Abstract
We introduce flows of branching processes with competition, which describe the evolution of general continuous state branching populations in which interactions between individuals give rise to a negative density dependence term. This generalizes the logistic branching processes studied by Lambert (Ann Appl Probab 15(2):1506–1535, 2005). Following the approach developed by Dawson and Li (Ann Probab 40(2):813–857, 2012), we first construct such processes as the solutions of certain flow of stochastic differential equations. We then propose a novel genealogical description for branching processes with competition based on interactive pruning of Lévy-trees, and establish a Ray–Knight representation result for these processes in terms of the local times of suitably pruned forests.
This is a preview of subscription content, access via your institution.

Notes
The reader is referred to [13] for further topological background.
The full result is actually that the Poisson point process can be constructed to obtain a strong solution with respect to a given noise to Eq. (1.15).
Notice that \(d_0\) corresponds to the metric on \(\mathscr {M}(\mathbb {R}_+)\).
Notice that results in Section 4.1 in [12] are stated in terms of the process \(((\rho _t, W_t): t\ge 0)\) with \(W_t\) corresponding here to the increasing \({{\mathcal {M}}}_{at}(\mathbb {R}_+)\)-valued path \(W_t =(h\mapsto {{\mathcal {N}}}_t([0,h], d\nu ): h\in [0,H_t) \, )\), which was denoted \(\xi ^{(t)}\) in Sect. 3.
In the notation of Section 4.1 of [12], \({{\mathcal {N}}}^{(i)} \) would correspond to the “increment” of an excursion of the path-valued process \(W_t\).
References
Abraham, R., Delmas, J.-F.: Fragmentation associated with Lévy processes using snake. Probab. Theory Relat. 141, 113–154 (2008)
Abraham, R., Delmas, J.-F., He, H.: Pruning of CRT-sub-trees. Stoch. Process. Appl. 125(4), 1569–1604 (2015)
Abraham, R., Delmas, J.-F., Voisin, G.: Pruning a Lévy continuum random tree. Electron. J. Probab. 15(46), 1429–1473 (2010)
Abraham, R., Serlet, L.: Poisson snake and fragmentation. Electron. J. Probab. 7, 1–15 (2002)
Aldous, D.: The continuum random tree I. Ann. Probab. 19, 1–28 (1991)
Aldous, D.: The continuum random tree III. Ann. Probab. 21, 248–289 (1993)
Ba, M., Pardoux, E.: Branching processes with competition and generalized Ray–Knight Theorem. Annales de l’Institut Henri Poincaré, Probabilités et Statistiques 51, 1290–1313 (2015)
Berestycki, J., Döring, L., Mytnik, L., Zambotti, L.: On exceptional times for generalized Fleming–Viot processes with mutations. Stoch. Partial Differ. Equ. Anal. Comput 2(1), 84–120 (2014)
Bertoin, J., Le Gall, J.-F.: Stochastic flows associated to coalescent processes. Probab. Theory Relat. Fields 126, 261–288 (2003)
Caballero, M.E., Lambert, A., Uribe-Bravo, G.: Proof(s) of the Lamperti representation of continuous-state branching processes. Probab. Surv. 6, 62–89 (2009)
Dawson, D.A., Li, Z.: Stochastic equations, flows and measure-valued processes. Ann. Probab. 40(2), 813–857 (2012)
Duquesne, T., Le Gall, J.-F.: Random trees, Lévy processes and spatial branching processes. Astérique, vol. 281. Société mathématique de France, Paris (2002)
Duquesne, T., Le Gall, J.-F.: Probabilistic and fractal aspect of Lévy trees. Probab. Theory Relat. Fields 131, 553–603 (2005)
Fu, Z .F., Li, Z.: Stochastic equations of non-negative processes with jumps. Stoch. Process. Appl. 120, 306–330 (2010)
Itô, K.: Poisson point processes attached to Markov processes. In: Proceedings Sixth Berkeley Symposium on Mathematical Statistic and Probability, vol.III, pp. 225–239 (1972)
Knight, F.B.: Random walks and a sojourn density process of Brownian motion. Trans. Am. Math. Soc. 109(1), 56–86 (1963)
Kyprianou, A.E.: Introductory Lectures on Fluctuations of Lévy Processes with Applications. Springer, Berlin (2006)
Lambert, A.: The branching process with logistic growth. Ann. Appl. Probab. 15(2), 1506–1535 (2005)
Lamperti, J.: Continuous-state branching processes. Bull. Am. Math. Soc. 73, 382–386 (1967)
Le, V., Pardoux, E., Wakolbinger, A.: “Trees under attack” a Ray–Knight representation of Feller’s branching diffusion with logistic growth. Probab. Theory Relat. Fields 155(3–4), 583–619 (2013)
