The Setting
Discretizing a trilinear form on the real line produces a triple of complex matrices acting on a finite set of real nodes at three cyclic vertices. Localizing each vertex by a Gaussian kernel scales the matrix entries across edges and yields a vertex star matrix whose -power energy varies with the vertex centers and variances. Tracking those variations leads to score-weighted local Gram matrices and the Hessian quadratic cost of the matrix energy, while integrating over constant-sum center planes connects the total heat energy to a cubic array norm.
Matrix Energy and Gram Preliminaries
Let be a finite index set, and write for the space of complex square matrices indexed by . For , the real trace is the real number , and the real trace functional on is the continuous real-linear functional defined by . Given , the real Hilbert-Schmidt inner product is the real number .
Given , the Sylvester operator on is the linear operator mapping to , and the Sylvester penalty of and is the real number .✦ The variational functional on associated with is the real-valued function , and the root variational cost is the real number .
With denoting the square root of defined by the continuous functional calculus, the matrix energy (written ) is the real number . At , the matrix energy gradient is the continuous real-linear functional on given for by .
The Hessian quadratic cost of is the real number .✦ For , the Hessian bilinear form is the real number .
Given and a real weight vector , the weighted Gram matrix is the matrix in .
Triangle Block Matrices
Index the three vertices by , equipped with cyclic addition and . For a finite index set , the triangle block space is , on which any array of matrices in assembles into a complex square matrix in .
Given a matrix triple , the triangle matrix is the block matrix on whose -block is when , when , and when . For each , the vertex block projection is the diagonal matrix in equal to on entries with first coordinate and elsewhere. For and , the vertex star row is the block matrix on , and the vertex star Gram matrix is the block matrix on .
For a vertex weight family , the vertex root matrix is the diagonal matrix in with real square-root entry at . Given and , the weighted matrix array is the triple with entry . For , , and , the vertex star base row is the block matrix on , and the vertex star base column is the block matrix on .✦
The Definition
Definition 1. For real numbers , the Gaussian density is the real number defined by (1): (1)✦ Given , the Gaussian weight is the real number (2): (2)✦ The Gaussian score for is the real number given by (3): (3)✦ For sets and , an assignment , scalar , coordinates , and parameter functions , the grouped heat weight is the function mapping to . Given , , , , and for sets and , the grouped first-variation weight is the function mapping to if and to otherwise, while the grouped second-variation weight is the function mapping to if and to otherwise.
Fix a finite index set . For , , and , the vertex heat weight family assigns to and the node weight defined in (4): (4)✦ Given a matrix triple together with , , and , the heat matrix array is the weighted matrix array (5): (5)✦ At a vertex , the heat matrix energy for , , , and is the real number given by (6): (6)✦
For , vertex weights , and vertices , the fixed base matrix is the block matrix on defined by (7): (7)✦ The coordinate projection is the map given by , and for , the extended coordinate function is the function given by . For , , , , and , the local Gram matrix is the block matrix on given by (8): (8)✦ The local first variation for and is the block matrix on in (9): (9)✦ At and , the local second variation is the block matrix on defined by (10): (10)✦
For , , , , and , with , , and as in (8), (9), and (10), the heat energy gradient is the real number (11): (11)✦ The pure heat second variation at and is the real number given by (12): (12)✦ At and , the heat variance term is the real number (13): (13)✦ The heat Hessian cost for and is the real number defined by (14): (14)✦ Given and three vertices , the mixed heat Hessian pairing is the real number (15): (15)✦
For real level , the constant-sum center plane is the plane , and for an omitted coordinate and , the plane coordinate chart is the point in defined by (16): (16)✦ Given a function and , the plane integral is the real number defined using the chart by (17): (17)✦ The total plane heat energy for , , , and is the real number (18): (18)✦ For a matrix triple and step size , the cubic array norm is the real number (19): (19)✦
In the paper, the result reads: for , integer , nodes (), arrays , centers , and variances , one defines , , , , , , (), , , , the plane , , and , where , , , (), , (), , (), and stand for , , , , , , , , and . By contrast, a general finite index set , arbitrary node coordinates , vertex indices , and unconstrained real parameters , , and replace the symmetric lattice and positive variances, while the fixed base matrix and local Gram matrix unify the diagonal () and off-diagonal () cases.
What Is Checked
The definitions are stated over an arbitrary finite index set , complex square matrix triples , real node coordinates , arbitrary real scalars , real vertex triples , vertex weights , and real-valued functions , together with arbitrary sets and for the grouped weights. For the underlying constructions, the library proves that the real trace on complex square matrices indexed by a finite set is additive and real-homogeneous.