The Peirce Decomposition and the Corner Recursion of $M_2(\mathbb{C})$
Introduction
At order two the corner recursion has one step: an idempotent of rank one cuts the algebra into four corner lines, and each corner is a copy of $M_1(\mathbb{C})=\mathbb{C}$, so the recursion that runs $M_n\to M_{n-1}\to\cdots$ stops at the first move.
The article states the spectral flag at order two, the Peirce lines of a root with simple spectrum, the coarse split by a single idempotent with the dimensions $p^{2}$, $p(2-p)$, $(2-p)p$ and $(2-p)^{2}$, the corner isomorphisms, and the recursion, and it states what the decomposition is not. It assumes The Peirce Decomposition in Two Dimensions, The Peirce Basis and the Corners at $n=2$ and Lagrange Idempotents of a $2\times2$ Matrix with Simple Spectrum.
It defers the full general-order recursion to The Peirce Decomposition and the Corner Recursion of $M_n(\mathbb{C})$, and the general frame to The Frame, the Simple Nilpotents and the Gauge of $M_n(\mathbb{C})$.
The Spectral Flag at Order Two
A root of unity with simple spectrum at order two is an involution of trace zero, and it produces the flag of two lines. Let $H^{2}=I$ have the distinct eigenvalues $\omega_{+}=1$ and $\omega_{-}=-1$, with the projectors $P_{\pm}=\tfrac12(I\pm H)$. Put $V_{\pm}=\operatorname{im}P_{\pm}$, a line each. The two lines are independent and sum to $\mathbb{C}^{2}$, and the partial sums give the flag
$$ 0\subset F_{1}=V_{+}\subset F_{2}=\mathbb{C}^{2},\qquad \dim F_{k}=k . $$
Proposition. The root of simple spectrum at order two is exactly an involution with trace zero, and every such involution gives a complete flag.
Proof. An involution has the eigenvalues $\pm1$; the spectrum is simple exactly when both eigenvalues occur; and the trace is the sum of the eigenvalues, so it is zero. Conversely an involution with trace zero has the eigenvalues $1$ and $-1$, and its eigenspaces are two independent lines.
Corollary. The flag at order two is the single choice of a line, so the space of flags is the projective line.
Proof. The flag is $0\subset V_{+}\subset\mathbb{C}^{2}$, and it is determined by $V_{+}$, which is a line; the lines form the projective line.
Corollary. The standard flag is the case $H=\operatorname{diag}(1,-1)$, and every other involution of trace zero gives a flag in general position.
Proof. At the diagonal involution the two lines are the coordinate axes, and a non-diagonal involution rotates them off the axes.
The Peirce Lines
The Peirce corner $A_{kl}$ is the space of maps from the $l$-th line to the $k$-th line. Put
$$ A_{kl}=P_{k}M_2(\mathbb{C})P_{l},\qquad k,l\in\{+,-\}. $$
Proposition. Each $A_{kl}$ is a complex line, the four lines are independent and sum to the algebra, and the dimensions are $1,1,1,1$.
Proof. Each corner is the image of a nonzero linear map between two one-dimensional spaces, so it is a line; the four lines are independent because their sum has the dimension $1+1+1+1=4$, which is the dimension of the algebra; and the sum is the whole algebra by the Peirce decomposition.
Corollary. The Peirce line $A_{kl}$ multiplies by the index rule, $A_{kl}A_{lm}\subseteq A_{km}$ and $A_{kl}A_{lm}=0$ when $l\ne l'$.
Proof. $P_{k}XP_{l}\cdot P_{l}YP_{m}=P_{k}XP_{l}YP_{m}$ because $P_{l}^{2}=P_{l}$, and the product is zero when the intermediate projectors are orthogonal.
Corollary. The four lines are the four corners of The Peirce Basis and the Corners at $n=2$, with $\upsilon_{+}$ a generator of $A_{+-}$ and $\upsilon_{-}$ a generator of $A_{-+}$.
Proof. The two constructions use the same projectors, and the generators are the elements of the mixed lines normalised so that their product is the idempotent.
The Coarse Split
A single idempotent of rank $p$ cuts the algebra into four blocks of dimensions $p^{2}$, $p(2-p)$, $(2-p)p$ and $(2-p)^{2}$. Let $e$ be an idempotent of rank $p$.
Proposition. At order two the four dimensions are $p^{2}$, $p(2-p)$, $(2-p)p$ and $(2-p)^{2}$, which gives $1,1,1,1$ for $p=1$ and $4,0,0,0$ for $p=2$.
Proof. The corner $eMe$ consists of the matrices that are the composition of a map into the image of $e$, a map of the algebra and a map out of the kernel of $1-e$; its dimension is $p\cdot p$. The other three are the same count with one or both factors replaced by the rank of the complementary idempotent, which is $2-p$.
Corollary. The sum of the four dimensions is four in both cases, and the two extreme idempotents give a single nonempty block.
Proof. $p^{2}+2p(2-p)+(2-p)^{2}=(p+(2-p))^{2}=4$, and for $p=2$ the three other entries vanish.
Corollary. A rank-one idempotent gives the finest split of the algebra, since all four blocks are one-dimensional.
Proof. At $p=1$ the four entries are one each, so no block can be split further.
