Two-type continuous-state branching processes in varying environments

Junyan Zhang; Zenghu Li

REVIEW 2 major objections 4 minor 1 cited by

Two-type continuous-state branching processes in varying environments

T0 review · 2 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read This paper proves that a two-type continuous-state branching process in a varying environment is determined by the unique bounded solution of a backward integral equation system, so its transition semigroup exists even when parameters…

desk verdict Solid first construction of two-type CBVE processes with cadlag parameters, but the proof of Theorem 1.1 contains a false uniform bound that needs a small, local fix. read the letter →

arxiv 2502.02852 v1 pith:IP65BCI2 submitted 2025-02-05 math.PR

Zenghu Li , Junyan Zhang This is my paper · ORCID

classification math.PR MSC 60J80

keywords branchingprocesstwo-typecontinuous-statevaryingenvironmentbackwardequationsystemcumulantsemigroupbottlenecksmomentcondition

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 constructs a two-type continuous-state branching process in a varying environment (a TCBVE-process), an inhomogeneous Markov process on the positive quadrant whose transition probabilities are governed by a cumulant semigroup. The central result is that the backward integral equation system (1.6) for the cumulant semigroup has a unique bounded positive solution under a moment condition on the jump measures, even though the branching parameters may be càdlàg in time and may have bottlenecks. This gives a two-type version of a construction that was previously open in the one-dimensional CBVE literature and serves as the foundation for building more general multi-type branching processes in varying environments.

What carries the argument

The central object is the cumulant semigroup $v_{r,t}(\lambda)$, the Laplace exponent in (1.1); the paper characterizes it as the unique solution of the system of backward integral equations (1.6), whose coefficients are the branching mechanism $(b_{ij},b_{ii},c_i,m_i)$. The moment condition (1.4) makes the jump integrals in (1.10) locally Lipschitz, which is the property that forces uniqueness of solutions. Bottlenecks are times $s$ where $\Delta b_{ii}(s)=1$, at which the solution can jump to the boundary $\partial R^2_+$; condition (1.5) keeps the solution inside the nonnegative quadrant. Existence is proved by approximating the mechanism $\varphi$ by simpler mechanisms $\varphi_n$ of the form (2.2), applying an inhomogeneous nonlinear $h$-transformation, and taking a monotone limit that preserves the Lévy-Khintchine structure.

What would settle it

In the affine case $m_i=0$, $c_i=0$, the backward system (1.6) reduces to the linear system (1.9), whose solution can be computed explicitly as a product integral. Checking the Chapman-Kolmogorov identity $v_{r,t}(\lambda)=v_{r,s}(v_{s,t}(\lambda))$ for $r<s<t$ with a cadlag $b_{ii}$ having a jump of size $1$ at a bottleneck time would directly test the claimed semigroup; a single violation would refute Theorem 1.1.

Watch

Extended reading notes

Core claim

The paper proves that, for each $t\ge 0$ and $\lambda\in R^2_+$, there is a unique bounded solution $r\mapsto v_{r,t}(\lambda)\in R^2_+$ to the backward integral equation system (1.6). The family $(Q_{r,t})_{t\ge r\ge 0}$ defined by the Laplace transform $\int_{R^2_+} e^{-\langle\lambda,y\rangle}Q_{r,t}(x,dy)=e^{-\langle x,v_{r,t}(\lambda)\rangle}$ is an inhomogeneous transition semigroup on $R^2_+$. The companion result, Theorem 1.2, differentiates this semigroup at the origin to obtain the first-moment formula $\int_{R^2_+}\langle\lambda,y\rangle Q_{r,t}(x,dy)=\langle x,\pi_{r,t}(\lambda)\rangle$, where $\pi_{r,t}(\lambda)$ solves the linear backward equation (1.9) with the compensated immigration parameter $\bar b_{ij}(t)=b_{ij}(t)+\int_0^t\int z_j\,m_i(ds,dz)$. Existence is obtained by approximating the general branching mechanism by simpler ones, proving convergence through a two-dimensional Gronwall inequality, and identifying the limit with a Lévy-Khintchine representation.

