Bose Einstein condensates in the Lowest Landau

Level: Hamiltonian dynamics.

F. Nier
IRMAR, UMR-CNRS 6625,
Campus de Beaulieu
Universite Rennes 1
35042 Rennes Cedex,France.
March 15, 2006
Abstract
In a previous article with A. Aftalion and X. Blanc, it was shown that the
hypercontractivity property of the dilation semigroup in spaces of entire functions
was a key ingredient in the study of the Lowest Landau Level model for fast rotating
Bose Einstein condensates. That former work was concerned with the stationnary
constrained variational problem. This article is about the nonlinear Hamiltonian
dynamics and the spectral stability of the constrained minima with motivations
arising from the description of Tkatchenko modes of Bose-Einstein condensates.
Again the hypercontractivity property provides a very strong control of the nonlinear
term in the dynamical analysis.
1 Introduction
The Lowest Landau Level energy functional of rapidly rotating Bose-Einstein condensates
in a harmonic trap can be written as
G
h
(f) =
_
C
[z[
2

f(z)e

|z|
2
2h

2
+
Na
2
h
2

f(z) e

|z|
2
2h

4
L(dz) . (1.1)
The number N of atoms in the condensate and the scattering length a are xed and
h =
_
1
2
h
, where
h
is the ratio of two rotational velocities, is a small parameter.
Here and in the sequel L(dz) denotes the Lebesgue measure on C R
2
. In this scaling,
the set of admissible f in the Lowest Landau Level approximation, is the semiclassical
1
Fock-Bargmann space
T
h
=
_
f L
2
(C, e

|z|
2
h
L(dz)), s.t f entire
_
(1.2)
with |f|
2
F
h
=
_
C
[f(z)[
2
e

|z|
2
h
L(dz) . (1.3)
The equilibrium states which are experimentally observed are within this modelling the
solution of the constrained minimization problem
inf
_
G
h
(f), f T
h
, |f|
F
h
= 1
_
. (1.4)
We follow the presentation of [ABN1, ABN2] and additional information about the physics
of the problem can be found in [ARVK, Aft, ABD, Ho, WBP].
This article focuses on the nonlinear Hamiltonian dynamics
i
t
f = 2
f
G
h
(f) (1.5)
and on the spectral analysis of the linearized Hamiltonian around an equilibrium cong-
uration. This last problem is motivated by the experimental and numerical study of the
vibration modes of condensates, also known as Tkachenko modes, which can be found in
the physics literature [BWCB, SCEMC, Son].
More precisely, after specifying the functional framework and the results of [ABN2]
about the minimization problem, the following topics will be studied:
1. Owing the hypercontractivity property already used in [ABN2], the Hamiltonian
ow associated with the nonlinear equation (1.5) will be globally dened on T
h
.
2. The spectral stability,
1
of constrained minimizers will be proved after a natural
modication of the energy functional. Since a minimum of the functional G
h
on
the sphere,
_
|f|
F
h
= 1
_
, is not always a local minimum of G
h
and due to some
degeneracy related with the rotational invariance, the spectral stability is better
studied by considering a modied energy
G
h

0
,
1
(f) = G
h
(f) +
0
(|f|
2
F
h
) +
1
(f [ zh
z
f)
F
h
)
in which the two additional terms are functions of quantities left invariant by the
Hamiltonian dynamics. This change of energy functional, amounts to a simple
explicit time-dependent gauge transformation f

0
,
1
(z, t) = e
i
1
t
f(e
i
2
t
z, t) where
the real numbers
1
,
2
are determined by the choice of
0
,
1
, and the initial
data f(z, t = 0) T
h
. A good choice of the function
0
,
1
allows to prove that
the spectrum of the modied linearized Hamiltonian is purely imaginary. From
the physical point of view, the introduction of an articially modied dynamics is
related with the idea that one is interested in the linearized dynamics up to uniform
solid rotations of the whole condensate. A good intuitive picture is provided by the
vibration modes of a swinging bell.
1
i.e. the spectrum of the linearized Hamiltonian is purely imaginary
2
3. The relevance of numerical approximations is considered: the numerical simula-
tions in [ABD] consist in minimizing the energy G
h
on a space of polynomials with
bounded degree, instead of the space T
h
. It was checked in [ABN2] that when the
degree of the polynomials is taken suciently large, the solution to the nite dimen-
sional problem provides a suitable approximation of the minimizer in T
h
. Here it is
proved that a similar result holds for the linearized Hamiltonian in a norm resolvent
sense. We end with some comments about the numerical stability of the spectral
elements in such an approximation, by pointing out that the linearized Hamiltonian
is a non anti-adjoint operator with a purely imaginary spectrum.
The appendix gathers standard tools about the classical Hamiltonian dynamics in
innite dimension and some results are adapted to our specic framework.
2 Preliminaries
In this section, we set up the functional framework. The basic tools introduced in [ABN2]
and the initial result on the minimization problem (1.4) are reviewed. Then, the Kaehler
structure of T
h
, which is the combination of its (real) euclidean structure and its (real)
symplectic structure and which is the natural framework for the study of the Hessian
and the linearized Hamiltonian ow, is explicitely written. Finally, the Hessian of the
functional G
h
is computed and an estimate is given for Hess G
h
(f) when f is a solution
to (1.4).
2.1 The minimization problem
Here is a very short review of [ABN2]. The Bargmann transform (see for example [Bar,
Fol, Mar]) is used with the following normalization
[B
h
](z) =
1
(h)
3/4
e
z
2
2h
_
R
e

2zy)
2
2h
(y) dy ,
with z =
xi

2
C and o

(R). It denes a unitary operator B

h
: L
2
(R, dy) T
h
and
the orthogonal projection
h
= B
h
B

h
: L
2
(C, e

|z|
2
h
L(dz)) T
h
is given by
[
h
f](z) = [B
h
B

h
f](z) =
1
h
_
C
e
zz

h
e

|z

|
2
h
f(z

) L(dz

) .
The harmonic oscillator quantum Hamiltonian (or number operator in the Fock represen-
tation) given by:

N
h
=
1
2
(h
2

2
y
+ y
2
h)
D(

N
h
) =
_
u L
2
(R, dy), y

y
u L
2
(R, dy), + 2
_
,
3
is transformed into the generator of dilations:
N
h
= B
h

N
h
B

h
= z(h
z
) .
An element f = B
h
of T
h
considered as an element of L
2
(C, e

|z|
2
h
L(dz)), satises
h
z
f = h
z
(
h
f) =
h
(zf) .
For the spectral resolution of these two operators, the basis of normalized Hermite func-
tions (c
n
H
n
(y))
nN
is tranformed via B
h
into the the basis of monomials (c
n
z
n
)
nN
.
Then the spaces
T
s
h
=
_
f entire, s.t.
_
C
z)
2s
[f(z)[
2
e

|z|
2
h
L(dz) <
_
, (2.1)
where z) =
_
1 +[z[
2
can identied constructed via the spectral resolution of N
h
. The
union
zR
T
s
h
is nothing but B
h
[o

(R)] and T
s
h
is compactly embedded in T
h
as soon as
s > 0.
Another property which will be used also in this article, is the next consequence of the
hypercontractivity property of the semigroup (e
tN
1
)
t>0
(see for example [Car, Gro, Nel]):
Lemma 2.1. [ABN2] The quantity
_
C
f
1
(z)f
2
(z)f
3
(z)f
4
(z)e
2|z|
2
h
L(dz)
denes a continuous (2, 2)-linear functional
2
on T
h
with norm smaller than
1
2h
. Hence
for any , 0, 1, 2, the

f
derivative of the functional
f
_
C
[f(z)[
4
e
2|z|
2
h
L(dz)
denes a continuous (2 , 2 )-linear mapping from T
h
into T
h

h
with norm
4
2h(2)!(2)!
.
The above Lemma and the compactness of the imbedding T
1
h
T
h
lead naturally to
the next result (see the proof in [ABN2])
Theorem 2.2. [ABN2] For any xed h > 0, the minimization problem (1.4) admits a
solution in T
1
h
. Any minimizer is a solution to the Euler-Lagrange equation

h
__
[z[
2
+ Na
2
h
e

|z|
2
h
[f[
2

_
f
_
= 0 (2.2)
2
A (2, 2)-linear functional is an R-quadrilinear functional which is C-linear with respect to the two
rst arguments and C-linear with respect to the two last arguments.
4
where R is the Lagrange multiplier satisfying the uniform estimates
2
h
3
_
2Na

