HAL Id: hal-01693163
https://hal.archives-ouvertes.fr/hal-01693163
Submitted on 25 Jan 2018
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
of the planar Brownian curve
Vincent Beffara
To cite this version:
Vincent Beffara. On conformally invariant subsets of the planar Brownian curve. Annales de
l’Institut Henri Poincaré (B) Probabilités et Statistiques, Institut Henri Poincaré (IHP), 2003, 39
(5), pp.793 - 821. �10.1016/S0246-0203(03)00030-X�. �hal-01693163�
arXiv:math/0105192v1 [math.PR] 23 May 2001
the Planar Brownian Curve
Vinent BEFFARA
Département de Mathématiques
Université Paris-Sud
F-91405ORSAY Cedex, Frane
Vinent.Beffaramath.u-psud. fr
Abstrat
Wedeneandstudyafamilyofgeneralizednon-intersetionexponentsforplanarBrow-
nianmotionsthatisindexedbysubsetsoftheomplexplane: Foreah
A ⊂ C
,wedeneanexponent
ξ(A)
thatdesribesthedeayofertainnon-intersetionprobabilities. Toeahof theseexponents,weassoiateaonformallyinvariantsubsetoftheplanarBrownianpath,ofHausdordimension
2 − ξ(A)
. Aonsequeneofthisandontinuityofξ(A)
asafuntionofA
isthealmost sureexisteneofpivoting pointsofanysuientlysmallangle onaplanarBrownianpath.
Résumé
Nousdénissons et étudions unefamilled'exposantsde non-intersetiongénéralisés en-
tre mouvements browniens plans, indexée par lesparties du plan omplexe : pour haque
A ⊂ C
nous dénissonsunexposantξ(A)
dérivantladéroissanede probabilitésde non- intersetion.Àhaundees exposantsest assoiéeunepartiedelatrajetoirebrowniennequi est invariante sous l'ation des transformations onformes et qui a une dimension de
Hausdor égale à
2 − ξ(A)
. Une onséquene de e résultat et de la ontinuité deξ(A)
omme fontion de
A
est l'existene presque sûre de points pivotants de tout angle assezpetit surunetrajetoirebrownienneplane.
Contents
Introdution 2
1 Generalized intersetion exponents 4
1.1 Denition of the exponents . . . 4
1.2 Strong approximation . . . 5
2 Properties of the funtion
A 7→ ξ(A)
11 3 Hausdor dimension of the orresponding subsets of the path 13 3.1 Conformally invariant subsetsof the Brownian path . . . 133.2 Seondmoments . . . 15
3.3 Hausdor dimensions . . . 15
3.4 Remark about ritialases . . . 20
4 Bounds and onjetures on the exponent funtion 21
4.1 Knownexat values of
ξ
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214.2 An upperboundfor the exponent . . . 22
4.3 Conjetured andexperimentalvalues . . . 25
5 Appendix 26
5.1 Sub-additivity. . . 26
5.2 Extremaldistane . . . 26
5.3 Sometopologial tools . . . 27
Introdution
It hasbeen onjeturedfor more than twenty years by theoretialphysiiststhat onformalin-
varianeplaysanimportantrole tounderstandthebehaviourofritialtwo-dimensional models
of statistialphysis. They justifyby amathematially non-rigorous argument involving renor-
malization ideas that inthe saling limit these models behave in a onformally invariant way;
they have been able to lassify them via a real-valued parameter orresponding to the entral
harge of the assoiated Virasoro algebra, and to predit the exat value of ritial exponents
thatdesribe thebehaviourofthese systems. Dierent models (forinstane,self-avoidingwalks
and perolation) withthe sameentral hargehave thesame exponents.
Reently, Shramm [19℄ introdued new mathematial objets that give insight into these
onjetures. Thesearerandomset-valued inreasing proesses
(K t ) t>0
that healledStohastiLöwnerEvolutionproesses. Foreahpositivenumber
κ
,thereexistsonesuhproessofparam-eter
κ
, inshortSLE κ
. He proved thatfor various models,if they have a onformally invariantsalinglimit,thenitanbeinterpretedintermsofoneofthe
SLE κ
's(theparameterκ
isrelatedto the entral harge of the model). One an then interpret the onjetures from theoretial
physiists interms ofproperties ofthis proess.
Inpartiular,Lawler,ShrammandWerner [13,14 ℄showedthatforonespeivalueofthe
parameter
κ
(namelyκ = 6
)whih onjeturallyorrespondedto thesalinglimitofperolation luster interfaes, theSLE 6
has the remarkable restrition property that enables to relate itsritial exponents to the so-alled intersetion exponents between planar Brownian motions.
This lead [13, 14,15 , 12 ℄ to the derivation of the exat value of theexponents between planar
Brownian paths. Furthermore, it turned out [23 ℄ that in fat, the outer boundary of a planar
Brownian urve hasexatlythesame lawthan that of an
SLE 6
. Inother words, thegeometryof ritialtwo-dimensional perolation lusters intheir saling limit should be exatlythat ofa
planar Brownian outer frontier.
In a very reent paper Smirnov [20℄ showed that ritial site perolation in the triangular
lattieisonformallyinvariantinthesalinglimitsothatthegeometryofritialtwo-dimensional
perolation lusters boundariesintheir salinglimitisidentialthatofaplanarBrownianouter
frontier.
