Multiresolution analysis and QM ∞
Fabio Bagarello
Montecatini Terme, Assemblea scientifica del GNFM, 2004
Amultiresolution analysis of L2(R) is an increasing sequence of closed subspaces
. . . ⊂ V−2 ⊂ V−1 ⊂ V0 ⊂ V1 ⊂ V2 ⊂ . . . ⊂ L2(R), (1) with S
j∈ZVj dense in L2(R) and T
j∈Z Vj = {0}, and such that (1) f (x) ∈ Vj ⇔ f (2x) ∈ Vj+1
(2) There exists a function φ ∈ V0, called a scaling function, such that {φ(x − k), k ∈ Z} is an o.n. basis of V0.
(1) + (2) ⇒ {φj,k(x) ≡ 2j/2φ(2jx − k), k ∈ Z} is an o.n. basis of Vj.
Each Vj can be interpreted as an approximation space: the approximation of f ∈ L2(R) at the resolution 2j is defined by its projection onto Vj. The additional details needed for increasing the resolution from 2j to 2j+1 are given by the projection of f onto the orthogonal complement Wj of Vj in Vj+1:
Vj ⊕ Wj = Vj+1, (2)
and we have:
M
j∈Z
Wj = L2(R). (3)
Now, there exists a function ψ, the mother wavelet, explicitly computable from φ, such that {ψj,k(x) ≡ 2j/2ψ(2jx − k), j, k ∈ Z} constitutes an orthonormal basis of L2(R).
The construction of ψ proceeds as follows. First, the inclusion V0 ⊂ V1 yields the relation
φ(x) = √ 2
X∞ n=−∞
hnφ(2x − n), hn = hφ1,n|φi. (4) Then one uses these coefficients to define the function ψ as
ψ(x) = √ 2
X∞ n=−∞
(−1)n−1h−n−1φ(2x − n). (5) ψ(x) can be used to generate the wavelet transform of a given L2-function f (x) as
(Twavf )(a, b) = |a|−1/2 Z
dxf (x)ψ
µx − b a
¶ ,
where a ∈ R\{0} and b ∈ R. a and b are the zooming and the translation parameter.
EXAMPLES: (1) Haar MRA:
h(x) =
1, if 0 ≤ x < 1/2
−1, if 1/2 ≤ x < 1 0, otherwise
(6)
This function gives an explicit example of a wavelet o.n. basis in L2(R), obtained via the usual formula
hmn(x) = 2−m/2h(2−mx − n). (7) (2) Littlewood Paley
Ψ(x) = (πx)−1(sin 2πx − sin πx). (8) The behavior of this function is, in a sense, complementary to that of the Haar wavelet: it is very well localized in frequency space (it has a compact support)
Ψ(ω) =b
(2π)−1/2, if π ≤ |ω| ≤ 2π 0, otherwise,
(3) Journ´e basis
Ψ(ω) =b
(2π)−1/2 if 4π7 ≤ |ω| ≤ π and 4π ≤ |ω| ≤ 32π7
0 otherwise
(9) (3) More Examples:
• splines (linear, quadratic, cubic,...),
• Daubechies’ wavelets,
• sinc wavelets,...
Problem: how to find new scaling functions φ and MR?
Definition:– We call relevant any sequence h = {hn, n ∈ Z}
which satisfies the following properties:
(r1) P
n∈Z hnhn+2l = δl,0; (r2) hn = O(1+|n|1 2), n À 1;
(r3) P
n∈Z hn = √ 2;
(r4) H(ω) = √12P
n∈Zhne−iωn 6= 0 ∀ω ∈ [−π2, π2].
Relevant sequences ⇒ Scaling functions and Mother Wavelets (Mallat’s algorithm).
Main question: how to produce relevant sequences?
