Alphabeta Math
Pipeline-generated
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

✓ 4 results · all verified · 4 also independently AI-judged
Every result on this page is machine-checked by a proof checker and read in full and owner-audited; the judge is an additional, independent cross-model AI review of the proofs; all 4 also cleared it.

Parabolic Subgroups and Double Coset Geometry

1 · Prerequisites

2 · Summary

Coset factorization is the algebraic form of moving to a face of a chamber. Unique minimum representatives, intersections and double cosets need separate arguments; an arbitrary subgroup is not automatically parabolic.

This page fixes the parabolic calculus of a Coxeter system: standard parabolic subgroups and their one- and two-sided transversals, the parabolic root subsystem, and the unique minimum of a parabolic double coset with its additive normal form. It is the algebraic counterpart of moving between faces of a chamber on the geometric pages, and it supplies the quotients WI, IW and IWJ used by the later Bruhat, Davis-complex and growth pages.

Definition and conventions

Standard parabolic subgroups, descent-free one- and two-sided representatives, parabolic and reflection subgroups introduces WI=⟨s:s∈I⟩, the descent sets DL,DR, the quotients WI={w:ℓ(ws)>ℓ(w) (s∈I)} and IW={w:ℓ(sw)>ℓ(w) (s∈I)}, their two-sided intersection IWJ, and the notions of a parabolic subgroup (a conjugate of a standard one) and of a reflection subgroup. The left/right conventions are fixed once and used consistently; no converse from reflection subgroups to parabolics is asserted, and the companion page computes the two test cases that separate the three classes.

Main results

Intersections of standard parabolics, the parabolic root subsystem, and global minimality of coset representatives proves WI∩WJ=WI∩J, identifies the parabolic root subsystem ΦI as the roots lying in the span VI of the simple roots indexed by I, so ΦI=Φ∩VI, and shows that the transversal elements of the one-sided quotients are global minima of their cosets, not merely locally descent-free words. The converse containment Φ∩VI⊆ΦI is proved by induction on the length of the reflection with a given root, using the root sign criterion and the root-length criterion twice.

Descent reduction, minimum-length elements, and the additive factorization in a double coset supplies the minimality input the rest of the page needs: descent reduction shows that every double coset contains a descent-free element, and the middle-block argument on a reducing word shows that for d∈IWJ the element d is the unique minimum of WIdWJ, with an additive factorization of every element of the coset. The parabolic intersection W_I cap dW_Jd inverse for d in ^IW^J then proves WI∩dWJd−1=WK for K={s∈I:d−1sd∈J} by strong exchange at the first letter of a reduced expression, and records the warning that a conjugated reflection d−1sd lying in WJ, with s∈I and d∈IWJ, is necessarily a simple root reflection, not an arbitrary positive combination of simple roots.

Unique minimal double coset representatives and the additive normal form u-d-v assembles these pieces: each double coset has a unique descent-free element, its minimum, and every element of the coset has a unique representation x=udv with u∈WIK, v∈WJ and length additivity ℓ(x)=ℓ(u)+ℓ(d)+ℓ(v). The transversal restriction on the first factor is recorded explicitly, together with the computation showing that an arbitrary u∈WI destroys uniqueness.

Prerequisites and reading

Required earlier pages: coxeter-presentations-exchange-and-reduced-word-theorems for the presented group, its length function and the exchange, deletion and parabolic-minimal-representative theorems, and canonical-roots-signs-and-faithful-reflections for the root sign criterion, the root-length criterion and the root-reflection dictionary. The companion parabolic-subgroups-and-double-coset-geometry-examples tests the constructions in S4 and in the infinite dihedral group.

3 · Logical flowchart

4 · Definitions, theorems and proofs

DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-6.1-sol)audited 2026-10-08Open item page →

Standard parabolic subgroups, descent-free one- and two-sided representatives, parabolic and reflection subgroups

Definition

Let (S,m) be a finite Coxeter matrix, let W be the presented group with length function ℓ (Coxeter matrices, the presented Coxeter group, reduced words, length, and standard parabolic subgroups), and let T={wsw−1:w∈W, s∈S} be the set of reflections of W (The canonical reflection homomorphism, roots, reflections, and the positive cone). Throughout, I and J denote subsets of S.

(1) Standard parabolic subgroups. For I⊆S put

WI:=⟨s:s∈I⟩≤W,

the subgroup generated by I (The subgroup ⟨S⟩ generated by a subset, the cyclic subgroup ⟨g⟩, and cyclic groups); WI is the standard parabolic subgroup of (W,S) of type I. Thus W∅={1} and WS=W. By Support, intrinsic parabolic presentations, minimal coset representatives and length additivity, with the type-A identification (1),(2): for every w∈W one has w∈WI if and only if S(w)⊆I, where S(w) is the (well-defined) set of letters of any reduced expression of w; (WI,I) is a Coxeter system; its intrinsic length function agrees with ℓ∣WI; and WI∩S=I.

(2) Descents and the one-sided quotients. For w∈W put

DL(w):={s∈S:ℓ(sw)<ℓ(w)},DR(w):={s∈S:ℓ(ws)<ℓ(w)};

each of ℓ(sw)−ℓ(w) and ℓ(ws)−ℓ(w) lies in {±1} (Length parity, exchange, two-letter deletion, and faithfulness of the signed reflection action (1)). Define

WI:={w∈W:ℓ(ws)>ℓ(w) for all s∈I}={w∈W:DR(w)∩I=∅},

IW:={w∈W:ℓ(sw)>ℓ(w) for all s∈I}={w∈W:DL(w)∩I=∅}.

Conventions are fixed here: in IW the subset I is tested on the left, in WI it is tested on the right; ℓ(sw) means left multiplication by s and ℓ(ws) right multiplication. The interpretation of these sets as sets of coset representatives is not part of the definition; it is quoted from Support, intrinsic parabolic presentations, minimal coset representatives and length additivity, with the type-A identification (3) and re-examined in Intersections of standard parabolics, the parabolic root subsystem, and global minimality of coset representatives ↗ (3): every w∈W has a unique factorization

w=u d(u∈WI, d∈IW),ℓ(w)=ℓ(u)+ℓ(d),

where d is the unique element of minimal length in the right coset WIw, and it satisfies ℓ(ud)=ℓ(u)+ℓ(d) for all u∈WI; symmetrically every w∈W has a unique factorization

w=d v(d∈WI, v∈WI),ℓ(w)=ℓ(d)+ℓ(v),

where d is the unique element of minimal length in the left coset wWI, and it satisfies ℓ(dv)=ℓ(d)+ℓ(v) for all v∈WI. Thus IW consists exactly of the elements of minimal length in the right cosets WIw={uw:u∈WI} and WI consists exactly of the elements of minimal length in the sets wWI={wv:v∈WI}. Inversion gives WI=(IW)−1, because ℓ(x)=ℓ(x−1) and (sw)−1=w−1s.

(3) The two-sided quotients. Define

IWJ:=IW∩WJ={w∈W:ℓ(sw)>ℓ(w) for all s∈I  and  ℓ(ws)>ℓ(w) for all s∈J};

thus ∅WJ=WJ, IW∅=IW and ∅W∅=W, and the members of IWJ are exactly the elements of W with no left descent in I and no right descent in J. Whether IWJ is a set of representatives of the double cosets WIwWJ is again a theorem, not a definition; it is proved in Unique minimal double coset representatives and the additive normal form u-d-v ↗.