Before all these reent developments, geometri properties of planar Brownian paths had
alreadybeensubjetofnumerousstudies(seee.g.[18℄forreferenes). Inpartiular,theHausdor
dimension of various subsets ofthe planar Brownian urvedened ingeometri termshadbeen
determined. For instane,Evans[4℄showedthatthe Hausdordimension ofthesetoftwo-sided
one points of angle
θ
(i.e.pointsB t
suh thatbothB [0,t]
andB [t,1]
are ontained inthe sameone of angle
θ
with endpoint atB t
) is2 − 2π/θ
. In a series of papers (see [9℄ for a review),LawlerprovedthatthedimensionofvariousimportantsubsetsoftheplanarBrownianurve an
be related to Brownian intersetionexponents. In partiular [8℄, he showed that thedimension
oftheset
C
of utpoints (i.e.pointsB t
suhthatB [0,1] \ { B t }
isnot onneted) is2 − ξ
whereξ
istheBrownian intersetionexponent dened byp R = P(B [0,T 1 1
R ] ∩ B [0,T 2 2
R ] = ∅) = R − ξ+o(1)
(1)(for independent Brownian paths
B 1
andB 2
starting respetively from1
and− 1
,T R 1
andT R 2
standing for theirrespetive hittingtimes oftheirle
C (0, R)
).In order to derive suh results and in partiular the more diult lower bound
d > 2 − ξ
,the strategy is rst to rene the estimate (1) into
p R ≍ R − ξ
(we shall use this notation todenotetheexisteneoftwo positive onstants
c
andc ′
suhthatcR − ξ 6 p R 6 c ′ R − ξ
), toderiveseond-momentestimatesandtousethesefatstoonstrutarandommeasureofnite
r
-energysupportedon
C
, for allr < 2 − ξ
. The determination of the value of theritial exponents viaSLE 6
[13, 14℄ then implies that the dimension ofC
is3/4
. Similarly, in [7℄ the HausdordimensionoftheouterfrontierofaBrownianpathanbeinterpretedintermsofanotherritial
exponent,andthedeterminationofthisexponentusing
SLE 6
thenimplied(see[16 ℄forareview)thatthis dimensionis
4/3
asonjeturedby Mandelbrot.Inthe present paper, we dene andstudy a family of generalizationsof theBrownian inter-
setion exponent
ξ
parameterized bysubsets of the omplex plane. For eahA ⊂ C
, we deneanexponent
ξ(A)
asfollows. LetB 1
andB 2
betwoindependent planarBrownianpathsstarting fromuniformlydistributedpointson theunitirle : thenξ(A)
isdened byp R (A) = P (B [0,T 1 1
R ] ∩ A.B [0,T 2 2
R ] = ∅) = R − ξ(A)+o(1)
(2)(withthenotation
E 1 .E 2 = { xy : x ∈ E 1 , y ∈ E 2 }
). Note that theaseA = { 1 }
orrespondstothe usualintersetionexponent. InSetion 1,we rst showthatfor a wide lassofsets
A
p R (A) ≍ R − ξ(A) .
(3)In Setion 2, we study regularity properties of the mapping
A 7→ ξ(A)
. In partiular, weproveuniform ontinuity (withrespetto theHausdormetri) onertainfamiliesof sets. One
important tool for this result is the fatthat the onstants impliit in (3) an in fatbe taken
uniform overthese familiesof sets.
InSetion 3,weassoiate toeah set
A
a subsetE A
oftheplanarBrownian urve denedingeometri terms:
E A = { B t : ∃ ε > 0, (B [t − ε,t] − B t ) ∩ A.(B (t,t+ε] − B t ) = ∅} .
Using the strong approximation and ontinuity of the mapping
A 7→ ξ(A)
, we then show thatthe Hausdor dimension of this subset of theplanar Brownian urve is almost surely
2 − ξ(A)
(and
0
inaseξ(A) > 2
). For example,whenA = { e iθ , 0 6 θ 6 α }
,theorrespondingsubsetC α
of the Brownian urve isthe set of (loal) pivoting points, i.e.pointsaround whih one half of
thepath anrotate of any anglesmallerthan
α
without intersetingtheother half.When
A ⊂ A ′
, thenE A ′ ⊂ E A
. In partiular, whenA
ontains1
, thenE A
is a subset ofthe set of (loal) ut points, and therefore theshape of thepath ina neighbourhood of suh a
point is the same as the Brownian frontier in the neighbourhood of a ut-point. This shows in
partiular that (atleast some of) the exponents
ξ(A)
desribe also the Hausdor dimension ofsets ofexeptional points ofthesaling limit ofritialperolation lusters.
In Setion 4, we derive some bounds on the exponents
ξ(A)
for small setsA
in the samespirit as theupperbounds for disonnetion exponents derived in [22 ℄. In partiular, for small
α
, we show that the exponentξ(C α )
is stritly smaller than2
, whih implies the existene ofpivotingpoints(ofsmallangle)ontheplanarBrownianurve. Wethenbrieypresentresultsof
simulations thatsuggestthatthere existpivoting pointsof angle upto an anglelose to
3π/4
.Atually, itis easy to dene other generalized exponentsin a similarfashion, bystudying
non-intersetionpropertiesbetweenBrownianmotionsandsomeoftheirimagesunderisometries
and salings, i.e. one an view
A
as a subset of the linear group. Also, one an onsider non-intersetionpropertiesbetween
B
anditsimagef (B)
byaonformalmap. Itiseasytoseeusingthefuntion
z 7→ z 2
that theexponent desribing the non-intersetionbetweenB
and− B
is infat twie thedisonnetion exponent. The methods of thepresent paper an then be adapted
to suh situations.
