OLIVER MATTE
Mathematisches Institut, LudwigMaximiliansUniversit
at,
Theresienstrae 39, D80333 M unchen, Germany
matte@math.lmu.de
EDGARDO STOCKMEYER
Institut f
ur Mathematik, Johannes GutenbergUniversit
at,
Staudingerweg 9, D55099 Mainz, Germany
stockm@mathematik.unimainz.de
We prove an HVZ theorem for a general class of nopair Hamiltonians describing an atom
or positively charged ion with several electrons in the presence of a classical external
magnetic field. Moreover, we show that there exist infinitely many eigenvalues below the
essential spectrum and that the corresponding eigenfunctions are exponentially localized.
The novelty is that the electrostatic and magnetic vector potentials as well as a non
local exchange potential are included in the projection determining the model. As a main
technical tool, we derive various commutator estimates involving spectral projections of
Dirac operators with external fields. Our results apply to all coupling constants e2 Z < 1.
1. Introduction
The relativistic dynamics of a single electron moving in the potential of a static
nucleus, VC 0, in the presence of an external classical magnetic eld B = curl A
is generated by the Dirac operatora
DA,VC := (i + A) + VC . (1.1)
of light. Length is measured in units of /(mc), which is the Compton wave length divided by 2.
is Plancks constant divided by 2. In these units, the square of the elementary charge equals
the fine structure constant, e2 1/137.037.
1
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
1 1
+
A,VC = + sgn(DA,VC ). (1.2)
2 2
This is conrmed by Diracs interpretation of the negative spectral subspace as
a completely lled sea of virtual electrons which, on account of Paulis exclusion
principle, forces an additional electron to attain a state of positive energy. On the
other hand, it is well known that there is no canonical a priori given atomic or
molecular Hamiltonian generating the relativistic time evolution of N > 1 inter
acting electrons. Guided by nonrelativistic quantum mechanics one might naively
propose to start with the formal expression
N
(j)
DA,VC + Wjk , (1.3)
j=1 1j<kN
where the superscript (j) means that the operator below acts on the jth electron
and Wjk 0 is the interaction potential between the jth and kth electron. It then
turns out, however, that (1.3) suers from the phenomenon of continuum dissolution
which is also known as the BrownRavenhall disease [9]. That is, the eigenvalue
problem associated to (1.3) has no normalizable solutions; see, e.g., [43, 48]. A
frequently used ansatz to nd a reasonable and semibounded Hamiltonian for an
N electron atomic or molecular system again incorporates the concept of a Dirac
sea. Namely, one projects (1.3) onto the N fold antisymmetric tensor product of a
suitable oneelectron subspace, i.e. one considers operators of the form
N
(j)
HN := +,N
A,V
DA,VC + Wjk +,N
A,V , (1.4)
j=1 1j<kN
where
N
+(j)
+,N
A,V := A,V . (1.5)
j=1
Here + A,V is dened as in (1.1) and (1.2) but with VC replaced by a new potential
V . A Hamiltonian of this kind has been introduced rst by Brown and Raven
hall in [9]. We emphasize that HN can formally be derived from quantum elec
trodynamics (QED) by a procedure that neglects the creation and annihilation
of electronpositron pairs [47], the latter being dened with respect to +A,V . For
this reason operators of the form (1.4) are often called nopair Hamiltonians. Mod
els of this type are widely used as a starting point for numerical computations in
quantum chemistry. We refer the reader to the recent textbook [43] for a detailed
exposition of the application of nopair models in quantum chemistry, for examples
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
consider: Namely, for vanishing magnetic elds, it is known that the oneparticle
BrownRavenhall operator is stable if and only if Z Zc := e12 (2/)+(/2) 2
124.2 [17]. In the presence of an exterior Beld one can show that H1 is still
bounded below in the free picture, for all Z Zc , provided the vector potential
is locally bounded and Lipschitz continuous in a neighborhood of the nuclei [36].
(For smaller values of the coupling constant, one can actually prove the stability of
matter of the second kind in the free picture treating a gauge xed vector potential
as a variable in the minimization and adding the eld energy to the multiparticle
Hamiltonian; see [35, 34], where the quantized electromagnetic eld is considered.
In this situation it is essential that the vector potential is included in the projection
for otherwise the model is always unstable if N > 1 [21].)
Finally, we comment on some closely related recent work. In the free picture
and for vanishing exterior magnetic eld, an HVZ theorem and the existence of
innitely many eigenvalues below the essential spectrum have been proved in [40],
for nuclear charges Z Zc . The case N = 2 is also treated in [27]. A more general
HVZ theorem that applies to dierent particle species and a wider class of interac
tion potentials and exterior elds in the free picture has been established in [37].
Moreover, the reduction to irreducible representations of the groups of rotation
and reection and permutations of identical particles is considered in [37]. The L2 
exponential localization of eigenvectors in the BrownRavenhall model is studied
in [38,39] improving and generalizing earlier results from [2]. In all these works, the
authors employ explicit position space representations of + 0,0 . An HVZ theorem
in the free picture with constant magnetic eld is established in [28] again using
explicit representations for the projection based on Mehlers formula. By employing
somewhat more abstract arguments we are able to study a wider class of projections
in this paper. Similar results on the spectral projectors are used in [36] to study
the regularity of the eigenvectors of H1 in the free picture and to derive pointwise
exponential decay estimates for their partial derivatives of any order (for Z Zc
and assuming that all partial derivatives of A increase more slowly than any expo
nential function). The rate of exponential decay found in [36] is actually the same
as it is known for the Chandrasekhar operator and, hence, seems to be the optimal
one. For manyelectron atoms it is, however, more dicult to prove the exponential
localization and as in [2,38] we shall only get suboptimal bounds on the decay
rate in the present article.
For recent developments and numerous references to the literature on HVZ
theorems we refer to [33].
threshold energy. Section 6 deals with Weyl sequences and, nally, in Sec. 7 we
show that HN possesses innitely many eigenvalues below the threshold energy.
Some frequently used notation. Open balls in R3 with radius R > 0 and center
z R3 are denoted by BR (z). Spectral projections of a selfadjoint operator, T ,
on some Hilbert space are denoted by E (T ) and EI (T ), if R and I is an
interval. D(T ) denotes the domain of the operator T and Q(T ) its form domain.
The characteristic function of a subset M Rn is denoted by 1M . C, C , C , . . .
denote constants whose values might change from one estimate to another.
{i , j } = 2ij 1, 0 i, j 3. (2.1)
H := L2 (R3 , C4 )
with domain H 1 (R3 , C4 ). Its spectrum is purely absolutely continuous and given
by the union of two halflines,
Moreover, there exist (0, 1) and > 0 such that the balls B (y), y Y, are
mutually disjoint and, for 0 < x < and y Y,
VC (y + x) . (2.4)
x
Example 2.1. The main example for a potential satisfying Hypothesis 2.1 is cer
tainly the Coulomb potential generated by a nite number of static nuclei,
e2 Zy
VC (x) = 1, x R3 \Y.
x y
yY
VE is compact and has the following property: There exist m > 0 and some
increasing function c : [0, m) (0, ) such that, for every F C 1 (R3 , R) with
F  a < m,
Example 2.2. (i) Possible choices for VH and VE satisfying the conditions of
Hypothesis 2.2 are the Hartree and nonlocal exchange potentials corresponding
to a set of exponentially localized orbitals 1 , . . . , M H , i (x) Cemx ,
1 i M , for some C (0, ). Their Hartree potential is given as
M
2 2 1
VH (x) := e i  (x), x R3 .
i=1

It incorporates the presence of M electrons in a xed state into the Dirac sea
by a smeared out background density. The exchange potential corresponding
to 1 , . . . , M is the integral operator with matrixvalued kernel
M
i (x) (y)
VE (x, y) := e2 i
.
i=1
x y
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
emxy
VE (x, y) C ,
x x yy
DA,V = (i + A) + + V,
+
A,V := E[e0 ,) (DA,V ), +
A,V := 1 A,V , (2.9)
where
Many of our technical results on DA,V , for instance, various commutator estimates
of Sec. 3 hold actually true under the mere assumption that the components of A
are locally bounded. Of course, if not all eigenvalues of DA,V are larger than 1
and e0 is chosen between 1 and the lowest eigenvalue, the physical relevance of
the N particle Hamiltonian HN becomes rather questionable. We remark that such
situations are not excluded by our hypotheses. For instance, if VC is the Coulomb
potential and the intensity of a constant exterior magnetic eld is increased, then
the lowest eigenvalue of DA,VC eventually reaches the lower continuum [16]. Never
theless, our theorems hold for any choice of e0 as in (2.10).
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
where +,N 3 3
A,V is given by (1.5) and (2.9). We let W : R R [0, ] denote the
interaction potential between two electrons.
Proposition 2.1. Assume that V fullls Hypothesis 2.2, W fullls Hypothesis 2.3,
and that A L 3 3
loc (R , R ) satises e
0 x
A < , for some 0 0 <
min{0 , m}. Then the operator H N given by (2.13) and (2.14) is welldened,
symmetric, and semibounded from below.
Proof. The only claim that is not obvious is that Wij +,N A,V is again square
integrable, for every DN . This follows, however, from Corollary 3.3.
where SN is the group of permutations of {1, . . . , N }, and dene the nopair Hamil
tonian by
A
HN := HN AN H + .
N
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
Remark 2.1. In the case N = 1 the assertion of Theorem 2.1 still holds true under
the assumptions of Proposition 2.1. This follows from the proof of Theorem 2.1.
In fact, for N = 1, we do not have to control error terms involving the interaction
W which is the only reason why B is assumed to be bounded in Theorem 2.1. If also
V = VC , then we obtain an exponential localization estimate for an eigenvector,
E , with eigenvalue E (1, 1) of the Dirac operator DA,VC . The estimate on the
decay rate which could be extracted from our proof is, however, suboptimal due to
error terms coming from the projections; see also [7] for decay estimates for Dirac
operators.
Next, we introduce a hypothesis which is used to prove the easy part of the
HVZ theorem below.
Hypothesis 2.4. (i) For every 1, there exist radii, 1 R1 < R2 < ,
Rn , and 1 , 2 , . . . D such that
(ii) A C 1 (R3 , R3 ) and B = curl A has the following property: There are b1
(0, ) and 0 < min{0 , m} (m and 0 are the parameters appearing in
Hypothesis 2.2 and (2.11)) such that, for all x, y R3 ,
Example 2.3 ([22]). We recall a result from [22] which provides a large class
of examples where Hypothesis 2.4(i) is fullled: Suppose that A C (R3 , R3 ),
B = curl A, and set, for x R3 and N,
 B(x)
=
0 (x) := B(x), (x) := .
1+  B(x)
<
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
Proof. (i) follows directly from Lemma 6.3 and (ii) follows from Theorems 5.1
and 2.1.
To show the existence of innitely many eigenvalues below the bottom of the
A
essential spectrum of HN we certainly need a condition on the relationship between
VC , W , and the magnetic eld. To formulate it we set, for , R > 0,
S (R) := {x R3 : (1 )R x (1 + )R} (2.19)
v
(, R) := sup sup v  VC (x)vC4 . (2.20)
xS (R) v=1
or
where C0 (R3 , [0, 1]) equals 1 on B/2 (0) and 0 outside B (0). Here is the
parameter appearing in Hypothesis 2.1. We let VCs (x) = S(x)VCs (x) denote the
polar decomposition of VCs (x). By Hardys inequality we know that VCs is a bounded
operator from H 1 (R3 , C4 ) to L2 (R3 , C4 ). By duality and interpolation it possesses
a unique extension VCs L (H 1/2 , H 1/2 ). Given some A L 3 3
loc (R , R ) we set
s 3
A := (1 )A, where C0 (R , [0, 1]) is equal to 1 on some ball containing
supp(VCs ). We let As (x) = U (x) As (x) denote the polar decomposition of
As (x) and note that the operator sum D 0 + As + V s is welldened as an
C
1/2 1/2
element of L (Hc , Hc ), for every A L loc (R 3
, R 3
). So V s s
C and A have disjoint
support by denition. As a consequence the application of the following lemma
eventually becomes more convenient.
Proof. In [44, Proposition 4.3] it is observed that the claim follows from [8, The
orem 1.3] and [41, 42].
on the domain D(DA,V ) = D(DAs ,VCs ). Notice that in (3.2) we only add bounded
operators to DAs ,VCs . We state some of its properties in the following lemma where
De := { Hc1/2 (R3 , C4 ) : D
0 + A + V s L2 (R3 , C4 )}
C (3.4)
0 + A + VCs + (V VCs ).
DA,V = D
[DA,V , ] = i( ) + [VE , ].
Proof. Since VE is compact it is clear that all assertions hold true as soon
as they hold for VE = 0, which we assume in the following. To prove (a)
we write
1BR (0) (DA,V i)1 = (1BR (0) D0 1/2 )(D0 1/2 (DA,V i)1 ),
where C0 (R3 , [0, 1]) equals 1 in a neighborhood of BR (0). Then we use that
1BR (0) D0 1/2 is compact and that D0 1/2 (DA,V i)1 is bounded by Lemma 3.1
and the closed graph theorem. By standard arguments, we obtain the identity
ess (DA,V ) = ess (DA ) from (a) since V drops o to zero at innity; see, e.g., [49,
4.3.4]. The inclusion (DA ) (, 1] [1, ) follows from supersymmetry
arguments; see, e.g., [49, 5.6]. The assertions in (c) follow from [8, 2], (d) follows
from [8, Lemma G], and (e) from [41].
Next, we recall the useful resolvent identity (3.6) (see, e.g., [18, 53]) which is
used very often in the sequel. It should be regarded as a substitute for the second
resolvent identity which is typically not applicable in order to compare two dierent
Dirac operators in this paper. For, in general, the domain of one of these Dirac
operators is not included in the domain of the other. The vector potential A in
Eq. (3.6) below could for instance be the gradient of some gauge potential or just be
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
equal to zero. We recall another wellknown resolvent identity [41] in the beginning
of Sec. 3.5.
Lemma 3.3. Assume that V fullls Hypothesis 2.2, and that A, A L (R3 , R3 ).
loc
s
Let V be either VC (given by (3.1)) or 0, let z (DA, e ) (DA,V ) and
eV
C (R3 , R) be constant outside some ball in R3 , and assume that (VC V ) and
are bounded. Then
(A A)
RA,
eV
e (z)(V V + (A A))RA,V (z). (3.6)
DA,
eV
e = DA,V + {V + V (A A)}.
Therefore, we obtain
(RA,
eV e (z) RA,V (z))(DA,
eV e z)
= (RA,
eV e (z) RA,V (z))((DA,
eV e z) + i ())
+ (RA,
eV e (z) RA,V (z))i ()
+ (RA,
eV e (z) RA,V (z))i ()RA,
eV e (z)(DA,
eV e z).
As DA,
eV
e is essentially selfadjoint on De , we know that (DA,
eV
e z)De is dense,
which together with the calculation above implies
(RA,
eV e (z) RA,V (z)) = (RA,
eV e (z) RA,V (z))i ()RA,
eV e (z)
Notice that the decay rate in the following lemma is determined only by the decay
rate m appearing in Hypothesis 2.2 and the number 0 dened in (2.11). In the
next proof and henceforth we shall often use the abbreviations
We remark that, for V = 0, the bound (3.11) below follows from a wellknown
computation (see, e.g., [7]) which is recalled in the next proof. For nonvanishing,
singular potentials V a bound on the operator norm of the conjugated resolvent
seems to be less well known and the Neumanntype argument we use to prove it
might be a new observation.
Ca
eF RA,V (z)eF . (3.11)
1 + 2
Proof. First, we assume that F is constant outside some ball in R3 . Then it suf
ces to treat the case F 0, since otherwise we could consider the adjoint of
eF RA,V (z)eF .
Since F is smooth and constant outside some compact set a straightforward
calculation (see [7]) using Lemma 3.2(d) yields, for z C and D(DA ) E ,
1 F
e (DA z)eF 2 + 3 (i + A) 2
4
+ 3(1 + z2 ) 2 + 3  F 2
eF (DA + z)eF  eF (DA z)eF
= (i + A) 2 +  (1 z 2 F 2 ).
1 Ca,
eF RA (z)eF
2
. (3.12)
2 2
4 1 e0 a 9 + /2 1 + 2
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
We choose := (1 e20 a2 )/10 in what follows. Next, we pick some R > max{y :
y Y} and C (R3 , [0, 1]) such that 0 on BR (0), 1 on R3 \BR+2 (0),
and 1. We set := 1 . Furthermore, we let denote the character
istic function of R3 \BR (0). We choose R so large (depending on a, but not on F ;
recall (2.8)) that
1
eF VE eF
sup VC (x) + sup VH (x) + . (3.13)
xR xR 2Ca,
Conjugating (3.6) with exponential weights and rearranging the terms we nd,
for z ,
{1 + eF RA (z)eF ( VH + eF VE eF
VC + )}eF RA,V (z)eF
= eF RA (z)eF (eF RA (z)eF )(eF i )(RA,V (z) RA (z))eF
(eF RA (z)eF )(eF VE eF )(1BR+2 (0) eF )RA,V (z)eF .
Here the operator { } on the left side can be inverted by means of a Neumann
series and { }1 2 by (3.12) and (3.13). Furthermore, we recall the identity
which follows from the Cliord algebra relations (2.1), and observe that, by the
choice of , the assumption on F , and (3.14),
eF i ea(R+2) , eF 1.
C eR+2
eF RA,V (z)eF a , z = e0 + i .
1 + 2
This estimate implies the assertion if F is constant outside some ball since, certainly,
eF RA,V (z)eF ea(R+2) (02 + 2 )1/2 .
Let us now assume that F 0 is not necessarily bounded. Let F1 , F2 , . . .
C (R3 , [0, )) be constant near innity and such that Fn = F on Bn (0) and
Fn F . Then eFn RA,V (z)eFn eF RA,V (z)eF by the dominated conver
gence theorem, for every D. Since eFn RA,V (z)eFn obeys the estimate (3.11)
uniformly in n, we see that the densely dened operator eF RA,V (z)eF D is
bounded and satises (3.11), too. But this is the case if and only if its adjoint,
eF RA,V (z)eF = (eF RA,V (z)eF ) , is an element of L (H ) and satises (3.11)
as well.
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
[RA,V (z), eF ] = RA,V (z){i ( + F ) + [eF , VE ]eF }eF RA,V (z). (3.18)
eF RA,V (z)eF = RA,V (z) RA,V (z){i F + [eF , VE ]eF }eF RA,V (z)eF .
(3.19)
Using (3.21) and (3.22) with T = DA,V e0 and taking (2.8), (3.11), (3.14),
and (3.15) into account, we readily derive the asserted estimate (3.16) from (3.19).
The second estimate (3.17) it obtained analogously by means of (3.20).
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
3.3. Commutators
In this subsection, we derive the crucial technical prerequisites for the spectral anal
ysis of HN , namely various commutator estimates involving the projection + A,V ,
cuto functions, and exponential weights eF . Roughly speaking, these estimates
allow to adapt many arguments known from the spectral analysis of partial dier
ential operators that involve partitions of unity and conjugations with exponential
weights to our nonlocal model.
Our standard assumptions on the cuto and weight functions are
and
F C (R3 , R), F 0 or F 0, F (0) = 0, F  a,
(3.24)
F is constant outside some ball.
DA,V 1/2 [+ F F
A,V , e ]e DA,V 1/2 Ca0 ( + a). (3.25)
DA,V 1/2  [+ F F
A,V , e ]e DA,V 1/2 
DA,V 1/2 RA,V ( z ) i ( + F ) + [eF , VE ]eF
dz
eF RA,V (z)eF DA,V 1/2
2
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
1/2
2 dz
Ca 0 ( + F + a) DA,V  1/2
RA,V (z)
2
1/2
dz
eF RA,V (z)eF DA,V 1/2 2 , (3.28)
2
for , D(DA 1/2 ) Ran(RA (z)). By virtue of (3.16) and (3.17), we rst infer
that
[+ F F
A,V , e ]e DA,V 1/2 D((DA,V 1/2 ) ) = D(DA,V 1/2 ).
Proof. First, we assume that F satises (3.24). In this case the claim follows from
Proposition 3.1 because [eF , +
A,V ]e
F +
A,V A,V . If F is unbounded, then
= F
we apply an approximation argument similar to the one at the end of the proof of
Lemma 3.4.
(a + ) inf {  + +
A,V DA,VC A,V + C
1
2 }. (3.32)
0<1
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
Proof. The proof of (3.31) is a rather obvious application of Proposition 3.1 and in
fact a simpler analogue of the derivation of (3.32) below. Once (3.31) is veried, it
suces to prove (3.32) with DA,VC replaced by DA,V since VH and VE are bounded.
Without loss of generality we may further assume that DA,V is positive on the range
of +A,V . For otherwise we could add a suitable constant to DA,V . To prove (3.32)
we rst recall that +
A,V maps the domain of DA,V into itself and by Lemma 3.2(d)
we know that multiplication with or eF leaves D(DA,V ) invariant, too. We thus
have the following identity on D(DA,V ),
+
A,V DA,V A,V A,V DA,V
F F
= eF [+
A,V , e
F
]DA,V + + F +
A,V + A,V DA,V [e , A,V ]e
F
+ eF [+
A,V , e
F
]DA,V [eF , +
A,V ]e
F
.
It follows that the absolute value on the lefthand side of (3.32) is less than or
equal to
DA,V 1/2 +
A,V DA,V 1/2 [+ F F
A,V , e ]e
=
+ DA,V 1/2 [+ F F
A,V , e ]e 2 ,
=
which together with Proposition 3.1 implies (3.32). The last statement of the lemma
is valid since the argument above works equally well with DA,V  in place of DA,V
because + +
A,V DA,V  = A,V DA,V .
+
A,V H
1/2
(R3 , C4 ). Furthermore, we pick a (smooth, locally nite) partition of
3
unity on R , {J }N , =1 J = 1, such that =1 J  C, for some constant
C (0, ). Setting := J , N, := + A,V (DA,V i), and using (3.6), we
obtain
+
A,V = RA,V (i)+
A,V (DA,V i)
=1
= R0 (i) R0 (i)i ( )(RA,V (i) R0 (i)) (3.33)
=1
R0 (i) (V + A)RA,V (i). (3.34)
=1
Here the sum in (3.33) commutes with the rst resolvent and the strong limit
( ) denes a bounded operator on L2 (R3 , C4 ). To treat (3.34) we rst
=1 i
3
use that =1 V = V is bounded. Next, we pick some F C (R , [0, ))
vanishing on some ball containing 0 and supp() and satisfying F (x) = ax a ,
for x outside some suciently large ball with 0 < a < min{m, 0 }, a > 0. Then
we write
Using (2.7), Lemma 3.4 and Corollary 3.1 we see that ( A)RA,V (i) is an ele
ment of L2 (R3 , C4 ). These remarks imply that +
A,V belongs to Ran(R0 (i)) +
1 3 4
Ran(R0 (i)) = H (R , C ).
in what follows. The proof of the next proposition looks somewhat lengthy and
is hence postponed to Sec. 3.5. This is due to the fact that the singularity of Wy
may be located anywhere and that we allow for unbounded magnetic elds. We
remark that, even in the case V = 0, a diamagnetic inequality is not very useful in
this context since, for unbounded magnetic elds, one cannot compare i + A
with DA . We tackle this problem by a procedure that involves a partition of
unity, local gauge transformations, and exponential decay estimates which control
the correlation between dierent regions in position space. As a result we obtain a
commutator estimate which can be chosen to depend only on the local magnitude
of B either at the singularity y or on the support of the involved cuto function.
For any function on R3 we use the notation
Wy1/2 [+ F
A,V , e ]e
F
Ca0 ,N (1 + min{B(y), B , })(a + ). (3.37)
Corollary 3.4. Assume that A C 1 (R3 , R3 ) and B = curl A is bounded and that V
fullls Hypothesis 2.2. Then we nd, for every > 0, some constant Ca0 , (0, )
such that, for all F satisfying (3.24), DN , and 1 i N,
  1 eF +,N +,N
A,V WiN A,V 1 e
F
 +,N +,N
A,V WiN A,V 
Proof. This corollary is proved by means of Proposition 3.2 in the same way as
Corollary 3.2. We also recall that +,N
A,V DN D(Wij ).
The technique used in the proof of Proposition 3.2 also yields the following
result whose proof can be found in Sec. 3.5, too:
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
Lemma 3.7. Assume that A C 1 (R3 , R3 ) and B = curl A satises (2.18) and
that V fullls Hypothesis 2.2. Then there is some constant C (0, ) such that,
for all D(DA,V ),
Wy1/2 +
A,V C(1 + min{B(y), B , }) (DA,V i) . (3.38)
+ +
A := A,0 .
DA 1/2 (+ +
A A,V ) C( V + ).
In particular, +
A,V D(DA 
1/2
), for every D(DA ).
Proof. Due to (3.27) the norm in the statement (if it exists) is bounded from
above by
dz
sup DA 1/2  (RA (z) RA,V (z)) .
D(DA 1/2
), H
=1
We next use (3.6), (3.15), and (3.17) to conclude that the asserted bound holds
true.
DA 1/2 (1 R )(+
A +
A,V ) C (1 R )V + 0, (3.39)
R
as R tends to innity.
every D,
  + + + +
A,V DA,VC A,V  A DA A 
C
( V + ) inf  +
A D A  +
A + 2
, (3.40)
0<1
and
  + + + +
A,V DA,VC A,V  A DA A 
+ + C 2
( V + ) inf  A DA A + . (3.41)
0<1
The last estimate still holds true (with a new constant C), if DA and DA,VC are
replaced by DA 1 and DA,VC 1, respectively.
Proof. Let D and let R be the cuto function constructed in the paragraph
preceding (3.39). On account of Lemma 3.6 we know that + A,V H
1/2
and,
1/2
hence, we infer from Lemma 3.2(c) that R + A,V Hc belongs to the domain
of DA . Applying also the formula appearing in (ii) of Lemma 3.1 and using VCs = 0
we obtain
 + s
+ + s
+
A,V DA,VC A,V = lim R A,V  DA,VC A,V
R
= lim DA R + +
A,V  A,V .
R
Writing + := + +
A,V A we further get
DA R + +
A,V  A,V
= DA R + + + +
A  A + DA R A 
+ R +  DA + + +
A + DA R  .
By virtue of Lemma 3.8, we know that + D(DA 1/2 ) and it is easy to see
that DA R + +
A DA A , as R . Using also (3.39) we arrive at
  + + + +
A,V DA,VC A,V DA A  A 
s
2 DA 1/2 +
A DA 
1/2
+ + DA 1/2 + 2 .
Therefore, we obtain (3.40) by applying Lemma 3.8 once again. (3.41) follows from
a straightforward combination of (3.40) and Corollary 3.2. The last statement of
Corollary 3.5 follows from (3.41) and Lemma 3.8.
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
Lemma 3.9 ([32, 41, 42]). Assume that VC fullls Hypothesis 2.1 and let VCs
be given by (3.1). Then there exist 0 > 0 and 0 (, 1) such that, for every
R\(0 , 0 ), we have M (i) 0 and
R0,VCs (i) = R0 (i) R0 (i)VCs 1/2 (1 + SM (i))1 SVCs 1/2 R0 (i). (3.43)
Proof. The inequality  1/2 R0 (i) 1/2 1 has been conjectured in [41] and
proved in [32]. By means of this inequality and the arguments of [42, pp. 2 and 3
(with ks 1)] we nd some 0 (, 1) such that M (i) 0 provided  is
large enough. The resolvent formula (3.43) then follows from [41, Lemma 2.2 and
Theorem 2.2].
Proof of Proposition 3.2. We pick some C0 (R3 , [0, 1]) such that = 1
in a neighborhood of Y and supp() N . We set := 1 . Moreover, we pick
3
some C0 (R , [0, 1]) with 1 on B1/2 (0) and supp() B1 (0) and set
y (x) := (x y), x R3 . On account of Proposition 3.1 it suces to consider
y Wy1/2  [+ F
A,V , e ]e
F
1/2 dz dz
= y Wy  T (z) + y Wy1/2  T (z) , (3.44)
2 2
RA,V (z) = R0,VCs (z) + R0,VCs (z)i (R0,VCs (z) RA,V (z))
R0,VCs (z){V VCs + A}RA,V (z),
where VCs = SVCs  is dened in (3.1). Since D(D0,VCs ) H 1/2 (R3 , C4 ) due to
Lemma 3.2(e) we nd C, C (0, ) such that, for all y R3 and all z ,
To treat the remaining part of the rst integral in (3.44) we employ the rst resol
vent formula and (3.43) to obtain, for R with  0 and z = e0 + i (0 is
the parameter appearing in Lemma 3.9),
R0 (i)VCs 1/2 (1 + SM (i))1 SVCs 1/2 D0 1/2 D0 1/2 R0 (i). (3.49)
Since the remaining part of the integral over { < 0 } does not pose any further
problem, we see altogether that the rst integral in (3.44) is absolutely convergent
and of order O( + a).
Next, we treat the second integral on the righthand side of (3.44). Since we
eventually have to change the gauge locally, we pick a (smooth) partition of unity,
{J }Z3 , on R3 such that supp(J ) B1 (), for every Z3 . We can certainly
assume that Z3 J  C, for some C (0, ). We set
G () := { Z3 : J = 0}, := J , := 1 ,
G ()
T = i (F + ) + [eF , VE ]eF + [, VE ]
= {i (F + ) + [eF , VE ]eF + [, VE ]} {VE }
=: U1 U2 = (J U1 U2 J ).
G ()
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
For every Z3 , there is some gauge potential, g C 2 (R3 , R), such that
A g = A , where A is dened by
1
A (x) := B( + t(x )) t(x )dt, x R3 . (3.51)
0
:= g ,
By virtue of (3.6) we obtain with A
y RA,V (z)T = y RA,V (z)(J U1 U2 J )
G ()
= y RAe (z)(J U1 U2 J )
G ()
+ RAe (z)i (y )(RAe (z) RA,V (z))(J U1 U2 J )
G ()
RAe (z)y (V + A )RA,V (z)(J U1 U2 J )
G ()
=: S1 + S2 S3 . (3.52)
The proof of Proposition 3.2 is nished as soon as we have shown Lemma 3.10
below.
Lemma 3.10. In the situation above there exists a  and F independent constant,
Ca0 (0, ), such that, for j = 1, 2, 3,
Ij := Wy1/2  Sj eF RA,V (z)eF dz
Proof of Lemma 3.10. In our estimations below, that involve nonlocal operators,
we exploit the fact that the interference between spatially separated regions decays
exponentially. Therefore, we start by introducing appropriate exponential weight
functions: We pick some a (, min{0 , m}) and a convex, even function f
C (R, [0, )) such that f 0 on [2, 2], f(t) = a a, for t 4, and 0 f 1
t3
on (0, ). We dene f (x) := f(x ), x R3 , so that f 0 on B2 () and
f a dist(, B2 ()) a
with equality outside B4 (). Moreover, f  a . We
further pick some nondecreasing C (R, R) such that (t) = t, for t 1, 2
on [3, ) and 1. We set ,y (t) := ( y + 1)(t/( y + 1)), t R,
and f,y := ,y f , so that f,y is bounded, f,y = f a dist(, B2 ()) a
on
B1 (y) supp(y ), and f,y  a f,y
. By construction e J = J . Setting
we thus have
and employing (2.6) and (3.11) we further nd some constant C (0, ) such that,
for all z = e0 + i , Z3 , and y R3 ,
a +
,y (z) C . (3.53)
1 + 2
) C D e  ,
Wy C (eig ) = C (i + A A
1/2
f,y 1/2 2
I1 e RAe (
z )e f,y
DAe  dz
G ()
1/2
2
,y (z) dz
1/2
d
C (a + )
R 1 + 2
G ()
C sup{ef,y (x) : x B1 (y)} (a + )
Z3
C (a + ) .
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
for all Z3 and x R3 , which follows from (2.18). Since also B()
B(y) + B() B(y) and x y 1, if y (x) = 0, we infer again from (2.18)
that
y ef,y A 
G ()
ay
C e 1 + min B(y), sup B() , (3.55)
G ()
G ()
for some suciently small a > 0. Using these observations and the uniform bound
f,y f,y
edness of e V e , which is implied by Hypothesis 2.2 and the choice of , we
nd some , F , and yindependent constant C (0, ) such that
I3 z )Wy1/2 { y ef,y ef,y V ef,y
RAe(
G ()
This completes the proof of Lemma 3.10 and, at the same time, the proof of Propo
sition 3.2. (The last assertion of Proposition 3.2 follows by inspecting the arguments
above.)
Proof of Lemma 3.7. We use the notation introduced in the proofs of Propo
sition 3.2 and Lemma 3.10 in the following. We already know from Corollary 3.6
1/2
that the vector +A,V belongs to D(Wy ), but we do not have any control on the
norm on the left in (3.38) yet. It is certainly sucient to derive the asserted bound
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
1/2 1/2
with Wy replaced by y Wy . As in the proof of Corollary 3.6, we rst pick some
C0 (R3 , [0, 1]) (independent of ) such that 1 on some large open ball
containing Y and set = 1 . By the closed graph theorem D0 1/2 RA,V (i) is
bounded whence
y Wy1/2 +
A,V C D0 
1/2
RA,V (i) (DA,V i) .
:= +
A,V (DA,V i)J ,
Wy1/2  y +A,V 
Wy1/2  y RA,V (i) 
G ()
Wy1/2 y  RAe (i) 
G ()
+ Wy1/2  RAe (i) (y )(RA,V (i) RAe (i)) 
G ()
+ Wy1/2  RAe (i)y (V + A )RA,V (i) 
G ()
=: Q1 + Q2 + Q3 ,
,y := (ef,y +
A,V e
f,y f,y
)e (DA,V i)ef,y J ,
,y C (DA,V i) + O( J + a
) C (DA,V i) ,
where C, C (0, ) neither depend on nor y. Writing also ef,y RAe (i)ef,y =
RAe (i)(1 i f,y ef,y RAe (i)ef,y ) and using (3.11) we thus obtain
Q1 DAe 1/2 Wy1/2 y ef,y
G ()
4. Exponential Localization
In this section, we prove Theorem 2.1. To this end, we adapt an argument from [1]
and some useful improvements of the latter from [19] to our nonlocal situation.
In the proof below, we present the general strategy of the argument. In doing so,
we refer to three technical lemmata whose proofs are postponed to the end of this
section. The argument from [1] is advantageous here since it does not require any
a priori knowledge on the spectrum of HN . It rather gives the possibility to prove
the exponential localization of spectral projections directly and to infer results
on the nature of the spectrum from the localization estimate. In particular, the
argument avoids the use of eigenvalue equations which are, for instance, exploited
in Agmon type estimates.
Throughout this section we always assume that the assumptions of Theorem 2.1
are fullled.