< e
h
LLL
2e
h
LLL
2
2
h
3
_
2bNa

+ o
Na
(h
0
) , (2.3)
with
e
h
LLL
= min
_
G
h
(f) , f T
h
, |f|
F
h
= 1
_
.
The Euler-Lagrange equation (2.2) can also be written as
zh
z
f + Na
2
h

h
_
e

|z|
2
h
[f[
2
_

h
f ( h)f = 0 , (2.4)
or zh
z
f +
Na
2
h
2

f(h
z
)[f
2
(2
1
.)] ( h)f = 0 , (2.5)
the operator

f(h
z
) being dened as the limit lim
K

K
k=0
a
k
(h
z
)
k
if f(z) =

k=0
a
k
z
k
.
2.2 The Kaehler structure
The Hessian of the energy functional and the linearized Hamiltonian vector eld, are
objects associated with the real euclidean structure or with the real symplectic structure
of T
h
. Some of their properties are more obvious in this presentation. We specify here
the Kaehler structure associated with the complex Hilbert space T
h
.
The space T
h
can be viewed as a real Hilbert space with the scalar product:
Re f
1
[ f
2
)
F
h
and as a symplectic space with the form
(f
1
, f
2
) = 1mf
1
[ f
2
)
F
h
.
The Euclidean structure on T
h
is more convenient when one studies the second variation,
while the symplectic structure will be used in Section 3 for the Hamiltonian ow. These
structures are completely claried once the complex conjugation is dened on T
h
. The
simplest way of writing can be done in the orthonormal basis B
h
[H
h
n
] = c
n,h
z
n
with
c
n,h
=
1
(h)
1/2
h
n/2

n!
. Let
f
k
=

nN
f
k,n
c
n,h
z
n
=

nN
(f
R
k,n
+ if
I
k,n
)c
n,h
z
n
, k 1, 2
we get
f
1
[ f
2
)
F
h
=

nN
f
1,n
f
2,n
f
1
[ f
2
)
F
h
,R
=

nN
f
R
1,n
f
R
2,n
+ f
I
n,1
f
I
n,2
=
_
f
R
1
, f
I
1
_
_
f
R
2
f
I
2
_
,
and (f
1
, f
2
) =

nN
f
I
1,n
f
R
2,n
f
R
1,n
f
I
2,n
=
_
f
R
1
, f
I
1
_
_
0 Id
Id 0
__
f
R
2
f
I
2
_
.
5
For the real scalar product Ref
1
[ f
2
), we will use the notation f
T
1
f
2
which refers to the
matrix representation.
The natural denition of f (which has to stay in a space of holomorphic functions) follows
from the writing f(z) =

nN
(f
R
n
+ if
I
n
)c
n,h
z
n
= f
R
(z) + if
I
(z):
f(z) = f(z) .
This denition allows to check the relationships (with f, f
i
T
h
and v(z, z) non necessarily
holomorphic)
f
1
[ f
2
)
F
h
=
_
C
e

|z|
2
h
f
1
(z)f
2
(z) L(dz) , (2.6)

h
[v(z, z)] =
h
_
v(z, z)
_
,
h
[f(z)] =
h
f , (2.7)
G
h
(f) =
_
C
_
e

|z|
2
h
[z[
2
f(z)f(z) +
Na
2
h
2
e

2|z|
2
h
f
2
(z)f
2
(z)
_
L(dz) . (2.8)
2.3 Hessian
Of course there are several ways to study the second variation of a function G
h
(f), by
computing the (f, f) coordinates or with the coordinates (f
R
, f
I
) introduced before. The
second choice put the stress on the real euclidean structure of T
h
in which the notion of
Hessian makes sense.
Proposition 2.3. Let f = f
R
+ if
I
T
h
, the Hessian of G
h
is a bounded perturbation
of 2N
h
with form domain T
1
h
= Q(N
h
). It denes a closed operator with domain T
2
h
and,
after writing =
_

R

I
_
, is equal to
Hess G
h
(f) =
_
A 0
0 A
_
+
_
B 0
0 B
_
+
_
0 C
C 0
_
(2.9)
where A, B and C are the real operators
A = 2(N
h
+ h) + 4Na
2
h

h
_
e

|z|
2
h
[f(z)[
2
_

h
(2.10)
B = Na
2
h

h
_
e

|z|
2
h
_
f(z)
2
+ f(z)
2

(z)
_
(2.11)
C = Na
2
h

h
_
e

|z|
2
h
i
_
f
2
(z) f
2
(z)
_
(z)
_
. (2.12)
In the complexied Hilbert space T
h
T
h
, it denes a real self-adjoint operator which
has a compact resolvent (and therefore a discrete spectrum going to innity) and whose
spectrum is bounded from below by C
0
h
1
|f|
2
.
6
Remark 2.4. a) An operator / is said to be real when = implies / = /. This
denition is used in T
h
with the previous denition of the conjugation f f and in the
complexied space T
h
T
h
, where the conjugation is dened componentwise.
b) Beside the conservation of reality, the explicit expression will also be useful in the nal
discussion of Subsection 4.3.
Proof: The last expression of G
h
(f
R
+if
I
) suggests to start with the second variation
with respect to f and f:
G
2
= (
R
,
I
)Hess G
h
(f)
_

R

I
_
= 2
f

f
G
h
(f).. +
f

f
G
h
(f).. +
f

f
G
h
(f)..
= 2 [ (N
h
+ h))
F
h
+ 4Na
2
h
_
C
e

2|z|
2
h
(z) [f[
2
(z) L(dz)
+Na
2
h
_
C
e

2|z|
2
h
f
2
(z)
2
(z) L(dz) + Na
2
h
_
C
e

2|z|
2
h

2
(z)f
2
(z) L(dz),
hence
G
2
=
_

R
[
_
2(N
h
+ h) + 4Na
h

h
_
e

|z|
2
h
[f[
2
_

h
_

R
_
F
h
+
_

I
[
_
2(N
h
+ h) + 4Na
h

h
_
e

|z|
2
h
[f[
2
_

h
_

I
_
F
h
+Na
2
h
_
C
e

2|z|
2
h
f
2
(z)
_

R
(z)
2

I
(z)
2
+ 2i
R
(z)
I
(z)

L(dz)
+Na
2
h
_
C
e

2|z|
2
h
f
2
(z)
_

R
(z)
2

I
(z)
2
2i
R
(z)
I
(z)

L(dz).
We thus get:
G
2
=
R
[ A
R
)
F
h
+
I
[ A
I
)
F
h
+Na
2
h
_
C
e

2|z|
2
h
_
f(z)
2
+ f(z)
2

2
R
(z) L(dz)
Na
2
h
_
C
e

2|z|
2
h
_
f(z)
2
+ f(z)
2

2
I
(z) L(dz)
+Na
2
h
_
C
e

2|z|
2
h
i
_
f(z)
2
f(z)
2

I
(z)
R
(z) L(dz)
+Na
2
h
_
C
e

2|z|
2
h
i
_
f(z)
2
f(z)
2

R
(z)
I
(z) L(dz) .
The functions
R
and
I
are real elements of T
h
and the relations

R,I
(z) =
R,I
(z)
7
lead to
(
R
,
I
)Hess G
h
(f)
_

R

I
_
=
R
[ A
R
)
F
h
+
I
[ A
I
)
F
h
+
R
[ B
R
)
F
h

I
[ B
I
)
F
h
+
R
[ C
I
)
F
h
+
I
[ C
R
)
F
h
.
The expression of Hess G
h
is deduced from this once A, B and C are real operators. This
is a straightforward consequence of (2.7). Lemma 2.1 implies that Hess G
h
(f) 2N
h
is a
bounded self-adjoint operator with a norm controlled by C
0
h
1
|f|
2
F
h
.
Proposition 2.5. Assume that f T
1
h
is a solution to the minimization problem (1.4)
and of the Euler-Lagrange equation (2.2) with Lagrange multiplier > 0. Then we have
P
f
Hess G
h
(f)P
f
2P
f
,
where P
f
denotes the orthogonal projector on f

for the (real) euclidean structure on

T
h
.
Proof: The result is standard for nite dimensional problems. We write a proof in
order to check that Lemma 2.1 provides the suitable norm estimates. Since the derivatives
of G
h
with order larger than 2 are bounded according to Lemma 2.1, we obtain in the
real representation of elements of T
h
G
h
(f + t) = G
h
(f) +G
h
(f).t +
1
2
t
T
Hess G
h
(f)t + O(|t|
3
F
h
)
for all t T
1
h
. If f solves (2.2) and |f + t|
F
h
= |f| = 1, one obtains, while noting that
the real gradient is equal to 2f,
G
h
(f + t) = G
h
(f) +
1
2
t
T
_
Hess G
h
(f) 2 Id

t + O(|t|
3
F
h
) .
By Proposition 2.3 the operator P

(Hess f 2)P

is a bounded from below self-adjoint

operator with a compact resolvent. If the proposition is not true, it admits a negative
eigenvalue
0
< 0 with a normalized eigenvector . This eigenvector solves the equation
Hess G
h
(f) = (2
0
) +
1
f,
1
R.
It implies T
2
h
f

while the Euler-Lagrange (2.2) equation also gives f T

2
h
. We
take
t = t() = |f + |
1
(f + ) f = + ((1 +
2
)
1
1)f +
_
(1 +
2
)
1
1
_

and we obtain
G
h
(f + t()) = G
h
(f)

