Biased Annihilating Branching Process

• Biased annihilating branching process is a stochastic model where particles diffuse with bias, reproduce, and annihilate upon interaction.
• The models use rigorous probabilistic analysis, field-theoretic techniques, and large-scale simulations to quantify critical exponents and phase transitions.
• Bias influences universal behavior: local biases preserve directed percolation scaling while long-range biases yield continuously varying exponents and reentrant transitions.

A biased annihilating branching process is a class of interacting particle systems in which particles undergo both branching (reproduction) and annihilation (mutual destruction upon encounter), with an additional bias in their motion or interaction rates. In these nonequilibrium stochastic models, bias—whether in diffusion, branching, or interaction—plays a fundamental role in shaping the system's critical behavior, phase transitions, and universality class. Research in this area combines rigorous probabilistic methods, field-theoretic techniques, scaling analyses, and large-scale numerical simulations to characterize scenarios ranging from discrete lattice models to continuum settings.

1. Fundamental Mechanisms and Mathematical Framework

The prototypical annihilating branching process involves particles (denoted $A$) performing random walks, branching (e.g., $A \to (k+1)A$), and annihilating upon encounter (typically $2A \to \emptyset$). Bias is introduced via modification of the hopping rates (motion), branching rates, or through spatial inhomogeneity. The motion can be biased locally (e.g., hopping towards or away from nearest particles) or have long-range dependencies in the bias, characterized by a decay law such as $x^{-\sigma}$. The general evolution can be formalized in terms of Markov generators or master equations, for example:

$$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$

where terms are included for random walks ($q(i, j)$), annihilation ($a$), branching ($b$), and possibly coalescence or death (Athreya et al., 2012).

The field-theoretical (Doi–Peliti) formalism allows encoding the dynamics in path integral representations:

$$S^{(\mathrm{PA})}[\phi, \hat{\phi}] = \int_x \{ \hat{\phi}(\partial_t - D \nabla^2)\phi + \lambda (\hat{\phi}^2 - 1)\phi^2 \} \,,$$

with extension to include branching terms, where $D$ is the diffusion coefficient and $A \to (k+1)A$0 the annihilation rate (Benitez et al., 2012, Benitez et al., 2012). Here, bias can be incorporated as asymmetry in the diffusion term or through explicit functional dependence of reaction rates on the configuration.

2. Bias Implementation and Model Variants

Bias in annihilating branching processes has been implemented in several mathematically precise ways:

• Local Bias: Hopping rate from site $A \to (k+1)A$1 to site $A \to (k+1)A$2 becomes $A \to (k+1)A$3, with $A \to (k+1)A$4. In one dimension, biased hopping towards the nearest occupied site can be introduced with probability $A \to (k+1)A$5 (Mullick et al., 2018, Daga et al., 2018).
• Long-Range Bias: Hopping bias decaying with distance as $A \to (k+1)A$6, where $A \to (k+1)A$7 is the distance to the nearest particle. The transition probability then reads $A \to (k+1)A$8 ($A \to (k+1)A$9 for right/left) (Park, 2020, Park, 2020).
• Range-Limited Attraction/Repulsion: Bias is active only if the nearest particle is found within a range $2A \to \emptyset$0; otherwise, motion is unbiased. Infinite range ($2A \to \emptyset$1) corresponds to persistent, scale-free bias (Park, 2020).
• Dependent Rates: Jump or branching rates explicitly depend on the local occupation configuration (nearest neighbors or longer-range), as in dependent double branching annihilating random walks (Balázs et al., 2015, Nagy, 2016).

Physical realization includes both the bias in particle motion (e.g., self-generated chemotactic movement, local potential gradients) and bias in branching rates (e.g., environmental preference or competition effects).

3. Phase Transitions and Critical Behavior

The interplay between branching and annihilation, modulated by bias, yields diverse phase transition scenarios:

Even vs. Odd Offspring

• Odd-offspring processes ($2A \to \emptyset$2): Typically fall into the Directed Percolation (DP) universality class (Benitez et al., 2012, Benitez et al., 2012, Daga et al., 2018).
• Even-offspring processes ($2A \to \emptyset$3): With parity conservation, typically belong to the Directed Ising (DI)/Parity Conserving (PC) universality class, exhibiting distinct scaling exponents and critical behavior (Benitez et al., 2012, Park, 2020).

Effect of Bias

• DP Class Robustness: Local diffusion bias does not change the universality class or critical exponents of DP; scaling relations such as $2A \to \emptyset$4 are preserved (Daga et al., 2018).
• PC/DI Class Fragility: Bias alters the critical exponents, and, in several cases, fundamentally changes the universality class. For example, the PC class crosses to a new class under local bias (Daga et al., 2018). Infinite-range attraction produces non-DI exponents, while finite-range restores DI scaling (Park, 2020).
• Continuously Varying Exponents: For branching annihilating random walks with long-range attraction or repulsion decaying as $2A \to \emptyset$5, the critical decay exponent $2A \to \emptyset$6 changes continuously with $2A \to \emptyset$7 for $2A \to \emptyset$8, reaching the DI value for $2A \to \emptyset$9 (Park, 2020, Park, 2020).

