On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6....
Transcript of On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6....
![Page 1: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/1.jpg)
On property (T) for Aut(Fn)
Piotr Nowak
Institute of MathematicsPolish Academy of SciencesWarsaw
1
![Page 2: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/2.jpg)
G - group generated by a finite set S
Definition (“Easy definition”)
G has property (T) if every action of G on a Hilbert space byaffine isometries has a fixed point
⇐⇒
H1(G, π) = 0
for every π - unitary representation of G on some Hilbert space.
2
![Page 3: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/3.jpg)
G - group generated by a finite set S
Definition (“Easy definition”)
G has property (T) if every action of G on a Hilbert space byaffine isometries has a fixed point
⇐⇒
H1(G, π) = 0
for every π - unitary representation of G on some Hilbert space.
2
![Page 4: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/4.jpg)
Property (T)
G - group generated by a finite set S
Definition (Kazhdan 1966)
G has property (T) if there is κ = κ(G,S) > 0 such that
sups∈S‖v − πsv‖ ≥ κ‖v‖
for every unitary representation without invariant vectors.
Kazhdan constant = optimal κ(G,S)
⇐⇒ the trivial rep is an isolated point in the unitary dual of G withthe Fell topology
3
![Page 5: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/5.jpg)
Property (T)
Finite groups have (T)
Examples of infinite groups with (T):
• higher rank simple Lie groups and their latticesSLn(R), SLn(Z), n ≥ 3 (Kazhdan, 1966)
• Sp(n, 1) (Kostant 1969)
• automorphism groups of certain buildings(Cartwright-Młotkowski-Steger 1996, Pansu, Zuk,Ballmann-Swiatkowski, 1997-98)
• certain random hyperbolic groups in the triangular andGromov model (Zuk 2003)
Examples without (T): amenable, free, groups with infinite abelianization,groups acting unboundedly on finite-dim. CAT(0) cube complexes(Niblo-Reeves)
4
![Page 6: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/6.jpg)
Why is property (T) interesting?
• Margulis (1977): G with (T), Ni ⊆ G family of finite indexsubgroups with trivial intersection =⇒ Cayley graphs of G/Ni
form expander graphs (e.g. SL3(Zpn ) for p-prime)
• Fisher-Margulis (2003): G has (T), acts smoothly on asmooth manifold via an action ρ. Then any smooth action ρ′
sufficiently close to ρ on the generators is conjugate to ρ.
• rigidity for von Neumann algebras, Popa’s deformation/rigiditytechniques
• counterexamples to Baum-Connes type conjectures(Higson-Lafforgue-Skandalis 2003): K -theory classesrepresented by Kazhdan-type projections do not lie in theimage of certain versions of the Baum-Connes assembly map
5
![Page 7: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/7.jpg)
Why is property (T) interesting?
• Margulis (1977): G with (T), Ni ⊆ G family of finite indexsubgroups with trivial intersection =⇒ Cayley graphs of G/Ni
form expander graphs (e.g. SL3(Zpn ) for p-prime)
• Fisher-Margulis (2003): G has (T), acts smoothly on asmooth manifold via an action ρ. Then any smooth action ρ′
sufficiently close to ρ on the generators is conjugate to ρ.
• rigidity for von Neumann algebras, Popa’s deformation/rigiditytechniques
• counterexamples to Baum-Connes type conjectures(Higson-Lafforgue-Skandalis 2003): K -theory classesrepresented by Kazhdan-type projections do not lie in theimage of certain versions of the Baum-Connes assembly map
5
![Page 8: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/8.jpg)
Why is property (T) interesting?
• Margulis (1977): G with (T), Ni ⊆ G family of finite indexsubgroups with trivial intersection =⇒ Cayley graphs of G/Ni
form expander graphs (e.g. SL3(Zpn ) for p-prime)
• Fisher-Margulis (2003): G has (T), acts smoothly on asmooth manifold via an action ρ. Then any smooth action ρ′
sufficiently close to ρ on the generators is conjugate to ρ.
