A NNALES DE L ’I. H. P., SECTION A
V. G RECCHI M. M AIOLI
Borel summability beyond the factorial growth
Annales de l’I. H. P., section A, tome 41, n
o1 (1984), p. 37-47
<http://www.numdam.org/item?id=AIHPA_1984__41_1_37_0>
© Gauthier-Villars, 1984, tous droits réservés.
L’accès aux archives de la revue « Annales de l’I. H. P., section A » implique l’accord avec les conditions générales d’utilisation (http://www.numdam.
org/conditions). Toute utilisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
37
Borel summability beyond the factorial growth
V. GRECCHI M. MAIOLI Istituto Matematico, Universita di Modena,
41100 Modena, Italy
Henri Poincaré,
Vol. 41,~1,1984, Physique theorique
ABSTRACT. A criterion for a
generalized
Borelsummability
isgiven.
In this way we define a
unique
sum for seriesdiverging
faster than any factorial power. Possibleapplications
inquantum
mechanics and quantum field theories are discussed.RESUME. 2014 On donne un critere de sommabilite de Borel
generalisee.
De cette
façon,
on peut definir une sommeunique
pour des seriesdivergeant plus
vite que toutepuissance
de factorielle. On discute desapplications possibles
enmecanique quantique
et en theoriequantique
deschamps.
I INTRODUCTION
Non-polynomial
anharmonic oscillators have beenrecently
treatedby Doglov
andPopov [3] and
one of us[10 ].
Inparticular
for the operatorp2 + x2
+zex, by using
WKBmethods,
adivergency
of theperturbation expansion
coefficients of the order ofek2~4
is shown[3 ].
On the basis of this resultDoglov
andPopov
draw the conclusion that theperturbation
series is not Borel summable : in fact the extension of the Borel sum intro- duced
by
LeRoy (see
e. g. Ref.6)
does notapply
if thegrowth
of the coeffi- cients is faster than(nk) !
for all n E N.As a matter of fact there exists a
general
class of summationmethods,
called
by Hardy
« moment constant methods »(Ref. 6,
pp.81-86),
definedas follows. Let
Eakzk
be a formal powerseries,
whose « sum » we wishto define. be a sequence of
strictly positive
numbersgiven
l’Institut Henri Poincaré - Physique theorique - Vol. 41, 0246-0211
84/01/37/ll/$ 3,10 (Ç) Gauthier-Villars
38 V. GRECCHI AND M. MAIOLI
as the moments of some function on ,uk = We say that the formal power series
Eakzk
is «,u-p-Borel
summable » if :00
a) B(t) =
converges in some disk 8;k=0
b) B(t)
has ananalytic
continuation to aneighbourhood
of thepositive
real
axis ;
-
f~
c) g(z)
= converges for some z 7~ 0.B(t)
is called the «-03C1-Borel
transform » of the seriesEakzk
andg(z)
is its «
,u-p-Borel
sum ». Theordinary
Borel summation method takesp(t) = e - t.
The Borel-LeRoy
method of order((x,~8)
takespk =
r(ak
+~),/~)
=exp ( -
In the present article we
study
the method(Ref. 3,
pp.84-86)
definedby
Pk =
(7r/~~~ p(t)
= for fixed a > 0. We prove a criterion for this summation methodanalogous
to the Watson-Nevanlinna theorem(Ref. 6, 11, 15, 17)
for theordinary Borel (or
Borel-LeRoy)
summationmethod. More
precisely,
letf(z)
be a functionanalytic
on the Riemannsurface of the
logarithm
for 0! I z Ro,
which has03A3akzk
as itsasymptotic
series in a
suitably
strong sense(condition (2) below).
Then the-03C1-Borel
sum
g(z) (see
theexpression ( 14),
or theequivalent
one(5))
exists andequals
the
original
functionf (z). Moreover,
a remainder estimate of the type(2)
is a necessary
condition,
too, so alarge
class of suchlogarithmic
Borelsums is characterized.
The
techniques
of theproof
are similar to the ones in Ref.11,
15 after the choice of a suitable new variable and a redefinition of the function and of its transform.Besides,
for what concerns the estimate on the remain-ders,
in this case theproblem
is to combine thedivergence
in the index and the radial behaviour with a necessaryangular divergence.
For what concerns the above
quoted models,
we do notgive
here theapplicability
result because the usual operatortheory techniques
appearinadequate
toverify
all the criterionhypotheses.
On the otherhand, by
a heuristicanalytis
andby partial
results it seemsplausible
to us thatthe
proved
criterion can beapplied
in these cases. We quote some summa-bility examples
from the context ofsimplified
modelsrecently
introduced[9 ],
and for the characteristic function of a
probability
distribution introducedby Kolmogorov [4 ], relating
to the dimensions ofparticles yielded by
a
body pulverization.