MRA ⇔ FQHE
• F. Bagarello, J.P. Antoine, Wavelet-like Orthonormal Basis of the Lowest Landau Level, J. Phys. A, 27, 2471-2481 (1994)
• F. Bagarello, More Wavelet-like Orthonormal Bases for the Lowest Landau Level: Some Considerations, J. Phys. A, 27, 5583-5597 (1994)
• F. Bagarello, Applications of Wavelets to Quantum Mechanics: a Pedagogical Example, J. Phys. A, 29, 565-576 (1996)
• F. Bagarello Multi-Resolution Analysis and Fractional Quantum Hall Effect: an Equivalence Result, J. Math. Phys., 42, 5116-5129, (2001)
• F. Bagarello Multi-Resolution Analysis and Fractional Quantum Hall Effect: More Results, J. Phys. A, 36, 123-138 (2003)
• F. Bagarello, J.P. Antoine, Localization properties and wavelet-like orthonormal bases for the lowest Landau level, in Advances in Gabor Analysis, H.G. Feichtinger, T. Strohmer Eds., Birkh¨auser, Boston, 2003
• F. BagarelloMulti-Resolution Analysis generated by a seed function, J. Math. Phys., 44, 1519-1534 (2003)
The old results: FQHE
Full hamiltonian:
H(N) = H0(N) + λ(HC(N) + HB(N)), (10) H0(N) =
XN i=1
H0(i). (11)
Here H0(i) describes the minimal coupling of the i−th electron with the magnetic field:
H0 = 1 2
¡p + A(r)¢2
= 1 2
³
px − y 2
´2 + 1
2
³
py + x 2
´2
. (12)
HC(N) is the Coulomb interaction between charged particles, and HB(N) is the interaction of the charges with the background. Both are small compared with H0(N)!
Let us introduce the following operators:
P0 = px−y/2, Q0 = py+x/2, P = py−x/2, Q = px+y/2.
(13) Then we have
H0 = 1
2(Q02 + P02). (14)
and
[Q, P ] = [Q0, P0] = i, [Q, P0] = [Q0, P ] = [Q, Q0] = [P, P0] = 0.
(15) Therefore, defining the magnetic translation operators T (~ai) for a square lattice with basis ~a1 = a(1, 0), ~a2 = a(0, 1), a2 = 2π by T1 := T ( ~a1) = eiaQ, T2 := T ( ~a2) = eiaP, (16) we find
[T ( ~a1), T ( ~a2)] = [T ( ~a1), H0] = [T ( ~a2), H0] = 0, (17) and, given a generic function f (x, y) ∈ L2(R2),
fm,n(x, y) := T1mT2nf (x, y) = (−1)mneia2(my−nx)f (x+ma, y+na).
(18) The wave function in the (x, y)-space is related to its P P0- expression by the formula
Ψ(x, y) = eixy/2 2π
Z ∞
−∞
Z ∞
−∞
ei(xP0+yP +P P0)Ψ(P, P0) dP dP0, (19)
which can be easily inverted:
Ψ(P, P0) = e−iP P0 2π
Z ∞
−∞
Z ∞
−∞
e−i(xP0+yP +xy/2)Ψ(x, y) dxdy.
(20)
The ground state of H0 must have the form f0(P0)h(P ), where f0(P0) = π−1/4e−P02/2, (21) while the function h(P ) is arbitrary, which manifests the degen- eracy of the LLL, and should be fixed by the interaction. With f0 as above formula (19) becomes
Ψ(x, y) = eixy/2
√2π3/4
Z ∞
−∞
eiyPe−(x+P )2/2h(P ) dP, (22)
which produces the lattice as ψm,n(x, y) = T1mT2nΨ(x, y).
Our main result is that, if the following ONC holds,
< ψl1,2l2, ψ0,0 >=
Z ∞
−∞
dxe−2il2axh(x − l1a)h(x) =
=
Z ∞
−∞
dpeil1apˆh(p − 2l2a)ˆh(p) = δl1,0δl2,0, (23) for all l1, l2 ∈ Z, where ˆh(p) is the Fourier transform of h(x),
then, defining
hn = 1
√a
Z ∞
−∞
dpe−inxah(x), (24)
it follows automatically that X
n∈Z
hnhn+2l = δl,0. (25) We also have proved the opposite implication: given any rel- evant sequence we can produces an o.n. set in the LLL.
Therefore: At that stage we believed in the existence of a deep relationship between FQHE and MRA. We also analyzed the other requirements of a relevant sequence in that perspective...
New results
...but this is not strictly necessary: the procedure is only related to the unitary map between two pairs of conjugate oper- ators. All the results deduced are model independent !
Consider the operators ((ˆx, ˆpx), (ˆy, ˆpy)) and ((ˆx1, ˆp1), (ˆx2, ˆp2)), satisfying
[ˆx, ˆpx] = [ˆy, ˆpy] = i, [ˆx1, ˆp1] = [ˆx2, ˆp2] = i
Let ξx and ηy be (generalized) eigenstates of ˆx and ˆy: ˆxξx = xξx, ˆyηy = yηy, and ξx01 and ηx02 eigenstates of the new position operators ˆx1 and ˆx2. We have
Z dx
Z
dy |ξx,y >< ξx,y| = Z
dx1 Z
dx2|ξx01,x2 >< ξx01,x2| = 11,
where ξx,y = ξx ⊗ ηy and ξs,t0 = ξs0 ⊗ ηt0. Any Ψ ∈ H can be written in the (x, y)-coordinates or in the (x1, x2)-coordinates:
Ψ(x, y) =< ξx,y|Ψ > and Ψ0(x1, x2) =< ξx01,x2|Ψ >,
which are related as Ψ(x, y) =
Z
dx1 Z
dx2 < ξx,y|ξx01,x2 > Ψ0(x1, x2) Ψ0(x1, x2) =
Z dx
Z
dy < ξx01,x2|ξx,y > Ψ(x, y)
Remark:– Ψ(x, y) and Ψ0(x1, x2) are different representa- tions of the same element of H.
