The Symmetric and Antisymmetric Parts of a Product at $n=2$
Introduction
Every product of two matrices splits into a symmetric and an antisymmetric part, the two parts have dimensions $3$ and $1$, and the product of a symmetric with an antisymmetric matrix exchanges the roles of the anticommutator and the commutator.
The article proves the split $XY=\operatorname{sym}(XY)+\operatorname{antisymmetric}(XY)$, shows that the trace reads only the symmetric part, treats the three kinds of product by the halves of the two factors, proves that $\operatorname{sym}(SK)=\tfrac12[S,K]$ and $\operatorname{antisymmetric}(SK)=\tfrac12\{S,K\}$, and records the Clifford identity $\{u,v\}=2\operatorname{tr}(uv)I$ for traceless symmetric matrices. It assumes The Two Halves over $\mathbb{R}$: Symmetric and Antisymmetric at $n=2$ and The Two Halves as Orthogonal Complements at $n=2$.
It defers the collection of the four tables to The Four Multiplication Tables: $SS$, $SK$, $KS$, $KK$ at $n=2$, the algebra axioms of the symmetric half to The Jordan Algebra $\operatorname{Sym}_2$, and the bracket of the antisymmetric half to The Lie Algebra $\operatorname{Skew}_2$: One Dimension, Hence Abelian.
The Two Parts of a Product
Definition. For a square matrix $P$ the symmetric part is $\operatorname{sym}P=\tfrac12(P+P^{\mathsf T})$ and the antisymmetric part is $\operatorname{antisymmetric}P=\tfrac12(P-P^{\mathsf T})$.
Theorem. Every matrix splits in one way as a symmetric plus an antisymmetric matrix, so every product splits too:
$$ XY=\operatorname{sym}(XY)+\operatorname{antisym}(XY),\qquad \operatorname{sym}(XY)=\tfrac12\bigl(XY+(XY)^{\mathsf T}\bigr),\qquad \operatorname{antisym}(XY)=\tfrac12\bigl(XY-(XY)^{\mathsf T}\bigr). $$
Proof. $P=\tfrac12(P+P^{\mathsf T})+\tfrac12(P-P^{\mathsf T})$ and the two terms are symmetric and antisymmetric. The sum is direct because a matrix that is both symmetric and antisymmetric satisfies $P=-P$.
Corollary. The product $XY$ is symmetric exactly when $XY=Y^{\mathsf T}X^{\mathsf T}$, and antisymmetric exactly when $XY=-Y^{\mathsf T}X^{\mathsf T}$.
Proof. The condition $P=P^{\mathsf T}$ reads $XY=Y^{\mathsf T}X^{\mathsf T}$ after applying the transpose to each factor.
The transpose of a product turns the order around. $(XY)^{\mathsf T}=Y^{\mathsf T}X^{\mathsf T}$, so the two parts of a product are written with the transposes of the factors and not with the factors themselves.
The Trace Reads Only the Symmetric Part
Proposition. $\operatorname{tr}(P)=\operatorname{tr}(\operatorname{sym}P)$ and $\operatorname{tr}(\operatorname{antisym}P)=0$.
Proof. The trace of a matrix equals the trace of its transpose, so $\operatorname{tr}(P^{\mathsf T})=\operatorname{tr}(P)$ and the antisymmetric part has trace $\tfrac12(\operatorname{tr}P-\operatorname{tr}P)=0$.
Corollary. $\operatorname{tr}(XY)=\tfrac12\operatorname{tr}\bigl(XY+Y^{\mathsf T}X^{\mathsf T}\bigr)$, and $\operatorname{tr}([X,Y])=0$ while $\operatorname{tr}(\{X,Y\})=2\operatorname{tr}(XY)$.
Proof. The first identity is the proposition applied to $XY$; the second follows from the cyclicity of the trace, $\operatorname{tr}(YX)=\operatorname{tr}(XY)$.
The determinant is not blind to the antisymmetric part. For $X=I$ and $Y=J$ the sum $I+J$ has determinant $2$ and the symmetric part alone has determinant $1$, so the antisymmetric part does change the determinant.