(4) Parabolic subgroups and reflection subgroups. A subgroup H≤W is a parabolic subgroup of (W,S) if

H=wWIw−1for some w∈W, I⊆S;

a parabolic subgroup with w=1 is standard. A subgroup H≤W is a reflection subgroup if H=⟨H∩T⟩, i.e. if H is generated by the reflections it contains. Every parabolic subgroup is a reflection subgroup, since wWIw−1=⟨wsw−1:s∈I⟩ and wsw−1∈T for all s∈S, w∈W. No converse is asserted: it is not claimed here that a subgroup generated by reflections is parabolic, and the companion examples page computes test cases for the two distinctions created by this definition --- standard parabolic versus parabolic, and reflection subgroup versus parabolic subgroup.

Remarks

  • Well-definedness. WI is the subgroup generated by a subset of the group W and exists by The subgroup ⟨S⟩ generated by a subset, the cyclic subgroup ⟨g⟩, and cyclic groups; W∅={1} is the trivial subgroup and WS=W. The sets WI, IW and IWJ are subsets of W specified by a predicate and are therefore well-defined sets. The two displayed factorizations of (2) and the description of WI, IW as sets of minimal representatives are quoted from Support, intrinsic parabolic presentations, minimal coset representatives and length additivity, with the type-A identification (1)-(3), which proves them from the deletion and exchange properties before they are used here; the inversion identity WI=(IW)−1 uses ℓ(x)=ℓ(x−1), (sw)−1=w−1s (Group and abelian group) and the equivalence ℓ(sw)<ℓ(w)⇔ℓ(w−1s)<ℓ(w−1). The only genuinely two-sided assertion is deferred: this item introduces IWJ as a set of descent-free elements and asserts nothing about double cosets; the property that every double coset WIwWJ contains exactly one element of IWJ, and that this element is its minimum, is established in Unique minimal double coset representatives and the additive normal form u-d-v ↗, the later proof that justifies the definition for its applications.
  • Left and right conventions are not symmetric by accident. In IW the descent is tested by left multiplication, in WI by right multiplication; the right coset WIw is represented by IW and the left coset wWI by WI. Reversing either convention produces false statements already in S3 with I={s1} (for example s2s1 has no factorization ud with d∈WI), which is why they are recorded here once and used consistently.
  • No converse and no Choice. The inclusion "parabolic subgroups are reflection subgroups" in (4) is immediate from the definitions, since a conjugate wsw−1 of a simple reflection is a reflection; nothing is asserted in the other direction. No use of the Axiom of Choice is made: all constructions are subgroups and subsets of the fixed group W, and the minima in (2) are minima of nonempty subsets of the natural numbers.
LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-08Open item page →

Descent reduction, minimum-length elements, and the additive factorization in a double coset

Statement

Let (S,m), W, ℓ be as in Standard parabolic subgroups, descent-free one- and two-sided representatives, parabolic and reflection subgroups, let I,J⊆S, and let WI, WJ, IW, IWJ be as defined there; for d∈W write

Ω(d):=WIdWJ={udv:u∈WI, v∈WJ}

for the double coset of d.

(1) Descent reduction. Fix a total order on the finite set S. Every set Ω(d) contains an element of IWJ. More precisely, starting from x:=d and repeatedly replacing the current element x by sx, where s is the least element of I with ℓ(sx)<ℓ(x), if such an s exists, and otherwise by xs, where s is the least element of J with ℓ(xs)<ℓ(x), if such an s exists, the process stops after at most ℓ(d) replacements at an element of Ω(d)∩IWJ.

(2) Elements of IWJ have minimum length. Let d∈IWJ. Then

ℓ(d)≤ℓ(x)for every x∈Ω(d),andℓ(x)=ℓ(d)  ⟺  x=d.

Consequently every double coset WIwWJ contains exactly one element of IWJ, and it is the unique element of minimum length of that double coset; conversely, every element of minimum length in its double coset WIxWJ lies in IWJ.

(3) Additive factorization. Let d∈IWJ and x∈Ω(d). Then there exist u∈WI and v∈WJ with

x=udv,ℓ(x)=ℓ(u)+ℓ(d)+ℓ(v).

Facts & Assumptions

Given: a finite Coxeter matrix (S,m) with presented group W and length ℓ, subsets I,J⊆S, an element d∈W, and a fixed total order on S.

[F1]

For all w∈W and s∈S one has ℓ(sw)=ℓ(w)±1 and ℓ(ws)=ℓ(w)±1 (Length parity, exchange, two-letter deletion, and faithfulness of the signed reflection action).

[F2]

ℓ(w)=min⁡{k∈N:there are s1,…,sk∈S with w=s1⋯sk}; hence a word of length k representing v gives ℓ(v)≤k, and concatenating words gives ℓ(uv)≤ℓ(u)+ℓ(v) (Coxeter matrices, the presented Coxeter group, reduced words, length, and standard parabolic subgroups).

[F3]

WI=⟨s:s∈I⟩, IW={w:ℓ(sw)>ℓ(w) for all s∈I}, WJ={w:ℓ(ws)>ℓ(w) for all s∈J} and IWJ=IW∩WJ; for d∈W, Ω(d)=WIdWJ is the set of all products udv with u∈WI, v∈WJ (Standard parabolic subgroups, descent-free one- and two-sided representatives, parabolic and reflection subgroups).

[F4]

Every nonempty subset of N has a least element (The well-ordering principle, The natural numbers N (von Neumann)).

[F5]

WJ={w∈W:S(w)⊆J}, where S(w) is the set of letters of any reduced expression of w; in particular every element of WJ has a reduced expression all of whose letters lie in J (Support, intrinsic parabolic presentations, minimal coset representatives and length additivity, with the type-A identification).

[F6]

If w=s1⋯sk is any word in S and i<j are positions whose deletion does not change the value, in the sense that s1⋯si^⋯sj^⋯sk=w, then the deletion is an equality in the group W; conversely a word is reduced if and only if no two-letter deletion preserves its value (Length parity, exchange, two-letter deletion, and faithfulness of the signed reflection action).

[F7]

Multiplication in W is associative, and equalities may be multiplied on the left or on the right by group elements and cancelled (Group and abelian group).

Proof

technique · direct; a deletion process with a case analysis, then minima of double cosets
1.1F1F2F3F7

Descent reduction. A total order on S exists because S is finite (transport the order from any finite enumeration). Use the fixed order in every least-generator selection. Suppose x∈Ω(d) and ℓ(sx)<ℓ(x) for some s∈I; then s∈WI and sx=s⋅x⋅1∈WIxWJ=WIdWJ=Ω(d), because WIxWJ=WI(WIdWJ)WJ=WIdWJ by [F3] and [F7]. Likewise, if ℓ(xs)<ℓ(x) for some s∈J, then xs∈Ω(d) by [F3] and [F7]. By [F1] each replacement lowers ℓ by exactly 1, so a process that starts at x=d, replaces x by sx with the least s∈I of ℓ(sx)<ℓ(x) if one exists and otherwise by xs with the least s∈J of ℓ(xs)<ℓ(x) if one exists, and stops when neither exists, performs at most ℓ(d) replacements and terminates, because the values of ℓ lie in N by [F2]. At termination x∈Ω(d) and there is no s∈I with ℓ(sx)<ℓ(x) and no s∈J with ℓ(xs)<ℓ(x); since the differences lie in {±1} by [F1], this says ℓ(sx)>ℓ(x) for all s∈I and ℓ(xs)>ℓ(x) for all s∈J, that is, x∈IWJ by [F3].