Let us choose Ψ0(x1, x2) = ϕ(x1)h(x2) and let us define three commuting operators: H = H(ˆx1, ˆp1) = H†, T1 = eiaˆx2 and T2 = eiaˆp2. Here a2 = 4π. We still call the unitary operators magnetic translations and H the hamiltonian.
But there is no physical system behind, now.
As for the FQHE we require orthonormality of the functions Ψh,~l(x, y) = T1l1T2l2Ψh(x, y), which generate a lattice with cell area = 4π.
Remark:– In the FQHE the function ϕ(x1) was the ground state of H. We will show that this is unessential.
We find that, independently of the model, that is, of the ex- plicit expressions for H, T1 and T2, as well as of the choice of ((ˆx1, ˆp1), (ˆx2, ˆp2)), (mainly) due to the resolution of the identity above, if ϕ has L2(R)-norm 1, then:
< Ψh,(l1,l2), Ψh,(0,0) >=
Z
R
ds h(s) h(s + al2) e−isal1 =
= Z
R
dp ˆh(p) ˆh(p − al1) e−ipal2 =: Sl(h)
1,l2,
which does not depend on the choice of ϕ, [B, JMP, 2001], [B, JPA, 1996]. We call ONC the equation
Sl(h)
1,l2 = δ~l,~0.
Solutions of the ONC can be easily found using a relevant sequence h = {hn, n ∈ Z} (P
n∈Zhnhn+2l = δl0):
h(s) =
√1 a
P
n∈Zhne−isna/2, s ∈ [0, a[,
0 otherwise
(26) Remark 1: any MRA produces an o.n. basis in the LLL (as well as in all the other Landau levels!).
Remark 2: If we have a d-MRA, we still find solutions of the ONC if a2 = 2πd.
Remark 3: Since H, T1, T2 may have no physical meaning, it is not very relevant to use MRA to produce o.n. sets in the Landau levels. We do the opposite: we use the ONC to produce MRA!!!!
The recipe:
(1) Find a solution h(s) of the ONC.
(2) Put
hn = 1
√a Z
R
h(s) eisna/2ds =
r2π a ˆh³
−na 2
´
Then, using the Poisson summation formula, we can easily see that P
n∈Zhnhn+2l = δl0.
Remark 4: It is much simpler to find solutions of the ONC than, directly, sequences satisfying (r1).
Examples: solutions of the ONC and related sequences
Obvious solutions of the ONC are the ones arising from a MRA. Other solutions are:
(1) the trivial ones:
h(s) =
√1
a, s ∈ [0, a[, 0 otherwise
ˆh(p) =
q
2
a, p ∈ [0, a/2[, 0 otherwise then hn = δn0. Trivially we have P
n∈Z hnhn+2l = δl0. (2) the effects of a phase..:
h(s) =
e√−is
a , s ∈ [0, a[, 0 otherwise then hn = i(1 − e−ia)2πn−a1 . P
n∈Zhnhn+2l = δl0 is not trivial!
(3) close to the sinc wavelets:
h(s) =
q2
a, s ∈ [0, a/2[, 0 otherwise,
h(s) =
√1
2a, s ∈ [0, a/2[∪[2a, 3a[,
−√12a, s ∈ [a/2, a[, 0 otherwise and
h(s) =
√1
2a, s ∈ [0, a/2[∪[a, 2a[,
−√12a, s ∈ [a/2, a[, 0 otherwise
they all satisfy the ONC and produce the following sequence:
h0 = √12, h2n = 0 if n 6= 0, and h2n+1 = π(2n+1)i√2 , which satisfies (r1).
(4) the Haar wavelet:
ˆh(p) =
√1
a, p ∈ [0, a[, 0 otherwise
then h0 = h−1 = √12, which are the Haar coefficients.
