Cluster Representation of Renormalization Group Transformations and a Rigorous Proof for Convergence of the RG-Flow of the Ising Model to Trivial Fixed Points away from Criticality

Abstract: Many rigorous results in the modern era of statistical mechanics have been obtained through geometrical representations. A much studied tool in this setting is the random cluster representation of lattice spin models. Another area of statistical mechanics that is of great interest but lacking rigorous results is the theory of the renormalization group. This paper investigates the idea to find a common ground between these two concepts in order to obtain rigorous results on the renormalization group flow. A need for negative cluster-weights weakens the success of this approach. Nevertheless, we managed to establish a relation between the scaling limit of the renormalization group flow of the nearest-neighbour Ising model away from criticality with a simple one-dimensional dynamical system that follows the cluster connectivity of the renormalization group transformation. This allows for a rigorous proof of the convergence of this flow to the zero- and infinite-temperature fixed point respectively for a large family of renormalization group transformations. Explicit results will be established on $\mathbb{Z}<sup>2$ followed by a discussion of the available generalizations to higher dimensions.

• The paper develops a cluster-based representation of real-space renormalization transformations, using signed cluster weights and exponential bounds on single-spin influence to convert RG-flow questions into variance estimates.
• For the two-dimensional nearest-neighbor Ising model, it rigorously proves convergence away from criticality: flows reach the zero-temperature ordered fixed points when α∈(0,1/4), while α∈[1/4,1/2) drives every phase to the infinite-temperature fixed point.
• The results challenge the standard Wilsonian picture for part of the kernel family, while leaving critical behavior for α<1/4, endpoint parameters, higher-dimensional extensions, and broader universality questions unresolved.

Overview

This paper develops a geometrical, random-cluster-style representation of real-space renormalization group transformations (RGTs) and uses it to prove rigorous convergence statements for the RG-flow of the nearest-neighbour Ising model on $\mathds{Z}^2$ away from criticality (2608.18862). The central object is a family of $2\times2\to1$ block-spin kernels $T_\alpha$ on the square lattice, parametrized by $\alpha\in[0,1/2]$ and constrained by symmetry, probabilistic consistency, monotonicity preservation, and the requirement that a fully aligned block maps to an aligned coarse spin (conditions R1–R4). The main theorem establishes that for $\alpha\in(0,1/4)$,

$$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$

while for $\alpha\in[1/4,1/2)$ the flow collapses to the infinite-temperature fixed point for all $\beta$, including the entire low-temperature phase — thereby refuting the expected Wilsonian picture in that parameter range. The critical point $\beta_c$ remains open for $\alpha<1/4$.

Cluster representation of RGTs

The key structural idea is to reinterpret an iterated RGT not as a map on measures but as a measure on a multi-layered "RG-admissible" graph $2\times2\to1$0 with vertex set $2\times2\to1$1, where horizontal edges replicate the base lattice $2\times2\to1$2 at each level and vertical edges encode the block geometry. Spins live on all levels; open vertical bonds within each block enforce that connected spins are parallel, exactly as in the Edwards–Sokal coupling. A signed product measure over blocks, with weights matched to the transition probabilities of the kernel, then has level-$2\times2\to1$3 marginals equal to $2\times2\to1$4.

Conditioning on the bottom-layer configuration reduces correlation functions of the renormalized measure to expectations of local polynomials under the input measure:

$2\times2\to1$5

where the evolution map $2\times2\to1$6 computes, via cluster weights $2\times2\to1$7, the probability that a coarse spin connects down to the set $2\times2\to1$8 of fine spins. Two structural facts drive the analysis: any nonlinear probability kernel forces negative cluster weights (so $2\times2\to1$9 is necessarily a signed measure), and the derivative bound

$T_\alpha$0

provides exponential suppression of single-spin influence, which is the engine of every variance estimate in the paper.

Decimation and linear transformations

Decimation with transition parameter $T_\alpha$1 yields $T_\alpha$2, so the flow trivializes to $T_\alpha$3 from any input; imposing R4 ($T_\alpha$4) instead gives correlations decaying as $T_\alpha$5, i.e. convergence to product measures $T_\alpha$6 built from the percolation probability $T_\alpha$7. The same limit holds for the linear case $T_\alpha$8, proved via a variance computation using mixing of $T_\alpha$9. The paper argues, though without a formal proof, that any linear, symmetry- and monotonicity-preserving RGT produces $\alpha\in[0,1/2]$0 in the scaling limit — even at $\alpha\in[0,1/2]$1, where $\alpha\in[0,1/2]$2. Hence nonlinearity is necessary for saddle-point behaviour.

High-temperature phase

For $\alpha\in[0,1/2]$3 and all $\alpha\in[0,1/2]$4, Proposition 2 proves $\alpha\in[0,1/2]$5 via an Efron–Stein-type martingale argument applied to the Edwards–Sokal coupling. Because the finite-volume free-boundary measure is $\alpha\in[0,1/2]$6-symmetric, the conditional mean $\alpha\in[0,1/2]$7 is constant, leaving only the intra-configuration variance term, bounded by $\alpha\in[0,1/2]$8. Since $\alpha\in[0,1/2]$9 with $\alpha\in(0,1/4)$0 while $\alpha\in(0,1/4)$1, the variance vanishes geometrically. Consequently all correlations collapse to those of $\alpha\in(0,1/4)$2.