0
2

2
+ O(
3
)
with |f + t()| = 1, which is impossible.
8
3 The Hamiltonian ow
The Hamilton equations associated with the energy G
h
(f) in the Bargmann space T
h
simply reads
i
t
f = 2
f
G
h
(f) = 2(N
h
+ h)f + 2Na
2
h

h
(e

|z|
2
h
[f[
2
)f . (3.1)
We recall that the symplectic structure is given by
(f
1
, f
2
) = 1mf
1
[ f
2
)
F
h
according to Subsection 2.2.
We rst refer to the Appendix for some notations and results about the nonlinear Hamil-
tonian ow on T
h
. Then, we introduce a modied energy
G
h

0
,
1
(f) = G
h
(f) +
0
(|f|
2
F
h
) +
1
(f [ N
h
f)
F
h
) .
which allows to handle easily the restriction to the sphere
_
|f|
F
h
= 1
_
and the degeneracy
of the minimization problem due to the rotational symmetry. Finally, we tackle the
linearization problem: Owing to Proposition 2.5 and results of Appendix A, we are able
to select properly the modier
0
and
1
so that the linearized Hamiltonian has a discrete
purely imaginary spectrum (i
n
)
nZ

with
n
=
n
and lim
n
[
n
[ = +.
3.1 The Cauchy problem
We check here that the Hamiltonian ow associated with (3.1) is well dened and admits
two preserved quantities. We consider the Cauchy problem
_
i
t
f = 2(N
h
+ h)f + 2Na
2
h

h
(e

|z|
2
h
[f[
2
)f
f(t = 0) = f
0
.
(3.2)
Proposition 3.1. For any f
0
T
2
h
, the Cauchy problem (3.2) admits a unique solution
in (
0
(R; T
2
h
) (
1
(R; T
h
).
The ow (t)dened by (t)f
0
= f(t) admits a unique continuous extension to T
h
, so that
(t)f
0
(
0
(R; T
h
) for all f
0
T
h
.
This ow admits three preserved quantities. By setting f(t) = (t)f
0
, they are given by:
f
0
T
h
, t R, |f(t)|
2
F
h
= |f
0
|
2
F
h
,
f
0
T
2
h
, t R, G
h
(f(t)) = G
h
(f
0
) ,
f
0
T
2
h
, t R, f(t) [ N
h
f(t))
F
h
= f
0
[ N
h
f
0
)
F
h
.
Proof: We use the results of Proposition A.8 in Appendix A. We simply specify
here the corresponding notations. The Kaehler space H is the underlying real vector
space of T
h
. An element f = f
R
+ if
I
of T
h
is identied with
_
f
R
f
I
_
according to
9
Subsection 2.2. The real scalar product and the symplectic form on H are the ones dened
in Subsection 2.2, while the involution J is the multiplication by i in T
h
corresponding
to the matrix
_
0 Id
Id 0
_
. The complexied space H
C
is simply T
h
T
h
where a new
multiplication by i is authorized componentwise. The operator A is 2(N
h
+h) in T
h
, which
corresponds
_
2N
h
+ 2h 0
0 2N
h
+ 2h
_
in the real space H T
h
or its complexication
H
C
= T
h
T
h
. It is real, A(D(A) H) H, it has a compact resolvent and it commutes
with J. The function h(f) is the nonlinear part of G
h
(f):
h(f) =
Na
2
h
2
_
C
[f(z)[
4
e

2|z|
2
h
L(dz)
=
Na
2
h
2
_
C
(f
R
(z) if
I
(z))
2
(f
R
(z) + if
I
(z))e

2|z|
2
h
L(dz)
=
Na
2
h
2
_
C
_
f
R
(z) i f
I
(z)
_
2
(f
R
(z) + if
I
(z))e

2|z|
2
h
L(dz) .
Its expression in terms of
_
f
R
f
I
_
and the continuity property of Lemma 2.1 show that
it is real analytic in H. The last expression, due to f
R
= f
R
and f
I
= f
I
when f T
h
,
allows its extension as a real valued, real analytic function, with h(e
J
f) = h(f), to
the complexied space H
C
= T
h
T
h
as it is required in Hypothesis A.2. The relation
h(e
iN
h
f) = h(f) for R and f T
h
, which is the invariance of the above functional
with respect to rotations, provides the gauge invariance required in Proposition A.8.
Remark 3.2. Although we are working with an innite dimensional system, the con-
servation of |f(t)|
F
h
and f(t) [ N
h
f(t))
F
h
can be viewed as a consequence of Noethers
Theorem (see [AbMa][Arn]). These quantities are associated with the invariance with re-
spect to the multiplication by a phase factor f e
i
f and with respect to the rotations
f e
iN
h
f, R, of the energy functional G
h
.
3.2 A modied Hamiltonian
Modifying the energy functional with the help of preserved quantities in order to study
the stability of equilibrium in Hamiltonian systems is a standard process. This can be
viewed for our specic Hamiltonian dynamics as a variation of the Casimir functional
method. We refer for example to [HMRW] where many applications are discussed.
Here a modied Hamiltonian is introduced for two reasons:
1) We are interested in the stability of a constrained minimum.
2) The minimization problem is degenerate due to the rotational invariance.
We verify at the end of this paragraph that this modication of the energy functional
makes sense and allows to catch relevant information on the dynamics.
10
First of all, we note that a solution f to the Euler-Lagrange equation (2.2) with
Lagrange multiplier , is nothing but a critical point of the functional
G
h

0
() = G
h
() +
0
(||
2
F
h
) ,
where
0
is (
2
function on R
+
such that

0
(1) = . Due to the conservation of |f(t)|
F
h
by the Hamiltonian ow (t) associated with G
h
, the new dynamics is well understood in
terms of classical solutions: for f
0
T
2
h
, the Cauchy problem
_
i
t
g = 2
g
G
h

0
(g)
g(t = 0) = f
0
.
(3.3)
has a unique classical solution g (
1
(R; T
h
) (
0
(R; T
2
h
). It is equal to
g(t) = e
2it

0
(f
0

2
F
h
)
(t)f
0
.
Hence a solution f to the Euler-Lagrange equation (2.2) with Lagrange multiplier ,
is transformed by taking

0
(1) = into an (unconstrained) equilibrium for the new
Hamiltonian dynamics. Moreover the new dynamics for all other initial data is obtained
after applying an elementary change of phase.
The second modication will help to get rid of the degeneracy problem due to the
rotational invariance.
Denition 3.3. Let f T
2
h
be a solution to the Euler-Lagrange equation (2.2) with
Lagrange multiplier R and |f|
F
h
= 1. The functionals
0
(||
2
F
h
) and

1
([ N
h
)
F
h
) are said to be adapted to (f, ) if

0
and
1
are (
2
real-valued functions on [0, +).

0
(1) = (|f|
F
h
= 1) and

1
(f [ N
h
f)
F
h
) = 0 .
In such a case, the adapted energy is dened by
G
h

0
,
1
() = G
h
() +
0
(||
2
F
h
) +
1
([ N
h
)
F
h
) .
Remark 3.4. The condition f T
2
h
is not a restriction because an element f T
h
which
solves the Euler-Lagrange equation (2.2) necessarily belongs to T
2
h
.
The same argument as above leads to:
Proposition 3.5. Let f be a solution to (2.2) with Lagrange multiplier . Let
0
(||
2
F
h
)
and
1
([ N
h
)
F
h
) be adapted to the pair (f, ) according to Denition 3.3. Then for
any f
0
T
2
h
, the function
(z, t) = e
2it

0
(f
0

2
F
h
)
[(t)f
0
](e
2iht

1
(f
0
| N
h
f
0

F
h
)
z)
is the unique classical solution ( (
0
(R; T
2
h
) (
1
(R; T
h
)) of the Cauchy problem
_
i
t
= 2

G
h

0
,
1
()
(t = 0) = f
0
.
(3.4)
11
Proof: It is sucient to write

G
h

0
,
1
() =

G
h
() +

0
(||
2
F
h
) +

1
([ N
h
)
F
h
)zh
z
.
If is a classical solution to (3.4) then
e
2i
R
s
0

0
((s)
2
F
h
) ds
(e
2ih
R
s
0

1
(| N
h

F
h
) ds
z)
is a classical solution to (3.2). Proposition 3.1 yields the result.
The modication of the energy functional does not change the dynamics of the solu-
tion f to the Euler-Lagrange equation (2.2). For other initial data, it is changed by a
multiplication by a phase factor and by the action of a uniform solid rotation. These ad-
ditional time dependence are slow when the initial data f
0
is close to the critical point f.
In the spectral stability analysis which follows, this means that we consider the stability
up to the multiplication by a phase factor (which does not change the modulus of the
wave function [u[
2
= [f(z)[
2
e

|z|
2
h
) and up to a uniform (and slow) solid rotation.
3.3 Linearized Hamiltonian
Here we consider a solution f to the minimization problem (1.4). Although it is a de-
generate constrained minimization problem in innite dimension, the introduction of a
modied energy allows to recover the expected properties of the linearized Hamiltonian.
Theorem 3.6. Assume that f T
1
h
is a solution to the minimization problem (1.4) with
Lagrange multiplier . Assume moreover that the functional G
h

