Theorem The adjoint image is dual to a split summand
✓ theorem proved

The adjoint image is dual to a split summand

Let \(M,N\) be modules over a commutative ring \(R\). Suppose \[ D:M\xrightarrow{\sim}M^*,\qquad i:N\to M,\quad p:M\to N,\quad pi=1_N. \] Put \(e=ip\), an idempotent endomorphism of \(M\), and let \(e^*\) be its adjoint for \(D\).

Then \[ \operatorname{im}e^*\xrightarrow{\sim}N^*, \qquad m\longmapsto\bigl(n\mapsto\langle m,i(n)\rangle\bigr) \] is an \(R\)-linear isomorphism. Its inverse sends \(\lambda\in N^*\) to \(D^{-1}(\lambda\circ p)\).

Proof

Step 1. Construct two maps between the dual and the ambient module

Define \[ j:N^*\to M,\qquad j(\lambda)=D^{-1}(\lambda\circ p), \] \[ q:M\to N^*,\qquad q(m)(n)=\langle m,i(n)\rangle. \] These maps are linear by linearity of \(D^{-1}\), precomposition, and the pairing.

For \(\lambda\in N^*\) and \(n\in N\), \[ q(j(\lambda))(n) =D(D^{-1}(\lambda\circ p))(i(n)) =\lambda(p(i(n)))=\lambda(n). \] Thus \(qj=1_{N^*}\).

Step 2. Identify the adjoint projector

For \(m,z\in M\), \[ \begin{aligned} \langle jq(m),z\rangle &=q(m)(p(z))\\ &=\langle m,i(p(z))\rangle\\ &=\langle m,e(z)\rangle. \end{aligned} \] This is the defining equation of \(e^*\). Perfection gives \(e^*=jq\). In particular, \[ (e^*)^2=jqjq=j(qj)q=jq=e^*. \]

Step 3. Identify its image

Every \(e^*m=jq(m)\) lies in \(\operatorname{im}j\). Conversely, \[ e^*(j(\lambda))=jqj(\lambda)=j(\lambda), \] so every element of \(\operatorname{im}j\) lies in \(\operatorname{im}e^*\). Hence \(\operatorname{im}e^*=\operatorname{im}j\).

The map \(q\) restricts to this image. The identity \(qj=1\) proves it is surjective. If \(m=j(\lambda)\) and \(q(m)=0\), then \(\lambda=qj(\lambda)=0\), so \(m=0\). This proves injectivity and the asserted inverse.

Step 4. Specialize to a projector's own image

For an idempotent \(f:M\to M\), take \(N=\operatorname{im}f\), let \(i\) be inclusion, and let \(p\) send \(m\) to \(f(m)\) viewed in its image. If \(n=f(z)\), then \(f(n)=f^2(z)=f(z)=n\), so \(pi=1_N\).

The result becomes \[ \operatorname{im}f^*\simeq(\operatorname{im}f)^*, \] with pairing given by the ambient form on the two image modules.

Step 5. Evaluate the coordinate-line example

For \(M=R^2\) with \(\langle(a,b),(c,d)\rangle=ad+bc\), take \(N=R\), \(i(c)=(c,0)\), and \(p(a,b)=a\). Then \(e^*(a,b)=(0,b)\), and the isomorphism sends \[ (0,b)\longmapsto(c\mapsto bc). \] The functional \(c\mapsto bc\) pulls back to \((0,b)\). Thus the dual summand is realized by the other coordinate line.

References

The local source declarations below give the constructions and proofs.

Lean formalization

OAI.AdjointSummandDuality.dualSection: BalancedPairings.lean:1025.

OAI.AdjointSummandDuality.dualProjection: BalancedPairings.lean:1028.

OAI.AdjointSummandDuality.projection_section: BalancedPairings.lean:1031.

OAI.AdjointSummandDuality.adjointSummand: BalancedPairings.lean:1049.

OAI.AdjointSummandDuality.adjointEquivDual: BalancedPairings.lean:1067.

OAI.PerfectAdjoint.adjoint_projector_eq: BalancedPairings.lean:1136.

OAI.PerfectAdjoint.rangeAdjointEquivDual: BalancedPairings.lean:1142.

OAI.PerfectAdjoint.rangeAdjointEquivDualRange: BalancedPairings.lean:1173.

Used by