14.19.4 Tensor Map Product Kronecker Representation Change
Tensor Map Product Kronecker Representation Change explains how tensor products transform under linear maps using Kronecker products.
Tensor Map Product Kronecker Representation Change is the study of what happens to the Kronecker product matrix representing when the basis of the combined space is changed by a matrix that is not itself of Kronecker product form, revealing that the recognizable Kronecker structure of the representation is a special feature tied to bases compatible with the tensor decomposition, not a property preserved under arbitrary changes of basis on the combined space.
Two Different Kinds of Basis Change
Changes Compatible With the Tensor Decomposition
A change of basis on given by a matrix of the form , for invertible acting on the -side and invertible acting on the -side, respects the decomposition of as a tensor product, since the new basis vectors are still simple tensors of new basis vectors of and individually.
Changes Not Compatible With the Tensor Decomposition
An arbitrary invertible matrix on the -dimensional space need not have this form; generic invertible matrices of size are not expressible as for any pair of smaller matrices, since the set of Kronecker products has strictly lower dimension, as a subvariety, than the full space of invertible matrices once and are both greater than one.
Behavior Under Compatible Changes
Kronecker Structure Is Preserved
When the basis change is on the domain and on the codomain, the transformed matrix
remains a Kronecker product, equal to , by the mixed-product property of the Kronecker product.
Explicit Verification of the Mixed-Product Property
This preservation rests on the identity
applied twice, first to combine the two left factors and then to combine the two right factors of the conjugation, and this identity holds for any conformable matrices, not only invertible ones.
Behavior Under Incompatible Changes
General Similarity Destroys Kronecker Form
If is an invertible matrix of size that is not a Kronecker product, the conjugated matrix
is still similar to , and therefore still represents the same linear map in a different basis of , but it need not itself be expressible as a Kronecker product of two smaller matrices, since the new basis vectors of produced by need not be simple tensors at all.
What Is Preserved Regardless
Even when the Kronecker form is lost, invariants of the underlying linear map that do not depend on any particular matrix representation, such as its rank, its trace, its determinant, and its eigenvalues, remain exactly the same, since these are similarity invariants; only the visibly factored block structure of the matrix, not the map itself, is sensitive to whether the change of basis respects the tensor decomposition.
Illustrative Contrast
A Basis of Entangled Vectors
Replacing the basis input elements of a four-dimensional with a basis containing a vector such as , which is not a simple tensor, produces a change of basis matrix that is not of Kronecker form; the matrix of relative to this new basis is a perfectly ordinary matrix representing the same map, but no block decomposition of it into a Kronecker product of two two-by-two matrices exists in general.
Recovering the Kronecker Form Requires Reverting
The recognizable block structure can only be recovered by returning to a basis built from simple tensors of individually chosen bases of and , confirming that the Kronecker representation is a convenience tied to a specific, tensor-compatible choice of basis rather than an intrinsic feature of the map itself.