• rigidity for von Neumann algebras, Popa’s deformation/rigiditytechniques
• counterexamples to Baum-Connes type conjectures(Higson-Lafforgue-Skandalis 2003): K -theory classesrepresented by Kazhdan-type projections do not lie in theimage of certain versions of the Baum-Connes assembly map
5
![Page 9: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/9.jpg)
Why is property (T) interesting?
• Margulis (1977): G with (T), Ni ⊆ G family of finite indexsubgroups with trivial intersection =⇒ Cayley graphs of G/Ni
form expander graphs (e.g. SL3(Zpn ) for p-prime)
• Fisher-Margulis (2003): G has (T), acts smoothly on asmooth manifold via an action ρ. Then any smooth action ρ′
sufficiently close to ρ on the generators is conjugate to ρ.
• rigidity for von Neumann algebras, Popa’s deformation/rigiditytechniques
• counterexamples to Baum-Connes type conjectures(Higson-Lafforgue-Skandalis 2003): K -theory classesrepresented by Kazhdan-type projections do not lie in theimage of certain versions of the Baum-Connes assembly map
5
![Page 10: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/10.jpg)
2 main directions in proving (T):
- algebraic: G has rich algebraic structure + knowledge ofrepresentation theory
- Garland’s method (geometric or spectral):G acts on a complex whose links are sufficiently expanding(λ1 > 1/2)
6
![Page 11: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/11.jpg)
Question
Does Aut(Fn) or Out(Fn) have property (T)?
Neither of the earlier methods of proving (T) applies
Small values of n:
• Aut(F2) maps virtually onto Out(F2) ' GL2(Z)
• Aut(F3) maps onto Z (McCool 1989)and virtually onto F2 (Grunewald-Lubotzky 2006)
=⇒ no (T)
7
![Page 12: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/12.jpg)
Question
Does Aut(Fn) or Out(Fn) have property (T)?
Neither of the earlier methods of proving (T) applies
Small values of n:
• Aut(F2) maps virtually onto Out(F2) ' GL2(Z)
• Aut(F3) maps onto Z (McCool 1989)and virtually onto F2 (Grunewald-Lubotzky 2006)
=⇒ no (T)
7
![Page 13: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/13.jpg)
Ozawa’s characterization
Laplacian in the real group ring RG:
∆ = |S | −∑s∈S
s
=12
∑s∈S
(1 − s)(1 − s)∗ ∈ RG
Property (T) ⇔ 0 ∈ spectrum(∆) in C∗max(G) is isolated
⇔ (∆ − λI)∆ ≥ 0 in C∗max(G) for some λ > 0
Theorem (Ozawa, 2014)
(G,S) has property (T) iff for some λ > 0 and a finite collectionof ξi ∈ RG
∆2 − λ∆ =n∑
i=1
ξ∗i ξi
This is a finite-dim condition
8
![Page 14: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/14.jpg)
Ozawa’s characterization
Laplacian in the real group ring RG:
∆ = |S | −∑s∈S
s =12
∑s∈S
(1 − s)(1 − s)∗ ∈ RG
Property (T) ⇔ 0 ∈ spectrum(∆) in C∗max(G) is isolated
⇔ (∆ − λI)∆ ≥ 0 in C∗max(G) for some λ > 0
Theorem (Ozawa, 2014)
(G,S) has property (T) iff for some λ > 0 and a finite collectionof ξi ∈ RG
∆2 − λ∆ =n∑
i=1
ξ∗i ξi
This is a finite-dim condition
8
![Page 15: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/15.jpg)
Ozawa’s characterization
Laplacian in the real group ring RG:
∆ = |S | −∑s∈S
s =12
∑s∈S
(1 − s)(1 − s)∗ ∈ RG
Property (T) ⇔ 0 ∈ spectrum(∆) in C∗max(G) is isolated
⇔ (∆ − λI)∆ ≥ 0 in C∗max(G) for some λ > 0
Theorem (Ozawa, 2014)
(G,S) has property (T) iff for some λ > 0 and a finite collectionof ξi ∈ RG
∆2 − λ∆ =n∑
i=1
ξ∗i ξi
This is a finite-dim condition
8
![Page 16: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/16.jpg)
