Alec Kriebel

A vertex maximum principle for subcritical reinforcement

Read the paper (PDF) · Download PDF · Supporting ZIP · Zenodo record

Plain-English explanation · Read on this page · Collaboration


Abstract

We give a self-contained alternative proof of positive-equilibrium uniqueness and almost-sure coordinatewise homogenization for subcritical reinforcement on countable graphs of bounded degree. Positive vertex firing rates are bounded above but may have infimum zero. The proof combines an exact compensated jump transform with a maximum principle for bounded positive complete trajectories in vertex coordinates. Initial edge counts are arbitrary finite positive integers. Qin has also proved the full homogenization theorem, with broader positive real initialization; that result is prominently credited here. Unrestricted uniqueness among nonnegative equilibria is false, and is kept separate from the corrected positive statement. No first-priority claim is made.

The idea in plain English

On a network with subcritical reinforcement, each edge count divided by time converges to a deterministic positive value. This note proves convergence through a maximum principle on the vertices, even when some positive clock rates are arbitrarily small.

Each vertex of a countable simple graph has an independent Poisson clock. When its clock rings, it chooses an incident edge with probability proportional to the edge’s current weight raised to the power α, and adds one to that weight.

The degrees are bounded by a finite constant D, and the firing rates satisfy 0 < pv ≤ P < ∞. Their infimum may be zero. Initial edge counts are arbitrary finite positive integers, with no common bound across the graph. Isolated vertices may be discarded; disconnected graphs are allowed.

Positive equilibrium and convergence

For every 0 ≤ α < 1, there is exactly one equilibrium positive on every edge. Writing Sv(x) = ∑e∋v xeα, its coordinates satisfy

xuv = xuvα(pu/Su(x) + pv/Sv(x)).

If Ne(t) is the edge count, then, almost surely,

Ne(t)/t → xe for every edge e.

All edge coordinates converge on one probability-one event. This is coordinatewise convergence; the theorem supplies no uniform spatial convergence or stochastic convergence rate.

The theorem is already established in broader form by Qin. Qin’s Theorem 1.1 allows arbitrary finite positive real initial weights. This note treats positive integer initialization and presents a different proof.

The vertex maximum principle

For 0 < α < 1, write zv = pv/Sv(x). Positive equilibria correspond to vertex solutions through xuv = (zu + zv)1/(1−α).

  1. A strictly convex finite-volume minimization constructs a positive equilibrium, and an approximate maximum argument proves its uniqueness.
  2. An exact compensated jump transform controls the stochastic counts and gives local positive growth bounds.
  3. Limits in logarithmic time are bounded positive complete trajectories in vertex coordinates. The maximum principle forces every such trajectory to be stationary, giving convergence.

The argument uses coordinate-dependent positive lower bounds. It does not assume a common lower bound on the firing rates or differentiate an unattained infinite supremum. At α = 0, Poisson thinning gives the limit directly.

Scope and limitations

Uniqueness is asserted among equilibria positive on every edge. Unrestricted uniqueness among nonnegative equilibria is false. For example, on the integer line with unit firing rates and 0 < α < 1, assigning weight two to a perfect matching and zero to the other edges gives an equilibrium; the constant-one array is another.

The theorem assumes positive initial counts, bounded degree and bounded firing rates. It makes no assertion for zero initial edges, the critical exponent, unbounded degree or uniform convergence across vertices.

Background and prior work

Status and AI use

Unrefereed preprint. AI tools assisted substantially in developing, writing and checking the work. Independent human review is welcome.

Read the paper

Open the published PDF · Download PDF · Original PDF on Zenodo

Supporting materials

Download source-and-verification.zip · Original support archive on Zenodo

Collaborate on this result

I welcome researchers in reinforcement processes or probability on infinite graphs who would like to independently check the exact martingale transform, the complete-trajectory maximum principle or the passage through coordinate-dependent lower bounds. Comparing the proof with Qin’s broader theorem or contributing to a revised manuscript and journal submission would also be useful.

Authorship and responsibilities would be discussed in relation to substantive contributions. Contact Alec.

Cite this paper

Kriebel, Alec (2026). A vertex maximum principle for subcritical reinforcement. Version 1.0. Zenodo. 10.5281/zenodo.23408971.

Download BibTeX

Licensing

The deposited work is available under Creative Commons Attribution 4.0.