Proof. $\det(I+J)=\det\begin{pmatrix}1&1\\ -1&1\end{pmatrix}=2$, and $\operatorname{sym}(I+J)=I$.
Products of Two Symmetric Matrices
Theorem. For symmetric $S,T$,
$$ \operatorname{sym}(ST)=\tfrac12\{S,T\},\qquad \operatorname{antisym}(ST)=\tfrac12[S,T]. $$
Proof. $(ST)^{\mathsf T}=T^{\mathsf T}S^{\mathsf T}=TS$, so $\operatorname{sym}(ST)=\tfrac12(ST+TS)$ and $\operatorname{antisym}(ST)=\tfrac12(ST-TS)$.
Corollary. The symmetric part of a product of two symmetric matrices is symmetric and the antisymmetric part is antisymmetric, and both are nonzero in general.
Proof. The two formulas express the parts by the anticommutator and the commutator, which are symmetric and antisymmetric respectively.
Example. With $S=\sigma_{3}$ and $T=\sigma_{1}$ one has $\{S,T\}=0$ and $[S,T]=2J$, so the product $\sigma_{3}\sigma_{1}=J$ is purely antisymmetric.
Proof. $\sigma_{3}\sigma_{1}=J$ and $\sigma_{1}\sigma_{3}=-J$, so the sum vanishes and the difference is $2J$.
The product is symmetric exactly when the two matrices commute. For symmetric $S,T$ the product $ST$ is symmetric iff $[S,T]=0$.
Proof. $ST$ is symmetric iff $ST=TS$, which is the vanishing of the commutator.
Products of a Symmetric and an Antisymmetric Matrix
Theorem. For symmetric $S$ and antisymmetric $K$,
$$ \operatorname{sym}(SK)=\tfrac12(SK-KS)=\tfrac12[S,K],\qquad \operatorname{antisym}(SK)=\tfrac12(SK+KS)=\tfrac12\{S,K\}. $$
Proof. $(SK)^{\mathsf T}=K^{\mathsf T}S^{\mathsf T}=(-K)S=-KS$, so the symmetric part is $\tfrac12(SK-KS)$ and the antisymmetric part is $\tfrac12(SK+KS)$.
Corollary (the roles are exchanged). For a mixed product the symmetric part is the commutator term and the antisymmetric part is the anticommutator term, which is the reverse of the rule for two symmetric factors.
Proof. Compare the two theorems: for $S,T$ symmetric the symmetric part carries the anticommutator, and for the mixed product it carries the commutator.
Example. With $S=\sigma_{3}$ and $K=J$ one has $SK=\sigma_{1}$ and $KS=-\sigma_{1}$, so the product is symmetric, the commutator term is $\sigma_{1}$ and the anticommutator term vanishes.
Proof. $\sigma_{3}J=\sigma_{1}$ and $J\sigma_{3}=-\sigma_{1}$, hence $SK-KS=2\sigma_{1}$ and $SK+KS=0$.
Corollary. A product $SK$ with $S$ symmetric and $K$ antisymmetric is symmetric iff the two matrices anticommute.
Proof. By the theorem, the antisymmetric part vanishes exactly when $\{S,K\}=0$.
The reversed product. For $KS$ the same computation gives $\operatorname{sym}(KS)=-\tfrac12[S,K]$ and $\operatorname{antisymmetric}(KS)=\tfrac12\{S,K\}$, so the two mixed products have opposite symmetric parts and equal antisymmetric parts.
Proof. Apply the transpose to the theorem, or recompute with $(KS)^{\mathsf T}=S(-K)=-SK$.
Products of Two Antisymmetric Matrices
Proposition. The antisymmetric matrices of order two form the line $\mathbb{R}J$ with $J^{2}=-I$, so every product of two of them is a scalar multiple of the identity.
Proof. The antisymmetric matrices are the multiples of $J$ and $J^{2}=-I$.
Corollary. Every product $KK'$ of two antisymmetric matrices is symmetric, and its antisymmetric part vanishes.
