Recovering terminal-kernel freeness by reverse traversal
If every lower edge image in a schedule equals its corrected upper edge image, then all terminal kernels at every vertex are free.
Proof
Step 1. State the induction on suffixes
Let \(p\) be a finite chronological edge list. Assume the embedded image equality for every suffix \(e::q\) of \(p\). We prove that \(K_x(F_q(x))\) is free for every suffix \(q\) and every vertex \(x\). This is an induction on the length of \(p\), starting from the empty list and adding conditions back.
For \(p=[]\), the only suffix is empty and \(F_{[]}(x)=\varnothing\). Its kernel imposes no equations, so it is the whole stalk \(B^x\), which is free.
Step 2. Reduce a nonempty list to its tail
Write \(p=e::r\), with \(e:x\to y\). Every suffix of \(r\) is also a suffix of \(p\), so the given equalities restrict to \(r\). The induction hypothesis gives freeness at every vertex for every suffix of \(r\).
A suffix \(q\) of \(p\) is either a suffix of \(r\), already handled, or the whole list \(p\). It remains to handle \(p\).
Step 3. Identify the only changed kernel
At a vertex \(z\ne x\), adding \(e\) does not change its outgoing conditions; its kernel is the already free tail kernel. At \(x\), put \[ S=K_x(F_r(x)),\qquad T=K_y(F_r(y)). \] Both modules are free by the tail induction hypothesis. The new lower kernel is \(S\cap\ker\rho_{x,e}\).
Step 4. Check every hypothesis of reverse projectivity
Apply Recovering a projective kernel from scaled image equality with \(f=\rho_{x,e}\), \(g=\rho_{y,e}\), and \(\psi\) multiplication by the edge factor \(f_e\).
The given equality is exactly \(f(S)=\psi(g(T))\). The upper quotient rule gives \(\ker g=\alpha_eB^y\). The label \(\alpha_e\) is nonzero and acts injectively on the free stalk \(B^y\). Multiplication by \(f_e\) is injective on the edge module by the established regularity of the actual edge factor.
For saturation, suppose \(\alpha_ev\in T\). For each remaining outgoing edge \(h\) at \(y\), \(\alpha_e\rho_{y,h}(v)=\rho_{y,h}(\alpha_ev)=0\). The incoming label \(\alpha_e\) acts injectively on each such edge module, so \(\rho_{y,h}(v)=0\). Thus \(v\in T\). All hypotheses are now checked, and the new lower kernel is projective.
Step 5. Upgrade projectivity to freeness
The new kernel is a homogeneous submodule of a finite nonnegatively graded polynomial-ring module. It is finitely generated because the polynomial ring is Noetherian. Its grading is nonnegative because it inherits the stalk grading.
The homogeneous-basis theorem From graded projectivity to a homogeneous free basis therefore applies to this projective module and provides a homogeneous free basis. This gives freeness at \(x\), completing the suffix induction.
If image equalities were obtained from endpoint equality, first use Endpoint equality forces equality at every transfer step and Hilbert detection to establish them. They are inputs to Step 4, not conclusions assumed earlier in the comparison.
References
6. Reciprocity and simultaneous induction.
Lean formalization
OAI.KLInvariance.BruhatGraph.schedule_kernels_free_of_images_eq: SimultaneousInvariance.lean:181.
OAI.KLInvariance.BruhatGraph.partial_kernel_free_of_image_eq: PositiveTransfer.lean:1052.