1.2F2F3F5F6F7

The middle block of a reducing word is never deleted. Let m be an element of minimum length in Ω:=WIwWJ and let x∈Ω, say x=ama′ with a∈WI, a′∈WJ by [F3]. Choose reduced words A=(a1,…,ar) for a, M=(m1,…,mq) for m and A′=(a1′,…,ar′′) for a′; by [F5] the letters of A lie in I and those of A′ in J. While the current word, which begins as A followed by M followed by A′, is not reduced, choose the lexicographically least pair of positions i<j whose deletion preserves the value, which exists by [F6], and delete the two positions; write the current word as AiMAi′, where Ai and Ai′ are the surviving subwords of A and A′ and, by induction over the rounds, the full block M survives. Such a deletion pair never removes a letter of M: if both deleted letters lay in M, then the surviving word AiM1Ai′ has the value x of AiMAi′, so by [F6] and [F7] the word M1 has the same value m as M, while M1 has length q−2, contradicting ℓ(m)=q by [F2]; if exactly one deleted letter lay in M and the other in Ai, then the surviving word is Ai+1M1Ai′ with Ai+1 equal to Ai with one letter deleted and M1 equal to M with one letter deleted, so by [F6] and [F7] the value of M1 equals val⁡(Ai+1)−1 x val⁡(Ai′)−1, and this lies in WIxWJ=Ω because val⁡(Ai+1)∈WI and val⁡(Ai′)∈WJ by [F3]; but M1 has length q−1, so by [F2] the value of M1 has length ≤q−1<q=ℓ(m), contradicting the minimality of m in Ω; the case of one deleted letter in M and one in Ai′ is the same with the roles of Ai and Ai′ exchanged. A deletion pair with both letters outside M is not excluded; it simply shortens Ai or Ai′. Each deletion lowers the number of letters by 2, so the process terminates at a reduced word of x of the form A0MA0′, where A0 and A0′ arise from A and A′ by deletions and M is untouched; the words A0 and A0′ are themselves reduced, since a two-letter deletion inside A0 preserving the value of A0 would, by [F6] and [F7], preserve the value x of the reduced word A0MA0′. Writing a0:=val⁡(A0)∈WI and a0′:=val⁡(A0′)∈WJ, the reducedness of A0MA0′ gives x=a0ma0′ and ℓ(x)=ℓ(a0)+ℓ(m)+ℓ(a0′).

2.1F1F3F4F7step 1.1

A minimum has no descents. Let w∈W and let Ω:=WIwWJ; this set is nonempty, so the set {ℓ(x):x∈Ω}⊆N has a least element by [F4], and we choose m∈Ω with ℓ(m) that least element. If s∈I and ℓ(sm)<ℓ(m), then sm∈Ω by [F3] and [F7], contradicting minimality; hence ℓ(sm)>ℓ(m) for all s∈I, because ℓ(sm)=ℓ(m)±1 by [F1]. Symmetrically ℓ(ms)>ℓ(m) for all s∈J. Therefore m∈IWJ by [F3]: by step 1.1 every double coset WIwWJ contains an element of IWJ, and taking m of minimum length shows each double coset also has a minimum, which lies in IWJ.

3.1F2F3F5F7step 1.2step 2.1

Every element of IWJ is a minimum of its double coset. Let d∈IWJ and let m be a minimum-length element of Ω(d), which exists by step 2.1. Applying step 1.2 with w:=d, x:=d gives d=a0ma0′ with a0∈WI, a0′∈WJ and ℓ(d)=ℓ(a0)+ℓ(m)+ℓ(a0′). If a0≠1, then a0 has a reduced expression whose first letter s lies in I by [F5], so ℓ(sa0)=ℓ(a0)−1, and by [F2] and [F7] ℓ(sd)=ℓ(sa0ma0′)≤ℓ(sa0)+ℓ(m)+ℓ(a0′)=ℓ(d)−1<ℓ(d), contradicting ℓ(sd)>ℓ(d) for s∈I, which holds because d∈IW by [F3]. Hence a0=1, and symmetrically a0′=1, so d=m: the element d of IWJ is a minimum-length element of Ω(d).

4.1F1step 1.2step 2.1step 3.1∎

Minimality, the equality case, and the additive factorization. Let d∈IWJ and x∈Ω(d). By step 3.1 the element d is a minimum-length element of Ω(d), so step 1.2 applied with m:=d exhibits x=udv with u∈WI, v∈WJ and ℓ(x)=ℓ(u)+ℓ(d)+ℓ(v), which is the additive factorization (3); in particular ℓ(x)≥ℓ(d), with equality if and only if ℓ(u)=ℓ(v)=0, that is, u=v=1 and x=d. If d′∈IWJ∩Ω(d) is a second element, then applying the factorization with x:=d′ gives ℓ(d′)=ℓ(u)+ℓ(d)+ℓ(v) and, since d′ is also a minimum of Ω(d) by step 3.1, ℓ(d′)=ℓ(d), so u=v=1 and d′=d: each double coset contains at most one element of IWJ, and by step 1.1 it contains one, namely its unique element of minimum length. Finally, if x∈WIwWJ has minimum length in WIwWJ, then x∈IWJ by step 2.1. This proves (2) and completes the proof.

Remarks

  • The deletion step is Tits deletion, not a cancellation of equal letters: the theorem of Length parity, exchange, two-letter deletion, and faithfulness of the signed reflection action provides positions i<j whose deletion preserves the value of a non-reduced word, and the two deleted letters can be distinct simple reflections. The argument uses only the positions.
  • The argument is choice-free: each deletion is the lexicographically least admissible pair of a finite nonempty set of positions, and the minima of double cosets are minima of nonempty subsets of N.
TheoremStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-08Open item page →

Intersections of standard parabolics, the parabolic root subsystem, and global minimality of coset representatives

Statement

Let (S,m), W, ℓ and I,J⊆S be as in Standard parabolic subgroups, descent-free one- and two-sided representatives, parabolic and reflection subgroups, so that WI, WI, IW and the descents DL,DR have the meaning fixed there. Let V=RS with its Coxeter form B and positive cone V+={∑s∈Sλses:λs≥0} (The real Coxeter form, its radical, reflections, and form-preserving maps), let ρ:W→GL(V) be the canonical reflection homomorphism with root system Φ=Φ+⊔Φ− (The canonical reflection homomorphism, roots, reflections, and the positive cone, Root sign coherence and the action of simple reflections on positive roots (2)), and for I⊆S put

VI:=span⁡{es:s∈I},ΦI:={ρ(w)es:w∈WI, s∈I}

(Linear combination of a finite list, and the span span⁡(S) as the smallest linear subspace containing S); here V∅={0} and Φ∅=∅.

(1) Intersections. For all I,J⊆S,

WI∩WJ=WI∩J.

(2) Parabolic root subsystems. For every I⊆S one has ρ(w)VI=VI for all w∈WI; consequently

ΦI=Φ∩VI={α∈Φ:α is a real linear combination of the simple roots es, s∈I}.