0
,
1
is adapted to the pair
(f, ) according to Denition 3.3 and set
0
=

0
(|f|
2
F
h
) and
1
=

1
(f [ N
h
f)
F
h
).
Then f is a spectrally stable equilibrium for the energy G
h

0
,
1
provided that

0

1

1
2h 2 ,
1
> 0 .
The spectrum (JHess G
h

0
,
1
(f)) is made of a discrete set of eigenvalues (i
n
)
nZ
with
nite multiplicity such that
n
R,
n
=
n
and lim
n
[
n
[ = +.
According to [HMRW], the spectral stability simply says that the spectrum of the
linearized Hamiltonian
_
0 Id
Id 0
_
Hess G
h

0
,
1
(f) = JHess G
h

0
,
1
(f)
is purely imaginary. According to the notation of Appendix A, J denotes the matrix
_
0 Id
Id 0
_
in the real symplectic space H T
h
. We recall that the spectrum of a
linear Hamiltonian has two symmetries with respect to R and iR. Note however that in
the spectrally stable case, a pure imaginary spectrum does not mean that the Hamiltonian
is anti-adjoint (see Appendix A).
12
Lemma 3.7. With the same assumptions as in Theorem 3.6, the Hessian Hess G
h

0
,
1
(f)
equals
Hess G
h

0
,
1
(f) = Hess G
h
(f) 2 Id +4
0
_
[f
R
)f
R
[ [f
R
)f
I
[
[f
I
)f
R
[ [f
I
)f
I
[
_
+ 4
1
_
[N
h
f
R
)N
h
f
R
[ [N
h
f
R
)N
h
f
I
[
[N
h
f
I
)N
h
f
R
[ [N
h
f
I
)N
h
f
I
[
_
with
0
=

0
(|f|
2
F
h
) and
1
=

1
(f [ N
h
f)
F
h
) Except for the lower bound which now
depends on
0
and
1
it shares the properties of Hess G
h
stated in Proposition 2.3.
Proof: A direct calculation leads to

T
Hess
0
_
|f|
2
F
h
_
= 2

0
(|f|
2
F
h
) ||
2
F
h
+

0
_
|f|
2
F
h
_
4 (Ref [ )
F
h
)
2

T
Hess
1
(f [ N
h
f)
F
h
) = 2

1
(f [ N
h
f)
F
h
) [N
h
)
F
h
+

1
(f [ N
h
f)
F
h
) 4 (ReN
h
f [ )
F
h
)
2
We conclude with

0
(|f|
2
F
h
) = ,

1
(f [ N
h
f)
F
h
) = 0 and with the identication
between =
R
+ i
I
with
_

R

I
_
.
Proof of Theorem 3.6: We refer again to the general framework reviewed in Ap-
pendix A, namely Proposition A.5 with A = 2N
h
and B = Hess G
h

0
,
1
2N
h
. The
above expression for Hess G
h

0
,
1
combined with Proposition 2.3 and the fact that f
T
2
h
= D(N
h
), implies that B is a bounded real operator. Proposition A.5 states that the
spectral stability can be deduced from
T
1
h
,
T
Hess G
h

0
,
1
(f) 0 .
Any element of T
1
h
can be written + f with f

(i.e. Re f, )
F
h
= 0) and R.
We get
( + f)
T
Hess G
h

0
,
1
(f)( + f) =

T
(Hess G
h
(f) 2) + 2
T
(Hess G
h
(f) 2)f +
2
f
T
(Hess G
h
(f) 2)f
+4
0

2
|f|
4
F
h
+4
1
(Re[ N
h
f)
F
h
)
2
+8
1
f [ N
h
f)
F
h
Re[ N
h
f)
F
h
+4
1

2
f [ N
h
f)
2
F
h
.
The rst term
T
(Hess G
h
(f) 2) is non negative according to Proposition 2.5. Since
is real and f

the scalar products

T
f vanish. Proposition 2.3 (it is shorter to
reproduce the calculation of G
2
in its proof) leads to

T
(Hess G
h
(f) 2)f = 2 Re[ (N
h
+ h)f)
F
h
+
(4 + 2)Na
h
Re[
h
_
e

|z|
2
h
[f[
2
f
_
)
F
h
2 Re[ f)
F
h
13
Then the Euler-Lagrange equation (2.2) implies for f

2
T
(Hess G
h
(f) 2)f = 8 Re[ (N
h
+ h )f)
F
h
= 8 Re[ N
h
f)
F
h
and

2
f
T
(Hess G
h
(f) 2)f = 8
2
( h) 8
2
f [ N
h
f)
F
h
.
Adding all the terms leads to
( + f)
T
Hess G
h

0
,
1
(f)( + f)
8 Re[ N
h
f)
F
h
+ 8
2
( h) 8
2
f [ N
h
f)
F
h
+ 4
0

2
+4
1
(Re[ N
h
f)
F
h
)
2
+ 8
1
f [ N
h
f)
F
h
Re[ N
h
f)
F
h
+ 4
1

2
f [ N
h
f)
2
F
h
4
2
(2 2h +
0
)
+4
1

2
_
r
2
+ 2
_
f [ N
h
f)
F
h

1

1
_
r +f [ N
h
f)
F
h
_
f [ N
h
f)
F
h

2

1
__
4
2
(2 2h +
0
) + 4
1

2
_
_
r +f [ N
h
f)
F
h

1

1
_
2

2
1
_
by setting r =
Re| N
h
f
F
h

and for
1
,= 0 . Finally the last right-hand side is non negative
for
1
> 0 and
0

1
1
2h + 2 > 0.
4 Approximation by a nite dimensional problem
The approximation of the optimization problem (1.4) by nite dimensional ones, that is T
h
is replaced by a set of polynomials with bounded degree, was studied in [ABN2]. Here we
complete this information by showing that such a convergence result can be extended to
the linearized Hamiltonian in the norm resolvent sense. We end this section by recalling
that a more quantitative estimate of the convergence of spectral elements, in such a
discretization process, is a real issue because the linearized Hamiltonian JHess G
h
(f)
is not anti-adjoint.
4.1 Preliminaries
For K N, C
K
[z] denotes the set of polynomials with degree smaller than or equal to
K. Since (c
n,h
z
n
)
nN
, c
nh
=
1
(h)
1/2
h
n/2

n!
, is an orthonormal spectral basis for N
h
with
N
h
z
n
= hnz
n
, the orthogonal projection
h,K
onto C
K
[z] coincides with the orthogonal
spectral projection:

h,K
= 1
[0,hK]
(N
h
) .
We shall use the notation

h,K
for the imbedding from C
K
[z] into T
h
:

h,K

h,K
= Id
C
K
[z]

h,K

h,K
=
h,K
. (4.1)
14
We introduce the reduced minimum of the nite dimensional optimization problem:
e
h
LLL,K
= min
PC
K
[z],P
F
h
=1
G
h
(P) . (4.2)
Theorem 4.1. 1) The minima e
h
LLL
and e
h
LLL,K
satisfy
K N (h
1
C
2
(h), +), 0 < e
h
LLL,K
e
h
LLL

C
2
(h)
2
+ C
2
(h)
3
(1 C
2
(h)(hK)
1
)
4
(hK)
1
where C
2
(h) =
8
h
3
_
2bNa
h
+ o
Na
(h
1/2
) does not depend on K.
If f solves the minimization problem (1.4) then the sequence (f
K
)
KN
dened by f
K
=
|
h,K
f|
1

h,K
f, which satises f
K
C
K
[z] is a minimizing sequence for (1.4).
2) If for any K N, P
K
C
K
[z] denotes any solution to (4.2) then the sequence (P
K
)
KN
is a minimizing sequence for (1.4). Its accumulation points for the | |
F
h
topology are
solutions of (1.4). Moreover if a subsequence (P
Kn
)
nN
converges to f in T
h
then the
convergence also holds in T
2
h
according to:
lim
n
|f P
K
|
F
h
+|N
h
(f P
Kn
)|
F
h
= 0 .
4.2 Convergence of the linearized Hamiltonian
We now consider the question of the convergence of the linearized Hamiltonians associated
with the functional G
h

0
,
1
. We forget the term related to the gradient of the functionals
since in the end it will be applied with critical points. The linearized Hamiltonian at a
point T
h
is dened as
H

() := JHess G
h

0
,
1
()
with J =
_
0 Id
Id 0
_
=
nN
_
0 1
1 0
_
in T
h
T
h
=
nN
(Cz
n
Cz
n
) . We keep
the notation
h,K
for the diagonal operator

h,K
:=
_

h,K
0
0
h,K
_
in T
h
T
h
.
Due to the commutation
h,K
J = J
h,K
the restricted linearized Hamiltonian at a point
of G
h