In Section II the
summability
theorem isproved
and anexample
andsome remarks are
given.
In Section IIIpossible applications
toquantum
mechanics are discussed.l’Institut Henri Poincaré - Physique theorique
39 BOREL SUMMABILITY BEYOND THE FACTORIAL GROWTH
II LOGARITHMIC BOREL SUMMABILITY
THEOREM. 2014 Let
f(z)
beanalytic
on the Riemann surface of thelogarithm
for - oo Re
(log z)
Co(for
some Co E[R)
and let itsatisfy
thefollowing properties :
where the
positive
constantsA, £5,
a,~o rc)
areindependent of z
of 6 =
arg (z)
and of N E N. Then the seriesis convergent
for t ~
~ -1 and has ananalytic
continuation to aregion
~ -1
or
arg(t) ~
Thisregion corresponds,
inthe variable x = - 2
log t,
toSx(~,
x > 2log ~ or
Imx ~
If is the
analytic
continuation of( 1/2)e - °‘x2~4B(e - x~2)
then :uniformly
for m E Imx | 203C61
and c E( -
oo,co) (for fixed 03C61
Moreover
uniformly
for(for
fixed xo > 2log £5).
Besides, f(z)
can berepresented
as anabsolutely
convergentintegral :
for all z such that - oo Re
(log z)
Co.Conversely,
if the function is such thatB(t)
=log t)
is
analytic (for
suitablea)
in theregion
and if it satisfies(4a)-(4b),
then the
function f(z)
definedby (5)
isanalytic
in theregion
and satisfies a bound of the type
(2)
with formalexpansion
coefficientsProof
2014 Let usconsider
theintegral
obtained from(5) by
the Riemann-40 V. GRECCHI AND M. MAIOLI
Fourier inversion formula of the two-sided
Laplace
transformation(see
e. g. Ref.
2) :
where the
integration
variable is u =log
z, andce(2014
oo,co). By hypo-
theses
(1), (2)
theintegral
in(6)
is convergent for all x suchthat
Imx ~ 2~0,
and for such x is
independent
ofc E (-
oo,co).
In factby using (1), (2)
with N =
1,
andsetting
u = c +?,
we have :if we consider all jc such
that | Im x| ~ 24>1’
with4>1 4>0’
and C E(-
00,Co).
Notice that for Re jc > -
2co,
we can choose c= - - 2
Re x andthereby
obtain
Inserting (1)
into(6) (now
witharbitrary N),
The first term in
(8), computed by writing
u = c + ? andperforming
N-l
the Gaussian
integral, gives (1/2)~’~ ~ Therefore
k=0
by hypothesis (2) again setting
u = c +i6, (8) implies :
as N -+ oo whenever both Re x > 2
log ~ and
it is suffi-cient to choose c = -
N/2a,
so thatOn the other hand 0 this series is
just convergent
in thehalf-plane
"Re x > 2
log ð, by
thebound I ak implicit
in(2).
So the func-Annales de Henri Poincaré - Physique ’ theorique ’
41
BOREL SUMMABILITY BEYOND THE FACTORIAL GROWTH
tion
J3(x)
definedby (6)
isanalytically
continuable to aright half-plane,
where it is
uniquely
determinedby
thesequence {ak}k~N.
Let us consider the m-th derivative of
The
integral
in(10)
exists for any x suchthat
Imx ~ 24>0 and, by hypo-
theses
(1), (2)
with N =1,
it satisfies the boundwhere
Ai, A2
areindependent
of m, c and x in thestrip | Im x 203C61,
for any
fixed 411 ~o’
Thus the function is differentiable to all orders in the
strip I 2~o
and theTaylor
seriesexpansion
is convergent in some
neighbourhood of x 1,
for any x 1 in thestrip.
Moreover= if x
belongs
to both the convergence disks. So theanalytic
continuation to the whole
strip 24>0
issingle
valuedand, by (9),
is
uniquely
determinedby
the coefficients of the formalexpansion (1).
Besides,
satisfies the bounds(4) :
indeed(4a)
isproved by ( 11 ),
whileby (9)
we haveuniformly
for Re . x > xo(with
xo > 2log (5).
As a consequence,
inserting (3(x)
in(5),
theintegral
isabsolutely
convergentat - oo
by (4 a)
andabsolutely
convergent at + ooby (4 b).
On the other hand theintegral
in(6)
isabsolutely
convergentby (7) : therefore, by
theinversion formula of the
Laplace
transform[2 ], inserting (6)
in(5)
we havean
identity
and the criterion isproved.
Conversely,
let begiven,
such thatB(t)
=log t) is analytic
in theregion St(~, (for
suitable a >0),
and let itsatisfy (4 a)-(4 b).
Then the function defined
by (5)
isclearly analytic
for - oo Re(log z)
Co.By
the substitution x = - 2log t, f(z)
can berepresented
in the form1-1984.