Similarly, one ould also extend the denitions to higher dimensions (the ases
d > 4
analso be interesting if the set
A
is suiently large), but onformal invariane an not be usedanymore, sothat some ofthetoolsthat we useinthe present paper do not apply.
Aknowledgments
IthankWendelinWernerforsuggestingmetolookforpivotingpointsontheplanarBrownian
urve andfor neverrefusinghelp andadvie.
Notations
Throughout this paper, we will use the following notations for the asymptoti behaviour of
positive funtions(and sequenes, withthesame meaning):
• f ∼ g
iflim
t →∞
f (t)
g(t) = 1
andf
andg
aresaid to be equivalent;• f ≈ g
iflog f ∼ log g
,i.e.iflim
t →∞
log f (t)
log g(t) = 1
f
andg
arethenlogarithmiallyequivalent;• f ≍ g
iff /g
is bounded and bounded by below, i.e. if there exist two positive niteonstants
c
andC
suh that for allt
,cg(t) 6 f (t) 6 Cg(t)
whih we all strongapproximation of
f
byg
.1 Generalized intersetion exponents
1.1 Denition of the exponents
Proposition and Denition :
Let
A
be a non-empty subset of the omplex plane andB 1
,B 2
be two independent Brownian paths starting uniformlyon theunit irleC (0, 1)
; dene the hittingtimeT R i
ofC (0, R)
byB i
and letτ n i = T exp(n) i
,E n = E n (A) = { B [0,τ 1 1
n ] ∩ AB [0,τ 2 2
n ] = ∅ } , q n (A) = P (E n )
andp R (A) = P (E log R ).
Then,assumingtheexisteneofpositiveonstants
c
andC
suhthatp R (A) > cR − C
,thereexistsareal number
ξ(A)
suhthat, whenR → ∞
,p R (A) ≈ R − ξ(A) .
//
This is a standard sub-multipliativity argument. IfB
is a Brownian pathstartingon
C (0, 1)
withany lawµ
,thenthelawofB τ 1 (B)
ontheirleC (0, e)
hasadensity (relative to the Lebesgue measure)bounded and bounded awayfrom zero
byuniversalonstants (i.e.independently of
µ
). Combining this remark withtheMarkovpropertyat thehittingtimes oftheirle ofradius
e n
showsthat:∀ m, n > 1 q m+n 6 cq n q m − 1 .
Hene the family
(cq n − 1 )
is sub-multipliative and using Proposition 7 we haveq n ≈ e − ξn
,withξ ∈ (0, ∞ )
,aswell asa lowerboundq n > c − 2 e − ξ(n+1)
.//
Remarks: Forsomehoiesof
A
thereisaneasygeometri interpretationoftheeventE n (A)
:ξ( { 1 } )
isthelassialintersetionexponent;ifA = (0, ∞ )
,theE n (A)
istheeventthatthepathsstayindierent wedges.
If
A
is suh thatno lower boundp R (A) > cR − C
holds, we letξ(A) = ∞
. However, inmostoftheresultspresentedhere,wewill restritourselvesto alassof sets
A
forwhih itiseasytoderivesuhlower bounds:
Denition :
A non-empty subset
A
of the omplex plane is said to be nie if it is ontained in theintersetion of an annulus
{ r < | z | < R }
(with0 < r < R < ∞
) with a wedge of anglestritly lessthan
2π
and vertexat0
.Indeed, let
A
be suh a set and letα < 2π
be the angle of a wedge ontainingA
:B 1
andAB 2
will not interset provided eah path remains ina well-hosen wedgeof angle(2π − α)/2
,andthenit isstandard to derive the following bound:
p R (A) > cR − 4π/(2π − α) .
(4)The fat that
A
be ontained in an annulus will be needed in the following proof. The onlyusualasewhere thisdoesnot holdiswhen
A
is awedgeitself;butinthisasea diretstudy ispossible,basedonthe derivationofone exponents in[4 ℄ andtheexatvalueof
ξ
isthenknown(f.next setionfor details).
We will often onsider the ase where
A
is a subset of the unit irle. For suh sets,A
isnieifand only if
A ¯ ∂U
(it isinfateasy to prove thatforA ⊂ ∂U
,ξ(A) = ∞
ifandonly ifA ¯ = ∂U
).1.2 Strong approximation
Thiswhole subsetion will be dediated to the renement of
p R ≈ R − ξ
intop R ≍ R − ξ
. This isnot anedotial, sine thisstrong approximation will be neededon several oasionslater.
Theorem 1 :
For every nie
A
,p R (A) ≍ R − ξ(A)
,i.e. there exist positive onstantsc(A) < C(A)
suhthat
cR − ξ(A) 6 p R (A) 6 CR − ξ(A) .
Moreover,theonstants
c(A)
andC(A)
an betakenuniformlyonaolletionA
ofsubsetsof theplane, provided the elementsof
A
are ontained inthesame nieset.//
Note that sineA ∈ A
is nie, the exponentsξ(A)
exists and is uniformlybounded(for
A ∈ A
). Thesubadditivityargumentshowed thatq n > ce − ξ(A).(n+1)
,whihimpliesreadilythelowerboundinthetheorem. Itismorediultto derive
the upper bound. By Proposition 7, it will be suient to nd a nite onstant
c − (A)
(thatan bebounded uniformlyforA ∈ A
) suh that∀ n, n ′ q n+n ′ > c − q n q n ′ .