Ozawa’s characterization
Laplacian in the real group ring RG:
∆ = |S | −∑s∈S
s =12
∑s∈S
(1 − s)(1 − s)∗ ∈ RG
Property (T) ⇔ 0 ∈ spectrum(∆) in C∗max(G) is isolated
⇔ (∆ − λI)∆ ≥ 0 in C∗max(G) for some λ > 0
Theorem (Ozawa, 2014)
(G,S) has property (T) iff for some λ > 0 and a finite collectionof ξi ∈ RG
∆2 − λ∆ =n∑
i=1
ξ∗i ξi
This is a finite-dim condition
8
![Page 17: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/17.jpg)
Ozawa’s characterization
Laplacian in the real group ring RG:
∆ = |S | −∑s∈S
s =12
∑s∈S
(1 − s)(1 − s)∗ ∈ RG
Property (T) ⇔ 0 ∈ spectrum(∆) in C∗max(G) is isolated
⇔ (∆ − λI)∆ ≥ 0 in C∗max(G) for some λ > 0
Theorem (Ozawa, 2014)
(G,S) has property (T) iff for some λ > 0 and a finite collectionof ξi ∈ RG
∆2 − λ∆ =n∑
i=1
ξ∗i ξi
This is a finite-dim condition 8
![Page 18: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/18.jpg)
Property (T) quantified
If the equation is satisfied:
∆2 − λ∆ =n∑
i=1
ξ∗i ξi
then relation to Kazhdan constants:
√2λ|S |≤ κ(G,S)
we can also define the following notion
Definition
Kazhdan radius of (G,S) = smallest r > 0 such thatsupp ξi ⊆ B(e, r) for all ξi above.
9
![Page 19: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/19.jpg)
Implementation for SLn(Z)
Implementation of this strategy pioneered by Netzer-Thom (2015):
a new, computer-assisted proof of (T) for SL3(Z)
(generators: elementary matrices)
Improvement of Kazhdan constant:
' 1/1800 → ' 1/6
Later also improved Kazhdan constants for SLn(Z) for
n = 3, 4 (Fujiwara-Kabaya)
n = 3, 4, 5 (Kaluba-N.)
10
![Page 20: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/20.jpg)
Setup
We will work with SAut(Fn)
subgroup of Aut(Fn), generated by Nielsen transformations:
R±ij (sk ) =
sk s±1j if k = i
sk oth., L±ij (sk ) =
s±1j sk if k = i
sk oth.
Equivalently,SAut(Fn) = ab−1(SLn(Z))
under the map Aut(Fn)→ GLn(Z) induced by the abelianizationFn → Z
n.
SAut(Fn) has index 2 in Aut(Fn)
11
![Page 21: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/21.jpg)
Main results
Theorem (Kaluba - N. - Ozawa)
SAut(F5) has property (T) with Kazhdan constant
0.18 ≤ κ(SAut(F5))
Theorem (Kaluba - Kielak - N.)
SAut(Fn) has property (T) for n ≥ 6 with Kazhdan radius 2 andKazhdan constant√
0.138(n − 2)
6(n2 − n)≤ κ(SAut(Fn)).
12
![Page 22: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/22.jpg)
Certifying positivity viasemidefinite programming
![Page 23: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/23.jpg)
η ∈ RG positive if η is a sum of squares:
η =k∑
i=1
ξ∗i ξi
where supp ξi ⊆ E - finite subset of G. (here always E = B(e, 2))
⇐⇒
there is a positive definite E ×E matrix P such that for b = [g1, . . . , gn]gi∈E
η = bPbT = bQQT bT = (bQ)(bQ)T
i-th column of Q = coefficients of ξi
13
![Page 24: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/24.jpg)
Using the solver
To check if η is positve we can use a semidefinite solver to performconvex optimization over positive definite matrices:
find P ∈ ME×E such that:
η = bPbT
positive semi-definite
14
![Page 25: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/25.jpg)
Assume now that a computer has found a solution P - we obtain
η '∑
ξ∗i ξi
this is not a precise solution
However this can be improved if the error is small in a certainsense and η belongs to the augmentation ideal IG
Lemma (Ozawa, Netzer-Thom)∆ is an order unit in IG: if η = η∗ ∈ IG then
η + R∆ =∑
ξ∗i ξi
for all sufficiently large R ≥ 0.
