The Left Ideals and the Image of a Left Multiplication at $n=4$
Introduction
The subject. The left multiplication $L_A(X)=AX$ is a linear operator on the sixteen-dimensional algebra, and its image is the left ideal generated by $A$. The image has complex dimension $4\operatorname{rank}A$, the kernel has dimension $16-4\operatorname{rank}A$, and the assignment $A\mapsto\operatorname{im}L_A$ passes to a bijection between the left ideals of $M_4(\mathbb{C})$ and the subspaces of $\mathbb{C}^4$.
The order of the article. The first section computes the image, the kernel and the rank law. The second proves that the left ideals are exactly the subspaces of matrices whose columns lie in a fixed subspace, and that this description is a bijection. The third treats the right multiplication and the two-sided image. The fourth reads the minimal ideals, and the fifth separates this article from its neighbours.
It assumes The Two-Sided Operators and the Segre Cone at $n=4$ for the notation $T_{A,B}$, and The Matrix Algebra $M_4(\mathbb{C})$ for the matrix units $E_{ij}$. The model at order two is the same computation with $n=2$, and the classification of the minimal ideals is The Modules and the Minimal Left Ideals of $M_4(\mathbb{C})$.
The result in one line. The lattice of left ideals of the matrix algebra is the lattice of subspaces of the module, with dimensions multiplied by four, and the operator $L_A$ is the map that reads the column space of $A$.
Boundaries. The article treats images and not operators. The One-Sided Operators at $n=4$ treats the operators themselves: their multiplication laws, their matrices in the axis-free frame, and the group they generate. Nothing here is about the biquaternion display.
The Image of a Left Multiplication
Definition. For $A\in M_4(\mathbb{C})$ the left multiplication by $A$ is the operator $L_A$ on $M_4(\mathbb{C})$ given by $L_A(X)=AX$.
Proposition. $L_A$ is linear in $X$ and linear in $A$, the assignment $A\mapsto L_A$ is injective, and $L_{AB}=L_A L_B$.
Proof. Linearity is bilinearity of the product. Injectivity: $L_A(\mathrm I_4)=A$. The product law: $L_A L_B(X)=A(BX)=(AB)X=L_{AB}(X)$.
Proposition. The image of $L_A$ is the set of matrices whose columns lie in the column space of $A$,
$$ \operatorname{im}L_A=A\,M_4(\mathbb{C})=\{\,Y:\operatorname{col}Y\subseteq\operatorname{col}A\,\}, $$
and its complex dimension is $4\operatorname{rank}A$.
Proof. If $Y=AX$ then every column of $Y$ is $A$ applied to the corresponding column of $X$, hence lies in $\operatorname{col}A$. Conversely, if the columns of $Y$ lie in $\operatorname{col}A$, choose for each column $y_j$ a preimage $x_j$ with $Ax_j=y_j$, and let $X$ be the matrix with columns $x_j$; then $AX=Y$. The columns are chosen independently, each in a space of dimension $\operatorname{rank}A$, so the dimension of the image is $4\operatorname{rank}A$.
Corollary (the rank law). $\operatorname{rank}L_A=4\operatorname{rank}A$. The operator has rank $0,4,8,12,16$ according to the rank of $A$, and it is invertible exactly when $A$ is.
Proof. The rank of an operator is the dimension of its image.
Proposition. The kernel of $L_A$ is the set of matrices whose columns lie in $\ker A$, of dimension $4(4-\operatorname{rank}A)=16-4\operatorname{rank}A$.
Proof. $AX=0$ holds exactly when every column of $X$ lies in $\ker A$, and the columns are again chosen independently.
Corollary. The rank and the nullity add to sixteen, and the two numbers are $4\operatorname{rank}A$ and $16-4\operatorname{rank}A$.
Proof. The rank–nullity theorem, together with the two dimensions just computed.
The Left Ideals
Definition. A subspace $I\subseteq M_4(\mathbb{C})$ is a left ideal when $M_4(\mathbb{C})I\subseteq I$, that is, when $AY\in I$ for every $A$ and every $Y\in I$.
Theorem. For a subspace $U\subseteq\mathbb{C}^4$ put
$$ I_U=\{\,Y:\operatorname{col}Y\subseteq U\,\}. $$
Then $I_U$ is a left ideal of complex dimension $4\dim U$, and every left ideal is $I_U$ for exactly one $U$.
Proof. If $Y$ has columns in $U$ then $AY$ has columns in $U$, so $I_U$ is a left ideal; the columns are free, so $\dim I_U=4\dim U$. For the converse let $I$ be a left ideal and let
$$ U=\sum_{Y\in I}\operatorname{col}Y . $$
Every $Y\in I$ has its columns in $U$, so $I\subseteq I_U$. For the other inclusion let $Y$ have columns in $U$. Each column of $Y$ is a finite sum of columns of elements of $I$, so $Y$ is a sum of terms $A_iY_i$ with $Y_i\in I$: choose $A_i$ carrying the relevant column of $Y_i$ to the relevant column of $Y$ and killing the others. A left ideal is closed under left multiplication and under sums, so $Y\in I$. Hence $I=I_U$. Uniqueness: $U$ is recovered from $I$ as the union of the column spaces, and a subspace is the union of the column spaces of the elements of $I_U$.