Moreover ΦI=ΦI+⊔ΦI− with ΦI±:=ΦI∩Φ±=Φ±∩VI, and the reflections lying in WI are exactly the elements tα∈T with α∈ΦI+ in the root-reflection dictionary of The inversion formula ∣N(w)∣=ℓ(w), the root-reflection dictionary and strong exchange (1).

(3) Coset minima are global minima. Let I⊆S, w∈W, and let d∈WI be the unique element of WI∩wWI, i.e. the unique d∈WI with w∈dWI (Standard parabolic subgroups, descent-free one- and two-sided representatives, parabolic and reflection subgroups (2)). Then

ℓ(d)≤ℓ(x)for every x∈wWI,andℓ(d)=ℓ(x)  ⟺  x=d;

equivalently

WI={d∈W:ℓ(d)≤ℓ(dv) for all v∈WI},IW={d∈W:ℓ(d)≤ℓ(vd) for all v∈WI}.

Thus each set wWI contains exactly one element of WI, namely its unique element of minimum length, and symmetrically each right coset WIw contains exactly one element of IW, namely its unique element of minimum length.

Facts & Assumptions

Given: a finite Coxeter matrix (S,m) with presented group W and length ℓ, subsets I,J⊆S, the space V=RS with its Coxeter form B, and the canonical reflection homomorphism ρ with root system Φ and reflection set T.

[F1]

For J⊆S: WJ=⟨J⟩={w∈W:S(w)⊆J}, (WJ,J) is a Coxeter system with intrinsic length ℓ∣WJ, and WJ∩S=J; every right coset WJa has a unique element d of minimal length, characterized by ℓ(sd)>ℓ(d) for all s∈J, satisfying ℓ(ud)=ℓ(u)+ℓ(d) for all u∈WJ; by inversion every left coset aWJ has a unique minimal element d, characterized by ℓ(ds)>ℓ(d) for all s∈J and satisfying ℓ(du)=ℓ(d)+ℓ(u) for all u∈WJ (Support, intrinsic parabolic presentations, minimal coset representatives and length additivity, with the type-A identification).

[F2]

WI=⟨s:s∈I⟩, WI={w:ℓ(ws)>ℓ(w) for all s∈I} and IW={w:ℓ(sw)>ℓ(w) for all s∈I}, and WI=(IW)−1 (Standard parabolic subgroups, descent-free one- and two-sided representatives, parabolic and reflection subgroups).

[F3]

B(es,es)=1 and B(es,et)=−cos⁡(π/m(s,t)) for s≠t with m(s,t)<∞, B(es,et)=−1 when m(s,t)=∞; for a with B(a,a)≠0 the reflection with normal a is ra(v)=v−2B(v,a)B(a,a)a (The real Coxeter form, its radical, reflections, and form-preserving maps).

[F4]

ρ(s)=res for every s∈S, so ρ(s)es=−es and ρ(s)et=et−2B(et,es)es for t≠s (The canonical reflection homomorphism, roots, reflections, and the positive cone, The real Coxeter form, its radical, reflections, and form-preserving maps); ρ preserves B, every root α satisfies B(α,α)=1, and ρ(wsw−1)=rρ(w)es for all w∈W, s∈S (Descent of the reflection representation, unit root norms, and conjugation of reflections).

[F5]

Every root lies in V+∖{0} or in −V+∖{0}, but not in both; hence with Φ+=Φ∩V+ and Φ−=Φ∩(−V+) one has Φ=Φ+⊔Φ−, and every α∈Φ+ has all its coefficients ≥0 while every α∈Φ− has all its coefficients ≤0 (Root sign coherence and the action of simple reflections on positive roots).

[F6]

For all w∈W and s∈S: ℓ(ws)>ℓ(w)  ⟺  ρ(w)es∈Φ+, and ℓ(ws)<ℓ(w)  ⟺  ρ(w)es∈Φ− (The root-length criterion and faithfulness of the canonical reflection representation).

[F7]

The root-reflection dictionary: tα:=wsw−1 for any w∈W, s∈S with α=ρ(w)es is well defined, satisfies ρ(tα)=rα and tρ(w)α=wtαw−1, and restricts to a bijection Φ+→T, α↦tα, with tα=tβ if and only if α=±β. Strong exchange: if w=s1⋯sn is reduced and ℓ(tw)<ℓ(w) for a reflection t, then there is a unique i with tw=s1⋯si^⋯sn and t=s1⋯si−1sisi−1⋯s1, and the positive root α with t=tα is ρ(s1⋯si−1)esi (The inversion formula ∣N(w)∣=ℓ(w), the root-reflection dictionary and strong exchange).

[F8]

For all w∈W and s∈S one has ℓ(ws)=ℓ(w)±1 (Length parity, exchange, two-letter deletion, and faithfulness of the signed reflection action).

[F9]

For w=s1⋯sk one has w−1=sk⋯s1, because every generator satisfies s2=1; hence ℓ(w−1)=ℓ(w) and (ab)−1=b−1a−1 for all a,b∈W (Coxeter matrices, the presented Coxeter group, reduced words, length, and standard parabolic subgroups, Group and abelian group).

[F10]

WI is the smallest subgroup containing I (The subgroup ⟨S⟩ generated by a subset, the cyclic subgroup ⟨g⟩, and cyclic groups). The set of finite products of elements of I and their inverses is a subgroup containing I (concatenate words and reverse and invert a word), and is contained in WI; thus it equals WI.

[F11]

VI=span⁡{es:s∈I} is the set of all real linear combinations ∑s∈Iλses; in particular V∅={0} and the coefficients of a vector of VI in the basis (es)s∈S vanish outside I (span⁡(S) is exactly the set of linear combinations of finite lists of elements of S, and span⁡(∅)={0V}); the coordinate assertion follows by evaluating the finite combinations at each s∉I.

Proof

technique · direct, with a strong induction on the length of a reflection for the converse containment of (2)
1.1F1F2

Intersections of standard parabolics. By [F1], w∈WI if and only if S(w)⊆I, and w∈WJ if and only if S(w)⊆J; hence w∈WI∩WJ if and only if S(w)⊆I∩J, if and only if w∈WI∩J. This proves (1) for all subsets I,J, including I∩J=∅, where W∅={1} by [F1] and [F2].

1.2F3F4F5F10F11

Invariance of VI and one containment. Let s∈I. By [F4], ρ(s)es=−es and ρ(s)et=et−2B(et,es)es for t∈I, t≠s; every vector of VI is a linear combination of the et, t∈I, so ρ(s)VI⊆VI, and since ρ(s) is invertible with ρ(s)−1=ρ(s), also ρ(s)VI=VI. Every w∈WI is a product of elements of I and their inverses by [F10], and the latter are again elements of I because s2=1; multiplying the identities ρ(s)VI=VI along such a product gives ρ(w)VI=VI. Consequently, for w∈WI and s∈I the root ρ(w)es lies in Φ∩VI, so ΦI⊆Φ∩VI; for I=∅ this says Φ∅=∅=Φ∩{0}, since 0∉Φ by [F5] and V∅={0} by [F11].