0
,
1

C
K
[z]
equals
H
K
(
K
) =
h,K
_
JHess G
h

0
,
1
(
K
)
_

h,K
= J
h,K
_
Hess G
h

0
,
1
(
K
)
_

h,K
.
For any holomorphic function in an open subset C and any compact regular contour
which does not meet the spectrum (H

()) the holomorphic functional calculus

provides the operators

(H

()) =
1
2i
_

(z)(z H

())
1
dz (4.3)
with a corresponding denition for

(H
K
(
K
)) .
15
Theorem 4.2. With the notations of Theorem 4.1, let (P
Kn
)
nN
denote a converging
subsequence of solutions of (4.2) with K = K
n
, and let f denote the limit f = lim
n
P
Kn
which is a solution to (1.4). Then for all z C (H

(f)), the convergence

lim
n

h,Kn
(z Id
C
Kn
[z]C
Kn
[z]
H
Kn
(P
Kn
))
1

h,Kn
= (z H

(f))
1
holds in the norm topology. Hence for any pair (, ) (see (4.3)) such that (H

(f)) =
the convergence
lim
n

h,Kn

(H
Kn
(P
Kn
))
h,Kn
=

(H

(f))
holds in the norm topology.
We start with a lemma derived after the introduction of a Grushin problem (see [SjZw]
and references therein), a more exible variation of the Feshbach method (see [DeJa] and
references therein).
Lemma 4.3. In a complex Hilbert space (H
C
, . , .)), let A be a self-adjoint operator with
domain D(A) and with a compact resolvent. Let (
n
)
nN
be a sequence of complex numbers
such that lim
n

n
= 1 . Let (B
n
)
nN
be a sequence of bounded operators with limit B
in the norm topology as n : lim
n
|B
n
B| = 0 . Let
n
be the spectral projection
1
[Tn,Tn]
(A) with the assumption lim
n
T
n
= +. The imbedding Ran
n
H is
denoted by

n
according to (4.1). Then the limit
lim
n

n
(z Id
Ran n

n
(
n
A + B
n
)

n
)
1

n
= (z (A+ B))
1
holds in the norm topology for all z C (A+ B).
Proof: 1) We rst consider the case
n
= 1 for all n N. We set
= sup |B
n
| , n N |B| .
For z C, we set A
nz
= z (A + B
n
) : D(A) H. After the decomposition H =
Ran
n
Ran(1
n
) and D(A) = Ran
n
(D(A) Ran(1
n
), it is written:
A
nz
=
_
A
nn
nz
A
n n
nz
A
nn
nz
A
n n
nz
_
=
_
A
nn
nz
B
n n
n
B
nn
n
A
n n
nz
_
with X
nn
=
n
X

n
, X
n n
=
n
X(1
n
)

, X
nn
= (1
n
)X

n
and X
n n
= (1

n
)X(1
n
)

. Accordingly we use the notation A

,z
for z (A + B) with for n N
xed the corresponding restrictions A
nn
,z
, A
n n
,z
= B
n n
, A
nn
,z
= B
nn
and A
n n
,z
. We follow
[SjZw] for the introduction of the Grushin problem and we set:
/
nz
=
_
A
nz
R
n

R
n
+
0
_
=
_
_
A
nn
nz
B
n n
n
Id
Ran n
B
n n
n
A
n n
nz
0
Id
Ran n
0 0
_
_
: D(A) Ran
n
HRan
n
16
with
R
n

=
_
Id
Ran n
0
_
and R
n
+
= (Id
Ran n
, 0) .
The Schur complement formula says that if
/
1
nz
=
_
E
n
E
n
+
E
n

E
n
+
_
the operator A
nz
is invertible if and only if E
n
+
is invertible with
A
1
nz
= E
n
E
n
+
(E
n
+
)
1
E
n

, (E
n
+
)
1
= R
n

A
1
nz
R
n
+
. (4.4)
We now compute /
1
nz
. First note that for any z C there exists n(z) N such that A
n n
nz
is invertible for n n(z) . The second resolvent formula gives:
(A
n n
nz
)
1
= (z (1
n
)(A+ B
n
)(1
n
)

)
1
=
_
1 + (z A
n n
)
1
B
n n
n

1
(z A
n n
)
1
where the self-adjoint operator A
n n
= (1
n
)A(1
n
)

has a spectrum included in

R [T
n
, T
n
] . This also implies for n n(z):
_
_
_
_
1 + (z A
n n
)
1
B
n n
n

1
_
_
_ 2 and
_
_
_(A
n n
nz
)
1
_
_
_
2
[[z[ T
n
[
n
0 .
In the former calculation B
n
can be replaced by B so that
_
_
_(A
n n
nz
)
1
_
_
_
2
[[z[ T
n
[
,
_
_
_(A
n n
z
)
1
_
_
_
2
[[z[ T
n
[
(4.5)
hold for n n(z) .
The inverse /
1
nz
is computed by Gauss elimination:
/
1
nz
=
_
_
0 0 1
0 (A
n n
nz
)
1
(A
n n
nz
)
1
B
nn
n
1 B
n n
n
(A
n n
nz
)
1
A
nn
nz
+ B
n n
n
(A
n n
nz
)
1
B
nn
n
_
_
The Schur complement formula (4.4) yields:
A
1
nz
= (1
n
)

(A
n n
nz
)
1
(1
n
)+
_
1
(A
n n
nz
)
1
B
nn
n
_
_
A
nn
nz
B
n n
n
(A
n n
nz
)
1
B
nn
n
_
1
_
1, B
n n
n
(A
n n
nz
)
1
_
(4.6)
when
E
n
+
= A
nn
nz
B
n n
n
(A
n n
nz
)
1
B
nn
n
is invertible.
When z C(A+B), A
,z
is invertible and the Schur complement formula (4.4) applied
with B
n
replaced by B implies that the operator

E
n
+
= A
nn
z
B
n n
(A
n n
z
)
1
B
nn
17
is invertible with _
_
_
_
_

E
n
+
_
1
_
_
_
_
=
_
_
R
n

A
1
z
R
n
+
_
_
uniformly bounded for n n(z) . The second inequality of (4.5) implies
_
_
_A
nn
z


E
n
+
_
_
_
2
2
[[z[ T
n
[
, for n n(z)
and a uniform bound for |(A
nn
z
)
1
| with respect to n n(z). Owing to the convergences
|A
nn
nz
A
nn
z
| |B B
n
|
n
0
and
_
_
E
n
+
A
nn
nz
_
_

_
_
_B
n n
n
_
A
n n
n,z
_
1
B
nn
n
_
_
_
2
2
[[z[ T
n
[
n
0 ,
the same is true for |(A
nn
nz
)
1
| and
_
_
E
n
+
_
_
and we get
_
_
(E
n
+
)
1
(A
nn
nz
)
1
_
_
C
z
_
_
_B
n n
n
_
A
n n
n,z
_
1
B
nn
n
_
_
_
2C
z

2
[[z[ T
n
[
n
0 . (4.7)
We infer from (4.5), (4.6) and (4.7) the estimate
_
_
A
1
nz

n
(A
nn
nz
)
1

n
_
_

C

z
[[z[ T
n
[
n
0 .
The second resolvent formula also gives
_
_
A
1
nz
A
1
z
_
_
=
_
_
(z AB
n
)
1
(z A B)
1
_
_
n
0 ,
which yields the result for
n
= 1 .
2) For the general case, it is enough to write
(z
n
(
n
A+B
n

n
)
1
=
1

n
(
z

n
(A+
1

n
B
n
)

n
)
1
=
1

n
(z
n
(A+

B
n
)

n
)
1
with

B
n
=
1
n
B
n
+ (1
1
n
) Id and to apply the result of Part 1).
Proof of Theorem 4.2: we recall that
G
h

0
,
1
(f) = G
h
(f) +
0
_
|f|
2
F
h
_
+
1
(f [ N
h
f)
F
h
)
and its Hessian at a point f T
+2
h
equals according to Lemma 3.7 (without assuming

1
(f [ N
h
f)
F
h
) = 0):
Hess G
h

0
,
1
(f) = Hess G
h
(f) 2

0
(|f|
2
F
h
) Id 2

1
(f [ N
h
f)
F
h
)N
h
+ 4

0
(|f|
2
F
h
)
_
[f
R
)f
R
[ [f
R
)f
I
[
[f
I
)f
R
[ [f
I
)f
I
[
_
+ 4

1
(f [ N
h
f)
F
h
)
_
[N
h
f
R
)N
h
f
R
[ [N
h
f
R
)N
h
f
I
[
[N
h
f
I
)N
h
f
R
[ [N
h
f
I
)N
h
f
I
[
_
.
18
When f solves (1.4) and for well chosen
0
and
1
, it writes:
Hess G
h

0
,
1
(f) = Hess G
h
(f) 2 Id+4
0
_
[f
R
)f
R
[ [f
R
)f
I
[
[f
I
)f
R
[ [f
I
)f
I
[
_
+4
1
_
[N
h
f
R
)N
h
f
R
[ [N
h
f
R
)N
h
f
I
[
[N
h
f
I
)N
h
f
R
[ [N
h
f
I
)N
h
f
I
[
_
= 2
_
N
h
0
0 N
h
_
+

