = 1/2 involves one negative argument (x = −1/8) and is reassuring, but it is not a proof for all ρ ∈ (−1,1). Because V(S4) is the central new result, this is a load-bearing gap. The author should add a short argument (e.g., the integrand is odd, so I(h,−x) = I(h,x) if the right-hand side is continued accordingly) or cite a source that covers −h^2 < x < 0, and state the branch of Li2 used for negative x.
minor comments (4)
  1. [Section 2, Figure 1] The figure is not included in the manuscript text, so the quantitative claims about the largest separation between theory and IIA cannot be verified by the reader; providing the figure or a data table would improve reproducibility.
  2. [Introduction, paragraph 1] The sentence 'Closed-form variance expressions become impossible for n ≥ 5 (see the appendix)' overstates what is shown; the appendix demonstrates only that one particular integral J(ρ, ρ^2, ρ^4) is not evaluated in closed form. Suggest rewording to 'we were unable to obtain' or 'no closed form is known to us'.
  3. [Abstract and body] There are minor typographical errors, e.g., 'mom ents' in the abstract, and some covariance matrices are typeset with misalignments; these should be corrected.
  4. [Section 1 and Acknowledgments] The acknowledgments state that R's pmvnorm was used for verification, but no direct numerical comparison is shown in the text. A small table comparing the closed-form V(S4) with pmvnorm over a grid of ρ values, including negative ρ, would substantially strengthen the evidence for the extension of I(h,x).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the exact variance formula is derived from orthant probabilities and an independent cited integral evaluation, not from the model it is compared against.

full rationale

The paper's central derivation obtains V(S4) by enumerating all sign-change patterns for a four-dimensional Gaussian segment, expressing the required orthant probabilities as f(a,b) and g(a,b), and substituting Cheng's independently published dilogarithm evaluation of I(h,x). No parameter is fitted to V(S4), and the independent-interval approximation (IIA) from Nyberg, Lizana and Ambjörnsson is used only as a benchmark for comparison, not as an ingredient of the theoretical formula. The paper explicitly derives E(S2), V(S2), E(S3), and V(S3) from elementary bivariate and trivariate orthant probabilities before invoking the more difficult quadrivariate case, and the n=5 appendix treats further orthant integrals as unresolved rather than importing the target result. The only substantive concern is a proof gap, not circularity: the cited Cheng integral is stated for 0 < x < h^2 < 1, whereas several f and g terms in V(S4) require I(a,a^2 b) with a^2 b = -rho^3 < 0, and the paper does not prove the needed extension to negative x. This affects correctness and rigor of the analytic continuation, but it does not make the derivation equivalent to its inputs or reduce a prediction to a fitted parameter. Self-citations in the references are background pointers and do not carry the load-bearing argument. Accordingly, the circularity score is 0.

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

The paper introduces no new fitted parameters or invented entities. The central exact claim depends on standard orthant probability results and Cheng's integral evaluation, plus the IIA recursion from Nyberg et al. for the model comparison. Rho is the autocorrelation parameter of the model, not fitted.

assumptions (4)
  • domain assumption The segment (X1,...,Xn) is multivariate normal with mean vector 0 and covariance matrix R_ij = ρ^{|i-j|}.
    This defines the AR(1) model used throughout the paper (Section 1, first paragraph).
  • domain assumption Cheng's evaluation of I(h,x) as a sum of complex dilogarithms is correct for 0 < x < h^2 < 1 and extends appropriately to negative ρ.
    This is the load-bearing integral formula used to define f(a,b) and g(a,b) in Section 1; the paper cites [6], notes a typo in the source, and does not prove it.
  • standard math Orthant probability formulas for the quadrivariate normal with covariance structures R+ and R− are valid.
    These formulas, from the orthant probability literature cited in [1-5], are used to express all 16 sign probabilities p_e as f and g functions in Section 1.
  • domain assumption The independent interval approximation recursion for c_n from [17] describes the second moment of sign-change counts in AR(1).
    Used in Section 2 to generate model predictions; the paper notes the assumption is not universally valid and that negative ρ behavior is unresolved.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Number of Sign Changes: Segment of AR(1)." pith.science (2026). https://pith.science/paper/C2THH45Q

@misc{pith2026190902556,
  author       = {Pith},
  title        = {Pith review of: Number of Sign Changes: Segment of AR(1)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C2THH45Q}},
  note         = {Machine review of arXiv:1909.02556}
}
abstract

Let $X_{t}$ denote a stationary first-order autoregressive process. Consider $n$ contiguous observations (in time $t$) of the series (e.g., $X_{1}, ..., X_{n}$). Let its mean be zero and its lag-one serial correlation be $\rho$, which satisfies $|\rho| < 1$. Rice (1945) proved that $(n-1) \arccos(\rho)/\pi$ is the expected number of sign changes. A corresponding formula for higher-order moments was proposed by Nyberg, Lizana & Ambj\"ornsson (2018), based on an independent interval approximation. We focus on the variance only, for small $n$, and see a promising fit between theory and model.

Figures

Figures reproduced from arXiv: 1909.02556 by the authors.