These features are summarized in the following table:

Bias Type	Universality Class at $x^{-\sigma}$0	Universality Class at $x^{-\sigma}$1	Nature of Critical Exponent Δ
Long-range attr.	Non-DI (continuously varying $x^{-\sigma}$2)	DI (fixed exponent)	Varies with $x^{-\sigma}$3
Local bias (DP)	DP	DP	Unchanged
Local bias (PC)	New / altered (not PC)	New / altered	Changed by bias

4. Analytical and Numerical Techniques

Exact Solutions and Expansions

• Expansion about the Pure Annihilation (PA) fixed point: Solving the PA model exactly yields nontrivial IR scaling and enables controlled expansions in the branching rate $x^{-\sigma}$4 (Benitez et al., 2012, Benitez et al., 2012).
• Nonperturbative Renormalization Group (NPRG): Used to validate and connect field-theoretic expansions with numerical results, matching thresholds for the annihilation rate $x^{-\sigma}$5 at high accuracy (Benitez et al., 2012).
• Embedded Random Processes and Coupling: Systems with dependent rates or regulated by a width process (span between leftmost/rightmost particles) admit careful coupling to auxiliary processes, allowing analysis of phase transitions even in non-attractive cases (Nagy, 2016).

Monte Carlo and Finite Size Scaling

• Large-scale stochastic simulations determine critical points, exponents, and crossover behaviors. For models with spatial or temporal nonlocality, collapsing density and correlation data as functions of scaled variables quantifies crossover exponents (e.g., $x^{-\sigma}$6 in branching-annihilating attracting walks (Park, 2020) and branching-annihilating random walks with long-range repulsion (Park, 2020)).

5. RG Flows, Fixed Point Structure, and Crossover Behavior

Bias can fundamentally reshape the renormalization group (RG) structure:

• Threshold Phenomena: For odd-offspring DP-class models with bias, a nonuniversal (lattice-dependent) threshold $x^{-\sigma}$7 for the annihilation rate is computed exactly; for $x^{-\sigma}$8 the threshold vanishes (no active transition), but is finite for higher $x^{-\sigma}$9 (Benitez et al., 2012, Benitez et al., 2012).
• Fixed Point Instability: In PC/DI models, the branching rate's scaling dimension remains positive for all $$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$0 when treated at the exact non-Gaussian PA fixed point, implying the absorbing phase fixed point is generally unstable in the direction of branching (Benitez et al., 2012).
• Crossover Exponents: When a bias is introduced, the system can cross over from one universality class to another. For example, the crossover from DI to non-DI class under attractive bias has a measured exponent $$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$1 for $$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$2 (Park, 2020).

6. Reentrant and Nonmonotonic Transitions

Systems with long-range or dependent bias can exhibit reentrant phase transitions—a nonmonotonic sequence from active to absorbing and back to active phases as the branching rate or bias strength is varied. This is quantified via survival probabilities and threshold criteria (e.g., survival probability $$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$3 determines the bias threshold for phase transition under long-range repulsion (Park, 2020)). The system may display robust stability of the absorbing state for small or large parameter values, with intervening windows of activity (Park, 2020, Nagy, 2016).

7. Genealogical and Ancestral Bias

The “inspection paradox” applies to the genealogy of particles in branching processes: sampling a particle uniformly from a distant generation and tracking its ancestry preferentially selects more prolific lineages, inducing a bias in the observed reproduction events along these lineages (Cheek et al., 2022). In spatially regulated models, ancestral lineages can be systematically analyzed as random walks in dynamic random environments, establishing diffusive scaling (central limit theorems) for lineages, and revealing that even in locally regulated (annihilating) settings, the emergent large-scale genealogical dispersal is unbiased and Gaussian (Oswald, 2024).

8. Extensions, Applications, and Implications

Machine Learning Approaches

Recent advances deploy ML—such as convolutional neural networks and autoencoders—for supervised and unsupervised detection of phase transitions, critical exponents, and order parameters in BAW models, often surpassing traditional Monte Carlo in accuracy for small systems (Wang et al., 2023).

Practical Relevance

• Population Genetics: Genealogical structure and spatial dispersal in populations with local regulation.
• Statistical Physics: Nonequilibrium phase transitions, universality class identification in reaction-diffusion systems.
• Epidemiology/Ecology: Spread and extinction of species, infection dynamics under local bias or environmental gradients.
• Material Science: Nucleation and competition of domains in driven or regulated systems.

Summary Table: Key Features of Biased Annihilating Branching Processes

Feature	Effect	Reference(s)
Local diffusion bias	Alters PC/DI scaling; DP robust	(Daga et al., 2018)
Long-range bias ($$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$4)	Critical exponent $$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$5 varies for $$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$6; DI for $$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$7	(Park, 2020)
Reentrant transitions	Active $$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$8 absorbing $$G f(x) = \sum_{i, j} q(i, j) x(i) [f(x + 1_j - 1_i) - f(x)] + \sum_i a x(i)(x(i) - 1)[f(x - 2_i) - f(x)] + \cdots$$9 active as parameters vary	(Park, 2020)
Parity conservation	PC/DI universality or new class under bias	(Benitez et al., 2012, Park, 2020)
Ancestral bias	Lineages sample prolific ancestors (“inspection paradox”)	(Cheek et al., 2022, Oswald, 2024)

Concluding Remarks

Biased annihilating branching processes represent a rich class of stochastic systems in which local and long-range biases, parity constraints, and annihilation-induced non-attractiveness entwine to produce diverse and sometimes counterintuitive critical behaviors. Analytic control is possible through expansions about nontrivial fixed points, duality, stochastic coupling, and numerical approaches. The sensitivity of universality class, phase transition thresholds, and scaling exponents to the form of bias underscores the subtlety of nonequilibrium pattern formation and absorbing phase transitions in interacting particle systems.