Random Current Representation

• Random Current Representation is a probabilistic framework for the Ising model, expressing observables as random currents that satisfy specific vertex parity constraints.
• Double random currents, obtained by summing two independent currents, enable the switching lemma which connects Ising correlations to percolation, dimer models, and φ⁴ field theory.
• The framework facilitates measure-preserving correspondences with dimer models and rigorous analysis of critical phenomena, including the vanishing of spontaneous magnetization at criticality.

The random current representation is a probabilistic formulation for the Ising and related models, where statistical-mechanical observables are expressed in terms of random configurations ('currents') obeying prescribed parity constraints at vertices. Double random currents, obtained as the sum of two independent currents, form a central tool to connect Ising correlation structures to combinatorial and probabilistic properties, facilitate proofs of key results such as the random current switching lemma, and illuminate connections to models such as dimers and $\varphi^4$ field theory. The framework enables direct, measure-preserving correspondences between random current cluster statistics, the dimer model height function, and the field-theoretical bosonization rules, as well as the analysis of critical phenomena, such as the vanishing of spontaneous magnetization at the critical point.

1. Random Current Representation: Definition and Structure

Let $G = (V, E)$ be a finite, connected planar graph with positive ferromagnetic Ising couplings $\{J_e>0 : e\in E\}$ and inverse temperature $\beta > 0$. Define

$$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$

For a fixed subset of 'sources' $B \subseteq V$, a $B$–current $\omega = (\omega_{\rm odd}, \omega_{\rm even})$ is a partition of a subset of edges $\omega \subseteq E$, with edges classified as 'odd' or 'even' such that $\omega_{\rm odd}$ has odd degree at $G = (V, E)$0 and even degree elsewhere; $G = (V, E)$1.

The law of a single random current with sources $G = (V, E)$2, $G = (V, E)$3, is given by

$G = (V, E)$4

where $G = (V, E)$5 normalizes the measure. Equivalently,

$G = (V, E)$6

The high-temperature expansion of the Ising partition function corresponds to summing over sourceless currents ($G = (V, E)$7).

2. Double Random Currents and Switching Lemma

Consider two independent currents, $G = (V, E)$8 and $G = (V, E)$9. Their sum $\{J_e>0 : e\in E\}$0 is defined by

$\{J_e>0 : e\in E\}$1

The law induced on $\{J_e>0 : e\in E\}$2 is the double current measure $\{J_e>0 : e\in E\}$3, given by

$\{J_e>0 : e\in E\}$4

with $\{J_e>0 : e\in E\}$5 the number of connected components or 'clusters' of the subgraph $\{J_e>0 : e\in E\}$6.

On the complete graph $\{J_e>0 : e\in E\}$7, the double random current representation underlies the switching lemma, which enables the computation of products/ratios of Ising correlations and provides a connection to the percolation cluster structure arising from random current configurations. The switching lemma also generalizes in the context of $\{J_e>0 : e\in E\}$8 field theory to include tangling probabilities (Krachun et al., 2023).

3. Measure-Preserving Correspondence to the Dimer Model

A canonical measure-preserving correspondence is established between double random currents on planar graphs and the dimer model on associated bipartite graphs. The construction utilizes a 'three-edge refinement' $\{J_e>0 : e\in E\}$9, where each undirected edge of $\beta > 0$0 is replaced by three parallel directed arcs with assigned weights:

• $\beta > 0$1
• $\beta > 0$2

Alternating flows—subsets of arcs alternating in–out in cyclic planar order—are mapped via $\beta > 0$3 onto double current configurations, such that the pushforward of the alternating flow measure yields the double current law:

$\beta > 0$4

Proceeding, a bipartite 'dimer graph' $\beta > 0$5 is constructed by local vertex refinements and the addition of weighted edges. The dimer measure $\beta > 0$6 on $\beta > 0$7 projects onto the double current measure via explicit surjective maps, allowing local functionals of double currents to be computed as expectations in the dimer model (Duminil-Copin et al., 2017).

4. Nesting Field and Correspondence to Dimer Height Function

For a double current sampled from $\beta > 0$8, the subgraph decomposes into clusters $\beta > 0$9, each assigned an i.i.d. sign $$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$0. The nesting field $$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$1 is defined by

$$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$2

where 'odd around $$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$3' refers to the parity of the cluster boundary in the low-temperature contour representation.

Under the dimer correspondence, the standard height function $$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$4 on faces of $$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$5 coincides in law, restricted to $$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$6, with the nesting field $$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$7 under $$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$8. This is formalized in Theorem 1.2 (Nesting–height correspondence), showing that under the coupling $$x_e = \tanh(\beta J_e) \in (0,1), \quad p_e = 1 - \sqrt{1 - x_e^2}.$$9, the distribution of $B \subseteq V$0 matches $B \subseteq V$1, thus equating dimer height fluctuations with the nesting field structure of double random currents (Duminil-Copin et al., 2017).

5. Scaling Limits, Clusters, and Tangling in Random Currents

On complete graphs, the configuration of clusters formed by open edges (those with positive current) in random current realizations encapsulates the percolation structure associated with the Ising model near criticality. In the near-critical Curie–Weiss Ising model, cluster sizes associated with double currents, properly normalized, exhibit nontrivial scaling limits as $B \subseteq V$2:

$B \subseteq V$3

where $B \subseteq V$4 is a random object capturing limit cluster sizes and random partitions (tanglings) of sources. Tangling probabilities, generalizations of connections between sources, interpolate between Wick pairings in the Gaussian regime and the Ising switching lemma in strong coupling. Explicit formulas for the limiting densities involve sums over backbone multigraphs and exhibit Gaussian tails for cluster sizes (Krachun et al., 2023).

6. Applications: Bosonization, Correlation Formulas, and Critical Phenomena

The random current representation, particularly in its double current form, yields bosonization identities for the Ising model, relating squared spin–disorder correlations to combinatorial properties of double current clusters. Specifically, for spins at vertices $B \subseteq V$5 and disorder operators along lines ending at faces $B \subseteq V$6,

$B \subseteq V$7

where $B \subseteq V$8 enforces cluster parity with respect to the insertion set. The dimer model formulation translates this into determinantal correlation formulas involving the sine/cosine of the height function (Duminil-Copin et al., 2017).

For planar biperiodic graphs, the variance of the nesting field increments grows logarithmically at criticality, and the expected number of sign-changes of the current field diverges, establishing the vanishing of the spontaneous magnetization at the Ising critical point—a fundamental result in statistical mechanics (Duminil-Copin et al., 2017).

7. Unified Framework: From Ising to $B \subseteq V$9 and Gaussian Field

Random current representations provide a unified combinatorial framework under which switching lemmas connect observables in the Ising model, the discrete $B$0 model, and the Gaussian free field. In the large-$B$1 limit on the complete graph, the law of currents interpolates:

• As $B$2 (Gaussian limit), tangling measures collapse to uniform pairings, and the switching lemma becomes Wick's theorem.
• As $B$3 with $B$4, all tanglings join into a single block, recovering the Ising switching lemma.

This unification illuminates deep connections between discrete, combinatorial stochastic processes and continuum field-theoretic frameworks, leveraging the tractability of random currents for rigorous results across models (Krachun et al., 2023).