Moreover, R = 22r−2‖η‖1 is sufficient, where supp η ⊆ B(e, 2r).
15
![Page 26: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/26.jpg)
Assume now that a computer has found a solution P - we obtain
η '∑
ξ∗i ξi
this is not a precise solution
However this can be improved if the error is small in a certainsense and η belongs to the augmentation ideal IG
Lemma (Ozawa, Netzer-Thom)∆ is an order unit in IG: if η = η∗ ∈ IG then
η + R∆ =∑
ξ∗i ξi
for all sufficiently large R ≥ 0.
Moreover, R = 22r−2‖η‖1 is sufficient, where supp η ⊆ B(e, 2r).
15
![Page 27: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/27.jpg)
With this in mind we can try to improve our search of P to showthat η is “strictly positive”:
maximize λ ≥ 0 under the conditions
η − λ∆ = bPbT
P ∈ ME×E positive semi-definite
16
![Page 28: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/28.jpg)
Assume now that numerically on B(e, 2r)
η − λ∆ ' bPbT
and P = QQT .
Correct Q to Q , where columns of Q sum up to 0.∥∥∥∥∥η − λ∆ − bQ QT
bT∥∥∥∥∥
1≤ ε
Then for R = ε22r−2 or larger:
η − λ∆ + R∆︸ ︷︷ ︸(λ−R)∆
= bQ QT
bT︸ ︷︷ ︸∑η∗i ηi
+ (η − λ∆ − bQ QT
bT + R∆)︸ ︷︷ ︸≥0 by lemma
If λ − R > 0 then η − (λ − R)∆ ≥ 0 and in particular, ξ ≥ 0.
17
![Page 29: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/29.jpg)
Assume now that numerically on B(e, 2r)
η − λ∆ ' bPbT
and P = QQT .
Correct Q to Q , where columns of Q sum up to 0.∥∥∥∥∥η − λ∆ − bQ QT
bT∥∥∥∥∥
1≤ ε
Then for R = ε22r−2 or larger:
η − λ∆ + R∆︸ ︷︷ ︸(λ−R)∆
= bQ QT
bT︸ ︷︷ ︸∑η∗i ηi
+ (η − λ∆ − bQ QT
bT + R∆)︸ ︷︷ ︸≥0 by lemma
If λ − R > 0 then η − (λ − R)∆ ≥ 0 and in particular, ξ ≥ 0.
17
![Page 30: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/30.jpg)
Assume now that numerically on B(e, 2r)
η − λ∆ ' bPbT
and P = QQT .
Correct Q to Q , where columns of Q sum up to 0.∥∥∥∥∥η − λ∆ − bQ QT
bT∥∥∥∥∥
1≤ ε
Then for R = ε22r−2 or larger:
η − λ∆ + R∆︸ ︷︷ ︸(λ−R)∆
= bQ QT
bT︸ ︷︷ ︸∑η∗i ηi
+ (η − λ∆ − bQ QT
bT + R∆)︸ ︷︷ ︸≥0 by lemma
If λ − R > 0 then η − (λ − R)∆ ≥ 0 and in particular, ξ ≥ 0.
17
![Page 31: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/31.jpg)
Important:The `1-norm is computed in interval artihmetic.
This gives a mathematically rigorous proof of positivity of η.
18
![Page 32: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/32.jpg)
We want to use this strategy in SAut(F5).
