The One-Sided Operators at $n=2$
Introduction
The left multiplication $L_{X}:Z\mapsto XZ$ is an operator of rank twice the rank of $X$, with the eigenvalues of $X$ each repeated twice, and its image is the set of matrices whose columns lie in the image of $X$; the right multiplications are the mirror image.
The article studies the two one-sided families of operators on the matrix space, computes their rank, kernel, spectrum and adjoint, identifies the images of the rank-one elements as the minimal left ideals, and shows how the two families sit inside the endomorphism algebra. It assumes $\operatorname{End}(M_2(\mathbb{C}))$ as an Algebra: the Regular Representation, The Minimal Left Ideals of the Biquaternion Algebra and The Hilbert–Schmidt Form at $n=2$.
It follows the biquaternion article The Endomorphisms of the Biquaternion Algebra for the same operators on the eight real coordinates.
The Left Regular Representation
The map $X\mapsto L_{X}$ is an injective algebra homomorphism, so the matrix algebra acts on itself by left multiplication.
Proposition. The map $X\mapsto L_{X}$ is a homomorphism of algebras and its kernel is zero, so the left multiplications form a copy of $M_2(\mathbb{C})$.
Proof. $L_{X}L_{X'}=L_{XX'}$ by associativity, and $L_{X}=0$ forces $X=L_{X}I=0$.
Corollary. The eight real dimensions of the matrix algebra sit inside the thirty-two real dimensions of the endomorphism algebra as the left multiplications.
Proof. The copy is a real subspace of dimension eight.
Corollary. The left multiplication by the identity is the identity operator, and the left multiplication by a scalar is the scalar operator.
Proof. Both statements are immediate from the definition.
Rank and Kernel
Theorem. For a matrix $X$ of rank $r$ the operator $L_{X}$ has rank $2r$ over $\mathbb{C}$ and kernel of dimension $4-2r$.
Proof. The image of $L_{X}$ is the set of matrices $XZ$ with $Z$ arbitrary; the columns of $XZ$ are the images under $X$ of the columns of $Z$, so the image consists exactly of the matrices whose columns lie in $\operatorname{im}X$, a space of dimension $2r$. The rank-nullity theorem then gives the kernel dimension.
Corollary. The operator $L_{X}$ is invertible exactly when $X$ is invertible, and its kernel is the set of matrices whose columns lie in the kernel of $X$.
Proof. The rank is the full four exactly when $r=2$; and $XZ=0$ means that each column of $Z$ is annihilated by $X$.
Corollary. For a rank-one matrix the operator $L_{X}$ has rank two, so its image is a minimal left ideal of the algebra.
Proof. The image is a two-dimensional left ideal, and the minimal left ideals are exactly the two-dimensional left ideals.
The Spectrum
Proposition. The eigenvalues of $L_{X}$ are the eigenvalues of $X$, each with multiplicity two; in particular the trace of $L_{X}$ is twice the trace of $X$.
Proof. The operator $L_{X}$ is the tensor product of $X$ with the identity on the two-dimensional column space, so every eigenvalue of $X$ occurs once for each basis vector of that space.
Corollary. The determinant of $L_{X}$ is the square of the determinant of $X$, and the operator is singular exactly when $X$ is.
Proof. The product of the eigenvalues with multiplicity is the square of the product of the eigenvalues of $X$.
Corollary. A left multiplication is diagonalisable exactly when $X$ is, and it is nilpotent exactly when $X$ is.
Proof. The Jordan form of a tensor product with the identity repeats the Jordan blocks of $X$.
The Right Multiplications
Proposition. The right multiplication $R_{Y}$ has rank $2\operatorname{rank}Y$, kernel the matrices whose rows are annihilated by $Y$, and the eigenvalues of $Y$ each twice.
Proof. The image consists of the matrices $ZY$, whose rows are linear combinations of the rows of $Y$; the same computation as for the left family applies with the two factors interchanged.
Corollary. The right multiplication is the transpose of the left multiplication under the identification of a matrix with its transpose, so the two families have the same rank and spectrum.
Proof. $R_{Y}(Z)^{\mathsf T}=Y^{\mathsf T}Z^{\mathsf T}$, which is the left multiplication by $Y^{\mathsf T}$ applied to the transpose.
Corollary. The two families have the same singular values, and the same minimal left and right ideals up to transposition.
Proof. Transposition is an isometry for the Hilbert–Schmidt form and exchanges left and right.
