Pith. sign in

REVIEW 3 major objections 4 minor 39 references

Functional Limit Theorems for Marked Hawkes Point Measures

T0 review · 3 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 Marked Hawkes point measures fluctuate as Gaussian white noise plus a common Brownian factor.

desk verdict A genuine new functional CLT for marked Hawkes measures, held back by a load-bearing typo in the Volterra representation and an unchecked dependency on an unpublished preprint. read the letter →

arxiv 1908.06703 v1 pith:JWX7LQVW submitted 2019-08-19 math.PR

Ulrich Horst , Wei Xu This is my paper · ORCID
classification math.PR MSC 60G5760F1792B0592D25
keywords HawkespointmeasuremarkedprocessfunctionalcentrallimittheoremlawoflargenumbersshotnoisestochasticVolterraequationGaussianwhitebuddingmicrobesinahost
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

The paper establishes a functional law of large numbers and a functional central limit theorem for marked Hawkes point measures: random counting measures in which each event carries a mark and the arrival rate depends on all past events and their marks. Under a stability condition, the rescaled point measure converges to a deterministic linear growth, and the fluctuation process converges to a Gaussian white noise plus a lifting of a one-dimensional Brownian motion. The Brownian term is a common factor across all mark sets and encodes the self-exciting feedback of event arrivals. If the result is correct, it gives explicit covariance formulas for the fluctuations of Hawkes-type counting processes and their shot noise, and it upgrades known central limit theorems to functional ones with applications to microbial population dynamics.

What carries the argument

The central object is a stochastic Volterra representation of the intensity process: $Z(t)=\mu_0(t)+\int_0^t R_H(t-s)\mu_0(s)\,ds+\int_0^t\int_U R(t-s,u)\,N_I(ds,du)+\int_0^t\int_U\int_0^{Z(s-)} R(t-s,u)\,\tilde N_0(ds,du,dz)$, with $R=\varphi+R_H*\varphi$ the mean-impact kernel that resolves the branching structure. This representation rewrites the intensity as drift plus stochastic integrals against martingale measures; the martingale integrals are approximated by $\|R(u)\|_{L^1}$-weighted martingales, whose weak convergence yields both the Gaussian white noise and the Brownian component. The function $R(t,u)$ therefore carries the decomposition of fluctuations into mark-specific white noise and a common Brownian factor.

What would settle it

Set $U=\{1,2\}$, take $\varphi(t,u)=a_u e^{-b_u t}$ and constant immigration, and simulate the marked Hawkes process under Condition 3.2; the theorem predicts that the covariance of $\sqrt{T}\,\bar N^T_{H,t}(\{1\})$ and $\sqrt{T}\,\bar N^T_{H,t}(\{2\})$ converges to $\sigma_Z^2\,\nu_H(\{1\})\nu_H(\{2\})\,t$, so a systematic discrepancy beyond Monte Carlo error would refute the claim.

Watch

Extended reading notes

Core claim

The central claim is that, under Condition 3.2, $\sqrt{T}\,\bar N^T_{H,t}(A)$ converges weakly in $D([0,\infty),S(U))$ to $W_{H,t}(A)+\sigma_Z B_Z(t)\nu_H(A)$, where $W_H$ is a Gaussian white noise on $U$ with intensity $\lambda_I\|R_I\|_{L^1}\,dt\,\nu_H(du)$ and $B_Z$ is a standard Brownian motion built from both the Hawkes white noise and an independent immigration white noise. The variance is $\sigma_Z^2=\lambda_I\big(\|R_I\|_{L^1}\nu_H(\|\varphi(\cdot)\|_{L^1}^2)+\nu_I(\|\varphi(\cdot)\|_{L^1}^2)\big)\big/|1-\|\varphi_H\|_{L^1}|^2$. The same decomposition drives functional central limit theorems for shot noise processes and, in the budding-microbe application, for the joint process of cumulative budding rate and total toxin release.

Load-bearing premise

The load-bearing premise is the stochastic Volterra representation of the intensity process taken from an earlier preprint; if the hypotheses of that theorem fail for a given kernel, the proof of the functional limit theorems no longer goes through.

