A gap-labelling conjecture for the magnetic

A gap-labelling conjecture for the magnetic Schrödinger
operator on euclidean space
KK-theory, Gauge Theory and Topological Phases
Lorentz center-Leiden
March 2017
Moulay Tahar BENAMEUR
Institut Montpellierain Alexander Grothendieck
Statement of the conjecture
Conjecture
The range of the K -theory of the Bellissard C ∗ -algebra for the
magnetic Schrödinger operator is contained in Λ(Σ, Θ, µ) where:
X
Λ(Σ, Θ, µ) =
Pf(ΘI )ZI [µ].
0≤|I|≤p
joint work with:
• Varghese Mathai (University of Adelaide).
M.-T. Benameur and V. Mathai,
Gap-labelling conjecture with non-zero magnetic field.
[1508.01064]
+ ongoing research
Outline of talk
1
Magnetic Schrödinger operators
2
The conjecture
3
A twisted measured index theorem for laminations
4
What is known about the conjecture.
5
Proof in the 3D case and comments about the general
case.
Magnetic Schrödinger operators
- Consider Euclidean space Rp equipped with its usual
P
Riemannian metric pj=1 dxj2 .
P
- the magnetic field is B = 12 j,k Θjk dxj ∧ dxk = 21 dx t Θdx.
Main example: constant skew-symmetric matrix Θ (p × p).
-Pick a 1-form η such that dη = B. η is a magnetic potential,this
defines the connection ∇ = d + iη on the trivial line bundle.
Using the Riemannian metric the Hamiltonian of an electron in
this field is given by
H=
1 †
∇ ∇ + V = Hη + V ,
2
with V smooth bounded real-valued.
acting on L2 (Rp )
Magnetic Schrödinger operators
Define a unitary T σ of Zp (and Rp ) on L2 (Rp ) as follows.
Uγ (f )(x) = f (γ −1 x) = f (x − γ)
(1)
Sγ (f )(x) = exp(−iψγ (x))f (x)
(2)
Tγσ = Uγ ◦ Sγ .
(3)
where ψγ satisfies γ ∗ η − η = dψγ and ψγ (0) = 0+compatibility.
One has Tγσ Hη = Hη Tγσ . The magnetic translations Tγσ satisfy
Teσ = Id and Tγσ1 Tγσ2 (f )(x) = σ(x; γ1 , γ2 )Tγσ1 γ2 (f )(x),
where σ(x; γ1 , γ2 ) satisfies a groupoid cocycle condition.
If B is constant then σ(γ1 , γ2 ) = exp (−iψγ1 (γ2 )) is a multiplier
on Zd :
1
σ(γ, e) = σ(e, γ) = 1 for all γ ∈ Zd ;
2
σ(γ1 , γ2 )σ(γ1 γ2 , γ3 ) = σ(γ1 , γ2 γ3 )σ(γ2 , γ3 ) for γj ∈ Zd .
The Bellissard gap-labelling theorem
The C ∗ -algebra of observables is a twisted groupoid algebra
(Bellissard). We are interested in the trace-range of its
K -theory. Too hard in general! Assume from now on
B constant.
∃ a Cantor complete transversal Σ so that holonomy =
minimal action of Zp , preserving a probability measure µ.
Then the C ∗ -algebra of observables is Morita equivalent to the
twisted discrete crossed product algebra A = C(Σ) oσ Zd .
Twisted crossed product algebra
Recall for a, b ∈ A0 = Cc (Σ × Zd ):
Product:
P
ab(ω, γ) = γ 0 ∈Zd a(ω, γ 0 )b(γ 0−1 ω, γ − γ 0 )σ(γ − γ 0 , γ 0 );
Adjoint: a∗ (ω, γ) = a(γ −1 ω, −γ)σ(−γ, γ);
Regular representation for a ∈ A0 and ψ ∈ `2 (Zd ):
P
πω (a)ψ(γ) = γ 0 ∈Zd a(γ 0−1 ω, γ − γ 0 )ψ(γ 0 )σ(γ − γ 0 , γ 0 );
Norm: ||a|| = supω∈Σ ||πω (a)||;
The twisted crossed product C ∗ -algebra is then: A = A0
||·||
For f C0 on spec(H), f (H) ∈ C(Σ) oσ Zd ⊗ K.
