The joint critical Ashkin–Teller current limit

The explanation

Put two layers of tiny magnets on a square grid. Each magnet points up or down. In each layer, neighbors like to agree, with strength J. A second strength, U, couples the layers through the product of four magnets. This is the Ashkin–Teller model, which goes back to Ashkin and Teller in 1943Source 1.

The model has a critical line, where it sits exactly at a phase change. On that line sinh(2J) = e^(−2U). Take U = 0. Then the layers ignore each other, and J = ½ × asinh(1), about 0.4407. That is the critical point of the ordinary Ising model. Take U = J instead. Then e^(4J) = 3, so J is about 0.2747. That end point is the four-state Potts model, where each site has four options.

Along the line, one number g sets the large-scale picture. Put c = √(1 + e^(4U)). Then g = (8/π) × arcsin(c/2). At U = 0, c = √2 and g = 2. At the Potts end, c = 2 and g = 4. As U becomes very negative, c falls toward 1 and g falls toward 4/3. At U = −1, g is already about 1.35.

The model can be redrawn in two ways. One is a height function, a random landscape over the grid. The other is a pair of currents, random networks of edges. Their connected pieces are called current clusters, one family for the grid and one for its dual grid. The question is what all of these look like together when the grid spacing shrinks to zero. Alcalde López, Heeney and Lis stated a precise conjecture for thisSource 2.

At U = 0 the currents are double random currents, and g = 2Source 3. There, Duminil-Copin, Lis and Qian proved limits for the height and the cluster boundariesSource 4. The conjecture asks for the same picture along the whole critical line, with √2 replaced by √gSource 5. For part of the line, Duminil-Copin, Kozlowski, Lammers and Manolescu proved that a related height has a limitSource 6.

The paper claims the full conjecture at every point of the critical line, including the Potts endSource 7. The height converges to a Gaussian free field with the predicted strengthSource 8. A Gaussian free field is a random surface, a two-dimensional cousin of Brownian motion. The clusters converge to sets cut out of that same surface by a fixed recursive ruleSource 9. Roughly, the rule stops exploring the surface when it first reaches one of two set heights. The limit keeps every level of nesting, clusters inside holes inside clusters, all at onceSource 10. It holds in every bounded domain with a simple closed boundary.

The proof first identifies the height, then recovers the clusters from itSource 11. For the height, the paper looks at its correlations in several reflected frames. In frames turned by 45 degrees, certain fourth-order wiggles cancel exactlySource 12. That pins down the covariance of a free field. The harder step is new to this coupling. The paper must decide which small holes belong to the same current clusterSource 13. It must also rule out large pieces that no hole can see.

Some related questions stay outside the result. A stronger joint picture at the Ising point includes extra fields and measures. Those extra limits are not part of this theoremSource 14.

How sure is this? No Lean formalization is listed for this result. The paper relies on several earlier results, including the six-vertex work above, and proves the remaining steps. No specialist review of the paper has been published.

What it doesn't show. This does not include the extra fields and measures studied at the Ising point.