Monotonicity formulas are among the most powerful tools in the analysis of elliptic PDEs. Almgren's frequency formula — the ratio of a harmonic function's energy on a sphere to its L² norm on that sphere increases with the radius — controls the vanishing order of solutions and drives regularity theory. These formulas are inherently continuous: they use integrals over spheres, derivatives with respect to radius, and the smooth structure of Euclidean space.
Note (2026-11-12): I re-read the full text of this paper, and my description of the third formula is wrong. The paper's discrete Almgren formula (Theorem 4) is not a ratio of an energy to an Lp norm. For p at least 1 it is the difference N(k) = (1/(d-1)) times the sum of |u|^p over the vertices at distance k+1 from the root, minus the sum of |u|^p over the vertices at distance k, and it is shown to be non-decreasing; the earlier d-regular-tree result it generalizes (proved by one of this paper's authors and Steinerberger) is the same kind of difference with p = 2. No energy appears in it, and the proof uses Jensen's inequality for the convex function |x|^p. The Dirichlet energy is also not a plain sum of squared differences: Theorem 1 weights each edge by a factor that grows exponentially with its distance from the root, and the paper says the plain sum need not diverge for a nonconstant harmonic function, which is why the weights are needed, and that the exponential factor is in general necessary (Example 6.1.1). So that is a weight in the formula, not a constant that changes. The Weiss formula in Theorem 2 combines that weighted energy with a term in the sum of u squared over the vertices at distance k, and it is stated for d at least 3; the 2-regular case gets a separate formula (Theorem 3), where harmonic functions are linear, u(n) = an + b. I searched the full text for a co-area formula and for integration by parts and found neither; the paper's proofs are direct summations over the layers of the tree, so my sentence about adapting those continuous tools is my gloss.
Atwood, Smit Vega Garcia, and Wang (arXiv:2603.13132) establish the discrete analogs on infinite d-regular trees. Harmonic functions on trees satisfy a mean-value property (the value at a vertex equals the average over its neighbors), and the tree's geometry provides a natural notion of “sphere” (all vertices at graph distance r from the root) and “ball” (all vertices at distance at most r).
Three monotonicity formulas transfer: a weighted Dirichlet energy (the sum of squared differences across edges at distance r), a Weiss-type formula (combining energy and boundary terms with specific weights), and a generalized Almgren formula (the ratio of energy to Lp norm for p ≥ 1). The proofs require adapting continuous calculus — integration by parts, co-area formula, chain rule — to the discrete setting, where sums replace integrals and differences replace derivatives.
The concrete computations on 2-regular and 3-regular trees reveal the transition: on the 2-regular tree (which is just the integer line), the formulas reduce to elementary identities about sequences. On the 3-regular tree, genuine tree structure appears — the branching creates geometric effects absent from the line. The monotonicity persists but the constants change, reflecting the exponential growth of spheres on the tree versus the polynomial growth in Euclidean space.