Proof of Theorem 2.1. Since HN is bounded from below we may suppose that
inf I > . By assumption we have sup I < ENA1 + 1. Moreover, we consider HN
as an operator on the unprojected N particle space HN . In this case we have to
keep in mind that 0 becomes an innitely degenerated
eigenvalue of HN . Our goal
2bX
is to show that there are b, C (0, ) such that R3N e (X)2 dX C, for all
normalized Ran(EI (HN )) such that = AN and = +,N A,V . Borrowing an
idea from [38] we simplify the problem by using the bounds
N
e2bX max e2b N xj 
e2bN xj 
, X = (x1 , . . . , xN ) (R3 )N ,
j=1,...,N
j=1
and the antisymmetry of = AN . (We are not aiming to derive good estimates
on the decay rate here.) Indeed, it suces to show that there exist a, C (0, )
such that
e2axN  (X)2 dX C, (4.1)
R3N
+ PN 1 + A +
A,V (1 E1 )R A,V + QN
A A
= HN 1 1 + EN 1 PN 1 1 (4.3)
+ PN 1 + A +
A,V {DA,VC + (1 E1 )R }A,V + QN (4.4)
N 1
+ QN WiN QN . (4.5)
i=1
We denote the Friedrichs extension of H N again by the same symbol. (The idea to
introduce an additional cuto function R in (4.4) to compensate for the Coulomb
singularity in the last variable xN is borrowed from [19]; together with the other
N stays away
additional terms in (4.3) and (4.4) it ensures that the spectrum of H
from the interval I.) Notice that on DN we have HN 1 1 + EN 1 PN1 1
A A
PN 1 + A +
A,V {DA,VC + (1 E1 )R }A,V + QN 1 o(1)PN 1 1,
as R tends to innity. We now pick some > 0 with sup I < ENA1 + 1 . Then
the above remarks imply
N (E A + 1 /2) 2,
 H DN , (4.6)
N 1
Then H N and H have the same domain since they dier by a bounded operator
N
on their common form core DN . We further pick some I C0 (R, [0, 1]), such that
I 1 on I and supp(
I ) (, ENA1 + 1 ). Then N ) = 0 by (4.6). As
I (H
in [1] we now observe that
QN EI (HN )QN = QN EI (HN )QN = (
I (HN ) N ))QN .
I (H (4.7)
I ) (, ENA1 + 1 ) + i(, ),
supp(
(4.8)
I (z) = ON (zN ),
z N N,
which holds for every selfadjoint operator T on some Hilbert space. By means
of (4.7), we then nd the representation
QN EI (HN )QN =
[(HN N z)1 ]d
z)1 (H I (z)QN . (4.9)
C
For some F as in (3.24) (which acts only on the last variable in what follows), we
abbreviate
F +,N F
F,N
A,V := e A,V e .
Then (4.9) and the second resolvent identity together with the trivial identities
Q
N QN = 0 = (PN 1 1)QN yield
RF (z) := eF (H N eF z)1 ,
N z)1 eF = (eF H N ).
z (H (4.11)
In particular, eF H N eF is closed because its resolvent set is not empty. Using the
1 N eF ) = eF H
N eF .
identity RF (z) = RF (z)1 we readily verify that (eF H
Since eF maps DN into itself we further have
N eF ) = eF D(H
DN D(eF H N ) eF Q(H
N ). (4.12)
The following two lemmata, whose proofs are postponed to the end of this section,
show that eF H N eF is a small form perturbation of H
N . We dene T : DN
HN by
N eF H
T := eF H N , DN . (4.13)
N + O(a)  ,
  T  a  H DN . (4.14)
Lemma 4.2. There exist constants c1 , c2 (0, ) such that, for all F : R3 R
satisfying (3.24) and all DN ,
N eF  c1 eF 2  H
eF  H N + c2 eF 2 2 . (4.15)
N ) Q(H
In particular, eF Q(H N ).
If a < 1/2 then Lemma 4.1 implies that (eF H N eF )DN has a distinguished,
F
sectorial, closed extension, HN , that is the only closed extension having the prop
erties D(H F ) Q(H N ), D(H
F ) Q(H
N ), and i (H F ), for all R with
N N N
suciently large absolute value; see [31]. Thanks to (4.11), (4.12) and Lemma 4.2,
we know that eF H N eF is a closed extension enjoying these properties, whence
N
H F N eF .
= eF H
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
We are now prepared to derive a uniform bound on the norm under the integral
I ) and DN , we obtain
sign in (4.10). For z supp(
F z) =  (H
 (H N z) +  T
N
(1 a) H N z O(a) 2 . (4.16)
1a
By (4.6) and (4.8), we thus nd a (0, 1/2) and R [1, ) such that, for all
I ) and DN ,
z supp(
F z)
 (H N 2 .
4
N 4
(H F
z)1 , z supp(
I ),
which together with (4.10) proves Theorem 2.1.
Lemma 4.3. For every suciently large R 1, there is some cR (0, ) such
that cR 0, as R , and, for all D,
 + A + + 2 2
A,V [DA,VC + (1 E1 )R ]A,V A,V cR . (4.17)
 + A +
A,V [DA,VC + (1 E1 )R ]A,V
=  + A 2 +
A,V [j DA,VC j + (1 E1 )j R ]A,V
j=1,2
=: Yj . (4.18)
j=1,2
and D,
Y1 (1 1/R)1  + A + + 2
A,V [DA,VC E1 ]A,V 1 + 1 A,V O(1/R)
2
1 + 2 2
A,V O(1/R) . (4.19)
 + + + +
A,V 2 DA,VC 2 A,V (1 )  2 A DA A 2 o (1)
2
(1 ) + 2
A 2 o (1)
2
(1 )2 2 + 2
A o (1)
2
(1 )3 2 + 2 2
A,V o (1) , (4.20)
eF + +
A,V DA,VC A,V e
F
A,V (DA,VC + i F )A,V
= F F
= (1 + a)+ +
A,V DA,VC A,V + O(a).
The term in (4.4) involving the cuto function R yields a contribution of order
O(a), too, due to Corollary 3.2 (with L = (1 E1A )R and = 1). To account for
+
the projection on the right in (4.4) we write QN = 1HN PN 1 A,V and use
F F
Proposition 3.1 to obtain e QN e QN = O(a). Finally, we apply Corollary 3.4
to all terms in (4.5) this is the only place in this section where we use the
assumption that B is bounded and arrive at
!
N 1
(N )
  T  a QN DA,VC + WiN QN + O(a) 2 .
i=1
A A
Since HN 1 EN 1 this completes the proof of Lemma 4.1.
Proof of Lemma 4.2. We drop the signs in (4.15) since the they do not play
any role in this proof. It is clear that we only have to comment on those terms in
(4.3)(4.5) that involve unbounded operators. Since HN 1 1 commutes with eF
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
(If B is unbounded, then the Osymbol in (4.22) depends on the supremum of B
on supp(F ).) It remains to prove that there are constants c3 , c4 (0, ) such
that
 eF + + F
A,V DA,VC A,V e
c3 eF 2  + + F 2 2
A,V DA,VC A,V + c4 e , (4.23)
for D. Moreover, since VH and VE are bounded it suces to prove this estimate
with DA,VC replaced by D := DA,V e0 , which is positive on the range of + .
A,V
We abbreviate := A,V in the rest of this proof and seek for bounds on both
terms on the right side of
+ )1/2 eF 2 2
(D + )1/2 eF 2 .
(D (4.24)
=
Here the norm with = equals (D + )1/2 [eF , ] and is not greater than
some O(a) eF due to Proposition 3.1. We next dene
+ + 1 1.
:= + (DA,V e0 )+ + 1 = + D
D
+ )1/2 D
In fact, because of (D 1/2 1 and
1/2 eF + 2 eF D
D 1/2 + 2 + [D
1/2 , eF ]+ 2
eF 2 + D
1/2 2 + D
1/2 [D
1/2 , eF ]+ D
1/2 2
we shall see that (4.23) holds true as soon as we have shown that
1/2 [D
D 1/2 , eF ]+ = O(a) eF . (4.25)
eF ] + [+ , eF ]D
eF ] = + [D,
[D,
= + i F eF + ([VE , eF ]eF )eF + [+ , eF ]D. (4.26)
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
which holds for any strictly positive operator, T , on some Hilbert space. For ,
D, it implies
1/2 1/2 F + 1 1/2 1 F + dt
D  [D
, e ] = D [D, e ] . (4.28)
0
D+t
D+t t
We estimate the contribution of the rst term on the right side of (4.26) to (4.28) as
+ F
+ O(a) eF
D 1/2 i F e , t 0. (4.29)
D +t D +t (1 + t)3/2
In view of (2.7) the second term in (4.26) can be dealt with similarly. To account
for the third term in (4.26) we apply Proposition 3.1 and obtain, for t 0,
!
1 O(a) eF
1/2 + F D +
{D
[ , e ]e }e
F F
. (4.30)
D +t +t
D 1+t
Equations (4.28)(4.30) show that (4.25) holds true, which completes the proof of
Lemma 4.2.
A
5. The Lower Bound on inf ess (HN )
In order to prove the hard part of the HVZ theorem, Theorem 2.2(ii), we employ
an idea we learned from [19]: One may use a localization estimate for spectral
projections to prove their compactness. Of course one might try to derive a lower
bound on the ionization threshold by a more direct argument, for instance, by
following the general strategy presented in [33]. Since we have already derived an
exponential localization estimate we nd it, however, more convenient here to adapt
the observation from [19] to our nonlocal model. Another advantage of the proof
below is that we can work with the square root of HN . This is important since only
form bounds on perturbations of HN are available.
Theorem 5.1. Let the assumptions of Theorem 2.2(ii) be fullled and let I R
be an interval sup I < 1 + ENA1 . Then the spectral projection EI (HN
A
) is a compact
+
operator on AN HN . In particular,
A
ess (HN ) [1 + ENA1 , ).
A
Proof. Let g C(R, (0, )) satisfy g(r) , r , and g(X)EI (HN )
L (AN HN ) and set h := 1/g. We let R denote a smoothed characteristic function
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
of the closed ball in R3 with radius R > 0 and center 0 and set R (X) :=
R (x1 ) R (xN ), for X = (x1 , . . . , xN ) (R3 )N . First, we argue that it suces to
A
show that EI (HN )R (+1)1/8 is a (densely dened) bounded operator from HN
+
to AN HN . In fact, let us assume that this is the case. Since ( + 1)1/8 h(X)
A
is compact and g(X)EI (HN ) is bounded, it then follows that
A A
EI (HN )[R h(X)]g(X)EI (HN )
A
= EI (HN )R ( + 1)1/8 [( + 1)1/8 h(X)]g(X)EI (HN
A
)
Here the left curly bracket in (5.1) is a bounded operator from HN to HN+ since
+,N +,N
A,V SA,V HN + c, provided c is large enough, due to the positivity of the
interaction potentials and the boundedness of VH and VE . To see that the right
curly bracket in (5.1) is a bounded operator in HN we rst notice that it is a
restriction of S 1/2 T , where T := ( + 1)1/8 R is closed. It thus remains to
show that T S 1/2 = T S 1/2 = (S 1/2 T ) belongs to L (HN ). To this end
(i) (i)
we recall that ((i) + 1)1/4 R (DA,V  + 1)1 is bounded on L2 (R3i , C4 ) since
1/2
D(DA,V ) Hloc (R3 , C4 ). It follows that
(i) (i)
((i) + 1)1/4 R S 1 = ((i) + 1)1/4 R (DA,V  + 1)1 (DA,V  + 1)S 1
6. Weyl Sequences
In this section, we prove the easy part of our HVZ theorem, namely Part (i) of
Theorem 2.2 asserting that
A
ess (HN ) [ENA1 + 1, ).
A
This is done by constructing suitable Weyl sequences for HN . The diculties we
encounter are similar to those in [40] where the BrownRavenhall model (free pic
ture without magnetic eld) is considered. We have, however, to replace those
arguments in [40] that require explicit momentum or position space representations
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
of the free projection +0 by more abstract ones; see, e.g., Lemma 6.2. Another
new technical complication is caused by the related facts that +
A,V maps the dense
subspace D merely into H 1/2 when V has a strong Coulombic singularity and that,
compared to the free picture, it is more dicult to control the singularities of the
interaction potentials. For this reason we shall eventually study the square root of
HN rather than HN itself.
We x some spectral parameter 1 throughout the whole section and
{n }nN will always denote a corresponding Weyl sequence as in Hypothesis 2.4(i).
In this and the following section we shall repeatedly employ the following
sequence of cuto functions: We pick some C (R, [0, 1]) such that 0
on (, 1 /4] and 1 on [1, ). Here (0, 1) is a xed parameter whose
value becomes important only in Sec. 7. We set n := (x/Rn ), for x R3 and
n N, where Rn is given by Hypothesis 2.4(i). Then it holds n n = n and
n = Rn1 0, as n .
To begin with we draw two simple conclusions from our hypotheses:
Proof. The rst identity is clear from the hypotheses. To treat the second we
employ the cuto functions dened in the paragraph preceding the statement of
this lemma and abbreviate VHE := VH + VE . By means of Proposition 3.1 and
VHE n 0 we then obtain
VHE + + +
A,V n VHE n A,V n + VHE [A,V , n ]n 0,
as n tends to innity. Therefore, the second identity follows from the rst.
EI (DA,V )n 1, in particular, +
A,V n 1. (6.2)
Next, we pick 0 < a < min{m, 0 }, r (0, 1 /4), and r (0, 1) such that
(1 r)a > (1 + r), s := r + r 1 > 0. (6.5)
Here appears in (2.18). We further pick some cuto function, C (R, [0, 1]),
such that 0 on (, s/2] and 1 on [s, ). By Lemma 3.6 we know that
(1)
D0 1/2 n HN 1 , where the superscript (1) again indicates that the operator
acts on the rst variable. Therefore, we nd a subsequence, {Rkn }nN , of {Rk }kN
such that, for every n N,
(1) 1
D0 1/2 (x1 /Rkn )n (X)2 dX < , (6.6)
R3(N 1) n
As a candidate for a Weyl sequence we then try {AN n }nN , where
+,N
n := n +
A,V kn A,V DN , n N. (6.7)
To simplify the notation we again write n instead of kn in the following. Finally,
we pick some c > 1 and set
f (t) := (t + c)1/2 (t ENA1 ), t > c.
Lemma 6.3. Let the assumptions of Theorem 2.2(i) be fullled. If, in the situation
described above, c > 1 is suciently large, then AN n D(f (HN )), for every
n N, and
wlim AN n = 0, lim inf AN n > 0, lim f (HN )AN n = 0. (6.8)
n n n
Proof. First, suppose that (6.8) holds true. If c > 1 is chosen suciently large,
then f is strictly monotonically increasing on (HN ). If I is some small open
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
interval around ENA1 + we thus get EI (HN ) = Ef (I) (f (HN )). By (6.8) and the
Weyl criterion applied to f (HN ) it follows that = dim Ran(Ef (I) (f (HN ))) =
dim Ran(EI (HN )).
To verify (6.8) we rst notice that n 0, as n , because of (2.17).
Exactly as in [40, 4] we can also check that lim inf AN n > 0. So it suces
to show that f (HN )n 0, as HN commutes with AN . Since n and n are
+,N
normalized and n = n + A,V n A,V DN we obtain
f (HN )n
1 1 1
(HN + c) 2 (HN 1 ENA1 ) 2 1H + (HN 1 ENA1 ) 2 n (6.9)
+ (DA,VC )+
A,V n (6.10)
N 1
1 1 1
+ (HN + c) 2 +,N 2 2 +
A,V WiN WiN (n A,V n ) . (6.11)
i=1
We rst show that the operator norm in (6.11) is actually nite. In fact,
1/2 1/2
(HN + c)1/2 +,N +,N
A,V WiN = WiN A,V (HN + c)
1/2
1,
+,N 1
since W 0, + + + A
A,V DA,VC A,V C A,V , HN 1 EN 1 A,V , and, hence,
1/2
WiN +,N 2 +,N +,N
A,V =  A,V WiN A,V  (HN + c),
for +,N
A,V DN . Using similar estimates and (6.3) it is straightforward to check
that the term in (6.9) converges to zero provided c > 1 is suciently large. The
norm in (6.10) tends to zero by Lemma 6.1. The claim now follows from Lemma 6.4
below which implies that the remaining norm in (6.11) tends to zero, too.
The rst inequality of the following lemma is used in the proof of Lemma 6.3
and the second one in Sec. 7.
sup W (x, y) + 2
A,V n + C (1 + B )e
aRn /2
xy
(1)Rn
+C n (y)dy(1 + B ) (DA,V + i)n 2 .
{yRn /2}
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
In the last two steps we make use of Proposition 3.2 and (2.18). Next, if x y
(1 r )Rn and 1BrRn (0) (x) = 0, then y (r + r 1)Rn = sRn , and by the choice
of (see the paragraph below (6.5)) it follows that
(1 1BrRn (0) (x))W (x, y)n (y)+ 2
A,V n (x) d(x, y)
{xy(1r )Rn }
sup W (x, y)(y/Rn )n (y)dy + 2
A,V n . (6.13)
xrRn R3
On account of (6.5), Katos inequality, and (6.6) the rst asserted estimate follows
from (6.12) and (6.13). The second one is derived similarly by means of Lemma 3.7
and the replacements r 1 /2, r . Note that 1B(1+/2)Rn (0) n = 0, which is
used to derive the analogue of (6.12).
7. Existence of Eigenvalues
A
In this section, we prove Theorem 2.3 which asserts that HN possesses innitely
A A
many eigenvalues below inf ess (HN ) = 1 + EN 1 . We proceed along the lines of [40,
6] with a few changes. In particular, as in the previous section we replace the
arguments of [40] that employ explicit position or momentum space representations
of +0 by more abstract ones. A crucial observation is the new argument used to
prove Lemma 7.1.
Throughout this section we always assume without further notice that the
assumptions of Theorem 2.3, i.e. Hypothesis 2.5, are fullled.
Proof of Theorem 2.3. We proceed by induction on N and start with the induc
A
tion step. So, we pick N N, N 2, and assume that HN 1 possesses innitely
A
many eigenvalues below EN 2 + 1. In particular, we can pick a normalized ground
A
state of HN 1 , which we denote by . Moreover, we denote the transposition oper
ator which ips the ith and N th electron variable by iN , 1 i < N , and set
N N := 1. The vectors 1 , 2 , . . . are the elements of the sequence appearing in
Hypothesis 2.5.
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
Now, let d N. By Lemma 7.7 below we know that, for all suciently large
m0 N, the set {AN ( + m0 +d
A,V n )}n=m0 , where
1
N
AN ( +
A,V n ) = (1)N i iN ( +
A,V n ), n N,
N i=1
m 0 +d
N
 H cn 2 + +
A,V n  HN ( A,V n ) (7.2)
n=m0
m 0 +d
m 0 +d
N
 H cn 2 ,
0
n=m0
for all cm0 , . . . , cm0 +d C, if m0 is suciently large (depending on d). This con
cludes the induction step.
Finally, the case N = 1 is treated in the same way as the induction step N
N + 1 (setting E0A := 0 and ignoring , W , and the term (7.3)).
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
To show that the contribution coming from the (oneparticle) kinetic energy of
n decreases faster than its negative potential energy we make use of the require
ment that the n have vanishing lower spinor components, n = (n,1 , n,2 , 0, 0) ,
n N. This has also been used in [40] together with explicit formulas for +0 . We
replace these arguments by the following observation:
0 + + 2
A n  (DA 1)A n CRn , n N.
Proof. Since the last two components of n are zero we have ( 1)n = 0.
If we denote the projection onto the rst two spinor components, L2 (R3 , C4 )
(1 , 2 , 3 , 4 ) (1 , 2 , 0, 0) , by p then we also have pi n = 0 = pi i n ,
2
i = 1, 2, 3, and, therefore, p(DA 1)n = 0. Moreover, [p, DA ] = 0 and, hence,
1
[p, DA  ] = 0. This implies
+ +
A n  (DA 1)A n 
1 1
= n  p(DA 1)n + sgn(DA )n  (DA 1)n
2 2
1 1
= (DA 1)n  DA  (DA 1)n + n  DA  p(DA 1)n
1 1
2 2
1
(DA 1)n 2 = O(1/Rn2 ).
2
In the last step we apply Hypothesis 2.5.
In the following we split VC into a singular and regular part, VC = VCs + VCr ,
where VCs is dened in (3.1). By Hypothesis 2.1 VCr is bounded.
+ +
A,V n  (DA,VC 1)A,V n
= + r + + + 1
A,V n  VC A,V n + A n  (DA 1)A n + o(Rn ).
In the next lemma, we single out the leading order negative contribution to (7.2).
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
Lemma 7.3. There is some constant C (0, ) such that, for all suciently large
n N,
+ r + + 2
A,V n  VC A,V n v
(, Rn ) A,V n + Ce
Rn /C
,
where v
(, Rn ) is given by (2.20).
+ r + + r +
A,V n  VC A,V n A,V n  1S (Rn ) VC A,V n
+ r +
A,V n  1S (Rn ) VC A,V n
v
(, Rn )( + 2 + 2
A,V n 1R3 \S (Rn ) A,V n )
v
(, Rn ) + 2
A,V n + v
(, Rn ) 1R3 \S (Rn ) e
Fn 2 Fn +
e A,V eFn 2 .
By (2.19), (2.21), and the choice of Fn we know that 1R3 \S (Rn ) eFn
CeaRn /2 , which implies the assertion of the lemma.
From now on, we always assume that the induction hypothesis made in the proof
of Theorem 2.3 is fullled and that is a normalized ground state eigenvector of
A A A A A 1
HN 1 . So, HN 1 = EN 1 , EN 1 < EN 2 + 1. Given (0, N ) we pick some
(0, 1) as in Hypothesis 2.5(i). Then the following assertion is valid:
+ +
A,V n  WiN A,V n sup W (x, y) + 2
A,V n + O(Rn ).
xy
(1)Rn
Proof. This follows from Lemma 6.4 with n (y) = R3(N 2) (y, X )2 dX and
the exponential decay of , which is ensured by Theorem 2.1 and the induction
hypothesis.
nm := 1N ( + +
A,V n )  HN ( A,V m ) = O(Rmin{n,m} ).
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
Proof. We pick C0 (R3 , [0, 1]) such that 1 on B1/4 (0) and 0 outside
B1/2 (0) and set n := (/Rn ) and n := 1 n , for n N. As in [40] we nd
nm {n + +
A,V n }  {(DA,VC 1)A,V m }
+ {+ (1) +
A,V n }  {n } {(DA,VC 1)A,V m }
N 1
+ {n + +
A,V n }  WiN (A,V m )
i=1
N 1
+ {+ (1) +
A,V n }  WiN (n ) (A,V m )
i=1
N 1
N 1
=: Y1 + Y2 + Y3i + Y4i .
i=1 i=1
1/2
Y4i + (1) 1/2 +
A,V n WiN n sup Wy A,V m .
yR3
1/2
Here the norms WiN , i = 2, . . . , N 1, are actually nite since Ker(HN 1
ENA1 ) implies
1/2 +,N 1 +,N 1
WiN 2 =  A,V WiN A,V (ENA1 + C) 2 ,
for some constant C (0, ). Finally,
Y31 sup Wy1/2 n + 2 1/2 +
A,V n sup Wy A,V m ,
yR3 yR3
This estimate, the exponential decay of , and Lemma 3.7 imply that the terms
Y3i and Y4i , 1 i N 1, vanish of order O(Rmin{n,m} ) also.
 + +
A,V n  HN ( A,V m ) = O(Rn ).
Proof. We pick a family of smooth weight functions, {Fk }k,N , such that Fk 0
on supp(k ), Fk is constant outside some ball containing supp(k ) and supp( ),
Fk a < min{0 , m}, and
gk := eFk Fk Cea min{Rk ,R }
, k, N,
 + +
A,V n  HN ( A,V m )
+ + +
A,V n  (DA,V 1)m  + A,V n  (VH + VE )A,V m 
1/2 1/2
+ WiN + +
A,V n  WiN A,V m 
1i<N
gnm eFnm +
A,V e
Fnm
n { (DA,V 1)m
By virtue of Proposition 3.2 and Lemma 3.7, we know that all terms behind the
factors gnm appearing here are uniformly bounded which shows that the assertion
holds true.
Lemma 7.7. For every d N, there is some n0 N such that the set of vectors
{AN ( + m0 +d
A,V n )}n=m0 is linearly independent, for all m0 N, m0 n0 .
1N ( + +
A,V n )  A,V m  = O(Rmin{n,m} ).
Furthermore, by employing the exponential weights from the proof of Lemma 7.6
we see that
 + + +
A,V n  A,V m  = n  A,V m  Ce
min{Rn ,Rm }/C
.
Altogether we nd some C (0, ) such that, for d N and all suciently large
m0 N,
m0 +d m0 +d 0 +d
1 C (N 1)
m
C
2 cn 2 cn cm  cn cm .
2 n=m Rm0 n,m=m
Rm0
0 0 n, m=m0
n=m
Acknowledgments
It is a pleasure to thank Hubert Kalf, Sergey Morozov, and Heinz Siedentop for
useful remarks and helpful discussions. Moreover, we thank Sergey Morozov for
making parts of his manuscripts [38] available to us prior to publication.
References
[1] V. Bach, J. Fr ohlich and I. Sigal, Quantum electrodynamics of conned non
relativistic particles, Adv. Math. 137 (1998) 299395.
[2] V. Bach and O. Matte, Exponential decay of eigenfunctions of the BetheSalpeter
operator, Lett. Math. Phys. 55 (2001) 5362.
[3] A. A. Balinsky and W. D. Evans, On the virial theorem for the relativistic operator
of Brown and Ravenhall, and the absence of embedded eigenvalues, Lett. Math. Phys.
44 (1998) 233248.
[4] A. A. Balinsky and W. D. Evans, Stability of oneelectron molecules in the Brown
Ravenhall model, Comm. Math. Phys. 202 (1999) 481500.
[5] A. A. Balinsky and W. D. Evans, On the spectral properties of the BrownRavenhall
operator, J. Comput. Appl. Math. 148 (2002) 239255.
[6] P. Beiersdorfer, M. H. Chen, K. T. Cheng and J. Sapirstein, Transition energies of
the 3s3p3/2 resonance lines in sodiumlike to phosphoruslike uranium, Phys. Rev. A
68 (2003) 022507, 7 pp.
[7] A. Berthier and V. Georgescu, On the point spectrum of Dirac operators, J. Funct.
Anal. 71 (1987) 309338.
[8] A. M. Boutet de Monvel and R. Purice, A distinguished selfadjoint extension for the
Dirac operator with strong local singularities and arbitrary behavior at innity, Rep.
Math. Phys. 34 (1994) 351360.
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
[9] G. E. Brown and D. G. Ravenhall, On the interaction of two electrons, Proc. Roy.
Soc. London A 208 (1951) 552559.
[10] R. Cassanas and H. Siedentop, The groundstate energy of heavy atoms according
to Brown and Ravenhall: Absence of relativistic eects in leading order, J. Phys. A
39 (2006) 1040510414.
[11] K. T. Cheng, M. H. Chen and W. R. Johnson, Accurate relativistic calculations
including QED contributions for fewelectron systems, in Relativistic Electronic
Structure Theory. Part 2: Applications, ed. P. Schwerdtfeger, Theoretical and Com
putational Chemistry, Vol. 14 (Elsevier, 2002), pp. 120187.
[12] K. T. Cheng, M. H. Chen and J. Sapirstein, Potential independence of the solu
tion to the relativistic manybody problem and the role of negative energy states in
heliumlike ions, Phys. Rev. A 59 (1999) 259266.
[13] H. L. Cycon, R. G. Froese, W. Kirsch and B. Simon, Schr odinger Operators, Texts
and Monographs in Physics (Springer, BerlinHeidelberg, 1987).
[14] A. Derevianko, W. R. Johnson, D. P. Plante and I. M. Savukov, Negativeenergy
contributions to transition amplitudes in heliumlike ions, Phys. Rev. A 58 (1998)
44534461.
[15] M. Dimassi and J. Sj ostrand, Spectral Asymptotics in the SemiClassical Limit,
London Math. Soc. Lecture Note Series, Vol. 268 (Cambridge University Press,
Cambridge, 1999).
[16] J. Dolbeault, M. J. Esteban and M. Loss, Relativistic hydrogenic atoms in strong
magnetic elds, Ann. Henri Poincare 8 (2007) 749779.
[17] W. D. Evans, P. Perry and H. Siedentop, The spectrum of relativistic oneelectron
atoms according to Bethe and Salpeter, Comm. Math. Phys. 178 (1996) 733746.
[18] V. Georgescu and M. M antoiu, On the spectral theory of singular Dirac type
Hamiltonians, J. Operator Theory 46 (2001) 289321.
[19] M. Griesemer, Exponential decay and ionization thresholds in nonrelativistic quan
tum electrodynamics, J. Funct. Anal. 210 (2004) 321340.
[20] M. Griesemer and H. Siedentop, A minimax principle for the eigenvalues in spectral
gaps, J. London Math. Soc. (2) 60 (1999) 490500.
[21] M. Griesemer and C. Tix, Instability of a pseudorelativistic model of matter with
selfgenerated magnetic eld, J. Math. Phys. 40 (1999) 17801791.
[22] B. Heler, J. Nourrigat and X. P. Wang, Sur le spectre de lequation de Dirac (dans
R3 ou R2 ) avec champ magnetic, Ann. Sci. Ecole
Norm. Sup. 22(4) (1989) 515533.
[23] B. A. He, M. Reiher and A. Wolf, The generalized DouglasKroll transformation,
J. Chem. Phys. 117 (2002) 92159226.
[24] G. Hoever and H. Siedentop, Stability of the BrownRavenhall operator, Math. Phys.
Electr. J. 5 (1999) Paper 6, 11 pp.
[25] M. Huber and E. Stockmeyer, Perturbative implementation of the Furry picture,
Lett. Math. Phys. 79 (2007) 99108.
[26] G. Jansen and B. A. He, Revision of the DouglasKroll transformation, Phys. Rev. A
39 (1989) 60166017.
[27] D. H. JakubaaAmundsen, The HVZ theorem for a pseudorelativistic operator,
Ann. Henri Poincare 8 (2007) 337360.
[28] D. H. JakubaaAmundsen, Heat kernel estimates and spectral properties of a pseu
dorelativistic operator with magnetic eld, J. Math. Phys. 49 (2008) 032305, 22 pp.
[29] W. R. Johnson, Relativistic manybody theory applied to highlycharged ions, in
ManyBody Theory of Atomic Structure and Photoionization, ed. T. N. Chang (World
Scientic, 1993), pp. 1946.
February 11, 2010 10:0 WSPC/148RMP J070S0129055X10003874
[30] W. R. Johnson, Relativistic manybody perturbation theory for highly charged ions,
in ManyBody Atomic Physics, eds. J. J. Boyle and M. S. Pindzola (University Press,
1998), pp. 3964.
[31] T. Kato, Perturbation Theory for Linear Operators, Classics in Mathematics
(Springer, BerlinHeidelberg, 1995).
[32] T. Kato, Holomorphic families of Dirac operators, Math. Z. 183 (1983) 399406.
[33] Y. Last and B. Simon, The essential spectrum of Schr odinger, Jacobi, and CMV
operators, J. dAnalyse Math. 98 (2006) 183220.
[34] E. H. Lieb and M. Loss, Stability of a model of relativistic quantum electrodynamics,
Comm. Math. Phys. 228 (2002) 561588.
[35] E. H. Lieb, H. Siedentop and J. P. Solovej, Stability and instability of relativistic
electrons in classical electromagnetic elds, J. Statist. Phys. 89 (1997) 3759.
[36] O. Matte and E. Stockmeyer, On the eigenfunctions of nopair operators in classical
magnetic elds, Integr. Equ. Oper. Theory 65 (2009) 255283.
[37] S. Morozov, Essential spectrum of multiparticle BrownRavenhall operators in exter
nal eld, Documenta Math. 13 (2008) 5179.
[38] S. Morozov, Multiparticle BrownRavenhall operators in external elds, PhD thesis,
Universitat Munchen (2008).
[39] S. Morozov, Exponential decay of eigenfunctions of BrownRavenhall operators,
J. Phys. A 42 (2009) 475206, 16 pp.
[40] S. Morozov and S. Vugalter, Stability of atoms in the BrownRavenhall model, Ann.
Henri Poincare 7 (2006) 661687.
[41] G. Nenciu, Selfadjointness and invariance of the essential spectrum for Dirac oper
ators dened as quadratic forms, Comm. Math. Phys. 48 (1976) 235247.
[42] G. Nenciu, Distinguished selfadjoint extension for the Dirac operator with potential
dominated by multicenter Coulomb potentials, Helvetica Phys. Acta 50 (1977) 13.
[43] M. Reiher and A. Wolf, Relativistic Quantum Chemistry (WileyVCH, Weinheim,
2009).
[44] R. Richard and R. Tiedra de Aldecoa, On the spectrum of magnetic Dirac operators
with Coulombtype perturbations, J. Funct. Anal. 250 (2007) 625641.
[45] J. Sapirstein, Theoretical methods for the relativistic atomic manybody problem,
Rev. Modern Phys. 70 (1998) 5576.
[46] H. Siedentop and E. Stockmeyer, The DouglasKrollHe method: Convergence and
blockdiagonalization of Dirac operators, Ann. Henri Poincare 7 (2006) 4558.
[47] J. Sucher, Foundations of the relativistic theory of manyelectron atoms, Phys. Rev.
A 22 (1980) 348362.
[48] J. Sucher, Relativistic manyelectron Hamiltonians, Phys. Scripta 36 (1987) 271281.
[49] B. Thaller, The Dirac Equation, Texts and Monographs in Physics (Springer, Berlin
Heidelberg, 1992).
[50] C. Tix, Strict positivity of a relativistic Hamiltonian due to Brown and Ravenhall,
Bull. London Math. Soc. 30 (1998) 283290.
[51] C. Tix, Selfadjointness and spectral properties of a pseudorelativistic Hamiltonian
due to Brown and Ravenhall, preprint (1997) 20 pp.; mp arc 97441.
[52] C. Tix, Lower bound for the ground state energy of the nopair Hamiltonian, Phys.
Lett. B 405 (1997) 293296.
[53] J. Xia, On the contribution of the Coulomb singularity of arbitrary charge to the
Dirac Hamiltonian, Trans. Amer. Math. Soc. 351 (1999) 19892023.
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
ANGELO BASSI
Department of Physics, University of Trieste,
Strada Costiera 11, 34151 Trieste, Italy
and
Istituto Nazionale di Fisica Nucleare, Trieste Section,
Via Valerio 2, 34127 Trieste, Italy
bassi@ts.infn.it
DETLEF DURR and MARTIN KOLB
We discuss the time evolution of the wave function which is the solution of a stochastic
Schr
odinger equation describing the dynamics of a free quantum particle subject to spon
taneous localizations in space. We prove global existence and uniqueness of solutions.
We observe that there exist three time regimes: the collapse regime, the classical regime
and the diusive regime. Concerning the latter, we assert that the general solution con
verges almost surely to a diusing Gaussian wave function having a nite spread both
in position as well as in momentum. This paper corrects and completes earlier works
on this issue.
Keywords: Collapse models; GRWmodel; Hilbert space valued diusions; large time
behavior.
1. Introduction
Stochastic dierential equations (SDEs) in innite dimensional spaces are a subject
of growing interest within the mathematical physics and physics communities
working in quantum mechanics; they are currently used in models of sponta
neous wave function collapse [114], in the theory of continuous quantum meas
urement [15, 1725], and in the theory of open quantum systems [2628]. In the
rst case, the Schrodinger equation is modied by adding appropriate nonlinear
and stochastic terms which induce the (random) collapse of the wave function in
55
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
56 A. Bassi, D. D
urr & M. Kolb
space; in this way, one achieves the goal of a unied description of microscopic
quantum phenomena and macroscopic classical ones, avoiding the occurrence of
macroscopic quantum superpositions. Current research focuses on designing exper
iments which discriminate between collapse and noncollapse theories, see references
in [16]. In the second case, using the projection postulate, stochastic terms in the
Schrodinger equation are used to describe the eect of a continuous measurement.
