Recovering linear functionals through a Frobenius trace
Let \(\eta:R\to A\) be an algebra map between commutative rings. Let \(t:A\to R\) and \(D:A\xrightarrow{\sim}\operatorname{Hom}_R(A,R)\) be trace-duality data: \[ D(a)(b)=t(ab). \] Let \(M\) be an \(A\)-module, regarded as an \(R\)-module by \(r m=\eta(r)m\). Then \[ T:\operatorname{Hom}_A(M,A)\xrightarrow{\sim} \operatorname{Hom}_R(M,R),\qquad T(\lambda)=t\circ\lambda \] is an \(R\)-linear isomorphism.
If \(p:M\xrightarrow{\sim}\operatorname{Hom}_A(M,A)\) is a perfect pairing, composition with \(T\) gives a perfect \(R\)-pairing \[ B_R(m,n)=t(p(m)(n)). \] These statements hold for every \(A\)-module \(M\).
Proof
Step 1. Restrict a functional
If \(\lambda:M\to A\) is \(A\)-linear, then \[ t(\lambda(\eta(r)m)) =t(\eta(r)\lambda(m)) =r\,t(\lambda(m)). \] Thus \(t\circ\lambda\) is \(R\)-linear. Additivity and \(R\)-scalar linearity in \(\lambda\) follow from the same properties of \(t\). This defines the source's restrictDual map \(T\).
Step 2. Form a functional on the coefficient ring
Fix \(f\in\operatorname{Hom}_R(M,R)\) and \(m\in M\). Define \[ f_m:A\to R,\qquad f_m(a)=f(am). \] This is \(R\)-linear: for \(r\in R\), \(f_m(\eta(r)a)=f(\eta(r)(am))=r f_m(a)\), and additivity follows from the module laws.
Use \(D^{-1}\) to represent it by \[ c_f(m)=D^{-1}(f_m)\in A. \] The defining equation is \[ t(c_f(m)a)=f(am)\qquad(a\in A). \] This is trace_recover_fiber. It determines \(c_f(m)\) uniquely.
Step 3. Prove that the recovered map is \(A\)-linear
For addition, \(f_{m+n}=f_m+f_n\), and linearity of \(D^{-1}\) gives \(c_f(m+n)=c_f(m)+c_f(n)\).
For a scalar \(a\in A\), test the two candidates \(c_f(am)\) and \(a c_f(m)\) against an arbitrary \(b\in A\): \[ \begin{aligned} D(c_f(am))(b)&=f(b(am))=f((ba)m),\\ D(a c_f(m))(b)&=t(a c_f(m)b)\\ &=t(c_f(m)(ba))=f((ba)m). \end{aligned} \] Their \(D\)-images agree on every \(b\), so injectivity of \(D\) gives \(c_f(am)=a c_f(m)\). Thus \(c_f:M\to A\) is \(A\)-linear. It is the source's recoverDual map applied to \(f\).
Step 4. Verify both inverse identities
Taking \(a=1\) in the representing equation gives \[ T(c_f)(m)=t(c_f(m))=f(m). \] Hence restriction after recovery is the identity.
Conversely, start with \(\lambda\in\operatorname{Hom}_A(M,A)\) and put \(f=t\circ\lambda\). For every \(a\in A\), \[ f_m(a)=t(\lambda(am)) =t(a\lambda(m)) =D(\lambda(m))(a). \] Thus \(f_m=D(\lambda(m))\), so \(c_f(m)=\lambda(m)\). Recovery after restriction is also the identity.
Since \(T\) is \(R\)-linear and has this inverse, it is the asserted \(R\)-linear isomorphism.
Step 5. Restrict a perfect pairing
The map \(p\), viewed over \(R\), is still an isomorphism. Compose it with \(T\): \[ M\xrightarrow{p}\operatorname{Hom}_A(M,A) \xrightarrow{T}\operatorname{Hom}_R(M,R). \] Its value at \(m\), evaluated on \(n\), is \(t(p(m)(n))\). Its inverse sends \(f\) first to \(c_f\), then to \(p^{-1}(c_f)\). This gives the unique representing vector for every \(R\)-functional and proves perfection.
For \(M=A\) with multiplication pairing, this construction recovers \(c_f=D^{-1}(f)\), the trace representative from the reflection example.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.FrobeniusDuality.restrictDual: Frobenius.lean:112.
OAI.KLInvariance.FrobeniusDuality.fiber: Frobenius.lean:117.
OAI.KLInvariance.FrobeniusDuality.trace_recover_fiber: Frobenius.lean:124.
OAI.KLInvariance.FrobeniusDuality.recover_fiber_smul: Frobenius.lean:129.
OAI.KLInvariance.FrobeniusDuality.recoverDual: Frobenius.lean:139.
OAI.KLInvariance.FrobeniusDuality.restrict_recover: Frobenius.lean:152.
OAI.KLInvariance.FrobeniusDuality.recover_restrict: Frobenius.lean:158.
OAI.KLInvariance.FrobeniusDuality.restrictionEquiv: Frobenius.lean:168.
OAI.KLInvariance.FrobeniusDuality.restrictionEquiv_apply: Frobenius.lean:175.
OAI.KLInvariance.FrobeniusDuality.restrictPairing: Frobenius.lean:180.
OAI.KLInvariance.FrobeniusDuality.restrictPairing_apply: Frobenius.lean:183.