Symmetric Groups and Iwahori–Hecke Algebras
4.1 The symmetric groups
4.1.5 Equivalence of reduced expressions
We now state the so-calledexchange theorem.
Theorem 4.8.Let si1· · ·sir be a reduced expression for w ∈ Sn, where r = λ(w). If λ(wsj) < λ(w) for some j ∈ {1, . . . , n−1}, then there is k ∈ {1, . . . , r} such that wsj =si1· · ·s!ik· · ·sir. If λ(sjw) < λ(w) for some j∈ {1, . . . , n−1}, then there isk∈ {1, . . . , r}such thatsjw=si1· · ·s!ik· · ·sir. Proof. We saw in the proof of Lemma 4.7 (a) that if t1, . . . , tr are the trans- positions defined by (4.7), then I(w) = {t1, . . . , tr}. If λ(wsj)< λ(w), then sj ∈ I(w) by Lemma 4.7 (c). Therefore, sj = tk for some k ∈ {1, . . . , r}. By (4.8),
wsj=wtk =si1· · ·s!ik· · ·sir.
The second claim is deduced from the first one by replacingw withw−1. Corollary 4.9.Let w∈Sn. If λ(wsj)< λ(w) for some j ∈ {1, . . . , n−1}, then there is a reduced expression forwending withsj. Ifλ(sjw)< λ(w)for some j ∈ {1, . . . , n−1}, then there is a reduced expression for w beginning withsj.
This is a direct corollary of the previous theorem: ifλ(wsj)< λ(w), then wsj = si1· · ·s!ik· · ·sir and w = si1· · ·s!ik· · ·sirsj is a reduced expression for w, since its length is equal to r = λ(w). The second claim is proven similarly.
We conclude with a lemma needed in the proof of Lemma 4.18 below.
Lemma 4.10.If λ(siwsj) = λ(w) and λ(siw) = λ(wsj) for w ∈ Sn and somei, j∈ {1, . . . , n−1}, thensiw=wsj andsiwsj =w.
Proof. (a) Suppose first that λ(siw) = λ(wsj) > λ(siwsj) = λ(w). By Lemma 4.6,
I(siw) =w−1I(si)wΔI(w) ={w−1siw}ΔI(w).
Sinceλ(siwsj) < λ(siw) and λ(wsj)> λ(w), Lemma 4.7(c) implies that sj
belongs toI(siw), but not toI(w). Thereforesj=w−1siw; hencesiw=wsj
andsiwsj=ws2j =w.
(b) If λ(siw) = λ(wsj) < λ(siwsj) = λ(w), then we apply a similar argument, using thatI(w) =I(si(siw)) ={w−1siw}ΔI(siw).
S1(i, j)S2∼S1(j, i)S2 (4.9) for allS1, S2∈Mn and alli, j∈ {1, . . . , n−1} such that|i−j| ≥2, and
S1(i, j, i)S2∼S1(j, i, j)S2 (4.10) for allS1, S2∈Mn and alli, j∈ {1, . . . , n−1} such that|i−j|= 1. Observe that equivalent sequences have the same length. The equivalence relation∼ has been devised so that
(i1, . . . , ik)∼(j1, . . . , jk)∈Mn =⇒ si1· · ·sik=sj1· · ·sjk∈Sn. Lemma 4.11.Ifsi1· · ·sik andsj1· · ·sjk are reduced expressions for the same permutationw∈Sn, then(i1, . . . , ik)∼(j1, . . . , jk)inMn.
This lemma shows that for any w ∈ Sn, we can pass from one reduced expression forwto any other reduced expression forwusing only the relations
sisj =sjsi if|i−j| ≥2, sisjsi=sjsisj if|i−j|= 1.
Proof. We prove the lemma by induction onk. Ifk= 0, thenw= 1 has only one reduced expression. Ifk= 1, thenw=si for someiandw has only one reduced expression.
Assume that k ≥2. From the equality si1· · ·sik = sj1· · ·sjk we deduce thatsi2· · ·sik =si1sj1· · ·sjk. Sincesi2· · ·sik is reduced,
λ(si1sj1· · ·sjk) =λ(si2· · ·sik) =k−1< k=λ(sj1· · ·sjk). Therefore, by Theorem 4.8, there is an integerpwith 1≤p≤ksuch that
si2· · ·sik=si1sj1· · ·sjk=sj1· · ·s!jp· · ·sjk. (4.11) Sincesi2· · ·sik andsj1· · ·s!jp· · ·sjkrepresent the same permutation and have the same lengthk−1 and sincesi2· · ·sikis reduced,sj1· · ·s!jp· · ·sjk is also re- duced. By the induction hypothesis, (i2, . . . , ik)∼(j1, . . . ,jp, . . . , jk). Hence,
(i1, i2, . . . , ik)∼(i1, j1, . . . ,jp, . . . , jk). (4.12) The wordsj1· · ·sjp, being a part of the reduced word sj1· · ·sjk, is reduced.