Load-bearing premise

The load-bearing premise is the moment condition (1.4): for each type $i$, the jump measure $m_i$ must satisfy $\int_0^t\int (z_i^2 1_{\{\|z\|\le 1\}}+z_i 1_{\{\|z\|>1\}}+z_j)\,m_i(ds,dz)<\infty$; if this integral diverges, the backward equation can become irregular near the boundary and uniqueness can fail.

Editorial extensions

If this is right

For every $\lambda$ and $t$, the backward equation (1.6) has exactly one bounded positive solution, so the Laplace transform (1.1) defines a well-defined inhomogeneous transition semigroup on $R^2_+$.
The branching parameters may be càdlàg in time; together with the bottleneck condition (1.5), the moment condition (1.4) controls the jumps to the boundary at bottleneck times.
First moments are governed by the linear equation (1.9): $\int_{R^2_+}\langle\lambda,y\rangle Q_{r,t}(x,dy)=\langle x,\pi_{r,t}(\lambda)\rangle$ with $\bar b_{ij}$ given by (1.7).
The constructed processes form a basic class, and the paper states that the construction is the basis for building more general TCBVE-processes in a forthcoming work.
The result gives a two-type, varying-environment resolution of the uniqueness problem for backward equations of multi-type continuous-state branching processes.

Reading between the lines

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

The same backward-equation strategy should extend to $d$ types: the two-dimensional Gronwall inequality and the $h$-transformation are not dimension-specific, so a $d$-type version would follow from analogous estimates.
Because condition (1.4) is needed for the local Lipschitz property, a natural scaling-limit test is whether two-type Galton-Watson processes with heavy-tailed small jumps converge to a process whose cumulant semigroup is nonunique; if so, a pathwise stochastic-equation construction would be needed instead.
The semigroup is built without constructing the underlying genealogical structure; a likely extension, not explored here, is to build the associated flows of subordinators or historical processes for the two-type CBVE setting.
Adding a linear immigration term to the Laplace exponent would fit the same backward-equation framework and should produce two-type CBVE processes with immigration, mirroring the one-dimensional theory.

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

The paper does what it says: it builds a basic class of two-type continuous-state branching processes in varying environments, allowing cadlag parameters and handling bottleneck times. That is a genuine extension of the one-type work by Bansaye–Simatos and Fang–Li, and the two-dimensional coupling is not just cosmetic. The backward equation system, the two-dimensional Gronwall lemma, and the inhomogeneous h-transformation are the right tools, and the approximation scheme follows the Fang–Li template without introducing fitted parameters. Theorems 1.1 and 1.2 are new and, I believe, correct in statement.

The proof is mostly careful. The moment condition (1.4) is explicit and does the needed work near the boundary; it is stronger than in the one-dimensional papers, but the authors say so and the reason is clear. The citations to Rhyzhov–Skorokhod, Watanabe, Li–Li, and the Loewner papers look appropriate.

There is one real gap, and the stress-test note identifies it correctly. In the proof of Theorem 1.1, right after (3.6), the authors claim v(n) ≤ 2A e^{ρ(t)} with ρ(t)=2||b11||+2||b22||+b12+b21. The bound that actually follows from Proposition 2.4 has \bar b12+\bar b21 in the exponent, not b12+b21; the missing ∫ z_j m_i terms matter. The specific counterexample in the note is not exactly right as written (v2 is not constant), but the defect is real: in that example the limiting v1(0) for t=10 is around 3.5, so the printed bound with A=1 fails. The fix is simple: replace b12+b21 by \bar b12+\bar b21 in ρ. Then the inequality follows from (2.5), and the rest of the Gronwall argument works because (1.4) makes the \bar bij finite. So this is a repairable flaw in the written proof, not a flaw in the theorem.

Other issues are minor: a few estimates are compressed, there is a likely sign typo after (3.2), and the differentiability step in Theorem 1.2 is not fully expanded. None of these threaten the main result.

Bottom line: this deserves a serious referee. The construction is a natural and useful extension, the proof strategy is sound, and the one genuine gap is local and easily fixed. I would send it to review with a request that the authors correct the ρ bound. If I were working on multi-type branching in varying environments, I would cite it.

