Restricting an order isomorphism to a subinterval
Let \[ \phi:[u,b]\xrightarrow{\sim}[u',b'] \] be an order isomorphism, and let \(x,y\in[u,b]\). Restriction gives an order isomorphism \[ \phi_{x,y}:[x,y]\xrightarrow{\sim}[\phi(x),\phi(y)]. \] On underlying elements it is the original map \(\phi\).
Relation to the OpenAI notes
Explicit order-theoretic prerequisites.
OpenAI notes: The introduction describes Bruhat intervals as finite graded posets. The comparison induction restricts its isomorphism to smaller upper intervals and uses the equal relative ranks to normalize both vectors. sec:introduction, sec:induction.
Lean (primary): Subinterval restriction is constructed on the underlying interval elements. A simple-descent induction gives a length-one successor; rank induction shows that a strictly monotone map cannot decrease rank. Applying it to an isomorphism and its inverse proves rank preservation. Dihedral:19, CanonicalKL:297, CanonicalKL:367, CanonicalKL:415.
Role in the proof: These lemmas justify both the smaller induction instances and the common powers used when transporting interval-indexed vectors.
Proof
Step 1. Include the smaller interval in the original one
For \(z\in[x,y]\), we have \(x\le z\le y\). The ambient memberships of \(x,y\) give \[ u\le x\le z\le y\le b. \] Thus the same group element \(z\) is a member of \([u,b]\). Apply \(\phi\) to that member.
Step 2. Check the new endpoint inequalities
Monotonicity gives \[ \phi(x)\le\phi(z)\le\phi(y). \] Hence the image belongs to the target subinterval. This defines \(\phi_{x,y}\).
Step 3. Construct its inverse from the original inverse
For \(z'\in[\phi(x),\phi(y)]\), its endpoint inequalities place it in \([u',b']\). Apply \(\phi^{-1}\). Monotonicity of the inverse gives \[ x=\phi^{-1}(\phi(x)) \le\phi^{-1}(z') \le\phi^{-1}(\phi(y))=y. \] Thus its inverse image belongs to \([x,y]\).
The two composites return the same underlying element, because \(\phi^{-1}\phi\) and \(\phi\phi^{-1}\) are identities. Equality of interval members is equality of their underlying elements, so these are mutually inverse maps.
Step 4. Preserve and reflect the smaller order
The order in each subinterval is its ambient order. For \(z,w\in[x,y]\), \[ \phi_{x,y}(z)\le\phi_{x,y}(w) \quad\Longleftrightarrow\quad z\le w \] by the original order isomorphism. Therefore the restricted bijection is an order isomorphism.
Step 5. Transport indexed data through the restriction
For a function \(f\) on \([x,y]\), define \(f'(z')=f(\phi_{x,y}^{-1}(z'))\). Then \(f'(\phi_{x,y}(z))=f(z)\). This is the substitution used when an induction hypothesis compares data on the two smaller intervals.
The construction applies to any chosen \(x,y\) in the original interval. Their endpoint inequalities characterize the domain and target subintervals, including when those subintervals are empty.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.intervalIso_restrict: Dihedral.lean:19.
OAI.KLInvariance.intervalIso_restrict_val: Dihedral.lean:58.