Problem #1:|B(e, 2)| = 4 641
P has 10 771 761 variables - too large for a solver to handle
19
![Page 33: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/33.jpg)
Symmetrization
We reduce the number of variables via symmetries
Γ = Z2 o Symn acts on generators of Fn by inversions andpermutations
Γ also acts on Aut(F5) by conjugation
preserves SAut(Fn) and ∆2 − λ∆
There is an induced action onME - E × E matrices
LemmaIf there is a solution P then there is also a Γ-invariant solution P.
20
![Page 34: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/34.jpg)
Symmetrization
Problem #2: solvers do not work with Γ-invariant matrices...
ρE = permutation representation of Γ on `2(E)
ρE =⊕π∈Γ
mππ
where mπ=multiplicity of π
Theorem (Wedderburn decomposition)
C∗(ρE) '⊕π∈Γ
Mdim π ⊗ 1mπ
Thus: MΓE ' (C∗(ρE))′ '
⊕π∈Γ 1dim π ⊗Mmπ
We need this isomorphism to be explicit - this can be done using asystem of minimal projections provided by representation theory
21
![Page 35: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/35.jpg)
Symmetrization
Problem #2: solvers do not work with Γ-invariant matrices...
ρE = permutation representation of Γ on `2(E)
ρE =⊕π∈Γ
mππ
where mπ=multiplicity of π
Theorem (Wedderburn decomposition)
C∗(ρE) '⊕π∈Γ
Mdim π ⊗ 1mπ
Thus: MΓE ' (C∗(ρE))′ '
⊕π∈Γ 1dim π ⊗Mmπ
We need this isomorphism to be explicit - this can be done using asystem of minimal projections provided by representation theory 21
![Page 36: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/36.jpg)
? •
? • ◦
◦ ? ◦
• • ◦
◦ ? •
? ?
−→
[∗
] ∗ ∗∗ ∗
. . . [
∗]
The reduction for SAut(F5):
from 10 771 761 variables→ 13 232 variables in 36 blocks
22
![Page 37: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/37.jpg)
Theorem (Kaluba-N.-Ozawa)
SAut(F5) has (T) with Kazhdan constant ≥ 0.18027.
Proof.
Find sum of squares decomposition for ∆2 − λ∆
on the ball of radius 2Data from solver: P and λ = 1.3
8.30 · 10−6 ≤
∥∥∥∥∆2 − λ∆ −∑
ξ∗i ξi
∥∥∥∥1≤ 8.41 · 10−6
=⇒ R = 8.41 · 10−6 · 24−2 suffices
λ − R = 1.2999 > 0
� 23
![Page 38: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/38.jpg)
In the case of SLn(Z):
(T) for SL3(Z) =⇒ (T) for SLn(Z) for all higher n
Is a similar strategy possible here?
24
![Page 39: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/39.jpg)
Property (T) for Aut(Fn), n ≥ 6
![Page 40: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/40.jpg)
Gn with generating set Sn will denote either one of the families
• SAut(Fn) generatred by Nielsen transformations R±i,j , L±i,j
• SLn(Z) generated by elementary matrices E±i,j
25
![Page 41: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/41.jpg)
Gn form a tower via the inclusions of
Fn ⊆ Fn+1 and Zn ⊆ Zn+1
obtained by adding the next generator
Cn - (n − 1)-simplex on {1, . . . , n}
En set of edges of Cn = unoriented pairs e = {i, j}
Altn acts on edges: σ(e) = σ({i, j}) = {σ(i), σ(j)}
Mapln : Sn → Cn,
R±ij , L±ij 7→ {i, j}, E±ij 7→ {i, j}
26
![Page 42: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/42.jpg)
Gn form a tower via the inclusions of
Fn ⊆ Fn+1 and Zn ⊆ Zn+1
obtained by adding the next generator
Cn - (n − 1)-simplex on {1, . . . , n}
En set of edges of Cn = unoriented pairs e = {i, j}
Altn acts on edges: σ(e) = σ({i, j}) = {σ(i), σ(j)}
Mapln : Sn → Cn,
R±ij , L±ij 7→ {i, j}, E±ij 7→ {i, j}
26
![Page 43: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/43.jpg)
In other words, a copy of G2 is attached at each edge.
