Where the Half-Integers Go in $M_n(\mathbb{C})$

Introduction

The question. The biquaternion algebra carries the labels $0$, $\tfrac12$ and $1$. Under the principal copy at general $n$, which labels does the algebra carry, and where does the half-integer go?

The answer in one line. The algebra carries the integer labels $0,1,\dots,n-1$ at every $n$; the half-integer $j=\tfrac{n-1}{2}$ is the label of the module $\mathbb{C}^n$, which is not a subspace of the algebra except when $j$ happens to be an integer.

Why the article is needed. This is the point at which $n=2$ habits do the most damage. At $n=2$ the module has label $\tfrac12$, the algebra has labels $0$ and $1$, and the three numbers together look like the spectrum of one object. At $n=4$ the module has label $\tfrac32$ and the algebra has labels $0,1,2,3$; a reader who imports the $n=2$ picture looks for spin $\tfrac32$ inside $M_4$ and does not find it, and then doubts the decomposition.

Boundaries. The decomposition is The Spin Decomposition of $M_n(\mathbb{C})$; the carrier of spin zero is The Spin-Zero Carrier of $M_n(\mathbb{C})$; the carrier of spin one is The Spin-One Carrier of $M_n(\mathbb{C})$; the case $n=4$ in four readings is The Spin Content of $M_4(\mathbb{C})$. The biquaternion originals are The Spin-Half Bispinor Spaces and From the Spin to the Carriers and the Closures.

Conventions. $j=\tfrac{n-1}{2}$; the module is $\mathbb{C}^n\cong V_j$; the algebra is $M_n\cong\bigoplus_{k=0}^{n-1}V_k$ under conjugation. $V_s\cong V_t$ if and only if $s=t$.

The Two Objects

The module. $\mathbb{C}^n=V_j$, of dimension $2j+1=n$, with weights $-j,-j+1,\dots,j$, which are half-integers when $n$ is even and integers when $n$ is odd.

The algebra. $M_n=\bigoplus_{k=0}^{n-1}V_k$, with $k$ an integer, at every $n$.

The arithmetic of the two labels. The algebra's labels are differences of the module's weights, $d=m_a-m_b$, and the difference of two numbers that are both half-integers, or both integers, is an integer. Hence the algebra's labels are integers whether the module's label is or not.

The consequence. The half-integer cannot appear in the algebra as a weight difference. It can appear only as the label of a carrier, and the carriers have integer labels by the theorem.

Is the Module Inside the Algebra?

Proposition. The module's carrier $V_j$ occurs as a summand of $M_n$ if and only if $j\in\{0,1,\dots,n-1\}$, that is if and only if $n$ is odd.

Proof. The summands of $M_n$ are $V_k$ with $0\le k\le n-1$. Since $V_j\cong V_k$ only for $j=k$, the carrier $V_j$ occurs exactly when $j$ is one of those integers. With $j=\tfrac{n-1}{2}$ this is $0\le\tfrac{n-1}{2}\le n-1$, which holds for every $n\ge1$, together with the integrality of $j$, which holds exactly for odd $n$. $\square$

The table.

$n$ $j$ $j$ integral $V_j\subseteq M_n$ module weights
$2$ $\tfrac12$ no no $\pm\tfrac12$
$3$ $1$ yes yes, $V_1$ $-1,0,1$
$4$ $\tfrac32$ no no $\pm\tfrac12,\pm\tfrac32$
$5$ $2$ yes yes, $V_2$ $-2,\dots,2$
$6$ $\tfrac52$ no no $\pm\tfrac12,\dots,\pm\tfrac52$

The reading. For odd $n$ the module's spin is an integer and the module's carrier is one of the algebra's carriers; the question "where is the half-integer" does not arise. For even $n$ the module's spin is a half-integer and no carrier of the algebra has that label.

The precise negative statement. For even $n$ there is no subspace of $M_n$ that is isomorphic to $\mathbb{C}^n$ as a module for the principal copy. The module is the space on which the algebra acts, not a part of it.