1.3F1F4F5F7F11

Setup of the converse containment. Let φ∈Φ∩VI. By [F5] the root φ lies in Φ+ or in Φ−; replacing φ by −φ, which is again a root in Φ∩VI because Φ and VI are stable under negation, we may assume φ∈Φ+, so by [F5] and [F11] we have φ=∑r∈Iλrer with λr>0 for every r∈supp⁡φ, and φ≠0. Let t:=tφ∈T be the reflection of the dictionary [F7], so that ρ(t)=rφ and B(φ,φ)=1 by [F4]. We prove t∈WI by strong induction on n:=ℓ(t)≥1. If n=1, then t is a word of length one, hence t=r for some r∈S; the dictionary gives ter=r, so t=tφ=ter and the clause "tα=tβ if and only if α=±β" of [F7] yields φ=±er; since φ and er both lie in Φ+ while Φ+ and Φ−=−Φ+ are disjoint by [F5], φ=er, and then r∈supp⁡φ⊆I by [F11], so t=r∈WI.

1.4F1F2

Additivity on the left coset. Let d∈WI. By [F2] one has ℓ(ds)>ℓ(d) for all s∈I, so d is the minimal representative of the left coset dWI characterized in [F1], and therefore ℓ(du)=ℓ(d)+ℓ(u) for all u∈WI.

2.1F3F4F5F6F7F8F9step 1.3

The induction step. Assume n:=ℓ(t)≥2 in the situation of step 1.3 and that the claim holds for all roots in Φ+∩VI whose reflection has length <n. Since 1=B(φ,φ)=∑r∈IλrB(φ,er) by [F3] and [F4], and λr>0 for r∈supp⁡φ, there is r∈supp⁡φ⊆I with B(φ,er)>0; fix such an r. If ∣supp⁡φ∣=1, then φ=λrer with λr>0, and 1=B(φ,φ)=λr2B(er,er)=λr2 gives λr=1, that is φ=er and t=ter=r; then ℓ(t)=1, against n≥2, so necessarily ∣supp⁡φ∣≥2. Put φ′:=ρ(t)er=rφ(er)=er−2B(φ,er)φ; its coefficient at each r′∈supp⁡φ∖{r} equals −2B(φ,er)λr′<0, and this set is nonempty, so φ′∈Φ− by [F5]. Put ψ:=ρ(r)φ=φ−2B(φ,er)er∈Φ∩VI; its coefficient at each r′∈supp⁡φ∖{r} is λr′>0, so ψ∉Φ− and hence ψ∈Φ+ by [F5], while ψ∈VI because φ,er∈VI. By the dictionary [F7], tψ=rtr. Since ρ(t)er=φ′∈Φ−, the root-length criterion [F6] gives ℓ(tr)<ℓ(t) and then ℓ(tr)=ℓ(t)−1 by [F8]; by [F9], ℓ(rt)=ℓ(tr)=ℓ(t)−1. Moreover ρ(rt)er=ρ(r)φ′ has at each r′∈supp⁡φ∖{r} the coefficient −2B(φ,er)λr′<0 of φ′, because ρ(r) alters only the coefficient of er; hence ρ(rt)er∈Φ− by [F5], and the criterion [F6] with w:=rt gives ℓ(tψ)=ℓ(rtr)<ℓ(rt), that is ℓ(tψ)=ℓ(t)−2 by [F8]. Since ψ∈Φ+∩VI and ℓ(tψ)<n, the induction hypothesis applies and yields tψ∈WI; then t=rtψr∈WI because r∈I.

2.2F2F8F9step 1.4

The two descriptions of the quotients. Let d∈W. If d∈WI, then by step 1.4 ℓ(dv)=ℓ(d)+ℓ(v)≥ℓ(d) for every v∈WI, with equality if and only if ℓ(v)=0, that is v=1. Conversely, if ℓ(d)≤ℓ(dv) for all v∈WI, take v=s∈I: then ℓ(ds)≥ℓ(d), and ℓ(ds)≠ℓ(d) by [F8], so ℓ(ds)>ℓ(d) for all s∈I, that is d∈WI by [F2]; hence WI={d:ℓ(d)≤ℓ(dv) for all v∈WI}. Applying this to d−1 and using WI=(IW)−1 from [F2] and the identities (d−1v)−1=v−1d and ℓ(x−1)=ℓ(x) from [F9], we get d∈IW if and only if ℓ(d)≤ℓ(v−1d) for all v∈WI; as v runs over WI so does v−1, so this is the second displayed description.

3.1F1F4F5F7step 1.3step 2.1

The converse containment. We prove by strong induction on n that every φ∈Φ+∩VI satisfies tφ∈WI, the cases n=1 and n≥2 being steps 1.3 and 2.1; the induction is well founded because the length values are natural numbers. First let φ∈Φ+∩VI. Then tφ∈WI, and its positive root is φ. Strong exchange furnishes a representation with letters in I: choosing a reduced expression tφ=s1⋯sm of tφ inside WI with all si∈I by [F1] and applying strong exchange [F7] to w:=tφ and the reflection tφ, whose square is 1 and has length 0<m, produces an index i with tφ=s1⋯si−1sisi−1⋯s1 and φ=ρ(s1⋯si−1)esi with s1⋯si−1∈WI and si∈I. Hence φ∈ΦI for positive φ. If instead φ∈Φ−∩VI, apply this argument to −φ=ρ(w)es with w∈WI, s∈I; then φ=ρ(ws)es∈ΦI, since ρ(s)es=−es and ws∈WI by [F4]. Therefore Φ∩VI⊆ΦI.

3.2F1F2F9step 1.4step 2.2

Every coset has a unique minimum. Let w∈W and let d∈WI be the unique element of WI∩wWI, so that w=dv with v∈WI by [F1] and wWI=dWI. Every x∈dWI has the form x=du with u∈WI, and by step 1.4, ℓ(x)=ℓ(d)+ℓ(u)≥ℓ(d), with equality if and only if ℓ(u)=0, that is u=1 and x=d. If d′∈WI∩dWI is a second such element, write d′=du′ and d=d′u with u,u′∈WI (possible since d′WI=dWI); step 1.4 gives ℓ(d′)=ℓ(d)+ℓ(u′) and ℓ(d)=ℓ(d′)+ℓ(u), so ℓ(u)=ℓ(u′)=0, u=u′=1 and d=d′. Thus dWI=wWI contains exactly one element of WI, namely its unique element of minimum length, which proves the right-handed assertions of (3). The left-handed assertions follow by the same argument applied to the inverses, using the description of IW in step 2.2 and the identities ℓ(x−1)=ℓ(x) of [F9].

4.1F1F5F7F11step 1.2step 3.1∎

Conclusion of (2). Combining steps 1.2 and 3.1 gives ΦI=Φ∩VI, which by [F11] is the displayed set of roots that are real linear combinations of the es, s∈I. Since Φ=Φ+⊔Φ− by [F5], intersecting with VI gives ΦI=ΦI+⊔ΦI− for ΦI±=ΦI∩Φ±=Φ±∩VI. For the reflection identification: if α∈ΦI+, say α=ρ(w)es with w∈WI, s∈I, then tα=wsw−1∈WI by [F7]; conversely, if t∈WI∩T, then t=tα for the unique α∈Φ+ by [F7], and choosing a reduced expression t=s1⋯sm inside WI with letters in I by [F1] and applying strong exchange [F7] to w:=t and the reflection t exhibits t=s1⋯si−1sisi−1⋯s1 and α=ρ(s1⋯si−1)esi, a root in ΦI because s1⋯si−1∈WI and si∈I. Hence the reflections of WI are exactly the tα with α∈ΦI+. Together with steps 1.1 and 3.2 this proves (1), (2) and (3); no Axiom of Choice is used: each proof fixes finitely many witnesses, and no simultaneous selection from an arbitrary family is required.

