The Setting
The graph of this construction is bipartite. On one side are points , on the other faces , each face a triple of points, and a point is joined to every face that contains it. Each of the edges is a corner : the face together with a slot that names one of its points.✦ A bipartite kernel is a real function on , where the finite sets and carry probability laws and , and a homomorphism of into it sends points to and faces to . Sidorenko’s conjecture for asserts that in every nonnegative bipartite kernel the homomorphism density of is at least the th power of the mean of .
The kernels of this section live on symmetric matrices over a finite field. Each vertex of is mapped to a type, an element of a finite set of types, together with a family of symmetric matrices indexed by a finite set of coordinates, and for each edge a random signed set of coordinates, drawn from a law that depends on the types at its two ends, selects the coordinates at which the two families are compared. A second construction combines the moments of these laws along the corners of into a coefficient .
Laws and Corners
A finite probability law on a finite set is a function with nonnegative values summing to .✦ The expectation of a real function on under is (1)✦ The uniform law on a nonempty finite set gives every point the weight .✦ The product of laws on and on is the law on with weight at .✦ The product law of a family of laws on , indexed by a finite set , is the law on maps with weight , under which the coordinates are independent with laws ; when every is and has elements it is written .✦
A fixed table lists the points of each face in slots ✦, and the point of a corner is the point in slot .✦ The sides of the faces are point pairs, and a second table lists the sides of each face in the order , , .✦ At a corner two sides of the face meet in : is the earlier of them in this order✦ and the later.✦
Signed Rank Layers
Let be an odd prime, the field with elements, and its quadratic character. For , is the subspace of symmetric matrices in the space of matrices over .✦ The radical of a symmetric bilinear form on a finite-dimensional space over is the subspace of vectors with for all , and induces on the nondegenerate radical quotient form .✦ The discriminant sign of is , with the Gram matrix of in a basis of , and it equals when has rank zero.✦ The sign of a matrix is the discriminant sign of the form on .✦
The signed rank layer of rank and sign is the set of matrices in of rank and sign ✦, and its layer mass is .✦ The rank kernel is the real function on (2)✦
The Claim
Definition 1. Fix a finite set of coordinates and a natural number . An activation is a pair of a subset and a sign vector ✦, and an activation law is a finite probability law on activations. For write ✦; the activated monomial is the function on activations (3)✦ and the moment function of an activation law is the function on pairs of subsets of (4)✦ The sample kernel of an activation is the function on pairs of families and in given by the product of rank kernels of rank on (5)✦ and the activation kernel of an activation law is its expectation .✦
Let and be laws on finite sets and of types, and let assign an activation law to each pair , . The typed matrix laws are on and on , where is the uniform law on ✦, and the typed kernel is the function on their product.✦ The kernel mean of a real function on under laws on and on is .✦ If and is odd, then for all , and the typed kernel has kernel mean one under the typed matrix laws, (6)✦
A corner table assigns to each corner a real function on pairs of subsets of , and a labeling assigns a subset to each point pair . The label product of at is the product over the corners (7)✦ and the active coefficient of is the sum over the labelings by nonempty sets (8)✦ For type maps from the points to and from the faces to , the typed coefficient is the active coefficient of the corner table whose entry at is the moment function of the activation law at the types of and : (9)✦ The type map law is the law of independent types, at each point and at each face, (10)✦ and the averaged coefficient is the mean of the typed coefficient under it, (11)✦ The bipartite moment of a real function on under laws on and on is the homomorphism density of in , (12)✦ Here is a family of values in indexed by the points , and a family of values in indexed by the faces .
In the paper, the result reads: types and have laws and , an activation law at is the law of a set with signs , , , with the zero-moment condition for nonempty , over the labelings of the point pairs by nonempty sets, and , where , and are the paper’s , and .
What Is Checked
The definitions are stated over finite probability laws on finite sets and over two pieces of fixed structure: the rank kernels (2) on symmetric matrices over , built from rank and discriminant sign, and the corners of with their points and sides , .