(5)In order to make the proof more readable, it is arried out here for a xed
A
;howeverit is easy to see that, at eah step, theonstantsan be taken uniformly
forall
A
ontained insome xednie setA 0
. Moreover,weshallrstassumethatA 0
isa subset ofthe unitirle: We briey indiate at theend of theproof whatarethe fewmodiations neededto adapt it to thegeneral ase.
ThebasimethodisadaptedfromLawler'sprooffornon-intersetionexponents
in[10℄,withsometehnialsimpliationsmadepossibleusingtheabseneofthe
λ
exponent. Themainideaistoobtain aweakindependenebetween thebehaviour
of the paths before and after they reah radius
e n
. The rst step is an estimateonerning the probability that the paths are well separated when they reah
radius
e n
(morepreisely,thattheyremainintwoseparatedwedgesbetweenradiuse n − 1
and radiuse n
):Lemma (Tehnial) :
Let
η > 0
andα < 2π − η
suhthatA
isontained inawedgeof anglelessthan
α
. DeneW α = n
re iθ : r > 0, | θ | < α 2
o , δ n = e − n [d(B τ 1 1
n , AB [0,τ 2 2
n ] ) ∨ d(AB τ 2 2 n , B [0,τ 1 1
n ] )]
andthefollowing events:U n 1 = n B [0,τ 1 1
n ] ∩ {| z | > e n − 1 } ⊂ − W 2π − α − η
o , U n 2 = n
AB [0,τ 2 2
n ] ∩ {| z | > e n − 1 } ⊂ W α o ,
and
U n = U n 1 ∩ U n 2
. Then:∃ c, β > 0 ∀ ε > 0 ∀ r ∈ 3
2 , 3
P (E n+r , U n+r | E n , δ n > ε) > cε β .
///
ThisisandiretonsequeneoflassialestimatesonerningBrownianmotioninwedges; the value of
β
is not important, sonot muh are isneeded in ndingthelowerbound. Notethattheexisteneof
α
requiresthatA
benie.///
If
F n
stands for theσ
-eld generated by both paths up to radiuse n
(so thatfor instane
E n
isinF n
),we now provethatpathsonditionednot to interset upto radius
e n+2
have a good hane to be well separated at this radius, uniformlywithrespetto their behaviourup to radius
e n
:Lemma (End-separation) :
There exists
c > 0
suh that, for everyn > 0
:P (U n+2 | E n+2 , F n ) > c
(i.e. the essential lower bound of
P (U n+2 | E n+2 , F n )
, as anF n
-measurable funtion,isnot lessthanc
).///
The tehnial lemma states that start-separation ours ifthe starting points aresuiently far fromeah other; morepreisely,we havefor allε > 0
:P(U n+2 | E n+2 , F n , δ n > ε) > cε β .
(6)Hene, what is to be proved is that two paths onditioned not to interset have
a positive probability to be far from eah other after a relatively short time. To
prove this fat,one hasto useonditioning on thevalueof
δ n
.Fix
k > 0
, and assumethat2 − (k+1) 6 δ n < 2 − k
; letτ k
be thesmallestr
suhthatone ofthefollowing happens: either
δ n+r > 2 − k
,orE n+r
doesnot hold. It iseasy to usesalingto prove thatfor some
λ > 0
,P (τ k > 2 − k ) 6 2 − λ ,
meaningthatwithpositiveprobability(independentof
k
andn
)thepathsseparateor meet before reahing radius
e n+2 −k
. Hene by the strong Markov property,applying this
k 2
timesleads toP (τ k > k 2 2 − k ) 6 2 − λk 2 .
(7)The tehnial lemmastates that
P (E n+2 | δ n > 2 − (k+1) ) > c2 − βk
: ombining bothestimates thenleads to
P (τ k > k 2 2 − k | E n+2 , δ n > 2 − (k+1) ) 6 c2 βk − λk 2 .
(8)Consider now a generi starting onguration at radius
e n
, satisfyingE n
andhene
δ n > 0
. Fixalsok 0 > 0
and introdue theradiiτ k
(fork 0 6 k < ∞
)denedby
τ k = Inf { r : δ n+r > 2 − k }
(sothat
τ k = 0
aslongas2 − k 6 δ
). Equation (8)an be rewritten(using thefatthatthe tehnial lemmais valid for all
r > 3/2
) asP (τ k − τ k+1 > k 2 2 − k | E n+2 , τ k+1 6 1
2 ) 6 c2 βk − λk 2 .
Fix
k 0
suhthat∞
X
k=k 0
k 2 2 − k < 1
2 ,
andsum thisestimate for
k 0 6 k < ∞
: this leadstoP ( ∀ k > k 0 , τ k − τ k+1 6 k 2 2 − k | E n+2 ) > 1 − c
∞
X
k=k 0
2 βk − λk 2 .
Inpartiular, if
k 0
istakenlarge enough, this probability isgreater than1/2
,andwe obtain
P(τ k 0 6 1
2 | E n+2 ) > 1 2 .