Given 2 edges they can
1. coincide2. be adjacent (share a vertex)3. be opposite (share no vertices) - corresponding copies of G2
commute
∆n ∈ RGn - Laplacian of Gn
For an edge e = {i, j} let Se = {s ∈ Sn : ln(s) = e} and
∆e = |Se | −∑t∈Se
t
is the Laplacian of the copy of G2 attached to e
We have σ(∆e) = ∆σ(e) for any σ ∈ Altn
27
![Page 44: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/44.jpg)
In other words, a copy of G2 is attached at each edge.
Given 2 edges they can
1. coincide2. be adjacent (share a vertex)3. be opposite (share no vertices) - corresponding copies of G2
commute
∆n ∈ RGn - Laplacian of Gn
For an edge e = {i, j} let Se = {s ∈ Sn : ln(s) = e} and
∆e = |Se | −∑t∈Se
t
is the Laplacian of the copy of G2 attached to e
We have σ(∆e) = ∆σ(e) for any σ ∈ Altn 27
![Page 45: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/45.jpg)
0 1
2
3
Altn permutes vertices
e = {1, 2}
SAut(F2) or SL2(Z)
∆e
f = {0, 3}
SAut(F2) or SL2(Z)
∆f
28
![Page 46: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/46.jpg)
0 1
2
3
Altn permutes vertices
e = {1, 2}
SAut(F2) or SL2(Z)
∆e
f = {0, 3}
SAut(F2) or SL2(Z)
∆f
28
![Page 47: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/47.jpg)
0 1
2
3
Altn permutes vertices
e = {1, 2}
SAut(F2) or SL2(Z)
∆e
f = {0, 3}
SAut(F2) or SL2(Z)
∆f
28
![Page 48: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/48.jpg)
0 1
2
3
Altn permutes vertices
e = {1, 2}
SAut(F2) or SL2(Z)
∆e
f = {0, 3}
SAut(F2) or SL2(Z)
∆f
28
![Page 49: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/49.jpg)
0 1
2
3
Altn permutes vertices
e = {1, 2}
SAut(F2) or SL2(Z)
∆e
f = {0, 3}
SAut(F2) or SL2(Z)
∆f
28
![Page 50: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/50.jpg)
∆e are building blocks of the Laplacians in Gn:
Lemma
∆n =∑e∈En
∆e
=1
(n − 2)!
∑σ∈Altn
σ(∆e).
Proposition (Stability for ∆n)For m ≥ n ≥ 3 we have∑
σ∈Altm
σ(∆n) =|Altn |
(n − 2)!
∑σ∈Altn
σ(∆2) =
(n2
)(m − 2)!∆m.
Main step in proving property (T) for Gn:a “stable” decomposition of ∆2
n − λ∆n
29
![Page 51: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/51.jpg)
∆e are building blocks of the Laplacians in Gn:
Lemma
∆n =∑e∈En
∆e =1
(n − 2)!
∑σ∈Altn
σ(∆e).
Proposition (Stability for ∆n)For m ≥ n ≥ 3 we have∑
σ∈Altm
σ(∆n) =|Altn |
(n − 2)!
∑σ∈Altn
σ(∆2) =
(n2
)(m − 2)!∆m.
Main step in proving property (T) for Gn:a “stable” decomposition of ∆2
n − λ∆n
29
![Page 52: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/52.jpg)
∆e are building blocks of the Laplacians in Gn:
Lemma
∆n =∑e∈En
∆e =1
(n − 2)!
∑σ∈Altn
σ(∆e).
Proposition (Stability for ∆n)For m ≥ n ≥ 3 we have∑
σ∈Altm
σ(∆n) =|Altn |
(n − 2)!
∑σ∈Altn
σ(∆2) =
(n2
)(m − 2)!∆m.