In the third case, slightly generalizing the notion of continuous measurement to
generic interactions with environments, SDEs are used as phenomenological equa
tions describing the interaction of a quantum system with an environment, the
stochastic terms encoding the eect of the environment on the system. Looking
directly at the stochastic dierential equation for the wave function, rather than
the deterministic equation of the Lindblad type for the statistical operator has some
advantages with respect to the standard master equation approach, e.g. for faster
numerical simulations [29].
Among the dierent SDEs which have been considered so far, the following
equation, dened in the Hilbert space H L2 (R), is of particular interest [17
19, 3037, 39]
i p2 2
dt = dt + (q qt )dWt (q qt ) dt t , 0 = . (1.1)
2m 2
The rst term on the righthand side represents the usual quantum Hamiltonian
of a free particle in one dimension, p being the momentum operator. The second
and third terms of the equation, as we shall see, induce the localization of the
wave function in space; q is the position operator and qt denotes the quantum
expectation t qt of q with respect to t . The parameter is a xed positive
constant which sets the strength of the collapse mechanism, while Wt is a standard
Wiener process dened on a probability space (, F , P) with ltration {Ft , t 0}.
Equation (1.1) plays a special role among the SDEs in Hilbert spaces because
it is the simplest exactly solvable equation describing the time evolution of a non
trivial physical system. Within the theory of continuous quantum measurement,
it describes a measurementlike process designed to measure the position of a free
quantum particle; within decoherence theory it represents one of the possible unrav
ellings of the master equation rst derived by Joos and Zeh [40]. Within collapse
models (like GRWmodels), it may describe the evolution of a free quantum particle
(or the center of mass of an isolated system) subject to spontaneous localizations
in space [1, 2] in the following sense. Realistic models of spontaneous wave func
tion collapse are based on a more complicated stochastic dierential equation: The
dierence between Eq. (1.1) and the equations of the standard localization models
such as GRW [1] and CSL [2] is most easily described on the level of the Lindblad
equations for the respective statistical operators t := EP [t t ], induced by the
stochastic dynamics of the wave function. By virtue of Eq. (1.1) (see, e.g., [9]):
d i
t = [p2 , t ] [q, [q, t ]], (1.2)
dt 2m 2
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
58 A. Bassi, D. D
urr & M. Kolb
that there is no fundamental dierence between Eqs. (1.8) and (1.6), since any
solution of Eq. (1.8) can be obtained from a solution of Eq. (1.6) simply by nor
malizing the wave function. In turn, Eq. (1.6) does not contain any information as
to why the wave function should collapse according to the Born probability rule,
i.e. the Wiener process t is not forced to pick most likely those values necessary to
reproduce quantum probabilities, during the collapse process.
The way to include such a feature into the dynamical evolution of the wave
function is to replace the measure Q with a new measure (which will turn out to
be the measure P previously introduced) so that the process t , according to the
new measure, is forced to take with higher probability the values which account
for quantum probabilities. This is precisely the key idea behind the original GRW
model of spontaneous wave function collapse [1]: the wave function is more likely
to collapse where it is more appreciably dierent from zero. The mathematical
structure of the GRW model suggests that the square modulus t 2 should be
used as density for the change of measure. We now formalize these steps.
In [31], Holevo has proven that for initial condition 0 2 = 1 the process
(t 2 )t0 is a martingale satisfying the equation
t
t 2 = 0 2 + 2 qs s 2 ds . (1.9)
0
We shall always work with normalized initial states. The martingale t 2 can be
used as a RadonNikodym derivative to generate a new probability measure P from
Q, according to the usual formula:
P[E] := EQ [1E t 2 ], E Ft , t < +, (1.10)
where 1E is the indicator function relative to the measurable subset E. We recall
that the martingale property, together with the property EQ [t 2 ] = 1, guarantee
consistency among dierent times, so that (1.10) denes indeed a unique probability
measure P on F . In the following, for simplicity we will write dP/dQ t 2 .
One can then show that Eq. (1.8), with the stochastic dynamics dened on the
probability space (, F , P) in place of (, F , Q), correctly describes the desired
physical situations.
A drawback of the change of measure is that the equation is dened in terms
of the stochastic process t , which is not anymore a Wiener process with respect
to the measure P, as it was with respect to the measure Q. This can be a source of
many diculties, e.g. when analyzing the properties of the solutions of the equa
tion. The disadvantage can be removed by resorting to Girsanovs theorem, which
connects Wiener processes dened on the same measurable space, but with respect
to dierent probability measures. According to this theorem, the process
t
Wt := t 2 qs ds, (1.11)
0
is a Wiener process with respect to (, F , P) and ltration {Ft , t 0}, and thus
is the natural process for describing the stochastic dynamics with respect to the
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
60 A. Bassi, D. D
urr & M. Kolb
As we shall see, the parameter , which has the dimensions of a frequency, will set
the time scales for the collapse of the wave function. The representation in terms of
the Greens function (1.12), as we have said, is particularly suitable for analyzing the
time evolution of the general solution of Eq. (1.6), and thus of Eq. (1.1), even though
we will see that, when studying the long time behavior, another representation is
more convenient.
Our rst result concerns the meaning of the solution of Eq. (1.6) in terms of
t (x) := dyGt (x, y)(y) (1.20)
Theorem 1.1. Let t be dened as in (1.20); then the following three statements
hold true with Qprobability 1:
(1) L2 (R) t L2 (R), (1.21)
(2) L2B (R) t is a topological strong solution of Eq. (1.6), (1.22)
2
(3) L (R) limt0 t = 0, (1.23)
where L2B (R) is the subspace of all bounded functions of L2 (R).
Having the explicit solution of the Eq. (1.6), and thus of Eq. (1.1), the next
relevant problem is to unfold its physical content. Previous analysis of similar equa
tions [2,8,10,14,39] have shown that one can identify three regimes, which are more
or less well separated depending on the value of the parameters and m.
(1) Collapse regime. A wave function having an initial large spread, localizes in
space; the localization occurring in agreement with the Born probability rule.
(2) Classical regime. The localized wave function moves in space like a classical free
particle, since the uctuations due to the Wiener process can be safely ignored.
That a well localized Schrodinger wave function should move along a classical
path is connected to the validity of Ehrenfest or Egorovtype theorems [38].
(3) Diusive regime. Eventually, the random uctuations become dominant and
the wave function starts to diuse appreciably.
It is not an easy task to spell out rigorously these regimes and their properties.
We shall, however, be a bit more specic on this in the following section. We shall
afterwards focus on the simplest regime, namely the diusive one, which in fact has
been intensively looked at in the previous years [7, 19, 26, 34, 35, 39] and we shall
prove a remarkable property of the solutions of Eq. (1.1): Any solution converges
almost surely to a Gaussian wave function having a xed spread.
Theorem 1.2. let t be a solution of Eq. (1.1); then under conditions which we
will specify, the following property holds true with Pprobability 1:
lim t t = 0, (1.24)
t
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
where t , dened in (5.2), is a Gaussian wave function with a xed spread both in
position and momentum.
Theorems 1.1 and 1.2 have been extensively discussed before in the literature
[7, 19, 26, 30, 3436, 39], proving that the community has devoted much attention
to the problem. However, these proofs are not complete or awed. Concerning
Theorem 1.1, in particular Statement (3) was not proven [30, 3436]. While State
ments (1) and (2) are rather straightforward conclusions from the Gaussian kernel
of the propagator, the third statement is much more subtle and does not follow
from purely analytical arguments. Concerning Theorem 1.2, none of the previous
proofs is decisive. In [35, 36], the major aw was that it was overlooked that the
eigenfunction expansion of the relevant dissipative operator (not selfadjoint) does
not give rise to an orthonormal basis. In [19], the long time behavior was analyzed
by expanding the general solution in terms of coherent states, while in [26,39] it was
analyzed by scrutinizing the time evolution of the spread in position of the solu
tion; in [45] it has been shown that both approaches are not conclusive. Finally, [7]
proposed Theorem 1.2 as a conjecture, but shows stability of t only against small
perturbations. Building on previous work of Holevo, Mora and Rebolledo recently
enhanced in [46, 47] the general theory of stochastic Schrodinger equations. In par
ticular, they developed criteria for the existence of regular invariant measures for
a large class of stochastic Schrodinger equations as an important step towards an
understanding of the large time behavior. Until now however the only complete and
detailed results on the large time behavior seem to be Theorems 1.1 and 1.2.
We conclude this introductory section by summarizing the content of the paper.
In Sec. 2, we will present a qualitative analysis of the time evolution of the general
solution of Eq. (1.1); we will discuss the three regimes previously introduced, giving
also numerical estimates, and we will set the main problems which we aim at solving.
In Sec. 3, we will analyze the structure of the Greens function (1.12) and prove
Theorem 1.1. In Sec. 4, we will introduce another representation of the general
solution of Eq. (1.1), which is more suitable for analyzing its long time behavior.
Sec. 5 will be devoted to the proof of Theorem 1.2. Finally, Sec. 6 will contain some
concluding remarks and an outlook.
Notation. We will work in the complex and separable Hilbert space L2 (R), with
the norm and the scalar product given, respectively, by and . We will also
consider the subspace L2B (R) of all bounded functions of L2 (R). Given an operator
O, we denote with D(O) its domain and with R(O) its range.
Since in some expression the real and imaginary parts of some coecients
appear, we introduce for ease of readability the symbols z R or zR will denote the
real part of the complex number z, while z I or zI will denote its imaginary part.
Given the linear SDE (1.6), a topological strong solution is an L2 valued process
such that for any t > 0,
t
i t p2 t 2
t = s ds + qs ds q s ds (1.25)
0 2m 0 2 0
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
62 A. Bassi, D. D
urr & M. Kolb
t2 2
t = t
= tanh t, (2.3)
t
tbt
a
t = a t + , (2.4)
t
b2
ct = ct + t , (2.5)
2t
t x + bt
Ytx = . (2.6)
t
64 A. Bassi, D. D
urr & M. Kolb
2 sinh t + sin t
R
t =
cosh t + cos t
2t (1.20 1025 m2 kg1 sec1 )mt t 1 ,
= 2 (2.10)
(2.39 1029 m2 kg1 )m t +.
As we see, the spread 1/ R
t is well below , for any t t1 . We can then conclude
that, for times greater than the collapse time, any state initially well spread out in
space is mapped into a very well localized wave function.
An important issue is where the wave function collapses to, given that the initial
state is spread out in space. We now show that the position of the wave function after
the collapse is distributed in very good agreement with the Born probability rule.
A reasonable measure of where the wave function is, after it has collapsed,
is given by the quantum average of the position operator qt . Accordingly, the
probability for the collapsed wave function to lie within a Borel measurable set A
of R can be simply dened to be Pcollt [A] := P[ : qt A]. Though this probability
is mathematically well dened for any Borel measurable subset A, it is physically
meaningful only when A represents an interval much larger than the spread of
the wave function itself, or a sum of such intervals. In such a case, as discussed
in [49], one can show that:
coll 2
Pt [A] EP [P t ] pt (x)dx, (2.11)
where P (x) is the characteristic function of the interval of the real axis and
pt = EP [t (x)2 ]. The idea behind the approximate equality (2.11) is that when t
lies within , then P t t , so that P t 2 is almost equal to 1, while when it
lies outside , it is practically 0. The critical situations, which require special care,
are those when the wave function lies at the edges of .
In [39] it has been proven that:
t 2
pt (x) = dy et y pSch
t (x + y), (2.12)
3mm0 m
t = (2.27 1043 m2 kg1 sec3 ) 3 , (2.13)
22 0 t3 t
times we are considering, the wave function solution of the freeparticle Schr
odinger
equation does not change appreciably, implying that pt (x) p0 (x) = (x)2 ,
Sch Sch
which means precisely that the collapse probability is distributed in agreement with
the Born probability rule.
We can easily recognize in the deterministic parts of the above equations the free
particle equations of motions of classical mechanics describing a particle moving
along a straight line with constant velocity; the remaining terms are the uctuations
around the classical motion, driven by the Brownian motion Wt . The important fea
ture of the above equations is that these uctuations, for macroscopic masses, are
very small, for very long times. As a matter of fact, if we estimate the Brown
ian motion uctuations by setting Wt t, we have for the stochastic terms in
Eq. (2.14):
t 2 3/2 t3/2
Ws ds t (1.63 1022 m kg1/2 sec3/2 ) , (2.16)
m t1 3 m m
t 17 1/2 1/2 t
(Wt Wt1 ) (1.02 10 m kg sec ) . (2.17)
m m m
We see that the random uctuations decrease with the square root of the mass
m of the particle, which means that the bigger the system, the more determin
istic its motion. This is how collapse models recover classical determinism at the
macroscopic level, from a fundamentally stochastic theory.
We can introduce a time t2 , dened as the time after which the uctuations
become larger than L; we can set, e.g., L 1.00 103 m. Since the uctuations
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
66 A. Bassi, D. D
urr & M. Kolb
of a new equation which does not have that stochastic term. To this end, let us
consider the operator Qa : D(Qa ) L2 (R) L2 (R) dened as follows:
Qa (x) = eax (x), a C; (3.1)
where D(Qa ) is dened as the set of all (x) L2 (R) such that eax (x) L2 (R).
It should be noted that, in general, the operator Qa is unbounded and its domain
D(Qa ) is dense in L2 (R) but does not coincide with it. We will settle all technical
issues in the second part of the section. We now dene the vector:
(1)
t = Qt t ; (3.2)
(1)
an easy application of It
o calculus shows that t satises the dierential equation:
(1) i p2 1 2 (1) (1)
dt = Q t Q q t dt, 0 = . (3.3)
2m t
The stochastic dierential q dt has disappeared; in turn, the free Hamiltonian
p2 /2m has been replaced by the operator Qt (p2 /2m)Q1
t
which, due to the
specic commutation relations between q and p, takes the simple form:
Qt p2 Q1
= p2 2i t p 2 t2 . (3.4)
t
68 A. Bassi, D. D
urr & M. Kolb
where D(Pa ) is the set of all (x) L2 (R) which can be analytically continued to
the line x + a in the complex space C, and such that (x + a) L2 (R). Similarly
to Qa , also Pa is in general an unbounded operator and its domain D(Pa ), though
being dense, does not coincide with L2 (R); we will come back to this point later in
this section. We dene the operator:
where the coecients at , bt and ct , yet to be determined, will turn out to be complex
random functions of time. One can easily verify that:
Vt qVt1 = q + bt , (3.10)
Vt pVt1 = p + ct , (3.11)
and similarly for higher powers of q and p. Let us dene the vector:
(2)
t = Vt t , (3.12)
and
1 2 i
a t + ib2t + c t ct = 0, a0 = 0. (3.15)
2m t m
The rst two equations form a nonhomogeneous linear system of rstorder dier
ential equations, which has a unique Qa.s. continuous random solution; the third
equation instead determines the global factor at , which is also random. With such
a choice for the three parameters, Eq. (3.13) becomes:
2
d p
i t = iq 2 t , 0 = , (3.16)
dt 2m
which is the equation of the socalled nonselfadjoint (NSA) harmonic oscillator,
whose solution and most important properties are well known. Before continuing,
we note that in the case of a more general Hamiltonian H = p2 /2m+V (q) appearing
in Eq. (1.1) in place of just the free evolution p2 /2m, the potential V (q) would have
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
been transformed, when going from Eq. (3.7) to Eq. (3.16), according to the rule:
Vt V (q)Vt1 = V (q + bt ); in this case, we would not be able to remove completely
the timedependent terms from the equation and we would not be able to reduce
the original equation to one, whose solution is known. However, besides the free
particle case, all equations containing terms at most quadratic in q and p (among
them, the important case of the harmonic oscillator) can be solved in a similar way.
The solution of Eq. (3.16) admit a representation in terms of the Greens func
tion, also known as Mehlers formula:
2 1
GNSA
t (x, y) = 2
exp (x + y ) coth t + 2 xy sinh t , (3.17)
sinh t
with and dened as in (1.19). In this way, we have established the link between
the solutions of the SDE (1.6) and those of the equation for the NSA harmonic oscil
lator (3.16), which we summarize in the following lemma, whose proof is straight
forward.
Lemma 3.1. Let TtNSA be the evolution operator represented by the Greens func
tion GNSA
t (x, y) and Tt the one represented by Gt (x, y); then:
where the two random functions bt and ct solve the linear system (3.14), and t ,
which includes all global, i.e. independent of x, phase factors, solves the equation:
1 2 i 2 2
t = ib2t ct + t ct + , 0 = 0. (3.19)
2m m 2m t
We now proceed to prove in which sense t := Tt is the topological strong
solution of Eq. (1.6) for the given initial condition . We rst need to set some
properties of the Greens function GNSA
t (x, y) which will be necessary for the sub
sequent theorem.
sinh t sin t
pt = , (3.21)
cosh t cos t
sinh t/2 cos t/2 cosh t/2 sin t/2
qt = ; (3.22)
cosh t cos t
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
70 A. Bassi, D. D
urr & M. Kolb
note that the function pt is positive for any t > 0. The integral of GNSA t (x, y)2
with respect to y is equal to:
NSA 2 2 p2t 4qt2 2
dyGt (x, y) = exp 2 x . (3.23)
(sinh t sin t) pt
A simple calculation shows that p2t 4qt2 > 0 for any t > 0; this means that
GNSA
t (x, ), taken as a function of y, belongs to L2 (R) for any x R and t > 0;
moreover :
dxGNSA
t (x, )2 < + for any t > 0. (3.24)
with
sinh t + sin t
pt = , (3.26)
cosh t cos t
sinh t/2 cos t/2 + cosh t/2 sin t/2
qt = . (3.27)
cosh t cos t
The above formulas imply that, for any a, b C, for any x R and for any
t > 0, the function ebx GNSA
t (x + a, ) belongs to L2 (R) and:
dxebx GNSA
t (x + a, )2 < +. (3.28)
We are now in a position to state and prove the main theorem of this section.
Theorem 3.1. Let Pa and Qa be dened, respectively, as in (3.8) and (3.1); let bt
and ct solve the linear system (3.14) and t be the solution of Eq. (3.19). Finally,
let t = Tt , with L2 (R) and Tt dened as in (3.18). Then the following three
statements hold true with probability 1:
Proof. Statement 1. Let belong to L2 (R); since also GNSA t (x, ) belongs to
L2 (R) for any x R and t > 0, H olders inequality implies that GNSA
t (x, )
belongs to L1 (R); accordingly, the operator TtNSA is well dened for any t > 0, and
maps any L2 (R)function into a measurable function. By using Schwartz inequality
together with relation (3.24), we have:
2
dx dy GNSA
t (x, y)(y) 2 dxGNSA (x, )2 < +;
t (3.32)
thus TtNSA belongs to L2 (R) for any in L2 (R) and for any t > 0.
In a similar way, since also GNSA
t (x + a, ) belongs to L2 (R) for any a C and
because of (3.28), one proves that Pa TtNSA belongs to L2 (R) for any L2 (R),
for any complex a and for any t > 0, i.e. that D(Pa ) contains R(TtNSA ). Using once
more the same inequalities and (3.28), one shows also that Qb Pa TtNSA belongs to
L2 (R) for any in L2 (R), for any a, b C and t > 0.
Remark. Actually a stronger statement is true, as can be readily seen from the
Gaussian form of the Greens function Gt of the operator Tt : For positive t, it
maps L2 (R) to Schwartz space S(R). We shall need this information in the proof
of Statement 3.
Statement 2. Let us consider the vector t := TtNSA , with L2B (R). By
construction, t solves Eq. (3.16), once one proves that the integration
dy GNSA
t (x, y)(y) (3.33)
can be exchanged with the rst and second partial derivatives with respect to
x and with the rst partial derivative with respect to t. We note that the
function GNSA
t (x, y)(y) satises the following two properties: (i) The function
y GNSAt (x, y)(y) is measurable and integrable on R for any t > 0 and for
any x R; (ii) The rst and second partial derivatives with respect to x and the
rst partial derivatives with respect to t are exists for any t > 0, x R and y R
and can be bounded uniformly with respect to t and x. Accordingly, one can apply,
e.g., [50, Theorem 12.13, p. 199] to conclude that the operations of integration and
dierentiation can be exchanged.
Having proved that t solve Eq. (3.16), a direct application of It
o calculus proves
that t , dened as in (3.18), is a topological strong solution of Eq. (1.6).
Statement 3. Let = 0 Cc (R) be given. Since t solves Eq. (1.6) in a strong
sense, it also solves the SDE in a weak sense; hence, using, e.g., [31, Eq. (1.1)], one
has:
We extend (3.34) to the general case of L2 (R). Being dense in L2 (R), there
exist a sequence {n Cc (R), n N} which approximates any L2 (R). By
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
72 A. Bassi, D. D
urr & M. Kolb
The rst term on the righthand side can be made arbitrarily small because
of (3.34); the second and third terms can also be made arbitrarily small by choosing
n suciently large, while t can be bounded as it converges to 0 for t 0,
due to Eq. (1.9). This proves that:
Statement 3 for test functions Cc (R) now follows directly from Eq. (1.9),
Eq. (3.36) and observing t 0 2 = t 2 +0 2 20 t R . It remains to extend
the strong continuity of Tt from the subspace Cc (R) to L2 (R). For this, observe
that for Cc (R) (t 2 )t0 denes a stochastic process with continuous paths
and by Holevos result (cf. Eq. (1.9)) it is a martingale. For given f L2 (R) choose
a sequence (n )nN Cc (R), which converges to f in L2 (R). Doobs inequality
for submartingales implies that for all n, m N, T > 0 and > 0
1
Q sup t t  > EQ [nT 2 m
n 2 m 2 2
T ]. (3.37)
0tT
lim EQ [nT 2 m 2
T ] = 0. (3.38)
n,m
nt 2 m 2 n m n m
t  (t + t )t t
implies that
EQ [nt 2 m 2 n m n m
t  EQ [(t + t )t t ]
1 1
(EQ [(nt + m 2 2 n m 2 2
t ) ]) (EQ [t t ])
2 12
2(EQ [nt 2 ] + EQ [m n m 2
t ])
1
= 2(n 2 + m 2 ) 2 n m .
1
d(X, Y ) = EQ min 1, sup (X Y )s  (X, Y D).
n=1
2n 0sn
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
it suces to show that for some T > 0 limn suptT ft nt  = 0. But
suptT ft nt  suptT ft nt . Therefore, we need only establish
that limn suptT ft nt = 0. This is done by a similar argument as above,
namely we show that for every > 0
n 2
lim Q sup ft t > = 0,
n tT
t (x) = exp[t (x xm 2 m
t ) + ikt x + t ], (3.39)
74 A. Bassi, D. D
urr & M. Kolb
I
dktm = Rt [dt 2 xm t ], (3.42)
t
I m
dtR = (xmt )2
+ t + R
dt + xm t [dt 2 xt dt], (3.43)
m 4t
I m 2 R tI tI m
dt = (kt ) t + R 2
dt + R xt [dt 2 xm
t dt]. (3.44)
2m m 4(t ) t
In particular, the solution of Eq. (3.40) is t = (/) coth(t + ), where sets the
initial condition. These results will be useful in the subsequent analysis.
It is easy to see that the sequences {(n) } and {(n) } form a biorthonormal
system; one then denes the (nonorthogonal) projection operators:
Pn (n) (n) = n (n) , (4.4)
which satisfy the relations:
lnPn
Pn Pm = n,m Pn , Pn = (n) 2 and lim = 2c, (4.5)
n+ n
where c is an appropriate constant [54]. As we see, although the states (n) are
normalized, in the sense that
+
(n) (x)(m) (x)dx = n,m , (4.6)
+
2 t x+t
xt )2 /2+ik
= ez (x
z n e(1+i)n t/2 H
n [z(x bt )], (4.9)
n=0
where n = (n) (see Eq. (4.4)), while the two real parameters xt , kt and the
complex parameter t are dened as follows:
t = bR
x I I
t + bt (2/m)ct + (/2 )t , (4.10)
kt = (m/)bIt + (1/)(cR I
t ct ) + t , (4.11)
t = (1 i)(m/4)(b2t x
2t ) + (i/)t . (4.12)
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
76 A. Bassi, D. D
urr & M. Kolb
By resorting to Eqs. (3.14) and (3.19), and after a rather long calculation, we obtain
the following set of SDEs for these parameters:
xt = kt dt +
d [dt 2 xt dt], (4.13)
m m
dkt = [dt 2 xt dt], (4.14)
dtR = x2t + dt + xt [dt 2 xt dt], (4.15)
4
I 2
dt = k + dt xt [dt 2 xt dt]; (4.16)
2m t 4
0 = k0 = 0 = 0. Note that these equations are equiv
the initial conditions are: x
alent to (3.41)(3.44), with t = = / = z 2 /2, x t = xm m
t , kt = kt and
t = t + (1 + i)/4; as a matter of fact, the above equations describe the time evo
lution (according to Eq. (1.6)) of the ground state of the NSA harmonic oscillator,
which is:
2
z 2 1+i
t (x) = exp (x xt ) + ikt x + t t , 0 (x) =
(0)
(x). (4.17)
2 4
As we shall prove in the next section, this is the state to which apart from
normalization any initial state converges to, in the long time limit, hence the
name t .
As we see, due to the stochastic part of the dynamics, the argument of the
Gaussian weighting factor and that of the Hermite polynomials of Eq. (4.9) are
dierent functions of time, while for analyzing the long time behavior of the wave
function, it is more convenient that both arguments display the same time depen
dence. We thus modify the argument of the Hermite polynomials, to make it equal
to that of the weighting factor. To this end, let us dene t = x t bt ; we can then
write:
n [z(x bt )] = 1
H Hn [z(x x
t ) + zt ]
2n n!
n
1 n
= (2zt )nm Hm [z(x x t )]
n
2 n! m=0 m
n
n!
= ( 2zt )nm H m [z(x x
t )], (4.18)
m=0 m!(n m)!
where Hm is the standard (not normalized) Hermite polynomial of degree m; in
going from the rst to the second line, we have used property (A.2). Resorting to
the above relation, we can rewrite Eq. (4.9) as follows:
+
(m) (1+i)mt/2 (m)
t (x) = eikt x+t (1+i)t/4
t e (x x
t ); (4.19)
m=0
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
the functions (m) are the eigenstates dened in (4.3), while the time dependent
(m)
coecients t are dened as follows:
+
(m)
(k + m)! k
t = k+m ( 2z t ) , (4.20)
k=0
m!k!
Lemma 4.1. Let L2 (R) and n = (n) , with (n) dened as in (4.3).
(m)
Then the series (4.20) dening t is a.s. convergent for any m and any t > 0.
Moreover, one has the following bound on the coecients:
+ k(c+1)
(m) (c+1/2)m e  2z t k

t  Nt e , Nt A a.s., (4.21)
k=0 kk
The cases k = 0 and m = 0 can be treated separately, giving the same bound, with
the only possible dierence of an overall constant factor. This proves convergence
of the series dened in (4.20) and the bound (4.21).
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
78 A. Bassi, D. D
urr & M. Kolb
Theorem 4.1. Let the conditions of Lemma 4.1 be satised; let moreover t
e(1+i)t/2 t , where t = x
t bt with x
t and bt solutions of Eqs. (4.13) and (3.14),
respectively. Then the series dened in (4.19) is a.s. norm convergent for t > t
(4c + 1)/. In addition, the following equality holds true:
+
(m) (1+i)mt/2 (m)
Tt = eikt x+t (1+i)t/4
t e (x xt ), t > t, (4.26)
m=0
Theorem 5.1. Let t be a strong solution of Eq. (1.6) that admits, for t > t
a representation as in (4.26). Let t t /t (when t = 0), which can be
written as follows:
+
(m)
t x+ I t/4)
t
t = t +e i(k t e(1+i)mt/2 m (x x
t ), (5.1)
m=1
r t
with:
(0)
t i(kt x+tI t/4)
t := e 0 (x xt ), (5.2)
rt
+
(m) (1+i)mt/2
rt :=
t e m (x xt ) . (5.3)
m=0
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
lim t t = 0. (5.4)
t
Note that, apart from global factors, t is the ground state of the NSA har
monic oscillator, randomly displaced both in position space as well as in momentum
space.
Proof. According to Eq. (5.1), all we need to prove is that, with Pprobability 1:
+
(m)
t (1+i)mt/2
lim e m (x xt ) = 0. (5.5)
t r t
m=1
thus all we need to set is the long time behavior of rt and Nt . Lemmas 5.1 and 5.2
(see Eqs. (5.7) and (5.12)) state that, with Pprobability 1, rt converges asymp
totically to a nite and nonnull random variable, while Nt converges to a nite
random variable. From the above properties, the conclusion of the theorem follows
immediately.
Proof. According to Eqs. (4.19) and (5.3), the following equality holds:
R
t = et t/4 rt ; (5.8)
resorting to the stochastic dierentials (1.9) and (4.15) for t 2 and tR , respec
tively, one can write the following stochastic dierential equation for rt2 :
drt2 = [2 (qt xt )dt + 4(
x2t qt x
t ) dt]rt2 , r02 = 1. (5.9)
By using relation (1.11), the above equation can be rewritten in terms of the
Wiener process Wt as follows:
drt2 = [2 (qt xt )dWt + 4(qt x
t )2 dt]rt2 , r02 = 1, (5.10)
80 A. Bassi, D. D
urr & M. Kolb
Proof. Looking back at Eq. (4.21), we see that in order to prove this lemma it is
sucient to show that t tends to a nite limit as t , with Pprobability 1.
According to our previous denition, t is equal to:
t = e(1+i)t/2 (
xt bt ). (5.13)
Equations (3.14) and (4.10), together with the change of measure (1.11), lead
to the following stochastic dierential equation for t in terms of the Wiener
process Wt :
dt = e(1+i)t/2 [dWt + 2 (qt x t )dt], 0 = 0. (5.14)
2
Once again, the large time behavior of qt x
t (see Eq. (5.15)) yields the conclusion
of the lemma.
t = t xG
a G
t + ikt , (5.21)
G
with ft qt xt . By integrating the second equation, by using the strong law of
large numbers applied to Wt , Eq. (B.1) for ft and the fact that t has an asymptotic
nite limit, one can show that, with Pprobability 1, the process ktG grows slower
than t2 , for t . By integrating now the rst equation, and by using the same
properties as before, one can show that x G 3
t grows slower than t , for t and
again with Pprobability 1. According to Eqs. (5.21) and (1.17), we then have, with
Pprobability 1:
t = o(t3 ) as t ,
a lim bt = b nite. (5.24)
t
82 A. Bassi, D. D
urr & M. Kolb
In this way, we have proven that any initial state is Pa.s. norm convergent to
the Gaussian state (5.2), which can be written as follows:
2
z
t 4 2 exp (x x t )2 + ikt x + i tI t , (5.25)
zR 2 4
which has a xed nite spread both in position and in momentum, given by [39]:
2 1/2
q = t (q xt ) t = , (5.26)
m
2 1/2 m
p = t (p kt ) t = . (5.27)
2
This corresponds almost
to the minimum allowed by Heisenbergs uncertainty rela
tions, as q p = / 2. Note also that, the more massive the particle, the smaller
the spread in position of the asymptotic Gaussian state: this is a well known eect
of the localizing property of Eq. (1.1). Finally, Eqs. (4.13) and (4.14), together with
the change of measure (1.11), tell how the average position x t and momentum kt
evolve in time, as a function of the Wiener process Wt :
d
xt = kt dt + ht dt + dWt , (5.28)
m 2
dkt = 2ht dt + dWt , (5.29)
which imply that there exist two random variables X and K such that [35]:
t
t = X + Kt +
x Ws ds + Wt + O(et/2 ), (5.30)
m m 0 m
kt = K + Wt + O(et/2 ). (5.31)
These parameters fully describe the time evolution of the Gaussian state (5.25).
I. Collapse regime
Let be the length which discriminates between a localized and a nonlocalized
wave function, i.e. such that, dening with q the spread in position of a wave
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
function , we
say that is localized
in space whenever q . In our case, we
must take > /m, where /m is the asymptotic spread (see Eq. (5.26)).
TCOL := min{t : t q }. (6.1)
How is TCOL distributed? Find best possible bounds (depending on the param
eters dening the model) for the distribution function.
The dependence of the collapse time on the parameters of the model is physically
relevant as for macroscopic bodies the collapse is supposed to happen at a much
shorter time, producing a classical macroscopic body. This time must be much
before diusion becomes eective. Bounds on the collapse time will lead hopefully
to experimentally testable deviations from linear quantum mechanics (i.e. where
the superposition principle holds on all scales.)
Acknowledgments
The work was supported by the EU grant No. MEIF CT 2003500543 and by DFG
(Germany). We thank GianCarlo Ghirardi, Lajos Diosi and the referees for helpful
comments and an anonymous referee for pointing out a aw in a previous version.
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
84 A. Bassi, D. D
urr & M. Kolb
n/2
(1)m (2z)n2m
Hn (z) = n! , (A.1)
m=0
m!(n 2m)!
where z is any complex number. These polynomials satisfy the following addition
rule
n
n
Hn (z1 + z2 ) = (2z2 )nm Hm (z1 ). (A.2)
m=0
m
When the argument is real (z = x R), they form an orthogonal set with respect
to the weight exp[x2 ]; the normalized Hermite polynomials are:
n
n (x) = 1 Hn (x),
H Nn = 2 n!. (A.3)
Nn
G
ft qt xt = O(e
t/2
), (B.1)
G
where qt = t qt , and xt has been dened in (5.18).
Proof. Using the expression (1.12) for Gt (x, y) together with Schwartz inequality,
we can derive the following bound on t :
2 2 p2t 4qt2 2
t (x) Kt  exp 2 x
Rt pt
bR qt
R t R (bR
t )
2
+2 a
t + 8 x + 2
ct + , (B.2)
pt 2 pt
which holds for any t > 0. The above inequality implies that it is sucient to
consider L2 (R) such that:
2
(x) CeAx , (B.3)
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
where C and A are random variables. A direct calculation leads to the following
expression for the quantum average t qt :
2
aR
( t )
t qt = Kt 2 exp 2c R
t + dy1 dy2 (y1 )(y2 )
Rt R
t
R
t y 1 + t y 2 a
t 1 t2 2 1 t2
+ exp t y 1 t y22
2R
t R
t 2 2 R
t 2 2 R
t
R
R
2
t a
a t 
exp bt + Rt y1 + bt + t Rt y2 + y1 y2 . (B.4)
t t 2R t
As we shall soon see, all exponential terms in the above expression can be controlled.