Itis thensuient to ombine thisand Equation(6) to get
P (U n+2 | E n+2 ) > c2 − βk 0 > 0,
andisanbeseenthattheobtainedonstantdoesnotdependontheonguration
at radius
e n
providedE n
issatised.///
Therstonsequene oftheend-separationlemmais
P (E n , U n ) ≍ q n
;but itiseasy to see, using estimates on Brownian motion inwedges again and the strong
Markovproperty,that
P(E n+1 | E n , U n ) > c > 0
(with
c
independent ofn
), and ombining both estimates leads toq n+1 > cq n
,i.e.
q n+1 ≍ q n
. Now ifq ¯ n
stands for the upper bound for the non-intersetion probabilities,namely¯
q n = ∧ Sup
B 1 0 ,B 2 0 ∈U
P(E n | B 0 1 , B 0 2 ),
thepreviousremark onerningthe lawof
W τ 1 (W )
anbe usedtoprove thatq ¯ n 6 cq n − 1
: hene,¯ q n ≍ q n .
Nowthatwe knowthatpaths onditionednot to intersethave agoodhane
to exit a disk at a large distane from eah other, what remains to be proven is
that paths starting from distant points on
C (0, e n )
remain well separated for asuiently long time and beome (in a sense to be speied later) independent
fromtheir behaviour beforeradius
e n
.Lemma (Start-separation) :
Let
α
andη
be asin thetehnial lemma,η ′ = η/2
andα ′ = (2π + α)/2
;introdue
J n 1 = n B [0,τ 1 1
n ] ∩ B (0, 2) ⊂ − W 2π − α ′ − η ′ \ B (0, 1 − η ′ ) o , J n 2 = n
AB [0,τ 2 2
n ] ∩ B (0, 2) ⊂ W α ′ \ B (0, 1 − η ′ ) o ,
and
E ˜ n = E n ∩ J n 1 ∩ J n 2
. Deneq ˜ n
as˜
q n (x, y) = P( ˜ E n | B 0 1 = x, B 0 2 = y).
Then there exists
c > 0
suh that, for alln > 2
and uniformly on all pairs(x, y)
satisfyingU 0
(i.e.suh thatU 0
holdswhenB 0 1 = x
andB 0 2 = y
):˜
q n (x, y) > cq n .
///
Introdue the following (forbidden)sets:K 1 = B (0, e) \ − W 2π − α ′ − η ′
∪ B (0, 1 − η ′ );
K 2 = ( B (0, e) \ W α ′ ) ∪ B (0, 1 − η ′ ).
For all
n
wehaveJ n 1 = { B [0,τ 1 1
n ] ∩ K 1 = ∅ }
andJ n 2 = { AB 2 [0,τ 2
n ] ∩ K 2 = ∅ }
. Fortherest of the proof we shall x
n
, and ondition thepaths by their starting points;introdue thefollowing stopping times (forpositive values of
k
):T 0 1 = Inf { t > 0 : B [0,t] 1 ∩ C (0, 3) 6 = ∅ } , S k 1 = Inf { t > T k 1 − 1 : B [T 1 1
k − 1 ,t] ∩ K 1 6 = ∅ } , T k 1 = Inf { t > S k 1 : B [S 1 1
k ,t] ∩ C (0, 3) 6 = ∅ } ,
and
S k 2
,T k 2
similarly, replaing allourrenes ofB 1
byAB 2
andK 1
byK 2
. Weshall also use the notation
N i
for the number of rossings byB 1
(resp.AB 2
)between
K i
andC (0, 3)
,dened asN i = Max { k : S k i < τ n i } .
Withthose notations,
J n i = J 1 i ∩ { N n i = 0 }
anda.s.N i < ∞
. Moreover, uniformlyon the startingpointsonsidered here(satisfying theondition
U 0
),we haveJ 1 i >
c > 0
bythe tehnial lemma, wherec
dependsonly onη
.First,wesplittheevent
E n
aordingtothevalueof,say,N 2
: wewriteP (E n ) = P ∞
k=0 P (E n , N 2 = k)
. By the Beurling estimate, on{ N > k }
, the probability thatB [0,τ 1 1
n ]
andAB [S 2 2
k ,T k 2 ]
do not interset is bounded by some universal onstantλ < 1
(whih an even be hosen independent ofA
), independently ofB 1
andthe two remaining partsof
B 2
. By the strong Markov propertyat timeT k 2
,whenN 2 = k
the probability thatAB 2
afterT k 2
does not intersetB 1
is bounded byP (B 1 ∩ AB [T 2 2
0 ,τ n 2 ] = ∅, N 2 = 0)
(i.e.thepath afterT k 2
whenN 2 = k
isthesameasthe entirepath when
N 2 = 0
). Introduing thosetwoestimateinthesumleadstoP (E n ) 6
∞
X
k=0
λ k P (E n , N 2 = 0) = 1
1 − λ P (E n , N 2 = 0).
Doing this deomposition again aordingto
N 1
(with the same onstantλ < 1
)we thenobtain
P (E n ) 6 1
(1 − λ) 2 P (E n , N 1 = N 2 = 0),
i.e.
P(N 1 = N 2 = 0 | E n ) > (1 − λ) 2 > 0
. This, and the previous remark thatP (J n i | N i = 0)
isboundedbybelowbyaonstantprovidedthatthestartingpointssatisfy
U 0
,gives:P ( ˜ E n | B 0 1 = x, B 0 2 = y) > cP (E n | B 0 1 = x, B 0 2 = y).