42 V. GRECCHI AND M. MAIOLI
for
any 03C6
such sinceB(t)
isanalytic
in thesector
arg(t)|
Setting
for some a =
~(t),
0 ~ ~ ~ 1.Now,
since isanalytic for |t| 03B4-1,
we
have, by Cauchy’s integral
formula(N !)-1 I
~Di(25)~
on~N
the set
Ai={~/0~(3~)’’}. Moreover, B~)=~(~~(-21og~))
can be bounded
by
means of(4 ~), taking
into account that ~ = - 2log
~.In
general, setting ~
=log
t, we haveOn the other hand
(e.
g. Ref.5,
p.1033)
By combining everything
one has the estimate :Annales de Henri Poincaré - Physique ’ theorique ’
43 BOREL SUMMABILITY BEYOND THE FACTORIAL GROWTH
Since the
right
hand side of( 16)
is monotoneincreasing
forlarge
t andmonotone
non-increasing
for t1,
and since 0 81, setting
by ( 15)
and( 16)
we have :where the constants are
independent
of z and N.Choosing ~ _ ~ 1 (0
when () > 0 when ()0, (17)
is an estimateof the
required
type and the theorem isproved.
REMARK 1. 2014 As a consequence, one has the
following
annulment crite- rion : if/(z)
isanalytic
for 0I z Ro
on alogarithmic
Riemann surfaceand,
forA, a, ~o independent
of z,then
f(z)
= 0. In fact( 18) implies that ak
=(~!)’~(0 + )
= 0 for allk,
while
(2)
is satisfied since ’v’N EAnalytic
non-zerofunctions on these
domains,
such as fail tosatisfy (18) uniformly
with respectto I z I
and 0.with formal
expansion, 03A3akzk, ak
=( - 1)ke(k+1)2/403C01/2,
satisfies thehypo-
theses of the above theorem and
03A3akzk
admits aunique
sumgiven by (5),
with a = 1 and =
(1/2)~-~(1
+et 1 - x)/2) - 1.
In the context of thetheory
ofStieltjes summability,
we can consider the function44 V. GRECCHI AND M. MAIOLI
which fails to
satisfy hypothesis (2).
The moments(in
Ref. 13 and16)
can bethought
as the formalexpansion
coefficientsof fo(z).
Both
f(z)
and/(z)
+fo(z)
areStieltjes functions,
andthey
have the sameasymptotic
seriesexpansion.
Thus03A3akzk
is aStieltjes series,
which turnsout not to be
Stieltjes
summable[13 ]. However,
as aboveremarked,
it is summable
by
thisgeneralized
Borel method.REMARK 2. 2014 This summation method is
regular,
since itobviously
sums a
convergent
series in the interior of the convergence circle. However ingeneral
it cannot be used toanalytically
continue the functionbeyond
the convergence radius. Indeed
Hardy
shows(Ref. 6,
p.85)
that it cannotsum
Ezk
for z = - 1. This is notsurprising,
since the Borel-LeRoy
sum,-~
00 ,
of index n >
2,
of this series :( 1 -
=~’~
isk=0
valid in the interior of a
compact
setfor 0! ~ ~:
see Ref.
6,
p.197),
which tends to the disk of radius 1 as n -+ oo .III. APPLICABILITY
(A)
The above mentioned operatorH(z) = p2 + x2
+ ze" in with its formalperturbation series, displays
thedifficulty
ofapplication
of this criterion in the usual framework of operator
theory,
while thisapplication
appears reasonableby
thepartial
results obtained andby
the
foregoing
discussion(in (B))
of asimplified
model.Recall
[70] that H(z)
is acompact
resolventoperator for I arg (z)
7r,and the
eigenvalues
are stable as z !-+ 0by
norm resolvent convergence.By
a translation x ->(x - log z),
or x -+(x - ?),
theeigenvalues
canbe continued to a
logarithmic
Riemann surface in such a way,however,
thatthey
are not controlledas z | is kept
constant and 0 = arg(z)
-+ ±00.A
growth
of thefunction, for I z fixed,
of the order of is allowedby hypothesis (2)
of the above theorem. On the other hand an usual estimateby projection
operators(which
areCauchy integrals)
would presuppose theperturbed eigenvalue
to be near theunperturbed
oneuniformly
as8 -+ :t oo .
However
by
an estimate in any fixedangular sector e ~ - 80,
one checksAnnales de l’Institut Henri Poincaré - Physique theorique
45
BOREL SUMMABILITY BEYOND THE FACTORIAL GROWTH
just
the bound on the remainderrequired by (2)
for the behaviour with respectto Indeed,
the actualeigenvalues
ofare the
analytic
continuation of theeigenvalues
ofH(z) beyond
the firstsheet.