The crucial factors are the two within brackets: the rst term decays exponentially
in time, since t = O(et/2 ), while t has a nite asymptotic limit; the term
R
a R
t /t , instead, does not decay in time (see the discussion in connection with the
proof of Lemma 5.3). Since t 2 is equal to the expression (B.4) without the terms
in square brackets, and because of (5.18), we have that
R
a t
ft t 2 = t qt t 2 (B.5)
R t
2
aR
( t )
t y 1 + t y 2
= Kt 2 exp 2
c R
t + dy 1 dy 2 (y 1 )(y 2 )
Rt R t 2R
t
2
2
1 1
exp t tR y12 t t R y22
2 2t 2 2t
R
t a aR t 2
exp bt + Rt y1 + bt + t Rt y2 + y y
1 2 . (B.6)
t t 2Rt
with:
1 R (tR )2 2 R tR a
Rt
g(y) exp t y + bt + y . (B.9)
2 Rt Rt
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
86 A. Bassi, D. D
urr & M. Kolb
Next, by using the inequality g(y1 ) + g(y2 ) (g(y1 )2 + g(y1 )2 )/2 and the symmetry
between y1 and y2 , we have:
2
2 t  2 R aR
( t )
ft t Kt  exp 2
ct + dy1 dy2 (y1 )(y2 )
2Rt Rt R t
the key point is that, since Gt (x, y) solves Eq. (1.6), then Gt (, y)2 is a positive
martingale with respect to the measure Q, for any value of y; we call MarQ (t, y)
this martingale. We can then write:
t 
ft t 2 dy1 dy2 (y1 )(y2 )(y1  + y2 ) MarQ (t, y)
2Rt
t  2
dy eAy (A1 y + A2 ) MarQ (t, y), (B.12)
2Rt
where A1 and A2 are suitable constants. In going from the rst to the second line,
we have used (B.3). The quantity
1 2
R
dyeAy (A1 y + A2 ) MarQ (t, y) (B.13)
2t
is another positive martingale with respect to Q, which we call Mar Q (t). We arrive
in this way at the inequality:
Mar Q (t)
ft  t  . (B.14)
t 2
Since Mar Q (t) is a positive martingale with respect to Q, then MarP (t) =
Mar Q (t)/t 2 is a positive martingale with respect to P which, by Doobs con
vergence theorem, has a Pa.s. nite limit for t +. The conclusion of the
lemma then follows from Eq. (1.15), according to which t = O(et/2 ).
References
[1] G. C. Ghirardi, A. Rimini and T. Weber, Unied dynamics for microscopic and
macroscopic systems, Phys. Rev. D 34 (1986) 470491.
[2] G. C. Ghirardi, P. Pearle and A. Rimini, Markov processes in Hilbert space and
continuous spontaneous localization of systems of identical particles, Phys. Rev. A
42 (1990) 7889.
[3] G. C. Ghirardi, R. Grassi and P. Pearle, Relativistic dynamical reduction models:
General framework and examples, Found. Phys. 20 (1990) 12711316.
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
[4] P. Pearle, Reduction of the state vector by a nonlinear Schrdinger equation, Phys.
Rev. D 13 (1976) 857868.
[5] P. Pearle, Combining stochastic dynamical statevector reduction with spontaneous
localization, Phys. Rev. A 39 (1989) 22772289.
[6] P. Pearle, Collapse Models, in Open Systems and Measurement in Relativistic Quan
tum Theory, eds. F. Petruccione and H.P. Breuer (SpringerVerlag, Berlin, 1999).
[7] L. Diosi, Localized solution of simple nonlinear quantum Langevinequation, Phys.
Lett. A 132 (1988) 233236.
[8] L. Diosi, Models for universal reduction of macroscopic quantum uctuations, Phys.
Rev. A 40 (1989) 233236.
[9] L. Diosi, Relativistic theory for continuous measurement of quantum elds, Phys.
Rev. A 42 (1990) 50865092.
[10] S. L. Adler, D. C. Brody, T. A. Brun and L. P. Hughston, Martingale models for
quantum state reduction, J. Phys. A 34 (2001) 87958820.
[11] S. L. Adler and T. A. Brun, Generalized stochastic Schr odinger equations for state
vector collapse, J. Phys. A 34 (2001) 47974809.
[12] S. L. Adler, Quantum Theory as an Emergent Phenomenon. The Statistical Mechan
ics of Matrix Models as the Precursor of Quantum Field Theory (Cambridge Univer
sity Press, Cambridge, 2004).
[13] A. Bassi, E. Ippoliti and S. L. Adler, Towards quantum superpositions of a mirror:
An exact open systems analysis, Phys. Rev. Lett. 94 (2005) 030401, 4 pp.
[14] A. Bassi, E. Ippoliti and B. Vacchini, On the energy increase in spacecollapse models,
J. Phys. A 38 (2005) 80178038.
[15] V. P. Belavkin, Nondemolition measurements, nonlinear ltering and dynamic pro
gramming in quantum stochastic processes, in Lecture Notes in Control and Infor
mation Science, ed. A. Blaqui`ere, Vol. 121 (SpringerVerlag, Berlin, 1988).
[16] S. L. Adler and A. Bassi, Is quantum theory exact? Science 325 (2009) 275276.
[17] V. P. Belavkin and P. Staszewski, A quantum particle undergoing continuous obser
vation, Phys. Lett. A 140 (1989) 359362.
[18] V. P. Belavkin and P. Staszewski, Nondemolition observation of a free quantum
particle, Phys. Rev. A 45 (1992) 13471357.
[19] D. Chruscinski and P. Staszewski, On the asymptotic solutions of Belavkins stochas
tic wave equation, Phys. Scripta 45 (1992) 193199.
[20] A. Barchielli, Direct and heterodyne detection and other applications of quantum
stochastic calculus to quantum optics, Quantum Opt. 2 (1990) 423441.
[21] A. Barchielli, On the quantum theory of measurements continuous in time, Proceed
ings of the XXV Symposium on Mathematical Physics (Toru n, 1992), Rep. Math.
Phys. 33 (1993) 2134.
[22] A. Barchielli and A. S. Holevo, Constructing quantum measurement processes via
classical stochastic calculus, Stochastic Process. Appl. 58 (1995) 293317.
[23] Ph. Blanchard and A. Jadczy, On the interaction between classical and quantum
systems, Phys. Lett. A 175 (1993) 157164.
[24] Ph. Blanchard and A. Jadczyk, Eventenhanced quantum theory and piecewise deter
ministic dynamics, Ann. Physik 4(8) (1995) 583599.
[25] Ph. Blanchard and A. Jadczyk, Events and piecewise deterministic dynamics in event
enhanced quantum theory, Phys. Lett. A 203 (1995) 260266.
[26] J. Halliwell and A. Zoupas, Quantum state diusion, density matrix diagonalization,
and decoherent histories: A model, Phys. Rev. D 52 (1995) 72947307.
[27] J. Halliwell and A. Zoupas, Postdecoherence density matrix propagator for quantum
Brownian motion, Phys. Rev. D 55 (1997) 46974704.
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
88 A. Bassi, D. D
urr & M. Kolb
[28] H.P. Breuer and F. Petruccione, The Theory of Open Quantum Systems (Oxford
University Press, New York, 2002).
[29] H.P. Breuer, U. Dorner and F. Petruccione, Numerical integration methods for
stochastic wave function equations, Comp. Phys. Comm. 132 (2000) 3043.
[30] D. Gatarek and N. Gisin, Continuous quantum jumps and innitedimensional
stochastic equations, J. Math. Phys. 32 (1991) 21522157.
[31] A. S. Holevo, On dissipative stochastic equations in a Hilbert space, Probab. Theory
Related Fields 104 (1996) 483500.
[32] V. P. Belavkin and V. N. Kolokoltsov, Quasiclassical asymptotics of quantum
stochastic equations, Teoret. Mat. Fiz. 89 (1991) 163177 (Russian); translation in
Theoret. and Math. Phys. 89(2) (1991) 11271138.
[33] V. N. Kolokoltsov, Application of quasiclassical methods to the study of Belavkins
quantum ltering equation, Mat. Zametki 50 (1991) 153156 (Russian); translation
in Math. Notes 50 (1991) 12041206.
[34] V. N. Kolokoltsov, Scattering theory for the Belavkin equation describing a quantum
particle with continuously observed coordinate, J. Math. Phys. 36 (1995) 27412760.
[35] V. N. Kolokoltsov, Localization and analytic properties of the solutions of the sim
plest quantum ltering equation, Rev. Math. Phys. 10 (1998) 801828.
[36] V. N. Kolokoltsov, Semiclassical Analysis for Diusion and Stochastic Processes,
Lecture Notes in Mathematics, Vol. 1724 (SpringerVerlag Berlin, 2000).
[37] S. Albeverio, V. N. Kolokoltsov and O. G. Smolyanov, Continuous quantum meas
urement: Local and global approaches, Rev. Math. Phys. 9 (1997) 907920.
[38] S. Teufel, Adiabatic Perturbation Theory in Quantum Dynamics, Lecture Notes in
Mathematics, Vol. 1821 (SpringerVerlag, Berlin, 2003).
[39] A. Bassi, Collapse models: Analysis of the free particle dynamics, J. Phys. A 38
(2005) 31733192.
[40] E. Joos and H. D. Zeh, The emergence of classical properties through interaction
with the environment, Z. Phys. B 59 (1985) 223243.
[41] W. Marshall, C. Simon, R. Penrose and D. Bouwmeester, Towards quantum super
positions of a mirror, Phys. Rev. Lett. 91 (2003) 130401, 4 pp.
[42] J. Z. Bernad, L. Di
osi and T. Geszti, Quest for quantum superpositions of a mirror:
High and moderately low temperatures, Phys. Rev. Lett. 97 (2006) 250404, 4 pp.
[43] S. L. Adler, A density tensor hierarchy for open system dynamics: Retrieving the
noise, J. Phys. A 40 (2007) 89598990.
[44] A. Barchielli, Some stochastic dierential equations in quantum optics and measure
ment theory: The case of diusive processes, in Contributions in Probability in
Memory of Alberto Frigerio, ed. C. Cecchini (Forum, Udine, 1996), pp. 4355.
[45] A. Bassi and D. Durr, On the longtime behavior of Hilbert space diusion, Europhys.
Lett. 84 (2008) 10005.
[46] C. M. Mora and R. Rebolledo, Regularity of solutions to linear stochastic Schr
odinger
equations, Inn. Dimens. Anal. Quantum Probab. Relat. Top. 10 (2007) 237259.
[47] C. M. Mora and R. Rebolledo, Basic properties of nonlinear stochastic Schr odinger
equations driven by Brownian motions, Ann. Appl. Probab. 18 (2008) 591619.
[48] R. S. Liptser and A. N. Shiryaev, Statistics of Random Processes (SpringerVerlag,
Berlin, 2001).
[49] A. Bassi, G. C. Ghirardi and D. G. M. Salvetti, The Hilbertspace operator formalism
within dynamical reduction models, J. Phys. A 40 (2007) 1375513772.
[50] R. G. Bartle, A Modern Theory of Integration, Graduate Studies in Mathematics,
Vol. 32 (American Mathematical Society, Providence, RI, 2001).
February 11, 2010 10:1 WSPC/148RMP J070S0129055X10003886
Within the framework of algebraic quantum field theory, we propose a new method
of constructing local generators of (global) gauge symmetries in field theoretic models,
starting from the existence of unitary operators implementing locally the flip automor
phism on the doubled theory. We show, in the simple example of the internal symmetries
of a multiplet of free scalar fields, that through the pointlike limit of such local generators
the conserved Wightman currents associated with the symmetries are recovered.
1. Introduction
One of the most important features of eld theoretic models is the existence of local
conserved currents corresponding to spacetime and internal (gauge) symmetries.
While in the framework of classical Lagrangian eld theory a clarication of this
issue comes from Noethers theorem (which provides an explicit formula for the
conserved current associated to any continuous symmetry of the Lagrangian itself),
it is well known that in the quantum case several drawbacks contribute to make
the situation more confusing. For example, symmetries which are present at the
classical level can disappear upon quantization due to renormalization eects.
In [1, 2], a dierent approach to the problem was outlined in the context of
algebraic quantum eld theory. It consisted of two main steps: (1) given double
cones O, O with bases B, B in the timezero plane centered at the origin and such
that O O, start from generators Q of global spacetime or gauge transformations
Q
and construct local ones, i.e. operators JO, generating the correct symmetry on
O
the eld algebra F (O) and localized in O (i.e. aliated to F (O));
and (2) these
local generators should play the role of integrals of (time components of) Wightman
91
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
currents over B with a smooth cuto in B and possibly some smearing in time, so
that one is led to conjecture that
1 Q Q
f (x)x (JO, )dx cj0 (f ) (1.1)
3 R4 O
vol(B) c vol(B).
(1.2)
It is important to note that there is a large ambiguity in the choice of the local
generators: since their action in O O
is not xed by the above requirements we are
free to add perturbations in F (O O). Thus, the limit (1.1) is not to be expected
to converge in full generality, but we can still hope that a canonical choice or
construction of the local generators might solve the problem (see below).
The rst problem above was completely solved in [1] for the case of Abelian
gauge transformation groups, while in [2,3] the general case (including discrete and
spacetime symmetries and supersymmetries) was treated. The nal result was that
in physically reasonable theories what was called by the authors a canonical local
unitary implementation of global symmetries exists and if a part of them actu
ally constitutes a Lie group the corresponding canonical local generators provide a
local representation of the associated Lie (current) algebras. A key assumption was
identied in the socalled split property (for double cones), which holds in theories
with a realistic thermodynamic behavior [4]. It expresses a strong form of statistical
independence between the regions O and O and is equivalent to the existence of
normal product states on F (O) F (O) such that (AB) = (A)(B) ( being
the vacuum state) for A F (O) and B F (O) [5].
However, the abovementioned construction crucially depends on such a highly
elusive object as the unique vector representative of the state in the (natural)
cone
) = 1/4 (F (O) F (O)
P (F (O) F (O) )+ (1.3)
(see [6]), where indicates the vacuum vector and the modular operator of
the pair (F (O) F (O) , ), so that nding an explicit expression of the local
generators appears as an almost hopeless task. This makes it extremely hard to
proceed to the abovementioned second step, i.e. the determination of the current
elds themselves. Notwithstanding this, the reconstruction of the energy momentum
tensor of a certain (optimal) class of 2dimensional conformal models was carried out
in [7], while partial results for the U(1)current in the free massless 4dimensional
case were obtained in [8], showing that for the local generators of [3] the drawbacks
briey discussed after Eq. (1.1) might be less severe. However, in both cases the
existence of a unitary implementation of dilations was crucial for handling the limit
0.
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
This was used in [3] to dene the universal localizing map : B(H ) B(H ),
(T ) = U (T 1)U , T B(H ),
where the standard typeI factor N = (B(H )) satises F (O1 ) N
F (O2 ). For the commutant standard inclusion = (F (O2 ) , F (O1 ) , ) [10], one
has (T ) = U (1 T )U .
For any unitarily equivalent triple 0 = (V0 F (O1 )V0 , V0 F (O2 )V0 , V0 ), one
nds U0 V0 = V0 V0 U . Notice that in the case of gauge transformations
= 0 and so
U V (g) = V (g) V (g) U . (1.7)
It is then straightforward to verify that, with Z1,3 the unitary interchanging the
rst and third factors in H H H H , the operator
W = (U U )Z1,3 (U U )
is a local implementation of the ip. Setting g = g in (1.7) and dierentiating with
respect to , a simple computation shows that
W (1 Q)W = (Q) 1 + 1 (Q) = JQ 1 + 1 JQ ,
where JQ , JQ are the canonical local implementations of [2, 3], which of course
satisfy (1.6). Choosing now = U ( ), we see that Q
O1 ,O2 (Q) = J . The
above construction (1.5) therefore includes the canonical one as a particular case.
As remarked above, the control of the limit (1.1) for such operators does not
seem within reach of the presently known techniques. However, we shall see in
Sec. 3 below that if Q is the (unbounded) generator of a 1parameter subgroup of
a compact Lie gauge group acting on a nite multiplet of free scalar elds of mass
m 0, it is possible to provide a dierent explicit (semi)local implementation
of the ip WO1 ,O2 such that the limit (1.1) can actually be performed for the
Q
corresponding generator JO 1 ,O2
(which is selfadjoint and satises (1.6) in the same
Q
sense as J ).
The rest of the paper is organized as follows. In Sec. 2, we introduce a new
class of test functions spaces and use it to obtain estimates concerning certain
free eld bilinears; as it is shown in the Appendix, these estimates also allow to
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
c,
i () = a( ei ),
c,+
i () = a( ei ) ,
where L2 (R3 ) and a(), K, is the usual Fock space annihilation operator.
Their commutation relations are
, ,
[ci (), cj ()] = ij , , dp (p)(p).
R3
we have
1 , 1 +,
i (f ) = ci (j f ), i (f ) = ci (j f ).
2 =+, 2 =+,
where : l c,
i k c+,
j : (f )(n) : K S (n) K S n is a bounded operator whose
expression can be obtained from the formal expression of i in terms of creation
and annihilation operators. For instance, if K S n ,
(: l ci,+ k cj+, : (f )(n) )i11...i
...n
n
(p1 , . . . , pn )
X
n
= r ,+ i,ir il+r (1)k
r=1
Z
dpm (p)k1/2 m (pr )l1/2 f(pr,+ , p+ )+1 ...
r ...n
(p, p1 , . . . , p
r , . . . , pn ),
ji1 ...ir ...in
R3
where the hat over an index means that the index itself must be omitted and
where we have introduced the convention (which we will use systematically in the
following) of denoting simply by q R4 the 4vector (m (q), q), = +, .
We now want to show that such operators can be extended to suitable spaces
of tempered distributions on R8 , which in turn are left invariant by the operation
induced by the commutator of eld bilinears.
Definition 2.1. We denote by C the space of functions f C (R8 ) such that for
all r N, , N40 ,
f
r,, = sup (1 + p + q)r p q f (p, q) < .
(p,q)R8
k,l,
f (p, q) := f (p+ , q+ )m (p)k1/2 m (q)l1/2 ,
f
k,l := max{
T k,l (f )
,
T l,k (f)
,
k,l,
f
L2 (R6 ) }.
The spaces Ck,l depend also on the mass m appearing in m , but we have
avoided to indicate this explicitly in order not to burden the notations. It is
clear that functions in C are bounded with all their derivatives and therefore
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
(2.2)
defines a bilinear map C l,k : Cl ,l Ck,k Cl ,k , such that
C l,k (f, g)
l ,k
2
f
l ,l
g
k,k .
where, by recursion,
1
r1
1 2(2r 3) (2r 2k + 1) 1
Jr (u) := dv 2 2
= 2k 2
1 (u v )r (2r 2) (2r 2k) u (u 1)rk
k=1
(2r 3)!! 1 u + 1
+
log ,
(2r 2)!! u2r1 u 1
which easily gives estimate (2.3) with s(r, 0) = r 3. Take now h = 1. Dividing
the integration region into the subregions {x 1}, {x > 1} and using the
CauchySchwarz inequality in the rst integral, one gets
1/2
I1,r () x1 dx I0,2r ()1/2 + I0,r () O(r3 ).
x1
Finally, for h = 1, taking into account the bound x1/2 /(1 + x + 1 e)(1 + x
1 e) 1/2, one gets I1,r () O(r4 ).
We now show that if f Cl ,l , g Ck,k , then C l,k (f, g) Cl ,k . We introduce
the notation K to denote the HilbertSchmidt operator on L2 (R3 ) with kernel
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
T k ,l (C l,k
(f, g))
2 (
T k ,k (
g )T l,l (f)
2 +
Kl ,l, Kk,k ,+ 
2 ),
f  g
so that T l ,k (C l,k (f, g)) and T k ,l (C l,k (f, g)) are bounded. Furthermore one has,
2 6
for L (R ),
l ,k ,+
C k,k ,+
l,k (f,g) , L2 (R6 )  (g , (T l ,l (f ) 1)L2 (R6 )
+ lf,l,+
 , (1 T
k ,k
g) )L2 (R6 ) )
(
l ,k , l ,l,
C l,k (f,g) , L2 (R6 )  (f  , (1 T
k,k
(g))L2 (R6 )
k,k ,
+ g , (T l,l (f) 1)L2 (R6 ) )
l ,k , 2 6
l,k (f,g) L (R ). The bound on
C
so that by Riesz theorem C l,k (f, g)
l,k now
follows at once from the above estimates.
For (f, g) C l ,l C k,k , we write C l,k (f, g) := C l,k (f, g) .
Proposition 2.1. The following statements hold for any i, j {1, 2}, k, l
{0, 1}, n N, , {+, }, with n 0.
(N , : l i k : (f )](N
+ 1)1/2 [N + 1)1/2
f
l,k , (2.6)
j
for some > 0. If furthermore (f, g) C l ,l C k,k , there holds, on D
0,
[: l i l i : (f ), : k j k j : (g)]
= ij : l i k j : (C l,k (f, g)) i ,j : k j l i : (C k ,l (g, f ))
,+
+ il+l +k+k 2 i ,j i,j ((1)l+l lf,l, , k,k
g L2 (R6 )
(1)k+k lf,l,+ ,
k,k ,
g L2 (R6 ) )1. (2.7)
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
n
: l ci,+ k cj+, : (f )(n) = il (i)k Vr ((T l,k (f) e+ +
i ej ) 1 1),
r=1
1,n
il+k
: l ci,+ k cj+,+ (n)
: (f ) = Wr,s (l,k,+
f
(e+
i ej )) ,
n(n 1) r=s
: l ci, k cj+, : (f )(n) = (i)l+k (n + 1)(n + 2)(l,k,
f
(e +
i ej )),
where for i K, i = 1, . . . , n,
Vr 1 n = 2 1 n ,
r th place
Wr,s 1 n = 3 1 2 n .
rth place sth place
Remark 2.1. It is not dicult to see that the above dened exten
sion of : l c,
i k c+,
j : ()(n) to C l,k is unique in the family of linear maps
S (n)
S : C l,k
B(K , K S n ) which are sequentially continuous when
S (n) S n
B(K ,K ) is equipped with the strong operator topology and C l,k
is equipped with the topology induced by the family of seminorms
f
k,l, = max{
T k,l (f)
,
T l,k (f)
,
k,l,
f
L2 (R6 ) }, L2 (R3 ),
with respect to which S (R8 ) is sequentially dense in C l,k . On the other hand,
we point out the fact that, according to Eq. (2.7), the linear span of extended
eld bilinears is stable under the operation of taking commutators. Together with
Proposition A.1 in the Appendix, this implies that in the construction of the local
symmetry generator carried out in the following section, Eq. (3.8), only the above
dened extensions are relevant.
According to the results in [12, Sec. X.5], the bounds (2.5) and (2.6) imply that
: l i k j : l i k j : (f ) can be extended to an operator, denoted by the same
symbol, whose domain contains D(N ).
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
and if we consider
i
e 0
V () := 1 ,
0 ei
(1 (f )) = cos 1 (f ) sin 2 (f ),
(3.1)
(2 (f )) = sin 1 (f ) + cos 2 (f ).
In Proposition A.1 in the Appendix it is shown that the Noether current of this
U(1) symmetry
J (x) = 1 (x) 2 (x) + 1 (x) 2 (x) 1 (x)2 (x) 1 (x) 2 (x) (3.2)
where Or , Or+ are the double cones with bases Br , Br+ , respectively. It
then follows easily that setting h := with (x) = (1 x) and
(t) = 1 (1 t), the unitary operator
WOr ,Or+ := (1 Z)ei 2 J0 (h ) F (Or+ )
B(H ), (3.4)
Proposition 3.1. Let the operator WOr ,Or+ be defined as above. The operator
Or ,Or+ (Q) defined on D(N ) by
Or ,Or+ (Q) = P1 WOr ,Or+ (1 Q)WOr ,Or+
, D(N ), (3.5)
Proof. We start by observing that, for all H for which the righthand side
of (3.5) is dened, one has
Or ,Or+ (Q) = P1 ei 2 J0 (h ) (1 Q)ei 2 J0 (h ) . (3.9)
It follows from this formula that Or ,Or+ (Q) is welldened (and symmetric) on
D(N ): according to formula (A.1) in the Appendix for J0 (h ), Proposition 2.1(2)
and [11, Lemma 2], we have ei 2 J0 (h ) D(N
) D(N
) and D(N ) D(Q) as remarked
above.
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
)
Recalling now the denition of Q one has on D(N
2
Q1 () := i[J0 (h ), 1 Q] = [: j j : ((h ) ) : j j : ((h ) )]
j=1
0,1
2
= : l j k j : (K1,m
l,k
()),
j=1 l,k
where j = 3 j. Proceeding now inductively using formula (2.7), one veries that
0,
there are operators Qn () such that, on D
Qn+1 () = i[J0 (h ), Qn ()], (3.10)
0,1
2
Q2n () = (1)j+1 : l j k j : (K2n,m
l,k
()),
j=1 l,k
(3.11)
0,1
2
Q2n+1 () = : l j k j l,k
: (K2n+1,m ()),
j=1 l,k
l,k
where the distributions Kn,m () C l,k satisfy (3.7). It is also easy to verify induc
tively that the distributions K l,k () are real (g S being real if g, f = g, f),
n,m
so that Qn () is symmetric. Arguing again by induction, it follows from (3.7) and
Lemma 2.1, that
K
n,m
l,k
()
l,k (8)n1 (max{
(h
)
0,1 ,
(h )
1,0 }) (8)
n n1
h
nS ,
where
h
S is some xed Schwartz norm of h. The last inequality above follows
from Lemma A.1 and from the observation that, switching for a moment to the
(m)
notation
l,k in order to make explicit the dependence on the mass m of the
seminorms
l,k , one has
(h
(m) (m)
)
l,1l =
h
l,1l , l = 0, 1.
Using now the bounds in Proposition 2.1(2) and the results in [12, Sec. X.5], we see
that Qn () can be extended to an operator (denoted by the same symbol) which
. The domain D
is essentially selfadjoint on any core for N 0 being such a core,
Eq. (3.10) can be assumed to hold weakly on D(N ) D(N
) and we are therefore
in the position of applying [11, Theorem 1 ] to obtain
1 n
+
ei 2 J0 (h ) (1 Q)ei 2 J0 (h ) = 1 Q + Qn ()
n=1
n! 2
). Combining this with (3.9),
and the series converges strongly absolutely on D(N
and the fact that
P1 : l j k j : (K2n+1,m
l,k
()) = 0 = P1 : l 2 k 2 : (K2n,m
l,k
()) ,
Eq. (3.8) readily follows, upon identication of 1 (f ) = (f ) 1 with (f ).
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
We now show that the unitary group generated by the operator Or ,Or+ (Q)
dened in the above proposition provides a local implementation of the U(1) sym
metry.
Proof. Since the free eld enjoys Haag duality property, it is sucient to show
that
eiOr ,Or+ (Q) ei[(f )+(f ) ]
eiOr ,Or+ (Q) = ei[(f )+(f ) ]
if supp f Or+ and that
i
(f )+ei (f ) ]
eiOr ,Or+ (Q) ei[(f )+(f ) ]
eiOr ,Or+ (Q) = ei[e
if supp f Or . Applying once again [11, Theorem 1 ] and keeping in mind the
previously obtained N bounds for Or ,Or+ (Q), Eq. (3.12), one sees that in order
to achieve this, it is enough to show that for all 1 , 2 D(N )
Or ,Or+ (Q)1 , (f )2 (f ) 1 , Or ,Or+ (Q)2 = 0 (3.13)
for supp f Or+ and
Or ,Or+ (Q)1 , (f )2 (f ) 1 , Or ,Or+ (Q)2 = 1 , (f )2 (3.14)
for supp f Or . In order to prove the latter equation we compute
Or ,Or+ (Q)1 , (f )2
= (1 Q)ei 2 J0 (h) 1 , ei 2 J0 (h) ((f ) 1)2
= (1 Q)ei 2 J0 (h) 1 , (1 (f ))ei 2 J0 (h) 2
= (1 (f ) )ei 2 J0 (h) 1 , (1 Q)ei 2 J0 (h) 2
+ ei 2 J0 (h) 1 , (1 (f ))ei 2 J0 (h) 2
= (f ) 1 , Or ,Or+ (Q)2 + 1 , (f )2 ,
where in the second and fourth equalities we used (3.1) and (3.3), and in the third
equality the fact that, as noted in the proof of Proposition 3.1, ei 2 J0 (h) i
D(N ) and that for 2 D(N
1, ), there holds
(1 Q) 2 (1 (f ) )
1 , (1 (f )) 1 , (1 Q)
2 =
1 , (1 (f ))
2
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
[Q, (f )] = (f ), D(N ),
is the closure of N 1 + 1 N and of the N
of the fact that N bounds hold
ing for 1 Q and 1 (f ). The proof of (3.13) being analogous, we get the
statement.
l,k
In the following lemma, we collect some properties of the distributions Kn,m :=
l,k
Kn,m (1) which will be needed further on. We will use systematically the notations
f
:= sup (1 + p0  + p) f (p),
pR4
1, := max{
,
},
1, := max{
,
1
, . . . ,
3
},
n,m
K l,k
(p, q) = K
n,m
k,l
(q, p), n,m
K l,k n,m
(p0 , Rp, q0 , Rq) = K l,k
(p, q) (3.15)
n N, (3.16)
4 2
for all p = (p0 , p), q = (q0 , q) R4 .
(3) For each n N, the function (p, q, m) R8 [0, 1] K l,k (p, q) is continuous.
n,m
(4) For each n N, the function (p, q, m) R8 [0, 1/e] K n,m
l,k
(p, q) is of
1
class C . Moreover, given > 5, there exists a constant C2 C1 such that
uniformly for all m [0, 1/e] and all smearing functions DR (Br+ ),
DR ((, )),
C1n1 n
K l,k
(p, q) n1, (1 + p)2l (1 + q)2k ,
1,
(3.17)
u n,m 4 2
C2n1 n
K l,k
(p, q) mlog m
1,
Proof. (1) Both properties in (3.15) follow easily by induction from the recursive
n,m
denition of K l,k
, taking into account rotational invariance of the function .
(3.19)
where we recall that k = (m (k), k). Since > 5, there exists a xed constant
dk ks dk
B1 > , , s = 0, 1, 2, p R3 .
R3 k(1 + p k) R3 (1 + k)
so that estimate (3.16) follows by induction from (3.7) and the above expression
for C l,k , provided one denes C1 := 14B1 / and keeps in mind that h (p, q) =
1
4 2 (p 0 + q0 )(p
+ q).
(3) Using (3.16) and the fact that h S (R4 ), we obtain a bound to the inte
1l,r , K ) with an integrable func
grands in C (h r,k
n,m ) and C
r,1k (K
l,r , h
n,m
tion of k, uniformly for (p, q, m) in a prescribed neighborhood of any given
( R8 [0, 1]. By a straightforward application of Lebesgues domi
p, q, m)
nated convergence theorem, the continuity of (p, q, m) K l,k (p, q) follows
n,m
then by induction from the recursive relation (3.7).
(4) Since K l,k Cl,k , we already know that it is dierentiable with respect to the
n,m
components of p and q. The estimate (3.17) and the continuity of (p, q, m)
l,k
u Kn,m (p, q) then follow by an easy adaptation of the inductive arguments
of points (2) and (3) above, using also (3.16). In order to show that K n,m
l,k
K r,k
k )
h(p
n,m k ) K
(k , q) + m (k)rl h(p r,k (k , q).
p0 m n,m
It is now straightforward to verify, using (3.16), (3.17) and the inductive hypoth
esis (3.18), that it is possible to bound the last three terms in the above
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
1
B3 log(1 + 1 + m2 ) , m [0, 1/e].
1 + m2
2l
n+1
C2n1
1, 1, (1 + p)
n+1
(1 + q)2k ,
r,1k l,r
holds for all m [0, 1/e] together with a similar one for m C (Kn,m , h ).
2
Choosing C2 := [16(1 + B3 ) + 16B2 + 7B1 ] C1 , one nally gets (3.18) for
l,k
K n+1,m .
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
In the next theorem, which is our main result, we denote by D0,S the dense
subspace of H of nite particle vectors such that the nparticle wave functions are
in S (R3n ) for each n N.
Theorem 3.1. There holds, for each f S (R4 ) and each D0,S ,
1
lim dx f (x)x (Or ,Or+ (Q)) = cj0 (f ), (3.20)
0 3 R4
and
x (: l k : (K2n,m
l,k
())) y (: l k : (K2n,m
l,k
()))
(8)2n1
h
2n
S
(U (x) U (y) )(N + 1)
+
(U (x) U (y)): l k : (K2n,m
l,k
())U (y)
,
so that the function x x (Or ,Or+ (Q)) is continuous and bounded in norm
for each D0,S , the integral in (3.20) exists in the Bochner sense and furthermore
it is possible to interchange the integral and the series.
Given now K C l,k , it is easy to see that the pointwise product K f still
1
belongs to C l,k
and
K f
l,k (2)2
f
K
l,k so that we can dene K f :=
4
(2) (K f ) C l,k . It is then straightforward to check that
dx f (x)x (: l k : (K2n,m
l,k
())) = : l k : (K2n,m
l,k
() f ).
R4
Furthermore one has K l,k ()(p, q) = 2+l+k K l,k (p, q) and, with the nota
2n,m 2n,m
tion ( K)b(p, q) = K(p,
q), we see that we are left with the calculation of
0,1
+
2n
lim l+k1
n (2n)!
: l k : ( K2n,m
l,k
f ). (3.22)
0
n=1
4
l,k
treated in a similar way. Using then the notations for creation and annihilation
operators and for wave functions introduced in Sec. 2 and the formulas in the proof
of Proposition 2.1, we have
: l c,+ k c+, : ( K2n,m
l,k
f )(N )
16 5 N
((T l,k (( K2n,m
l,k
)bf ) e+ e+ ) 1 1)
,
f
4 2
(1 + p2 ) (1 + pN )
m (q)k1/2 (1 + q)2k
dq ,
R3 (1 + q) (1 + p1 q)
where we have used (3.16) and the fact that D0,S (which gives the constant
B1 > 0). It is now easy to see that the right hand side is a square integrable function
of (p1 , . . . , pN ) if > 3/2, > 3, > 15/2 and therefore we get
: l c,+ k c+, : ( K2n,m
l,k
f )(N )
B2 C1n1
n
,
: l c, k c+, : ( K2n,m
l,k
f )(N 2)
16 5 N (N 1)
l,k, ( K l,k
2 6
,
)bf L (R )
2n,m
with
2(n1)
C1 B3 2n
l,k, l,k b
2L2 (R6 ) 2n
2
f
( K2n,m ) f 16 4
(1 + p)3 (1 + q)3
dp dq ,
R6 (1 + p + q)2
: l k : ( K2n,m
l,k n
f )
B4 C1n1
n
,
=
3C2n1
1, (m + p + q)(1 + p)2 (1 + q)2 ,
1,
4 2
valid for , = and for [0, 0 ], with 0 := min{1/em, 1}. Then a straight
forward adaptation of the above arguments easily gives, uniformly for [0, 0 ],
1
: : ( K2n,m
l,k
f )
B5 C2n1
n
1, 1, , (3.23)
with B5 > 0 a constant independent of and n.
The same estimates above, being uniform in [0, 1/m], together with use of
Lemma 3.1(3), allow us also to conclude that
lim : l k : ( K2n,m
l,k
f ) = (2)4 K
l,k (0, 0): l k : (f ).
2n,0 (3.24)
0
0,0
1 0,0 K 2n,0
lim : : ( K2n,m f ) = (2)4 i (0, 0): : (f ).
