The Setting
Take an index set , a family of complex square matrices indexed by on the edges of a triangle whose vertices form the triangle vertex set (the three-element set under cyclic addition, so ), real node coordinates , and a step size . Modulating the matrix entries by Gaussian heat weights with center and variance at the nodes () produces a star Gram matrix at each vertex , whose matrix energy varies with the center vector and variance vector .
The problem is to express the first and second partial derivatives in the centers and the derivatives in the variances (and along positive rays with ) of this vertex star energy in terms of the matrix energy gradient and Hessian quadratic cost, and to relate the variance derivative to the center second derivatives and the Hessian cost after integrating over a constant-sum center plane.
Matrix Energy and Hessian Costs
The real trace of a complex square matrix indexed by an index set is the real number . The real Hilbert-Schmidt inner product of two complex square matrices and indexed by an index set is the real number (equal to ).✦ The matrix energy of a complex square matrix indexed by with matrix square root is the real number (1)✦ At a complex square matrix indexed by with square root , the energy gradient is the continuous real-linear functional on complex square matrices indexed by that sends each matrix to (2)✦
For a complex square matrix indexed by an index set , the Sylvester operator is the operator mapping each complex square matrix indexed by to ✦, and the Sylvester penalty of and is the real number .✦ Given complex square matrices , , and indexed by , the Sylvester variational functional of , , and is the real number ✦, and the root variational cost of and is the real number (3)✦ where the supremum ranges over all complex square matrices indexed by . For complex square matrices , , and indexed by , the Hessian quadratic cost of and is the real number ✦, and the polarized Hessian pairing of and at is the real number (4)✦
The triangle block space on an index set is the Cartesian product .✦ For a family of complex square matrices indexed by , the triangle matrix is the complex square matrix on whose block in position (for ) equals when , equals when , and equals when : (5)✦ For each vertex , the vertex block projection is the diagonal complex matrix on whose diagonal entry at is when and otherwise.✦ Given a family of complex square matrices indexed by and a vertex , the vertex star row of at is the complex matrix on ✦, and the vertex star Gram matrix of at is the complex matrix .
Given a family of complex square matrices indexed by and a family of real node weights on , the weighted matrix array is the family of complex square matrices indexed by whose entry at (for ) is (6)✦ For a family of real node weights on , the vertex root matrix is the diagonal complex matrix on with diagonal entry at each .✦ For a family of complex square matrices indexed by , a family of real node weights on , and a vertex , the vertex star base row of and at is the complex matrix on ✦, and the vertex star base column of and at is the complex matrix on .✦ The weighted Gram matrix of a complex square matrix indexed by an index set and a real weight function is the complex matrix .✦
Gaussian Heat Weights and Plane Integrals
For a real variance and a real argument , the Gaussian density is the real number (7)✦ For real numbers , , and with , the Gaussian score is the real number ✦, and for a real step size , the Gaussian weight is the real number .✦ Given a step size , node coordinates , centers , and variances , the vertex heat weights form the family of real functions on given at each and by (8)✦
Given a step size , coordinates on an index set , a group assignment to a group index set , centers , and variances , the grouped heat weight is the real function on defined at each by (9)✦ Given , , , , and a group , the grouped first variation weight and grouped second variation weight are the real functions on that vanish at when and satisfy (10) when .
For a family of complex square matrices indexed by , step size , node coordinates , centers , and variances , the heat matrix array is the weighted matrix array formed from and the vertex heat weights ✦, and for each the heat matrix energy is the real number (11)✦ For a family of complex square matrices indexed by , a family of real node weights on , and vertices , the fixed base matrix is the complex matrix on that equals when and when .✦ For , , , , , and , write with the vertex heat weights , and let be the grouped weights on for group set , coordinates , group assignment , centers , and variances ; the local Gram matrix , local Gram first variation , and local Gram second variation are the complex square matrices on given by (12)
For and vertices , writing , , and , the heat first variation , heat variance term , heat Hessian cost , pure heat second variation , and mixed heat Hessian pairing are the real numbers defined by (13)
The constant-sum center plane at sum level is the subset of . Given and , the plane coordinate chart is the map sending each to the triple equal to at and to and at the remaining two vertices in increasing order.✦ For a function and , the plane integral is the Lebesgue integral (14)✦ For a matrix family on , step size , node coordinates , a sum level , a vector , and , writing (for any ) for the rescaled variance vector with -th entry , the total plane time derivative is the real number (15)✦
The Claim
Theorem 1. Fix an index set , a family of complex square matrices indexed by , a step size , and real node coordinates ; for a vector , a vertex , and , write for the vector equal to at and for .
Given , with for every , and , the function is differentiable at with derivative given by (16): (16)✦ For those same , the function is differentiable at and satisfies (17): (17)✦ In the variance variable , for the same , the map is differentiable at , yielding (18): (18)✦ When , satisfies for every , , and , the scaled-variance function is differentiable at with (19): (19)✦ If , satisfies for every , and satisfy , , and , then is differentiable at as in (20): (20)✦ For any sum level , any with for all , and any , the plane integrals over satisfy (21): (21)✦ Given a sum level , a vector with for all , and , the total plane time derivative obeys (22): (22)✦
In the paper, the result reads: fix a positive vector and set ; then for each , all centers , and all , one has and , where stands for the heat matrix energy, stands for the heat Hessian cost , () stands for , and stands for the mixed heat Hessian pairing The library separates the pointwise derivative identities (16) through (20) from the dissipation identities (21) and (22), stating (21) after integrating over the constant-sum plane and (22) after also summing over the vertices.
What Is Checked
The seven claims are stated over an index set , a family of complex square matrices indexed by , a step size , real node coordinates , center vectors , positive variance vectors (including rescaled vectors with and ), and sum levels , together with the Gaussian heat weights, vertex star Gram matrices, matrix energy and energy gradient, variational Hessian costs and pairings, and plane integrals over . For these constructions, the library proves that the real trace on complex square matrices indexed by an index set is additive and homogeneous under multiplication by real scalars.