B,
where

B L(T
h
T
h
) . Meanwhile we obtain for P
Kn
(with lim
n
P
Kn
= f in T
2
h
):
Hess G
h

0
,
1
(P
Kn
) = Hess G
h
(P
Kn
) 2

0
(|P
Kn
|
2
F
h
) Id 2

1
(P
Kn
[ N
h
P
Kn
)
F
h
)N
h
+4

0
(|P
Kn
|
2
F
h
)
_
[P
Kn,R
)P
Kn,R
[ [P
Kn,R
)P
Kn,I
[
[P
Kn,I
)P
Kn,R
[ [P
Kn,I
)P
Kn,I
[
_
+4

1
(P
Kn
[ N
h
P
Kn
)
F
h
)
_
[N
h
P
Kn,R
)N
h
P
Kn,R
[ [N
h
P
Kn,R
)N
h
P
Kn,I
[
[N
h
P
Kn,I
)N
h
P
Kn,R
[ [N
h
P
Kn,I
)N
h
P
Kn,I
[
_
= 2(1

1
(P
Kn
[ N
h
P
Kn
)
F
h
))
_
N
h
0
0 N
h
_
+

B
n
with
lim
n

1
(P
Kn
[ N
h
P
Kn
)
F
h
) = 0, lim
n
_
_
_

B
n

B
_
_
_ = 0 .
By recalling
H

(f) = JHess G
h

0
,
1
(f) and H

(P
Kn
) = JHess G
h

0
,
1
(P
Kn
)
and by applying Lemma 4.3 with
A = 2iJ
_
N
h
0
0 N
h
_
=
_
0 2iN
h
2iN
h
0
_
,

n
= (1

1
(P
Kn
[ N
h
P
Kn
)
F
h
))
B = iJ

B , B
n
= iJ

B
n
and
n
=
_

h,Kn
0
0
h,Kn
_
=
h,Kn
, T
n
= 2hK
n
,
one gets
lim
n

h,Kn
(z Id
C
Kn
[z]C
Kn
[z]

h,Kn
H

(P
Kn
)

h,Kn
)
1

h,Kn
= (z H

(f))
1
for all z C (H

(f)) . We conclude with

H
Kn
(P
Kn
) =
h,Kn
H

(P
Kn
)

h,Kn
.
Finally the convergence of the spectral elements

h,Kn

(H
Kn
(P
Kn
))
h,Kn
comes from the
fact that all the convergence estimates are locally uniform in z for h > 0 xed.
19
4.3 Remarks about the stability of spectral quantities
Contrary to Theorem 4.1, the results of Theorem 4.2 about the stability of spectral quan-
tities does not provide any quantitative estimate. For a xed not so small value of h > 0
and when only a xed nite number of spectral elements are computed, such a quanti-
tative estimate is not crucial. It becomes denitely an issue when h > 0 gets small or
if one is interested in a large number of spectral elements. For example, the behaviour
of the sequence (i
n
)
nZ
of eigenvalues of the linearized Hamiltonian JHess G
h

0
,
1
(f)
stated in Theorem 3.6, lim
n
[
n
[ = +, can be stated more accurately since the prob-
lem amounts to looking at a bounded perturbation of the harmonic oscillator quantum
Hamiltonian. Nevertheless this behaviour seems dicult to recover in numerical simula-
tions
3
A very likely explanation is that the linearized Hamiltonian JHess G
h

0
,
1
(f) is
not anti-adjoint. Whatever the choices of the functions
0
and
1
are, the Hessian
Hess G
h

0
,
1
(f) is a nite rank perturbation of Hess G
h
(f) according to Lemma 3.7. But
the commutator
_
J, Hess G
h
(f)

can be computed from (2.9) and equals

_
2C 2B
2B 2C
_
.
The operators B and C, dened in (2.11)(2.12), are non vanishing Hilbert-Schmidt op-
erators but with innite rank. Hence the linear Hamiltonian JHess G
h

0
,
1
(f) is not
anti-adjoint in spite of a purely imaginary spectrum.
The stability of the spectrum of non self-adjoint (or non normal) operators with respect
to perturbations enters in the theory of pseudospectral estimates and it is known that
there can be a big gap between the knowledge of the spectrum and a good control of resol-
vent estimates with dramatic consequences in numerical computations. Such an analysis
for pseudo-dierential operators has had a great development in the recent years and we
refer the reader to [Tre][Dav1][Dav2][DSZ][Hag1][Hag2][HerNi][HelNi][Pra][Zwo]. In order
to perform such an analysis of the linearized Hamiltonian JHess G
h
(f), a better infor-
mation on the minimizer f than the one provided in [AfBl][ABN1][ABN2] is necessary.
A Specic innite dimensional Hamiltonian systems
Our aim is not here to give a complete account on innite dimensional Hamiltonian
systems. We refer the reader for example to [ChMa][Kuk][BHK] for a more general pre-
sentation or dierent points of view. We simply briey point out the properties which are
relevant to our problem.
We consider a separable Kaehler space (H, (. [ .), ): (H, (. [ .)) is a real Hilbert space
while is a symplectic form compatible with (. [ .). We recall that the last condition
means that there exists a continuous R-linear (skew-adjoint) operator on H such that
3
according to discussions with A. Aftalion and X. Blanc.
20
J
2
= 1 and
(X, Y ) = (JX [ Y ) = (X[ JY ) .
Before going further, it is useful to introduce the complexied Hilbert space H
C
with the
scalar product
(f
1
+ if
2
[ g
1
+ ig
2
)
C
= (f
1
[ g
1
) + (f
2
[ g
2
) + i (f
1
[ g
2
) i (g
1
[ f
2
) .
In this framework the operator J becomes a skew-adjoint bounded involution
J

= J, J
2
= Id
which diers from i Id.
Remark A.1. It is important to note here that the complexied Hilbert space H
C
has
nothing to do with the natural complex structure associated with J. In fact, the complexi-
ed space has no relationship with the symplectic structure on H. It is introduced only in
order to provide the framework for spectral theory. More precisely, consider the example
where H = R
2n
= R
n
x
R
n

is endowed with
the scalar product: (X[ X

) =
_
x

_
.
_
x

_
= xx

and the symplectic form: (X, X

) = x

.
Let J =
_
0 1
1 0
_
. After the identication between X R
2n
and z = x + i C
n
, the
real scalar product happens to be the real part of the complex scalar product (X [ X

) =
Re z.z

and the symplectic form the opposite of the imaginary part (X, X

) = 1mz.z

,
while the operator J is translated into the multiplication by i . Instead, the complexied
space H
C
equals C
2n
and allows the action of J =
_
0 1
1 0
_
and the componentwise
multiplication by the complex scalar i .
In the case of our analysis, the Kaehler space is the complex Hilbert space T
h
. We have,
as a set, H = T
h
, after the identication between f = f
R
+ if
I
T
h
and
_
f
R
f
I
_
H
while H
C
equals T
h
T
h
.
As in the study of second variations, some properties and symmetries of a linearized
Hamiltonian are more obvious when working with the real structure (the complexication
being added only in order to apply spectral theory) .
The energy functional is given by
H(f) =
1
2
(f [ Af)
C
+
1
2
(f [ Bf)
C
+ h(f), f D(A) H
C
where the operators A, B and the nonlinear function h satisfy the next assumptions:
21
Hypothesis A.2. The operator (A, D(A)) is a non negative self-adjoint opera-
tor on H
C
, with a compact resolvent, which commutes with J and which is real:
A(D(A) H) H.
The operator B is a bounded real (BH H) self-adjoint operator on H
C
(non
necessarily commuting with J).
The function h : H
C
R is real analytic and satises the gauge invariance
h(e
J
f) = h(f) for all R and all f H
C
.
The Hamilton equation can be written as
_

t
f = JH(f) = JAf JBf Jh(f)
f(t = 0) = f
0
H (or H
C
) .
(A.1)
where denotes the gradient with respect to the scalar product (. [ .) in the real case and
the gradient with respect to the real scalar product Re (. [ .)
C
in the complex case. As
usual an equilibrium is a critical point of H .
Proposition A.3. Assume Hypothesis A.2. Then the initial value problem (A.1) admits
a unique mild global solution for any f
0
H
C
. Moreover the ow dened by f(t) = (t)f
0
for f
0
H
C
and t R satises
f
0
H, t R, (t)f
0
H
and f
0
H
C
, t R, |(t)f
0
|
H
C
= e
[B,J]t/2
|f
0
|
H
C
.
Proof: 1) The linear case with B = 0: Since the operator A has a compact
resolvent, commutes with the involution J and is real, it admits an orthonormal basis of
real eigenvectors
n
H, J
n
H, n N with
A
n
=
n

n
and AJ
n
=
n
J
n
so that
n
R and lim
n
[
n
[ = +.
The operator iJA is self-adjoint with domain D(A) and writes in H
C
=
nN
(C
n