0 p0
Proposition 3.3. Assume that the timesmearing function used in the con
struction of Or ,Or+ (Q) satisfies (t) = 1 1 ( 1 t), where 1 DR ((1, 1))
is such that R 1 = 1, and that the spacesmearing function is such that
DR (Br+/2+ ), 0 1 and (x) = 1 for all x Br+/2 , with
< /2 . Then, denoting with c(, ) the corresponding constant given by
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
l,k :
Proof. By induction, it is straightforward to prove the following formula for K n,0
l,k (p, q)
K n,0
0,1
n1
(1)k+n1 inlk n r rj1
= j j
(2)n+1 r1 ,...,rn2 1 ,...,n1 j=1
n1
dkj kj rj rj1 h(p 1
1, k2, ) h(k
k1, )h(k
1 2
n1,
n1 + q),
R3(n1) j=1
in the limit 0 the dependence on the j s drops o the integral in the second
line of the above equation and therefore
0,1 (1)n 22n1 (1)n 4n
lim K2n,0 (0, 0) = (0)
= dx (x)2n .
0 (2)3n+1 2(2)4 R3
But, thanks to the estimates (3.16), (3.17), the convergence of the series (3.21) is
uniform in , so that one has
+
1 (1)n 2n
lim c(, ) = dx (x)2n .
0 2 n=1 (2n)! R 3
Since is bounded above by the characteristic function of the ball Br+ for < /2,
the convergence of the above series is also uniform in so that, taking into account
that converges to the characteristic function of the ball Br+/2 when 0, we
nally get (3.25).
It is straightforward to extend the above analysis to treat the case of the net
O F (O) generated by a multiplet of free scalar elds a , a = 1, . . . , d, with the
action of a compact Lie group G dened by
d
V (g)a (f )V (g) = v(g)ab b (f ), g G,
b=1
d
[Q , a (f )] = i t()ab b (f ),
b=1
l,k
where K2n,m () are the distributions dened in (3.6) and (3.7). Finally, the ana
logue of formula (3.20) holds, where on the righthand side the appropriate Noether
current
d
j0 (f ) = t()ab : a b a b : (f )
a,b=1
These results have been obtained in the spirit of giving a consistency check
towards a full quantum Noether theorem according to the program set down in [1]
and recalled in the introduction. In order to proceed further in this direction it is
apparent that two main problems have to be tackled. First, it is necessary to extend
the construction of local generators proposed in the Introduction to a suitably gen
eral class of theories. Second, it would be desirable to gain a deeper understanding
of the general properties granting the existence and nontriviality of the pointlike
limit of the free generators, which are presently under investigation. Among other
things, this is likely connected with the problem of clarifying if it is generally possi
ble, through a suitable choice of the local ip implementation, to gain control over
the boundary part of the local symmetry implementation, whose arbitrariness is
considered to be an important obstruction for the reconstruction of Noether cur
rents. The methods of [14] can be expected to be useful to put this analysis in a
more general framework.
Finally, we believe that our method could help to shed some light on the dicult
problem of obtaining sharply localized charges from global ones.
Acknowledgments
We would like to thank Sergio Doplicher for originally suggesting the problem to
one of us and for his constant support and encouragement, and Sebastiano Carpi
for several interesting and useful discussions. We also thank the referees for suggest
ing several improvements in the exposition. This work was supported by MIUR,
GNAMPAINDAM, the SNS, the Marie Curie Research Training Network MRTN
CT2006031962 EUNCG and the ERC Advanced Grant 227458 Operator Alge
bras and Conformal Field Theory.
Lemma A.1. Let h S (R4 ), and consider the tempered distribution h (x, y) =
h(x)(x y). Then h C 0,1 C 1,0 for all m 0, and
h
0,1 ,
h
1,0
h
S
where
S is some Schwartz norm independent of m varying in bounded intervals.
February 11, 2010 11:24 WSPC/148RMP J070S0129055X10003904
(p, q) = 1 C. We denote by
Proof. One has h (2)2 h(p
+ q), which implies h
w(p, q) the integral kernel dening T 1,0 (h
). It is easy to see that, for q 1,
m (p)
(1 + p q)1/2 ,
m (q)
2L2 (R3 ) ,
R3 (1 + p)r
where use was made of the Young inequality
f g
L2
f
L1
g
L2 . On the other
hand, there exist C2 > 0 and s > 2 such that, for p > 1,
m (p) C2
dq w(p, q)(q) dq (q)
q (1 + p q)s
q1 q1
m (p) dq
C2 (q)
ps
q1 q
2m (p)
C2
L2 (R3 ) ,
ps
and a C3 > 0 such that, for p 1,
m (p)
dq w(p, q)(q) C3 dq (q)
q
q1 q1
2C3 (1 + m2 )1/4
L2 (R3 ) .
2 1/4
+ 2 2/3C3 (1 + m )
L2 (R3 ) ,
This lemma, together with Proposition 2.1, shows that the timelike component
J0 (h) of the current (3.2) is welldened for h S (R4 ). Using the fact that pi 
m (p), the proof above shows that the spacelike components Ji (h), i = 1, 2, 3, are
welldened too.
2
J (h) := (1)j [: j j : (h ) : j j : (h )], (A.1)
j=1
,
0 is a
and we see that is an analytic vector for J (h). Since any element in D
nite sum of such vectors, essential selfadjointness of J (h) follows.
(2) A straightforward but lengthy calculation shows that, on D 0,
References
[1] S. Doplicher, Local aspects of superselection rules, Comm. Math. Phys. 85 (1982)
7386.
[2] S. Doplicher and R. Longo, Local aspects of superselection rules. II, Comm. Math.
Phys. 88 (1983) 399409.
[3] D. Buchholz, S. Doplicher and R. Longo, On Noethers theorem in quantum eld
theory, Ann. Phys. 170 (1986) 117.
[4] D. Buchholz and E. H. Wichmann, Causal independence and the energy level density
of states in local quantum eld theory, Comm. Math. Phys. 106 (1986) 321344.
[5] D. Buchholz, Product states for local algebras, Comm. Math. Phys. 36 (1974) 287
304.
[6] S. Stratila, Modular Theory in Operator Algebras (Abacus Press, Bucharest, 1981).
[7] S. Carpi, Quantum Noethers theorem and conformal eld theory: A study of some
models, Rev. Math. Phys. 11 (1999) 519532.
[8] L. Tomassini, Sul teorema di Noether quantistico: Studio del campo libero di massa
zero in quattro dimensioni, Masters thesis, Universit`
a di Roma La Sapienza (1999).
[9] C. DAntoni and R. Longo, Interpolation by type I factors and the ip automorphism,
J. Funct. Anal. 51 (1983) 361371.
[10] S. Doplicher and R. Longo, Standard and split inclusions of von Neumann algebras,
Invent. Math. 75 (1984) 493536.
[11] J. Fr
ohlich, Application of commutator theorems to the integration of representations
of Lie algebras and commutation relations, Comm. Math. Phys. 54 (1977) 135150.
[12] M. Reed and B. Simon, Methods of Modern Mathematical Physics. Vol. II: Fourier
Analysis, SelfAdjointness (Academic Press, New York, 1975).
[13] R. M. Corless, G. H. Gonnet, D. E. G. Hare, D. J. Jerey and D. E. Knuth, On the
Lambert W function, Adv. Comput. Math. 5 (1996) 329359.
[14] H. Bostelmann, Phase space properties and the short distance structure in quantum
eld theory, J. Math. Phys. 46 (2005) 052301, 17 pp.
[15] J. Langerholc and B. Schroer, On the structure of the von Neumann algebras gener
ated by local functions of the free Bose eld, Comm. Math. Phys. 1 (1965) 215239.
[16] S. Albeverio, B. Ferrario and M. W. Yoshida, On the essential selfadjointness of
Wick powers of relativistic elds and of elds unitary equivalent to random elds,
Acta Appl. Math. 80 (2004) 309334.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
IGOR KRIZ
Mathematics Department, University of Michigan,
Ann Arbor, MI 481091109 USA
ikriz@umich.edu
The purpose of this paper is to revisit the theory of perturbative deformations of con
formal field theory from a mathematically rigorous, purely worldsheet point of view. We
specifically include the case of N = (2, 2) conformal field theories. From this point of
view, we find certain surprising obstructions, which appear to indicate that contrary to
previous findings, not all deformations along marginal fields exist perturbatively. This
includes the case of deformation of the Gepner model of the Fermat quintic along cer
tain cc fields. In other cases, including Gepner models of K3surfaces and the free field
theory, our results coincides with known predictions. We give partial interpretation of
our results via renormalization and mirror symmetry.
1. Introduction
Recently, there has been renewed interest in the mathematics of the moduli space of
conformal eld theories, in particular, in connection with speculations about elliptic
cohomology. The purpose of this paper is to investigate this space by perturbative
methods from rst principles and from a purely worldsheet point of view. It is
conjectured that at least at generic points, the moduli space of CFTs is a manifold,
and in fact, its tangent space consists of marginal elds, i.e. primary elds of weight
(1, 1) of the conformal eld theory (that is in the bosonic case, in the supersym
metric case there are modications which we will discuss later). This then means
that there should exist an exponential map from the tangent space at a point to
the moduli space, i.e. it should be possible to construct a continuous 1parameter
set of conformal eld theories by turning on a given marginal eld.
There is a more or less canonical mathematical procedure for applying a Pexp
type construction to the eld which has been turned on, and obtaining a perturba
tive expansion in the deformation parameter. This process, however, returns certain
117
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
118 I. Kriz
does not happen in the case of the (compactied or uncompactied) free eld the
ory. Also, there is strong evidence that obstructions to deformations along cc elds
and the corresponding real elds are equivalent (see the remark after Example 2
in Sec. 6).
Since an nth order obstruction indeed means that the marginal eld gets
deformed into a eld of nonzero weight, which changes to the order of the nth
power of the deformation parameter, usually [30, 4547, 6365, 61], when obstruc
tions occur, one therefore concludes that the CFT does not possess continuous
deformations in the given direction. Other interpretations are possible. One thing
to observe is that our conclusion is only valid for purely perturbative theories where
we assume that all elds have power series expansion in the deformation param
eters with coecients which are elds in the original theory. This is not the only
possible scenario. Therefore, as remarked above, our results merely indicate that in
the case when our algebraic obstruction is nonzero, nonperturbative corrections
must be made to the theory to maintain the presence of marginal elds along the
deformation path.
In fact, evidence in favor of this interpretation exists in the form of the analysis
of Nemeschansky and Sen [55,35] of higher order corrections to the function of the
nonlinear model. Grisaru, Van de Ven and Zanon [35] found that the fourloop
contribution to the function of the nonlinear model for CalabiYau manifolds
is nonzero, and [55] found a recipe how to cancel this singularity by deforming
the manifold to metric which is nonRicci at at higher orders of the deformation
parameter. The expansion [4] used in this analysis is around the 0 curvature tensor,
but assuming for the moment that a similar phenomenon occurs if we expanded
around the Fermat quintic vacuum, then there are no elds present in the Gepner
model which would correspond perturbatively to these higher order corrections in
the direction of nonRicci at metric: bosonically, such elds would have to have
critical conformal dimension classically, since the model Lagrangian is classically
conformally invariant for nonRicci at target K ahler manifolds. However, quan
tum mechanically, there is a oneloop correction proportional to the Ricci tensor,
thus indicating that elds expressing such perturbative deformations would have
to be of generalized weight (cf. [3942]). Fields of generalized weight, however, are
not present in the Gepner model, which is a rational CFT, and more generally
are excluded by unitarity (see discussions in Remarks after Theorems 2 and 3 in
Sec. 3 below). Thus, although this argument is not completely mathematical, renor
malization analysis seems to conrm our nding that deformations of the Fermat
quintic model must in general be nonperturbative. It is also noteworthy that the
function is known to vanish to all orders for K3surfaces because of N = (4, 4)
supersymmetry. Accordingly, we also nd that the phenomenon we see for the Fer
mat quintic is not present in the case of the Fermat quartic (see Sec. 7 below). It is
also worth noting that other nonperturbative phenomena such as instanton correc
tions also arise when passing from K3surfaces to CalabiYau 3folds ( [14, 15, 17]).
Finally, one must also remark that the proof of [55] of the function cancellation
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
120 I. Kriz
require for example regularization of the deformation parameter, and are much
more dicult to calculate. Because of this, we work only with the case of one cc
and one ac eld. We will show that at least one cc deformation, whose real version
corresponds to the quintics
x5 + y 5 + z 5 + t5 + u5 + x3 y 2 = 0 (2)
for small (but not innitesimal) is algebraically obstructed. (One suspects that
similar algebraic obstructions also occur for other elds, but the computation is
too dicult at the moment; for the cc eld corresponding to xyztu, there is some
evidence suggesting that the deformation may exponentiate.)
It is an interesting question if nonlinear models of CalabiYau 3folds must
also contain nonperturbative terms. If so, likely, this phenomenon is generic, which
could be a reason why mathematicians so far discovered so few of these conformal
eld theories, despite ample physical evidence of their existence [3, 55, 23, 60, 44, 66,
67].
Originally prompted by a question of Igor Frenkel, we also consider the case of
the Fermat quartic K3 surface
x4 + y 4 + z 4 + t4 = 0
in CP 3 . This is done in Sec. 7. It is interesting that the problems of the Fermat
quintic do not arise in this case, and all the innitesimally critical elds exponentiate
in the purely perturbative sense. This dovetails with the result of AlvarezGaume
and Ginsparg [5] that the function vanishes to all orders for critical perturbative
models with N = (4, 4) supersymmetry, and hence from the renormalization point
of view, the nonlinear model is conformal for the Ricci at metric on K3surfaces.
There are also certain dierences between the ways mathematical considerations of
moduli space and mirror symmetry vary in the K3 and CalabiYau 3fold cases,
which could be related to the behavior of the nonperturbative eects. This will be
discussed in Sec. 8.
To relate more precisely in what setup these results occur, we need to describe
what kind of deformations we are considering. It is well known that one can obtain
innitesimal deformations from primary elds. In the bosonic case, the weight of
these elds must be (1, 1), in the N = 1supersymmetric case in the NSNS sec
tor the critical weight is (1/2, 1/2) and in the N = 2supersymmetric case the
innitesimal deformations we consider are along so called ac or cc elds of weight
(1/2, 1/2). For more specic discussion, see Sec. 2 below. There may exist innites
imal deformations which are not related to primary elds (see the remarks at the
end of Sec. 3). However, they are excluded under a certain continuity assumption
which we also state in Sec. 2.
Therefore, the approach we follow is exponentiating innitesimal deformations
along primary elds of appropriate weights. In the algebraic approach, we assume
that both the primary eld and amplitudes can be updated at all points of the defor
mation parameter. Additionally, we assume one can obtain a perturbative power
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
122 I. Kriz
a generalization of the coset construction. The formulas required are obtained from
the Coulomb gas approach (= FeiginFuchs realization), which is taken from [34].
The present paper is organized as follows: In Sec. 2, we give the general setup
in which we work, show under which condition we can restrict ourselves to defor
mations along a primary eld, and derive the formula for innitesimally deformed
amplitudes, given in Theorem 1. In Sec. 3, we discuss exponentiation theoretically,
in terms of obstruction theory, explicit formulas for the primary and higher obstruc
tions, and regularization. We also discuss supersymmetry, and in the end show a
mechanism by which nonperturbative deformations may still be possible when alge
braic obstructions occur. In Sec. 4, we give the example of the free eld theories, the
trivial deformations which come from 0 momentum gravitational deforming elds,
and the primary obstruction to deforming along primary elds of nonzero momen
tum. In Sec. 5, we will discuss the Gepner model of the Fermat quintic, and in
Sec. 6, we will discuss examples of nonzero algebraic obstructions to perturbative
deformations in this case, as well as speculations about unobstructed deformations.
In Sec. 7, we will discuss the (unobstructed) deformations for the Fermat quar
tic K3 surface, and in Sec. 8, we attempt to summarize and discuss our possible
conclusions.
124 I. Kriz
case for deformation theory. It should be noted that a physical CFT has still more
structure. Namely, we want to consider the operators (4) where is an annulus
approaching the degenerate annulus which is the unit circle with both inbound and
outbound parametrizations equal to the identity. In such limit, the operator (4)
should approach the identity
H H.
U + V
elements
u0 , . . . , un1 H, u0 = u
such that
m
U (m) = Ui i , U0 = U (6)
i=0
126 I. Kriz
The composition notation on the righthand side means gluing. Granted (9), we can
recover dUd(m) from dUDd(m) for the unit disk D. denotes the Lebesgue measure
(this is well dened on a worldsheet at least up to absolute continuity, which is
sucient for taking the limit in the above computation).
Now in the case of the unit disk, we get a candidate for u(m 1) in the following
way:
Assume that H is topologically spanned by subspaces H(m1 ,m2 ) of weight
(m1 , m2 ) where m1 , m2 0, H(0,0) = UD . Then UD (m) is invariant under rigid
rotation, so
UD (m) H(k,k) []/m+1 . (10)
k0
1 xk x
k
= . (14)
2 k
k1
We see that the element (14) is not an element of H, since its norm is k1 1 = .
This occurs despite the fact that the norm on H is preserved by the deformation,
i.e. the deformation is unitary. (This is because the inner product is conjugate by
reality to the quadratic form which is the operator associated with the degener
ate worldsheet with two outbound boundary components S 1 = {z C z = 1}
parametrized by z and 1/z; in the class of measures equivalent to the Lebesgue
measure by absolute continuity, this worldsheet has measure 0 and hence the defor
mation acts trivially on it.)
The explanation is simply that the innitesimally deformed vacuum is
1 xn x n
1+ . (15)
2 n>0 n
When computing the square norm of (15), the second summand is orthogonal to
the rst, hence its square norm occurs with coecient 2 , which disappears when
calculating up to linear order in (which is what we are doing in an innitesimal
deformation); such phenomena routinely occur when one attempts to dierentiate
unitary processes on Hilbert spaces. In our case, as we shall see, the situation is
further complicated by the fact that the process actually has to be regularized.
There are other problems as well. For one thing, we wish to consider theories
which really do not have Hilbert axiomatizations in the proper sense, including
Minkowski signature theories, where the Hilbert approach is impossible for physical
reasons. Therefore, we prefer a vertex operator algebra approach where we discard
the Hilbert completion and restrict ourselves to examining tree level amplitudes.
One such axiomatization of such theories was given in [41] under the term full eld
algebra. In the present paper, however, we prefer to work from scratch, listing the
properties we will use explicitly, and referring to our objects as conformal eld
theories in the vertex operator formulation.
As mentioned in the introduction, our approach in this paper is essentially to
build the minimal possible machinery in which we can phrase the concept of per
turbative deformation of a CFT along a primary eld of critical weight to arbitrary
degree, and identifying obstructions to obtaining such deformation. Actually identi
fying the deformed conformal eld theory upon plugging in a value of the deforma
tion parameter (provided the obstructions vanish) by means of a general abstract
machinery (i.e. not assuming we can recognize the theory by other means) is a
dicult problem which remains untreated in the present paper. Therefore, speak
ing purely mathematically, we are actually dening the concept of perturbative
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
128 I. Kriz
deformation along with nding our obstructions. It would be far superior to dene
rigorously the moduli space of conformal eld theories upfront, with enough geom
etry to allow us to dene paths. Such technology, however, is not mathematically
available at the present time.
Regarding the approach to constructing and treating elds, the vertex opera
tor approach is largely superior from the computational point of view. One can
convert to the more symmetric and foundationally more powerful Hilbert space
approach when we have appropriate convergence of the operators constructed. We
shall proceed by using either language according to what is more convenient at each
particular time.
For now, let us consider untopologized vector spaces
V = V(wL ,wR ) . (16)
Here (wL , wR ) are weights (we refer to wL , respectively wR , as the left, respectively
right, component of the weight), so we assume wL wR Z and usually
wL , wR 0, (17)
V(0,0) = UD . (18)
The no ghost assumptions (17), (18) will sometimes be dropped. If there is a
Hilbert space H, then V is interpreted as the subspace of states of nite weights.
We assume that for u VwL ,wR , we have vertex operators of the form
Y (u, z, z) = uvL wL ,vR wR z vL zvR . (19)
(vL ,vR )
Here ua,b are operators which raise the left (respectively, right) component of weight
by a (respectively, b). We additionally assume vL vR Z and that for a given
w, the weights of operators which act on w are discrete. Even more strongly, we
assume that
Y (u, z, z) = Yi (u, z)Yi (u, z) (20)
i
where
Yi (u, z) = ui;vL wL z vL ,
(21)
Yi (u, z) = i;vR wR zvR
u
where all the operators Yi (u, z) commute with all Yj (v, z). The main axiom which
elds (19) must satisfy is commutativity and associativity analogous to the
case of vertex operator algebras, i.e. there must exist for elds u, v, w V and
w V of nite weight, a 4point function
w Z(u, v, z, z, t, t)w (22)
which is realanalytic and unbranched outside the loci of z = 0, t = 0, z = ,
t = and z = t, and whose expansion in t rst and z second (respectively, z rst
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
Remark. Even the axioms outlined here are meant for theories which are initial
points of the proposed perturbative deformations, they are too restrictive for the
theories obtained as a result of the deformations themselves. To capture those
deformations, it is best to revert to Segals approach, restricting attention to genus
0 worldsheets with a unique outbound boundary component (tree level amplitudes).
Operators will then be expanded both in the weight grading and in the perturbative
parameter (i.e. the coecient at each power of the deformation parameter will
be an element of the productcompleted state space of the original theory). To
avoid discussion of topology, we simply require that perturbative coecients of all
compositions of such operators converge in the product topology with respect to
the weight grading, and the analytic topology in each graded summand.
In this section, we discuss innitesimal perturbations, i.e. the deformed theory
is dened over C[]/(2 ) where is the deformation parameter. One case where such
innitesimal deformations can be described explicitly is the following
where
Z, (u, v, z, z, t,
t) = Z,,i (u, v, z, t)Z,,i (u, v, z, t)
i
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
130 I. Kriz
and similarly for thes, the s and the s. (The denition makes sense when applied
to elds on which the term with denominator 0 vanishes.) Then
Y,,u (v, t, t)w = ,,i Y (v, t, t)w Y (,,i v, t, t)w Y (v, t, t),,i w
i
Remark 1. Note that technically, the integral (25) is not dened on the nonde
generate worldsheet described. This can be treated in the standard way, namely by
considering an actual worldsheet obtained by gluing on standard annuli on the
boundary components. It is easily checked that if we denote by Auq the innitesimal
deformation of Aq by u, then
Therefore, the theorem can be stated equivalently for the worldsheet . The only
change needs to be made in formula (31), where needs to be multiplied by s2n
and needs to be multiplied by r2n where r and s are radii of the corresponding
boundary components. Because however this is equivalent, we can pretend to work
on the degenerate worldsheet directly, in particular avoiding inconvenient scaling
factors in the statement.
132 I. Kriz
representation (in which case it factors through the abelianization of the fundamen
tal group the rst homology group). Then and only then is the function a sum
of contributions which can be rendered holomorphic by multiplication with appro
priate products of (zi zj )ij . A most basic example of a branched function with
nonabelian holonomy is the hypergeometric function, which occurs as the 4point
function of parafermions and N = 2minimal models.
Even for an abelian theory, the theorem only calculates the deformation in the
0 charge sector because of the assumption (29). Because of this, even for a free
eld theory, we will need to discuss an extension of the argument. Since in that
case, however, stating precise assumptions is even more complicated, we prefer to
treat the special case only, and to postpone the discussion to Sec. 4 below.
where c+ , c are the two parts of along the cut c, oriented from t to 0 and
back respectively. Before going further, let us look at two points x+ c+ , x c
which project to the same point on c. We have
C(e2i 1)
(x ) = C (x+ ) C (x )
(x )
(x+ ) (t + C)
= (t + C)
= (x+ ) (x )
(x )
(x+ ) (0 + D)
= (0 + D)
(x )
(x+ ) D
= D
(x )
= D(e2i 1)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
Comment. This is valid when the constants C,,i , D,,i are both taken at the
point x ; note that since the chiral forms are branched, we would have to adjust
the statement if we measured the constants elsewhere. This however will not be of
much interest to us as in the present paper we are most interested in the case when
the constants vanish.
In any case, note that (36) implies C,,i = 0 when = 0 mod Z and = 0
mod Z, and D,,i = 0 when = 0 mod Z and = 0 mod Z. There is an anlogous
,,i . Note that when = 0 = , all the forms in
relation to (36) between C,,i , D
sight are unbranched, and (32) follows directly. To treat the case = , proceed
analogously, but replacing ,,i, by ,,i,0 or ,,i,t . Thus, we have nished
proving (32) under its hypotheses.
Returning to the general case, let us study the righthand side of (35). Subtract
ing the rst two terms from (33), (34), we get
C,,i ,,i , D,,i ,,i , (37)
t 0
respectively. On the other hand, the sum of the last two terms, looking at points
x+ , x for each x c, can be rewritten as
C,,i (e2i + 1)
,,i = D,,i (e2i + 1)
,,i . (38)
c+ c
Now recall (30). Choosing ,,i, as the primitive function of ,,i , we see that
for the end point x of c ,
,,i, (x ) =
,,i, (x+ )
,,i,t (x )
,,i,t (x+ )
= (e2i 1)
,,i,t (x )
= (e2i 1)
,,i, (x ) + (e2i 1)C,,i .
(39)
Similarly, for the beginning point y of c ,
,,i, (y ) =
,,i, (y + ) + ,,i,0 (y )
,,i,0 (y + ) +
,,i,0 (y )
= (e2i 1)
,,i, (y ) (e2i 1)D
= (e2i 1) ,,i .
(40)
Then (39), (40) multiplied by C,,i are the integrals (37), while the integral (38) is
,,i,0 (y ) + C,,i (1 e2i )
D,,i (1 e2i ) ,,i,0 (x ). (41)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
134 I. Kriz
as claimed.
(Analogously for the s. In the following, we will work on the obstruction for the
chiral part, the antichiral part is analogous.) At rst, these equations seem very
overdetermined. Similarly as in the case of Gerstenhabers obstruction theory, how
ever, of course the obstructions are of cohomological nature. If we denote by A the
Lie algebra L0 1, L1 , L2 , . . . , then the system
Ln (m)u(m 1)
(46)
(L0 (m) 1)u(m 1)
is divisible by m in V []/m+1 , and is obviously a coboundary, hence a cocycle
with respect to L0 (m) 1, L1 (m), . . . . Hence, dividing by m , we get a 1cocycle
in H 1 (A, C). Solving (45) means expressing this Acocycle as a coboundary.
In the absence of ghosts (= elements of negative weights), there is another sim
plication we may take advantage of. Suppose we have a 1cocycle c = (x0 , x1 , . . .)
of A, representing an element of H 1 (A, C). (In our applications, we will be inter
ested in the case when the xi s are given by (46).) Writing out the cocycle condition
explicitly, we obtain the equations
Lk xj Lj xk = (k j)xj+k ,
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
are the conditions for solving (45), i.e. the actual obstruction.
For m = 1, we get what we call the primary obstruction. Calculating the integral
(5) over an annulus and passing to the appropriate limits (the innitesimal annuli
expressing the operators Ln ), we obtain
1k =
L1k = L ui,m+k ui,m , (52)
m,i
so (50) becomes
ui,0 u
i,0 u = 0. (53)
i
136 I. Kriz
Eq. (42) remains the same as in the N = 0 case, the second becomes
(L0 1/2)um = Li0 umi . (55)
i1
we additionally get
(x0 )k = (k 1/2)y,
so
(x0 )1/2 = 0.
so we see that in the absence of ghosts, all zr s are determined, with the exception of
(z1/2 )0 .
this is a superLie algebra on which the central extension vanishes canonically ([31],
Sec. 3.1). Looking at a 1cocycle whose value is xk ,zr , tk on Lk , G
r , Jk respectively,
we get Eq. (57), and additionally
G
r x0 = (L0 + r 1/2)zr , r 1/2 for , r 1/2 for + (67)
and
Jn x0 = (n 1/2)tn , n 0. (68)
+
We see that the cocycle is determined by x0 , with the exception of (z1/2 )0 , (z1/2 )1 .
Therefore, we get the condition
(x0 )1/2 = 0
(z1/2 )0 Im(G
1/2 ) (69)
+
(z1/2 )1 = G+
1/2 u where G+
1/2 u =0
and similarly for the s.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
138 I. Kriz
Remark. The relevant computation in verifying that (70), (71) (and the analogous
cases before) form a cocycle uses formulas of the following type ([72]):
Resz (a(z)v(w)z n ) Resz (v(w)a(z)z n ) = Reszw ((a(z w)v)(w)z n ). (72)
For example, when v is primary of weight 1, a = L2 , the right hand side of (72) is
Reszw (L0 v(z w)2 n(z w)wn1 + L1 v(z w)1 wn )
= nv(w)wn1 + L1 v(w)wn
= nvk wnk2 + (k 1)vk wnk2
= (n k)vn wnk2 .
The lefthand side is [Ln1 , vkn+1 ]wnk2 , so we get
[Ln1 , vkn+1 ] = (n k 1)vk ,
as needed.
Other required identities follow in a similar way. Let us verify one interesting
case when a = G 3/2 , u chiral primary. Then the righthand side of (72) is
Reszw (G 1/2 v(w)(z w) 1 n
w ) = (G
1/2 v)(w) = (G1/2 v)w
n1
.
This implies
[G
r , us ] = (G1/2 u)r+s , (73)
as needed.
We have now analyzed the primary obstructions for exponentiation of innites
imal CFT deformations. However, in order for a perturbative exponentiation to
exist, there are also higher obstructions which must vanish. The basic principle for
obtaining these obstructions was formulated above. However, in practice, it may
often happen that those obstructions will not converge. This may happen for two
dierent basic reasons. One possibility is that the deformation of the deforming eld
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
itself does not converge. This is essentially a violation of perturbativity, but may
in some cases be resolved by regularizing the CFT anomaly along the deformation
parameter. We will discuss this at the end of this section, and will give an example
in Sec. 4 below.
Even if all goes well with the parameter, however, there may be another problem,
namely the expressions for Lin etc. may not converge due to the fact that our
deformation formulas concern vacua of actual worldsheets, while Lin etc. correspond
to degenerate worldsheets. Similarly, vertex operators may not converge in the
deformed theories. We will show here how to deal with this problem.
The main strategy is to rephrase the conditions from the above part of this
section in terms of nite annuli. We start with the N = 0 (nonsupersymmetric)
case. Similarly as in (42), we can expand
m
UAr (m) = UAhr h . (74)
h=0
(76)
In particular, the obstruction is the vanishing of the sum (with the term mh + +m1
omitted from the denominator ) of the terms in (76) with mh + + m1 = 0.
Proof. The identity (75) is essentially by denition. The key point is that in the
higher deformed vacua, there are terms in the integrand obtained by inserting uk ,
k > 1 to boundaries of disjoint disks Di cut out of Ar . Then there are corrective
terms to be integrated on the worldsheets obtained by cutting out those disks. But
the point is that under our weight assumption, all the disks Di can be shrunk to
a single point, at which point the term disappears, and we are left with integrals
of several copies of u inserted at dierent points. If we are using vertex operators
to express the integral, the operators must additionally be applied in time order
(i.e. elds at points of lower modulus are inserted rst). There is an h! permutation
factor which cancels with the Taylor denominator. This gives (75).
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
140 I. Kriz
Now (76) is proved by induction. For h = 1, the calculation is done above ((52)).
Assume now the induction hypothesis, and evaluate the integral in the standard
way of taking primitive functions successively from the inside out. The primitive
function of ms is taken to be ms+1 /(s + 1) (by the induction hypothesis, and the
assumptions that lower order obstructions vanish, the case s = 1 never occurs.
Then the contributing term of the integral where the k 1 innermost integrals have
the upper bound and the kth innermost integral has the lower bound is equal to
UAhk
r
uk , h>k1
(and this term occurs with a minus sign because of the lower bound involved). The
summand which has all upper bounds except in the last integral is equal to
1 r2(m1 ++mh )
um ,m um1 ,m1 ur2 , (77)
2h (mh + + m1 )(mh1 + + m1 ) m1 h h
which is supposed to be equal to
UAr uh + r2 uh .
This gives the desired solution.
Remark. The formula (77) of course does not apply to the case m1 + + mh = 0.
In that case, the correct formula is
ln(r)
um ,m um1 ,m1 ur2 . (78)
(mh1 + + m1 ) m1 h h
So the question becomes whether there could exist a eld uh such that UAr uh r2 uh
is equal to the quantity (78). One sees immediately that such eld does not exist
in the productcompleted space of the original theory. What this approach does
not settle however is whether it may be possible to add such nonperturbative
elds to the theory and preserve CFT axioms, which could facilitate existence of
deformations in some generalized sense, despite the algebraic obstruction. It would
have to be, however, a eld of generalized weight in the sense of [3942].
In eect, written in innitesimal terms, the expression (78) becomes
1
L0 uh uh = um ,m um1 ,m1 u.
(mh1 + + m1 ) m1 h h
The righthand side wu is a eld of holomorphic weight 1, so we see that we have
a matrix relation
uh 1 w uh
L0 = .
u 0 1 u
This is an example of what one means by a eld of generalized weight. One should
note, however, that elds of generalized weight are excluded in unitary conformal
eld theories. By Wick rotation, the unitarity axiom of a conformal eld theory
becomes the axiom of reection positivity [59]: the operator U associated with a
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
where v is another eld of weight 1. Note that in case that u = v, (79) can be
interpreted as saying that u changes weight at order m of the perturbation param
eter. In the general case, we obtain a matrix involving all the (holomorphic) weight
1 elds in the unperturbed theory. Excluding elds of generalized weight in the
unperturbed theory (which would translate to elds of generalized weight in the
perturbed theory), the matrix must have other eigenvalues than 1, thus showing
that some critical elds will change weight.
and accordingly
1
uh =
mk
2h (mh + + m1 )(mh1 + + m1 ) m1
(G1/2 G
1/2 u)m ,m (G1/2 G
h h
1/2 u)m1 ,m1 u, (81)
so the obstruction again states that the term with mh + + m1 = 0 must vanish.
In the N = 2 case, when u is a cc eld, we simply replace G by G in (80) and (81).
But in the supersymmetric case, to preserve supersymmetry along the deforma
tion, we must also investigate the nite analogs of the obstructions associated
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
142 I. Kriz
exp(sG+
1/2 ). (82)
Now we wish to study the deformations of the operator associated with (82) along
a cc eld u as a perturbative expansion in .
Thinking of G+1/2 as an N = 2supervector eld, we have
+
G+
1/2 = (z + ) +
z . (83)
z
We see that (83) deforms innitesimally only the variables + and z, not . Thus,
more specically, (82) results in the transformation
z
exp(s )z,
(84)
.
This gives rise to the formula, valid when uk have weight > (1/2, 1/2) for 1 k < h,
1 th
mh1 1
mh 1
h
Uexp(sG +
)
= t h th1
1/2
mk th =exp(s ) th1 =exp(s )
t2
t1m1 1 dt1 dth vmh ,mh vm1 ,m1 Uexp(sG+ ) , (85)
1/2
t1 =exp(s )
in summands of (85) where the factor resulting from integrating the tk variable has
a factor, and
(G
1/2 G1/2 u)mk ,mk (87)
in other summands. (We see that each summand can be considered as a product of
factors resulting from integrating the individual variables tk ; in at most one factor,
(86) can occur, otherwise the product vanishes.)