(9)Conditioning on
B 2
shows thatthemapf : x 7→ P (E n | B 0 1 = x, B 0 2 = 1)
(10)isharmonianddoesnotvanishontheomplement of
A
. Moreover,itssupremumontheunitirleisequal to
q ¯ n
bydenition: ApplyingtheHarnakpriniple thenproves that
f
is bounded by below bycq n
on the set ofx
satisfyingU 0
, whihompletes the proof.
///
Anotherestimateanbeobtainedusingtheverysameproof: Onlykeepingthe
onditionsinvolvingdisksand relaxing those involvingwedges, we obtain
P B [0,τ 1 1
n ] ∩ B(0, 1 − η) = ∅, AB [0,τ 2 2
n ] ∩ B (0, 1 − η) = ∅
B 0 1 , B 0 2 , E n
> c > 0,
(11)where
c
does not depend on the initial positionsB 0 1
andB 0 2
, nor onn
(it learlydepends on
η
, though, and a loser look at the proof shows that we an ensurec > η β
asη → 0
,for someβ > 0
). Thisestimate will be needed inthederivationofHausdordimensions, f.Setion 3.
We know have all theneeded estimates to derive the lower bound in thesub-
additivityondition,andhenetheonlusionofthetheorem. Taketwopathswith
independent starting points uniformly distributed on theunit irle and killed at
radius
e m+n
, onditioned not to interset between radii1
ande n
. This happenswithprobability
q n
. Withlarge probability(i.e. witha positiveprobability, inde- pendentofm
andn
)thepathsuptoradiuse n
endupwellseparated inthesenseof the end-separation lemma. In partiular, the points where they reah radius
e n
, after suitable resaling, satisfy the hypothesis of the start-separation lemma:Hene with probability greater that
cq m
, the paths between radiie n
ande m+n
remainseparatedupto radius
e n+1
,donotreahradius(1 − η)e n
anymoreanddonot intersetup to radius
e m+n
. Underthoseonditions, itiseasy to seethat thepaths do not meet at all. So
q m+n > cq m q n
for some positivec
, and we get theonlusion.
Someadaptations areneededif
A
isinluded inanannulus, say{ r < | z | < R }
with
r < 1 < R
. First, replae all ourrenes ofe
bye 0
, withe 0
hosen largerthan
10R/r
,andin thestart-separationlemma, replaeB (0, 1 − η)
byB (0, r/2R)
inthe denitionof the
J n
. As long asr
andR
are xed, this hanges nothing totheproof,exeptthat the onstantswe obtainwill then depend on
R/r
whihitselfisbounded provided
A
remainsa subset ofsome xednie set.A more serious problemarises iftheomplement of
A
is not onneted, sinethe natural domain of the funtion
f
(as dened by Equation (10)) is itself notonneted. However, sine
A
is nie, its omplement has exatly one unboundedomponent, and it is easy to see that if
x
is not in this omponent thenf(x)
vanishes for
n > 1
. Hene, nothing hanges (as far as non-intersetion properties areonerned)whenA
isreplaedbytheomplement oftheinniteomponentofitsomplement (i.e.when llingtheholes in
A
).//
In fat, a stronger result an be derived: If the starting points
B 0 1
andB 0 2
are xed, thenP (E n | B 0 1 , B 0 2 )
is equivalent toce − nξ(A)
, wherec
is a funtion ofB 0 1
andB 0 2
satisfyingc 6
c 0 d(B 0 1 , AB 0 2 ) β
. This estimate is related to a strong onvergene result on the law of pathsonditioned by
B 1 ∩ AB 2 = ∅
. However, proving this result would be muh more involved(f.[17℄ for the proofinthease
A = { 1 }
).2 Properties of the funtion
A 7→ ξ(A)
We rstlista fewsimple properties ofthefuntion
A 7→ ξ(A)
. Forp ∈ Z
andA ⊂ C
,introdueA p = { z p , z ∈ A }
andletA ∗ = { z, z ¯ ∈ A }
.Proposition 1 :
Is these statements, all setsare assumedto be non-emptybut do not need to be nie:
(i).
ξ
is non-dereasing: ifA ⊂ A ′
thenξ(A) 6 ξ(A ′ )
;(ii).
ξ
is homogeneous: ifλ ∈ C ∗
thenξ(λA) = ξ(A)
;(iii).
ξ
is symmetri:ξ(A − 1 ) = ξ(A ∗ ) = ξ(A);
(iv).
ξ
hasthe following property: ifn > 1
thenξ [
e 2ikπ/n A
= nξ(A n ).
//
(i): Thisisa trivialonsequene ofp R (A) > p R (A ′ )
.(ii): Applying the saling property with fator
| λ |
toB 2
proves that one ansuppose
| λ | = 1
; in whih ase we havep R (A) = p R (λA)
(beause the startingpointsareuniformlydistributed ontheunit irle).