Editorial extensions

If this is right

  • The explicit limit gives covariance formulas: for disjoint mark sets $A_1,A_2$, the asymptotic covariance of the two scaled count fluctuations is $\sigma_Z^2\,t\,\nu_H(A_1)\nu_H(A_2)$, so correlatedness across marks is governed by the Brownian factor.
  • The functional CLT for shot noise processes does not rely on the law of large numbers; the LLN follows as a corollary, making the fluctuation result the primary statement.
  • For standard marked Hawkes processes with constant immigration, the result yields a functional CLT whose one-dimensional version recovers a known CLT for marked Hawkes processes.
  • In the budding-microbe model, the theorem provides a joint functional CLT for cumulative budding rate and total toxin, and, when toxins are released at unit rate, a functional CLT for the integral of population, extending a previously known CLT to a process-level statement.

Reading between the lines

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

  • The paper leaves implicit that the covariance between disjoint mark sets is purely Brownian, which gives a testable factor structure for Hawkes-type data.
  • The same Volterra-martingale route would likely produce an analogous decomposition for other self-exciting point processes whenever a resolvent kernel exists, even outside the Hawkes-specific branching structure.
  • The long-memory remark at the end of the paper points to a concrete open test: replace the light-tail kernel condition by regular variation and check whether the Brownian factor becomes a Gaussian process with Riemann-Liouville-type covariance.
  • A direct practical use is to turn the explicit variance formula into finite-sample confidence bands for estimated Hawkes intensities; the formula predicts the exact $1/\sqrt{T}$ scaling of estimation error under self-excitation.
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

3 major / 4 minor

Summary. The paper studies marked Hawkes point measures with homogeneous immigration and establishes functional laws of large numbers and functional central limit theorems. Under moment and tail conditions (Condition 3.2) and the stability condition ‖φ_H‖_{L1}<1, the main result (Theorem 3.11) states that the rescaled point-measure error process √T \bar N^T_{H,t}(A) converges weakly in D([0,∞),S(U)) to W_{H,t}(A)+σ_Z B_Z(t)ν_H(A), where W_H is a Gaussian white noise, B_Z is a standard Brownian motion, and the covariance is given explicitly. The proof proceeds through a stochastic Volterra representation of the intensity process (Proposition 4.1), followed by martingale approximations and a functional CLT for the cumulative intensity (Proposition 3.10). Functional CLTs are also proved for the associated shot noise processes (Theorem 3.13), and the results are applied to toxin accumulation in budding microbial populations in a host (Section 5).

Significance. If the main theorem is correct, this is a substantial contribution: it gives a genuinely measure-valued functional CLT for marked Hawkes processes with explicit covariance structure, separating the mark-wise white-noise fluctuations from the common Brownian fluctuation induced by self-excitation. It extends known CLTs for marked Hawkes processes (Karabash–Zhu) and provides new functional CLTs for Hawkes shot noise and for the integral of population in a microbial dynamics model. The paper contains detailed moment estimates and explicit constants, and the limiting objects are stated directly in terms of the model parameters with no fitted constants. The significance is tempered, however, by the fact that the central representation is imported from the authors' own unpublished preprint [39] and by an internal inconsistency in the displayed form of that representation.