Why the Biquaternion Case Looks Different

The real form. At $n=2$ the module is $\mathbb{C}^2=V_{1/2}$, and the algebra is $M_2\cong V_1\oplus V_0$. The biquaternion series reads both through a real structure: the algebra over $\mathbb{R}$ has the four-dimensional space $\mathbb{B}$, and the spinor plane and the bispinor space are real subspaces of it, and they are visible inside the algebra because the algebra is only four-real-dimensional.

Why half-integers appear there. The real algebra $\mathbb{B}$ acts on itself by left multiplication, and the left ideal $\mathbb{B}\tilde\Pi_+$ is a real subspace of the algebra isomorphic to the spinor module. So at $n=2$ the module is a subspace of the algebra over $\mathbb{R}$: the left ideals are inside the algebra.

The general-$n$ statement. For $n\ge3$ no left ideal of $M_n$ is a minimal carrier of the principal copy: the minimal left ideals are the spaces $M_nP$ for a rank-one idempotent $P$, of complex dimension $n$, and their module structure is the defining module, not a subspace of the carrier list.

Proof. $M_nP$ consists of the matrices whose image lies in the line $\operatorname{im}P$, so it is the space of all matrices with that image; under left multiplication it is the defining module. Its dimension is $n$, and its isomorphism type as a principal-copy module is $V_j$, which occurs in the carrier list only for odd $n$. $\square$

The honest conclusion. The biquaternion spin-half space is a left ideal of a four-dimensional real algebra. The general-$n$ object of the same shape is the left ideal $M_nP$, of complex dimension $n$, and it carries the label $j$ — which is a half-integer exactly when $n$ is even, and then it is not a summand of the algebra. The label lives on the ideal, and the ideal is not a carrier of the algebra.

Where Half-Integers Do Appear in the Algebra

Other groups. Under a group other than the principal copy the algebra can carry half-integer labels. At $n=4$, with the block reading on $\mathbb{C}^2\oplus\mathbb{C}^2$ and the group $SU(2)\times SU(2)$, the two off-diagonal corners carry two copies of $(\tfrac12,\tfrac12)$; the diagonal content of the whole algebra in that reading is $4V_1\oplus4V_0$, with no half-integer, but the pair labels $\tfrac12$ are visible in the corners.

The lesson. A half-integer label in the algebra is a statement about a group action, and the group must be named. Under the principal copy the answer is always "integers only".

The exception that is not one. The label $\tfrac12$ at $n=2$ is the principal copy's module label, and it is visible in the algebra only because the algebra happens to be four-real-dimensional and the left ideal is a real subspace. It is not a half-integer carrier of $M_2$.

What Not to Import

The wrong claim. "The algebra $M_4$ has spin content $\tfrac32 + \tfrac12 + \dots$ because the module has label $\tfrac32$." False: the algebra's labels are $0,1,2,3$, and $\tfrac32$ is not among them.

The wrong claim. "The spin-half carrier of $M_2$ generalises to a spin-half carrier of $M_n$." False: at even $n$ there is no carrier of label $\tfrac{n-1}{2}$; at odd $n$ the label is an integer.

The right claim. "The module has label $j$ and the algebra has labels $0$ to $n-1$; the algebra's top label is $2j$." True at every $n$, and it is the whole of the general statement.

The Finite Check

Computed for $n=2,3,4,5$: the Casimir $-\sum_i\operatorname{ad}_{J_i}^2$ (the $J_i$ anti-Hermitian) has the eigenvalues $k(k+1)$ with $k=0,\dots,n-1$, all even integers, and the eigenspace dimensions are $2k+1$; no half-integer eigenvalue occurs.

The module's weights were computed from $J_z=\operatorname{diag}(j,j-1,\dots,-j)$: they are $\pm\tfrac12$ at $n=2$, $-1,0,1$ at $n=3$, $\pm\tfrac12,\pm\tfrac32$ at $n=4$, $-2,\dots,2$ at $n=5$. The differences $m_a-m_b$ are integers in every case, with multiplicity $n-\lvert d\rvert$.

The left ideal was checked at $n=2,3,4,5$: for a rank-one idempotent $P$ the space $\{XP\}$ has dimension $n$ and is invariant under left multiplication, confirming that the module of the ideal is the defining module and not a carrier of the algebra.