The second equality in (4.11) implies that si1sj1· · ·sjp−1 and sj1· · ·sjp are equal in Sn. Since these words have the same length pand one of them is reduced, so is the other one. Ifp < k, then we apply the induction hypothesis to these words, obtaining (i1, j1, . . . , jp−1)∼(j1, . . . , jp). From this and (4.12), we obtain
(i1, i2, . . . , ik)∼(i1, j1, . . . ,jp, . . . , jk)
= (i1, j1, . . . , jp−1)(jp+1, . . . , jk)
∼(j1, . . . , jp)(jp+1, . . . , jk)
= (j1, . . . , jp, jp+1, . . . , jk), which was to be proven.
Ifp=k, then (4.12) becomes (i1, i2, . . . , ik)∼(i1, j1, . . . , jk−1). This equi- valence implies that
si1sj1· · ·sjk−1 =si1si2· · ·sik=sj1sj2· · ·sjk. Summarizing our argument, we see that to prove the implication
si1· · ·sik=sj1· · ·sjk =⇒ (i1, . . . , ik)∼(j1, . . . , jk), it is enough to prove the implication
si1sj1· · ·sjk−1 =sj1· · ·sjk =⇒ (i1, j1, . . . , jk−1)∼(j1, . . . , jk). (4.13) We now start the argument all over again with the reduced expressions sj1· · ·sjk = si1sj1· · ·sjk−1. Proceeding as above, we show that in order to prove (4.13), it is enough to prove the implication
sj1si1sj1· · ·sjk−2=si1sj1sj2· · ·sjk−1
=⇒ (j1, i1, j1, . . . , jk−2)∼(i1, j1, j2, . . . , jk−1). (4.14) We first prove (4.14) when|i1−j1| ≥2. Thensi1sj1 =sj1si1 and
si1sj1sj1· · ·sjk−2 =sj1si1sj1· · ·sjk−2 =si1sj1sj2· · ·sjk−1. Multiplying on the left bysj1si1 inSn, we obtain
sj1· · ·sjk−2=sj2· · ·sjk−1.
Both sides are reduced expressions of lengthk−2. By the induction assump- tion, (j1, . . . , jk−2)∼(j2, . . . , jk−1). From this and (4.9), we obtain
(j1, i1, j1, . . . , jk−2) = (j1, i1)(j1, . . . , jk−2)
∼(i1, j1)(j2, . . . , jk−1)
= (i1, j1, j2, . . . , jk−1), which was to be proven.
If |i1−j1| = 1, then we proceed again as above and reduce the proof of (4.14) to showing that the equality
si1sj1si1sj1· · ·sjk−3 =sj1si1sj1sj2· · ·sjk−2
implies that (i1, j1, i1, j1, . . . , jk−3) ∼ (j1, i1, j1, j2, . . . , jk−2). This and the equalitysj1si1sj1=si1sj1si1 imply that
sj1si1sj1sj1· · ·sjk−3 =si1sj1si1sj1· · ·sjk−3 =sj1si1sj1sj2· · ·sjk−2, which, after left multiplication bysj1si1sj1, gives
sj1· · ·sjk−3=sj2· · ·sjk−2.
Since both sides of this equality are reduced expressions of lengthk−3, we can apply the induction hypothesis and obtain (j1, . . . , jk−3)∼(j2, . . . , jk−2).
From this and (4.10),
(i1, j1, i1, j1, . . . , jk−3) = (i1, j1, i1)(j1, . . . , jk−3)
∼(j1, i1, j1)(j2, . . . , jk−2)
= (j1, i1, j1, j2, . . . , jk−2),
which was to be proven.
The following theorem is useful for defining maps from the symmetric groups to monoids.
Theorem 4.12.For any monoidM and anyx1, . . . , xn−1∈M satisfying the relations
xixj=xjxi if |i−j| ≥2, xixjxi=xjxixj if |i−j|= 1, there is a set-theoretic mapρ:Sn→M defined by
ρ(w) =xi1· · ·xik,
for anyw∈Sn and any reduced expression w=si1· · ·sik. Proof. Define a monoid homomorphismρ:Mn→M by
ρ(i1, . . . , ik) =xi1· · ·xik
for all (i1, . . . , ik)∈Mn. We claim thatρ(S) =ρ(S) for allS, S ∈Mn such that S ∼S. Indeed, by definition of the equivalence ∼, it suffices to prove the claim whenS =S1(i, j)S2 (resp.S=S1(i, j, i)S2) andS =S1(j, i)S2
(resp.S=S1(j, i, j)S2) forS1,S2∈Mn andi, j∈ {1, . . . , n−1}such that
|i−j| ≥2 (resp.|i−j|= 1). By the assumptions of the theorem, ρ(S1(i, j)S2) =ρ(S1)xixjρ(S2)
=ρ(S1)xjxiρ(S2)
=ρ(S1(j, i)S2) if|i−j| ≥2, and
ρ(S1(i, j, i)S2) =ρ(S1)xixjxiρ(S2)
=ρ(S1)xjxixjρ(S2)
=ρ(S1(j, i, j)S2) if|i−j|= 1.
