MULT[FRACTAL ANALYSIS FOR POINTWISE
HÖLDER
EXPONENTS OF THE COMPLEX TAKAGI FUNCTIONS IN RANDOM COMPLEX DYNAMICSJOHANNES JAERISCH AND HIROKI SUMI
ABSTRACT. We considerhyperbolicrandomcomplex dynamicalsystemsonthe Riemannspherewith sep‐
aratingcondition andmultiplermnimalsets. Weinvestigatethe Hölderregularityofthe function T of the
probabilityoftendingtooneminimal set, thepartialderivatives of T withrespecttotheprobabilityparameters,
whichcanberegardedascomplexanaloguesof theTakagifunction,and thehigher partialderivatives C of T.
Our main resultgivesadynamical descriptionof thepointwiseHölderexponentsof Tand C,which allowsus
todeterminethespectrumofpointwiseHölderexponentsby employingthe multifractal formalism inergodic theory. Also,weprove that the bottom of thespectrum$\alpha$_{-}isstrictlyless than1,which allowsustoshow that theaveragedsystemactschaoticallyonthe Banach spaceC^{ $\alpha$}of $\alpha$‐Hölder continuous functions for every $\alpha$\in($\alpha$_{-}, 1),thoughtheaveragedsystembehaves verymildly (e.g. wehavespectral gaps)onC^{ $\beta$}for small
$\beta$>0.
1. MAINRESULTS
Thisnoteisthe summary of the results from paper
[JS16].
Wedonotgive
anyproofs
of them in thisnote. Forthe
proofs
of theresults, see[JS16].
Inthis paper, weconsider randomdynamical
systemsof rational mapsonthe Riemann
sphere
\hat{\mathbb{C}}
:=\mathbb{C}\cup\{\infty\}\cong S^{2}
. Thestudy
ofrandomcomplex
dynamics
wasinitiated
by
J.E. Fomaess and N.Sibony
([FS91]).
There aremanynewinteresting
phenomena
in randomdynamical
systems,socalled randomness‐inducedphenomena
ornoise‐inducedphenomena,
which cannothold in the deterministic iterationdynamics.
For the motivations andrecentresearch of randomcomplex
dynamical
systemsfocusedontherandomness‐inducedphenomena,
seethe second authors works[Sumlla,
Sum13,\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{i}\mathrm{l}5\mathrm{a},
\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{l}5\mathrm{b}]. Inthese papers itwasshown that forageneric
randomdynamical
systemof
complex
polynomials,
thesystemactsverymildly
onthe space of continuous functionson\hat{\mathbb{C}}
and onthe spaceC^{ $\alpha$}(\hat{\mathbb{C}})
for small$\alpha$\in(0,1)
,whereC^{ $\alpha$}(\hat{\mathbb{C}})
denotes the Banach space of $\alpha$‐Hölder continuousfunctionson
\hat{\mathbb{C}}
endowed with $\alpha$‐Höldernonn,but under certain conditions thesystemstillactschaoticaly
onthe spaceC^{ $\beta$}(\hat{\mathbb{C}})
forsome$\beta$\in(0,1)
closeto 1. Thus, weinvestigate
thegradation
between chaos and order in random(complex)
dynamical
systems.In ordertoshow the main ideas of the paper, let Rat denote thesetof allnon‐constantrational mapson
\hat{\mathbb{C}}
.This isasemigroup
whosesemigroup operation
is thecomposition
of maps.Throughout
the paper, lets\geq 1and let
(fl,
\cdots,f_{s+1}
)
\in(\mathrm{R}\mathrm{a}\mathrm{t})^{s+1}
with\deg(f_{i})\geq 2,i=1
, s+1.Let\mathrm{p}=(p_{1}, \ldots,p_{s})\in(0,1)^{s}
with\displaystyle \sum_{i=1}^{s}p_{i}<1
and letp_{s+1}:=1-\displaystyle \sum_{i=1}^{s}p_{i}
. We consider the(i.i.
\mathrm{d}.)
randomdynamical
system
on\hat{\mathbb{C}}
such thatateverystepwechoose
f_{ $\iota$}
withprobability
p_{i}.This definesaMarkov chain withstatespace\hat{\mathbb{C}}
such thatfor each
x\in\hat{\mathbb{C}}
and for each Borel measurable subset A of\hat{\mathbb{C}}
,the transitionprobability
p(x,A)
fromxtoADate:8thMay2016. MSC2010:37\mathrm{H}10,37\mathrm{F}15.