Low-temperature phase

Below $\alpha\in(0,1/4)$3 the $\alpha\in(0,1/4)$4 symmetry is broken and two new difficulties arise: the conditional mean depends on the shape of the boundary (infinite) cluster, and bond variables are no longer independent. The first term of the total-variance decomposition is handled by restricting to finite clusters, whose susceptibility $\alpha\in(0,1/4)$5 is finite by planar duality. The second requires a coupling lemma (Lemma 4), constructed via a continuous-time Glauber-type Markov chain, showing that flipping one edge only perturbs clusters inside the dual cluster surrounding that edge's dual. Combined with subcriticality of the dual system, exponential decay of dual encirclement events, and the isoperimetric inequality $\alpha\in(0,1/4)$6, this yields $\alpha\in(0,1/4)$7 for all $\alpha\in(0,1/4)$8.

The limiting value of $\alpha\in(0,1/4)$9 is then determined by the one-dimensional dynamics $$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$0. For $$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$1, $$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$2 has unstable fixed point at $$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$3 and stable fixed points at $$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$4; for $$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$5 the roles reverse. An induction using stochastic domination shows $$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$6 stays below (resp. above) the value $$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$7 attained at $$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$8, keeping the trajectory in the correct basin of attraction. This yields convergence to $$\lim_{n\to\infty}T_\alpha^n\mu_\beta^\pm=\begin{cases}\mu_{\beta=0}&\beta<\beta_c\ \mu_{\beta=\infty}^\pm&\beta>\beta_c\end{cases},$$9 for $\alpha\in[1/4,1/2)$0 and to $\alpha\in[1/4,1/2)$1 for $\alpha\in[1/4,1/2)$2 throughout the ordered phase — the latter being the result contradicting the expected RG picture. A corollary extends the $\alpha\in[1/4,1/2)$3 trivialization to $\alpha\in[1/4,1/2)$4 itself by sandwiching $\alpha\in[1/4,1/2)$5 between $\alpha\in[1/4,1/2)$6 and $\alpha\in[1/4,1/2)$7 under the order-preserving kernel.

Generalizations

In three dimensions with cubic blocks and rotation-invariant kernels, the cluster-weight system becomes three-parameter ($\alpha\in[1/4,1/2)$8). The high-temperature proof carries over verbatim whenever $\alpha\in[1/4,1/2)$9, since it relies only on exponential decay of connectivities, which holds in all dimensions. The low-temperature proof used planar duality essentially, but the paper notes that what was actually needed — exponentially decaying influence of conditioning on a single edge — should be obtainable in higher dimensions, though this is not carried out. Notably, some 3D parameter choices produce five rather than three fixed points of $\beta$0, complicating the basin analysis.

For Potts models with $\beta$1, Proposition 3 proves a negative result: no block-size-$\beta$2 RGT admits a cluster representation using vertical bonds alone beyond the linear one, because specifying all spins in one state leaves unresolvable freedom among the others. Allowing horizontal bonds within blocks circumvents this obstruction — demonstrated explicitly for the 3-state Potts model on the triangular lattice with $\beta$3 coarse-graining — but whether the Ising proofs extend to such constructions is left open.

Limitations and open questions

The paper is explicit about several gaps. The endpoint cases $\beta$4 and $\beta$5 fall just outside the method, though they are expected to behave like their neighbours. Most significantly, the behaviour at criticality for $\beta$6 — the existence of a non-trivial fixed point $\beta$7 — is untouched: the simple connectivity estimates fail there, and the saddle-point structure of $\beta$8 is merely necessary, not sufficient, for such a fixed point. The signed nature of the cluster weights prevents full-strength connectivity arguments (e.g., crossing probabilities) from being deployed. Finally, all results are specific to the nearest-neighbour Ising model; nothing is said about universality across models with arbitrary couplings, nor about whether the lifted dynamics on a space of interactions is well-defined, an issue known to fail in general due to RG pathologies.

Conclusion

The paper supplies a workable bridge between random-cluster geometry and real-space renormalization, reducing the RG-flow away from criticality to the analysis of a one-dimensional dynamical system plus variance estimates controlled by percolation inputs (finite susceptibility, mixing, dual subcriticality). It delivers a complete, rigorous classification of the scaling limits of the $\beta$9 family on $\beta_c$0 off criticality, including the novel result that the entire low-temperature phase flows to the trivial $\beta_c$1 fixed point for $\beta_c$2 — and, strikingly, that for $\beta_c$3 even the critical point flows to infinite temperature. Whether these techniques can reach the critical fixed point, higher dimensions, or $\beta_c$4 Potts models remains unresolved.