Realizing that exp(ms ) = 1 + ms , this gives that the obstruction (under
the weight assumption for uk ) is that the summand for m1 + + mh = 0 (with
the denominator m1 + + mh omitted) in the following expression vanish:
h
1 1
(G G u)m ,m
mk k=1
m1 + + mh m1 1/2 1/2 h h
u)m
mk (G (G
1/2 k+1/2 ,mk 1/2 G1/2 u)m1 ,m1 u. (88)
To investigate the higher obstructions further, we need the language of corre
lation functions. Specically, the CFTs whose deformations we will consider are
RCFTs. The simplest way of building an RCFT is from chiral sectors H
where runs through a set of labels, by the recipe
H= H H
where denotes the contragredient label (cf. [38]). (In the case of the Gepner
model, we will need a slightly more general scenario, but our methods still apply
to that case analogously.) Further, we will have a symmetric bilinear form
B : H H C
with respect to which the adjoint to Y (v, z) is
(z 2 )n Y (ezL1 v, 1/z)
where v is of weight n. There is also a real structure
H
=H ,
144 I. Kriz
which can be dened by taking the vacuum operator associated with the degenerate
worldsheet obtained by cutting out unit disks with centers z0 , . . . , zm from the
unit disk with center z , applying this operator to v0 vm , and taking
inner product with u. Thus, the correlation function (89) is in fact the same thing
as applying the eld on either side of (89) to the identity, and taking the inner
product.
This object (89) is however not simply a function of z0 , . . . , z . Instead, there is
a nitedimensional vector space M depending holomorphically on (called the
modular functor) such that (89) is a linear function
M C.
However, now one assumes that M is a unitary modular functor in the sense
of Segal [59]. This means that M has the structure of a positivedenite inner
product space for not just the as above, but an arbitrary worldsheet. The inner
product is not valued in C, but in
det()2c
where c is the central charge. Since the determinant of as above is the same as
det(P1 ) (hence in particular constant), we can make the inner product Cvalued in
our case.
If the deforming eld is of the form
uu
, (90)
the higher L0 obstruction (under the weight assumptions given above) can be
further written as
v(0) u(zm ) u(z1 )u(0)
0 z1 zm 1
v 
u(zm ) u
(z1 ) z1 dzm d
u(0) dz1 d zm for w(v) 1 (91)
(w is weight) in the N = 0 case and
v(0) (G
1/2 u)(zm ) (G1/2 u)(z1 )u(0)
0 z1 zm 1
u)(zm ) (G
v (G u z1 dzm d for w(v) 1/2
1/2 1/2 )(
z1 )
u(0) dz1 d zm
(92)
u
v (G
)
1/2 )(zm ) (G1/2 u
(z1 ) z1 dzm d
u(0) dz1 d zm for w(v) 0, w(
v ) 1/2 (93)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
and similarly for the . We see that these obstructions vanish when we have
v(z ) u(zm ) u(z0 ) = 0, for w(v) 1 (94)
in the N = 0 case (and similarly for the s), and
v(z ) G
1/2 u(zm ) G1/2 u(z1 )u(z0 ) = 0, for w(v) 1/2, (95)
and similarly for the s. Observe further that when
u
=u
,
the condition for the s is equivalent to the condition for u, and further (94), (95)
are also necessary in this case, as in (91), (92) we may also choose v = v, which
makes the integrand nonnegative (and only 0 if it is 0 at each chirality). In the
N = 2 case, it turns that the condition (95) simplies further:
Theorem 3. Let u be a chiral primary eld of weight 1/2. Then the necessary
and sucient condition (95) for existence of perturbative CFT deformations along
the eld u u
is equivalent to the same vanishing condition applied to only chiral
primary elds v of weight 1/2.
Proof. In order for the elds (95) to correlate, they would have to have the same
Jcharge QJ . We have
QJ u = 1, QJ (G
1/2 u) = 0.
As QJ of the righthand side of (95) is 1. Thus, for the function (95) to be possibly
nonzero, we must have
QJ v = 1. (96)
But then we have
1 1
w(v) QJ v =
2 2
with equality arising if and only if
v is chiral primary of weight 1/2 (97)
(see [31, Sec. 3.3]).
146 I. Kriz
regularization along the ow parameter. This stems from the fact that Eqs. (43),
(44) only determine u() up to scalar multiple, where the scalar may be of the form
1+ Ki i = f (). (101)
i1
But the point is (as we shall see in an example in the next section) that we may
only be able to get a well dened value of
f 1 ()u() = v() (102)
when the constants Ki are innite. The obstruction then is
Ln (m)f (m)v(m) = 0 V []/m+1 for n > 0
(103)
(1 L0 (m))f (m)v(m) = 0 V []/ m+1
.
At rst, it may seem that it is dicult to make this rigorous mathematically with
the innite constants present. However, we may use the following trick. Suppose we
want to solve
c1 a11 + + cn a1n = b1
.. (104)
.
c1 am1 + + cn amn = bn
in a, say, nitedimensional vector space V . Then we may rewrite (104) as
(b1 , . . . , bn ) = 0 V (a11 , . . . , am1 ), . . . , (a1n , . . . , amn ) . (105)
m
This of course does not give anything new in the algebraic situation, i.e when the
aij s are simply elements of the vector space V . When, however the vectors
(a11 , . . . , am1 ), . . . , (a1n , . . . , amn )
are (possibly divergent) innite sums
(a1j , . . . , amj ) = (a1jk , . . . , amjk ),
k
In that sense, (105) always makes sense, while (104) may not when interpreted
directly. We interpret (103) in this way.
Let us now turn to the question of sucient conditions for exponentiation of
innitesimal deformations. Suppose there exists a subspace W V closed under
vertex operators which contains u and such that for all elements v W , we have
that
Yi (u, z)Yi (u, z)v
i
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
148 I. Kriz
and the sucient condition of exponentiability from the last section is met when
we take W the subspace consisting of states of momentum 0. Then W is closed
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
under vertex operators, u W and the n = 0 term of (110) drops out in this case.
However, this is an example where regularization is needed. It can be realized as
follows: Write
= n
n>0
where
xn x
n xn x
n
n = .
n n
We have
exp = exp n . (111)
n>0
150 I. Kriz
1 xn xn xn x
n xn x
n
: exp 1 + : exp tanh()
cosh n n n
(122)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
1 xn xn x
n x
n
: exp 1 + :
cosh n n
xn x
n
exp tanh() .
n
So expressing this as
1
n = , (124)
cosh n
the product
= n
n1
152 I. Kriz
eld, which in this case is determined by the momentum, does not vanish. The
answer is actually known in string theory to correspond to constant deformation of
the metric on spacetime, which ends up isomorphic to the original free eld theory.
From the point of view of string theory, what we shall give is a purely worldsheet
argument establishing this fact.
Let us look rst at the innitesimal deformation of the operator Y (v, t, t) for
some eld v V which is an eigenstate of momentum. We have three forms which
coincide where dened:
1 , z, z)Y (v, t, t)dzd
Y (x1 x z (125)
Y (v, t, t)Y (x1 x
1 , z, z)dzd
z (126)
1 , z t, z t)v, t, t)dzd
Y (Y (x1 x z. (127)
By chiral splitting, if we assume v is a monomial in the modes, we can denote
(125)(127) by (without forming a sum of terms). Again, integrating (125)
(127) term by term dz, we get forms , 0 , t , respectively. Here we set
1
dz = ln z.
z
Again, these are branched forms. Selecting points p0 , p , pt on the correspond
ing boundary components, we can, say, make cuts c0,t and c0, connecting the
points p0 , pt and p0 , p . Cutting the worldsheet in this way, we obtain well dened
branches , 0 , t . To complicate things further, we have constant discrepancies
C0t = 0 t
(128)
C0 = 0 .
These can be calculated for example by comparing with the 4 point function
Y+ (x1 , z)Y (v, t) + Y (v, t)Y (x1 , z) + Y (Y (x1 , z t)v, t) (129)
where Y (v, z) denotes the sum of the terms in Y (v, z) involving negative powers
of z, and Y+ (v, z) is the sum of the other terms. Another way to approach this is
as follows: one notices that
Y (x1 , z)dz = Y (1 , z)S =0 (130)
on the boundary components around 0, t and , and along both sides of the cuts
c0,t , c0, . To get the integrals of the terms in (132)(134) which do not involve the
discrepancy constants, we need to integrate
xn
n
z + x0 ln z x
m zm1
. (135)
n m
n
=0
which will cancel with the integral along the cuts (to calculate the integral over the
cuts, pair points on both sides of the cut which project to the same point in the
original worldsheet), and local terms
2i xn x n 1
(2i)2 x0 x0 . (139)
2 n 2
n
=0
The discrepancies play no role on the cuts (as the forms C0t , C0 are
unbranched), but using the formula (131), we can compensate for the discrepancies
to linear order in by applying on each boundary component
S2ix0 . (140)
In (138), however, when integrating , we obtain also discrepancy terms conjugate
to (140), so the correct expression is
S2ix0 S2ix0 . (141)
The term (141) is also local on the boundary components, so the sum of (139) and
(141) is the formula for the innitesimal isomorphism between the free CFT and
the innitesimally deformed theory. To exponentiate, suppose now we are working
in a Ddimensional free CFT, and the deformation eld is
M x1 x
1 . (142)
Then the formula for the exponentiated isomorphism multiplies left momentum by
exp M (143)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
154 I. Kriz
exp M T . (144)
But of course, in the free theory, the left momentum must equal to the right momen
tum, so this formula works only when M is a symmetric matrix. Thus, to cover the
general case, we must discuss the case when M is antisymmetric. In this case, it may
seem that we obtain indeed a dierent CFT which is dened in the same way as
the free CFT with the exception that the left momentum mL and right momentum
mR are related by the formula
mL = AmR
for some xed orthogonal matrix A. As it turns out, however, this theory is still iso
morphic to the free CFT. The isomorphism replaces the left moving oscillators xi,n
by their transform via the matrix A (which acts on this Heisenberg representation
by transport of structure).
Next, let us discuss the case of deforming gravitaitonal eld of nonzero momen
tum, i.e. when
u = M x1 x1 1 (145)
with = 0. Of course, in order for (145) to be of weight (1, 1), we must have
= 0. (146)
Clearly, then, the metric cannot be Euclidean, hence there will be ghosts and a part
of our theory does not apply. Note that in order for (145) to be primary, we also
must have
M= i
i (147)
i
where
i , =
i , = 0. (148)
Despite the indenite signature, we still have the primary obstruction, which is
xk
xk k
coe z1 z1 : n z m1 zn1 exp
M xm x zk + z :
m,n
k k
k
=0
M x1 x
1 1 (149)
(we omit the z ,x0 term, since the power is 0 by (146)). In the notation (147),
this is
(i x0 i x0 x1 x1 + i x1 x1 + x1 i x1 )
i,j
(
j x0
j x
0
x1
x1 +
j x
1
x1 +
x1
j x
1 )M x1 x
1 1
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
If we set
f () = 2 =0 M ()2
then the Fourier transform of f will be a function g satisfying
2g
2 =0
i
where the signs correspond to the metric, which we assume is diagonal with entries
1. The Fourier transform of the condition (152) is then
2
g = 0. (153)
i j
Assuming a decay condition under which the Fourier transform makes sense, (153)
implies g = 0, hence (151), so in this case also the obstruction is nonzero unless
(145) is a null state.
In this discussion, we restricted our attention to deforming elds of gravitational
origin. It is important to note that other choices are possible. As a very basic exam
ple, let us look at the 1dimensional Euclidean model. Then there is a possibility
of critical elds of the form
a12 + b12 . (154)
This includes the sineGordon interaction [69] when a = b. (We see hyperbolic
rather than trigonometric functions because we are working in Euclidean spacetime
rather than in the time coordinate, which is the case usually discussed.) The primary
obstruction in this case states that the weight (0, 0) descendant of (154) applied to
(154) is 0. Since the descendant is
(4ab)x1 ,
we obtain the condition a = 0 or b = 0. It is interesting to note that in the case of
the compactication on a circle, these cases where investigated very successfully by
Ginsparg [30], who used the obstruction to competely characterize the component
of the moduli space of c = 1 CFTs originating from the free Euclidean compactied
free theory. The result is that only free theories compactify at dierent radii, and
their Z/2orbifolds occur.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
156 I. Kriz
The point is that the innitesimal deformation is obtained by integrating the inser
tion operators of
1/2 M 1/2 1/2 = M x1 x
G1/2 G 1 .
Therefore, (155) behaves exactly the same as a deformation along the eld (142) in
the bosonic case. Again, if M is a symmetric matrix, exponentiating the deformation
leads to a theory isomorphic via scaling the momenta, while if M is antisymmetric,
the isomorphism involves transforming the left moving modes by the orthogonal
matrix exp(M ).
In the case of momentum = 0, we again have indenite signature, and the
eld
Once again, for (156) to be primary, we must have (147), (148). Moreover, again
the actual innitesimal deformation is obtained by applying the insertion operators
of G1/2 G 1/2 u, so the treatment is exactly the same as for the deformation along
the eld (145) in the bosonic case. Again, we discover that under a suitable decay
condition, the obstruction is always nonzero for gravitational deformations of non
zero momentum with suitable decay conditions.
It is worth noting that in both the bosonic and supersymmetric cases, one can
apply the same analysis to free eld theories compactied on a torus. In this case,
however, scaling momenta changes the geometry of the torus, so using deformation
elds of 0 momentum, we nd exponential deformations which change (constantly)
the metric on the torus. This seems to conrm, in the restricted sense investigated
here, a conjecture stated in [59].
Remark. Since one can consider CalabiYau manifolds which are tori, one sees
that there should also exist an N = 2supersymmetric version of the free eld theory
compactied on a torus. (It is in fact not dicult to construct such model directly,
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
The body parts of the bosonic and fermionic vertex operators are given by the
usual formulas
Y (x1 , z) = xn z n1 , Y (y1 , z) = yn z n1 ,
Y (1/2 , z) = s z ns1/2 , Y (1/2 , z) = s z ns1/2 ,
[r , r ] = [r , r ] = 1
[xn , xn ] = [yn , yn ] = n.
We have, say,
As usual,
1
G
3/2 =
(G13/2 iG23/2 ).
2
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
158 I. Kriz
where the sum on the right is over all permutations on the set {1, . . . , k}. Now in
the present case, we have the innitesimal isomorphism on the 0momentum part,
up to nonzero coecient,
(xn iyn )( xn + i
yn )
= (160)
n
and in the absence of regularization, the expansion of the exponentiated isomor
phism on the 0momentum parts is simply
exp().
(The + sign in the s is caused by the fact that we are in the complex conjugate
Hilbert space.) Applying this to (157), we see that we have formulas analogous to
(116)(122), and applying the exponentiated isomorphism to (157), all the terms
in normal order involving x>0 , y>0 will vanish, so we end up with
(xn iyn )(
xn + i
yn )
exp D u
n<0
n
must vanish identically. To see this, we simply observe (157), (158) that in the
present case, u is in the coset model with respect to G 1/2 u (see the discussion
below formula (250) below). Thus, in the N = 2free eld theory, the obstruction
theory works as expected, and in the case discussed, the obstructions vanish. It is
worth noting that in 2ndimensional N = 2free eld theory, we thus have an n2 
dimensional space of cc + aa real elds, and an n2 dimensional space of real ca + ac
elds, and although regularization occurs, there is no obstruction to exponentiating
the deformation by turning on any linear combination of those elds. For a free N =
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
The corresponding sector of the orbifold construction (formula (2.10) of [33]) has
the form
5
(Hik /5 H
(j+i
k )/5
Hik /5 H
(j+i
k )/5
). (162)
P
(ik ): ik 5Z jZ/5Z k=1
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
160 I. Kriz
Call the oscillator corresponding to the jth coordinate xj,m , j = 0, 1, 2. Then the
conformal vector is
1 2 1 2 i 2 1
x x + x1,2 + x22,1 . (163)
2 0,1 2 1,1 2 5 2
The superconformal algebra is generated by
1 5
+
G3/2 = i x2,1 x1,1 1( 5 ,0,i 2 ) ,
2 3 15 3
(164)
1 5
G
3/2 = i x2,1 + x1,1 1( 5 ,0,i 2 ) .
2 3 15 3
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
(, m) = 1( m 2 2 , (165)
15
, i
2
im
3, 2 3)
0 3, m = , + 2, . . . , 2, . (166)
It is obvious that to stay within the range (166), we must understand the fusion rules
and how they are applied. The basic principle is that labels are indentied as follows:
No identications are imposed on the 0th lattice coordinate. This means that upon
any identication, the 0th coordinate must be the same for the labels identied.
Therefore, the identication is governed by the 1st and 2nd coordinates, which give
the Coulomb gas (= FeiginFuchs) realization of the corresponding parafermionic
theory (the Z/3 parafermion model). The keypoint here are the parafermionic cur
rents
1 5
1,2/3 = i x2,1 x1,1 1(0,i 2 ) ,
2 3 3 PF
(167)
1 5
+
1,2/3 = i x2,1 + x1,1 1(0,i 2 )
2 3 3 PF
(the 0th coordinate is omitted). Clearly, the parafermionic currents act on the
labels by
The lattice labels (, m)P F allowed are those which have nonnegative weight. This
condition coincides with (166). Now we impose the identication for parafermionic
labels:
This implies
162 I. Kriz
Now in the Gepner model corresponding to the quintic, the cc elds allowed are
L /L (178)
The correspondence is
As described above, the fusion rules on level 0, 3 are determined by the lattice
theory of L. Additionally, multiplication preserves the correspondence (179), while
the level of the product is restricted only by requiring that any level added to level
0, 3 is the original level.
To put it in another way still, the Verlinde algebra is
where (, m) is a PF label, k Z. Two labels (181) are identied subject to identi
cations of PF labels, and also
(By we mean that the labels (i.e. VA modules) are identied, but we do not imply
that the states involved actually coincide; in the case (183), they have dierent
weights.) Recalling again that we abbreviate (m, , m)MM as (, m)MM , we get the
following labels for the c = 9/5 N = 2 SUSY MM:
Again, the left column (184) represents 0, 3 level labels, the right column represents
level 1, 2 labels. The left column labels multiply as the labels of the lattice super
CFT corresponding to the lattice in C C spanned by
5 2
( 15, 0), , i (185)
15 3
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
164 I. Kriz
(recall that a superCFT can be assigned to a lattice with integral quadratic form;
the quadratic form on C C is the standard one, the complexication of the
Euclidean inner product). The dual lattice of (185) is spanned by
3 5 2
,0 , ,i , (186)
15 15 3
which correspond to the labels (3, 0, 0)MM , (5, 3, 1)MM , respectively. We see that
/
= Z/5. (187)
In (184), the rows (counted from top to bottom as 0, . . . , 4) match the corresponding
residue class (187). The fusion rules for (2, 0)MM , (0, 0)MM are the same as in the
PF case. Hence, again, multiplication of labels preserves the rows (184), and the
Verlinde algebra is isomorphic to
Z[]/(2 1).
The real structure of this last modular functor can be changed by multiplying by
1 the complex conjugation in M for a worldsheet precisely when has an
odd number of boundary components labelled on level 1, 2. The resulting modular
functor of this operation is not positivedenite.
Let us now discuss the question of vertex operators in the PF realization of the
minimal model. Clearly, since the 0th coordinate acts as a lattice coordinate and is
not involved in renaming, it suces the question for the parafermions. Now in the
FeiginFuchs realization of the level 3 PF model, any state can be written as
u1 (189)
where is one of the labels (166) and u is a state of the Heisenberg representation of
the Heisenberg algebra generated by xi,m , i = 1, 2, m = 0. The situation is however
further complicated by the fact that not all Heisenberg states u are allowed for a
given label . We shall call the states which are in the image of the embedding
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
admissible. For example, since the = 0 part of the PF model is isomorphic to the
coset model SU (2)/S 1 of the same level, states
are not admissible for (a, b) = (0, 0). One can show that admissible states are exactly
those which are generated from the ground states (166) by vertex operators and
PF currents. Because not all states are admissible, however, there are also states
whose vertex operators are 0 on admissible states. Let us call them null states. For
example, since (190) is not admissible, it follows that
which is easily seen to be admissible for any choice of (a, b), is null.
Determining explicitly which states are admissible and which are null is
extremely tricky (cf. [34]). Fortunately, we do not need to address the question
for our purposes. This is because we will only deal with states which are explicitly
generated by the primary elds, and hence automatically admissible; because of
this, we can ignore null states, which do not aect correlation functions of admis
sible states.
On the other hand, we do need an explicit formula for vertex operators. One
method for obtaining vertex operators is as follows. We may rename elds using
the identications (169) and also PF currents: a PF current applied to a renamed
eld must be equal to the same current applied to the original eld. Note that this
way we may get Heisenberg states above labels which fail to satisfy (166). Such
states are also admissible, even though the corresponding ground states (which
have the same name as the label) are not. Now if we have two admissible states
ui 1( i ,mi ) , i = 1, 2
always satises our fusion rules, and (up to scalar multiple constant on each module)
is a correct vertex operator of the PF theory. This is easily seen simply by the fact
that (192) intertwines correctly with module vertex operators (which are also lattice
operators).
While in our examples, it will suce to always consider operators obtained
in the form (192), it is important to realize that they do not describe the PF
vertex operators completely. The problem is that when we want to iterate vertex
operators, we would have to keep renaming states. But when two ground states 1 ,
1 are identied via the formula (169), it does not follow that we would have
u1 = u1 (193)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
166 I. Kriz
for every Heisenberg state u. On the contrary, we saw for example that (190) is
inadmissible, while (191) is null. One also notes that one has for example the iden
tication
x1 1 = L1 1 = L1 1 = x1 1 , (194)
which is not of the form (193).
Because of this, to describe completely the full force of the PF theory, one needs
another device for obtaining vertex operators (although we will not need this in the
present paper). Briey, it is shown in [34] that up to scalar multiple, any vertex
operator
u(z)v = Y (u, z)(v)
where u, v are admissible states can be written as
(ak x1 1(0,2) )(tk ) (a1 x1 1(0,2) (t1 )u(z)vdt1 dtk (195)
where the operators in the argument (195) are lattice vertex operators and the
number k is selected to conform with the given fusion rule. While it is easy to show
that operators of the form (195) are correct vertex operators on admissible states
(again up to scalar multiple constant on each irreducible module), as the screening
operators
ax1 1(0,2)
commute with PF currents, selecting the bounds of integration (contours) is much
more tricky. Despite the notation, it is not correct to imagine these as integrals over
closed curves, at least not in general. One approach which works is to bring the
argument of (195) to normal order, which expands it as an innite sum of terms of
the form
(ti tj )ij tk k (196)
(where we put t0 = z) with coecients which are lattice vertex operators. Then
to integrate (196), for ij , k > 0, we may simply integrate ti from 0 to ti1 , and
dene the integral by analytic continuation in the variables ij , k otherwise.
The functions obtained in this way are generalized hypergeometric functions,
and fail for example the assumptions of Theorem 1 (see Remark 2 after the theo
rem). The explanation is in the fact that, as we already saw, the fusion rules are
not abelian in this case.
(In the remaining three coordinates, we will always put the vacuum, so we will
omit them from our notation.) First note that by Theorem 3, this is actually the
only relevant case (95), since the only other chiral primary eld of weight 1/2
with only two nonvacuum coordinates is (2, 2, 2) (3, 3, 3), which cannot correlate
with the righthand side of (95), whose rst coordinate is on level 0, 3. In any
case, we will show therefore that the Gepner model has an obstruction against
continuous perturbative deformation along the eld (170) in the moduli space of
exact conformal eld theories.
Now the chiral correlation function (95) is a complicated multivalued function
because of the integrals (196), which are generalized hypergeometric functions. As
remarked above, the modular functor has a canonical at connection on the space
of degenerate worldsheets whose boundary components are shifts of the unit circle
with the identity parametrization. The at connection comes from the fact that
these degenerate worldsheets are related to each other by applying exp(zL1 ) to
their boundary components. This is why we can speak of analytic continuation
of a branch of the correlation function corresponding to a particular fusion rule.
It can further be shown (although we do not need to use that result here) that
the continuations of the correlation function corresponding to any one particular
fusion rule generate the whole correlation function (i.e. the whole modular functor
is generated by any one nonzero section).
Let us now investigate which number m we need in (95). In our case, we have
G
1/2 (u) = G1/2 (3, 3, 3) (2, 2, 2) (3, 3, 3) G1/2 (2, 2, 2). (197)
(The sign will be justied later;it is not needed at this point.) The rst sum
mand (197) has x0,0 charge (2/ 15, 2/ 15), the second has x0,0 charge (3/ 15,
3/ 15). Thus, the charges can add up to 0 only if m is a multiple of 5. The smallest
possible obstruction is therefore for m = 5, in which case (95) is a 7 point function.
Let us focus on this case. This function however is too big to calculate completely.
Because of this, we use the following trick.
First, it is equivalent to consider the question of vanishing of the function
1(G
1/2 u)(z5 ) (G1/2 u)(z1 )u(z0 )
u(t) . (198)
Now by the OPE, it is possible to transform any correlation function of the form
 v(z)w(t) (199)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
168 I. Kriz
(where (u) means the parity of u). Regarding supersymmetry (if present), an
element H of the superconformal algebra also acts on a tensor product by
H(u v) = Hu v + (1)(H)(u) u Hv, (203)
in particular
G
1/2 (u v) = (G1/2 u) v + (1)
(u)
u (G
1/2 v). (204)
We see that because of (204), the elds a, b, c may have the form of sums of several
terms.
Example 1. Recall that the inner product (more precisely symmetric bilinear
form) of labels considered as lattice points is
r1 r2 s1 s2 t1 t2
(r1 , s1 , t1 ), (r2 , s2 , t2 ) = + . (205)
15 10 6
Recall also (from the denition of energymomentum tensor) that weight of the
label ground states is calculated by
r2 r2 s(s + 2) t2
w(r, s, t)MM = + w(s, t)P F = + . (206)
30 30 20 12
Now we have
u = (3, 3, 3) (2, 2, 2) = (3, 0, 0) (2, 1, 1). (207)
We begin by choosing the eld c. Compose rst u and
= (3, 3, 3) (2, 2, 2) = (3, 0, 0) (2, 1, 1).
u (208)
We choose the nonzero un u of the bottom weight for the fusion rule which adds
the lattice charges on the righthand side of (207), (208). The result is
u1/10 u = (0, 0, 0) (0, 2, 0). (209)
Next, apply G1/2 u to (209). Again, we will choose the bottom descendant. Now
G1/2 u has two summands,
(2, 3, 1) (2, 1, 1) (210)
and
(3, 0, 0) (0, 5, 3)x1 (3, 1, 3) (211)
(the term (211) involves renaming to stay withing noghost PF labels after compo
sition). Applying (210) to (209) gives bottom descendant
(2, 3, 1) (2, 3, 1) of weight 8/5, (212)
applying (211) to (209) gives bottom descendant
(3, 0, 0) (3, 0, 0) of weight 3/5. (213)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
170 I. Kriz
Since (213) has lower weight than (212), (212) may be ignored, and we can choose
c = (3, 0, 0) (3, 0, 0). (214)
Now again, using the formula (204), we see that in the sets of elds Gx , Gy we need
one summand (211) and three summands (210) to get to x00 charge 0. Thus, one of
the groups Gx , Gy will contain two summands of (210) and the other will contain
one. We employ the following convention:
We choose Gy to contain two summands (210) and Gx to contain one
summand (210) and one summand (211). (215)
This leads to the following:
We must choose the elds a and b of the same weights and symmetrize the
resulting correlation function with respect to x and y. (216)
We will choose b rst. Again, we will choose the bottom weight (nonzero) descendant
of (210) applied to itself renamed as
(0, 5, 3)x1 (2, 0, 2) (2, 2, 2), (217)
which is
(4, 3, 1) (4, 3, 1). (218)
We rename to level 0, which gives
b = (0, 5, 3)x1 (4, 0, 2) (0, 5, 3)x1 (4, 0, 2),
(219)
w(b) = 12/5.
Then a must have weight 12/5 to satisfy (216). When calculating a, however, there
is an additional subtlety. This time, we actually have to take into account two
summands, from applying (210) to (211) and vice versa, i.e. (211) to (210). In both
cases, we must rename to get the desired fusion rule. To this end, we may replace
(211) by
(3, 0, 0) (3, 2, 0). (220)
However, when applying (210) and (220) to each other in opposite order, the renam
ings then do not correspond, resulting in the possibility of wrong coecient/sign
(since renaming are correct only up to constants which we have not calculated). To
reconcile this, we must use exactly the same renamings step by step, related only
by applying PF currents. To this end, we may compare the renaming of applying
(0, 5, 3)x1 (2, 0, 2) (2, 2, 2) (221)
to
1
(3, 0, 0) (0, 5, 3)x1 (3, 1, 3) (222)
2
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
(the 1
2 comes from the PF current (5, 3)x1 (0, 2) which takes (2, 2) to 2(2, 0))
and
to
The expression (225) is the negative of (226). On the other hand, we see that the
bottom descendants of applying (210) to (220) and vice versa are the same. This
means that we are allowed to use the names (210) and (220) to each other in either
order, but we must take the results with opposite signs.
Now (226) has weight 7/5, so to get weight 12/5, we must take the descendant of
applying (210) to (220) and vice versa which is of weight 1 higher than the bottom.
This gives
((2, 3, 1) (3, 0, 0))x1 (1, 3, 1) (1, 3, 1)
+ (1, 3, 1) ((2, 1, 1) (3, 2, 0))x1 (1, 3, 1),
which is
Now the correlation function of a(x), b(y), c(z) given in (227), (219), (214) is an
ordinary lattice correlation function. The algorithm for calculating the lattice cor
relation function of elds ui (xi ) which are of the form
1i (xi )
or
i x1 1i (xi )
1P i
172 I. Kriz
by a certain factor, which is a sum over all the ways we may absorb any i x1
factors. Each such factor may either be absorbed by another j x1 , which results
in a factor
i , j (xi xj )2 , i = j (228)
i , j (xi xj )1 , i = j. (229)
Each i x1 must be absorbed exactly once (and the mechanism (228) is considered
as absorbing both i and j ), but one lattice label 1j may absorb several dierent
i x1 s via (229).
Evaluating the correlation function of a(x), b(y), c(z) with the vacuum using
this algorithm, we get
2(y z)
.
(x z)(x y)3
2(x 2z + y)
,
(y z)(x z)(x y)2
1
, (230)
(x z)(y x)
1
. (231)
yx
1
,
(x z)(x y)2
x 2z + y
, (232)
(x z)(x y)2 (y z)
Example 2. In this example, we keep the same a(x) and b(y) as in the pre
vious example, but change c(z). To select c(z), this time we start with G
1/2 u
represented as
and
Therefore, we may ignore (238) and select (237) only. Now applying (237) to u
written as
Recalling from the conjugate of (191) that weight 1 states above the label
(3, 3)P F = (0, 0)P F must vanish, we get
174 I. Kriz
(x 2z + y)2
. (241)
5(y z)2 (x z)2 (x y)2
Let us write again in more detail the contributions of the two summands (227). For
the rst summand, the contribution of the factor before is again (230) (remain
unchanged), and the contribution of the factor after is
y 3z + 4x
. (242)
15(y z)(x z)(x y)
Multiplying, we get
y 3z + 4x
,
15(y z)2 (x z)2 (x y)
and symmetrizing with respect to x, y,
y 2 + 6yz 6z 2 8xy + 6zx + x2
. (243)
15(y z)2 (x z)2 (x y)2
This is the total contribution of the rst summand (227).
For the second summand (227), the coordinate before contributes again (234),
and the coordinate after contributes
2(yx 3yz 3xz + 3z 2 + 2x2 + 2y 2 )
. (244)
15(y z)(x z)2 (x y)2
Multiplying, we get
yx 3yz 3xz + 3z 2 + 2x2 + 2y 2
.
15(y z)(x z)2 (x y)3
Symmetrizing with respect to x, y, we get
2(yx 3yz 3xz + 3z 2 + 2x2 + 2y 2 )
(245)
15(y z)2 (x z)2 (x y)2
which is the total contribution of the second summand (227). Adding the contribu
tions (243) and (245) (which are not equal in this case) gives (241).
Remark. When u is, say, a cc eld of weight (1/2, 1/2) in an N = (2, 2) CFT,
then we have a CPTconjugate aa eld v. Physical CFTs require a real structure,
and the elds u, v are not real. As already noted in the Remark at the end of Sec. 4
for the case of the free eld theory, deforming along the eld u (or v), which is the
case considered in this section, breaks real structure of the CFT. Truly physical
innitesimal deformations therefore occur not along the elds u, v but the eld
u + v. (246)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
(This, of course, explains why the dimension of the space of, say, innitesimal cc
deformations is the dimension of the space of deformations of the complex structure,
and not the double of that number.) In the literature, the contribution of the CPT
conjugate is often ignored (cf. [31, formulas (4.5) and (4.7)]). Nevertheless, from
the point of view of obstruction theory, considering (246) and the original cc eld u
should be equivalent. An argument can be sketched as follows: Let a be a nonzero
complex number. Then replacing u by au, (246) becomes
au + a
v. (247)
Since the obstructions are homogeneous, instead of (247), we may consider
a
u + bv, b = (248)
a
Then b is an arbitrary element of the unit circle S 1 . Thus, even when we restrict to
real deformations, the obstruction should vanish for every eld (248) with b S 1 .
But the chiral part of the obstruction is holomorphic in b, so vanishing for all b S 1
implies vanishing for b = 0 and hence if all of the real deformations along (247) are
unobstructed, so is the deformation along u.
While this argument is compelling, we learned in Sec. 4 that when deform
ing along elds of the form (246), regularization along the deformation parameter
is required. Therefore, to make the argument precise in the present setting, we
would either need to develop a general regularization scheme to the same order
to which obstructions vanish, or compute the regularization parameters explicitly
in the present case. Working this out would be a substantial improvement of the
present result.
The remainder of this section is dedicated to comments on possible perturbative
deformations along the elds (15 , 15 ), (15 , 15 ) (the exponent here denotes repeti
tion of the eld in a tensor product, and 1 again stands for (1, 1, 1)MM , etc.). We
will present some evidence (although not proof) that the obstruction might vanish
in this case. The results we do obtain will prove useful in the next section. Such
conjecture would have a geometric interpretation. In Gepners conjectured interpre
tation of the model we are investigating as the model of the Fermat quintic, the
eld (175) corresponds to the dilaton. It seems reasonable to conjecture that the
dilaton deformation should exist, since the theory should not choose a particular
global size of the quintic. Similarly, the eld (174) can be explained as the dilaton
on the mirror manifold of the quintic, which should correspond to deformations of
complex structure of the form
x5 + y 5 + z 5 + t5 + u5 + xyztu = 0. (249)
Therefore, our analysis predicts that the (body of) the moduli space of N = 2
supersymmetric CFTs containing the Gepner model is 2dimensional, and contains
models of the quintics (249), where the metric is any multiple of the metric for
which the model exists (which is unique up to a scalar multiple).
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
176 I. Kriz
To discuss possible deformations along the elds (15 , 15 ), (15 , 15 ), let us rst
review a simpler case, namely the coset construction: In a VOA V , we set, for u V
homogeneous,
Y (u, z) = un+w(u) z n ,
nZ
Y (u, z) = un+w(u) z n , (250)
n<0
Y (u, z)Y (v, t)w = Y+ (u, z)Y (v, t)w + Y+ (v, t)Y (u, z)w + Y (Y (u, z t)v)w.