major comments (3)
  1. [Section 4, Prop. 4.1 and Eq. (4.6)] The representation (4.6) as printed is not the representation used in the rest of the proof. Its second term is ∫_0^t R_H(t-s) µ0(t) ds, with the terminal time t inside the integrand, and the same terminal-time form appears in the first term of (4.3). Equation (4.7), which is obtained by integrating (4.6), uses instead the causal term ∫_0^{Tt} R_H(Tt-s) ds ∫_0^s µ0(r) dr, i.e. the integral of ∫_0^t R_H(t-s) µ0(s) ds. All subsequent estimates (Lemmas 4.2–4.4, Propositions 4.6–4.11) and the proof of Proposition 3.10 rely on the causal form. This is a load-bearing identity: the displayed equations must be corrected to the intended form, or, if the terminal-time form is genuinely what [39] supplies, the switch between the two representations must be justified. As it stands, the proof of the functional CLT is not self-consistent.
  2. [Section 4, Prop. 4.1 / reference [39]] Proposition 4.1 is imported from Theorem 2.2 of the authors' unpublished preprint [39], but the hypotheses of that theorem are neither stated nor checked against Condition 3.2. Because every martingale approximation in Section 4 starts from Eq. (4.6), the central proof is conditional on an unverified external result. The authors should prove Proposition 4.1 in this paper, or provide a complete verification that the conditions of [39, Theorem 2.2] follow from Condition 3.2, including the two-parameter kernel R and the relevant integrability conditions. Relying on an inaccessible self-cited result is not sufficient for a self-contained proof of the main theorem.
  3. [Section 4.1, proof of Proposition 3.9 / Theorem 3.11] The proof of Proposition 3.9 verifies convergence of the real-valued processes W^f_T for each f∈B(U) and identifies the covariance structure of the limit, but it does not explicitly establish tightness of the S(U)-valued processes needed for weak convergence in D([0,∞),S(U)). Since Theorem 3.11 is the central claim, the authors should supply the tightness argument for the measure-valued sequence, or give a precise reference showing that the one-dimensional projections established in the proof imply tightness in D([0,∞),S(U)).
minor comments (4)
  1. [Example 3.16] The formula for |c_m|² contains the denominator |1−∑_{i=1}^d ‖φ(i)‖_{L1}|³, whereas the stability condition and the displayed |c_k|² both involve the weighted norm ∑_{i=1}^d ν_H({i})‖φ(i)‖_{L1}. As printed, the two denominators are inconsistent and the common-factor variance should be computed with the weighted norm.
  2. [Section 4.1, proof of Proposition 3.9] The independence argument at the end of the proof refers to 'N0(ds,du,dz) and N1(ds,du)'; the process N1 is never defined and should be N_I(ds,du).
  3. [Section 4.2, Lemma 4.8] In the statement of inequality (4.29), the integration variable in the last term is written 'ν_H(dy)' while earlier in the same equation it is 'ν_H(du)'; this should be made uniform.
  4. [Section 5, Theorem 5.2] The symbol Φ T_i in the display following (5.9) appears to be a typo for Φ_i; the superscript T is otherwise unexplained.

Circularity Check

1 steps flagged · score 4.0 of 10

Central CLT statement is not fitted or renamed, but its proof rests on a load-bearing Volterra representation imported from the same authors' unpublished preprint [39].

  1. uniqueness imported from authors [Section 4, Proposition 4.1 / Eq. (4.6), used in Section 4.1 proofs of Proposition 3.10 and Theorem 3.11]
    "Applying Theorem 2.2 in [39], we see that {Z(t,u′) : t ≥ 0,u′ ∈ U} also solves the following stochastic Volterra-Fredholm integral equation: … Armed with the representation (4.6), we can now give the proofs of Lemma 3.4, 3.8 and Proposition 3.9, 3.10."

    The cumulative-intensity CLT (Prop. 3.10) and hence the point-measure CLT (Thm. 3.11) are proved by analyzing the stochastic integrals in (4.8), obtained by integrating (4.6). Equation (4.6) is not derived in this paper; the text says only 'Applying Theorem 2.2 in [39]', and the hypotheses of that theorem are never checked against Condition 3.2. Since [39] is an unpublished preprint by one of the present authors, the proof's load-bearing input is a same-author citation chain rather than an independently verified or internally derived result. The CLT limit itself is still expressed explicitly in model parameters, so this is partial, not total, circularity.