Referee Report

2 major / 4 minor

Summary. The paper constructs a class of two-type continuous-state branching processes in varying environments (TCBVE-processes). The main result, Theorem 1.1, states that under the assumptions on the increasing functions b_ij, the locally bounded variation functions b_ii, the increasing continuous functions c_i, and the measures m_i satisfying the moment condition (1.4) and the bottleneck condition (1.5), the backward integral equation system (1.6) has a unique bounded positive solution for every t and λ, and that (1.1) defines an inhomogeneous transition semigroup. Theorem 1.2 identifies the first moments of this semigroup with the solution of the linearized equation (1.9) with coefficients \bar b_ij. The proofs use a two-dimensional Gronwall inequality, an h-transform lemma, and an approximation of the general branching mechanism by special mechanisms of the form (2.2), for which existence is proved by Picard iteration.

Significance. If correct, the paper gives a natural and useful extension of the one-dimensional CBVE theory of Fang and Li to a genuine two-type setting with càdlàg parameters; it also provides the cumulant semigroup characterization that underlies further constructions. The method is transparent and mostly self-contained, and the construction is parameter-free: the semigroup is derived from the stated branching mechanism rather than tuned to match a target. The first-moment formula (1.8)-(1.9) is a concrete, checkable prediction. However, one displayed estimate in the proof of the main theorem is false as written, and a second proof step in Theorem 1.2 needs a missing justification, so I cannot recommend acceptance without revision.

major comments (2)

[§3, proof of Theorem 1.1, equation after (3.6)] The asserted n-independent bound v^{(n)}_{i,r,t}(λ) ≤ 2A e^{ρ(t)} with ρ(t)=2‖b11‖(t)+2‖b22‖(t)+b12(t)+b21(t) is false as stated because ρ omits the cross-type jump moments ∫ z_j m_i(ds,dz). For the approximate equation (3.6), the sum inequality in Proposition 2.4 gives coefficients of order \bar b_{ij}(ds)=b_{ij}(ds)+∫ z_j m_i(ds,dz), not b_{ij}(ds). A concrete failure is c_i=0, b_{ii}=b_{ij}=0, m_2=0, m_1(ds,dz)=ds δ_{(0,1)}(dz), λ=(1,1). Then v^{(n)}_{2}=1 and v^{(n)}_{1,r,t}(λ)=1+[e^{-n}+(1-e^{-n})(1-e^{-1})](t-r), which exceeds 2 for t-r large and n large. The printed bound is load-bearing: it is used to define C_1(t) and \tilde A(t) and to justify the passage to the limit in D_{k,n}. The gap is repairable by replacing b_{ij} with \bar b_{ij} in ρ; condition (1.4) ensures the resulting constants are finite. The theorem itself is not in doubt from this check, but the written proof needs correction.
[§3, proof of Theorem 1.2, Step 2] The proof introduces π_{i,r,t}(λ)=∂_a v_{i,r,t}(aλ)|_{a=0} and differentiates the integral equation, but differentiability of a↦v_{r,t}(aλ) at a=0 is not established. This is not immediate from continuity of λ↦v_{r,t}(λ): the integrand in (1.6) is nonlinear, and passing from difference quotients to the derivative requires a dominated convergence argument and a Gronwall bound for the derivative. A rigorous proof should derive the linear equation for the derivative and verify its integrability using Proposition 3.3 and condition (1.4). Without this, the mean formula (1.8) is not fully proved. This is also repairable by standard arguments.

minor comments (4)

[§2, Proposition 2.4, Step 3] In the displayed estimate for u(k,r,t,λ), the integral '∫_r^t ∫_{R2_+} u(k−1,t1,t,λ)ρ(dt1)' appears to have an extra space integration; it should be a single integral ∫_r^t u(k−1,t1,t,λ)ρ(dt1).
[§3, definition of φ_n, after (3.2)] The inequality 'γn,ij(ds) ≥ \bar b_{ij}(ds)' is inconsistent with the displayed formula, since a nonnegative term is subtracted from \bar b_{ij}. The proof only needs γn,ij to be a nondecreasing measure, which can be checked from the definition; the displayed inequality should be corrected or removed.
[§3, equation (3.2)] The notation δ_{e_i}(n dz) is nonstandard and should be defined explicitly; from the subsequent computation ∫(z1+z2) μ_{n,i}=2n c_i(t)+..., it appears to mean the point mass at e_i/n.
[§2, Proposition 2.4] The spelling 'Lévy-Kthintchine' should be 'Lévy-Khintchine'.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the construction is self-contained and the only self-citations are minor and non-load-bearing.