Corollary. The map $U\mapsto I_U$ is a bijection between the subspaces of $\mathbb{C}^4$ and the left ideals of $M_4(\mathbb{C})$, order preserving, with $\dim I_U=4\dim U$.
Proof. The theorem, together with the evident implication $U\subseteq U'\Rightarrow I_U\subseteq I_{U'}$ and its converse, read on column spaces.
Corollary. The assignation $A\mapsto\operatorname{im}L_A$ induces this bijection: $\operatorname{im}L_A=I_{\operatorname{col}A}$, and two matrices have the same image exactly when their column spaces agree.
Proof. The first proposition identifies the image, and the theorem identifies the ideal attached to a subspace. The kernel of the assignment $A\mapsto I_{\operatorname{col}A}$ is therefore the set of matrices with a prescribed column space, which is not a single matrix but a full set.
The four levels. There are five left ideals up to the dimension of $U$, of complex dimensions $0,4,8,12,16$: the zero ideal, the minimal ideals, the ideals attached to planes, the ideals attached to hyperplanes, and the whole algebra. Each level is a Grassmannian: the minimal ideals are parametrised by the lines of $\mathbb{C}^4$, that is, by $\mathbb{P}^3$.
Corollary. The minimal left ideals are exactly the $I_U$ with $\dim U=1$, and they are the images $M_4P$ of the rank-one idempotents.
Proof. A left ideal is minimal exactly when its dimension is four, which forces $\dim U=1$. A rank-one idempotent $P$ has column space a line and satisfies $P^2=P$, so $M_4P=I_{\operatorname{col}P}$. Conversely a line determines the orthogonal projection onto it, which is a rank-one idempotent, and $I_U=M_4P$ for it.
Corollary. Every left ideal is principal: $I_U=M_4P_U$ for the projection $P_U$ onto $U$, and it is generated by any element whose column space is $U$.
Proof. The identity $M_4P_U=I_U$ of the previous corollary, and the fact that $M_4Y$ depends only on the column space of $Y$.
The Right Version and the Two-Sided Image
Proposition. For $B\in M_4(\mathbb{C})$ let $R_B(X)=XB$. Then $\operatorname{im}R_B=M_4(\mathbb{C})B$ is the set of matrices whose rows lie in the row space of $B$, of dimension $4\operatorname{rank}B$, and the rank law is $\operatorname{rank}R_B=4\operatorname{rank}B$.
Proof. Transpose the proof of the left case: $XB=(B^{\mathsf T}X^{\mathsf T})^{\mathsf T}$, and the rows of $XB$ are combinations of the rows of $B$.
Corollary. The right ideals are the subspaces $J_W=\{\,Y:\operatorname{row}Y\subseteq W\,\}$ for a subspace $W$ of the dual space, of dimension $4\dim W$, and $B\mapsto\operatorname{im}R_B$ induces the bijection with the subspaces of the dual.
Proof. The same argument as for the left ideals, with rows in place of columns.
Proposition. For $A,B\in M_4(\mathbb{C})$ the two-sided operator $T_{A,B}(X)=AXB$ has image
$$ \operatorname{im}T_{A,B}=A\,M_4(\mathbb{C})\,B=\{\,Y:\operatorname{col}Y\subseteq\operatorname{col}A,\ \operatorname{row}Y\subseteq\operatorname{row}B\,\}, $$
of complex dimension $\operatorname{rank}A\cdot\operatorname{rank}B$.
Proof. A matrix $AXB$ has columns in $\operatorname{col}A$ and rows in $\operatorname{row}B$. Conversely a matrix with both properties is $AXB$: choose bases adapted to the two subspaces and read the entries. The space of matrices with columns in a fixed space of dimension $r$ and rows in a fixed space of dimension $s$ has dimension $rs$, since such a matrix is determined by the $r\times s$ block of its coordinates in adapted bases.
Corollary. The image of $T_{A,B}$ is neither a left ideal nor a right ideal in general, and it is both exactly when one of the two factors is invertible.
Proof. If $A$ is invertible, $\operatorname{col}A=\mathbb{C}^4$ and the image is $M_4B$, a right ideal; if $B$ is invertible it is $AM_4$, a left ideal. If both are singular, a matrix with columns in $\operatorname{col}A$ need not have its columns in $\operatorname{col}A$ after left multiplication by an arbitrary element, so the image is not a left ideal.
Corollary. The image of $T_{A,B}$ is contained in $\operatorname{im}L_A$ and in $\operatorname{im}R_B$, with equality only in the two invertible cases.