(iii): Simplyexhange
B 1
andB 2
forA − 1
,andsaythattheomplex onjugateof aBrownian path isstill aBrownian path to get
A ∗
.(iv): This is a onsequene of the analytiity of the mapping
z 7→ z n
(henethe fat that
((W t ) n )
is a Brownian path ifW
is one) together with the remarkthattheexisteneof
s
,t > 0
andz ∈ A n
with(B s 1 ) n = z(B t 2 ) n
isequivalent totheexisteneof
z ′
inS
e 2ikπ/n A
withB t 1 = z ′ B t 2
notethatthemappingalsohasaninuene on
R
,henethefatorn
.//
We now turn our attention toward regularity properties of the funtion
A 7→ ξ(A)
thefollowing resultbeingakeysteptowardsthederivation ofdimensionsinthenextsetion. Intro-
due theHausdordistane between ompat subsetsof theplane (f.Setion 5 for details). It
willbeonvenient hereto deneneighbourhoods by
V r (A) = { xe z , x ∈ A, | z | < r }
insteadoftheusual
A + B (0, r)
leadingto the logarithmi Hausdordistane. The(logarithmi)Hausdor topologyis the metritopology derivedfrom thisdistane.Proposition 2 :
ξ
is ontinuous on the olletion of nie sets, endowed withthe logarithmial Hausdor topology. For anyniesetA 0
,ξ
is uniformlyontinuousin{ A : A ⊂ A 0 }
.//
TheproofreliesontheuniformityofthestrongapproximationinTheorem1: x anie setA 0
andassumeall setsonsideredherearesubsetsofA 0
. Theonstantsc
,c −
andc +
appearing duringthe proof mayonly depend onA 0
.First,x
R > 1
andonditionalleventsbyB [0,T 2 2
R+1 ]
i.e.xtheseondpath.For all
A ⊂ A 0
,letd R (A) = d(B 1 [0,T 1
R ] , AB [0,T 2 2 R ] ) ;
for all
ε > 0
introdue the stopping timeS ε = Inf { t : d(B 1 t , AB [0,T 2 2
R ] ) < ε } .
Notethat
{ d R (A) < ε } = { S ε < T R 1 }
. Onthis event, thestrong Markovpropertyshowsthat
B S 1 ε + ·
isa Brownian path startingε
-losetoAB 2
. By Beurling'stheo-rem,the probabilitythattheydo notmeet beforeradius
R + 1
issmallerthan theorrespondingprobabilityfor a path nearahalfline; hene,
P(B [S 1
ε ,T R+1 1 ] ∩ AB [0,T 2 2
R+1 ] = ∅ | d R (A) < ε) 6 √ ε,
sothat, onsidering the wholepath,
P (E R+1 | d R (A) < ε) 6 √
ε
. Apply theBayesformula:
P (d R (A) < ε | E R+1 ) = P (d R (A) < ε)
P (E R+1 ) P (E R+1 | d R (A) < ε);
sineweknowthat
P (E R+1 ) > c − (R + 1) − ξ(A)
withξ(A) 6 ξ(A 0 )
wenallyobtainP(d R (A) < ε | E R+1 ) 6 cR ξ(A 0 ) √
ε.
Fromnowon,weshallassumethat
ε
issuientlysmall tomaketheobtainedbound smallerthat
1
. Taking the omplement leads toP(d R (A) > ε | E R+1 ) > 1 − cR ξ(A 0 ) √
ε.
Now, remark that when
d R (A) > ε
andd H (A, A ′ ) < ε/R
, we haveB [0,T 1 1 R ] ∩ A ′ B [0,T 2 2
R ] = ∅
: fromthis and the previous equationfollowsthat, aslongasA
andA ′
remain subsetsofA 0
,d H (A, A ′ ) < ε
R ⇒ p R (A ′ ) >
1 − cR ξ(A 0 ) √ ε
p R+1 (A).
We an apply the estimates on
p R
we derived in Theorem 1 i.e.p R (A) ≍ p R+1 (A) ≍ R − ξ(A)
: stillford H (A, A ′ ) < ε/R
andA
,A ′
insideA 0
wegetc + R − ξ(A ′ ) >
1 − cR ξ(A 0 ) √ ε
c − R − ξ(A) ,
andtakingthe logarithm ofeah sideof the inequality leadsto
log c + − ξ(A ′ ) log R > log c − + log
1 − cR ξ(A 0 ) √ ε
− ξ(A) log R,
heneaftersuitable transformations:
ξ(A ′ ) 6 ξ(A) + c
log R − log 1 − cR ξ(A 0 ) √ ε
log R .
(12)Fix
η > 0
,and hooseR
suhthatc/ log R < η/2
. Itisthenpossibleto takeε
suiently small sothat
| log(1 − cR ξ(A 0 ) √
ε) | < (η log R)/2
;ford H (A, A ′ ) < ε/R
we then have
ξ(A ′ ) 6 ξ(A) + η
, hene by symmetry| ξ(A ′ ) − ξ(A) | 6 η
. Thisproves that
ξ
is uniformly ontinuous onP c (A 0 )
, for allA 0
, heneontinuous onthe family ofnie sets.
//
Remark 1: Equation (12) allows the derivation of an expliit modulus of ontinuity for
ξ
inside
A 0
,ofthe form| ξ(A ′ ) − ξ(A) | 6 C(A 0 )
| log d H (A, A ′ ) |
(take
R = d − 1/2ξ(A 0 )
). But sineC(A 0 )
is not known, this does not provide numerial boundsfor
ξ
.Remark2: Insideanieset,theusualandlogarithmiHausdortopologiesareequivalent,so
theintrodutionofexponentialneighbourhoods inProposition2anseemartiial;however,it
leadsto onstants thatdonot varywhen
A
is multipliedbysome onstant (asinProposition1, point (ii)), hene uniform ontinuity holds on the olletion of nie sets ontained in a xedwedgeand insomeannulus
{ r < | z | < cr }
for xedc
whihis wrongfor theusual Hausdortopology,asaonsequene ofthe homogeneityof
ξ
applied for small| λ |
.Note that uniform ontinuity annot hold on the family of nie sets ontained in a given
annulussine
ξ
wouldthenbebounded(byaompaityargument),whihitisnot: theexponentassoiated to airle isinnite.
