Ising Broadcast Process on Recursive Trees

• Ising broadcast process is a binary spin-propagation model on random recursive trees where a root spin is transmitted along edges through a noisy channel.
• The process uses a degree-dependent attachment rule and a probabilistic flip mechanism to model spin inheritance as the tree grows dynamically.
• Majority reconstruction analysis reveals a spectral threshold via random-walk and Pólya urn formulations that determines when recovery outperforms random guessing.

Searching arXiv for the target paper and closely related prior work to ground the article in cited sources. The Ising broadcast process on random recursive or increasing trees is a binary spin-propagation model in which a root spin is transmitted along edges through a noisy channel while the underlying tree itself grows randomly over time. In the formulation studied in "A Random Walk Approach to Broadcasting on Random Recursive Trees" (Althaus et al., 2024), each vertex carries a spin in $\{-1,+1\}$, the initial vertex has spin $+1$ by symmetry, and every newly arriving vertex attaches to an existing vertex according to a degree-dependent attachment rule before inheriting its parent’s spin with an independent flip probability $q$. The associated inference problem is not the classical rooted-tree reconstruction problem: the observer sees only the final unrooted, unlabeled tree together with its spins, does not know which vertex is the root, and seeks to reconstruct the original spin using the global majority of spins (Althaus et al., 2024).

1. Model and Ising interpretation

The paper studies a growing tree process $(T_n)_{n\ge 1}$ in which, at time $n$, the tree $T_n$ has $n$ vertices labeled by arrival times $\{1,\dots,n\}$, and at time $n+1$ a new vertex attaches to an existing vertex according to an attachment rule depending on degrees (Althaus et al., 2024). Broadcasting is defined by assigning to each vertex $v$ a spin $+1$0, taking vertex $+1$1 as the unique starting point with spin $+1$2 by symmetry, and then propagating spins along newly created edges with noise level $+1$3: $+1$4

In Ising notation, with $+1$5, this is exactly the standard edge broadcast channel: $+1$6 The paper uses $+1$7 rather than $+1$8, but the two encodings are equivalent via

$+1$9

The reconstruction task differs from conventional formulations in two respects. First, the tree is not fixed in advance but generated dynamically by a recursive growth mechanism. Second, the root is unknown to the observer. The available data at size $q$0 are the unlabeled final tree $q$1 and the collection of spins $q$2; the arrival labels and root identity are hidden (Althaus et al., 2024). This makes the problem a variant of Ising reconstruction on trees in which global statistics, rather than root-to-leaf geometry, are central.

2. Random recursive tree classes

The underlying trees are random recursive trees in the sense that vertices arrive sequentially and attach to earlier vertices. The paper focuses on two parametric families: very simple increasing trees and shape exchangeable trees (Althaus et al., 2024).

For very simple increasing (v.s.i.) trees, with parameter

$q$3

the attachment probability at time $q$4 is

$q$5

where $q$6 is the out-degree of $q$7 at time $q$8 (Althaus et al., 2024). The case $q$9 gives uniform attachment, $(T_n)_{n\ge 1}$0 gives linear preferential attachment with respect to out-degree, and $(T_n)_{n\ge 1}$1 yields certain anti-preferential regimes with bounded degrees.

For shape exchangeable (s.e.) trees, the rule is

$(T_n)_{n\ge 1}$2

where $(T_n)_{n\ge 1}$3 is the total degree (Althaus et al., 2024). Because the rule uses full degree rather than out-degree, the root is structurally less distinguishable.

These attachment kernels generate a spectrum of recursively grown trees, from nearly uniform to strongly preferential. The paper states that these trees have logarithmic height and degree distributions that depend on $(T_n)_{n\ge 1}$4 (Althaus et al., 2024). For the broadcasting problem, the decisive feature is that attachment probabilities can be expressed in terms of low-dimensional color aggregates, which permits a Markovian reformulation.

3. Reconstruction objective and majority estimator

The estimator considered is the majority estimator, defined from the total spin

$(T_n)_{n\ge 1}$5

It outputs $(T_n)_{n\ge 1}$6 when $(T_n)_{n\ge 1}$7, and in case of a tie outputs a fair random bit: $(T_n)_{n\ge 1}$8 Its error probability at tree size $(T_n)_{n\ge 1}$9 is

$n$0

and the asymptotic error is

$n$1

The central question is: for which values of $n$2 and which tree models does $n$3 hold, so that majority beats random guessing (Althaus et al., 2024)?

To analyze this question, the paper introduces the color-difference process

$n$4

Since $n$5, one has

$n$6

(Lemma 2.1 in the paper) (Althaus et al., 2024). Thus, control of the sign of $n$7 is effectively equivalent to control of majority reconstruction performance.

A frequent misconception is to identify this directly with classical leaf-reconstruction on regular trees. The paper makes clear that the observable here is the global majority over all vertices of a recursively grown random tree, not the spin configuration on a prescribed generation (Althaus et al., 2024). This changes both the relevant state variables and the form of the threshold.

4. Random-walk and Pólya-urn formulations

A central contribution of the paper is the encoding of the process as both a two-dimensional inhomogeneous random walk with memory and a Pólya urn with randomized replacement (Althaus et al., 2024). This reduction is what makes precise asymptotic analysis possible.

Because $n$8 alone is not Markovian when $n$9, the paper introduces a second coordinate $T_n$0 measuring a degree-weight imbalance between colors. For v.s.i. trees,

$T_n$1

whereas for s.e. trees,

$T_n$2

Then

$T_n$3

with increment $T_n$4 (Althaus et al., 2024).

Defining

$T_n$5

the attachment probability to a red vertex is

$T_n$6

(Althaus et al., 2024). This identity exhibits the self-reinforcing structure: larger red imbalance increases future attachment probability to red.