full rationale

The paper derives the two-type cumulant semigroup by solving the backward integral equation (1.6) under the moment condition (1.4), with no fitted parameters and no target quantity appearing as an input. The semigroup property in Theorem 1.1 follows from uniqueness of the solution to (1.6), which is proved in Proposition 3.3 and the approximation argument via Proposition 2.4; this is a genuine mathematical derivation, not a definitional equivalence. The Laplace-transform representation (1.1) is a standard encoding, and the claim that (Q_{r,t}) is an inhomogeneous transition semigroup is justified by the flow property (1.3) rather than assumed. Theorem 1.2 is obtained by differentiating the solved cumulant equation, so the mean formula is a consequence, not an input. The citations to Fang-Li (2022) and Li-Li (2024) are for methodology and background; the present proof of the two-type result does not rest on those papers for the theorem itself. Separately, the proof of Theorem 1.1 after (3.6) contains a display asserting v^{(n)}_{i,r,t}(λ) ≤ 2A e^{ρ(t)} with ρ(t)=2||b_{11}||(t)+2||b_{22}||(t)+b_{12}(t)+b_{21}(t); this bound appears to omit the cross-type jump moments ∫ z_j m_i and may be false as printed, but that is a correctness gap, not circularity, and does not change the circularity verdict.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central construction depends only on explicit assumptions on the parameter functions and on standard tools from probability and analysis. There are no fitted numbers and no new entities are postulated. The most consequential input is the moment condition (1.4), which is stated as an assumption rather than derived.

assumptions (6)

standard math Levy-Khintchine representation for infinitely divisible distributions on R^2_+ as used in (1.2).
Identifies Laplace exponents of the transition semigroup; standard infinite-divisibility theory.
standard math Watanabe (1969), Lemma 1: the class of Levy-Khintchine functions is preserved under the monotone and iteration limits used in Proposition 2.4.
Used to conclude that the limiting function v_{r,t} is a cumulant semigroup; referenced in Step 3 of Proposition 2.4.
standard math Two-dimensional Gronwall inequality (Lemma 2.1), based on Mao (1990).
Main tool for uniqueness and bounds in Propositions 2.4 and 3.3; proved within the paper.
standard math Integration by parts for cadlag functions of locally bounded variation (Lemma 2.2).
Used for the inhomogeneous nonlinear h-transformation leading to equation (2.4).
domain assumption Moment condition (1.4) and bottleneck condition (1.5) on the parameters.
These restrict the class of processes constructed. (1.4) yields the local Lipschitz property of (1.10); (1.5) keeps solutions in R^2_+.
domain assumption The approximating branching mechanisms phi_n converge monotonically to phi (Lemma 3.2(1)) and dominated convergence applies in the limit.
Justifies passing from special equations to the general equation (1.6) in Theorem 1.1; relies on the moment condition for integrability.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Two-type continuous-state branching processes in varying environments." pith.science (2026). https://pith.science/paper/IP65BCI2

@misc{pith2026250202852,
  author       = {Pith},
  title        = {Pith review of: Two-type continuous-state branching processes in varying environments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IP65BCI2}},
  note         = {Machine review of arXiv:2502.02852}
}

read the original abstract

A basic class of two-type continuous-state branching processes in varying environments are constructed by solving the backward equation determining the cumulant semigroup. The parameters of the process are allowed to be c\`adl\`ag in time and the difficulty brought about by the bottlenecks are overcome by introducing a suitable moment condition.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Stochastic equations for two-type continuous-state branching processes in varying environments
math.PR 2025-02 conditional novelty 6.0 of 10

A two-type continuous-state branching process in a varying environment is constructed as the pathwise unique strong solution of a system of stochastic integral equations driven by white noise and Poisson random measures.

