Finite sets in a dihedral group have a common Bruhat upper bound
Let \(H\) be a two-generator Coxeter system, with length \(\ell\) and strong Bruhat order. For every finite subset \(F\subseteq H\), there is an element \(d\in H\) such that \[ x\le d\qquad\text{for every }x\in F. \] The common upper bound need not belong to \(F\).
The proof uses a general Coxeter fact: if every simple reflection is a left descent of \(w\), then every group element lies below \(w\).
Application to finite relative lower ideals
For a finite relative coset vertex set \(K\), apply the theorem to its underlying elements of the induced dihedral group. The resulting \(d\) gives \[ K\subseteq[1,d]_H. \] The inclusion preserves and reflects the intrinsic order because both use the same Bruhat relation on the same underlying elements. This is the interval used in the coset-model transport proof.
Relation to the OpenAI notes
Finite-interval construction of the relative models.
OpenAI notes: After the relative-flabbiness lemma, the notes restrict the localized model from the full coset to its finite relative lower ideal. They extend a section by zero outside the model support inside an upward closure, then use model flabbiness. Flat localization preserves the finite-support section conditions. loc:models, loc:relative-flabbiness.
Lean (primary): The finite relative ideal is embedded in a principal dihedral interval using a common upper bound. Explicit scalar-extension and pullback isomorphisms identify the structure models. The outgoing-lift theorem is applied to edges whose targets lie below the chosen model center; canonical positive-root equality identifies the actual labels. MinimalExtensions:1295, DihedralIntervals:667, PlaneLocalization:637, PlaneLocalization:724, PlaneLocalization:761.
Role in the proof: The Blueprint follows the finite-interval construction and verifies the support, outgoing lifts, scalar maps and label equality before assembling the local-model fields.
Proof
Step 1. Prove the general all-descents assertion by word induction
Assume \(\ell(sw)<\ell(w)\) for every simple reflection \(s\). We prove that the product of every simple-generator word lies below \(w\), by induction on the word.
The empty product is \(1\), and \(1\le w\).
Suppose the shorter word has product \(z\le w\), and prepend \(s\). Let \(U_s\) be the upper simple envelope. Its monotonicity and pair-constancy give \[ sz\le U_s(sz)=U_s(z)\le U_s(w)=w. \] The last equality uses the descent hypothesis at \(w\). This proves the induction step.
Every Coxeter-group element is the product of a simple-generator word, so every element is below \(w\).
Step 2. Handle bounded length
Suppose the set of lengths \(\{\ell(x):x\in H\}\subseteq\mathbb N\) is bounded. It is nonempty and finite, so it has a maximum. Choose \(d\) attaining that maximum.
For either simple reflection \(s\), maximality gives \(\ell(sd)\le\ell(d)\). Simple multiplication changes length by exactly one, so equality cannot occur. Hence \(\ell(sd)<\ell(d)\).
Both generators are descents. Step 1 gives \(x\le d\) for every \(x\in H\), and therefore for every \(x\in F\).
Step 3. Handle unbounded length
Suppose lengths are unbounded. Let \[ N=\max\bigl(\{0\}\cup\{\ell(x):x\in F\}\bigr). \] Choose \(d\) with \(\ell(d)>N\). Then every \(x\in F\) satisfies \(\ell(x)<\ell(d)\).
The dihedral Bruhat comparison theorem states that in a two-generator system every element of strictly smaller length lies below the longer element. Applying it to \(x,d\) gives \(x\le d\).
These two cases prove the assertion, including the empty finite set.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.bruhat_top_of_all_descents: DihedralIntervals.lean:644.
OAI.KLInvariance.Dihedral.exists_common_upper: DihedralIntervals.lean:667.