Theorem. Projectors of an involution on a module [fdip-003O]
AGENTDRAFTED
Let \(k\) be a commutative unital ring in which \(2\) is a unit,
put \(h=2^{-1}\), and let \(T:M\to M\) be linear with \(T^2=I\).
The maps \(E_\varepsilon =h(I+\varepsilon T)\), for
\(\varepsilon \in \{1,-1\}\), give
\[\begin {aligned}
E_\varepsilon ^2&=E_\varepsilon ,&E_+E_-&=E_-E_+=0,\\
E_++E_-&=I,&E_+-E_-&=T,\\
\operatorname {im}E_\varepsilon &=\ker (T-\varepsilon I)=M_\varepsilon .
\end {aligned}\]
Thus \(M\cong M_+\oplus M_-\), with map
\(v\mapsto (E_+v,E_-v)\) and inverse addition. No finite generation,
freeness or pairing is needed.
If an algebra acts by \(\alpha :A\to \operatorname {End}_k(M)\),
these projectors are \(A\)-linear precisely when
\[T\alpha (a)=\alpha (a)T\qquad (a\in A).\]
Indeed this is exactly \(A\)-linearity of \(T\); its linear combinations
with \(I\) are then \(A\)-linear. Conversely, either projector recovers
\(T=2E_+-I=I-2E_-\). When \(T=\alpha (\omega )\),
centrality of \(\omega \) is sufficient. It need not be
necessary: on \(M=\mathbb R\), let \(A=\mathbb R\times M_2(\mathbb R)\)
act by \(\alpha (a,B)=aI\), and put
\(\omega =(1,\operatorname {diag}(1,-1))\). Then \(\omega ^2=1\)
and \(T=I\), although \(\omega \) fails to commute with a matrix unit
in the second factor.
For comparison, the central-idempotent product
is an algebra statement. Central orthogonal idempotents \(e_+,e_-\)
with sum one give the mutually inverse unital algebra maps
\[a\longmapsto (e_+a,e_-a),\qquad (x,y)\longmapsto x+y.\]
Each ideal \(e_\varepsilon A\) has unit \(e_\varepsilon \) and scalar
map \(c\mapsto ce_\varepsilon \). Centrality makes each component
multiplicative and the mixed products zero; the sum and orthogonality
identities prove both inverse equations. This is stronger than the
module splitting above.
In \(k[C_2]\), writing \(g=[r]\) for \(r^2=1\), the
two central idempotents are \(e_\pm =h(1\pm g)\).
The explicit product is
\[\begin {aligned}
\Phi (a+bg)&=(a+b,a-b),\\
\Phi ^{-1}(x,y)&=h(x+y)+h(x-y)g.
\end {aligned}\]
The coefficient product
\((a+bg)(c+dg)=(ac+bd)+(ad+bc)g\) maps to
\(((a+b)(c+d),(a-b)(c-d))\). Thus \(\Phi \) is multiplicative;
its inverse equations use \(2h=1\), its unit image is \((1,1)\),
and scalars map diagonally. The images of \(e_+,e_-\) are
\((1,0),(0,1)\). An arbitrary coefficient ring need not be semisimple,
and an abstract idempotent summand need not be free.
[sengupta2010representations, Section 4.5, pp. 63-64]
discusses character projectors over a field with invertible group order;
its subsequent ideal argument assumes semisimplicity. The displayed
ring identities require only invertible two. Mathlib's
LinearMap.IsIdempotentElem.isCompl supplies the abstract
range/kernel complement. TauCeti's
TauCeti.isIdempotentElem_halfOneAdd supplies the
square-one idempotent identity, and
MonoidAlgebra.cyclicTwoEquivProd supplies the
\(C_2\) algebra product. Its generator is
\(\operatorname {Multiplicative.ofAdd}(1:\mathbb Z/2\mathbb Z)\).
None of these facts determines the ranks of a particular represented
sector; those require the coordinate bases.