Remarks

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-08Open item page →

The parabolic intersection W_I cap dW_Jd inverse for d in ^IW^J

Statement

Let (S,m), W, ℓ be as in Standard parabolic subgroups, descent-free one- and two-sided representatives, parabolic and reflection subgroups, let I,J⊆S, and let d∈IWJ. Put

K:=I∩dJd−1={s∈I:d−1sd∈J},

where dJd−1={djd−1:j∈J}⊆T. Let VI, ΦI be as in Intersections of standard parabolics, the parabolic root subsystem, and global minimality of coset representatives (2) and let α↦tα be the root-reflection dictionary of The inversion formula ∣N(w)∣=ℓ(w), the root-reflection dictionary and strong exchange (1).

(1) The intersection. WI∩dWJd−1=WK. Conjugating by d−1 gives

d−1WKd=Wd−1Kd,d−1Kd={d−1sd:s∈K}⊆J,

and equivalently WJ∩d−1WId=Wd−1Kd.

(2) The descent form. If y∈WI∩dWJd−1 and y=s1⋯sp is a reduced expression of y with letters s1,…,sp∈I, then

d−1sid∈Jfor every i=1,…,p.

(3) The conjugated positive root is simple. Let s∈I. Then z:=d−1sd is a reflection of W (that is, z∈T), its root is ρ(d)−1es∈Φ+, and z∈J if and only if this root is one of the simple roots ej, j∈J. Moreover, if z∈WJ then already z∈J: a reflection of the form d−1sd with s∈I that lies in WJ is necessarily a simple reflection of the parabolic root system ΦJ=Φ∩VJ of Intersections of standard parabolics, the parabolic root subsystem, and global minimality of coset representatives (2), never a non-simple positive combination such as ej+ej′.

Facts & Assumptions

Given: a finite Coxeter matrix (S,m) with presented group W and length ℓ, subsets I,J⊆S, an element d∈IWJ and the subset K={s∈I:d−1sd∈J}.

[F1]

WI=⟨s:s∈I⟩, IW={w:ℓ(sw)>ℓ(w) for all s∈I}, WJ={w:ℓ(ws)>ℓ(w) for all s∈J} and IWJ=IW∩WJ; in particular d∈IWJ satisfies ℓ(sd)>ℓ(d) for s∈I and ℓ(ds)>ℓ(d) for s∈J (Standard parabolic subgroups, descent-free one- and two-sided representatives, parabolic and reflection subgroups).

[F2]

WJ=⟨J⟩={w:S(w)⊆J}, WJ∩S=J, (WJ,J) is a Coxeter system with intrinsic length ℓ∣WJ, every left coset aWJ has a unique minimal element d, characterized by ℓ(ds)>ℓ(d) for s∈J and satisfying ℓ(du)=ℓ(d)+ℓ(u) for all u∈WJ, and dually ℓ(ud)=ℓ(u)+ℓ(d) for the minimal representative d∈JW of a right coset and all u∈WJ (Support, intrinsic parabolic presentations, minimal coset representatives and length additivity, with the type-A identification).

[F3]

For d∈IWJ, every x∈WIdWJ admits some factorization x=udv with u∈WI, v∈WJ and ℓ(x)=ℓ(u)+ℓ(d)+ℓ(v); moreover ℓ(d)≤ℓ(x) for all x∈WIdWJ (Descent reduction, minimum-length elements, and the additive factorization in a double coset).

[F4]

Strong exchange: if w=w1⋯wn is a reduced expression and t∈T satisfies ℓ(tw)<ℓ(w), then there is a unique index i with tw=w1⋯wi^⋯wn and t=w1⋯wi−1wiwi−1⋯w1 (The inversion formula ∣N(w)∣=ℓ(w), the root-reflection dictionary and strong exchange).

[F5]

The root-reflection dictionary: for α∈Φ and w∈W, s∈S with α=ρ(w)es the element tα:=wsw−1 is well defined, tρ(w)α=wtαw−1, the map Φ+→T, α↦tα, is a bijection and tα=tβ if and only if α=±β (The inversion formula ∣N(w)∣=ℓ(w), the root-reflection dictionary and strong exchange).

[F6]

For all w∈W and s∈S: ℓ(ws)>ℓ(w)  ⟺  ρ(w)es∈Φ+ and ℓ(ws)<ℓ(w)  ⟺  ρ(w)es∈Φ− (The root-length criterion and faithfulness of the canonical reflection representation).

[F7]

T={wsw−1:w∈W, s∈S}, so every conjugate d−1sd of a simple reflection is a reflection (The canonical reflection homomorphism, roots, reflections, and the positive cone).

[F8]

ΦJ=Φ∩VJ, and the reflections lying in WJ are exactly the tα with α∈ΦJ+; in particular the simple roots ej, j∈J, correspond to the simple reflections of WJ, and a reflection of WJ whose positive root is not any ej (j∈J) cannot have length one: by [F2] a length-one element lies in J, and [F5] then identifies its positive root with ej (Intersections of standard parabolics, the parabolic root subsystem, and global minimality of coset representatives).

[F9]

For w=s1⋯sk one has ℓ(w)≤k, so ℓ(uv)≤ℓ(u)+ℓ(v), and ℓ(w−1)=ℓ(w) with (ab)−1=b−1a−1 (Coxeter matrices, the presented Coxeter group, reduced words, length, and standard parabolic subgroups, Group and abelian group).

Proof

technique · direct; strong exchange at the first letter of a reduced expression, then iteration along the expression
1.1F1F7

The easy inclusion and the conjugation identities. If s∈K, then s∈I⊆WI and d−1sd∈J, so s=d(d−1sd)d−1∈dWJd−1; hence WK⊆WI∩dWJd−1. Conjugation by d−1 is an isomorphism of groups carrying the generating set K to d−1Kd, so d−1WKd=Wd−1Kd, and d−1Kd⊆J holds by the definition of K. Conjugating the coming equality WI∩dWJd−1=WK by d−1 will give WJ∩d−1WId=Wd−1Kd.

1.2F1F2F3F4F9

The key step: the first letter conjugated lies in WJ. Let y∈WI∩dWJd−1 and let y=s1⋯sp be a reduced expression with p≥1 and all si∈I by [F2]; put y~:=d−1yd∈WJ and write y~=y~1⋯y~p for a reduced expression of y~. Then ℓ(y)=ℓ(y~)=p: indeed yd=dy~, while ℓ(yd)=ℓ(y)+ℓ(d) because y∈WI, d∈IW, and ℓ(dy~)=ℓ(d)+ℓ(y~) because d∈WJ, y~∈WJ, both by [F2]. Consequently the word s1⋯sp followed by a reduced word a1⋯aq of d, and the word a1⋯aq followed by y~1⋯y~p, are two reduced expressions of the same element yd=dy~ of length ℓ(d)+p, where q=ℓ(d). The element s1 is a left descent of yd: s1y=s2⋯sp, so ℓ(s1⋅yd)=ℓ(s1y)+ℓ(d)=(p−1)+ℓ(d)<ℓ(yd), using the additivity of [F2] and [F9] for ℓ(s1y)=p−1. Applying strong exchange [F4] to the reduced expression a1⋯aqy~1⋯y~p of yd and the reflection s1 gives a unique index i with s1(yd) equal to that word with its i-th letter deleted, and s1=a1⋯ai−1aiai−1⋯a1 if i≤q, while s1=Xy~jX−1 with X:=dy~1⋯y~j−1 if i=q+j. The first case is impossible: then the deleted word exhibits di:=a1⋯ai−1ai+1⋯aq with diy~=(s2⋯sp)d, hence di=(s2⋯sp) d y~−1∈WIdWJ, while ℓ(di)≤q−1<q=ℓ(d) by [F9], contradicting the minimality of d in WIdWJ given by [F3]. Hence i=q+j for some j, and substituting X=dy~1⋯y~j−1 gives d−1s1d=y~1⋯y~j−1y~jy~j−1⋯y~1∈WJ, a conjugate in WJ of the letter y~j∈J.