Reference graph

Works this paper leans on

12 extracted references · 11 canonical work pages · cited by 1 Pith paper

[1]

and Shchurenkov, V.M

Aliev, S.A. and Shchurenkov, V.M. (1982). Transitional phenome na and the convergence of Galton– Watson processes to Jiˇ rina processes. Theory Probab. Appl. 27, 472–485

work page 1982
[2]

and Simatos, F

Bansaye, V. and Simatos, F. (2015). On the scaling limits of Galton –Watson processes in varying environment. Electron. J. Probab. 20, no. 75, 1–36

2015
[3]

and Li, Z

Fang, R. and Li, Z. (2022). Construction of continuous-state branching processes in varying envi- ronments. Ann. Appl. Probab. 32, 3645–3673

work page 2022
[4]

Grimvall, A. (1974). On the convergence of sequences of branc hing processes. Ann. Probab. 2, 1027– 1045

work page 1974
[5]

Loewner Theory for Bernstein functions I: evolution families and differential equations

Gumenyuk, P.; Hasebe, T. and P´ erez, J.-L. (2024): Loewner T heory for Bernstein functions I: evolution families and diﬀerential equations. Constr. Approx. Online: doi.org/10.1007/s00365-023- 09675-9; arXiv:2206.04753

work page Pith review arXiv 2024
[6]

and P´ erez, J.-L

Gumenyuk, P.; Hasebe, T. and P´ erez, J.-L. (2022+): Loewner Theory for Bernstein functions II: applications to inhomogeneous continuous-state branching proce sses. arXiv:2211.12442

work page arXiv 2022
[7]

Lamperti, J. (1967). The limit of a sequence of branching proces ses. Z. Wahrsch. verw. Ge. 7, 271–288

work page 1967
[8]

and Li, Z

Li, P. and Li, Z. (2024). Uniqueness problem for the backward diﬀ erential equation of a continuous- state branching process. Acta Math. Sin. (Engl. Ser.) 40, 1825–1836

work page 2024

Show all 12 references

[9]

Li, Z. (2020). Continuous-state branching processes with immig ration. In From Probability to Finance-Lecture Notes of BICMR Summer School on Financial M athematics. Math. Lect. Peking Univ. 1-69

2020
[10]

Mao, X. (1990). Lebesgue-stieltjes integral inequalities in sev eral variables with retardation. Proc. Indian Acad. Sci. (Math. Sci.) 100, 231-243

1990
[11]

and Skorokhod, A.V

Rhyzhov, Y.M. and Skorokhod, A.V. (1970). Homogeneous bra nching processes with a ﬁnite number of types and continuous varying mass. Theory Probab. Appl. 15, 704–707

1970
[12]

Watanabe, S. (1969). On two dimensional Markov processes w ith branching property. Trans. Amer. Math. Soc. 136, 447–466. 12

1969

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

and Shchurenkov, V.M

Aliev, S.A. and Shchurenkov, V.M. (1982). Transitional phenome na and the convergence of Galton– Watson processes to Jiˇ rina processes. Theory Probab. Appl. 27, 472–485

work page 1982

[2] [2]

and Simatos, F

Bansaye, V. and Simatos, F. (2015). On the scaling limits of Galton –Watson processes in varying environment. Electron. J. Probab. 20, no. 75, 1–36

2015

[3] [3]

and Li, Z

Fang, R. and Li, Z. (2022). Construction of continuous-state branching processes in varying envi- ronments. Ann. Appl. Probab. 32, 3645–3673

work page 2022

[4] [4]

Grimvall, A. (1974). On the convergence of sequences of branc hing processes. Ann. Probab. 2, 1027– 1045

work page 1974

[5] [5]

Loewner Theory for Bernstein functions I: evolution families and differential equations

Gumenyuk, P.; Hasebe, T. and P´ erez, J.-L. (2024): Loewner T heory for Bernstein functions I: evolution families and diﬀerential equations. Constr. Approx. Online: doi.org/10.1007/s00365-023- 09675-9; arXiv:2206.04753

work page Pith review arXiv 2024

[6] [6]

and P´ erez, J.-L

Gumenyuk, P.; Hasebe, T. and P´ erez, J.-L. (2022+): Loewner Theory for Bernstein functions II: applications to inhomogeneous continuous-state branching proce sses. arXiv:2211.12442

work page arXiv 2022

[7] [7]

Lamperti, J. (1967). The limit of a sequence of branching proces ses. Z. Wahrsch. verw. Ge. 7, 271–288

work page 1967

[8] [8]

and Li, Z

Li, P. and Li, Z. (2024). Uniqueness problem for the backward diﬀ erential equation of a continuous- state branching process. Acta Math. Sin. (Engl. Ser.) 40, 1825–1836

work page 2024