full rationale

Theorem 3.11 is assembled from Propositions 3.9 and 3.10 through the decomposition (3.26). Proposition 3.9 is proved from the martingale CLT using the Poisson representation (2.2) and Lemma 2.1, with no fitted or renamed input. Proposition 3.10 uses (4.8), where the stochastic integrals come from the Volterra representation (4.6); the moment and error estimates in Lemmas 4.2-4.4 and 4.8 all rely on that representation. The only load-bearing imported item is therefore Proposition 4.1/(4.6), obtained by citing the same author's Theorem 2.2 in [39]. This is a self-citation chain rather than a restatement of the target CLT, so the central claim retains independent mathematical content; no fitted parameters appear anywhere and the limit processes and covariances are functions of the model primitives (lambda_I, nu_H, nu_I, phi, mu_0). I also flag a non-circular correctness risk: as printed, Eq. (4.6) contains the noncausal term integral_0^t R_H(t-s)mu_0(t)ds, whereas the subsequent integration in Eq. (4.7) uses the causal form integral_0^{Tt} R_H(Tt-s) ds integral_0^s mu_0(r)dr. This inconsistency does not by itself make the derivation circular, but it strengthens the need to verify the imported representation before the proof is complete. Overall: partial circularity due to load-bearing self-citation, not due to fitting or definitional equivalence.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard point-process machinery, a subcriticality condition, moment-tail conditions that are technical and model-specific, and a stochastic Volterra representation from the authors' own unpublished preprint. No free parameters are fitted; the limit constants are explicit functions of the model primitives.

assumptions (5)
  • standard math There exists an extension of the probability space with a time-homogeneous Poisson random measure N0 such that Eq. (2.2) holds.
    Used to represent the Hawkes measure NH as an integral against N0 with intensity Z(s-)du dz; standard Poisson thinning, cf. Ikeda and Watanabe.
  • domain assumption Stability condition ||phi_H||_{L1} < 1, where phi_H(t) = E[phi(t, xi_1)].
    Ensures the mean intensity E[Z(t)] stays bounded (Lemma 2.1); without it the point process may be explosive and no LLN/CLT holds.
  • ad hoc to paper Condition 3.2: uniform 2alpha-moments of mu_0 and polynomial tail conditions on phi with theta_0 > alpha/(2alpha-2).
    Technical condition used throughout the proofs to control error terms; not derived from the model.
  • domain assumption Theorem 2.2 in Xu [39] gives the stochastic Volterra-Fredholm equation (4.3) for the Hawkes random measure representation.
    Imported from the same authors' preprint; the applicability to the two-parameter process in (4.1) is not verified in the text.
  • ad hoc to paper Condition 3.6 on the shot shape function psi: 2alpha-moments and tail decay with theta_1.
    Technical conditions for the shot noise CLT, analogous to Condition 3.2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Functional Limit Theorems for Marked Hawkes Point Measures." pith.science (2026). https://pith.science/paper/JWX7LQVW

@misc{pith2026190806703,
  author       = {Pith},
  title        = {Pith review of: Functional Limit Theorems for Marked Hawkes Point Measures},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JWX7LQVW}},
  note         = {Machine review of arXiv:1908.06703}
}
read the original abstract

This paper establishes a functional law of large numbers and a functional central limit theorem for marked Hawkes point measures and their corresponding shot noise processes. We prove that the normalized random measure can be approximated in distribution by the sum of a Gaussian white noise process plus an appropriate lifting map of a correlated one-dimensional Brownian motion. The Brownian results from the self-exiting arrivals of events. We apply our limit theorems for Hawkes point measures to analyze the population dynamics of budding microbes in a host.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages

  1. [39]

    Scaling Limits for Crump-Mode-Jagers Processes with Immigration via Stochastic Volterra Equations

    W. Xu, Scaling limits for Crump-Mode-Jagers processes with immig ration via stochastic Volterra equations, arXiv: 1809.05931v3

    work page Pith review arXiv

  2. [1]

    Ackermann, S

    M. Ackermann, S. C. Stearns, and U. Jenal , Senescence in a bacterium with asymmetric division , Science, 300 (2003), pp. 1920–1920. 27

    work page 2003

  3. [2]

    Bacry, S

    E. Bacry, S. Delattre, M. Hoffmann, and J. Muzy , Some limit theorems for Hawkes processes and application to financial statistics , Stochastic Process. Appl., 123 (2013), pp. 2475–2499

    work page 2013

  4. [3]

    Bacry, A

    E. Bacry, A. Iuga, M. Lasnier, and C.-A. Lehalle , Market impacts and the life cycle of investors orders, Market Microstructure and Liquidity, 02 (2015), pp. 1–46

    work page 2015

  5. [4]