1.3F1F5F6F7F9

The conjugated root is positive. Let s∈I and z:=d−1sd. Then z∈T by [F7], and z=tρ(d)−1es by the dictionary [F5], since d−1sd=d−1tesd=tρ(d−1)es. The root ρ(d)−1es lies in Φ+: because d∈IW we have ℓ(sd)>ℓ(d) by [F1], and ℓ(sd)=ℓ(d−1s) and ℓ(d)=ℓ(d−1) by [F9], so the root-length criterion [F6] applied to w:=d−1, s gives ρ(d−1)es=ρ(d)−1es∈Φ+.

2.1F1F2F9step 1.2

Length one forces simplicity. Let s∈I and suppose z:=d−1sd∈WJ. Repeating the length computation of step 1.2 with (y,y~) replaced by (s,z) is legitimate because s∈WI, z∈WJ and sd=dz: it gives ℓ(z)=ℓ(s)=1. An element of WJ of length one is a product of one generator, hence lies in WJ∩S=J by [F2]. Applying this to the first letter of step 1.2, s1∈K; indeed d−1s1d∈WJ there, so d−1s1d∈J and s1∈I.

2.2F5step 1.3

The root of a conjugate that lies in J. Let s∈I and let φ:=ρ(d)−1es∈Φ+ be the root of z=d−1sd from step 1.3, so that z=tφ. If z∈J, then z=j=tej for an element j∈J, so tφ=tej; by [F5] φ=±ej, and since both φ and ej lie in Φ+ while Φ+ and Φ−=−Φ+ are disjoint, φ=ej. Conversely, if φ=ej with j∈J, then z=tej=j∈J.

3.1F1F2F9step 1.1step 2.1

The intersection. Let y∈WI∩dWJd−1 with reduced expression y=s1⋯sp, letters in I, and write y=dy~d−1 with y~∈WJ. If p=0, then y=1∈WK and there is nothing to prove, so assume p≥1. By step 2.1 the first letter s1 lies in K, i.e. d−1s1d∈J; then y′:=s1y=s2⋯sp satisfies y′∈WI and y′=d (d−1s1d) y~ d−1∈dWJd−1, and it has length p−1, because ℓ(y′)≤p−1 by [F9] and p=ℓ(y)=ℓ(s1y′)≤1+ℓ(y′). Iterating this argument through the suffixes shows d−1sid∈J, hence si∈K, for every i=1,…,p. Thus y∈WK, which proves (2) and, together with the inclusion of step 1.1, gives WI∩dWJd−1=WK, that is (1) and its conjugation identities.

4.1F2F5F8step 1.3step 2.1step 2.2∎

Conclusion of (3). Let s∈I. By step 1.3 the element z=d−1sd lies in T with root φ=ρ(d)−1es∈Φ+; by step 2.2 one has z∈J if and only if φ=ej for some j∈J. If instead z∈WJ, then by step 2.1 (applied to this s) ℓ(z)=1 and z∈WJ∩S=J by [F2], so here too φ=ej for some j∈J: the conjugated positive root is then a simple root of ΦJ=Φ∩VJ and never a non-simple positive combination such as ej+ej′, because [F8] shows that every reflection with a non-simple positive root has length >1, whereas ℓ(z)=1. The illustrative sum ej+ej′ need not itself be a root; when it is a root for distinct j,j′, it is non-simple. This proves (3) for every s∈I.

Remarks

  • The key step is Lusztig's argument for the intersection of parabolic subgroups: strong exchange at the first letter of a reduced expression of y∈WI∩dWJd−1 either deletes a letter of the middle representative d (impossible by minimality) or exhibits d−1s1d as a conjugate inside WJ of a letter of J.
  • The warning in (3) is not vacuous: in a parabolic subsystem of type A2 the sum ej+ej′ is a positive root whose reflection tej+ej′ lies in WJ with length 3, so "lying in WJ" alone would not make the conjugated root simple; it is the length-one conclusion ℓ(d−1sd)=ℓ(s)=1 that forces z∈J and hence φ=ej.
TheoremStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-08Open item page →

Unique minimal double coset representatives and the additive normal form u-d-v

Statement

Let I,J⊆S, let d∈IWJ, and put

K:=I∩dJd−1={s∈I:d−1sd∈J},

so that WI∩dWJd−1=WK by The parabolic intersection W_I cap dW_Jd inverse for d in ^IW^J. Write

WIK:={u∈WI:ℓ(us)>ℓ(u) for all s∈K}

for the set of elements of WI with no right descent in K --- the same construction as in Standard parabolic subgroups, descent-free one- and two-sided representatives, parabolic and reflection subgroups (2), applied inside the Coxeter system (WI,I) with its standard parabolic subgroup WK.

(1) Unique minimum. Two elements of IWJ lie in a common double coset WIwWJ only if they are equal. Hence, by Descent reduction, minimum-length elements, and the additive factorization in a double coset (1),(2), every double coset WIwWJ contains exactly one element of IWJ, namely its unique element of minimum length; in particular d is uniquely determined by the double coset WIdWJ.

(2) The additive normal form. Every x∈WIdWJ has a unique representation

x=u d vwith u∈WIK, v∈WJ,

and every product udv with u∈WIK, v∈WJ lies in WIdWJ. Thus

WIK×WJ→WIdWJ,(u,v)↦udv

is a bijection, and for all u∈WIK and v∈WJ

ℓ(udv)=ℓ(u)+ℓ(d)+ℓ(v).

(3) The size of a double coset. ∣WIdWJ∣=∣WIK∣⋅∣WJ∣; and if WI is finite then ∣WIdWJ∣=(∣WI∣/∣WK∣)⋅∣WJ∣, with the quotient a finite integer and the product interpreted as cardinal multiplication.

(4) The restriction to WIK is necessary. For arbitrary u∈WI the representation x=udv is not unique in general: if u=u0z is the factorization of u with u0∈WIK and z∈WK, and z′:=d−1zd∈WJ, then

udv=u0 d (z′v),u0∈WIK, z′v∈WJ,

and z′≠1 whenever z≠1; so the same x has two representations of the form udv with different first factors unless u∈WIK already. This is why the transversal restriction in (2) is recorded explicitly and may not be dropped.

Facts & Assumptions

Given: a finite Coxeter matrix (S,m) with presented group W and length ℓ, subsets I,J⊆S, an element d∈IWJ, the subset K={s∈I:d−1sd∈J} and the transversal WIK of the statement.