To prove the theorem, we need only check that ρ is well defined, i.e., if si1· · ·sik andsj1· · ·sjk are reduced expressions forw∈Sn, then
xi1· · ·xik =xj1· · ·xjk.
By Lemma 4.11, (i1, . . . , ik)∼(j1, . . . , jk) inMn. By the claim above, xi1· · ·xik =ρ(i1, . . . , ik) =ρ(j1, . . . , jk) =xj1· · ·xjk. 4.1.6 The longest element ofSn
Letw0∈Sn be the permutationi→n+ 1−ifor alli∈ {1, . . . , n−1}: w0=
1 2 . . . n−1 n n n−1 . . . 2 1
. (4.15)
It is clear thatw0is the only permutationw∈Sn such thatw(i)> w(j) for alli,j∈ {1, . . . , n−1}withi < j. In other words,w=w0 if and only if the setI(w) consists of all transpositions. By Lemma 4.7 (a),λ(w0) =n(n−1)/2 andλ(w)< n(n−1)/2 for w=w0. Because of this, w0 is called thelongest element ofSn. We record two other properties ofw0 (the second one will be used in Section 6.5.2).
Lemma 4.13.If w∈Sn satisfiesλ(wsi)< λ(w) for all i ∈ {1, . . . , n−1}, thenw=w0.
Proof. By Lemma 4.7 (c), si ∈I(w) for alli. Thenw(i)> w(i+ 1) for alli.
The only permutation satisfying these inequalities isw0. Lemma 4.14.For any u, v∈Sn such thatuv=w0,
λ(u) +λ(v) =λ(w0). Proof. The lemma trivially holds for u=w0 andv= 1.
We claim that for any u∈Sn,u=w0, there is a sequencesi1, . . . , sir of simple transpositions such thatusi1· · ·sir =w0andλ(usi1· · ·sir) =λ(u)+r.
Before we prove the claim, let us show that it implies the lemma foru and v=u−1w0=si1· · ·sir. Clearly,λ(v)≤r and
λ(w0) =λ(uv)≤λ(u) +λ(v)≤λ(u) +r=λ(usi1· · ·sir) =λ(w0). Therefore,λ(u) +λ(v) =λ(w0).
Let us now establish the claim. Since u = w0, by Lemma 4.13, there issi1 such thatλ(usi1)≥λ(u). By Lemma 4.5, we haveλ(usi1) =λ(u) + 1. If λ(usi1) =λ(w0), thenusi1=w0, sincew0is the unique element ofSnof max- imal length, and we are done. Ifλ(usi1)< λ(w0), then again by Lemma 4.13, we can findsi2 such thatλ(usi1si2)≥λ(usi1). Then
λ(usi1si2) =λ(usi1) + 1 =λ(u) + 2.
Ifλ(usi1si2) =λ(w0), thenusi1si2=w0and we are done. If not, we continue as above until we find the required sequencesi1, . . . , sir.
Exercise 4.1.1.Using Theorem 4.8, prove that ifλ(si1· · ·sir)< r, then there arep, q∈ {1, . . . , r} such thatp < q and
si1· · ·sir =si1· · ·s!ip· · ·s!iq· · ·sir, wheres!ip ands!iq are removed on the right-hand side.
Exercise 4.1.2.Deduce Theorem 4.1 from Theorem 4.12, using the latter to construct a left inverseSn→Gn ofϕ:Gn→Sn.
Exercise 4.1.3.(a) Show thatwk,=sksk−1· · ·sis a reduced word for each pair (k, ) such that 1≤≤k≤n−1.
(b) Prove that the wordwk1,1wk2,2· · ·wkr,r obtained by concatenating words as in (a) is reduced fork1< k2<· · ·< kr.
Exercise 4.1.4.For any integerk≥1, set [k]q = 1 +q+· · ·+qk−1 ∈Z[q].
Show that
w∈Sn
qλ(w)= [1]q[2]q[3]q· · ·[n]q. (Hint: Use Exercise 4.1.3 and Corollary 4.4.)
Exercise 4.1.5.Prove that for any w ∈ Sn and any i = 1, . . . , n−1, the equalityλ(siw) =λ(w) + 1 holds if and only ifw−1(i)< w−1(i+ 1).