Main step in proving property (T) for Gn:a “stable” decomposition of ∆2
n − λ∆n
29
![Page 53: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/53.jpg)
Define three elements of RGn:
1. Sqn =∑
e∈En ∆2e
2. Adjn =∑
e∈En
∑f∈Adjn(e) ∆e∆f
3. Opn =∑
e∈En
∑f∈Opn(e) ∆e∆f
Lemma
Sqn + Adjn + Opn = ∆2n
30
![Page 54: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/54.jpg)
LemmaThe elements Sqn and Opn are sums of squares.
Proof.Obvious for Sqn.
For Opn:
∆e =12
∑t∈Se
(1 − t)∗(1 − t)
e, f opposite edges then the generators associated to themcommute and ∆e∆f can be rewritten as a sum of squares using
(1 − t)∗(1 − t)(1 − s)∗(1 − s) =((1 − t)(1 − s)
)∗((1 − t)(1 − s)
)�
31
![Page 55: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/55.jpg)
LemmaThe elements Sqn and Opn are sums of squares.
Proof.Obvious for Sqn.
For Opn:
∆e =12
∑t∈Se
(1 − t)∗(1 − t)
e, f opposite edges then the generators associated to themcommute and ∆e∆f can be rewritten as a sum of squares using
(1 − t)∗(1 − t)(1 − s)∗(1 − s) =((1 − t)(1 − s)
)∗((1 − t)(1 − s)
)�
31
![Page 56: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/56.jpg)
Stability for Adj and Op:
PropositionFor m ≥ n ≥ 3 we have∑
σ∈Altm
σ (Adjn) =
(12
n(n − 1)(n − 2)(m − 3)!
)Adjm
PropositionFor m ≥ n ≥ 4 we have∑
σ∈Altm
σ (Opn) =
(2(n2
) (n − 2
n
)(m − 4)!
)Opm
32
![Page 57: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/57.jpg)
Stability for Adj and Op:
PropositionFor m ≥ n ≥ 3 we have∑
σ∈Altm
σ (Adjn) =
(12
n(n − 1)(n − 2)(m − 3)!
)Adjm
PropositionFor m ≥ n ≥ 4 we have∑
σ∈Altm
σ (Opn) =
(2(n2
) (n − 2
n
)(m − 4)!
)Opm
32
![Page 58: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/58.jpg)
The following allows to prove (T) for SLn(Z) for n ≥ 3 and forSAut(Fn), n ≥ 7.
TheoremLet n ≥ 3 and
Adjn +k Opn −λ∆n =∑
ξ∗i ξi
where supp ξi ⊂ B(e,R), for some k ≥ 0, λ ≥ 0.
Then Gm has property (T) for every m ≥ n such that
k(n − 3) ≤ m − 3.
Moreover, the Kazhdan radius is bounded above by R.
33
![Page 59: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/59.jpg)
Proof: Rewrite ∆2m − λ∆m using stability of Op, Adj and ∆.
Eventually arrive at:
∆2m −
λ(m − 2)
n − 2∆m = Sqm + Adjm + Opm
= Sqm +
(1 −
k(n − 3)
m − 3
)Opm +
2n(n − 1)(n − 2)(m − 3)!
∑σ∈Altm
σ (Adjn +k Opn −λ∆n)
When 1 −k(n − 3)
m − 3≥ 0 we obtain the claim. �
34
![Page 60: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/60.jpg)
Results
![Page 61: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/61.jpg)
Theorem
SAut(Fn) has (T) for n ≥ 6, with Kazhdan radius 2 and Kazhdanconstant estimate √
0.138(n − 2)
6(n2 − n)≤ κ(SAutn).
Proof.The case n ≥ 7:
Adj5 +2 Op5 −0.138∆5
certified positive on the ball of radius 2. �
The case n = 6 needs a different computation.The case n = 5 so far can only be proven directly.
Remark: Currently we also can certify Adj4 +100 Op4 −0.1∆ inSAut(F4) proving (T) for n ≥ 103.
