13.10.5 Tensor Matrix Multiplication Composition Meaning

Tensor matrix multiplication composition meaning explores how tensor operations combine linear transformations in multi-dimensional spaces.

Tensor Matrix Multiplication Composition Meaning is the interpretation of the matrix multiplication contraction case as representing the sequential application of two linear maps, one after the other, so that the resulting tensor encodes the single combined map obtained by first applying the transformation represented by one factor and then applying the transformation represented by the other. It captures what the contraction underlying matrix multiplication actually signifies beyond its formal index manipulation, namely the composition of linear transformations into a single equivalent transformation.

Conceptual Basis

Linear Maps as the Objects Being Composed

Each rank-two mixed tensor entering a matrix multiplication can be understood as representing a linear map between vector spaces. The composition meaning of the multiplication is that the product tensor represents the linear map obtained by applying the second factor's transformation to the output of the first factor's transformation, or vice versa depending on convention.

The Contracted Index as the Intermediate Space

The shared index summed over in the contraction corresponds to the intermediate vector space through which the composed map passes: a vector is first mapped into this intermediate space by one factor, and then mapped from the intermediate space into the final space by the other factor.

Composition Rather Than Mere Combination

The composition meaning distinguishes matrix multiplication from operations that merely combine two matrices entrywise, such as addition, emphasizing instead that multiplication produces an object encoding the effect of performing one transformation followed by another, a fundamentally different kind of combination.

Formal Description

Composing Two Linear Maps

Given linear maps $f : U \to V$ and $g : V \to W$ , represented respectively by mixed tensors $F_{j}^{i}$ and $G_{k}^{j}$ , the composed map $g \circ f : U \to W$ is represented by the matrix multiplication contraction:

P_{k}^{i} = F_{j}^{i} G_{k}^{j}

with the intermediate space $V$ corresponding exactly to the range of the shared index $j$ .

Action on an Arbitrary Vector

For a vector $u^{i}$ in $U$ , the composition meaning asserts that applying the product tensor directly:

P_{k}^{i} u^{i}

yields the same result as first applying $F$ to obtain an intermediate vector in $V$ , and then applying $G$ to that intermediate vector to obtain the final vector in $W$ .

Associativity as a Consequence of Composition

The associativity of matrix multiplication corresponds directly to the associativity of function composition, since applying three maps in sequence in either grouping, first two then the third, or first one then the remaining two, produces the same overall composed map.

Consequences of the Composition Meaning

Order Dependence

Because composing $f$ followed by $g$ generally differs from composing $g$ followed by $f$ , even when both compositions are individually well defined, the composition meaning directly explains why matrix multiplication is non-commutative in general.

Identity and Inverses

The composition meaning clarifies why the identity matrix behaves as a neutral element under multiplication, since composing any linear map with the identity map leaves the original map unchanged, and why an invertible matrix corresponds to a linear map possessing an inverse map that composes with it to yield the identity.

Iterated Composition and Powers

Repeatedly multiplying a square matrix by itself corresponds, under the composition meaning, to applying the same linear map repeatedly in succession, so that a matrix power represents the effect of iterating the underlying transformation that many times.

Applications

Modeling Sequential Processes

The composition meaning underlies the use of matrix multiplication to model any process consisting of sequential linear stages, such as successive coordinate transformations, successive states of a linear dynamical system, or successive layers of a linear transformation pipeline.

Change of Basis as Composition

Expressing a tensor in a new basis is itself understood through the composition meaning, as the sequential application of a change-of-basis map either before or after the original transformation, explaining the sandwich structure seen in similarity transformations.

Foundation for Tensor Network Composition

The composition meaning of matrix multiplication generalizes directly to the broader composition role of contraction discussed for tensors of higher rank, situating matrix multiplication as the simplest and most concrete example of composing transformations through shared contracted indices.