For 0! ~ Rz
=(H( -
isuniformly
norm con-vergent as z !-+ 0. Thus it is sufficient to evaluate
(see
e. g. Ref.12, 7)
thenorm of the vectors :
in order to estimate the
perturbation
series remainders.Setting . f ’(x)
=(i/2j)(x2 - (k
+l)log(2
+x2))
we have :and
analogously
forNow,
one checksthat ~w-sRzws~ ~
c, for some c > 0 and for 1 ~ s ~j.
Indeed we haveand
has invariant domain and numerical ranges contained in the
half-plane Moreover Ij
It follows that the remainders are
uniformly
boundedby
zIN, when 0! 0o, for
any fixedBo,
so that therequired
bound(2)
holdsat least in this
region.
(B)
Instead ofsingle eigenvalues,
let us consider anotherunitary
inva-riant of
H(z),
that is To calculateit,
one can use the Trotter formula for the kernel ofwhere the are defined in Ref. 14
(p. 6). Following
G.’t Hooft[9 ]
we can consider a
simplified model,
in which K isreplaced by K(1).
Thetrace
corresponding
to y ; -it)
isgiven by
For each t >
0,
the formalexpansion
coefficients are46 V. GRECCHI AND M. MAIOLI
Tt(z)
satisfies thehypotheses
of the above theorem and admits the repre- sentation(5) (with
a = t,/3(x) _ (2~ct) -1 ~2( 1/2)e - t"2~4e - te - "~2) uniquely
determined
by
theak’s.
Notice that to ourknowledge
all the quantum mechanical modelsadmitting
usual Borelsummability
after suchsimpli- fication,
areactually proved
to be Borel summable(e.
g.p2 + x2
+[12 ].
As far as quantum field theories areconcerned,
thesummability
method could
perhaps
beapplied
toexponential
interaction models introduced in Ref.8,
1.(C)
Let usfinally
quote anexample provided by
the characteristic function of thelogarithmic
normalprobability
distribution defined asintroduced
by Kolmogorov (see
Ref.4,
p.32,
and relatedreferences)
forthe dimensions of
particles yielded by
abody pulverization.
If we considerthe characteristic function of such
probability, /(z)
= it satisfiesthe
hypotheses
of the theorem. As a consequence, the moments of such measure,simply
related to the formalexpansion
coefficients off (z),
uni-quely
determine the characteristic function via thelogarithmic
Borel sum.ACKNOWLEDGEMENT
It is a
pleasure
to thank Prof. A. Sokal for a useful discussion.[1] S. ALBEVERIO and R. HOEGH-KROHN, J. Funct. Anal., t. 16, 1974, p. 39.
[2] G. DOETSCH, Introduction to the Theory and Application of the Laplace Transforma- tion, Springer-Verlag, 1974.
[3] A. D. DOGLOV and V. S. POPOV, Phys. Lett., t. 79 B, 1978, p. 403.
[4] B. V. GNEDENKO, Yu. K. BELYAYEV and A. D. SOLOVYEV, Méthodes Mathématiques
en Théorie de la Fiabilité, Éditions MIR, 1972.
[5] I. S. GRADSHTEYN and I. M. RYZHIK, Table of Integrals, Series and Products, Acade- mic Press, 1980.
[6] G. H. HARDY, Divergent Series, Clarendon Press, Oxford, 1949.
[7] I. HERBST, Comm. Math. Phys., t. 64, 1979, p. 279.
[8] R. HØEGH-KROHN, Comm. Math. Phys., t. 21, 1971, p. 244.
[9] G.’T HOOFT, private communication via B. Simon.
[10] M. MAIOLI, J. Math. Phys., t. 22, 1981, p. 1952.
[11] F. NEVANLINNA, Ann. Acad. Sci. Fennicae, Ser. A, t. 12, n° 3, 1918-1919.
[12] M. REED and B. SIMON, Methods of Modern Mathematical Physics, vol. IV, Academic Press, 1978.
[13] J. A. SHOHAT and J. D. TAMARKIN, The Problem of Moments, A. M. S., New York, 1943.
Annales de Henri Poincare - Physique theorique
47
BOREL SUMMABILITY BEYOND THE FACTORIAL GROWTH
[14] B. SIMON, Functional Integration and Quantum Physics, Academic Press, 1979.
[15] A. D. SOKAL, J. Math. Phys., t. 21, 1980, p. 261.
[16] T. J. STIELTJES, Annales de la Faculté des Sciences de Toulouse, (1), t. 8, 1894, T, p. 1-122, (1), t. 9, 1895, A, p. 5-47.
[17] G. N. WATSON, Phil. Trans. Roy. Soc. London, Ser. A, t. 211, 1912, p. 279.
(Manuscrit reçu le 25 juin 1983) (Version revisee reçue le 12 octobre 1983)