Proof. The two inclusions are the two conditions of the proposition, and their dimensions are $\operatorname{rank}A\operatorname{rank}B$, $4\operatorname{rank}A$ and $4\operatorname{rank}B$.
A Worked Case
Example. Let $A=E_{00}+E_{11}$, of rank two. Then $\operatorname{im}L_A$ is the set of matrices whose last two rows vanish: the column space of $A$ is spanned by $e_0$ and $e_1$, so every column of $AX$ is a combination of the first two coordinate vectors. The dimension is eight, the kernel has dimension eight, and the kernel consists of the matrices whose last two rows vanish as well, since $\ker A$ is spanned by $e_2$ and $e_3$.
Corollary. For this $A$ the image and the kernel are complementary left ideals, attached to the coordinate plane and to its complement.
Proof. The two subspaces $\operatorname{col}A$ and $\ker A$ are complementary in $\mathbb{C}^4$, and the two ideals $I_{\operatorname{col}A}$ and $I_{\ker A}$ have dimensions eight and eight, meeting at zero.
Example. Let $A=E_{03}$, of rank one. Then $\operatorname{im}L_A$ is the minimal left ideal $I_{\mathbb{C}e_0}$, of dimension four, and the kernel has dimension twelve. The same minimal ideal is the image of $L_A$ for every matrix $A$ whose column space is the line $\mathbb{C}e_0$.
Proof. The column space of $E_{03}$ is the line spanned by $e_0$, and the previous corollary applies.
The dependence on the column space alone. The examples show the general rule: the image of $L_A$ depends on $A$ only through its column space, so the rank law is a statement about a subspace and not about a matrix. This is why the left ideals are the natural recipients of the construction, and why the assignment is a bijection on subspaces rather than on matrices.
The Boundary
Against the classification of the minimal ideals. The Modules and the Minimal Left Ideals of $M_4(\mathbb{C})$ classifies the minimal left ideals as modules: it proves that each is isomorphic to $\mathbb{C}^4$ as a left module, exhibits the idempotent that projects onto it, and computes the action of the algebra on it. The present article does not classify: it proves that the assignment of an ideal to a subspace is a bijection, and computes the dimension of every member of the family, minimal or not.
Against the one-sided operators. The One-Sided Operators at $n=4$ treats the operators $L_A$ and $R_B$ as elements of the endomorphism algebra: their multiplication laws, their matrices in the axis-free frame, and the finite group they generate for the units of the biquaternion subalgebra. The present article treats the images of these operators. The two articles meet in the rank law, which the other article states for the units and this one states for every matrix.
Against the Segre cone. The Two-Sided Operators and the Segre Cone at $n=4$ treats the two-sided operators with factors in the biquaternion algebra. There the factors have rank at most two, and the images have dimensions $0,1,2,4$. Here the factors range over the whole matrix algebra, and the two-sided image has dimension $\operatorname{rank}A\cdot\operatorname{rank}B$ with both factors up to four.
Summary
The left multiplication $L_A$ has image the matrices whose columns lie in the column space of $A$, of dimension $4\operatorname{rank}A$, and kernel the matrices whose columns lie in $\ker A$, of dimension $16-4\operatorname{rank}A$. The rank law is $\operatorname{rank}L_A=4\operatorname{rank}A$.
The left ideals of the matrix algebra are exactly the subspaces of matrices with columns in a fixed subspace $U$, of dimension $4\dim U$; the assignment $U\mapsto I_U$ is a bijection, and $A\mapsto\operatorname{im}L_A$ induces it. The minimal left ideals are the members with $\dim U=1$, parametrised by the lines of $\mathbb{C}^4$, and each is $M_4P$ for a rank-one idempotent.
The right version gives the matrices with rows in a fixed subspace, and the two-sided image $AM_4B$ is the set of matrices with columns in $\operatorname{col}A$ and rows in $\operatorname{row}B$, of dimension $\operatorname{rank}A\cdot\operatorname{rank}B$.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $L_A(X)=AX$ | the left multiplication by $A$ |
| $R_B(X)=XB$ | the right multiplication by $B$ |
| $T_{A,B}(X)=AXB$ | the two-sided operator |
| $\operatorname{col}A,\operatorname{row}A$ | the column space and the row space of $A$ |
| $I_U$ | the left ideal $\{Y:\operatorname{col}Y\subseteq U\}$ |
| $J_W$ | the right ideal $\{Y:\operatorname{row}Y\subseteq W\}$ |
| $M_4P$ | the minimal left ideal of a rank-one idempotent |
Further Reading
- The Modules and the Minimal Left Ideals of $M_4(\mathbb{C})$, for the classification of the minimal ideals.
- The One-Sided Operators at $n=4$, for the operators and their matrices.
- The Two-Sided Operators and the Segre Cone at $n=4$, for the products of the two families.
- The Ten Ranks of the Two-Sided Operators at $n=4$, for the rank of $T_{A,B}$ as an operator.
- The Matrix Algebra $M_4(\mathbb{C})$, for the conventions and the matrix units.