35
![Page 62: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/62.jpg)
Theorem
SAut(Fn) has (T) for n ≥ 6, with Kazhdan radius 2 and Kazhdanconstant estimate √
0.138(n − 2)
6(n2 − n)≤ κ(SAutn).
Proof.The case n ≥ 7:
Adj5 +2 Op5 −0.138∆5
certified positive on the ball of radius 2. �
The case n = 6 needs a different computation.The case n = 5 so far can only be proven directly.
Remark: Currently we also can certify Adj4 +100 Op4 −0.1∆ inSAut(F4) proving (T) for n ≥ 103.
35
![Page 63: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/63.jpg)
Theorem
SLn(Z) has (T) for n ≥ 3, with Kazhdan radius 2 and Kazhdanconstant estimate√
0.157999(n − 2)
n2 − n≤ κ(SLn(Z)).
Proof.
Adj3 −0.157999∆3
certified positive on the ball of radius 2. �
This gives a new estimate on the Kazhdan constants of SLn(Z).
36
![Page 64: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/64.jpg)
We obtain an even better estimate by certifying positivity of
Adj5 +1.5 Op5 −1.5∆5
in R SL5(Z):
√0.5(n − 2)
n2 − n≤ κ(SLn(Z)) for n ≥ 6.
Previously known bounds:
(Kassabov 2005)1
42√
n + 860≤ κ(SLn(Z)) ≤
√2n
(Zuk 1999)
Asymptotically, the new lower bound is 1/2 of the upper bound:
Zuk’s upper boundour lower bound
= 2
√n − 1n − 2
−→ 2 (n ≥ 6)
37
![Page 65: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/65.jpg)
We obtain an even better estimate by certifying positivity of
Adj5 +1.5 Op5 −1.5∆5
in R SL5(Z):
√0.5(n − 2)
n2 − n≤ κ(SLn(Z)) for n ≥ 6.
Previously known bounds:
(Kassabov 2005)1
42√
n + 860≤ κ(SLn(Z)) ≤
√2n
(Zuk 1999)
Asymptotically, the new lower bound is 1/2 of the upper bound:
Zuk’s upper boundour lower bound
= 2
√n − 1n − 2
−→ 2 (n ≥ 6)
37
![Page 66: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/66.jpg)
Some applications
Product Replacement Algorithm generates random elements infinite groups
Lubotzky and Pak in 2001 showed that property (T) for Aut(Fn)
explains the fast performance of the Product ReplacementAlgorithm
=⇒ now proven
38
![Page 67: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/67.jpg)
Some applications
Question (Lubotzky 1994)Is there a sequence of finite groups such that their Cayley graphsare expanders or not for different generating sets (uniformlybounded)?
Kassabov 2003: yes for a subsequence of symmetric groups
Gilman 1977: Aut(Fn) for n ≥ 3 residually alternating
=⇒ sequences of alternating groups with Aut(Fn) generators areexpanders
This gives an alternative answer to Lubotzky’s questions withexplicit generating sets
39
![Page 68: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/68.jpg)
Some applications
Question (Lubotzky 1994)Is there a sequence of finite groups such that their Cayley graphsare expanders or not for different generating sets (uniformlybounded)?
Kassabov 2003: yes for a subsequence of symmetric groups
Gilman 1977: Aut(Fn) for n ≥ 3 residually alternating
=⇒ sequences of alternating groups with Aut(Fn) generators areexpanders
This gives an alternative answer to Lubotzky’s questions withexplicit generating sets
39
![Page 69: On property (T) for Aut(Fn)math.ecnu.edu.cn/RCOA/events/Special Week on Operator... · 2021. 6. 8. · Why is property (T) interesting? ‹ Margulis (1977): G with (T), Ni G family](https://reader036.fdocument.pub/reader036/viewer/2022071605/6140f53883382e045471c8a0/html5/thumbnails/69.jpg)
For Aut(F4) property (T) was confirmed recently by M. Nitsche(arxiv 2020)
With Uri Bader we generalized Ozawa’s chatacterization to highercohomology.
40