(252)
When v, w Vu , the last two terms of the righthand side of (252) vanish, which
proves that
Now in the case of N = 2superVOAs, let us stick to the NS sector. Then (250) still
correctly describes the body of a vertex operator. The complete vertex operator
takes the form
n
Y (u, z, + , ) = un+w(u) z n + u+ n +
n1/2+w(u) z + un1/2+w(u) z
nZ
+ u n +
n1+w(u) z . (253)
D+ Y (u, z, + , ) = Y (G+ +
1/2 u, z, , ),
(254)
D Y (u, z, + , ) = Y (G +
1/2 u, z, , ),
where
D+ = +
+ + , D =
+ . (255)
z z
Now using (252) again, for u V homogeneous, we will have a subN = 2VOA Vu
dened by (251), which is further endowed with the operators G +
1/2 , G1/2 .
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
In the case of lack of locality, only a weaker conclusion holds. Assume rst we
have abelian fusion rules in the same sense as in Remark 2 after Theorem 1.
where each (zi zj )ij are expanded in zj is a power series whose coecients
involve nonnegative integral powers of z0 , . . . , zn only.
n1
zj
nj
+ zn n0 expand1/zn 1
j=1
zn
178 I. Kriz
In (258), expand? (?) means that the argument is expanded in the variable given
as the subscript. The symbol (?)?<0 (respectively, (?)?0 ) means that we take only
terms in the argument, (which is a power series in the subscript), which involve
negative (respectively, nonnegative) powers of the subscript.
In any case, by the assumption of the lemma, all summands (258) vanish with
the exception of the last, which is the induction step.
in the 5fold tensor product of the N = 2 minimal model of central charge 9/5.
Before proving this, let us state the following consequence:
Indeed, assuming Lemma 5 and setting w = ((1, 1, 1), . . . , (1, 1, 1)), the obstruc
tion is
w (G
1/2 )w(zn ) (G1/2 w)(z1 )w . (262)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
(The antichiral primary case is analogous.) But using the fact that
G
1/2 ((G1/2 )w(zn ) (G1/2 w)(z1 )w) = (G1/2 )w(zn ) (G1/2 w)(z1 )G1/2 w
G
1/2 (1, 1, 1) = (4, 1, 1). (263)
so we see that with the fusion rules which stays on levels 1, 2, the vertex operators
u(z)u have only nonsingular terms.
However, this is not sucient to verify (259). In eect, when we use the fusion
rule which goes to levels 0, 3,
(1, 1, 1)(z)(4, 1, 1)
and
(4, 1, 1)(z)(1, 1, 1)
will have most singular term z 2/5 , so when we write again 1 instead of (1, 1, 1)
and G instead of G1/2 (1, 1, 1), with the least favorable choice of fusion rules, it
seems u(z)u can have singular term z 4/5 , coming from the expressions
(G1111)(z)(1G111) (265)
and
(1G111)(z)(G1111) (266)
(and expression obtained by permuting coordinates). Note that with other combi
nations of fusion rules, various other singular terms can arise with z >4/5 .
Now the point is, however, that we will show that with any choice of fusion rule,
the most singular terms of (265) and (266) come with opposite signs and hence
cancel out. Since the z exponents of other terms are higher by an integer, this is all
we need.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
180 I. Kriz
Recalling the KoszulMilnor sign rules for the minimal model, recall that 1 is
odd and G is even, so
We have
with some nonzero coecients M, N , so the bottom descendants of (267) and (268)
are
M N (3, 0, 0) (3, 0, 0)
respectively,
N S N S = N S,
N S R = R, (269)
R R = N S + N S .
We shall again nd it useful to use the free eld realization of the N = 2 minimal
model, which we used in the last two sections. In the present case, the theory is a
subquotient of a lattice theory spanned by
1 1 i i 1
, 0, 0 , , , , , 0, i .
2 8 8 2 2
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
182 I. Kriz
are even, the remaining fusion rules (adding 4 to the kcoordinate on the righthand
side) are odd.
Now the c elds of the MM are
(0, 0, 0), (1, 1, 1), (2, 2, 2)
and the a elds are
(0, 0, 0), (1, 1, 1), (2, 2, 2).
If we denote by H1,2k+1 the state space of label (2k + 1, 1, 1), 0 k 1, and by
H0,2k the state space of label (2k, 0, 0), 0 k 3, then the state space of the 4fold
tensor product of the level 2 minimal model is
3 3 1
0,2ki
H0,2ki H H1,2ki +1 H
1,2ki +1 . (271)
i=0 ki =0 ki =0
The Gepner model is an orbifold with respect to the Z/4group which acts by i on
products in (271) where the sum of the subscripts 2ki or 2ki + 1 is congruent to
modulo 4. Therefore, the state space of the Gepner model is the sum over Z/4
and i Z/4,
3
i = 0 Z/4,
i=0
of
3
H0,2ki H
0,2ki +2
H1,2ki +1 H
1,2ki +1+2
.
i=0 2ki i mod 4 2ki +1i
(272)
It is important to note that each summand (272) in which all the factors have the
odd subscripts 1, 2ki + 1 occurs twice in the orbifold state space.
If we write again for (, , ) and for (, , ), then the critical cc elds
are chirally symmetric permutations of
(2, 2, 0, 0), (2, 2, 0, 0)
(2, 1, 1, 0), (2, 1, 1, 0) (273)
(1, 1, 1, 1), (1, 1, 1, 1).
Note that applying all the possible permutations to the elds (273), we obtain only
19 elds, while there should be 20, which is the rank of H 1,1 (X) for a K3surface
X. However, this is where the preceeding comment comes to play: the last eld
(273) corresponds to a term (272) where all the factors have odd subscripts, and
hence there are two copies of that summand in the model, so the last eld (273)
occurs twice.
By the fact that the Fermat quartic Gepner model has N = (4, 4) worldsheet
supersymmetry (se e.g. [9, 54] and references therein), the spectral ow guarantees
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
that the number of critical ac elds is the same as the number of critical cc elds.
Concretely, the critical ac elds are the permutations of
(0, 0, 2, 2), (2, 2, 0, 0)
(0, 1, 1, 2), (2, 1, 1, 0) (274)
(1, 1, 1, 1), (1, 1, 1, 1).
As above, the last eld (274) occurs in 2 copies, thus the rank of the space of critical
ac elds is also 20.
We wish to investigate whether innitesimal deformations along the elds (273),
(274) exponentiate perturbatively. To this end, let us rst see when the coset
type scenario occurs. This is sucient to prove convergence in the present case.
This is due to the fact that in the present theory, there is an even number of
fermions, in which case it is well known by the bosonfermion correspondence that
the correlation functions follow abelian fusion rules, and therefore Lemma 4 applies.
To prove that the coset scenario occurs, let us look at the chiral c elds
u = (2, 2, 0, 0), (2, 1, 1, 0), (1, 1, 1, 1)
and study the singularities of
G
1/2 (z)(G1/2 u). (275)
By Lemma 4, if (275) are nonsingular, the obstructions vanish. The inner product is
kk mm
(k, , m), (k , , m ) = + ,
8 8 4
k2 ( + 2) m2
w(k, , m) = + .
16 16 8
Next, (2, 2, 2) = (2, 0, 0),
G
1/2 (2, 0, 0) = (0, 4, 2)x1 (2, 0, 2),
G
1/2 (1, 1, 1) = (3, 1, 1),
22 1
The most singular zpower of 2(z)2 is = , (276)
8 2
2 2 1
The most singular zpower of G2(z)2 is = . (277)
8 2
For G2(z)G2, rename the rightmost G2 as (2, 2, 0). We get
(2) (2) 1
The most singular zpower of G2(z)G2 is 1 + = , (278)
8 2
The most singular zpower of 1(z)1 is 0 for the even fusion rule and 1/2
for the odd fusion rule, (279)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
184 I. Kriz
3 1 1
The most singular zpower of G1(z)1 is + + = 0 for the even,
8 8 4
fusion rule and 1/2 for the odd fusion rule, (280)
9 1 1
The most singular zpower of G1(z)G1 is + = 1 for the even
8 8 4
fusion rule and 3/2 for the odd fusion rule. (281)
One therefore sees that for the eld u = (1, 1, 1, 1), (275) is nonsingular: In the
case of the least favorable (odd) fusion rules, the most singular term appears to be
1, coming from
(G1, 1, 1, 1) (1, G1, 1, 1). (282)
However, this term cancels with
(1, G1, 1, 1) (G1, 1, 1, 1). (283)
To see this, note that the last two coordinates do not enter the picture. We have
an odd (respectively, even) pair of pants P respectively, P+ in the MM with input
1, 1. They add up to a pair of pants in MMMM. On (282), we have pairs of pants
Pi {P , P+ },
P (G1 1) (1 G1) = (P1 P2 )(G1 1) (1 G1)
= sP1 (G1 1) P2 (1 G1) (284)
where s is the sign of permuting P2 past G1 1. Here we use the fact that 1 is
even. On the other hand,
P (1 G1) (G1 1) = (P1 P2 )(1 G1) (G1 1)
= sP1 (1 G1) P2 (G1 1) (285)
(as G1 is odd, so there is a by permuting it with itself). From (73), the lowest
term of Pi (1 G1) and Pi (G1 1) have opposite signs, so (284) and (285) cancel
out.
The situation is simpler for u = (2, 2, 0, 0), in which case all the fusion rules are
even, and the most singular term of
(G2 2)(z)(2 G2)
appears to have most singular term z 1 . However, again note that 2 is even and
G2 is odd, so
(G2 s)(z)(2 G2) = G2(z)2 2(z)G2, (286)
while
(2 G2)(z)(G2 2) = 2(z)G(z) G2(z)2. (287)
Renaming G2 as (2, 2, 0), the bottom descendant of both G2(z)2 and 2(z)G2 is
(0, 2, 0) with some coecient, so (286) and (287) cancel out. Thus, the deformations
along the rst and last elds of (273) and (274) exponentiate.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
The eld u = (2, 1, 1, 0) is dicult to analyse, since in this case, (275) has
singular channels and the cosettype scenario does not occur. We do not know how
to calculate the obstruction directly in this case. It is however possible to present
an indirect argument why these deformations exist.
In one precise formulation, the bosonfermion correspondence asserts that a
tensor product of two copies of the 1dimensional chiral fermion theory considered
bosonically (= the level 2 parafermion) is an orbifold of the lattice theory 2 , by
the Z/2group whose generator acts on the lattice by sign.
This has an N = 2supersymmetric
version. We tensor with two copies of the
lattice theory associated with 8 , picking out the sector
m n p
, , where m n p mod 2. (288)
8 8 4
Then the elds (291), (292) are Z/2invariant. In the case of (291), we can proceed
in the lift instead of the orbifold, because the fusion rules in the orbifold are abelian
anyway. In the case of (292), the choice amounts to choosing a particular fusion
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
186 I. Kriz
(we saw some indication, although not proof, that it may vanish for the eld cor
responding to adding the symmetric term xyztu to the superpotential, and for the
unique critical ac eld). By comparison, the obstruction vanishes for the critical
cc elds and ac elds in the Gepner model of the Fermat quartic K3surface. Our
calculations are not completely physical in the sense that cc elds are not real: real
elds are obtained by adding in each case the complexconjugate aa eld, in which
case the calculation is more complicated and is not done here.
Assuming (as seems likely) that the real eld case exhibits similar behavior as
we found, why are the K3 and 3fold cases dierent, and what does the obstruction
in the 3fold case indicate? In the K3case, our perturbative analysis conforms
with the AspinwallMorrison construction [9] of the big moduli space of K3s, and
corresponding (2, 2) (in fact, (4, 4)) CFTs, and also with the ndings of Nahm
and Wendland [54, 62].
In the 3fold case, however, the straightforward perturbative construction
of the deformed nonlinear model fails. This corresponds to the discussion of
NemeschanskySen [55] of the renormalization of the nonlinear model. They
expand around the 0 curvature tensor, but it seems natural to assume that similar
phenomena would occur if we could expand around the Fermat quintic vacuum.
Then [55] nd that nonRicci at deformations must be added to the Lagrangian
at higher orders of the deformation parameter in order to cancel the function.
Therefore, if we want to do this perturbatively, elds must be present in the orig
inal (unperturbed) model which would correspond to nonRicci at deformation.
No such elds are present in the Gepner model. (Even if we do not a priori assume
that the marginal elds of the Gepner model correspond to Ricci at deforma
tions, we see that dierent elds are needed at higher order of the perturbation
parameter, so there are not enough elds in the model.) More generally, ignoring
for the moment the worldsheet SUSY, the bosonic superpartners are elds which
are of weight 1 classically (as the classical nonlinear model Lagrangian is con
formally invariant even in the nonRicci at case). A 1loop correction arises in
the quantum picture [4], indicating that the corresponding deformation elds must
be of generalized weight (cf. [3942]). However, such elds are excluded in unitary
CFTs, which is the reason why these deformations must be nonperturbative. One
does not see this phenomenon on the level of the corresponding topological models,
since these are invariant under varying the metric within the same cohomological
class, and hence do not see the correction term [68]. Also, it is worth noting that
in the K3case, the function vanishes directly for the Ricciat metric by the
N = (4, 4) supersymmetry ( [5]), and hence the correction terms of [55] are not
needed. Accordingly, we have found that the corresponding perturbative deforma
tions exist.
From the point of view of mirror symmetry, mirrorsymmetric families of hyper
surfaces in toric varieties were proposed by Batyrev [10]. In the case of the Fermat
quintic, the exact mirror is a singular orbifold and the nonlinear model deforma
tions corresponding to the Batyrev dual family exist perturbatively by our analysis.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
188 I. Kriz
To obtain mirror candidates for the additional deformations, one uses crepant res
olutions of the mirror orbifold (see [57] for a survey). In the K3case, this approach
seems validated by the fact that the mirror orbifolds can indeed be viewed as a
limit of nonsingular K3surfaces [6]. In the 3fold case, however, this is not so
clear. The moduli spaces of CalabiYau 3folds are not locally symmetric spaces.
The crepant resolution is not unique even in the more restrictive category of alge
braic varieties; dierent resolutions are merely related by ops. It is therefore not
clear what the exact mirrors are of those deformations of the Fermat quintic where
the deformation does not naturally occur in the Batyrev family, and resolution
of singularities is needed. In other words, the McKay correspondence sees only
topological invariants, and not the ner geometrical information present in the
whole nonlinear model.
In [21], Fan, Jarvis and Ruan constructed exactly mathematically the Amodels
corresponding to LandauGinzburg orbifolds via GromovWitten theory applied
to the Witten equation. Using mirror symmetry conjectures, this may be used to
construct mathematically candidates of topological gravitycoupled Amodels as
well as Bmodels of CalabiYau varieties. GromovWitten theory, however, is a
rich source of examples where such gravitycoupled topological models exist, while
a full conformally invariant (2, 2)model does not. For example, GromovWitten
theory can produce highly nontrivial topological models for 0dimensional orbifolds
(cf. [56, 43]).
Why does our analysis not contradict the calculation of Dixon [19] that the
central charge does not change for deformation of any N = 2 CFT along any linear
combination of ac and cc elds? Zamolodchikov [70,71] dened an invariant c which
is a nondecreasing function in a renormalization group ow in a 2dimensional
QFT, and is equal to the central charge in a conformal eld theory. It may therefore
appear that by [19], all innitesimal deformations along critical ac and cc elds in
an N = 2CFT exponentiate. However, we saw that when our obstruction occurs,
additional counterterms corresponding to those of NemeschanskySen are needed.
This corresponds to nonperturbative corrections of the correlation function needed
to x c, and the functions [19] cannot be used directly in our case.
Finally, let us briey discuss the signicance of our result to the relationship
between classical and quantum geometry. One of the well known eects (and also
great puzzles) of string duality (as reviewed, say, in [31]) is that a smooth path in
the moduli space of conformal eld theories corresponding to CalabiYau varieties
can correspond to a discontinuous path in the classical moduli space of the Calabi
Yau varieties themselves, and more specically that the topology of the underlying
CalabiYau variety can change along such path. In view of our result, it is possible
that this picture needs to be rened. Namely, what we perceive as a smooth path
in quantum geometry may actually consist of discrete steps tunneling across the
changes of topology. An explanation of such phenomenon could be that the moduli
space of quantum geometries should itself be quantized, and can have a discrete
rather than continuous spectrum.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
Acknowledgments
The author thanks D. Burns, I. Dolgachev, I. Frenkel, Doron Gepner, Y. Z. Huang,
I. Melnikov, K. Wendland and E. Witten for explanations and discussions. Spe
cial thanks to H. Xing, who contributed many useful ideas to this project before
changing his eld of interest.
The author is supported by grants from the NSA and the MCTP.
References
[1] I. Aeck and A. W. Ludwig, Universal noninteger ground state degeneracy in critical
quantum systems, Phys. Rev. Lett. 67 (1991) 161164.
[2] I. Aeck and A. W. Ludwig, Exact conformal eld theory results on the multichannel
Kondo eect: Single Fermion Greens function, selfenergy and resistivity, J. High
Energy Phys. 11 (2000) 21.
[3] L. AlvarezGaume, S. Coleman and P. Ginsparg, Finiteness of Ricci at N = 2
supersymmetric models. Comm. Math. Phys. 103(3) (1986) 423430.
[4] L. AlvarezGaume and D. Z. Freedman, K ahler geometry and renormalization of
supersymmetric models, Phys. Rev. D 22 (1980) 846853.
[5] L. AlvarezGaume and P. Ginsparg, Finiteness of Ricci at supersymmetric models,
Comm. Math. Phys. 102 (1985) 311326.
[6] M. T. Anderson, The L2 structure of moduli spaces of Einstein metrics on
4manifolds, Geom. Funct. Anal. 2 (1992) 2989.
[7] D. V. Anosov and A. A. Bolibruch, The RiemannHilbert Problem, Aspects of Math
ematics, E22 (Friedr. Vieweg and Sohn, Braunschweig, 1994).
[8] J. Ashkin and G. Teller, Statistics of twodimensional lattices with four components,
Phys. Rev. 64 (1943) 178184.
[9] P. S. Aspinwall and D. R .Morrison, String theory on K3 surfaces, in Mirror Sym
metry, eds. B. R. Greene and S. T. Yau, Vol. II, AMS/IP Stud. Adv. Math. (Amer.
Math. Soc., 1994), pp. 703716.
[10] V. V. Batyrev, Dual polyhedra and mirror symmetry for CalabiYau hypersurfaces
in toric varieties, J. Alg. Geom. 3 (1994) 493535.
[11] R. J. Baxter, Eightvertex model in lattice statistics, Phys. Rev. Lett. 26 (1971)
832833.
[12] P. Bouwknecht and D. Ridout, A note on the equality of algebraic and geometric
Dbrane charges in WZW, J. High Energy Phys. 0405 (2004) 029.
[13] R. Cohen and D. Gepner, Interacting bosonic models and their solution, Mod. Phys.
Lett. A 6 (1991) 2249.
[14] M. Dine, N. Seiberg, X. G. Wen and E. Witten, Nonperturbative eects on the string
world sheet, Nucl. Phys. B 278 (1986) 769969.
[15] M. Dine, N. Seiberg, X. G. Wen and E. Witten, Nonperturbative eects on the string
world sheet, Nucl. Phys. B 289 (1987) 319363.
[16] L. J. Dixon, V. S. Kaplunovsky and J. Louis, On eective eld theories describing
(2, 2) vacua of the heterotic string, Nucl. Phys. B 329 (1990) 2782.
[17] J. Ellis, C. Gomez, D. V. Nanopoulos and M. Quiros, World sheet instanton eects
on noscale structure, Phys. Lett. B 173 (1986) 5964.
[18] J. Distler and B. Greene, Some exact results on the superpotential from CalabiYau
compactications, Nucl. Phys. B 309 (1988) 295316.
[19] L. Dixon, Some worldsheet properties of superstring compactications, on orbifolds
and otherwise, in Superstrings, Unified Theories and Cosmology, Proc. ICTP Summer
school, 1987, ed. G. Furlan (World Scientic, 1988), pp. 67126.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
190 I. Kriz
[20] M. R. Douglas and W. Taylor, The landscape of intersecting brane models, J. High
Energy Phys. 0701 (2007) 031.
[21] H. Fan, T. J. Jarvis and Y. Ruan, The Witten equation, mirror symmetry and quan
tum singularity theory, arXiv:0712.4021.
[22] S. Fredenhagen and V. Schomerus, Branes on group manifolds, Gluon condensates,
and twisted Ktheory, J. High Energy Phys. 104 (2001) 7.
[23] D. Friedan, Nonlinear models in 2+ dimensions, Ann. Phys. 163(2) (1985) 318419.
[24] D. Gepner, Spacetime supersymmetry in compactied string theory and supercon
formal models, Nucl. Phys. B 296 (1988) 757778.
[25] D. Gepner, Exactly solvable string compactications on manifolds of SU (N ) holon
omy, Phys. Lett. B 199 (1987) 380.
[26] M. Gerstenhaber, On the deformation of rings and algebras I, Ann. of Math. 79
(1964) 59103.
[27] M. Gerstenhaber, On the deformation of rings and algebras II, Ann. of Math. 84
(1966) 119.
[28] M. Gerstenhaber, On the deformation of rings and algebras III, Ann. of Math. 88
(1968) 134.
[29] M. Gerstenhaber, On the deformation of rings and algebras IV, Ann. of Math. 99
(1974) 257276.
[30] P. Ginsparg, Curiosities at c = 1, Nucl. Phys. B 295 (1988) 153170.
[31] B. R. Greene, String theory on CalabiYau manifolds, hepth/9702155.
[32] B. R. Greene and M. R. Plesser, Duality in CalabiYau moduli space, Nucl. Phys. B
338 (1990) 1537.
[33] B. R. Greene, C. Vafa and N. P. Warner, CalabiYau manifolds and renormalization
group ows, Nucl. Phys. B 324 (1989) 371390.
[34] P. A. Grin and O. F. Hernandez, Structure of irreducible SU (2) parafermion mod
ules derived vie the FeiginFuchs construction, Int. J. Modern Phys. A 7 (1992)
12331265.
[35] M. T. Grisaru, A. E. M. Van Den and D. Zanon, Fourloop function for the N = 1
and N = 2 supersymmetric nonlinear sigma model in two dimensions, Phys. Lett. B
173 (1986) 423.
[36] P. Hu and I. Kriz, Conformal eld theory and elliptic cohomology, Adv. Math. 189
(2004) 325412.
[37] P. Hu and I. Kriz, Closed and open conformal eld theories and their anomalies,
Comm. Math. Phys. 254 (2005) 221253.
[38] P. Hu and I. Kriz, A mathematical formalism for the Kondo eect in WZW branes,
J. Math. Phys. 48 (2007) 072301, 31 pp.
[39] Y. Z. Huang, J. Lepowsky and L. Zhang, Logarithmic tensor product theory for
generalized modules for a conformal vertex algebra, Part I, math/0609833.
[40] Y. Z. Huang, J. Lepowsky and L. Zhang, A logarithmic generalization of tensor
product theory for modules for a vertex operator algebra, Internat. J. Math. 17
(2006) 9751012; math/0311235.
[41] Y. Z. Huang and L. Kong, Full eld algebras, QA/0511328.
[42] Y. Z. Huang and A. Milas, Intertwining operator superalgebras and vertex tensor
categories for superconformal algebras, II, Trans. Amer. Math. Soc. 354 (2002) 363
385.
[43] P. Johnson, Equivariant GromovWitten theory of one dimensional stacks, Ph.D.
thesis, Univ. of Michigan (2009).
[44] S. Kachru and E. Witten, Computing the complete massless spectrum of a Landau
Ginzburg orbifold, Nucl. Phys. B 407 (1993) 637666.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X10003916
192 I. Kriz
We consider the fundamental solution for the Cauchy problem for perturbations of the
harmonic oscillator by time dependent potentials which grow at spatial infinity slower
than quadratic but faster than linear functions and whose Hessian matrices have a fixed
sign. We prove that the fundamental solution at resonant times grows indefinitely at
spatial infinity with an algebraic growth rate, which increases indefinitely when the
growth rate of perturbations at infinity decreases from the near quadratic to the near
linear ones.
1. Introduction
We consider ddimensional time dependent Schrodinger equations
u 1
i = + V (t, x) u(t, x), (t, x) R1 Rd . (1)
t 2
We assume throughout this paper that V (t, x) is smooth with respect to the x vari
ables, and V (t, x) and its derivatives x V (t, x) are jointly continuous with respect
to (t, x). Under the conditions to be imposed on V (t, x) in what follows Eq. (1)
generates a unique unitary propagator {U (t, s) : t, s R} in the Hilbert space
H = L2 (Rd ), so that the solution in H of (1) with the initial condition
u(s, x) = (x) H
193
March 10, 2010 10:14 WSPC/S0129055X 148RMP
J070S0129055X10003928
is uniquely given by u(t) = U (t, s). The distribution kernel E(t, s, x, y) of U (t, s)
is called the fundamental solution (FDS for short) of Eq. (1):
U (t, s)(x) = E(t, s, x, y)(y)dy.
We write E(t, x, y) = E(t, 0, x, y). It is well known that the FDS of the free
Schr
odinger equation, viz. Eq. (1) with V = 0, is given by
e 4
id
2
E0 (t, x, y) = ei(xy) /2t , t0 (2)
2td/2
and that of the harmonic oscillator, viz. Eq. (1) with V (t, x) = x2 /2, is given for
nonresonant times m < t < (m + 1), m Z via Mehlers formula:
eid(1+2m)/4 (i/sin t)((x2 +y2 )cos t/2xy)
Eh (t, x, y) = e , (3)
2 sin td/2
and, for resonant times t s = m by
Note that the FDS for the free Schr odinger equation is smooth and spatially
bounded for any t = 0; for the harmonic oscillator the FDS has this property
only at nonresonant times; at resonant times t = m singularities of the initial
function Eh (0, x, y) = (x y) recur at x = (1)m y, however, it is smooth and
decays rapidly at spatial innity. Actually, it vanishes outside the singular point
x = (1)m y when t = m.
We begin with a brief review on properties of the FDS for (1) with general
potentials V (t, x) laying emphasis on its smoothness and boundedness with respect
to the spatial variables (x, y). We denote the classical Hamiltonian and Lagrangian
corresponding to (1), respectively, by
and (x(t, s, y, k), p(t, s, y, k)) is the solution of the initial value problem for
Hamiltons equations
x(t)
= p H(t, x, p), = x H(t, x, p);
p(t) x(s) = y, p(s) = k. (5)
Then, in the seminal work [4], Fujiwara has shown that there exists a T depending
only on V such that the following results hold for the time interval 0 (ts) < T :
The map Rd k x(t, s, y, k) Rd is a dieomorphism for every xed y Rd
March 10, 2010 10:14 WSPC/S0129055X 148RMP
J070S0129055X10003928
and, therefore, there exists a unique path of (5) such that x(s) = y and x(t) = x;
if we write
t
S(t, s, x, y) = 2 /2 V (r, x(r)))dr
(x(r)
s
for the action integral of the path, the FDS E(t, s, x, y) has the form
e 4
id
where k is the (unique) point such that x = x(t, s, y, k). In particular, E(t, s, x, y)
is smooth and bounded with respect to the spatial variables (x, y) Rd Rd for
every 0 < t s < T (see [9] for a generalization to the case when magnetic elds
are present). For the free Schrodinger equation or for the harmonic oscillator the
relation (9) holds without the error term O(t s(d2)/2 ).
Under the condition (6) the structure (7) of the FDS in general breaks down
at later times because singularities of the initial data (x) may recur in nite time
as the FDS of the harmonic oscillator (4) explicitly demonstrates. If V (t, x) is
subquadratic at spatial innity in the sense that
lim sup x V (t, x) = 0,  = 2,
x t
(10)
x V (t, x) C , for all  3,
then this recurrence of singularities does not take place, however, and the FDS is
of the form (7) for any nite time ([12]). More precisely, if V satises (10), then for
any T > 0, there exists R > 0 such that, for any t and s with 0 < (t s) T
and for any pair (x, y) Rd Rd with x2 + y 2 R2 , there exists a unique path
of (5) such that x(s) = y and x(t) = x and the FDS for 0 < (t s) T may
be written in the form (7), where, for (x, y) with x2 + y 2 R2 , S(t, s, x, y) is the
action integral of this path. Moreover, we have a(t, s, x, y) 1 as x2 + y 2 . In
particular, E(t, s, x, y) is smooth and bounded with respect to (x, y) for any t = s.
On the other hand, if d = 1 and V (t, x) does not depend on t, and if V is convex
and V (x) Cx2+ for large x for some > 0 and C > 0, then, under certain
additional techinical assumption on the derivatives, E(t, x, y) is nowhere C 1 with
respect to (t, x, y) ([10]). It is also known that, if V satises C1 x V (x) C2 x
near innity with constants > 10 and 0 < C1 C2 < , then E(t, 0, x, y) is
March 10, 2010 10:14 WSPC/S0129055X 148RMP
J070S0129055X10003928
unbounded with respect to (x, y) for any t R ([13]). These results have been
proven only in one dimension so far, however, it is believed that similar results hold
in all dimensions.
In this way, properties of the FDS experience a sharp transition when the
growth rate at spatial innity of the potential V (t, x) changes from subquadratic
to superquadratic. Thus, the FDS for the borderline case, viz. perturbations of the
harmonic oscillator
u 1 1
i = + x2 + W (t, x) u(t, x), (t, x) R1 Rd , (11)
t 2 2
where W (t, x) is subquadratic in the sense it satises (10) with W in place of V ,
has attracted particular interest of many authors, and the following properties of
E(t, s, x, y) have been established (see, e.g., [14, 5, 12, 2, 3]). We may set s = 0,
which we will do, and we will write E(t, x, y) for E(t, 0, x, y); x
= (1 + x2 )1/2 .
(a) The structure of the FDS Eh (t, x, y) at nonresonant times as stated in (3) is
stable under perturbations and E(t, x, y) is smooth and spatially bounded for
m < t < (m + 1).
E(m, x, y) CN x y
N , x y 1. (12)
E(m, x, y) CN x y
N , x y 1. (13)
(d) If W is superlinear and satises the following sign condition on the Hessian
matrix x2 W = ( 2 W/xj xk ) that
C1 x
x2 W (t, x) C2 x
, (t, x) R1 Rd (14)
for some constants 0 < < 1 and 0 < C1 < C2 < or < C1 < C2 < 0,
then E(m, x, y), m Z, is C with respect to (x, y), viz. singularities at
resonant times t = m are swept away.
March 10, 2010 10:14 WSPC/S0129055X 148RMP
J070S0129055X10003928
This paper is concerned with the properties of the FDS E(t, x, y), when t is
at resonant times t Z. We show that, in the last case (d) above, E(m, x, y)
increases indenitely as x at the algebraic rate Cxd/(22) , exhibiting
a sharp contrast to the decay result (12) or (13) for the case when W is at
most linearly increasing at spatial innity. More precisely we prove the following
theorem:
Theorem 1.1. Suppose that W (t, x) is subquadratic and satisfies the sign condition
(14) for some 0 < < 1 and 0 < C1 < C2 < or < C1 < C2 < 0. Let m Z
and y Rd be fixed. Let C0 (Rd \{0}) be such that (x) = 1 for a x b,
0 < a < b < being constants. Then there exist constants 0 < M1 < M2 ,
independent of R 1, such that
2 1/2
2 x dx
M1 R d/(22)
E(m, x, y) M2 Rd/(22) . (15)
Rd R Rd
1/2
x
E(m, x, y) det kd/2 xd/(22) , x ,
k
which is consistent with (15). Here is another remark, which claries that
Theorem 1.1 is more or less consistent with the known results. We should note that
2
if = 0, then W = c x
, and m is no longer a resonant time for V = x2 /2 + W ,
and the corresponding E(m, x, y) is bounded as x y ; on the other hand,
if = 1, then W = c x
and, as in (c) above, a large portion of E(m, x, y) is
concentrated in a bounded domain x y 2cm, which may be represented as the
extreme case of C x
d/(22) as 1.
We mention here that the result of the theorem has been conjecture by Martinez
and the second author in [7], where a similar problem is studied in the semiclassical
March 10, 2010 10:14 WSPC/S0129055X 148RMP
J070S0129055X10003928
setting. More precisely, they consider the FDS of the semiclassical Schrodinger
equation
2
u h 1
ih = + x2 + h W (x) u,
t 2 2
where W (x) is t independent and satises the same conditions as in this paper,
(10) and (14); and they prove that the FDS at the resonant times may be written
in the form
E(m, x, y) = hd(1+)/2 a(x, y, h)eiS(x,y)/h , = /(1 ), (16)
where S(x, y) is the action integral of the path of (5) connecting x(0) = y and
x(m) = x and a(x, y, h) satises C 1 a(x, y, h) C uniformly with respect h
on every compact subset K of R2d \{(x, (1)m x) : x Rd }. Thus, E(m, x, y) has
the extra growing factor hd/2 as h 0 compared to E(t, x, y) at nonresonant
times t = m and they remark that, if their arguments applied for nonsmooth
potentials, (16) would imply the estimate (15) of Theorem 1.1 for the homogeneous
potential W (x) = Cx2 .
It is well known that the boundedness of E(t, s, x, y) with respect to (x, y)
implies the so called Lp Lq estimates of the propagator U (t, s) (hence, also nite
time Strichartz estimates). There are examples of Schr odinger equations with
smooth coecients, which exhibit break down of the estimates, e.g., the harmonic
oscillator at resonant times. However, to the best knowledge of the authors, in all
known examples they are broken because of local singularities and, Theorem 1.1 is
the rst example in which they are broken because of the growth at spatial inn
ity of the FDS (see [8] for Lp Lq estimates for potentials which are singular but
decay at innity). For the microlocal smoothing estimate which may be applied
for proving the smoothness of the FDS, see for example [1] or [6].
The rest of the paper is devoted to the proof of this theorem. We prove it only
in the m = 1 case. The proof for the other cases is similar. In Sec. 2, we recall
several known facts, which will be used in Sec. 3, where the theorem is proved. We
often omit some of the variables of functions, if no confusion is to be feared. For
functions f of several variables, we write f C k (x) or f C k (t, x) etc., if f is of
class C k with respect to x or (t, x), etc.
2. Preliminaries
We rst recall some results on the Hamiltonian ow generated by (5) when V (t, x) =
x2 /2 + W (t, x) and W is subquadratic. We set the initial time s = 0 and omit the
variable s. The solutions (x(t), p(t)) = (x(t, y, k), p(t, y, k)) of (5) satisfy the integral
equations
t
x(t) = y cos t + k sin t sin(t s)x W (s, x(s))ds, (17)
0
t
p(t) = y sin t + k cos t cos(t s)x W (s, x(s))ds. (18)
0
March 10, 2010 10:14 WSPC/S0129055X 148RMP
J070S0129055X10003928
p(t, y, k) = . (24)
(t,
, y) = (t, , y) for t I, 2 + y 2 R2 .
We call integrals of the form (26) oscillatory integrals and often write them
simply as
i(m+1)d eixi(t,,y)
a(t, , y)
d.
Rd (2) cos t
d d/2
When W satises the sign condition (14), the phase function (, , y) satises
the following properties which are essential for the proof of the theorem. From now
on we let m = 1.
Proposition 2.2. Let W be subquadratic and satisfy (14). Let L > 0. Then, there
exist constants C > 0 and R > 0 depending only on L such that for every  R
and y L:
Proof. The upper bound in estimate (27) is obvious from (25), (17) and (20);
the lower bound is proved in [11, pp. 6163] for time independent perturbations
W (t, x) = W (x), and the proof applies to the time dependent case as well, if we
use [12, Lemmas 2.1 and 2.2] instead of [11, Lemmas 4.2 and 4.3]. From [11, pp. 61
63], we also have for  R and k such that p(, y, k) =
k x(, y, k)  . (29)
and, applying the second result of (23) and (29), we obtain (28) for the case  = 2.