Figure 1
Figure 1. The red curve is an IIA-based model prediction of varianc [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

  1. [1]

    S. S. Gupta, Probability integrals of multivariate normal and multiv ariate t, Annals Math. Statist. 34 (1963) 792–828; MR0152068 [formulas (41) & (42)]

    work page 1963

  2. [2]

    S. S. Gupta, Bibliography on the multivariate normal integrals and related topics, Annals Math. Statist. 34 (1963) 829–838; MR0152069

    work page 1963

  3. [3]

    D. B. Owen, Orthant probabilities, Encyclopedia of Statistical Sciences , v. 6, ed. S. Kotz, N. L. Johnson and C. B. Read, Wiley, 1985, pp. 521–523; M R0873585

    work page 1985

  4. [4]

    Y. L. Tong, The Multivariate Normal Distribution , Springer-Verlag, 1990, pp. 188-190; MR1029032

    work page 1990

  5. [5]

    Genz and F

    A. Genz and F. Bretz, Computation of Multivariate Normal and t Probabilities , Lect. Notes in Statist 195, Springer-Verlag, 2009, pp. 11-12; MR 2840595

    work page 2009

  6. [6]

    M. C. Cheng, The orthant probabilites of four Gaussian variates , Annals Math. Statist. 40 (1969) 152–161; MR0235596 [error in formula (2.21) requires c orrec- tion: product arcsin( ρ12) arcsin(ρ23) should be arcsin( ρ12) arcsin(ρ34)]

    work page 1969

  7. [7]

    M. C. Cheng, Output autocorrelation functions of smooth and h ard limiters, Internat. J. Control 7 (1968) 223–240

    work page 1968

  8. [8]

    M. C. Cheng, On a class of integrals expressible in terms of arcsin( x), Li 2(x) and Li2(r, θ), Nanta Math. 4 (1970) 113–116; MR0304709

    work page 1970

Show all 25 references
  1. [9]

    Ni and B

    Z. Ni and B. Kedem, On normal orthant probabilities, Chinese J. Appl. Probab. Statist., v. 15 (1999) n. 3, 262–275; MR1771106

    1999

  2. [10]

    S. O. Rice, Mathematical analysis of random noise, Bell System Tech. J. 23 (1944) 282–332; 24 (1945) 46–156; also in Selected Papers on Noise and Stochastic Processes, ed. N. Wax, Dover, 1954, pp. 133–294; MR0010932 and MR00119 18

    1944

  3. [11]

    S. R. Finch, Zero crossings, Mathematical Constants II , Cambridge Univ. Press, 2019, pp. 479–485; MR2003519

    2019

  4. [12]

    Kedem, Time Series Analysis by Higher Order Crossings , IEEE Press, 1994, pp

    B. Kedem, Time Series Analysis by Higher Order Crossings , IEEE Press, 1994, pp. 115–143; MR1261636

    1994

  5. [13]

    J. T. Barnett, Zero-crossings of random processes with app lication to estimation and detection, Nonuniform Sampling: Theory and Practice , ed. F. Marvasti, Kluwer/Plenum, 2001, pp. 393–435; MR1875683. Number of Sign Changes: Segment of AR(1) 12

    2001

  6. [14]

    A. J. F. Siegert, On the first passage time probability problem, Phys. Rev. 81 (1951) 617–623; MR0054192

    1951

  7. [15]

    J. A. McFadden, The axis-crossing intervals of random functio ns. II, IEEE Trans. Inform. Theory (1958) 14–24; MR0098437

    1958

  8. [16]

    Nyberg, T

    M. Nyberg, T. Ambj¨ ornsson and L. Lizana, A simple method to c alculate first- passage time densities with arbitrary initial conditions, New J. Phys. 18 (2016) 063019

    2016

  9. [17]

    Nyberg, L

    M. Nyberg, L. Lizana and T. Ambj¨ ornsson, Zero-crossing st atistics for non- Markovian time series, Phys. Rev. E 97 (2018) 032114; arXiv:1711.02926 [beware of typo in preprint version: plus sign should be minus sign in formulas (3 ) & (25)]

    2018 arXiv

  10. [18]

    F. N. David, A note on the evaluation of the multivariate normal in tegral, Biometrika 40 (1953) 458–459; MR0058314

    1953

  11. [19]

    Guillaume, Computation of the quadrivariate and pentavariat e normal cumu- lative distribution functions, Comm

    T. Guillaume, Computation of the quadrivariate and pentavariat e normal cumu- lative distribution functions, Comm. Statist. Simulat. Comput. 47 (2018) 839– 851; MR3810596

    2018

  12. [20]

    Dieckmann, Table of Indefinite Integrals, http://www-elsa.physik.uni-bonn.de/˜dieckman/IntegralsIndefinite/IndefInt.html

    A. Dieckmann, Table of Indefinite Integrals, http://www-elsa.physik.uni-bonn.de/˜dieckman/IntegralsIndefinite/IndefInt.html

  13. [21]

    X. Mi, T. Miwa and T. Hothorn, New numerical algorithm for multiva riate normal probabilities in package mvtnorm, The R Journal 1 (2009) pp. 37–39; http://journal.r-project.org/archive/2009-1/

    2009

  14. [22]

    S. M. Stigler, Estimating serial correlation by visual inspection o f diagnostic plots, Amer. Statist. 40 (1986) 111–116; MR0841577

    1986

  15. [23]

    Ku and E

    S. Ku and E. Seneta, The number of peaks in a stationary sample and orthant probabilities, J. Time Series Analysis 15 (1994) 385–403; MR1292616

    1994

  16. [24]

    Finch, Moments of maximum: Segment of AR(1), arXiv:1908.04 179

    S. Finch, Moments of maximum: Segment of AR(1), arXiv:1908.04 179

    1908

  17. [25]

    Finch, Another look at AR(1), arXiv:0710.5419

    S. Finch, Another look at AR(1), arXiv:0710.5419. Steven Finch MIT Sloan School of Management Cambridge, MA, USA steven finch@harvard.edu

    arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.