Two-Spine Decomposition in Branching Processes

• Two-spine decomposition is defined via a double size-biasing of critical branching systems, marking two distinct lineages with correlated fluctuations.
• It is applied in discrete, measure-valued, and diffusion settings to derive moment identities and establish exponential limit laws like Yaglom's theorem.
• The method enhances classical one-spine decompositions by revealing second-moment effects and providing a robust framework for spectral and genealogical analysis.

Two-spine decomposition is a refined probabilistic representation which introduces two distinguished lineages—“spines”—within critical branching systems such as Galton–Watson trees, superprocesses, and branching diffusions. The construction captures the effects of second-moment biasing (quadratic size-bias) and is pivotal in deriving moment identities and limiting theorems, notably the exponential law in Yaglom-type results for critical systems. This approach generalizes classical size-biased (“one-spine”) decompositions by encoding not only the survival of lineages but also the correlated fluctuation structure arising at criticality (Schertzer et al., 2023); (Ren et al., 2017); (Ren et al., 2017).

1. Conceptual Foundations

The two-spine decomposition is defined via a double size-biasing of the measure describing the evolution of the branching system. In the case of a critical Galton–Watson tree with offspring variance $\sigma^2<\infty$, the process is altered by weighting trees according to $Z_n(Z_n-1)$, where $Z_n$ is the generation-$n$ population size. This biases towards trees in which two marked lineages (spines) survive to generation $n$. The method can be extended to continuous-mass superprocesses via double size-biased Kuznetsov measures $N_\mu^{(w_T(\phi))^2}$ and to branching diffusions via martingale change of measure, thereby constructing two interacting (or branching) spine trajectories (Ren et al., 2017); (Ren et al., 2017); (Schertzer et al., 2023).

2. Construction in Discrete and Measure-Valued Settings

For discrete branching processes such as the critical Galton–Watson tree, the two-spine system is constructed as follows. A random split generation $K_n$ is chosen uniformly in $\{0,\ldots,n-1\}$. Up to generation $K_n$, a single marked lineage propagates under the size-biased law; at $K_n$, a splitting occurs via an $Z_n(Z_n-1)$0–size-biased reproduction, and the two resulting spines then evolve independently (but under size-biased mechanisms) until generation $Z_n(Z_n-1)$1. All unmarked individuals branch according to the original law (Ren et al., 2017).

In measure-valued superprocesses, the analogous construction uses a double size-bias (see below for table), where the principal eigenfunction $Z_n(Z_n-1)$2 of the mean semigroup determines the $Z_n(Z_n-1)$3-transform, and the splitting time $Z_n(Z_n-1)$4 is a random variable with density proportional to the second-moment functional. This introduces a main spine and an auxiliary spine with associated independent “immigration” processes for mass (see Table below for structural comparison).

Model class	Biasing factor	Spines	Splitting time distribution
GW tree	$Z_n(Z_n-1)$5	2, at $Z_n(Z_n-1)$6	Uniform on $Z_n(Z_n-1)$7
Superprocess	$Z_n(Z_n-1)$8	2, at $Z_n(Z_n-1)$9	$Z_n$0
BBM (diffusions)	Martingale $Z_n$1	2, at $Z_n$2	Rate $Z_n$3 (continuous time)

This unified framework facilitates the analysis of high-order moments and conditional distributions under criticality (Ren et al., 2017); (Schertzer et al., 2023); (Ren et al., 2017).

3. Generators, Martingales, and Change of Measure

The two-spine framework realizes a change of measure by martingales which encode the dynamics of both spines. For instance, in branching Brownian motion (BBM) with inhomogeneous branching rate $Z_n$4 and critical drift $Z_n$5, the two-spine martingale is

$Z_n$6

where $Z_n$7 (principal harmonic), and $Z_n$8, with $Z_n$9 the invariant measure (Schertzer et al., 2023).

Under the new (two-spine) measure, the distinguished spines evolve either jointly (before splitting) or separately (after splitting), governed by a Markov generator of the form:

$n$0

This structure holds in diffusion, discrete, and measure-valued settings, with model-adapted modifications (Schertzer et al., 2023); (Ren et al., 2017); (Ren et al., 2017).

4. Limit Theorems and Applications

Two-spine decompositions directly drive the analysis of Yaglom-type laws and limit theorems. For the critical Galton–Watson process, the $n$1-size-bias corresponding to the two-spine construction enables a probabilistic proof that $n$2 (conditioned on survival) converges to $n$3. The fixed-point equation

$n$4

for $n$5, $n$6, is derived by decomposing $n$7 into subtree sizes along the two spines. This approach has been generalized to critical superprocesses, where analogous exponential limit laws arise for total mass (Ren et al., 2017); (Ren et al., 2017).

In BBM with general branching rates, the two-spine method yields, after suitable scaling and conditionalization, convergence of genealogies to the Brownian Coalescent Point Process—a random ultrametric tree described by a Poisson process of binary split points (Schertzer et al., 2023).

5. Spectral Structure, Ergodicity, and Mixing

The identification of the principal eigenfunction of the linearized mean-field operator allows explicit construction of the h-transform and thus of the spine dynamics. Ergodicity of the spine process arises under fairly general conditions; for instance, in BBM on $n$8, the invariant measure for the spine process is

$n$9

where $n$0 is the principal eigenfunction. The spectral gap $n$1 determines the exponential mixing time of order $n$2 for the spine chain. This spectral connection is critical for the full characterization of fluctuation behavior and relaxation to equilibrium under the two-spine measure (Schertzer et al., 2023).

6. Generalizations, Limitations, and Scope

The two-spine approach extends in principle to multitype branching processes, more general branching mechanisms, and measure-valued or spatial branching systems, provided the second-moment condition is satisfied and a canonical eigenfunction exists. In all such cases, the two-spine decomposition provides a probabilistic mechanism to uncover the effect of quadratic biasing and to connect directly to second-order martingale identities (Ren et al., 2017); (Ren et al., 2017).

A key limitation is the lack of projective consistency: in contrast to the single-spine construction, the family of two-spine measures does not admit a consistent infinite-horizon extension, which limits applications to finite-horizon or asymptotic results. Moreover, the method relies crucially on the system having finite variance and a well-behaved second-moment structure; different biasing and decomposition schemes would be required for infinite-variance or non-critical regimes (Ren et al., 2017).

7. Significance in Probability Theory

Two-spine decomposition underpins modern probabilistic proofs of classical exponential limit laws (Yaglom's theorem), facilitates the derivation of genealogical scaling limits, and provides a natural route to multispine and $n$3-spine decompositions required for higher moment and fluctuation analysis. It is an essential tool in understanding criticality and universality in branching systems, as well as for spectral and genealogical analysis in complex models of population dynamics (Schertzer et al., 2023); (Ren et al., 2017); (Ren et al., 2017).