(4) more compactly supported solutions (in p)
ˆh(p) =
√1
2a, p ∈ [0, a/2[∪[2a, 3a[,
−√12a, s ∈ [a/2, a[, 0 otherwise
produces h0 = h−4 = h−5 = −h−1 = 12, while
ˆh(p) =
√1
2a, s ∈ [0, a/2[∪[a, 2a[,
−√12a, s ∈ [a/2, a[, 0 otherwise produces h0 = h−2 = h−3 = −h−1 = 12.
Remark: The ONC satisfies an useful symmetry property:
if h(s) is a solution of the ONC, another solution is obtained simply by taking the inverse Fourier transform of a ˆg(p) ≡ h(p).
More solutions: the orthonormalization trick
More solutions of the ONC, and more relevant sequences as a consequence, can be found using this ONT:
let h(s) be a generic L2 function and let us compute Sl(h)
1,l2 =<
ψl(h)
1,l2, ψ0,0(h) >. If Sl(h)
1,l2 6= δl1,0δl2,0 then the ψl(h)
1,l2(~r) generate a non o.n. lattice (in L2(R2)). However, [BMS, PRB (1993)], we can find an o.n. set in L2(R2), invariant under magnetic translations, simply by considering
ψl(H)
1,l2(~r) = T1l1T2l2ψ(H)0,0 (~r), where ψ0,0(H)(~r) = X
~n∈Z2
f~nψ~n(h)(~r),
and the coefficients f~n are fixed by requiring that Sl(H)
1,l2 =< ψl(H)
1,l2, ψ(H)0,0 >= δl1,0δl2,0 In this way we deduce that H(s) = P
~n∈Z2 f~neisan1h(s + an2), and the related coefficients are
Hn =
r2π a Hˆ
³
−na 2
´
= . . . = 1
√a Z
R
ds h(s)
pS(h)(as, 0)eiasn/2,
where
S(h)(~p) = X
~n∈Z2
S~n(h)ei~p·~n.
Then P
n∈Z HnHn+2l = δl,0.
Examples of this construction are contained in [B., JMP, 2003]
Remarks: (1) if < ψl(h)
1,l2, ψ0,0(h) >= δl1,0δl2,0 then S(h)(~p) = 1 and we go back to the original result: Hn = hn and no o.n. trick is needed!
(2) again the procedure is model independent! It only de- pends on the unitary map from ((ˆx, ˆpx), (ˆy, ˆpy)) to ((ˆx1, ˆp1), (ˆx2, ˆp2)) and not on the particular expressions for H, T1 and T2.
The other conditions
• Condition (r2), hn = O(1+|n|1 2), n À 1
For this to be satisfied we have a very easy criterion: since hn =
q2π a ˆh¡
−na2 a¢
, it is enough to look for solutions of the ONC such that ˆh(p) is compactly supported.
It would be enough to find solutions h(s) of the ONC regular enough: the more regular h(s) is, the more rapidly hn goes to zero when n → ∞.
If we adopt the o.n. trick, then we can repeat the same considerations with h(s) replaced by ˜h(s) = √ h(s)
S(h)(as,0).
• Condition (r3), P
n∈Z hn = √ 2.
A necessary condition for (r3) to hold h(s) must satisfies the ONC and one of the following equalities: P
n∈Zh(na) = q
2
a or P
n∈Z ˆh(na2 ) = pa
π. If we adopt the o.n. trick then we can relax the assumption that h(s) satisfies the ONC. It
is enough to have one of the following equalities satisfied:
X
n∈Z
h(an) = vu utX
~l∈Z2
h
µal1 2
¶ h
µal1
2 + al2
¶
or
X
n∈Z
ˆh³an 2
´
= vu
ut2 X
~l∈Z2
ˆh
µal2 2
¶ ˆh
µal2
2 + al21
¶
(The Haar example above satisfies both!)
• Condition (r4), P
n∈Z hne−iωn 6= 0 for all ω ∈ £
−π2, π2¤ . For that we find that the following condition must be satis- fied by the seed function h(s):
X
n∈Z
h
³ a
³
n + ω 2π
´´
6= 0, ω ∈ h
−π 2, π
2 i
whether h(s) satisfies the ONC or not.
Future Plains
1. We have to consider still the examples from the point of view of MRA: what kind of mother wavelets do we get? For that we need to find more examples: this is work in progress!
2. How to impose stronger conditions (e.g. on the support of the mother wavelets or its regularity)?
3. More physical applications
4. Explore the underlying symmetry between (ˆx1, ˆp1) and (ˆx2, ˆp2):
this produces examples of the Tomita-Takesaki procedure for constructing, e.g., a modular operator. A first concrete application has been explored in [Ali, B., JMP, submitted]
where the FQHE has been analyzed.
5. ...