CJ
n
) as the block diagonal operator
iJA =
nN
_
0 i
n
i
n
0
_
.
Hence the equation
i
t
f = iJAf
is solved by the unitary strongly continuous group
_
e
it(iJA)
_
tR
= (e
tJA
)
tR
, which
admits the explicit block diagonal expression
e
tJA
=
nN
_
cos(t
n
) sin(t
n
)
sin(t
n
) cos(t
n
)
_
.
22
Hence this linear evolution preserves the H
C
-norm, the domain D(A) and reality.
2) Local existence for the nonlinear case: The Duhamel formula
f(t) = e
tJA
f
0

_
t
0
e
(ts)JA
J(Bf(s) +h(f(s))) ds .
and the analyticity assumption on h allow to use the standard xed point argument
in (
0
([0, T
f
0
] ; H
C
) . The xed point provides the real analyticity of f(t) with respect
to f
0
H
C
. Finally, the uniqueness in (
0
([0, T
f
0
]; H
C
) and the fact that the integral
equation can be solved in (
0
([0, T
f
0
]; H) ensures that f(t) H for all t [0, T
f
0
] as soon
as f
0
H.
3) Approximation with a bounded generator: In order to establish the preserved
quantities, we approximate the linear operator by bounded ones. Let A

= 1
[0,]
(A)A.
By the spectral theorem, we get
_
_
(e
tJA

e
tJA
)f
_
_
2
_
_
(e
it(iJA

)
e
it(iJA)
)f
_
_
2
=
_

1 e
it

2
d
f
()
where d
f
is the spectral measure of the given element f T
h
with respect to the self-
ajoint operator iJA. Hence by dominated convergence we get:
t R, s-lim

e
tJA

= e
tJA
.
We write the dierence between the two integral equations:
f

(t) = e
tJA

f
0

_
t
0
e
(ts)JA

J(Bf

(s) +h(f

(s))) ds
and f(t) = e
tJA
f
0

_
t
0
e
(ts)JA
J(Bf(s) +h(f(s))) ds
as
f(t)f

(t) =
_
e
tJA
e
tJA

_
f
0

_
t
0
_
e
(ts)JA
e
(ts)JA

_
J(Bf(s)+h(f(s))) ds

_
t
0
e
(ts)JA

JB(f(s) f

(s)) ds
_
t
0
e
(ts)JA

J (h(f(s)) h(f

(s))) ds .
For a xed f
0
and a xed t [0, T
f
0
], the analyticity assumption on h and the fact that
e
tJA
and e
tJA

are unitary operators, lead to

|f f

|
2

f
0
() +C
f
0
_
0,t
|f(s) f

(s)|
2
ds
with lim

f
0
() = 0 . By the Gronwall Lemma, we obtain for any f
0
H
C
the
existence of T
f
0
such that
t [0, T
f
0
], lim

|f(t) f

(t)| = 0 .
23
4) Upper bound for the norm and global existence: According to the rst step,
we can reduce the analysis to the case where A is a bounded operator. Then the local in
time mild solution is a classical solution for any f
0
H
C
. We compute

t
|f|
2
= (
t
f [ f)
C
+ (f [
t
f)
C
= 2 Re (JAf + JBf + Jh(f) [ f)
C
= 2 Re (f [ Jh(f))
C
(JBf [ f)
C
+ (f [ JBf)
C
2 Re (f [ Jh(f))
C
+|[J, B]| |f|
2
Here we dierentiate the gauge invariance of h, h(e
J
f) = f,:
0 =
d
d

h(e
J
f)

=0
= Re (h(f) [ Jf)
C
.
We have proved
t
|f|
2
|[B, J]| |f|
2
when f solves (A.1) with A L(H
C
). The
inequality |f(t)| e
[B,J]t/E
|f
0
| can be extended to the case of unbounded A according
to step 3).
Finally this norm control provides the existence of a global in time solution, T
f
0
= +,
for any f
0
H
C
.
We now consider the conservation of energy under the additional assumption that the
ow (t) preserves the domain D(A). This will be checked in the proof of Proposition A.8
below. We refer to [ChMa] for a more general statement.
Proposition A.4. Under Hypothesis A.2 and if the ow (t) preserves the domain D(A)
in the sense that the solution f to (A.1) belongs to (
0
(R; D(A)) when f
0
D(A), then
the equality
H(f(t)) = H((t)f
0
) = H(f
0
)
holds for any t R and any f
0
D(A) .
Proof: If f (
0
(R; D(A)), then the mild solution to (A.1) is a strong solution,
f (
1
(R; H
C
). Since the gradient of H equals Af + Bf +h(f) we write:

t
H(f) = Re (Af + Bf +h(f) [
t
f)
C
= Re ((A+ B)f [ J(A+ B)f)
C
Re ((A+ B)f [ Jh(f))
C
Re (h(f) [ J(A+ B)f)
C
Re (h(f) [ Jh(f))
C
= 0 .
We now give some applications in specic cases arising in our analysis.
Proposition A.5. Under Hypothesis A.2 with h = 0, then the Hamiltonian vector J(A+
B) denes a linear closed unbounded operator on H
C
with domain D(J(A+B)) = D(A).
It has a compact resolvent and its spectrum has symmetries with respect to the two axes
R and iR.
Moreover if the energy H(f) =
1
2
(f [ (A+B)f) is non negative for all f D(A) H then
(J(A + B)) iR.
24
Remark A.6. Note that although (J(A+B)) iR, the operator J(A+B) is not anti-
adjoint (even in nite dimension), except when J and A + B commute, [J, B] = 0. The
nite dimensional version of the nal result is a very specic case of the classication of
quadratic Hamiltonian functions which is reviewed in [Arn]-Appendix 6.
Especially when applying this Proposition, a good identication of the Kaehler structure
on H and the two dierent complex structures on H
C
is useful.
Proof: Since the operator is a bounded perturbation of JA, the rst statements
are standard (see [ReSi]). Concerning the symmetries of the spectrum, the following
equivalences hold:
( , (J(A + B))) ((J(A + B) ) invertible)

_
J((A + B)J )J
1
invertible
_

_
((A + B)J )

= (J(A+ B) ) invertible
_

_
, (J(A + B))
_
and provide the symmetry with respect to iR. For the second symmetry, we introduce the
conjugate u of any vector u H
C
as the symmetric vector with respect the real subspace
H. Since J, A and B are real operators, we obtain
4
for any u H, u ,= 0,
(J(A + B)u = u)
_
J(A + B)u = u
_
.
Since J(A + B) has a compact resolvent, its spectrum is thus symmetric with respect
to R.
Finally, assume that the energy H(f) =
1
2
(f [ (A + B)f) is non negative for all f
D(A) H. Since (A + B, D(A)) is self-adjoint on H
C
, this means that A + B is a non
negative operator and we get
(H(u) = 0, u H
C
) (u Ker(A+ B)) .
Let (J(A + B)) be a non zero eigenvalue with eigenvector u
0
,= 0. Since u
0

D(A+ B) = D(A), u(t) = e
tJ(A+B)
u
0
(
0
(R; D(A)) and according to Proposition A.4,
the energy is conserved:
e
2 Re t
(u
0
[ (A+ B)u
0
)
C
=
_
e
t
u
0
[ (A+ B)e
t
u
0
_
C
= (u(t) [ (A+ B)u(t))
C
= (u
0
[ (A+ B)u
0
)
C
.
Since ,= 0, we get u
0
, Ker(A+ B), (u
0
[ (A+ B)u
0
)
C
,= 0 and Re = 0 .
Proposition A.7. Under Hypothesis A.2, a sucient condition for an equilibrium to be
spectrally stable, is that it is a local minimum.
4
Especially for this argument, it is preferable to forget the complex structure on H identifying J with
the multiplication by i .
25
Proof: According to [HMRW], an equilibrium is spectrally stable when the spectrum
of the linearized Hamiltonian is included in iR. At a critical point f of H, the linearized
Hamiltonian equals
JA JB JHess h(f) = J(A + B + Hess h(f)) .
If f is a minimum for H then A+B +Hess h(f) is non negative owing to the analyticity
property of h:
H(f + ) =
1
2
([ (A+ B + Hess h(f))) + O(||
3
) .
By replacing B by B+Hess h(f) in Proposition A.5, we get (J(A+B+Hess h(f)))
iR.
We end this appendix with a specic nonlinear Hamiltonian for which Noethers the-
orem ([Arn][AbMa]) can be stated in a very explicit form.
Proposition A.8. Assume Hypothesis A.2 with B = 0 and with the additional gauge
invariance for h:
R, f H, h(e
JA
f) = h(f) .
Then for any f
0
D(A) H, the solution f to (A.1) satises
f (
1
(R; H) (
0
(R; D(A))
t R, H(f(t)) = H(f
0
) , (f(t) [ Af(t))
C
= (f
0
[ Af
0
)
C
h(f(t)) = h(f
0
) and |f(t)| = |f
0
|.
Proof: The new gauge invariance implies that for any f, H and R:
h(f) + (h(f) [ ) + O(||
2
) = h(f + ) = h(e
JA
(f + ))
= h(e
JA
f) +
_
h(e
JA
f) [ e
JA