The Corner Isomorphisms
Theorem. Each corner at order two is isomorphic to $M_1(\mathbb{C})=\mathbb{C}$, and the isomorphism sends the unit of the corner to the idempotent.
Proof. The corner $eMe$ is one-dimensional, so every element is a scalar multiple of $e$; the product is $(\lambda e)(\mu e)=\lambda\mu e$, which is the multiplication of scalars, and $e$ is the unit of the corner.
Corollary. The two diagonal corners are copies of the scalars, and the two mixed corners are one-dimensional bimodules over them.
Proof. The diagonal corners are the ends of the algebra and their products are scalar products; the mixed corners are modules over the diagonal corners on the two sides, and each is one line.
Corollary. The algebra is generated by the four corner lines together with the products that pair them.
Proof. The four lines span the algebra, and every product of two elements is a sum of products of corner elements, each of which is again a corner element or zero.
The Recursion Stops
The general corner recursion replaces a frame of order $n$ by a diagonal corner isomorphic to $M_{n-1}(\mathbb{C})$ and a mixed corner, and the process continues until the order is one. At order two the first step already reaches order one.
Proposition. The corner $eMe$ of an idempotent of rank $n-1$ is isomorphic to $M_{n-1}(\mathbb{C})$; at order two with $p=1$ this is $M_1(\mathbb{C})=\mathbb{C}$.
Proof. Choose a basis whose first $n-1$ vectors span the image of $e$; in that basis $e$ is the diagonal idempotent with the first $n-1$ entries equal to one, and $eMe$ is the block of the first $n-1$ rows and columns.
Corollary. The recursion at order two has exactly one step, and the terminal algebra is the field itself.
Proof. The first step gives $M_1(\mathbb{C})$, which is the field, and the corner of the identity of the field is the field again, so nothing further happens.
Corollary. The recursion embeds $M_1(\mathbb{C})$ in $M_2(\mathbb{C})$ twice, once in each diagonal corner, and the two embeddings are conjugate.
Proof. The two diagonal corners are the corners of $e$ and of $1-e$, which are idempotents of the same rank, and a conjugating matrix carries one to the other.
The Biquaternion Case
Proposition. In the biquaternion language the coarse split is the splitting of the algebra into the four corners $\mathbb{B}_{st}$, and the recursion at order two is the statement that each corner is a copy of the scalars.
Proof. The corners $\mathbb{B}_{st}$ are the four products of the idempotent pair $\tilde\Pi_{\pm}$ with the algebra on the two sides, which is the same construction as the four corners here, and each corner is one complex line over the field.
Corollary. The companion nilpotents $\vec\Upsilon_{+-}$ and $\vec\Upsilon_{-+}$ generate the two mixed corners, and their product is the idempotent of the diagonal corner.
Proof. The companion pair of the corpus is the pair of generators of the mixed corners normalised so that the product is $\tilde\Pi_{+}$, which is the normalisation used here.
What the Decomposition Is Not
Proposition. The Peirce decomposition is not a direct-sum decomposition of the algebra into subalgebras, and the corners other than the diagonal ones are not closed under multiplication.
Proof. A mixed corner is square-zero, so it is not a subalgebra; and the product of two elements of different mixed corners lands in a diagonal corner, so the sum of the mixed corners is not closed either.
Corollary. The decomposition is a decomposition of the vector space with a multiplication rule across the pieces, not an isomorphism with a product of four algebras.
Proof. The four pieces span the space directly, and the multiplication rule is the index rule, which mixes the pieces.
Corollary. The only subalgebras among the four pieces are the two diagonal corners, and they are unital subalgebras with their own units.
Proof. The diagonal corners are closed under multiplication and contain $e$ and $1-e$, which are idempotents acting as units there.
Summary
At order two a root of unity with simple spectrum is an involution of trace zero, and it gives the flag $0\subset V_{+}\subset\mathbb{C}^{2}$, a single line. The Peirce corners are the four lines $A_{kl}=P_{k}MP_{l}$, each of dimension one, and a single idempotent of rank $p$ gives the dimensions $p^{2}$, $p(2-p)$, $(2-p)p$ and $(2-p)^{2}$, which are $1,1,1,1$ at $p=1$.
Each corner is isomorphic to $M_1(\mathbb{C})=\mathbb{C}$, so the corner recursion has one step and stops at the field. The decomposition is a splitting of the vector space with a multiplication rule across the pieces, not a product of subalgebras, and in the biquaternion language it is the splitting into the four corners $\mathbb{B}_{st}$.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $H$ | the involution of simple spectrum |
| $P_{\pm}$ | its projectors, $\tfrac12(I\pm H)$ |
| $A_{kl}$ | the Peirce corner $P_{k}MP_{l}$ |
| $p$ | the rank of the idempotent of the coarse split |
| $\mathbb{B}_{st}$ | the corners of the biquaternion algebra |
Further Reading
- Richard S. Pierce, Associative Algebras (Springer, 1982), for the corner decomposition of a finite-dimensional algebra and the corner isomorphisms.
- Kevin McCrimmon, A Taste of Jordan Algebras (Springer, 2004), for the recursive use of the Peirce decomposition.
- The Peirce Decomposition in Two Dimensions, for the four spaces and the index rule.
- The Peirce Basis and the Corners at $n=2$, for the corners and the gauge.
- The Peirce Decomposition and the Corner Recursion of $M_n(\mathbb{C})$, for the recursion at general order.