Proof. $KK'=cc'J^{2}=-cc'I$, which is symmetric, and a scalar matrix has zero antisymmetric part.
The trace of the product is negative. For $K=cJ$ and $K'=c'J$ one has $\operatorname{tr}(KK')=-2cc'$, so the trace form is negative definite on the antisymmetric line.
Proof. $\operatorname{tr}(J^{2})=\operatorname{tr}(-I)=-2$.
The mixed and the pure cases are the whole list. Every product of two matrices falls in one of the three cases according to the halves of its factors, since every factor is symmetric or antisymmetric.
The Traceless Symmetric Matrices
Theorem. For traceless symmetric $u,v$ of order two,
$$ \{u,v\}=2\operatorname{tr}(uv)\,I . $$
Proof. In the Pauli basis $u=\vec a\cdot\vec\sigma$ and $v=\vec b\cdot\vec\sigma$ with real vectors $\vec a,\vec b$, the anticommutator of two Pauli matrices is $\{\sigma_{i},\sigma_{j}\}=2\delta_{ij}I$, hence $\{u,v\}=2(\vec a\cdot\vec b)I$ and $\vec a\cdot\vec b=\tfrac12\operatorname{tr}(uv)$.
Corollary. The anticommutator of two traceless symmetric matrices is a scalar, and its determinant vanishes unless the scalar is zero.
Proof. The result is a multiple of the identity, whose determinant is the square of the scalar.
Corollary. The commutator $[u,v]$ is antisymmetric and its size is governed by the cross product: $[u,v]=2i\,(\vec a\times\vec b)\cdot\vec\sigma$, a multiple of the antisymmetric unit, since a commutator of two symmetric matrices is antisymmetric and the line is one-dimensional.
Proof. $[\sigma_{i},\sigma_{j}]=2i\epsilon_{ijk}\sigma_{k}$, so the commutator is $2i$ times the Pauli combination of the cross product.
The traceless part is closed under the anticommutator up to the scalar term. The product $\{u,v\}$ leaves the traceless space and lands in the scalars, which is why the symmetric half with its trace is the right object and not the traceless part alone.
Summary
Every product splits into a symmetric and an antisymmetric part, of dimensions $3$ and $1$, and the trace reads only the symmetric part, so commutators are traceless and $\operatorname{tr}\{X,Y\}=2\operatorname{tr}(XY)$.
For two symmetric factors the symmetric part of the product is the anticommutator and the antisymmetric part is the commutator; for a mixed product the roles are exchanged, with $\operatorname{sym}(SK)=\tfrac12[S,K]$ and $\operatorname{antisymmetric}(SK)=\tfrac12\{S,K\}$; for two antisymmetric factors the product is a scalar matrix because $J^{2}=-I$. For traceless symmetric matrices the anticommutator is the scalar $2\operatorname{tr}(uv)I$, and the commutator is a cross product in the antisymmetric direction.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $\operatorname{sym}P$ | $\tfrac12(P+P^{\mathsf T})$ |
| $\operatorname{antisymmetric}P$ | $\tfrac12(P-P^{\mathsf T})$ |
| $\{X,Y\}$ | the anticommutator $XY+YX$ |
| $[X,Y]$ | the commutator $XY-YX$ |
| $\sigma_{1},\sigma_{3}$ | the two real Pauli matrices, symmetric |
| $J$ | $i\sigma_{2}$, the antisymmetric unit |
Further Reading
- Roger A. Horn and Charles R. Johnson, Matrix Analysis (Cambridge University Press, 2013), for the symmetric and antisymmetric parts of a matrix and their traces.
- V. S. Varadarajan, Lie Groups, Lie Algebras, and Their Representations (Springer, 1984), for the anticommutator identities of the Pauli matrices.
- The Two Halves over $\mathbb{R}$: Symmetric and Antisymmetric at $n=2$, for the two halves and their bracket.
- The Two Halves as Orthogonal Complements at $n=2$, for the trace form that separates the two parts.
- The Four General Products and Their Relations over the Biquaternion Algebra, for the same product splitting in the biquaternion language.