_
+ O(||
2
)
= h(f) +
_
e
JA
h(e
JA
f) [
_
+ O(||
2
) .
Hence we get
f H, R , h(e
JA
f) = e
JA
h(f) .
Thus the solution to (A.1) with f
0
H, f(t) = (t)f
0
, also satises
R, t R, e
JA
f(t) = (t)(e
JA
f
0
) .
The regularity of the ow with respect to initial data allows to say that for any t R,
e
JA
f(t) is dierentiable with respect to when f
0
D(A). This yields
(f
0
D(A))
_
f(.) = (.)f
0
(
0
(R; D(A))
_
.
Proposition A.4 gives
t R, H(f(t)) = H(f
0
)
26
when f
0
D(A) . It is enough to check that the quantity (f(t) [ Af(t)) does not vary. By
dierentiating the new gauge invariance with respect to we get now for any g D(A)H:
0 =
d
d
h(e
JA
g)

=0
= (h(g) [ JAg) .
For f(t) = (t)f
0
with f
0
D(A) H we compute:

t
(f(t) [ Af(t)) = 2 Re (
t
f [ Af(t))
= 2 Re (JAf(t) [ Af(t)) + 2 Re (Jh(f(t)) [ Af(t))
= 0 + 2 (h(f(t)) [ JAf(t)) = 0 .
Finally, the fact that |f(t)| = |f
0
| is a direct consequence of the equality h(e
J
f) = h(f).
Remark A.9. For the specic Hamiltonian considered in Proposition A.8, one can think
about several criteria for the formal stability and the nonlinear stability. This would
require additional discussions and again specic assumptions according to [HMRW]. We
do not consider such criteria in our analysis.
Acknowledgments: This work started while the author had a sabbatical semester at
CNRS and after the organization of the workshop Nonlinear spectral problems and mean
elds models supported by the French ACI Syst`emes hors-equilibre classiques et quan-
tiques. He was also supported by the ESF network SPECT for a short visit in the Erwin
Schrodinger Institute and he thanks J. Yngvason for his hospitality. Finally, the author
warmly thanks A. Aftalion and X. Blanc for the numerous discussions which stimulated
his interest in the models for Bose-Einstein condensates.
References
[ARVK] Abo-Shaeer JR, Raman C,Vogels JM, Ketterle W, Observation of Vortex Lattices
in Bose-Einstein Condensates (2001) Science 292, 476-479.
[AbMa] R. Abraham, J.E. Marsden Foundations of Mechanics. Addison-Wesley Publish-
ing Company (1985).
[Aft] A. Aftalion, Vortices in Bose Einstein condensates. Progress in Nonlinear Dieren-
tial Equations and Their Applications, Vol. 67, Birkhauser, 2006.
[AfBl] A. Aftalion, X. Blanc Vortex lattices in rotating Bose Einstein condensates.
Preprint.
[ABD] A. Aftalion, X. Blanc and J. Dalibard, Vortex patterns in fast rotating Bose Ein-
stein condensates, Phys. Rev. A 71, 023611 (2005).
27
[ABN1] A. Aftalion, X. Blanc and F.Nier, Vortex distribution in the lowest Landau Level,
Phys. Rev. A (2006) to appear.
[ABN2] A. Aftalion, X. Blanc and F. Nier A Lowest Landau Level functional and
Bargmann spaces in Bose Einstein Condensates. preprint (2006).
[Arn] V. Arnold Mathematical Methods of Classical Mechanics. Graduate Texts in Math-
ematics, Springer (1989).
[Bar] V. Bargmann, On a Hilbert space of analytic functions and an associated integral
transform. Comm. Pure Appl. Math. 14 1961 187214.
[BHK] G. Benettin, J. Henrard and S. Kuksin Hamiltonian dynamic theory and applica-
tions. Lect. Notes in Mathematics 1861, CIME Foundation Springer-Verlag (2005).
[BWCB] L. O. Baksmaty, S. J. Woo, S. Choi, and N. P. Bigelow, Tkachenko Waves in
Rapidly Rotating Bose-Einstein Condensates, Phys. Rev. Lett. 92, 160405 (2004).
[Car] E. Carlen Some Integral Identities and Inequalities for Entire Functions and Their
Application to the Coherent State Transform. J. Funct. Analysis 97 (1991).
[ChMa] P.R. Cherno and J.E. Marsden Properties of innite dimensional Hamiltonian
systems. Lect. Notes in Mathematics 425, Springer-Verlag (1974).
[Dav1] E.B. Davies. Pseudospectra, the harmonic oscillator and complex resonances. R.
Soc. London Ser. A Math. Phys. Eng. Sci. 455, p. 585-599 (1999).
[Dav2] E.B. Davies. Non-self-adjoint dierential operators. Bull. London Math. Soc.
34(5), p. 513-532 (2002).
[DeJa] J. Derezinski and V. Jaksic Spectral theory of Pauli-Fierz operators. J. Funct.
Anal. 180(2) (2001), pp 243327.
[DSZ] N. Dencker, J. Sjostrand, and M. Zworski. Pseudospectra of semi-classical
(pseudo)dierential operators. Comm. Pure Appl. Math. 57 (3), p. 384-415 (2004).
[Fol] G.B Folland Harmonic Analysis in Phase Space. Princeton University Press, 1989.
[GiJa] S.M.Girvin, T.Jach, Formalism for the quantum Hall eect: Hilbert space of ana-
lytic functions, Phys. Rev. B, 29 (1984) 5617-5625.
[Gro] L. Gross Hypercontractivity on complex manifolds. Acta Math. 182, (1999).
[Hag1] M. Hager Instabilite spectrale semiclassique pour des operateurs non auto-
adjoints I : un mod`ele. Preprint http://hal.ccsd.cnrs.fr
[Hag2] M. Hager Instabilite spectrale semiclassique pour des operateurs non auto-
adjoints II. Preprint http://hal.ccsd.cnrs.fr
28
[HelNi] B. Heler and F. Nier Hypoelliptic Estimates and Spectral Theory for Fokker-
Planck Operators and Witten Laplacians. Lect. Notes in Math. Vol. 1862, Springer
(2005).
[HerNi] F. Herau and F. Nier. Isotropic hypoellipticity and trend to the equilibrium
for the Fokker-Planck equation with high degree potential. Archive for Rational
Mechanics and Analysis 171 (2), p. 151-218 (2004).
[Ho] T. L. Ho, Bose-Einstein Condensates with Large Number of Vortices, Phys. Rev.
Lett. 87, 060403 (2001).
[HMRW] D. Holm, J.E. Marsden, T. Ratiu and A. Weinstein Nonlinear stability of uid
and plasma equilibria. Physics Reports 123(1-2) (1996), pp 1116.
[Kuk] S. Kuksin Hamiltonian perturbations of innite dimensional linear systems with
imaginary spectrum. Funct. Anal. Appl. 21(3) (1987) pp 192205.
[Mar] A. Martinez An Introduction to Semiclassical Analysis and Microlocal Analysis.
Universitext, Springer-Verlag, (2002).
[Nel] E. Nelson A quartic interaction in two dimensions in Mathematical Theory of Ele-
mentary Particles 6973, MIT Press (1966), 6973.
[Pra] K. Pravda-Starov Pseudospectrum for a class of semiclassical operators. preprint.
[ReSi] M. Reed and B. Simon. Method of Modern Mathematical Physics. Academic press,
(1975).
[SCEMC] V. Schweikhard, I. Coddington, P. Engels, V. P. Mogendor, and E. A. Cornell,
Rapidly Rotating Bose-Einstein Condensates in and near the Lowest Landau Level,
Phys. Rev. Lett. 92, 040404 (2004).
[SHMacD] J. Sinova, C. B. Hanna, and A. H. MacDonald, Quantum Melting and Absence
of Bose-Einstein Condensation in Two-Dimensional Vortex Matter, Phys. Rev. Lett.
89, 030403 (2002).
[SjZw] J. Sjostrand and M. Zworski Elementary linear algebra for advanced spectral
problems. eprint arXiv:math/0312166 (2003).
[Son] E.B. Sonin, Ground state and Tkachenko modes of a rapidly rotating Bose Einstein
condensate in the lowest Landau level state, Phys. Rev. A 72, 021606(R) (2005).
[Tre] L.N. Trefethen. Pseudospectra of linear operators. Siam Review 39 (3), p. 383-406
(1997).
[WBP] G. Watanabe, G. Baym and C. J. Pethick, Landau levels and the Thomas-Fermi
structure of rapidly rotating Bose-Einstein condensates, Phys. Rev. Lett. 93, 190401
(2004).
29
[Zwo] M. Zworski. Numerical linear algebra and solvability of partial dierential equa-
tions. Comm. Math. Phys. 229 (2), p. 293-307 (2002).
30