Recovering a projective kernel from scaled image equality
Let \(A\) be a commutative ring. Suppose \(S\subseteq M\) and \(T\subseteq P\) are projective \(A\)-modules. Let \[ f:M\to N,\qquad g:P\to N,\qquad \psi:N\to N \] be linear maps, with \(\psi\) injective. Assume \[ f(S)=\psi(g(T)),\qquad \ker g=\alpha P, \] multiplication by \(\alpha\) on \(P\) is injective, and \(T\) is saturated with respect to \(\alpha\). Then \(S\cap\ker f\) is projective.
Relation to the OpenAI notes
The kernel comparison through explicit splitting maps.
OpenAI notes: After image equality, the notes use a length-one graded free resolution of the edge image and graded Schanuel comparison to obtain a projective kernel. eq:image-equality, sec:induction.
Lean (primary): The module comparison lifts each presentation map through the other and writes an explicit retraction from the sum of one source and the other kernel. The scalar-quotient presentation and actual-image equality supply the required projective kernel. The graded projective-to-free theorem is applied separately. ModuleAlgebra:246, GradedModules:538, PositiveTransfer:970.
Role in the proof: The KB follows the maps used by Lean, with the common image, its two surjections, the regular scalar kernel and the retraction equation all displayed.
Proof
Step 1. Construct the two maps onto the common image
Put \(U=f(S)\). The first map is \[ u:S\to U,\qquad s\mapsto f(s), \] which is surjective by the definition of \(U\).
Image equality gives a second well-defined map \[ v:T\to U,\qquad t\mapsto\psi(g(t)). \] Every \(z\in U=\psi(g(T))\) has a representation \(z=\psi(g(t))\) with \(t\in T\), proving its surjectivity.
Step 2. Identify the second kernel as a scalar multiple
If \(v(t)=0\), injectivity of \(\psi\) gives \(g(t)=0\). Since \(\ker g=\alpha P\), write \(t=\alpha p\), \(p\in P\).
The vector \(t\) belongs to \(T\), so saturation implies \(p\in T\). Hence \(t\in\alpha T\).
Conversely, for \(t=\alpha p\) with \(p\in T\), the scalar-kernel assumption gives \(g(t)=0\), so \(v(t)=0\). Thus \(\ker v=\alpha T\).
Step 3. Transfer projectivity to this kernel
Multiplication by \(\alpha\) defines \[ T\longrightarrow\ker v,\qquad t\longmapsto\alpha t. \] It is surjective by Step 2, and injective because the action on the ambient \(P\) is injective. It is therefore a linear isomorphism.
Since \(T\) is projective, this makes \(\ker v\) projective.
Step 4. Apply the two-presentation construction
The common-target theorem applies with \[ F=S,\qquad E=T,\qquad I=U,\qquad f=u,\qquad g=v. \] Both maps are surjective; \(S,T\) are projective by hypothesis; and \(\ker v\) is projective by Step 3.
It chooses lifts \(a:S\to T\), \(b:T\to S\) with \(va=u\), \(ub=v\), and exhibits \(\ker u\) as a retract of \(S\oplus\ker v\) by \[ x\longmapsto(x,a(x)),\qquad (s,k)\longmapsto s-ba(s)+b(k). \] The two-presentation proof checks both the kernel equation and the identity of the composite. Thus \(\ker u\) is projective.
Step 5. Identify the submodule in the statement
A vector of \(\ker u\) is a vector \(s\in S\) with \(f(s)=0\). Forgetting the subtype identifies it linearly with \(S\cap\ker f\); the inverse equips such a vector with those same two membership conditions.
Transporting projectivity along this equivalence proves the result.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.ReverseFiltration.scalarKernelEquiv: PositiveTransfer.lean:931.
OAI.KLInvariance.ReverseFiltration.projective_kernel_of_scalar_presentation: PositiveTransfer.lean:947.
OAI.KLInvariance.ReverseFiltration.imageMapKernelEquiv: PositiveTransfer.lean:958.
OAI.KLInvariance.ReverseFiltration.inf_kernel_projective_of_image_eq: PositiveTransfer.lean:970.