[F1]

Inside the Coxeter system (WI,I) with standard parabolic subgroup WK: WIK={u∈WI:ℓ(us)>ℓ(u) for all s∈K} is the set of minimal-length representatives of the left cosets uWK in WI; every u∈WI has a unique factorization u=u0z with u0∈WIK, z∈WK, and then ℓ(u)=ℓ(u0)+ℓ(z), while ℓ(u0z′)=ℓ(u0)+ℓ(z′) for all z′∈WK; more generally WJ={w:S(w)⊆J} for J⊆S, every left coset aWJ has a unique minimal element, and ℓ(du)=ℓ(d)+ℓ(u) for a minimal representative d∈WJ of aWJ and u∈WJ, with the mirrored statement for right cosets: for the minimal representative d of a right coset WJa one has ℓ(ud)=ℓ(u)+ℓ(d) for all u∈WJ (Support, intrinsic parabolic presentations, minimal coset representatives and length additivity, with the type-A identification).

[F2]

The element x=d−1zd of the statement lies in WJ: indeed WI∩dWJd−1=WK and d−1WKd=Wd−1Kd⊆WJ (The parabolic intersection W_I cap dW_Jd inverse for d in ^IW^J).

[F3]

Every double coset WIwWJ contains an element of IWJ; for d∈IWJ one has ℓ(d)≤ℓ(x) for all x∈WIdWJ with equality if and only if x=d; and every x∈WIdWJ has the form x=adc with a∈WI, c∈WJ and ℓ(x)=ℓ(a)+ℓ(d)+ℓ(c) (Descent reduction, minimum-length elements, and the additive factorization in a double coset).

[F4]

For w=s1⋯sk one has ℓ(w)≤k, so ℓ(uv)≤ℓ(u)+ℓ(v) for all u,v∈W (Coxeter matrices, the presented Coxeter group, reduced words, length, and standard parabolic subgroups).

[F5]

Multiplication in W is associative and equalities in W may be multiplied and cancelled; in particular conjugation w↦cwc−1 is injective (Group and abelian group).

Proof

technique · direct; existence by factoring inside $W_I$ and $W_J$, uniqueness by the intercoset intersection $W_K$, then the cardinality and necessity clauses
1.1F1F2F3F5

Existence of the normal form. Let x∈WIdWJ. By [F3] write x=adc with a∈WI, c∈WJ and ℓ(x)=ℓ(a)+ℓ(d)+ℓ(c); by [F1] factor a=u0z with u0∈WIK, z∈WK and ℓ(a)=ℓ(u0)+ℓ(z). Put z′:=d−1zd∈WJ by [F2], which satisfies zd=dz′ because z=dz′d−1, and put v:=z′c∈WJ. Then x=u0zdc=u0dz′c=u0dv with u0∈WIK, v∈WJ, so every element of the double coset has a representation of the required form.

1.2F1F2F5

Uniqueness of the first factor I: the intersection. Suppose udv=u′dv′ with u,u′∈WIK and v,v′∈WJ. Left-multiplying by u−1 and right-multiplying by (v′)−1d−1 and cancelling gives u−1u′=d v (v′)−1d−1∈WI∩dWJd−1, which equals WK by [F2]; hence u′=u z for some z∈WK. Since u,u′∈WIK and WIK consists of the minimal representatives of the left cosets uWK in WI by [F1], u=u′; substituting back gives dv=dv′ and hence v=v′ by cancellation, so the representation has at most one pair of factors and the map of (2) is injective.

2.1F1F3F4step 1.1step 1.2

Additivity along the representation. Let x∈WIdWJ and write x=adc with a∈WI, c∈WJ as in [F3], and factor a=u0z, z′=d−1zd, v=z′c as in step 1.1. First ℓ(z)=ℓ(z′): indeed zd=dz′, while ℓ(zd)=ℓ(z)+ℓ(d) because z∈WK⊆WI and d∈IW, and ℓ(dz′)=ℓ(d)+ℓ(z′) because d∈WJ and z′∈WJ, both by [F1]. Estimating with subadditivity [F4], ℓ(x)≤ℓ(u0)+ℓ(d)+ℓ(v)≤ℓ(u0)+ℓ(d)+ℓ(z′)+ℓ(c)=ℓ(u0)+ℓ(z)+ℓ(d)+ℓ(c)=ℓ(a)+ℓ(d)+ℓ(c)=ℓ(x), where the last equality is [F3]; hence every inequality is an equality, so ℓ(u0dv)=ℓ(u0)+ℓ(d)+ℓ(v) for the representation of step 1.1, and also ℓ(v)=ℓ(z′)+ℓ(c). For any prescribed u∈WIK, v∈WJ, apply this construction to x=udv; uniqueness in step 1.2 identifies the constructed pair with (u,v), proving additivity for every such pair.

2.2F1step 1.1step 1.2

The size formulas. The map WIK×WJ→WIdWJ, (u,v)↦udv, is surjective by step 1.1 and injective by step 1.2, hence a bijection, so ∣WIdWJ∣=∣WIK∣⋅∣WJ∣. If WI is finite, the factorization u=u0z of [F1] is a bijection WIK×WK→WI, so ∣WIK∣=∣WI∣/∣WK∣ and therefore ∣WIdWJ∣=(∣WI∣/∣WK∣)⋅∣WJ∣, also when WJ is infinite.

3.1F3step 1.1step 2.1

Uniqueness of the minimum. Suppose d,d′∈IWJ lie in a common double coset Ω:=WIdWJ=WId′WJ. By [F3] each of d,d′ is a minimum-length element of Ω, so ℓ(d)=ℓ(d′); applying the representation of steps 1.1 and 2.1 with x:=d′ and base point d gives d′=udv with u∈WIK, v∈WJ and ℓ(d′)=ℓ(u)+ℓ(d)+ℓ(v). Hence ℓ(u)+ℓ(v)=0, so u=v=1 and d′=d. Consequently each double coset contains at most one element of IWJ; by [F3] (or its part (1)) it contains at least one, namely its unique minimum.

4.1F1F2F5step 1.1step 1.2step 2.1∎

The normal form and the necessity of the restriction. By steps 1.1, 1.2 and 2.1 every x∈WIdWJ has a unique representation x=udv with u∈WIK, v∈WJ, and then ℓ(x)=ℓ(u)+ℓ(d)+ℓ(v); conversely every product udv with u∈WIK⊆WI and v∈WJ lies in WIdWJ by the definition of the double coset, so the displayed map is a bijection and (2) holds. For (4) let u∈WI and v∈WJ be arbitrary and factor u=u0z with u0∈WIK, z∈WK by [F1]; put z′:=d−1zd∈WJ by [F2]. Then udv=u0zdv=u0dz′v=u0d(z′v) with u0∈WIK and z′v∈WJ, so x=udv has two representations with first factors u and u0; and z′≠1 whenever z≠1, because z=dz′d−1 and conjugation by d is injective by [F5], so u≠u0 whenever z≠1, that is, whenever u∉WIK. This is (4) and completes the proof of (1)-(4); no Axiom of Choice is used.

Remarks

  • The theorem is the algebraic heart of the double coset calculus: (1) and (3) say that the double cosets WIwWJ are parameterized by IWJ, and (2) upgrades the transversal to a normal form with exact length additivity. The counterexample in (4) is the classical failure of uniqueness once the transversal condition on the first factor is dropped.
  • The finite formula of (3) is the parabolic analogue of ∣WIwWJ∣=∣WI∣∣WJ∣/∣WI∩wWJw−1∣; the intersection is written WK by The parabolic intersection W_I cap dW_Jd inverse for d in ^IW^J.

5 · Examples, counterexamples and false statements

None yet.

Sources