Le Gall, J.-F., Le Jan, Y.: Branching processes in Lévy processes. Ann. Probab. 26, 213–252 (1998)
Li, Z.: Measure-Valued Branching Markov Processes. Springer, Berlin (2011)
Pardoux, E.: Probabilistic Models of Population Evolution. Scaling Limits and Interactions. Springer, Berlin (2016)
Pardoux, E., Wakolbinger, A.: From Brownian motion with local time drift to Feller’s branching diffusion with logistic growth. Electron. Commun. Probab. 16, 720–731 (2011)
Pardoux, E., Wakolbinger, A.: From exploration paths to mass excursions–variations on a theme of Ray and Knight. In: Surveys in Stochastic Processes, Proceedings of the 33rd SPA Conference in Berlin, 2009, EMS (2011)
Ray, D.: Sojourn times of diffusion processes. Ill. J. Math. 7(4), 615–630 (1963)
Walsh, J.B.: An introduction to stochastic partial differential equations. In: École d’été de probabilités de Saint-Flour, XIV—1984, Volume 1180 of Lecture Notes in Mathematics, pp. 265–439. Springer, Berlin (1986)
Acknowledgements
J.B. was partially supported by ANR grants ANR-14-CE25-0014 (ANR GRAAL) and ANR-14-CE25-0013 (ANR NONLOCAL). M.C.F.’s research was supported by a doctoral grant of BASAL-Conicyt Center for Mathematical Modeling (CMM) at University of Chile and a postdoctoral grant from the National Agency for Science and Technology Promotion, Argentina. She thanks also ICM Millennium Nucleus NC120062 (Chile) for travel support, a doctoral mobility grant of the French Embassy in Chile for a three months visit to Laboratoire de Probabilités et Modèles Aléatoires of University Paris 6 in 2012, and hospitality of the latter institution during her stay. J.F. acknowledges partial support of ICM Millennium Nucleus NC120062 and BASAL-Conicyt CMM. The authors thank Jean-François Delmas for enlightening discussions at the beginning of this work and Victor Rivero and Juan Carlos Pardo for kindly providing some useful mathematical insights. We also thank two anonymous referees for their careful reading of the manuscript, for valuable remarks and comments which allowed us to clarify the presentation of our results, and for pointing out to us references [7, 23].
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
1.1 Proof of Proposition 1.2
Recall from [3, Theorem 0.3] that \((\rho ^\theta _t: t\ge 0) : = (\rho _{C^\theta _t}: t\ge 0)\), with \(C^\theta _t \) the right-continuous inverse of \(A^\theta _t : = \int _0^t \mathbf {1}_{\{m^{\theta }_s =0\}} \mathrm {d}s\), has the same law as the exploration process associated with a Lévy process of Laplace exponent (1.11). Denote by \(\left( {\bar{L}}_s^a :t\ge 0 \right) \) its local time at level a. Applying (3.4) to \(\rho ^{\theta }\) and performing the change of variable \(C^{\theta }_r\mapsto u\), we deduce that
in the \(L^1(\mathbb {P})\) sense. Let us check that this limit is equal to \(L_{C^{\theta }_t}^a(m^\theta )\). Since \(C^{\theta }_t<\infty \) a.s. (see [3]), we deduce from (3.4) applied to the original exploration process \((\rho _t: t\ge 0)\) that, on the interval \([0,C^{\theta }_t ]\), the finite measures \(\epsilon ^{-1} \mathbf {1}_{\{a-\epsilon <H_s\le a\}}\mathrm {d}s\) converge weakly in probability towards \(\mathrm {d}L_s^a\) as \(\epsilon \rightarrow 0\). By Lemma 3.3 the function \(s\mapsto \mathbf {1}_{\{ m_s^{\theta } ([0,H_s))=0\}}\) has a.s. at most countably many discontinuities, therefore a.s. it is continuous a.e. with respect to the measure \(\mathbf {1}_{[0,C^{\theta }_t]}(s)\mathrm {d}L_s^a\). We thus deduce that:
By right-continuity in \(t\ge 0\) of the first and third expressions, \(\left( {\bar{L}}_t^a :t\ge 0 \right) \) and \(\left( L_{C^{\theta }_t}^a(m^{\theta }) :t\ge 0 \right) \) are indistinguishable for each \(a\ge 0\). In particular, if we set \({\bar{T}}_x=\inf \{ s>0 : \, {\bar{L}}_t^0 >x\}\), for each \(a\ge 0\) we a.s. have
since \({\bar{L}}^0_{\cdot }=L_{C^{\theta }_{\cdot }}^0(m^\theta )\) (by (A.1) with \(a=0\)) and \(L^0_{\cdot }(m^{\theta })=L^0_{\cdot }\). The result follows from the above identities and Theorem 1.1 applied to the local times of \(\rho ^{\theta }\).
