Noncommutative de Leeuw theorems
Éric Ricard
Laboratoire de Mathématiques Nicolas Oresme Université de Caen Normandie
June, 2016
Joint work with M. Caspers, J. Parcet and M. Perrin
Motivation
Work by M. Junge, T. Mei, J. Parcet on transference of smooth multipliers To decide if some Idempotent multipliers are bounded on R
Ω
slope/∈Q
γ 6
-
De Leeuw's Theorems
General setting :H ⊂Rd =Rcd subgroup,H =Zandd =1 to simplify.
Question : Relating Lp-multipliers with symbols m:Rd →C: Tcm :Lp(Rcd)→Lp(Rcd)
Tdmf(ξ) = m(ξ)fb(ξ), Tmf =
Z
Rn
m(ξ)fb(ξ)λ(ξ)dξ, to H⊂Rd.
Recall that Rd Hb.
De Leeuw's Theorems
The restriction theorem (1965)
H ⊂Rd closed subgroup, m:Rd →Ccontinuous, let m˜ =m|H then
Tm˜ :Lp(H)b →Lp(H)b 6
Tm:Lp(Rcd)→Lp(Rcd) .
Basic case : Transference of multipliers on Lp(R) toLp(T) m continuous can be weakened :
Riesz Transforms on Lp(R) ⇒Riesz Transforms on Lp(T)
De Leeuw's Theorems
The periodization theorem
H ⊂Rd closed subgroup, m:Rd/H→C, letmp :Rd →Cwhere mp =m◦q with q :Rd Rd/H then
Tmp :Lp(Rcd)→Lp(Rcd) 6
Tm:Lp(R\d/H)→Lp(R\d/H) .
Basic case : Transference of convolutors on `p(Z) toLp(R)
De Leeuw's Theorems
Let Rddisc beRd with the discrete topology as a LCA group.
R[ddisc =Rdbohr
The compactication theorem Let m:Rd →C continuous then
Tm:Lp(Rdbohr)→Lp(Rdbohr) =
Tm:Lp(Rcd)→Lp(Rcd) .
Compactication is somehow the strongest result.
It implies the restriction thm as it is clear for discreteG. All these theorems are about transference.
Ideas behind the proof of compactication
Ris commutative
Ris well approximated by discrete subgroups : R= ∪
k>02−kZ.
Ris amenable : existence of Folner sets (intervals) Convolution with Gaussians γ(x) = √1
2πe−x22,γ = 1γ
x
Good approximations on both frequency and space sides.
Nice semi-group of convolution
Extensions
Saeki 1970
Those 3 theorems hold more generally for LCA groups.
Proof : De Leeuw thms + structure theory of LCA groups.
There are also related works by Igari
There is another approach to compactication :
Extension of multipliers from a closed subgroup to the whole group
Extension of multipliers Let F :R→R,F =1[−1
2,12]∗1[−1
2,12] be a Fejer-type kernel Fnd =F⊗d:Rd →R
Jodeit (1969)
Let m:Zd →C and letm˜ :Rd →Cbem∗Fd, then
Tm˜ :Lp(Rd)→Lp(Rd) 6Cd
Tm :Lp(Td)→Lp(Td) . Let m:Rd →C and letm˜ be its periodic extension from[−π, π], then
Tm :`p(Zd)→`p(Zd) 6Cd
Tm˜ :Lp(Rd)→Lp(Rd) . He also treated restrictions.
This gives the hard way in the compactication theorem.
G LCA, H⊂G be closed.
Figà-Talamanca and Gaudry (1970)
AssumeH is discrete, letm:H→C and letm˜ =m∗F ∗F for a Fejer-type kernel, then
Tm˜ :Lp( ˆG)→Lp( ˆG) 6
Tm :Lp(H)b →Lp(H)b . More satisfactory : Better constant butF ∗F instead of F. Cowling (1975)
AssumeH is closed, let m:H→Cand letm˜ =m∗F ∗F for a Fejer-type kernel, then
Tm˜ :Lp( ˆG)→Lp( ˆG) 6
Tm :Lp(H)b →Lp(H)b . Use of disintegration and structure theories for LCA groups.