In the s.e. case, the transition probabilities for $T_n$7 are written explicitly as

$T_n$8

The paper describes this as reminiscent of an Elephant Random Walk, because the sign of the increment repeats or flips a past direction with probabilities influenced by accumulated degrees (Althaus et al., 2024).

The Pólya-urn representation uses four types: $T_n$9 for weight types that determine attachment probabilities, and $n$0 for count types representing the numbers of red and blue vertices. The activities are

$n$1

For v.s.i. trees, the expected replacement matrix is

$n$2

and for s.e. trees,

$n$3

(Althaus et al., 2024). The urn variables $n$4 recover the imbalance coordinates through

$n$5

This dual formulation has methodological significance. The random-walk picture supports supermartingale and hitting-time arguments, while the urn picture enables the use of limit theorems controlled by the leading eigenvalues of the replacement matrix (Althaus et al., 2024).

5. Detectability threshold for majority

The principal negative result is a non-reconstruction theorem for the majority estimator. For each allowed $n$6, the paper defines

$n$7

It then proves that for both tree classes,

$n$8

(Theorem 2.1 / Theorem 5) (Althaus et al., 2024). Thus, above the threshold $n$9, majority is asymptotically no better than random guessing.

The mechanism is spectral. For v.s.i. trees, the eigenvalues of $\{1,\dots,n\}$0 are

$\{1,\dots,n\}$1

and for s.e. trees, the eigenvalues of $\{1,\dots,n\}$2 are

$\{1,\dots,n\}$3

The critical relation is

$\{1,\dots,n\}$4

where $\{1,\dots,n\}$5 are the top two eigenvalues (Althaus et al., 2024). In that regime, the color-count fluctuations are of order $\{1,\dots,n\}$6 or $\{1,\dots,n\}$7 with zero mean, and the normalized red-blue count difference converges to a symmetric normal distribution. Consequently,

$\{1,\dots,n\}$8

which yields asymptotic failure of majority reconstruction.

The paper explicitly compares this with the classical Kesten–Stigum condition for regular $\{1,\dots,n\}$9-ary trees,

$n+1$0

while noting that the present threshold is not identical (Althaus et al., 2024). A plausible implication is that, in recursively growing trees, the effective analogue of branching is encoded by the urn spectrum rather than by a fixed offspring number.

6. Small-noise regime and quantitative error bounds

The main positive result establishes a uniform small-noise bound. For each allowed $n+1$1, there exists a constant $n+1$2 such that for all $n+1$3,

$n+1$4

(Theorem 2.2 / Theorem 6) (Althaus et al., 2024). Hence, as $n+1$5,

$n+1$6

and for large $n+1$7,

$n+1$8

Thus the majority estimator becomes asymptotically almost perfectly accurate in the low-noise regime.

The proof proceeds through the random-walk formulation rather than the urn CLT. The paper analyzes the combined process $n+1$9 using supermartingales and hitting times, with bounds showing that the probability of trajectories leading to a negative final majority is at most of order $v$0 (Althaus et al., 2024). Among the stated ingredients are the supermartingale

$v$1

hitting times $v$2 of suitable regions in the state space, concentration estimates for the number of flips along color-preserving or color-changing attachments, and event decompositions conditioning on the process remaining in a reinforcement-favorable region (Althaus et al., 2024).

The paper does not claim an exponential small-$v$3 error law. Instead, it provides a polynomial-order upper bound. This distinction matters: the result certifies asymptotic recoverability by majority as noise vanishes, but does not identify the exact asymptotic form of the success probability.

7. Relation to prior work and broader significance

The study is situated within a broader literature on broadcasting on trees while extending earlier work on recursive tree models. It explicitly builds on Addario-Berry, Devroye, Lugosi, and Véber, who studied the same majority estimator for a subset of v.s.i. trees, namely uniform attachment and linear preferential attachment, using Pólya-urn methods in a more model-specific manner (Althaus et al., 2024). The present work extends the analysis to all very simple increasing trees and to shape exchangeable trees, and introduces a more model-agnostic random-walk formulation connected to inhomogeneous random walks with memory and Elephant Random Walks (Althaus et al., 2024).

Within the general Ising-on-trees literature, the setting differs from work on Galton–Watson or regular trees associated with Evans, Kenyon, Peres, Schulman, Lyons, and Mossel–Peres, where the standard focus is leaf-based reconstruction and Kesten–Stigum-type thresholds (Althaus et al., 2024). Here, the observation is the global spin configuration on the entire recursively generated tree, and the tree itself is part of the stochastic mechanism.

The paper also emphasizes links to Pólya urns and reinforced stochastic processes. The urn viewpoint draws on limit theorems such as those of Janson (2004) and DHW21, while the random-walk viewpoint relates broadcasting dynamics to memory-driven walk models (Althaus et al., 2024). This suggests a methodological bridge between information propagation on random structures and the theory of reinforced processes.

Several open directions are indicated implicitly by the methodology. These include extension of the random-walk and urn approach to growing graph models beyond trees, analysis of estimators more sophisticated than global majority, and identification of optimal reconstruction thresholds rather than thresholds specific to majority (Althaus et al., 2024). The discussion of “superdiffusive” Elephant Random Walk analogies for small $v$4 and large $v$5 further suggests that strong reinforcement may lead to non-Gaussian scaling and potentially sharper reconstruction phenomena (Althaus et al., 2024).

Taken together, these results show that the Ising broadcast process on broad classes of random recursive and increasing trees admits a majority-based detectability phase transition whose location is governed by spectral data of an associated urn. In this framework, the growth rule of the network is not incidental: it directly shapes the persistence of the root’s influence in global spin aggregates and therefore the asymptotic feasibility of reconstruction (Althaus et al., 2024).