Perfect pairings and balanced bimodules
Let \(A\) be a commutative ring and \(M\) an \(A\)-module. The linear dual \[ M^*=\operatorname{Hom}_A(M,A) \] consists of the additive maps \(\lambda:M\to A\) satisfying \(\lambda(am)=a\lambda(m)\). Its addition and scalar action are defined pointwise.
A perfect pairing is an \(A\)-linear isomorphism \[ p:M\xrightarrow{\sim}M^*,\qquad B(m,n)=p(m)(n). \] Linearity of \(p\) gives linearity of \(B\) in its first argument; each \(p(m)\) is a linear functional, giving linearity in its second argument. Perfection means that each functional \(\lambda\) has a unique representing vector \(m=p^{-1}(\lambda)\) with \(B(m,n)=\lambda(n)\) for every \(n\).
The finite free paired-module structure
The source's PairedModule packages these fields:
| Field | Mathematical content |
|---|---|
carrier |
The underlying set \(M\). |
add |
The abelian-group operations on \(M\). |
module |
The left \(A\)-module action. |
free |
Existence of an \(A\)-basis. |
finite |
Finite generation over \(A\); together with freeness, a finite basis is available. |
pairing |
The isomorphism \(p:M\to M^*\) representing the perfect pairing. |
A finite basis \(b_1,\ldots,b_r\) gives the pairing matrix \(G_{ij}=B(b_i,b_j)\). For \(m=\sum_i x_i b_i\), the functional \(p(m)\) has values \(\sum_i x_iG_{ij}\) on the \(b_j\). Thus perfection is equivalent to invertibility of \(G\) over \(A\).
The unit paired module is \(M=A\), with \(B(a,b)=ab\). For any \(A\)-linear functional \(\lambda:A\to A\),
\[
\lambda(b)=b\lambda(1).
\]
Its representing element is \(\lambda(1)\), so evaluation at \(1\) is the inverse of the pairing map. These are unitPairing and unitModule.
For the related example \(B(a,b)=cab\), this map is multiplication by \(c\) under the same dual identification. It is perfect exactly when \(c\) is a unit.
A right action commuting with the left action
Let \(\operatorname{End}_A(M)\) be the ring of \(A\)-linear endomorphisms of \(M\), with multiplication given by composition. A unital ring homomorphism \[ \rho:A\longrightarrow\operatorname{End}_A(M) \] specifies the right action \(m\cdot a=\rho(a)m\). Since \(A\) is commutative, its images commute, and \[ (m\cdot a)\cdot b=\rho(b)\rho(a)m=\rho(ab)m=m\cdot(ab). \] The \(A\)-linearity of each \(\rho(a)\) means that the two scalar actions commute: \[ c(m\cdot a)=(cm)\cdot a. \]
The balanced symmetric paired-bimodule structure
The source's PairedBimodule has four fields:
| Field | Mathematical content |
|---|---|
obj |
A finite free paired \(A\)-module \((M,p)\). |
right |
The ring homomorphism \(\rho\) defining the commuting right action. |
balanced |
\(B(m\cdot a,n)=B(m,n\cdot a)\) for every \(a,m,n\). |
symmetric |
\(B(m,n)=B(n,m)\) for every \(m,n\). |
Balance moves a right scalar from one argument to the other. Symmetry exchanges the arguments. These are the additional pairing identities in the bimodule structure.
For the unit bimodule, both actions are ordinary multiplication on \(A\), and \(B(m,n)=mn\). Its balance identity is \((ma)n=m(na)\), and symmetry is commutativity. This gives the empty-word datum from which repeated tensor induction starts.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.FrobeniusDuality.PairedModule: Frobenius.lean:226.
OAI.KLInvariance.FrobeniusDuality.unitPairing: Frobenius.lean:237.
OAI.KLInvariance.FrobeniusDuality.unitModule: Frobenius.lean:249.
OAI.KLInvariance.FrobeniusDuality.PairedBimodule: Frobenius.lean:295.
OAI.KLInvariance.FrobeniusDuality.unitBimodule: Frobenius.lean:301.
Used by
- Homogeneous functionals have homogeneous pairing representatives
- Integral adjoints for a perfect pairing
- Balanced perfect duality on global sections
- Perfect graded pairings and support orthogonality
- Vertex support equals the annihilator of an evaluation kernel
- Perfect graded pairings on reduced-word sections
- Frobenius traces and trace duality
- Right actions, balance and symmetry under tensor induction
- Extending a finite free dual and pairing by tensor product
- Tensor induction of a finite free paired module
- Recovering linear functionals through a Frobenius trace