He also looked at periodization.
Noncommutative G?
→ Lots of works on convolutors on Lp(G) andLp(G/H) (Doodley, Gaudry, Derighetti,...)
→ Lots of works on (vector valued) transference (Coiman-Weiss,...)
→ Another direction : Noncommutative analysis How to dene Lp( ˆG)when G is non abelian ? For p=∞
L∞( ˆG)⊂B(L2( ˆG))'B(L2(G))
where L∞( ˆG) : convolution operator onB(L2(G)) G l.c. group : left representation λ:G →B(`2G)
L(G) =λ(G)00=Span{λ(g) ;g ∈G}w Noncommutative Lp-space
Non commutative analysis
M ⊂B(`2) von Neumann algebra with a normal faithful seminite traceτ (M, τ) ↔ (L∞(Ω), µ)
τ(xy) =τ(yx) ↔ fg =gf normal ↔ monotone cv thm faithful ↔ full support
seminite ↔ σ-nite
L0(M, τ) ↔ L0(Ω, µ)
via functional calculus on L0(M, τ),|y|= (y∗y)12 Lp(M, τ) ={x∈L0(M, τ)| kxkpp =τ(|x|p)<∞}
Not all vN admit a trace.
If not, equivalent denitions by Connes and Haagerup
Noncommutative G?
We denie Lp( ˆG) =Lp(L(G))for G LC.
Denition of the trace τ via FT (G unimodular) x =R
fgλ(g)dg,f ∈Cc(G) :τ(x) =fe. Notion of multiplier with symbolφ:G →C:
Mφ( Z
fλ(g)dg) = Z
φgfgλ(g)dg
Restriction, periodization, compactication make sense in this setting.
Noncommutative G?
Pb : What is the best notion of boundedness ?
→ Boundedness of Mφ on Lp
→ Complete Boundedness : boundedness of Mφ×1 on G ×G0 for anyG0. If G andG0 are abelian :Mφ×1 bounded⇔Mφ bounded
Arhancet / Dutta-Mohanty-Tewari
For any innite LCA group G and 1<p 6=2<∞, there is an Lp-Fourier multiplier which is not completely bounded.
Extension of a result by Pisier for compact groups using transference.
Cb norm results
For m:G →C,Tm :Lp(L(G))→Lp(L(G))Fourier multiplier Sm :Sp(L2(G))→Sp(L2(G))the equivariant Schur multiplier φ(s,t) =m(s−1t).
Neuwirth-R (G discrete), Caspers-de la Salle (G LC) AssumeG is amenable andm is bounded
Tm:Lp(L(G))→Lp(L(G))kcb=
Sm:Sp(L2(G))→Sp(L2(G))kcb
It suces to look at restriction, periodization, compactication for Schur multipliers.
Using basic results by Haagerup, Laorgue-de la Salle An easy compactication theorem
AssumeG is amenable, 16p6∞, letm:G →C continuous then
Tm:Lp(L(Gdisc))→Lp(L(Gdisc)) cb =
Tm:Lp(L(G))→Lp(L(G)) cb.
An easy restriction theorem
AssumeH ⊂G with compatible modular functions andH amenable, 16p 6∞, letm:G →Ccontinuous, m˜ =m|H then
Tm˜ :Lp(L(H))→Lp(L(H)) cb 6
Tm :Lp(L(G))→Lp(L(G)) cb. When p =1,∞, one can removeH amenable (Bo»ejko-Fendler).
Similarly there are is an easy periodization theorem.
An easy Jodeit's theorem
AssumeG is amenable andH ⊂G be a lattice with fdX, 16p 6∞, let m:H →C, putm˜ =1X ∗m∗1X then
Tm˜ :Lp(L(G))→Lp(L(G)) cb=
Tm :Lp(L(H))→Lp(L(H)) cb.
1X ∗1X =F → cst 1 : better than Jodeit's result for Z⊂Rbut cb.
Question : What is the right constant in Jodeit's thm (not cb) ?
Drawbacks :
Amenability of G : hard to get rid of it
cb assumption, one does not recover the classical results general result for bounded maps⇒ cbversion Basic idea's :
to adapt de Leeuw's approach to use other transferences if possible to relateLp(L(H))andLp(L(G))
The basic restriction thm
AssumeG is LC andH⊂G amenable discrete with(∆G)|H =1, let m:G →Ccontinuous, m˜=m|H
Tm˜ :Lp(L(H))→Lp(L(H)) 6
Tm :Lp(L(G))→Lp(L(G)) .
Idea of the proof :
To embed Lp(L(H))approximately in Lp(L(G))in a way that intertwines multipliers
Lp(L(H)) −→φi Lp(L(G))
Tm˜ ↓ ↓ Tm
Lp(L(H)) −→φi Lp(L(G))
Take V a small symmetric neighborhood of 1∈G y = √1
µ(V)λ(1V)∈L2(L(G))
φpy :Lp(L(H))→Lp(L(G)) ; λ(h)7→λ(h)u|y|2/p
φpy is a contraction by interpolation if V is small enough.
One would think of limV→{e}kφpy(f)kp=kfkp. Obvious forp =2
UsingL2-duality → obvious whenG is commutative
To get it we need that V is almost invariant by conjugation byH
∀h∈H, µ(hVh−1∆V)/µ(V)→0 If this is true G is [SAIN]H
H amenable ⇒G is[SAIN]H
The commutation relation with Tm is a more delicate technical issue.
→ Obvious for p=2 : good space location of y inL2.
→ It suces to do it for ucpTm using continuity ofm.
What is the support of |y|t?
De Leeuw→ multiplication withγ instead of y : nice convolution semi-group : γt =γt
good approximations of identity
One needs a local control on approximations of identity for dierent values of p.
Almost multiplicative maps
Multiplicative domains
Let Abe a C∗-algebra, T :A→Abe ucp and x =x∗ ∈A, then T(x2) =T(x)2 ⇒ T(f(x)) =f(T(x)), f ∈C(σ(x))
Using an ultraproduct argument if kxk61 and f ∈C([−1,1]) kT(x2)−T(x)2k6 ⇒ kT(f(x))−f(T(x))k6δ
If A=C([0,1]), this is a strong quantitative Korovkin theorem.
AssumeT : (M, τ)→(M, τ) is ucp trace preserving Then T :Lp→Lp
What can we say if x∈Lp?
Almost multiplicativity on Lp
Let x∈L+p andT :M →M ucpτ-preserving then kT(x)−T(√
x)2k2p6 1
2kT(x2)−T(x)2kp. Local approximations of identity
Let y ∈L2 with y =u|y|andT :M →M ucpτ-preserving then kT(u|y|θ)−u|y|θk2
θ 6CkT(y)−yk
θ
24kyk3
θ
24 . This gives the commutation relation !
No easy ultraproduct argument (type III)
More elaborated versions Recall De Leeuw's idea R= ∪
k>02−kZ.
We say that G is ADS if there is are latticesΓi ⊂G with fdXi shrinking to {e}.
Examples : LCA, Heisenberg groups, Nilpotent matricial groups.
The restriction thm
AssumeG is LC andH⊂G with(∆G)|H =1 and H∈ADS, G ∈[SAIN]H, letm:G →C continuous,m˜ =m|H for 16p6∞:
Tm˜ :Lp(L(H))→Lp(L(H)) 6
Tm :Lp(L(G))→Lp(L(G)) .
The compactication theorem
Let 16p6∞, letm:G →C continuous, If G is ADS
Tm :Lp(L(G))→Lp(L(G)) 6
Tm :Lp(L(Gdisc))→Lp(L(Gdisc)) , If Gdisc is amenable
Tm :Lp(L(Gdisc))→Lp(L(Gdisc)) 6
Tm :Lp(L(G))→Lp(L(G)) . There is =for LCA, Heisenberg, Nilpotent triangular matricial groups.
One can also get some periodization results.