P
Recall the `2 trace τ µ ( fγ δγ ) = µ(fe ).
Theorem
(Bellissard) The spectral gap-labels for the magnetic
Schrödinger operator lie in the image under the trace τ µ of
K0 (C(Σ) oσ Zd ).
The range of
τ∗µ : K0 (C(Σ) oσ Zd ) −→ R,
is the group we are interested in here.
It will be our “magnetic gap-labelling group”.
Bellissard’s gap-labelling conjecture (B = 0)
Here gap-labels are expected to live in the frequencies group
Z[µ] := µ(K 0 (Σ)). This is the Bellissard conjecture.
Conjecture
Under the previous assumptions, the gap-labels group is
contained in the frequency module Z[µ].
τ∗µ (K0 (C(Σ) o Zd )) ⊂ µ(C(Σ, Z)) = Z[µ].
Many works in the 90’s: conjecture is true for d = 1, 2 and
3. Bellissard, Contensou, Kellendonk, Legrand, ...
For general d, 3 differents proofs were given in
J. Bellissard, R. Benedetti, J-M. Gambaudo;
J. Kaminker, I. Putnam;
M. B., H. Oyono-Oyono.
Comment on the gap-labelling conjecture
They used that the range of the top Chern character
chp : K p (X ) → H p (X , Q) lies in H p (X , Z).
This is true for low dimensions d = 1, 2, 3.
For higher dimensions, this is still true when the
cohomology of the quasi-crystal is torsion-free
(B.-Oyono-Oyono).
In work in progress with H. Oyono-Oyono, we started a
new approach to integrality.
Back to the magnetic gap-labelling conjecture - Goal
We need a candidate for the freqencies group when B 6= 0?
The magnetic gap-labelling conjecture concerns finding an
optimal subgroup Λ(Σ, Θ, µ) of R such that
Λ(Σ, Θ, µ) is computable and coincides with Z[µ] for B = 0.
The range of the trace lies inside Λ(Σ, Θ, µ), i.e.
τµ (K0 (C(Σ) oσ Zd )) ⊂ Λ(Σ, Θ, µ).
√
P
where σ(γ, γ 0 ) = exp(π −1 Θjk γj γk0 )
Statement of the conjecture
Let Λ[dx] = Λ[dx1 , . . . , dxd ] denote the exterior algebra with
generators dx1 , . . . , dxd . Given a skew-symmetric matrix Θ,
recall Pf(Θ) is given when d = 2m by
m
1 1 t
= Pf(Θ)dx1 ∧ · · · dxd .
dx Θdx
m! 2
1
In general, e 2 dx
t Θdx
can be expressed in terms of the Pfaffians:
Mathai-Quillen identity
1
e 2 dx
t Θdx
=
X
Pf(ΘI )dxI
I
where I runs over subsets of {1, . . . , p} with an even number of
elements, and ΘI = (Θij )i,j∈I .
Statement of the conjecture
Recall the subgroup Z[µ] generated by µ-measures of clopens.
Let I be an ordered subset of {1, . . . , d}, and let C(Σ, Z)ZI c
c
denote the coinvariants under the subgroup ZI of Zd . Set
ZI [µ] = µ
C(Σ, Z)ZI c
ZI .
So, for instance:
Z{1,··· ,p} [µ] = Z ⊂ ZI [µ] ⊂ Z[µ] = Z∅ [µ].
Statement of the conjecture
Conjecture
(MGL conjecture) The range of the trace is contained in
Λ(Σ, Θ, µ) where:
X
Λ(Σ, Θ, µ) =
Pf(ΘI )ZI [µ]
0≤|I|≤d
So
1
If d is even,
Λ(Σ, Θ, µ) = Z[µ] +
X
Pf(ΘI )ZI [µ] + Pf(Θ)Z.