The Adjoints
Proposition. The adjoint of $L_{X}$ for the Hilbert–Schmidt form is $L_{X^{*}}$, and the adjoint of $R_{Y}$ is $R_{Y^{*}}$.
Proof. $\langle L_{X}Z,W\rangle=\operatorname{tr}((XZ)^{*}W)=\operatorname{tr}(Z^{*}X^{*}W)=\langle Z,L_{X^{*}}W\rangle$.
Corollary. A left multiplication is self-adjoint exactly when $X$ is Hermitian, and unitary exactly when $X$ is unitary.
Proof. The adjoint is the left multiplication by the conjugate transpose, and the identity of the two operators is the identity of the parameters.
Corollary. The Hilbert–Schmidt norm of $L_{X}$ is $\sqrt2$ times the Hilbert–Schmidt norm of $X$.
Proof. The squared norm of $L_{X}$ is the sum of the squared norms of the images of the four matrix units, and the image of a matrix unit is the corresponding column of $X$, so each column is counted twice.
Corollary. The left multiplications by Hermitian matrices form a real subspace of dimension four inside the self-adjoint operators, and the left multiplications by anti-Hermitian matrices form the corresponding anti-Hermitian subspace.
Proof. The four real dimensions of the Hermitian half carry over to the four real dimensions of the subspace of self-adjoint one-sided operators.
Where the Two Families Meet
Proposition. The left multiplication by $X$ equals the right multiplication by $X$ exactly when $X$ is a scalar.
Proof. The identity $XZ=ZX$ for every $Z$ is the statement that $X$ is in the centre, and the centre of the algebra is the scalars.
Corollary. The intersection of the two copies is the two real dimensions of the scalars, and the sum of the two copies has real dimension fourteen.
Proof. The intersection is the scalars, and the dimensions add with the intersection subtracted: eight plus eight minus two.
Corollary. The product $L_{X}R_{Y}$ is the operator $Z\mapsto XZY$, and its rank is at most twice the smaller of the two ranks.
Proof. The image is contained in the set of matrices whose columns lie in the image of $X$, which has dimension at most $2\operatorname{rank}X$; and symmetrically for the rows and $Y$.
The One-Sided Ideals
Proposition. The image of $L_{X}$ is a left ideal of the algebra, and it is the left ideal generated by the columns of $X$.
Proof. For any $A$ the matrix $A(XZ)=(AX)Z$ is again in the image, so the image is closed under left multiplication by the algebra.
Corollary. The images of the rank-one elements are exactly the minimal left ideals, and they are parametrised by the lines of the plane.
Proof. The image of a rank-one element is the set of matrices with columns in a fixed line, which is a minimal left ideal, and the line is the parameter.
Corollary. The algebra is the direct sum of any two minimal left ideals carried by independent lines, so the four complex dimensions split into two halves of dimension two.
Proof. Choosing two independent lines gives two minimal left ideals whose intersection is zero, and their dimensions add to four.
Summary
The left multiplication by $X$ is an operator of rank twice the rank of $X$, with the eigenvalues of $X$ repeated twice, kernel the matrices whose columns lie in the kernel of $X$, and image the matrices whose columns lie in the image of $X$. Its adjoint for the Hilbert–Schmidt form is the left multiplication by $X^{*}$.
The right multiplications are the transposed mirror of the left ones, with the same rank, spectrum and singular values. The images of the rank-one elements are the minimal left ideals, parametrised by the lines of the plane, and two independent lines split the algebra into two halves.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $L_{X}$ | the left multiplication $Z\mapsto XZ$ |
| $R_{Y}$ | the right multiplication $Z\mapsto ZY$ |
| $\operatorname{rank}X$ | the rank of the matrix |
| $X^{*}$ | the conjugate transpose |
| $\langle,\rangle$ | the Hilbert–Schmidt form |
Further Reading
- Paul R. Halmos, A Hilbert Space Problem Book (Springer, 1982), for one-sided multiplication operators.
- Rajendra Bhatia, Matrix Analysis (Springer, 1997), for ranks and spectra of tensor products.
- $\operatorname{End}(M_2(\mathbb{C}))$ as an Algebra: the Regular Representation, for the ambient algebra of the operators.
- The Minimal Left Ideals of the Biquaternion Algebra, for the ideals that appear as images.
- The Hilbert–Schmidt Form at $n=2$, for the form used to take adjoints.