1.2 Proof of Lemma 2.1
Write \(M_t= (M^i_t)_{ i\le n}\), \(N_t= (N^k_t)_{ k\le m} \), let \(\lambda =(\lambda _1,\ldots ,\lambda _n)\in \mathbb {R}^n\), \(\mu =(\mu _1,\ldots ,\mu _m)\in \mathbb {R}^m\) and set \({\hat{\lambda }} =(\lambda _1 a_1^{-1},\ldots ,\lambda _n a_n^{-1})\). It is enough to show for all \(t,s\ge 0\) that
(where “\(\cdot \)” is the Euclidean inner product and \(i=\sqrt{-1}\)). Equivalently, we need to show that
is a \(({{\mathcal {S}}}_t)\)- martingale. By i), \( {{\mathcal {E}}}_t^{(M)}:=\exp \left\{ i {\hat{\lambda }} \cdot M_{t} +t\frac{|\lambda |^2}{2} \right\} = \exp \left\{ i {\hat{\lambda }} \cdot M_{t} - \frac{i^2}{2}\right. \left. [ {\hat{\lambda }} \cdot M]_t \right\} \) clearly is a \(({{\mathcal {S}}}_t)\)-martingale. On the other hand, writing \({{\mathcal {E}}}_t^{(N)}= \exp \left\{ i\mu \cdot N_t -t\sum _{k=1}^m b_k(e^{i\mu _k}-1) \right\} \) we have
by the first property in ii) and since \(\Delta N^k_s=0\) or 1 for each \(k=1,\dots ,m\). The second property in ii) then grants that \( {{\mathcal {E}}}_t^{(N)}- 1=\int _0^t \sum _{k=1}^m b_k(e^{i\mu _k}-1) {{\mathcal {E}}}_{s-}^{(N)} \mathrm {d}{\tilde{N}}^k_s\), with \({\tilde{N}}^k_t=N^k_t-b_k t\), is a \(({{\mathcal {S}}}_t)\)- martingale too. Since the martingales \({{\mathcal {E}}}^{(M)}\) and \({{\mathcal {E}}}^{(N)}\) respectively are continuous and pure jump, \([{{\mathcal {E}}}^{(M)},{{\mathcal {E}}}^{(N)}]=0\) and \({{\mathcal {E}}}^{(M)}{{\mathcal {E}}}^{(N)}\) is a \(({{\mathcal {S}}}_t)\)- martingale as required.
1.3 Proof of estimate (5.7)
We follow ideas in the proofs of [11, Theorem 2.1] and [14, Proposition 3.1]. Define a function \(\varrho \) on \(\mathbb {R}_+\) by \(\varrho (x):= [\sigma + \int _{0}^{1} r^{2}\Pi (\mathrm {d}r)]\sqrt{x}\). Notice that for all \(y\le x\) one has
Consider a real sequence \(\{a_{j}\}_{j\ge 1}\) defined by \(a_{0} = 1\) and \(a_{j}= a_{j-1} \,e^{-j[\sigma + \int _{0}^{1} r^2\Pi (\mathrm {d}r)]^{2}}\), so that \(a_{j}\searrow 0\) and \(\int _{a_{j}}^{a_{j-1}} \varrho (z)^{-2 }\mathrm {d}z = j\). For each \(j\ge 1\), let \(\psi _{j}\) be a non-negative continuous function on \(\mathbb {R}\) supported in \((a_{j} , a_{j-1})\) such that \(0 \le \psi _{j}(z)\le 2j^{-1}\varrho (z)^{-2}\) and \(\int _{a_{j}}^{a_{j-1}} \psi _{j}(z)\mathrm {d}z = 1\). Define also non-negative twice continuously differentiable functions \(\phi _{j}\) on \(\mathbb {R}\) by
Notice that for each \(x\in \mathbb {R}\) it holds for all \(j\ge 1\) that \( 0 \le \phi '_{j}(x)\text{ sign }(x) \le 1 \), \(\phi _{j}''(x)\ge 0\) and \( \phi ''_{j}(x)\sigma ^2|x| \le 2j^{-1}\). Moreover, we have \(\phi _{j}(x) \nearrow |x|\) as \(j \rightarrow \infty \).