The Four Lists at $n=4$

group carrier list module labels
principal $SU(2)$ $7+5+3+1$ the module $V_{3/2}$, not a summand
full $U(4)$ $1+15$ the defining module, not a summand
factorised on $\mathbb{C}^2\otimes\mathbb{C}^2$ $9+3+3+1$ $(\tfrac12,\tfrac12)$, a pair of half-integers
block on $S_+\oplus S_-$ $8+3+3+2$ two copies of $(\tfrac12,\tfrac12)$ plus $(1,0)\oplus(0,1)$

The observation. In every list the algebra's own labels are integers. The half-integers occur only as the pair $(\tfrac12,\tfrac12)$, never as a single half-integer label.

The reading. A single half-integer in the algebra would be a state of half-integer total spin. At $n=4$ there is none, and the factorised reading explains why: the two half-integers come in a pair, and the pair has integer total spin.

Why the Module Is Not a Subspace

The dimensions. The module has complex dimension $n$ and the algebra has complex dimension $n^2$; the carrier list of the algebra has $n$ summands of dimensions $1,3,\dots,2n-1$.

The labels. The module's label $j=\tfrac{n-1}{2}$ is half-integral for even $n$, and no summand of the algebra has that label, because the algebra's labels are $0,1,\dots,n-1$.

The conclusion. For even $n$ the module is not isomorphic to any submodule of the algebra, so the module does not occur as a submodule of the algebra at all.

The reading. At $n=4$ the module is a two-dimensional complex space of spin $\tfrac32$, and the algebra has carriers of spins $0,1,2,3$; there is no $\tfrac32$ anywhere in the algebra, and no subspace of the algebra carries it.

What a Half-Integer Carrier Would Require

The requirement. A half-integer carrier of the algebra would be a summand $V_{k/2}$ with $k$ odd, hence a summand of even dimension $2\cdot\tfrac{k}{2}+1=k+1$ with $k$ odd.

The obstruction. The decomposition is multiplicity-free with labels $0,1,\dots,n-1$, so no summand of that kind exists. A half-integer subspace that is not a summand is not a carrier, and it is not invariant.

The reading. The half-integer would have to be introduced by a different group action, and at $n=4$ the only place it can appear is a pair of labels coming from a product of two half-integer factors.

The Rule

The rule. In the algebra, integer labels only. Half-integer labels occur in the module, and in the algebra only as pairs coming from a product of two half-integer factors.

The reason. The algebra's labels are the differences of the module's weights, and the difference of two half-integers is an integer.

The reading. The rule is the article in one line, and it holds at every $n$: the parity of the labels is inherited from the parity of $n$, and the differences forget it.

The Counterexample That Is Not One

The apparent counterexample. At $n=4$ the block reading exhibits the labels $(\tfrac12,\tfrac12)$, and a reader may take them as half-integer labels of the algebra.

The correction. The pair has integer total label $\tfrac12+\tfrac12=1$, and the algebra's carrier $V_1$ is the corresponding summand. The half-integers are the two factors, not the label.

The reading. A half-integer label is a property of a single factor of a product group, and the principal copy has only one factor.

The Count

The numbers. At $n=4$ the carriers have dimensions $7,5,3,1$ and the module has dimension $2$. No summand has dimension $2$.

Summary

Under the principal copy the algebra carries the integer labels $0$ to $n-1$ at every $n$, and the module carries $j=\tfrac{n-1}{2}$, which is a half-integer for even $n$. The module's carrier is a summand of the algebra exactly when $n$ is odd. The biquaternion spin-half space is a left ideal of a real four-dimensional algebra, and its general analogue is the left ideal $M_nP$ of complex dimension $n$, which carries the label $j$ without being a carrier of the algebra. Half-integer labels inside the algebra require a different group, and the group must be named.

Summary of Notation

Symbol Meaning
$j=\tfrac{n-1}{2}$ the label of the module
$0,\dots,n-1$ the labels of the algebra
$V_j\subseteq M_n$ holds exactly for odd $n$
$M_nP$ the minimal left ideal, of complex dimension $n$
$(\tfrac12,\tfrac12)$ the pair label in the chiral corners at $n=4$

Further Reading