    Bacry, I

    E. Bacry, I. Mastromatteo, and J.-F. Muzy , Hawkes processes in finance , Market Microstructure and Liquidity, 01 (2015), p. 1550005

    work page 2015

  6. [5]

    M. S. Bartlett , The spectral analysis of point processes , J. Roy. Statist. Soc. Ser. B, 25(2) (1963), pp. 264–296

    work page 1963

  7. [6]

    Billingsley , Convergence of Probability Measures , Wiley series in probability and statistics, Wiley, New York, NY, 2ed ed., 1999

    P. Billingsley , Convergence of Probability Measures , Wiley series in probability and statistics, Wiley, New York, NY, 2ed ed., 1999

    work page 1999

  8. [7]

    Bordenave and G

    C. Bordenave and G. L. Torrisi , Large deviations of Poisson cluster processes , Stoch. Models, 23 (2007), pp. 593–625

    work page 2007

Show all 39 references
  1. [8]

    Br ´ emaud and M

    P. Br ´ emaud and M. L. , Power spectra of general shot noises and Hawkes point proces ses with a random excitation, Adv. in Appl. Probab., 34 (2002), pp. 205–222

    2002

  2. [9]

    Br ´ emaud, G

    P. Br ´ emaud, G. Nappo, and G. L. Torrisi , Rate of convergence to equilibrium of marked Hawkes processes, J. Appl. Probab., 39 (2002), pp. 123–136

    2002

  3. [10]

    Chavez-Demoulin and J

    V. Chavez-Demoulin and J. A. McGill , High-frequency financial data modeling using Hawkes pro- cesses, J. Bank. Finance, 36 (2012), pp. 3415–3426

    2012

  4. [11]

    Clements, R

    A. Clements, R. Herrera, and S. Hurn , Modelling interregional links in electricity price spikes , Energy Econ., 51 (2015), pp. 383–393

    2015

  5. [12]

    S. N. Ethier and T. G. Kurtz , Markov Processes: Characterization and Convergence , Wiley, New York, 1986

    1986

  6. [13]

    Gao and L

    X. Gao and L. Zhu , Large deviations and applications for Markovian Hawkes pro cesses with a large initial intensity, Bernoulli, 24 (2018), pp. 2875–2905. [14] , Limit theorems for Markovian Hawkes processes with a large i nitial intensity , Stochastic Process. Appl., 128 (201...

    2018

  7. [15]

    Gripenberg, S.-O

    G. Gripenberg, S.-O. Londen, and O. Staffans , Volterra integral and functional equations , Cam- bridge Univ. Press, 1990

    1990

  8. [16]

    A. G. Hawkes , Point spectra of some mutually exciting point processes , J. Roy. Statist. Soc. Ser. B, 33 (1971), pp. 438–443

    1971

  9. [17]

    A. G. Hawkes , Spectra of some self-exciting and mutually exciting point p rocesses, Biometrika, 58 (1971), pp. 83–90

    1971

  10. [18]

    M. A. Holbrook and J. R. Menninger , Erythromycin slows aging of Saccharomyces cerevisiae. , J. Gerontol. Series A, Biological Sciences and Medical Sciences, 57A (2 002), pp. B29–B36

  11. [19]

    Horst and W

    U. Horst and W. Xu , A scaling limit for limit order books driven by Hawkes proces ses, SIAM J. Financial Math., 10 (2019), pp. 350–393

    2019

  12. [20]

    Ikeda and S

    N. Ikeda and S. W atanabe , Stochastic Differential Equations and Diffusion Processes , North-Holland Kodansha, Amsterdam/Tokyo, 1989

    1989

  13. [21]

    Jacod and A

    J. Jacod and A. N. Shiryaev , Limit Theorems for Stochastic Processes , Springer, Berlin, 2003

    2003

  14. [22]

    Jagers , A plea for stochastic population dynamics , J

    P. Jagers , A plea for stochastic population dynamics , J. Math. Biol., 60 (2010), pp. 761–764

    2010

  15. [23]