Keywordsandphrases. Complex dynamicalsystems,rationalsemigroups,randomcomplex dynamics,multifractalformalism,Julia
set, random iteration.
Johannes Jaerisch
DepartmentofMathematics, Facultyof Science andEngineering,ShimaneUniversity,Nishikawatsu 1060Matsue,Shimane 690‐
8504,JapanE‐‐mail:jaerisch@nko.shimane‐u.ac.jpWeb:http://www.math.shimane‐u.ac.jp/\simjaerisch/
Hiroki Sumi(corresponding author)
DepartmentofMathematics,Graduate School ofScience,OsakaUniversity,1‐1Machikaneyama, Toyonaka,Osaka, 560‐0043,Japan
is
equal
to\displaystyle \sum_{i=1}^{s+1}p_{i}1_{A}(f_{i}(x))
,where1_{A}
denotes the characteristic function of A. LetG=(f_{1},\ldots,f_{s},f_{s+1})
be the rationalsemigroup
(i.e.,
subsemigroup
ofRat) generated Uy
\{f_{1}, f_{s+1}\}
. Moreprecisely,
G={f_{0\}_{l}}\circ\cdots\circ f_{$\varpi$_{1}}
:n\in \mathbb{N},
t)_{1},\cdots,$\omega$_{n}\in\{1
,. s+1 We denoteby
F(G)
the maximal open subset of\hat{\mathbb{C}}
on which G isequicontinuous
withrespecttothespherical
distanceon\hat{\mathbb{C}}
.ThesetF(G)
iscalled the Fatousetof G,and theset
J(G)
:=\hat{\mathbb{C}}\backslash F(G)
iscalled the Juliasetof G. We remark that in ordertoinvestigate
randomcomplex
dynamical
systems,it is veryimportant
toinvestigate
thedynamics
of associated rationalsemigroups.
The firststudy
ofdynamics
of rationalsemigroups
wasconductedby
A. Hinkkanenand G.J. Margin
([HM96]),
whowereinterested in the role ofpolynomial
semigroups
(i.e.,
semigroups
ofnon‐constant
polynomial
maps)
whilestudying
variousone‐complex‐dimensional
moduli spaces for discrete groups, andby
F. Rens group([GR96]),
who studied suchsemigroups
from theperspective
of randomdynamical
systems. For theinterplay
of randomcomplex dynamics
anddynamics
of rationalsemigroups,
see
[\mathrm{S}\mathrm{u}\mathrm{m}00]-[\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{l}5\mathrm{b}]
,[SSII
,SU13, \mathrm{J}\mathrm{S}15\mathrm{a},\mathrm{J}\mathrm{S}15\mathrm{b}].
Throughout
the paper,we assumethefollowing.
(1)
G ishyperbolic,
i.e.,wehaveP(G)\subset F(G)
,whereP(\mathrm{G})
:=\overline{\cup g(\bigcup_{i=1}^{s+1}\{\mathrm{c}\mathrm{n}\mathrm{t}\mathrm{i}\mathrm{c}\mathrm{a}1}
values off_{i}:\hat{\mathbb{C}}\rightarrow\hat{\mathbb{C}}
}).
Here,the closure is taken in\hat{\mathbb{C}}.
g\in G\cup\{\mathrm{i}\mathrm{d}\}
(2)
(fl,
\cdots,f_{s+1}
)
satisfies theseparaungcondition,
i.e.,f_{i}^{-1}(J(G))\cap f_{j}^{-1}(J(G))=\emptyset
wheneveri,j\in
\{1, s+1\},i\neq j.
(3)
There existatleasttwominimalsetsof G. Here,anon‐empty compactsubsetKof\hat{\mathbb{C}}
is calleda minimalsetof G ifK=\overline{\bigcup_{g\in G}\{g(z)\}}
for eachz\in K.Note that
by
assumption (2), [Sum97,
Lemma 1,1.4]
and[Sumlla,
Theorem3.15],
wehave that there existat most
finitely
many minimalsetsof G. Moreover,denoting by
S_{G}the union of minimalsetsof GandsemngI
:=\{1, s+1\}
,wehave that for eachz\in\hat{\mathbb{C}}
there existsaBorel subsetA_{z}
ofI^{\mathrm{N}}
with\tilde{p}_{\mathrm{p}}(A_{z})=1
such that
d(f_{\mathfrak{B}_{l}}\cdots f_{0)_{1}}(z),S_{G})\rightarrow 0
as n\rightarrow\inftyfor alla=($\omega$_{\dot{7}})_{i=1}^{\infty}\in A_{z}
,where\overline{p}_{\mathrm{p}} :=\otimes_{n=1}^{\infty}p_{\mathrm{p}}
denotes theproduct
measure onI^{\mathrm{N}}
given
by
p_{\mathrm{p}}:=\displaystyle \sum_{i=1}^{s+1}p_{i}$\delta$_{7}
.with$\delta$_{i} denoting
the Diracmeasureconcentratedati\in I.Throughout,
wefix aminimalsetLof G(e.g.