For higher derivatives, we further dierentiate (30) and apply (22) and (23) in
addition to (29). Estimate (28) follows inductively.
Lemma 2.3. Let L > 0 and 0 < a < b < be fixed arbitrarily and let C0 (Rd )
be supported by {x Rd : a x b}. Then, there exist R0 > 0 and C0 > 0, such
that for all R > R0 and y L
1
( (, , y)/R)2 d C0 Rd/(1) . (31)
R d Rd
If (x) > > 0 for a1 < x < b1 , a < a1 < b1 < b, then we also have the lower
bound:
1
C1 R d/(1)
d ( (, , y)/R)2 d. (32)
R Rd
Proof. For suciently large R > 0, we have by virtue of (21) that 1/2
p(, y, k)/k 2 for y L and k R, and (27) implies
D1 R1/(1)  D2 R1/(1)
and
1
( (, , y)/R)2 d CRd/(1) .
Rd
A similar argument yields the lower bound in the second case. We omit the obvious
details.
then the standard stationary phase method combined with a change of scale would
yield the pointwise estimate
However, (33) does not seem to hold in general and this required a weaker formu
lation of the theorem and a little complicated proof given below.
March 10, 2010 10:14 WSPC/S0129055X 148RMP
J070S0129055X10003928
We need to estimate
2
1
I(R) x E(, x, y) dx. (35)
Rd R
R d
In what follows, we omit the variable , the domain of integration Rd from integral
signs and write
as . Since y is xed in the following computation, we sometimes
omit the variables y as well. This should not cause any confusion. Then, by virtue
of (26), (35) may be written as an oscillatory integral
2
1
I(R) = lim x e i(x()) 2 /2
a()d dx
0 (2)2d Rd R
2
1 x 2 2
= lim 2d d
eix()+i(()())( + )/2 a()a()dddx
0 (2) R R
1 2 2
= lim 2 (R( ))ei(()())( + )/2 a()a()dd,
(36)
0 (2)d
In what follows we omit the limit sign lim0 and the damping factors which arise
from exp(( 2 + 2 )/2). In the righthand side of (36), we change variables to
= and expand by Taylors formula as
1 1
a( + ) = a() () + b (, )
! !
N =N +1
We take N such that (1 ) > d and apply integration by parts times to the
inner integral, which we denote by I(, R), by using the identity
1 i ( + )
ei(+) = ei(+) .
1 + ( ( + ))2
Thus, if we write M for the transpose of the dierential operator on the left, we
have
I(, R) = ei(+) M (
2 (R) b (, ))d. (38)
C 1 +
2(1) 1 + ( ( + ))2 C +
2(1)
I(, R) C +
(1) R (
2 )(R)  b (, )d.
++
(1)
(1) R
L R N +1 .
It follows, by changing variables to /R, and by taking L large enough, that for
R>1
(1)
I(, R) CRN 1d+
/R
(1)
N +1L d
(1)
C RN 1d+
.
Thus, for such that (1 ) > d we may estimate the remainder BN (R) in (37) by
C (1) C
BN (R) a()
d ,
RN +d+1 RN +d+1
March 10, 2010 10:14 WSPC/S0129055X 148RMP
J070S0129055X10003928
and we may ignore BN (R) by taking N large enough. We have next to deal with
the rst terms in (37), which are sum over  N of
1
A =
2 (R) i((+)())
e d a()a() ()d. (39)
(2)d !
By using Taylors formula, we write
ei((+)()) = ei () ei(,) ,
1
2
(, ) = (1 ) 2 ( + )d ,
0
where we take N large enough so that (N + 1) > d. We then insert this into the
righthand side of (39). Note that
(, ) C
2
CR(d++2N +2) .
Thus, we may again ignore this term and we are left for A with
N
1 1
e i ()
2 (R) (i(, )) d a()a() ()dd.
m
(2)d ! m=0 m!
Here we repeat the same argument as in the rst step to the inner integral. We
expand (, ) further by Taylors formula:
()
(, ) = () + LN (, ),
!
2N
1
LN (, ) = C (1 )N () ( + )d
=N +1 0
Acknowledgements
The rst author was partially supported by the Danish Natural Science Research
Council grant Mathematical Physics. The second author was supported by JSPS
grant in aid for scientic research No. 18340041. This work has been done while
the second author was visiting Department of Mathematical Sciences of Aalborg
University. He acknowledges the hospitality of the department.
References
[1] W. Craig, T. Kappeler and W. Strauss, Microlocal dispersive smoothing for the
odinger equation, Comm. Pure Appl. Math. 48 (1995) 769860.
Schr
[2] S. Doi, Dispersion of singularities of solutions for Schrodinger equations, Comm.
Math. Phys. 250 (2004) 473505.
[3] S. Doi, Smoothness of solutions for Schr
odinger equations with unbounded potentials,
Publ. RIMS Kyoto Univ. 41 (2005) 175221.
March 10, 2010 10:14 WSPC/S0129055X 148RMP
J070S0129055X10003928
[4] D. Fujiwara, Remarks on the convergence of the Feynman path integrals, Duke Math.
J. 47 (1980) 4196.
[5] L. Kapitanski, I. Rodnianski and K. Yajima, On the fundamental solution of a per
turbed harmonic oscillator, Topol. Methods Nonlinear Anal. 9 (1997) 77106.
[6] A. Martinez, S. Nakamura and V. Sordoni, Analytic smoothing eect for the
odinger equation with longrange perturbation, Comm. Pure Appl. Math. 59(9)
Schr
(2006) 13301351.
[7] A. Martinez and K. Yajima, On the fundamental solution of semiclassical Schrodinger
equations at resonant times, Comm. Math. Phys. 216 (2001) 357373.
odinger operators: A survey, in Mathematical
[8] W. Schlag, Dispersive estimates for Schr
Aspects of Nonlinear Dispersive Equations, Ann. of Math. Stud., Vol. 163 (Princeton
Univ. Press, Princeton, NJ, 2007), pp. 255285.
[9] K. Yajima, Schrodinger evolution equations with magnetic elds, J. dAnalyse Math.
56 (1991) 2976.
[10] K. Yajima, Smoothness and nonsmoothness of the fundamental solution of time
dependent Schr odinger equations, Comm. Math. Phys. 181 (1996) 605629.
[11] K. Yajima, On fundamental solution of time dependent Schr odinger equations, Con
temp. Math. 217 (1998) 4968.
[12] K. Yajima, On the behavior at innity of the fundamental solution of time dependent
odinger equation, Rev. Math. Phys. 13 (2001) 891920.
Schr
[13] G. P. Zhang and K. Yajima, Smoothing property for Schr odinger equations with
potential superquadratic at innity, Comm. Math. Phys. 221 (2001) 573590.
[14] S. Zelditch, Reconstruction of singularities for solutions of Schrodinger equation,
Comm. Math. Phys. 90 (1983) 126.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X1000393X
1. Introduction
The dynamics of a nite quantum system, i.e. one with a nite number of degrees
of freedom described by a Hilbert space H, is given by the Schr
odinger equation.
c 2010 by the authors. This paper may be reproduced, in its entirety, for noncommercial
purposes.
207
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X1000393X
where, for each t R, A(t) B(H) is a bounded linear operator on H. Its solutions
are given by a oneparameter group of automorphisms, t , of B(H):
A(t) = t (A(0)).
classical systems. Two classic papers are [10, 15]. Many properties of this classical
innite volume harmonic dynamics have been studied in detail, e.g., [23,9] and some
recent progress on locality estimates for anharmonic systems is reported in [5, 20].
The paper is organized as follows. We begin with a section discussing bounded
interactions. In this case, the existence of the dynamics follows by mimicking the
proof valid in the context of quantum spins systems. Section 3 describes the innite
volume harmonic dynamics on general graphs. It is motivated by an explicit example
on Zd . Next, in Sec. 4, we discuss nite volume perturbations of the innite volume
harmonic dynamics and prove that such systems satisfy a LiebRobinson bound. In
Sec. 5, we demonstrate that the existence of the dynamics and its continuity follow
from the LiebRobinson estimates established in the previous section.
2. Bounded Interactions
The goal of this section is to prove the existence of the dynamics for oscillator sys
tems with bounded interactions. Since oscillator systems with bounded interactions
can be treated as a special case of more general models with bounded interactions,
we will use a slightly more general setup in this section, which we now introduce.
We will denote by the underlying structure on which our models will be
dened. Here will be an arbitrary set of sites equipped with a metric d. For
with countably innite cardinality, we will need to assume that there exists a
nonincreasing function F : [0, ) (0, ) for which:
and
(ii) F satises
F (d(x, z))F (d(z, y))
C := sup < . (2.2)
x,y F (d(x, y))
z
Given such a set and a function F , by the triangle inequality, for any a 0
the function
where the union is over all nite subsets ; see [3, 4] for a discussion of these
issues in general.
The result discussed in this section corresponds to bounded perturbations of
local selfadjoint Hamiltonians. We x a collection of onsite local operators H loc =
{Hx }x where each Hx is a selfadjoint operator over Hx . In addition, we will
consider a general class of bounded perturbations. These are dened in terms of
an interaction , which is a map from the set of subsets of to A with the
property that for each nite set X , (X) AX and (X) = (X). As with
the LiebRobinson bound proven in [19], we will need a growth condition on the
set of interactions for which we can prove the existence of the dynamics in the
thermodynamic limit. This condition is expressed in terms of the following norm.
For any a 0, denote by Ba () the set of interactions for which
1
a := sup (X) < . (2.3)
x,y Fa (d(x, y))
Xx,y
Theorem 2.1. Under the conditions stated above, for all t R, A A , the norm
limit
Fix T > 0 and X nite. For any A AX , we will show that for any
n
nondecreasing, exhausting sequence {n } of , the sequence {t,int (A)} is Cauchy
in norm, uniformly for t [T, T ]. Moreover, the bounds establishing the Cauchy
property depend on A only through X and A. Since
loc loc P P
t (A) = t,int
(eitH AeitH ) = t,int
(eit xX Hx
Aeit xX Hx
),
an analogous statement then immediately follows for {tn (A)}, since they are all
also localized in X and have the same norm as A.
Take n m with X n m and calculate
t
m n d
t,int (A) t,int (A) = {Um (0, s)Un (s, t)AUn (t, s)Um (s, 0)} ds. (2.9)
0 ds
A short calculation shows that
d
U (0, s)Un (s, t)AUn (t, s)Um (s, 0)
ds m
= iUm (0, s)[(Hint
m
(s) Hint
n
(s)), Un (s, t)AUn (t, s)]Um (s, 0)
loc loc
n
= iUm (0, s)eisHn [B(s), st (A(t))]eisHn Um (s, 0), (2.10)
where
= eitHlocn AeitHlocn = eitHX
A(t)
loc loc
AeitHX (2.11)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X1000393X
and
loc loc
B(s) = eisHn (Hint
m
(s) Hint
n
(s))eisHn
loc loc
= eisHm \n (Z)eisHm \n (Z)
Zm Zn
loc loc
isH m \n
isH m \n
= e (Z)e . (2.12)
Zm :
Zm \n =
2A 2 a Ca ts
(e 1) (Z) Fa (d(x, z))
Ca
ym \n Zm : xX zZ
yZ
2A 2 a Ca ts
(e 1) (Z) Fa (d(x, z))
Ca
ym \n zm Zm : xX
y,zZ
2Aa 2 a Ca ts
(e 1) Fa (d(x, z))Fa (d(z, y))
Ca
ym \n xX zm
2Aa(e2 a Ca ts 1) Fa (d(x, y)). (2.14)
ym \n xX
With the estimate above and the properties of the function Fa , it is clear that
m n
sup t,int (A) t,int (A) 0 as n, m , (2.15)
t[T,T ]
and the rate of convergence only depends on the norm A and the set X where A
is supported. This proves the claim.
can be regarded as a state of the innite system dened on the norm comple
tion of A . The GNS Hilbert space H of can be constructed as the closure of
A x x . Let A x x . Then
(n ) (n ) (n )
(t (A) t0 (A)) (t (A) t (A)) + (t (A) t0 (A))
(n )
+ (t0 (A) t0 (A)). (2.17)
For suciently large n , the limtt0 of the middle term vanishes by Stones the
orem. The two other terms are handled by (2.5). It is clear how to extend the
continuity to H .
We will discuss this type of situation in more detail in the next three sections
where we consider models that include quadratic (unbounded) interactions as well.
(f, g) = Im[
f, g]. (3.2)
and
t (W (f )) = W (Tt f ), f D (3.5)
For the von Neumann algebra M, take the weakclosure of (W) in L(H ) and
let t be the weakly continuous, one parameter group of automorphisms of M
obtained by lifting t to M. The latter step is possible since is t invariant; see,
e.g., [3, Corollary 2.3.17].
holds for all t R and all f, g 2 (Zd ). In general, we will assume that the harmonic
dynamics satises an estimate of this type. Namely, we suppose that there exists a
number a0 > 0 for which given 0 < a a0 , there are numbers ca and va for which
1 ei(Tt f,g)  ca eva t f (x)g(y)Fa (d(x, y)) (3.10)
x,y
holds for all t R and all f, g 2 (). Here we describe the spatial decay in
through the functions Fa as introduced in Sec. 2. Since the Weyl operators are
unitary, the norm estimate
[t (W (f )), W (g)] ca eva t f (x)g(y)Fa (d(x, y)), (3.11)
x,y
readily follows.
Here the quantities px and qx , which appear in (3.12) above, are the single site
momentum and position operators regarded as operators on the full Hilbert space
HL by setting
d
px = 1l 1l i 1l 1l and qx = 1l 1l q 1l 1l,
dq
(3.14)
i.e. these operators act nontrivially only in the xth factor of HL . These operators
satisfy the canonical commutation relations
valid for all x, y L . In addition, {ej }dj=1 are the canonical basis vectors in
Zd , the numbers j 0 and 0 are the parameters of the system, and the
Hamiltonian is assumed to have periodic boundary conditions, in the sense that
qx+ej = qx(2L1)ej if x L but x + ej L . It is wellknown that Hamiltonians
of this form can be diagonalized in Fourier space. We review this quickly to establish
some notation and refer the interested reader to [19] for more details.
Introducing the operators
1 1
Qk = eikx qx and Pk = eikx px , (3.16)
L  xL L  xL
d
(k) = 2 + 4 j sin2 (kj /2), (3.17)
j=1
Using the above diagonalization, one can determine the action of the dynamics
corresponding to HLh on the Weyl algebra W(2 (L )). In fact, by setting
W (f ) = exp i (Re[f (x)]qx + Im[f (x)]px ) , (3.20)
xL
for each f 2 (L ), it is easy to verify that (3.3) and (3.4) hold with (f, g) =
Im[
f, g]. It is convenient to express these Weyl operators in terms of annihilation
and creation operators, i.e.
1 1
ax = (qx + ipx ) and ax = (qx ipx ), (3.21)
2 2
which satisfy
with (k) as in (3.17). Using the fact that is real valued and even, it is easy to
check that
U U V V = 1l = U U V V (3.32)
and
V U U V = 0 = V U U V (3.33)
where we stress that V is the adjoint of the antilinear mapping V . The relation
(3.29) is invertible, in fact,
and therefore
i
W (f ) = exp (b((U V )f ) + b ((U V )f )) . (3.35)
2
Clearly then,
t (W (f )) = W (Tt f ), (3.36)
Tt = (U + V )F 1 Mt F (U V ), (3.37)
i.e. Tt is sympletic in the sense of (3.6). Using [4, Theorem 5.2.8], there is a unique
oneparameter group of automorphisms on W(D), which we will denote by t ,
that satises
(1) 2 (/2)+1
(xc, max( ,e )t)
Ht (x) c1
, e
(3.43)
(1) 2 (/2)+1
Ht (x) c, e/2 e(xc, max( ,e )t)
d
hold for all t R and x Zd . Here x = j=1 xi .
Given the estimates in Lemma 3.1, Eq. (3.41) and Youngs inequality imply that
Tt can be dened as a transformation of p (Zd ), for p 1. However, the symplectic
form limits us to consider D = p (Zd ) with 1 p 2.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X1000393X
This implies an estimate of the form (3.9), and hence a LiebRobinson bound as
in (3.11).
A simple corollary of Lemma 3.1 follows.
(0)
Ht 0 1 0 as t 0, (3.45)
and
(m)
Ht 1 0 as t 0, for m {1, 1}. (3.46)
(m)
Proof. The estimates in Lemma 3.1 imply that the functions Ht are bounded
by exponentially decaying functions (in x). These estimates are uniform for t in
compact sets, e.g., t [1, 1], and therefore dominated convergence applies. It is
(0) (m)
clear that H0 (x) = 0 (x) while H0 (x) = 0 for m {1, 1}. This proves the
corollary.
Remark 3.3. We observe that, when = 0, the nite volume Hamiltonian HLh
(3.12) is translation invariant and commutes with the total momentum operator P0
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X1000393X
where we used the notation (3.16) and, for k = 0, we introduced the operators bk , bk
as in (3.19). In this case, the operator HLh does not have eigenvectors: its spectrum is
purely continuous. By a unitary transformation, the Hilbert space HL (see (3.13))
can be mapped into the space L2 (R, dP0 ; Hb ) of square integrable functions of
P0 R, with values in Hb . Here, Hb denotes the Fock space generated by all creation
and annihilation operators bk , bk with k = 0. It is then easy to construct vectors
which minimize the energy by a given distribution of the total momentum: for an
arbitrary (complex valued) f L2 (R) with f = 1, we dene f L2 (R, dP0 ; Hb )
by setting f (P0 ) = f (P0 ) (where is the Fock vacuum in Hb ). These vectors
are not invariant with respect to the time evolution. It is simple to check that the
2
odinger evolution of f is given by eiHL t f = ft with ft (P0 ) = eitP0 f (P0 )
h
Schr
is the free evolution of f . In particular, for = 0, HLh does not have a ground state
in the traditional sense of an eigenvector. For this reason, when = 0, it is not
a priori clear what the natural choice of state should be. As is discussed above, one
possibility is to consider rst = 0 and then take the limit 0. This yields a
ground state for the innite system with vanishing center of mass momentum of
the oscillators. By considering nonzero values for the center of mass momentum,
one can also dene other states with similar properties.
to the generator. This is a version of what is usually known as the Dyson or Duhamel
expansion. The following statement summarizes [4, Proposition 5.4.1].
Proposition 4.1. Let {M, t } be a W dynamical system and let denote the
infinitesimal generator of t . Given any P = P M, set P to be the bounded
derivation with domain D(P ) = M satisfying P (A) = i[P, A] for all A M. It
follows that + P generates a oneparameter group of automorphisms P of M
which is the unique solution of the integral equation
t
P
t (A) = t (A) + i P
s ([P, ts (A)]) ds. (4.1)
0
In addition, the estimate
t P
P
t (A) t (A) (e 1)A (4.2)
holds for all t R and A M.
Since the initial dynamics t is assumed weakly continuous, the norm estimate
(4.2) can be used to show that the perturbed dynamics is also weakly continuous.
Hence, for each P = P M the pair {M, P
t } is also a W dynamical system.
P1 +P2
Thus, if Pi = Pi M for i = 1, 2, then one can dene t iteratively.
Our rst estimate involves perturbations dened as nite sums of onsite terms.
More specically, the perturbations we consider are dened as follows.
To each site x , we will associate a nite measure x on C, and an element
Px W(D) which has the form
Px = W (zx )x (dz). (4.6)
C
We use Proposition 4.1 to dene the perturbed dynamics. Fix a nite set .
Set
P = Px , (4.8)
x
()
and note that (P ) = P W(D). We will denote by t the dynamics that
results from applying Proposition 4.1 to the W dynamical system {M, t0 } and P .
Before we begin the proof of our estimate, we discuss two examples.
Example 1. Let x be supported on [, ) and absolutely continuous with respect
to Lebesgue measure, i.e. x (dz) = vx (z) dz. If vx is in L2 ([, )), then Px is
proportional to an operator of multiplication by the inverse Fourier transform of
vx . Moreover, since the support of x is real, Px corresponds to multiplication by
a function depending only on qx .
Example 2. Let x have nite support, e.g., take supp(x ) = {z, z} for some
number z = + i C. Then
Px = W (zx ) + W (zx ) = 2 cos(qx + px ). (4.9)
Proof. Fix t > 0 and dene the function t : [0, t] W(D) by setting
It is clear that t interpolates between the commutator associated with the original
()
harmonic dynamics, t0 at s = 0, and that of the perturbed dynamics, t at s = t.
A calculation shows that
d
t (s) = i [s() ([Px , W (Tts f )]), W (g)], (4.13)
ds
x
where
Lts;x (f ) = Lts;x (f ) = W (zx ){ei(Tts f,zx ) 1}x (dz) W(D). (4.15)
C
Thus t satises
d
t (s) = i t (s)s() (Lts;x (f ))
ds
x
+i s() (W (Tts f ))[s() (Lts;x (f )), W (g)]. (4.16)
x
The rst term above is norm preserving. In fact, dene a unitary evolution Ut ()
by setting
d
Ut (s) = i s() (Lts;x (f ))Ut (s) with Ut (0) = 1l. (4.17)
ds
x
and therefore,
t
t (t)Ut (t) = t (0) + i s() (W (Tts f ))[s() (Lts;x (f )), W (g)]Ut (s) ds.
x 0
(4.19)
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X1000393X
t
+ ca f (x ) Fa (d(x, x )) eva (ts)
x x 0
z[s() (W (zx )), W (g)]x (dz) ds. (4.22)
C
Following the iteration scheme applied in [19], one arrives at (4.11) as claimed.
The proof of this result closely follows that of Theorem 4.2, and so we only
comment on the dierences.
holds similarly. With (4.28), it is easy to see that the integrand in (4.29) is
bounded by
ca eva (ts) f (x) Fa (d(x, x )) zx [s() (W (z X )), W (g)]X (dz),
x x X CX
(4.30)
The estimate claimed in (4.27) follows by iteration. In fact, the rst term in the
iteration is bounded by
ca f (x) Fa (d(x, x1 ))
x X x1 X
t
va (ts)
e zx1  ca e va s
zx2 g(y)Fa (d(x2 , y))
0 CX x2 X y
X (dz) ds
ca t ca eva t f (x)g(y) Fa (d(x, x1 ))Fa (d(x2 , y))
x,y x1 ,x2
zx1 zx2 X (dz)
X: CX
x1 ,x2 X
a ca t ca eva t f (x)g(y) Fa (d(x, x1 ))Fa (d(x1 , x2 ))Fa (d(x2 , y))
x,y x1 ,x2
a Ca2 ca t ca eva t f (x)g(y)Fa (d(x, y)). (4.32)
x,y
It is important to note that since the estimates in Theorem 4.3 are independent
of , the limiting dynamics also satises a LiebRobinson bound as in (4.27). We
now prove Theorem 5.1.
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X1000393X
Proof. Fix a Weyl operator W (f ) with f 1 (). Let T > 0 and take m n.
Iteratively applying Proposition 4.1, we have that
t
( ) ( ) ( )
t n (W (f )) = t m (W (f )) + i s(n ) ([P n \m , tsm (W (f ))]) ds, (5.3)
0
follows readily from Theorem 4.3 and assumption (5.1). For f 1 () and xed
t, the upper estimate above goes to zero as n, m . In fact, the convergence is
uniform for t [T, T ]. This proves (5.2).
By an /3 argument, similar to what is done at the end of Sec. 2, weak continuity
follows since we know it holds for the nite volume dynamics. This completes the
proof of Theorem 5.1.
Acknowledgments
The work reported in this paper was supported by the National Science Foundation:
B.N. under Grants #DMS0605342 and #DMS0757581, R.S. under Grant #DMS
0757424, and S.S. under Grant #DMS0757327 and #DMS0706927. The authors
would also like to acknowledge the hospitality of the Department of Mathematics
at U.C. Davis where a part of this work was completed.
References
[1] L. Amour, P. LevyBruhl and J. Nourrigat, Dynamics and LiebRobinson estimates
for lattices of interacting anharmonic oscillators, to appear in Colloq. Math., Special
volume dedicated to A. Hulanicki; arXiv:0904.2717.
[2] H. Araki and E. J. Woods, Representations of the canonical commutation relations
describing a nonrelativistic infinite free Bose gas, J. Math. Phys. 4 (1963) 637662.
[3] O. Bratteli and D. W. Robinson, Operator Algebras and Quantum Statistical Mechan
ics. Volume 1, 2nd edn. (SpringerVerlag, 1987).
March 10, 2010 10:13 WSPC/S0129055X 148RMP
J070S0129055X1000393X
[4] O. Bratteli and D. W. Robinson, Operator Algebras and Quantum Statistical Mechan
ics. Volume 2, 2nd edn. (SpringerVerlag, 1997).
[5] P. Butt` a, E. Caglioti, S. Di Ruzza and C. Marchioro, On the propagation of a per
turbation in an anharmonic system, J. Stat. Phys. 127 (2007) 313325.
[6] M. Cramer and J. Eisert, Correlations, spectral gap, and entanglement in harmonic
quantum systems on generic lattices, New J. Phys. 8 (2006) 71.
[7] M. Cramer, A. Serafini and J. Eisert, Locality of dynamics in general harmonic
quantum systems, in Quantum Information and Many Body Quantum Systems, eds.
M. Ericsson and S. Montangero (Edizioni della Normale, 2008).
[8] M. Hastings and T. Koma, Spectral gap and exponential decay of correlations, Comm.
Math. Phys. 265(3) (2006) 781804.
[9] J. L. van Hemmen, Dynamics and ergodicity of the infinite harmonic crystal, Phys.
Rept. 65 (1980) 45149.
[10] O. E. Lanford, J. Lebowitz and E. H. Lieb, Time evolution of infinite anharmonic
systems, J. Statist. Phys. 16(6) (1977) 453461.
[11] E. H. Lieb and D. W. Robinson, The finite group velocity of quantum spin systems,
Comm. Math. Phys. 28 (1972) 251257.
[12] E. H. Lieb and R. Seiringer, The Stability of Matter in Quantum Mechanics (Cam
bridge University Press, 2009).
[13] J. Manuceau and A. Verbeure, Quasifree states of the CCR algebra and Bogoliubov
transformations, Comm. Math. Phys. 9 (1968) 293302.
[14] J. Manuceau, M. Sirugue, D. Testard and A. Verbeure, The smallest C algebra for
canonical commutation relations, Comm. Math. Phys. 32 (1973) 231243.
[15] C. Marchioro, A. Pellegrinotti, M. Pulvirenti and L. Triolo, Velocity of a perturbation
in infinite lattice systems, J. Statist. Phys. 19(5) (1978) 499510.
[16] B. Nachtergaele and R. Sims, LiebRobinson bounds and the exponential clustering
theorem, Comm. Math. Phys. 265(1) (2006) 119130.
[17] B. Nachtergaele, Y. Ogata and R. Sims, Propagation of correlations in quantum
lattice systems, J. Statist. Phys. 124(1) (2006) 113.
[18] B. Nachtergaele and R. Sims, Locality estimates for quantum spin systems, in New
Trends in Mathematical Physics, Selected Contributions of the XVth International
Congress on Mathematical Physics (SpringerVerlag, 2009), pp. 591614.
[19] B. Nachtergaele, H. Raz, B. Schlein and R. Sims, LiebRobinson bounds for harmonic
and anharmonic lattice systems, Comm. Math. Phys. 286 (2009) 10731098.
[20] H. Raz and R. Sims, Estimating the LiebRobinson velocity for classical anharmonic
lattice systems, J. Statist. Phys. 137 (2009) 79108.
[21] M. Reed and B. Simon, Methods of Modern Mathematical Physics, II, Fourier Anal
ysis, SelfAdjointness (Academic Press, 1975).
[22] D.W. Robinson, The ground state of the bose gas, Comm. Math. Phys. 1 (1965)
159174.
[23] H. Spohn and J. L. Lebowitz, Stationary nonequilibrium states of infinite harmonic
systems, Comm. Math. Phys. 54 (1977) 97120.
[24] W. Thirring and F. Dyson (eds), The Stability of Matter: From Atoms to Stars:
Selecta of Elliott H. Lieb, 4th edn. (SpringerVerlag, 2005).
April 20, 2010 14:17 WSPC/S0129055X 148RMP J07000395
Contents
1. Introduction 234
2. GrandCanonical Pressure and Gap Equation 241
233
April 20, 2010 14:17 WSPC/S0129055X 148RMP J07000395
1. Introduction
Since the discovery of mercury superconductivity in 1911 by the Dutch physicist
Onnes, the study of superconductors has continued to intensify, see, e.g., [2]. Since
that discovery, a signicant amount of superconducting materials has been found.
This includes usual metals, like lead, aluminum, zinc or platinum, magnetic materi
als, heavyfermion systems, organic compounds and ceramics. A complete descrip
tion of their thermodynamic properties is an entire subject by itself, see [24] and
references therein. In addition to zeroresistivity and many other complex phe
nomena, superconductors manifest the celebrated Meiner or MeinerOchsenfeld
eect, i.e. they can become perfectly diamagnetic. The highesta critical tempera
ture for superconductivity obtained nowadays is between 100 and 200 Kelvin via
doped copper oxides, which are originally insulators. In contrast to most supercon
ductors, note that superconduction in magnetic superconductors only exists on a
nite range of nonzero temperatures.
Theoretical foundations of superconductivity go back to the celebrated BCS
theory appeared in the late fties (1957) which explains conventional type I
a In January 2008, a critical temperature over 180 Kelvin was reported in a Pbdoped copper oxide.
April 20, 2010 14:17 WSPC/S0129055X 148RMP J07000395
1
+ k, a
k,k a k, a
k , a
k , (1.1)

k,k
dened in a cubic box R3 of volume . Here is the dual group of seen as a
torus (periodic boundary condition) and the operator ak,s respectively ak,s creates
respectively annihilates a fermion with spin s {, } and momentum k .
The function k represents the kinetic energy, the real number is the chemical
potential and k,k is the BCS coupling function. The choice k,k = < 0 is often
used in the Physics literature and the case k = 0 is known as the strong coupling
limit of the BCS model.
The lattice approximation of the BCS Hamiltonian amounts to replace the box
R3 by Z3 (or, more generally, by Zd1 ) and the strong coupling
limit of the reduced BCS model is in this case known as the strong coupling (with
k,k = ) BCS model.b The assumptions k = 0 and k,k = are of interest,
because in this case the BCS Hamiltonian can be explicitly diagonalized. The exact
solution of the strong coupling BCS model is wellknown since the sixties [6, 7].
This model is in a sense unrealistic: among other things, its representation of the
kinetic energy of electrons is rather poor. Nevertheless, it became popular because
it displays most of basic properties of real conventional type I superconductors.
See, e.g., [8, Chap. VII, Sec. 4]. Even though the analysis of the thermodynamics of
the BCS Hamiltonian was rigorously performed in the eighties [9, 10] (see also the
innovating work of Bernadskii and Minlos in 1972 [11]), generalizations of the strong
coupling approximation of the BCS model are still subject of research. For instance,
strong couplingBCStype models with superconducting phases at arbitrarily high
temperatures are treated in [12].
In fact, a general theory of superconductivity is still a subject of debate, espe
cially for highTc superconductors. An important phenomenon ignored in the BCS
theory is the Coulomb interaction between electrons or holes, which can imply
strong correlations, for instance in highTc superconductors. To study these cor
relations, most of theoretical methods, inspired by Beliaev [5], use perturbation
theory or renormalization group derived from the diagram approach of Quantum
Field Theory. However, even if these approaches have been successful in explaining
many physical properties of superconductors [3, 4], only few rigorous results exist
on superconductivity.
For instance, the eect of the Coulomb interaction on superconductivity is not
rigorously known. This problem was of course adressed in theoretical Physics right
after the emergence of the Frohlich model and the BCS theory, see, e.g., [13].
for real parameters , h, , and 0. The operator ax,s respectively ax,s creates
respectively annihilates a fermion with spin s {, } at lattice position x Zd
whereas nx,s := ax,s ax,s is the particle number operator at position x and spin s.
The rst term of the righthand side of (1.2) represents the strong coupling limit
of the kinetic energy, with being the chemical potential of the system. Note that
this strong coupling limit explained above for the BCS Hamiltonian is also
called atomic limit in the context of the Hubbard model, see, e.g., [14, 15]. The
second term in the righthand side of (1.2) corresponds to the interaction between
spins and the magnetic eld h. The onesite interaction with coupling constant
represents the (screened) Coulomb repulsion as in the celebrated Hubbard model.
So, the parameter should be taken as a positive number but our results are also
valid for any real . The last term is the BCS interaction written in the xspace
since
ax, ax, ay, ay, = a
k, a
k, a
q, a
q, , (1.3)
N N
x,yN k,qN
with N being the reciprocal lattice of quasimomenta and where a q,s is the cor
responding annihilation operator for s {, }. Observe that the thermodynamics
of the model for = 0 can easily be computed. Therefore, we restrict the analy
sis to the case > 0. Note also that the homogeneous BCS interaction (1.3) can
imply a superconducting phase and the mediator implying this eective interaction
does not matter here, i.e. it could be due to phonons, as in conventional type I
superconductors, or anything else.
We show that the onesite repulsion suppresses superconductivity for large
0. In particular, the repulsive term in (1.2) cannot imply any superconducting
state if = 0. However, the rst elementary but nonetheless important property
of this model is that the presence of an electron repulsion is not incompatible with
superconductivity if  and (+h) are not too big as compared to the coupling
constant of the BCS interaction. In this case, the superconducting phase appears
at low temperatures as either a rst order or a second order phase transition.
More surprisingly, the onesite repulsion can even favor superconductivity at xed
chemical potential by increasing the critical temperature and/or the Cooper pair
condensate density. This contradicts the naive guess that any onesite repulsion
between electron pairs should at least reduce the formation of Cooper pairs. It is
however important to mention that the physical behavior described by the model
depends on which parameter, or , is xed. (It does not mean that the canonical
and grandcanonical ensembles are not equivalent for this model.) Indeed, we also
analyze the thermodynamic properties at xed electron density per site in the
grandcanonical ensemble, as it is done for the perfect Bose gas in the proof of
BoseEinstein condensation. The analysis of the thermodynamics of the strong
coupling BCSHubbard model is performed in details. In particular, we prove that
the Meiner eect is rather generic but also that the coexistence of superconducting
and ferromagnetic phases is possible (as in the VonsovkiiZener model [16, 17]), for
instance at large > 0 and densities near halflling. The later situation is related to
a superconductorMott insulator phase transition. This transition gives furthermore
a rigorous explanation of the need of doping insulators to obtain superconductors.
Indeed, at large enough coupling constant , the superconductorMott insulator
phase transition corresponds to the breakdown of superconductivity together with
the appearance of a gap in the chemical potential as soon as the electron density
per site becomes an integer, i.e. 0, 1 or 2. If the system has an electron density
per site equal to 1 without being superconductor, then any nonzero magnetic eld
h = 0 implies a ferromagnetic phase.
Note that the present setting is still too simplied with respect to real super
conductors. For instance, the antiferromagnetic phase or the presence of vortices,
which can appear in (type II) highTc superconductors [3,4], are not modeled. How
ever, the