The transported comparison for actual characters
Let \(W,W'\) be finite-generator Coxeter systems, \(a\le_W b\), and \(\phi:[a,b]_W\simeq[a',b']_{W'}\) an order isomorphism. Fix a finite real basis \(\Sigma\) of the source extended realization, put \(A=\mathbb R[X_i:i\in\Sigma]\), \(N=|\Sigma|\), and restrict the actual boundary sheaf on \([1,b]_W\) to \([a,b]_W\).
Write \(P=\operatorname{chronologicalSchedule}(\phi)\) and \(Q=\operatorname{transportedSchedule}(\phi)=P^{\mathrm{rev}}\). For every suffix \(e::q\preceq_{\mathrm{suf}}P\), assume that the upper selected kernel \(K_{r(e)}(F_q(r(e)))\) is free over \(A\).
Let \(C_{x,b}(q)\in\mathbb Z[q]\) be the actual stalk character of the source boundary sheaf. For nonzero real \(t\), its scaled actual-stalk character is
\[
c_x(t)=t^{\ell_W(x)-\ell_W(b)}C_{x,b}(t^2).
\]
For \(0<t<1\), put \(T=t^{-1}-t\). Then
\[
c(t)\le\mathcal D_Q(T)c(t^{-1})
\tag{1}
\]
coordinatewise on the source interval. This is the exact dispatch-valued conclusion of actual_comparison_of_upper_freeness.
Relation to the OpenAI notes
Different proof route.
OpenAI notes: The simultaneous induction compares normalized KL vectors, obtains P-invariance from the strict half-degree bound, and then recovers bottom terminal-kernel freeness from endpoint equality. sec:induction, eq:comparison, eq:reciprocal-vector.
Lean (primary): The simultaneous induction first proves normalized R-invariance together with suffix-kernel freeness in both transported directions. It cancels known bottom R-row entries against one actual source-character vector. Canonical KL equality is proved afterwards in a separate induction using the now common R-matrix and the canonical degree bound. Invariance:136, Invariance:190, Invariance:427.
Role in the proof: Proper upper intervals supply kernel freeness, and proper initial intervals supply known R-entries. The new R-equality gives exact endpoints and then current-rank suffix freeness. The canonical degree bound enters the subsequent P-induction.
Transfer comparison before the target path-matrix identification.
OpenAI notes: The notes identify the state endpoints with normalized KL vectors and present the accumulated bound using the target R-matrix. eq:endpoints, eq:comparison.
Lean (primary): The declared actual_comparison_of_upper_freeness theorem concludes an inequality with the dispatch operator applied to the actual source-character vector at the inverse parameter. Its proof substitutes the full- and empty-kernel endpoint identities into the schedule bound. The path-matrix identification is a separate input to later comparison results. TerminalKernels:1324, SimultaneousInvariance:255, SimultaneousInvariance:264.
Role in the proof: The entry states that exact dispatch conclusion, derives both endpoint substitutions, and links the subsequent geometric path identification. The actual source-character polynomial remains explicit.
Proof
Step 1. The conditional complete-schedule bound
The sheaf, interval isomorphism, coefficient basis and free upper kernels satisfy the hypotheses of the comparison under upper-kernel freeness. It gives \[ D_{[]}(t)\le\mathcal D_Q(T)D_P(t). \tag{2} \] We identify its two state vectors using the normalized endpoint identities.
Step 2. The empty-list state
At each \(x\), the empty-list selected kernel is the entire stalk, with Hilbert value
\[
H_{\mathcal B^x}(t^2)=\frac{C_{x,b}(t^2)}{(1-t^2)^N}.
\]
The stalk is the same as the corresponding stalk in \([1,b]_W\), since the sheaf was restricted along the upper-interval inclusion.
Multiplication by the state scale gives
\[
\begin{aligned}
D_{[]}(t)_x
&=(1-t^2)^N t^{\ell_W(x)-\ell_W(b)}
\frac{C_{x,b}(t^2)}{(1-t^2)^N}\\
&=t^{\ell_W(x)-\ell_W(b)}C_{x,b}(t^2)
=c_x(t).
\end{aligned}
\]
The cancellation is valid because \(1-t^2>0\). This is schedule_final_scaledStalk.
Step 3. The full-list kernel
The chronological schedule \(P\) contains every edge of the interval. Thus \(F_P(x)=\operatorname{Out}(x)\) at every vertex. Restriction to \([a,b]\) retains every upward edge from \(x\), so its full outgoing kernel equals the corresponding full kernel in \([1,b]\). In the actual boundary sheaf this is the costalk used by the normalized endpoint identities.
Put \(d_x=\ell_W(b)-\ell_W(x)\). The full-kernel formula established there is \[ H_{K_x(\operatorname{Out}(x))}(t^2) =\frac{(t^2)^{d_x} C_{x,b}(t^{-2})}{(1-t^2)^N}. \] Its derivation uses the actual homogeneous section duality and the count of incoming edges, as detailed in those endpoint proofs.
Step 4. Scaling at the initial endpoint
Substitute this full-kernel value:
\[
\begin{aligned}
D_P(t)_x
&=(1-t^2)^N t^{-d_x}
\frac{(t^2)^{d_x} C_{x,b}(t^{-2})}{(1-t^2)^N}\\
&=t^{-d_x+2d_x}C_{x,b}(t^{-2})\\
&=t^{d_x}C_{x,b}(t^{-2})\\
&=(t^{-1})^{-d_x}C_{x,b}((t^{-1})^2)
=c_x(t^{-1}).
\end{aligned}
\]
All inverse powers are defined because \(t>0\). This calculation is transported_initial_scaledStalk for the arbitrary interval isomorphism \(\phi\).
Step 5. Substitution of the endpoint vectors
Steps 2 and 4 hold at every vertex, hence give equalities of vectors \[ D_{[]}(t)=c(t),\qquad D_P(t)=c(t^{-1}). \] Replace the two states in (2) by these vectors to obtain (1).
The geometric path identity identifies the ordered product \(\mathcal D_Q(T)\) with the target normalized R-matrix transported through \(\phi\). That subsequent identification converts (1) to the matrix inequality used in the R-induction. The vector in (1) is already defined entirely from the actual source sheaf.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.BruhatGraph.schedule_final_scaledStalk: TerminalKernels.lean:1324.
OAI.KLInvariance.BruhatGraph.transported_initial_scaledStalk: SimultaneousInvariance.lean:255.
OAI.KLInvariance.BruhatGraph.actual_comparison_of_upper_freeness: SimultaneousInvariance.lean:264.