Support orthogonality makes integral cuts balanced
Let \(R\) be a commutative domain, \(M,N\) be \(R\)-modules, and \(b:M\times M\to R\) be a bilinear pairing. Let \(\pi:M\to N\) and \(C:M\to M\) be linear maps. Suppose \(p\in R\) is nonzero and \[ \pi(Cm)=p\,\pi(m)\qquad(m\in M). \] Assume the two orthogonality conditions \[ \begin{aligned} \pi(n)=0&\ \Longrightarrow\ b(Cm,n)=0,\\ \pi(m)=0&\ \Longrightarrow\ b(m,Cn)=0 \end{aligned} \qquad(m,n\in M). \] Then \(C\) is balanced for the pairing: \[ b(Cm,n)=b(m,Cn)\qquad(m,n\in M). \]
For a graded sheaf over a domain, a perfect graded section pairing satisfies these hypotheses for each integral cut \(C_x\) defined by a nonzero common scalar. Consequently its cuts are balanced.
Proof
Step 1. Two vectors in the evaluation kernel
Linearity and the scaling hypothesis give \[ \pi(pn-Cn)=p\pi(n)-\pi(Cn)=0. \] The same calculation with \(m\) gives \(\pi(pm-Cm)=0\).
Step 2. Compare through the twice-cut value
Apply the first orthogonality condition to \(m\) and \(pn-Cn\). Expanding in the second argument yields \[ 0=b(Cm,pn-Cn)=p\,b(Cm,n)-b(Cm,Cn). \] Hence \(p\,b(Cm,n)=b(Cm,Cn)\).
Apply the second condition to \(pm-Cm\) and \(n\). Expanding in the first argument yields \[ 0=b(pm-Cm,Cn)=p\,b(m,Cn)-b(Cm,Cn). \] Thus \(p\,b(m,Cn)=b(Cm,Cn)\) as well.
Step 3. Cancel the common scalar
The two equations imply \[ p\bigl(b(Cm,n)-b(m,Cn)\bigr)=0. \] Since \(R\) is a domain and \(p\ne0\), the parenthesized scalar is zero. This is the required balance equation.
Step 4. Apply the argument to a sheaf cut
Take \(M=\Gamma(B)\), \(N=B^x\), \(\pi=\pi_x\), and \(C=C_x\). By the cut definition, \(\pi_x(C_xm)=p\,m_x\), and \(C_xm\) is supported at \(x\).
If \(n_x=0\), left support orthogonality gives \(\langle C_xm,n\rangle=0\). If \(m_x=0\), right support orthogonality gives \(\langle m,C_xn\rangle=0\). All the hypotheses of Steps 1–3 therefore hold, proving balance for the section pairing.
References
The source declarations below give the definitions and constructions used here.
Lean formalization
OAI.BalancedCuts.balanced_two: CharacterReciprocity.lean:49.
OAI.KLInvariance.MomentGraph.GradedSheaf.PerfectPairing.cut_balanced: CharacterReciprocity.lean:76.