3 Hausdor dimension of the orresponding subsets of the path
3.1 Conformally invariant subsets of the Brownian path
It is well-known that the Brownian path is invariant in law under onformal transformations;
inthis setion, we study subsetsof theBrownian urve thatarealso invariant underonformal
maps. Arst example istheset of so-alled Brownian ut-points, i.e.points
B t
suhthatB [0,t)
and
B (t,1]
aredisjoint;thesepointsformasetofHausdordimension2 − ξ( { 1 } ) = 3/4
. Relatedtothoseareloal ut-points,i.e.pointssuhthatthereexists
ε > 0
satisfyingB [t − ε,t) ∩ B (t,t+ε] = ∅
thedimensionisthesame asforglobalut-points. Other examplesaregivenbyLawlerin[9℄:
inpartiularthe setofpioneer points (suhthat
B t
liesonthefrontieroftheinniteomponentofthe omplement of
B [0,t]
), relatedto the disonnexionexponentη 1
;frontier points (points oftheboundaryoftheinniteomponentoftheomplementof
B [0,1]
),relatedtothedisonnetion exponent for two paths inthe plane. Another exeptional subset of thepath is theset of onepoints (suh that
B [0,t]
is ontained in a one of endpointB t
), related to the one exponents(studiedin[18 ℄ forexample).
We will usethe exponent introdued inthe previous setions to desribe a family of exep-
tionalsets, indexed by a subset
A
of the omplex plane, having dimension2 − ξ(A)
, and thatareinvariant under onformaltransformations, as follows. Fixa Brownian path
B [0,1]
, asubsetA
of the omplex plane,and introdue thefollowing times for allt ∈ (0, 1)
andr > 0
:T r (t) = Inf { s > t : | B s − B t | = r } , S r (t) = Sup { s < t : | B s − B t | = r } .
Denition :
If
0 < ε < R
andt ∈ (0, 1)
,letZ t [ε,R] (B) =
B s − B t B s ′ − B t
: s ∈ [T ε (t), T R (t)], s ′ ∈ [S R (t), S ε (t)]
;
and introdue
E A [ε,R] = { B t : Z t [ε,R] ∩ A = ∅ }
. Then, lettingε
go to0
:Z t R = [ ↑
ε>0
Z t [ε,R] , Z t = \ ↓
R>0
Z t R , Z ˜ t = \ ↓
R>0
Z t R ;
dene
E A R
,E A
andE ˜ A
aordingly.Weshallalsousethenotation
T A = { t : B t ∈ E A }
,for thesetofA
-exeptional times,andT ˜ A = { t : B t ∈ E ˜ A }
,for thesetofA
-strongly exeptional times.Note that, sine
0
is polar for planar Brownian motion,Z
is well-dened for almost anyt
.For
A = { 1 }
,E A
isthe set ofloal ut-points; more generally,t
is inE A
if,and only if,for someε > 0
,we have(B (t,t+ε] − B t ) ∩ A.(B [t − ε,t) − B t ) = ∅,
so the setup looks similar to the denition of the exponent
ξ(A)
. It is easy to see that for allxed
t > 0
,a.s.Z t = C ∗
andZ ˜ t = C
,sothatforA 6 = ∅
,P (t ∈ T A ) = 0
,leading toE(µ( T A )) = 0
i.e.
µ( T A ) = 0
almost surely henethe term exeptional points.Theset
E A
ofA
-exeptional points is generallynot onformallyinvariant. However, itis the ase forstrongly exeptional points:Proposition 3 :
Let
Φ
be a onformal map on a neighbourhoodΩ
of0
, withΦ(0) = 0
, and letB Ω
beB
stopped at its rst hitting of∂Ω
. By onformal invariane of planar Brownian motion,Φ(B Ω )
isaBrownian path stopped at its rsthitting of∂Φ(Ω)
. Moreover, we haveE ˜ A (Φ(B Ω )) = Φ( ˜ E A (B Ω )).
//
We prove thatZ ˜
isinvariant. Itis suient to prove thefollowing harateri-zation:
z ∈ Z ˜ t (B) ⇐⇒ ∃ (s n ) ↓ 0, (s ′ n ) ↓ 0 : B t+s n − B t B t − s ′ n − B t → z,
asonformalmapsonservethe limitsofsuhquotients. Suhasequene iseasily
onstrutedusing thevery denitionof
Z ˜
.//
Note that nothing in the preeding uses the fat that
B
be a Brownian path, exept forthe remark about
P (t ∈ T A )
. The remaining of the present setion is dediated to derivingthe Hausdor dimension of
E A
andE ˜ A
. It will be more onvenient to work in the time set, sointrodue
T A [ε,R] = { t ∈ [0, 1] : B [t 1 − R,t − ε] ∩ A.B 2 [t+ε,t+R] = ∅} .
ThesalingpropertyofBrownianmotionanthenbeusedto show, asin[8,lemmas3.143.16℄,
thatTheorem 1implies the following: