= Orthogonal idempotent
{title2=$e_i^2=e_i,\quad e_ie_j=e_je_i=0\ (i\ne j)$}
A finite family of <idempotents> $e_1,\ldots,e_r$ in a unital <ring> $A$ is orthogonal when
$$
e_ie_j=e_je_i=0\qquad(i\ne j).
$$
Suppose also that $e_1+\cdots+e_r=1$. For a unital left <module> $M$, these elements give a decomposition of its underlying <abelian group>:
$$
M=e_1M\oplus\cdots\oplus e_rM.
$$
The decomposition exists because $m=\sum_i e_i m$. It is unique because applying $e_j$ to a relation $\sum_i m_i=0$, with $m_i\in e_iM$, isolates $m_j=0$. This argument works for <modules> over general <rings>, without requiring an underlying <vector space>.
If every $e_i$ is central, each $e_iM$ is also an $A$-<submodule>, since $a(e_im)=e_i(am)$. For a noncentral example, take $A=M_2(k)$ acting on $M=k^2$ and the <matrix units> $E_{11},E_{22}$. Their images are the two coordinate axes. The <matrix unit> $E_{12}$ sends $(0,1)$ to $(1,0)$, so the second axis is not an $A$-<submodule>. Over a <field> the decomposition is still a <vector space> decomposition. The vertex <idempotents> of a <path algebra> similarly supply the separate spaces in a <quiver representation>.
Back to article page