For any \(z,h, r, \nu ,x,y\ge 0\) and a differentiable function f, set now \(l(r,\nu ; ,x, y): = r\left[ \mathbf {1}_{\{\nu<x\}} - \mathbf {1}_{\{\nu <y\}}\right] \), \(\Delta _{h}f(z):= f(z+h)-f(z)\) and \(D_h f (\zeta ) :=\Delta _h f(\zeta ) - f' (\zeta )h\). Using (5.6) and Itô’s formula, we get
where \((M_{t \wedge \tau _{m}})_{t\ge 0}\) is a martingale. By properties of \(\phi '_j\), the integrands in the first and second lines are respectively bounded by \(|\alpha _0||Z_{s}(x) - Z^{\varepsilon ,\delta }_{s}(x)|\) and \( \int _1^{\infty } r \Pi (\mathrm {d}r) |Z_{s}(x) - Z^{\varepsilon ,\delta }_{s}(x)|\). The fact that
following from estimate (A.2), together with the mentioned properties of \(\phi ''_{j}\) ensure that the terms on the third and fourth lines vanish when \(j\rightarrow \infty \). Taking expectation in (A.3) and letting \(j\rightarrow \infty \), we get the desired bound, noting that g(m) is a Lipschitz constant for the function G in [0, m].
1.4 Proof of Lemma 6.1
In a similar way as for the process \(\left( (\rho _t,{{\mathcal {N}}}_t)\, : t\ge 0 \right) \), the trajectories of the process \(\left( (\rho ^a_t,{{\mathcal {N}}}_t^a)\, : t\ge 0 \right) \) are determined in a unique (measurable) way from the atoms of the point process (6.3). It is therefore enough to establish the first claim. To do so, one easily adapts first the arguments of the proof of Proposition 4.2.3 in [12] in order to prove that, under the excursion measure \(\mathbb {N}\), the process
with \(I_j:=\{ i \in I: (\alpha ^{(i)},\beta ^{(i)}) \subset (\alpha ^{j},\beta ^{j})\}\) denoting the sub-excursions above level a of the excursion away from 0 labeled j, is conditionally on \({{\mathcal {E}}}_a\) a Poisson point process of intensity \(\mathrm {d}x \mathbf {1}_{[L^a_{\alpha _j}, L^a_{\beta _j} ]} \otimes \mathbb {N}(\mathrm {d}\rho \, , \mathrm {d}{{\mathcal {N}}})\) (our superscripts “(i)” correspond to superscripts “i” in the statement of [12]). The only difference is that, in the computation analogous to the one at end of that proof, we must consider here test functions depending also on the components \(\ell ^{(i)}\) of the atoms, and depending on the excursions of the spatial component above level a only through their increments respect to their values at that a. Since I is equal to the disjoint union \(\bigcup _{j\in J} I_j\), one then concludes applying conditionally on \({{\mathcal {E}}}_a\) the additivity of Poisson point measures.
1.5 Proof of Lemma 6.3
By standard properties of Poisson processes, for any nonnegative predictable process f and stopping time \(\tau \) in the given filtration, it holds that \(\mathbb {E}\left[ e^{-u \int _0^{\tau \wedge t} f(\ell )N(\mathrm {d}\ell ) + \lambda \int _0^{\tau \wedge t } (1- e^{-u f (\ell )})\mathrm {d}\ell }\vert {{\mathcal {K}}}_0 \right] =1\) a.s. for all \(u\ge 0\) and \(t\ge 0\). If moreover \(\tau \) is such that \( \mathbb {E}\left[ e^{ \lambda \int _0^{\tau } (1- e^{-u f (\ell )})\mathrm {d}\ell }\right] <\infty \), by dominated convergence one gets after letting \(t\rightarrow \infty \) that \(\mathbb {E}\left[ e^{-u \int _0^{\tau } f(\ell )N(\mathrm {d}\ell ) + \lambda \int _0^{\tau } (1- e^{-u f (\ell )})\mathrm {d}\ell }\vert {{\mathcal {K}}}_0 \right] =1\) a.s. Now, from our assumptions and by a monotone class argument, we can check that \(\mathbf {1}_{E}\) is a predictable process. Since \(e^{\lambda \int _0^{\phi _x} (1- e^{-u \mathbf {1}_{E}(\ell )})\mathrm {d}\ell }=e^{\lambda \int _0^{\phi _x} (1- e^{-u}) \mathbf {1}_{E}(\ell )\mathrm {d}\ell }= e^{\lambda (1- e^{-u}) (x \wedge \int \mathbf {1}_{E}(\ell )\mathrm {d}\ell )} \le e^{\lambda (1- e^{-u}) x } \), from the previous we conclude that for all \(x\ge 0\),
Applying conditionally on \( {{\mathcal {K}}}_{\phi _{y}}\) this argument to increments \(\int _{\phi _y} ^{\phi _x} \mathbf {1}_{E}(\ell )N(\mathrm {d}\ell ) \), the proof is complete.
1.6 Proof of Lemma 6.5
(a) The proof is inspired by that of Lemma 1.3.2 in [12]. We have
The time integral in the above expressions can be written in terms of the excursion point process (6.2) as follows:
where \(J_y:=\{ j\in J: \, \ell ^j \le y\}\), \(H_s^j=H_{\alpha ^j+s}\), \(m_s^j =\mathcal {N}^j_s (\, \cdot \times [0, \Theta (\ell ^j) )) \) and \(\zeta ^j\) is the length of the excursion labelled j. By compensation, the desintegration \(\mathbb {N}(\mathrm {d}\rho \, , \mathrm {d}{{\mathcal {N}}}) = \mathbf {N}(\mathrm {d}\rho ) Q^{H(\rho )}( \mathrm {d}{{\mathcal {N}}} )\) and the very definition of the snake \((\rho ,\mathcal {N})\), we get
with \(\zeta \) the length of the canonical excursion and \((H_s(\rho ): 0\le s\le \zeta )\) its height process. Thus, the second term in the r.h.s. of (A.5) is bounded by
Using Proposition 1.2.5 in [12] to compute the integral with respect to \(\mathbf {N}\) for each \(\ell \in [0,x]\), the latter expression is seen to be equal to
where \((S_b)_{b\ge 0}\) is a subordinator of Laplace exponent \({\widehat{\psi }}(\lambda ):= \frac{\sigma ^2}{2}\lambda + \int _0^\infty (1 - e^{-\lambda } )\Pi ([r, \infty ))\mathrm {d}r\) which does not depend on the drift coefficient \(\alpha \) of the underlying Lévy process X. In particular, the expression on the r.h.s. of (A.7) goes to 0 with \(\varepsilon \).
Let now \(({{\mathcal {Q}}}_\ell )_{\ell \ge 0}\) denote the right continuous completion of the filtration \((\sigma (\mathbf { M}([0,x],d\rho ,d\mathcal {N}): 0 \le x \le \ell ))_{\ell \ge 0}\), with \(\mathbf { M}\) the point process defined in (6.2). Since the first term on the r.h.s. of (A.5) is the expected supremum of the absolute value of a \(({{\mathcal {Q}}}_{\ell })_{\ell \ge 0}\)-martingale, we can bound it using BDG inequality by some universal constant \(C_1\) times
Written in terms of the excursion Poisson point process (6.2), the previous quantity reads
and can be estimated by the same arguments as in the proof of Lemma 1.3.2 of [12], as follows (see also the proof of Lemma 1.1.3 therein for details on related arguments):
where \( \varepsilon \mapsto X_{L^{-1}(\varepsilon )}\) is a subordinator of Laplace exponent \(\exp \left( - t\left( {\widehat{\psi }}(\lambda )+ \alpha \right) \right) \). That is, the same subordinator S as above, but killed at an independent exponential time of parameter \(\alpha \). Thus, we have \(\mathbb {E}( X_{L^{-1}( \varepsilon )}\wedge K)\le \mathbb {E}(S_ \varepsilon \wedge K) + K(1-e^{-\alpha \varepsilon }) \rightarrow 0\) as \( \varepsilon \rightarrow 0\). The statement now follows by bringing together (A.5), (A.8) and (A.7) with the r.h.s. of the latter replaced by its supremum over \(\varepsilon '\in [0,\varepsilon ]\), which is an increasing function of \(\varepsilon \) going to 0 as \(\varepsilon \rightarrow 0\), as required.
(b) Observe first that by continuity of \(s\mapsto L_s^0\)
We deduce that
Therefore, we have \(\sup \limits _{t\in [0,T_x]}| L_t^0- L_t^{\varepsilon }(m) |\le \sup \limits _{y\in [0,x]} \left| L^{\varepsilon }_{T_y}(m)- y \right| \) and it is enough to establish the required upper bound for the quantity
We have
and the absolute value of the second term on the right hand side is bounded by
thanks to the inequalities \(\mathbf {1}_{\{ 0<H_s \le 2\varepsilon \}}\ge \mathbf {1}_{\{ 0<H_s \le 2 \varepsilon ,m_s ((0,\varepsilon ))=0\}} \ge \mathbf {1}_{ \{0<H_s \le 2\varepsilon ,m_s((0,2\varepsilon ))=0\}}\). It follows from part a) that expression (A.9) is upper bounded by
and it only remains us to bound this last expectation. Notice to that end that the inner supremum can be written in terms of the Poisson excursions point process pruned below the level \(a=\varepsilon \) as considered in (6.4), namely
where \(I^{\vartheta ^{\varepsilon }}= \{ i\in I \, : \, m^{\vartheta ^{\varepsilon }}_{\alpha ^{(i)}}([0,\varepsilon ))= 0\}\). As before, let \({\widehat{L}}^0_t\), \({\widehat{T}}_x\), \({\widehat{H}}_t\) and \({\widehat{\tau }}^K\) denote the corresponding local time at 0, inverse local time at 0, height process, and the stopping time \({\widehat{\tau }}^K:=\inf \{s>0 \, : \langle {\widehat{\rho }}_s, 1\rangle \ge K\}\), all of them now associated with \(({\widehat{\rho }}, {{\widehat{\mathcal {N}}}})\), the snake process associated with (A.11).
Then, writing in a similar way as in (A.6) the time integral in (A.10), that is, as a sum of integrals over (now) non marked excursion intervals above level \(\varepsilon \), we get
Since \(T_x\ge {\widehat{T}}_{L^\varepsilon _{T_x}(m)}\) and \(\sup \limits _{t\le T_x} \langle \rho _t, 1\rangle \ge \sup \limits _{t\le {\widehat{T}}_{L^a_{T_x}} (m)} \langle {\widehat{\rho }}_t, 1\rangle \), the expectation in (A.10) is seen to be less than
by applying, thanks to Lemma 6.2, the previous part a) (with \(m=0\) or equivalently \({{\bar{\theta }}} =0\)) conditionally on \({{\mathcal {E}}}_{\varepsilon }\). With the obvious bounds \(L^\varepsilon _{T_x}(m)\le L^\varepsilon _{T_x}\) a.s., \(\mathbb {E}\left[ \sqrt{L^\varepsilon _{T_x}}\right] \le \sqrt{\mathbb {E}\left[ L^\varepsilon _{T_x}\right] }\) and the equalities \(\mathbb {E}( L^{\varepsilon }_{T_x} )= x \mathbf {N}\left( L^{\varepsilon }_{\zeta } \right) = x e^{- \alpha {\varepsilon }} \) following from Corollary 1.3.4 in [12], the desired result is seen to hold with \({{\mathcal {C}}}({{\bar{\theta }}} ,K,\varepsilon )= \left( 2\hat{{\mathcal {C}}}({{\bar{\theta }}} ,K,2\varepsilon ) + 3\hat{{\mathcal {C}}}(0 ,K,\varepsilon ) + \hat{{\mathcal {C}}}({{\bar{\theta }}} ,K,\varepsilon ) \right) \).
1.7 Lamperti-type representation
Proposition A.1
Let \(X_t\) be a Lévy process with Laplace exponent satisfying (1.1) and (1.2), and assume that the locally bounded competition mechanisms g is such that \(\lim \limits _{z\rightarrow 0} \frac{g(z)}{z}\) exists. For each \(x>0\) there is a unique strong solution \(U_t\) to the SDE (2.4). Moreover, if we set
with \(C_t\) the right inverse of \(\eta _{t}:=\inf \left\{ s>0:\int _{0}^{s}\frac{\mathrm {d}r}{U_r }>t \right\} \), there exists in some enlarged probability space a Brownian motion \(B^V\) and an independent Poisson point process \(N^V\) in \([0,\infty )^3\) with intensity measure \(\mathrm {d}t\otimes \mathrm {d}\nu \otimes \Pi (\mathrm {d}r)\), such that
for all \(t\ge 0\). Last, pathwise uniqueness (and then also in law) holds for (A.13).
Since \((Z_t(x): t\ge 0)\) in (1.5) satisfies (A.13) with the Brownian motion
we conclude that \((Z_t(x): t\ge 0)\) and \((V_t :t\ge 0)\) are equal in law.
Proof
Let \(B^X\) and \(N^X\) respectively denote a standard Brownian motion and a Poisson random measure on \([0,\infty )^2\) with intensity \(\mathrm {d}s\otimes \Pi (\mathrm {d}r)\) such that \(\mathrm {d}X_t= -\alpha \mathrm {d}t + \sigma \mathrm {d}B^X_t + \int _0^{\infty } r{\widetilde{N}}^X (\mathrm {d}t,\mathrm {d}r)\). Standard localization arguments using a sequence \(G^R\) of globally Lipschitz functions equal to G on [0, R] (the local Lipschitz character of G following from the assumptions) show the existence of a unique strong solution U until some random explosion time. For each \(R,K\ge 0\), set \(\tau ^R=\inf \{s\ge 0 \, : U^2_s\ge R^2\}\) and \(\theta _K=\inf \{s\ge 0 \, : [U,U]_s \ge K\}\). Applying Itô’ s formula to the solution of equation (2.4) with \(G_R\) instead of G while keeping in mind the sign of G, we get using Gronwall’s lemma that \(\mathbb {E}(U_{t\wedge \theta _K\wedge \tau ^{R}}^2) \le c+ c'K\) for some finite constants \(c,c'>0\) depending on t, x and the characteristics of X but not on G. Fatou’s Lemma then yields \(t\wedge \theta _K\le \tau ^{\infty } =\sup _{R\ge 0}\tau ^{R}\) a.s., from where \(\tau ^{\infty }=\infty \) a.s. and
Let us now set \(T_0:= \inf \{t>0:U_t=0\}\) and \(T:= \inf \{t>0:V_t=0\}= \inf \{t>0: U_{C_t}=0\}\wedge \eta _{\infty }.\) As \(C_t\) is right-continuous, we have that \(U_{C_T}=0,\) so that \(C_T = T_0\) since \(\eta _{r}= \int _0^{r \wedge T_0} \dfrac{\mathrm {d}s}{U_s}= \eta _{\infty }\) for all \(r \ge T_0\). In order to show that the time-changed process \(V = (U_{C_{t}}: t \ge 0)\) is solution of (A.13), we follow the arguments of Caballero et al. [10, Proposition 4] providing the existence of a standard Brownian motion B such that
and of a Poisson random measure N with intensity \(\mathrm {d}s \otimes \mathrm {d}\nu \otimes \Pi (\mathrm {d}r)\) such that
where \(((\Delta _n, t_n) : n \in \mathbb {N})\) is a fixed but arbitrary labeling of the jump times and sizes of V and \(((r^X_n, t^X_n) : n \in \mathbb {N})\) are the atoms of \(N^X\). We have in an \(L^2\) sense that
so that the compensated measures satisfy
Inserting the identities (A.15), (A.16) and (A.17) into equation (A.14), we deduce that
By (A.12), \(\mathrm {d}C_t = V_t \mathrm {d}t \) and \( \dfrac{G(|U_{C_t}| )}{U_{C_t}} \mathrm {d}C_t = G(|V_t|)\mathrm {d}t\), thus \(V=(V_t: t\ge 0)\) is a solution of (A.13). Uniqueness for (A.13) follows from general results in [14]. \(\square \)
Rights and permissions
About this article
Cite this article
Berestycki, J., Fittipaldi, M.C. & Fontbona, J. Ray–Knight representation of flows of branching processes with competition by pruning of Lévy trees. Probab. Theory Relat. Fields 172, 725–788 (2018). https://doi.org/10.1007/s00440-017-0819-4
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00440-017-0819-4
Keywords
- Continuous state branching processes
- Competition
- Stochastic flows
- Lévy-tree exploration and local times
- Poisson-snake
- Pruning
- Ray–Knight representation
Mathematics Subject Classification
- 60J25
- 60G57
- 60J80