    Jaisson and M

    T. Jaisson and M. Rosenbaum , Limit theorems for nearly unstable Hawkes processes , Ann. Appl. Probab., 25 (2015), pp. 600–631. 28

    2015

  16. [24]

    J. C. Jiang, E. Jaruga, M. V. Repnevskaya, and S. M. Jazwinski , An intervention resembling caloric restriction prolongs life span and retards aging in yeast., The F ASEB J., 14 (2000), pp. 2135–2137

    2000

  17. [25]

    Karabash and L

    D. Karabash and L. Zhu , Limit theorems for marked Hawkes processes with applicatio n to a risk model , Stoch. Models, 31 (2015), pp. 433–451

    2015

  18. [26]

    Kl ¨uppelberg and T

    C. Kl ¨uppelberg and T. Mikosch , Delay in claim settlement and ruin probability approximati ons, Scand. Actuar. J., 1995.2 (1995), pp. 154–168

    1995

  19. [27]

    Kl ¨uppelberg and T

    C. Kl ¨uppelberg and T. Mikosch , Explosive Poisson shot noise processes with applications t o risk reserves, Bernoulli, 1 (1995), pp. 125–147

    1995

  20. [28]

    P. A. W. Lewis , A branching Poisson process model for the analysis of comput er failure patterns , J. Roy. Statist. Soc. Ser. B, 26 (1964), pp. 398–456

    1964

  21. [29]

    Maulik and S

    K. Maulik and S. Resnick , Small and large time scale analysis of a network traffic model , Queueing Syst., 43 (2003), pp. 221–250

    2003

  22. [30]

    Ogata, Statistical models for earthquake occurrences and residua l analysis for point processes , J

    Y. Ogata, Statistical models for earthquake occurrences and residua l analysis for point processes , J. Amer. Statist. Assoc., 1 (1988), pp. 228–281

    1988

  23. [31]

    A. G. Pakes , Limit theorems for the integrals of some branching processe s, Stochastic Process. Appl., 3 (1975), pp. 89–111

    1975

  24. [32]

    Samorodnitsky , A class of shot noise models for financial applications , in Athens Conference on Applied Probability and Time Series Analysis, Vol

    G. Samorodnitsky , A class of shot noise models for financial applications , in Athens Conference on Applied Probability and Time Series Analysis, Vol. I (1995), vol. 114 of L ect. Notes Stat., Springer, New York, 1996, pp. 332–353

    1995

  25. [33]

    Stabile and G

    G. Stabile and G. L. Torrisi , Risk processes with non-stationary Hawkes claims arrivals , Methodol. Comput. Appl. Probab., 12 (2010), pp. 415–429

    2010

  26. [34]

    Stephens , Senescence: even bacteria get old

    C. Stephens , Senescence: even bacteria get old. , Current Biol., 15 (2005), pp. R308–R310

    2005

  27. [35]

    G. J. Tortora, B. R. Funke, C. L. Case, and T. R. Johnson , Microbiology: An Introduction , Pearson Education, 2016

    2016

  28. [36]

    Vere-Jones , Stochastic models for earthquake occurrence , J

    D. Vere-Jones , Stochastic models for earthquake occurrence , J. Roy. Statist. Soc. Ser. B, 32 (1970), pp. 1–62

    1970

  29. [37]

    J. B. W alsh , An introduction to stochastic partial differential equatio ns, In ´Ecole d’´Et´ e de Probabilit´ es de Saint Flour XIV-1984, (1986), pp. 265–439

    1986

  30. [38]

    J. G. Wood, B. Rogina, S. Lavu, K. Howitz, S. L. Helfand, M. Tat ar, and D. Sinclair , Sirtuin activators mimic caloric restriction and delay ageing in me tazoans., Nature, 430 (2004), pp. 686–689

    2004

  31. [40]

    Zhu , Ruin probabilities for risk processes with non-stationary arrivals and subexponential claims , Insur- ance Math

    L. Zhu , Ruin probabilities for risk processes with non-stationary arrivals and subexponential claims , Insur- ance Math. Econom., 53 (2013), pp. 544–550. 29

    2013

Pith tools

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