--- title: CDFI in Stochastic Processes url: https://www.emergentmind.com/topics/coming-down-from-infinity-property type: topic --- # CDFI in Stochastic Processes The coming-down-from-infinity (CDFI) property describes the phenomenon whereby a stochastic population or partition-valued process initialized from an infinite state attains a finite state in any strictly positive time interval. This phenomenon serves as a qualitative demarcation between processes that remain “stuck at infinity” versus those for which infinity is merely an entrance boundary. CDFI criteria have been studied for coalescent processes (particularly Kingman’s coalescent and its generalizations), exchangeable fragmentation–coalescence processes, pure death and birth–death chains, interacting particle systems, diffusion and branching models, and stochastic partial differential equations. ## 1. Formal Definitions and Foundational Models Let $(X_t)_{t\geq0}$ be a Markov process on a state space augmented by an infinite state (often $\mathbb{N}_0 \cup \{\infty\}$, partitions of $\mathbb{N}$, or function-valued spaces), with $X_0=\infty$. We say the process comes down from infinity if \[ \Pr(X_t < \infty \text{ for all } t > 0) = 1. \] In coalescent and fragmentation–coagulation models, $X_t$ often represents the number of blocks at time $t$. For branching, birth–death, or particle systems, $X_t$ may denote a (possibly infinite) population size or measure. A representative example is Kingman's coalescent, where any pair of blocks merges at rate $c>0$. Starting from infinitely many singletons, for any $t>0$, \[ \Pr(\text{number of blocks at time } t < \infty) = 1. \] This property is classical for Kingman’s coalescent and its $\Lambda$-coalescent generalizations with suitable merger rates [1602.05241][1105.2409]. ## 2. Necessary and Sufficient Criteria for CDFI CDFI is governed by explicit integral or series criteria balancing drift, branching, coalescence, competition, and fragmentation rates: - **Pure death/birth-death chains**: The process comes down from infinity iff \[ S = \sum_{n=1}^\infty \pi_n < \infty, \quad \pi_1 = 1, \quad \pi_n = \frac{\lambda_1 \cdots \lambda_{n-1}}{\mu_2 \cdots \mu_n} \] where $\lambda_n$ are birth rates, $\mu_n$ are death rates [1310.7402][1504.08160]. For pure death rates $\lambda_n=0$, the criterion is $\sum_{n=2}^\infty 1/\mu_n < \infty$. - **$\Lambda$-coalescents**: Schweinsberg’s criterion: CDFI occurs iff \[ \sum_{b=2}^\infty \frac{1}{\gamma_b} < \infty, \quad \gamma_b = \sum_{k=2}^b k\binom{b}{k} \lambda_{b,k} \] or equivalently, $\int^\infty dq/\psi(q) < \infty$ for the associated branching mechanism $\psi$ [1105.2409][1611.06280][2506.07067]. - **Fragmentation-coalescence**: For EFC processes, there is a phase transition characterized by \[ \theta_\star = \liminf_{n\to\infty} \sum_{k=1}^\infty \frac{n\,\overline\mu(k)}{\Phi(n+k)}, \quad \Phi(n) = \text{coalescent speed function} \] and CDFI occurs if $\theta^\star < 1$ [1605.07039][1602.05241]. - **CB processes with competition/drift**: Integral conditions involve the competition/growth function $g(\cdot)$ or generalized drifts $I(\cdot)$, \[ \int_{z_0}^\infty \frac{dy}{g(y)} < \infty \] for competition [1801.04501], or analogous Lyapunov-based conditions on the drift/branching interaction [2510.05958]. - **Time-changed Lévy and diffusion processes**: The condition \[ \int^\infty \frac{dx}{x \Psi(1/x) R(x)} < \infty \] where $\Psi$ is the log-Laplace exponent of the input Lévy process and $R(x)$ the time change, ensures CDFI [1901.10689][1711.08603]. ## 3. Quantitative Rates and Scaling Laws When CDFI occurs, the process typically follows a deterministic “speed” function $v(t)$ for small $t$: - **Coalescents**: For Kingman, $v(t) = 2/c t$; for Beta$(a,b)$-coalescents ($a<1$), $v(t) \propto t^{1/(a-1)}$. - **Birth–death processes**: If death rates are regularly varying, $v(t) \sim ((p-1)t)^{1/(1-p)}$ for $\mu_n \sim n^p$ [1504.08160][1310.7402][1607.08794]. - **Fragmentation-coalescence**: In the "fast" regime of $c > 2\lambda$, the scaling is $t N(t) \to 2/c$ [1602.05241]. - **Nested coalescents**: For nested Kingman, $N(t) \sim 2\gamma/c t^2$ with $\gamma \approx 3.45$, characterizing the joint action of species- and individual-level mergers [1803.08973]. - **CB with competition**: For quadratic competition, solution of $\frac{dz}{dt} = -g(z)$ yields the CDFI speed [1801.04501][2510.05958]. These rates determine not just the deterministic descent but also the scaling of fluctuations and higher-order corrections, often with central limit theorems or moderate deviation principles [1607.08794][1504.08160][2501.06684]. ## 4. Phase Transitions and Universality Classes Several processes display sharp phase transitions separating regions where CDFI holds from those where it fails: - **Fragmentation-coalescence (extreme shattering)**: For the Kingman coalescent with fragmentation rate $\lambda$, CDFI holds iff $\lambda < c/2$; at the boundary $\lambda = c/2$, the system is critical and does not come down [1602.05241]. - **Simple EFC processes**: Threshold is set by the parameter $\theta$: CDFI if $\theta < 1$, stays infinite if $\theta>1$ [1605.07039]. For regular variation, thresholds can be explicit: e.g. if $\Phi(n)\sim d n^{1+\beta}$, $\overline\mu(n)\sim \lambda n^{-\alpha}$, then $\beta = 1-\alpha$ is critical. These phase diagrams are robust across classes—analogous transitions occur in general coordinated particle systems, nested coalescents, and processes with generalized “infinite dispersion” mechanisms [2506.15736][1803.08973][1605.07039]. ## 5. Excursion Theory, Entrance Laws, and Pathwise Structure The analytic structure of excursions from infinity is accessible in regimes exhibiting CDFI. For instance, in the fast fragmentation–coalescence process ($\lambda < c/2$), the local time at $0$ for $M(t) = 1/N(t)$ has a pure-jump subordinator inverse; the stationary law is Beta–Geometric, and explicit entrance laws and hitting times are computable: \[ \lim_{t\downarrow0} t\,N(t) = \frac{2}{c} \] from typical excursions [1602.05241]. The Hausdorff dimension of the set $\{t : N(t) = \infty\}$ is exactly the phase parameter $\theta$. In metric settings, such as $\Lambda$-coalescent trees, CDFI is equivalent to the compactness of the associated metric measure space in the Gromov-weak topology. Failure to come down from infinity results in noncompact (and not even locally compact) limiting spaces [1105.2409]. ## 6. Generalizations and Future Directions Current research extends CDFI results to multitype and spatially structured models (nested and coordinated coalescents, Fleming–Viot processes with initial infinite support), time-inhomogeneous transition rates, and SPDEs or field-theoretic limits (e.g. dynamic $\Phi^4_3$ models [1601.01234]). Open questions include: - CDFI for broad classes of exchangeable fragmentation–coalescence with arbitrary dislocation laws. - Coupling genealogical or metric properties (e.g. fractal dimension of the zero set when $\theta<1$) to biologically interpretable quantities. - Extensions to interacting diffusions and multiple interacting species, particularly in non-Lipschitz or non-expansive settings [1511.07396]. - Excursion theory and scaling at critical phase boundaries [1602.05241][1605.07039]. ## 7. Illustrative Models and Table of Criteria | Process/Class | CDFI Criterion | Speed/Scaling Function | |--------------------------------------------|--------------------------------------------------------|---------------------------------------------------| | Kingman coalescent | Always (binary mergers at rate $c>0$) | $v(t) = 2/c t$ | | $\Lambda$-coalescent | $\sum_b \gamma_b^{-1} < \infty$ | $v(t)$ via $\int_{v(t)}^\infty dq/\psi(q)=t$ | | Simple EFC | $\theta^\star < 1$ | Finer scaling via $\Phi(n)$, fragmentation tail | | Birth–death ($\lambda_n$,$\mu_n$) | $S=\sum_n \pi_n < \infty$ | $v(t)$ from $E_\infty[T_n]$ inversion | | Fragmentation–coalescence ($\lambda$,$c$) | $\lambda