0<|I|<d
2
If d is odd,
Λ(Σ, Θ, µ) = Z[µ] +
X
0<|I|<d
Pf(ΘI )ZI [µ].
Comments on the conjecture
For B = 0, MGL is the Bellissard GL conjecture.
For the periodic case (Σ = pt), MGL is Elliott’s theorem
(’82) about the range of the trace for the rotation algebras.
Integrality of Chern ⇒ MGL conjecture.
So, MGL is satisfied in low dimensions d ≤ 3.
Index approach ([B.-Oyono-Oyono]) unifies the previous cases
and explains the conjecture
More precisely...
We use the twisted Baum-Connes isomorphism to reduce
the problem to the computation of twisted measured
indices.
Indeed, twisting the Dirac operator on Rp with the
connection d + iη defines the twisted BC morphism
µBC : K p (X ) −→ K0 (C(Σ) oσ Zd ) by µBC (ξ) := Ind(/∂η ⊗ ξ).
A twisted version of the Connes measured index theorem
allows to compute the range of τµ ◦ µBC .
We then have reduced the problem to a cohomological
computation.
The index approach
The suspension X = Rp ×Zp Σ is a compact foliated space with
transversal the Cantor set Σ, and with invariant transverse
measure induced from µ. The monodromy groupoid is
G = (Rp × Rp × Σ)/Zp .
Let /∂ denote the Dirac operator on Rp and ∇ = d + iη the
connection on the trivial line bundle on Rp , ∇E the lift to Rp × Σ
of a connection on a vector bundle E → X . Consider the
twisted Dirac operator along the leaves of the lifted foliation,
DE = /∂ ⊗ ∇ ⊗ ∇E : L2 (Rp × Σ, S + ⊗ E) −→ L2 (Rp × Σ, S − ⊗ E).
Then one computes that Tγ ◦ DE = DE ◦ Tγ for γ ∈ Zp .The heat
kernel of D has a well defined µ-trace.
The index approach
For t > 0, define the Wasserman idempotent
et (DE+ ) ∈ M2 (C(Σ) oσ Zp ⊗ K)
as follows:

 e
et (DE+ ) = 

−tDE− DE+
t
+
−
e− 2 DE DE DE+
e

− +
− e−tDE DE ) +
DE 
,
DE− DE+

−
+
1 − e−tDE DE
− 2t DE− DE+ (1
Then the C ∗ -twisted foliated analytic index is defined as
Index(DE+ ) = [et (DE+ )] − [P0 ] ∈ K0 (C(Σ) oσ Zp ),
where t > 0 and P0 is the idempotent
!
0 0
P0 =
0 1
(4)
The index approach
The Baum-Connes map
µBC : K p (X ) −→ K0 (C(Σ) oσ Zp ).
is then E 7→ Index(DE+ ). It is an isomorphism (using
Packer-Raeburn).
Reduces the problem to computing the range of
τ∗µ ◦ µBC : K p (X ) −→ R.
The index theorem
Theorem (B.-Mathai ’15)
τµs (Index(DE+ ))
=
X
Z
Pf(ΘI )
I
Σ×(0,1)p
dxI ∧ ch` (FE )dµ(ϑ).
Here I runs over subsets of {1, . . . , p} with an even number of
elements, and ΘI is the submatrix of Θ = (Θij ) with i, j ∈ I.
A McKean-Singer type argument shows that the supertrace
2
τµs (e−tD ) = τµs (Index(DE+ ))
is independent of t > 0 and represents the measured twisted
foliated index.
The 2D case
The group cohomology class of [σ] ∈ H 2 (Z2 ; R/Z) ∼
= R/Z can
be identified with a real number θ, 0 ≤ θ < 1. More precisely,
we take σ = e2πiθω with ω the standard symplectic form on Z2 .
Theorem (2D case)
One has
τµ (K0 (C(Σ) oσ Z2 ) = Z[µ] + Zθ
NB: Previously proved by J. Kellendonk.
Index proof for 2D
∼
µBC : K 0 (X ) −→ K0 (C(Σ) oσ Z2 )
is an isomorphism, where X = Σ ×Z2 R2 .
Measured twisted index theorem ⇒
Z
τµ (µBC (E)) = θ rank(E) +
c1 (E)dΛµ
ZX
= θ rank(E) +
Σ
ΨZ2 (c1 (ξ))dµ
using the isomorphism ΨZ2 : H 2 (X , Z) → C(Σ, Z)Z2 .
The result follows.
Periodic potentials = Elliott’s theorem
Consider the special case when the self-adjoint p-dimensional
magnetic Schrödinger operator
H=
1
(d + iη)† (d + iη) + V ,
2
acting on L2 (Rp )
has (real-valued) periodic potential V , then much easier: Σ is
just a point.
The crossed product algebra is C oσ Zd = AΘ .
Periodic potentials = Elliott’s theorem
MGL becomes Elliott’s theorem:
Theorem (purely periodic case)
The range of the trace on the K-theory of AΘ is:
τ (K0 (AΘ )) =
X
Pf(ΘI )Z.
0≤|I|≤p
Proof.
∼
µΘ : K 0 (Tp ) −→ K0 (AΘ ) is an isomorphism. The twisted index
theorem gives (1 leaf here!):
Z
X
τ (µΘ (E)) =
Pf(ΘI )
dxI ∧ Ch(E)
I
The result follows.
Tp
The 3D case
Theorem (B.-Mathai ’15)
The magnetic GL group is contained in
Z[µ] + Θ12 Z12 [µ] + Θ13 Z13 [µ] + Θ23 Z23 [µ].
Theorem (B.-Mathai ’15)
If for instance T3 acts minimally then Θ12 Z12 [µ] ⊂ Range(τµ ).
Therefore, if each generator acts minimally, then the magnetic
GL group coincides with:
Z[µ] + Θ12 Z12 [µ] + Θ13 Z13 [µ] + Θ23 Z23 [µ].
Main steps of the proof for 3D
Again the twisted BC isomorphism
K 1 (X ) → K0 (C(Σ) oσ Z3 ) is the class of the twisted Dirac
operator.
Applying the gap-labelling theorem + the twisted measured
index theorem, we get:
Z
−1
u d` u
tr
Range(τ∗µ ) = Z[µ]+
∧ π ∗ B dΛµ , u ∈ K 1 (X ) .
2iπ
X
This index formula+compatibility of the isomorphism
H ∗ (X , Z) ' H ∗ (Z3 , C(Σ, Z)), reduces the MGL conjecture
to computing the range of the map
∪Θ
Ψ
µ
H 1 (Z3 , C(Σ, Z)) −→ H 3 (Z3 , C(Σ, R) −→ C(Σ, R)Z3 −→ R.
Concentrate on Θ12 ψ1 ∪ ψ2 .We have exact sequences:
0 → C hT1 ,T3 i
hT2 i
→H 1 (Z3 , C)→H 1 (hT1 , T3 i, C)hT2 i → 0
and
0 → C hT3 i
hT1 i
−→H 1 (hT1 , T3 i, C)−→ (C)hT3 i
hT1 i
→ 0.
Denote by A the image of H 1 (hT1 , T3 i, C)hT2 i in (C)hT3 i
hT ,T i
So A ⊂ (C)hT3 i 1 2 .
hT1 i
.
Main steps of the proof for 3D
The map ∪(ψ1 ∪ ψ2 ) : H 1 (Z3 , C) → (C)Z3 induces a
well-defined morphism α : H 1 (hT1 , T3 i, C)hT2 i −→ (C)Z3 .
α coincides with the composite map
∪ψ
H 1 (hT1 , T3 i, C)hT2 i →1 H 2 (hT1 , T3 i, C)hT2 i → (C)Z3 .
Therefore, α induces a morphism β : A → (C)Z3 .
Hence, the range of ∪(ψ1 ∪ ψ2 ) on H 1 (Z3 , C) coincides with
β(A).
But A ⊂ H 1 (hT3 i, C)hT1 ,T2 i , so the range of µ ◦ ∪(ψ1 ∪ ψ2 ) is
hT ,T i
contained in the range under µ of (C)hT3 i 1 2 .
QED