L=\{\infty\}
when G is apolynomial
semigroup).
Denoteby
T_{\mathrm{p}}(z)
theprobability
oftending
toLof the processon\hat{\mathbb{C}}
which startsinz\in\hat{\mathbb{C}}
and which isgiven
by drawing independently
withprobability
p_{i}the mapf_{i}
. Moreprecisely,
T_{\mathrm{p}}(\mathrm{z}) :=\tilde{p}_{\mathrm{p}}(\{ $\omega$=($\omega$_{i})_{i=1}^{\infty}\in
I^{\mathrm{N}}:d(f_{$\omega$_{n}}\circ\cdots\circ f_{0J_{1}}(z),L)\rightarrow 0
as n\rightarrow\infty Itwasshownby
the second author in[Sum13] that,
foreach\mathrm{p}=(p_{1}, \ldots,p_{s})
there exists$\alpha$\in(0,1)
such that\mathrm{x}=(x_{1}, \ldots,x_{s})\mapsto T_{(xx_{S},1-$\Sigma$_{i=1^{X_{i)}}}^{s}}1,\ldots,\in C^{ $\alpha$}(\hat{\mathbb{C}})
isreal‐analytic
ina
neighbourhood
of \mathrm{p}, whereC^{ $\alpha$}(\hat{\mathbb{C}})
denoted
the \mathbb{C}‐Banach space of $\alpha$‐Hölder continuous \mathbb{C}‐valuedfunctionson
\hat{\mathbb{C}}
endowed with $\alpha$‐Höldernorm\Vert\cdot\Vert_{ $\alpha$}
(Remark 1.17).
Thus it is very natural andimportant
toconsider the
following.
For\mathrm{N}_{0}
:=\mathbb{N}\cup\{0\}
and\mathrm{n}=(n_{1}, \ldots,n_{S})\in \mathbb{N}_{0}^{s}
wedenoteby
C_{\mathrm{n}}\in C^{ $\alpha$}(\hat{\mathbb{C}})
thehigher
orderpartial
derivative ofT_{\mathrm{p}}
oforder|\displaystyle \mathrm{n}|:=\sum_{i=1}^{s}n_{i}
withrespecttotheprobability
parametersgiven
Uy
\displaystyle \mathcal{C}_{\mathrm{n}}(z):=\frac{\partial^{|\mathrm{n}|}T_{1}(x\ldots jx_{s},1-.$\Sigma$_{i--1}^{s}x_{i})(z)}{\partial x_{1^{1}'}^{n}\partial x_{2^{2}}^{n}\cdot\cdot\partial x_{s}^{n_{s}}}|_{\mathrm{x}=\mathrm{p}}, z\in\hat{\mathbb{C}}.
These functionsareintroduced in
[Sum13]
by
the second author. We introduce the \mathbb{C}‐vectorspace\mathscr{C}:= span
\{\mathrm{C}_{\mathrm{n}}|\mathrm{n}\in \mathrm{N}_{0}^{s}\}\subset C^{a}(\hat{\mathbb{C}})
,which consists of all the finite
complex
linear combinations of elements from\{C_{\mathrm{n}}|\mathrm{n}\in \mathrm{N}_{0}^{s}\}
.The first order derivativesarecalledcomplex analogues
oftheTakagi
function in[Sum13].
Note thatC_{0}=T_{\mathrm{p}}.
ForanelementC\in \mathscr{C}and
z\in\hat{\mathbb{C}}
the HolderexponentHöl(C,z)
isgiven
by
Höl
(C,z)
:=\displaystyle \sup\{ $\alpha$\in[0,\infty)
:where d denotes the
spherical
distanceon\hat{\mathbb{C}}
.Itwasshown in[\mathrm{J}\mathrm{S}15\mathrm{a}]
that the levelsetsH(C_{0}, $\alpha$)
:={
z\in\hat{\mathbb{C}}
:Höl(Co, z)= $\alpha$},
$\alpha$\in \mathbb{R},
satisfy
the multifractal formalism. Inparticular,
there existsaninterval ofparameters($\alpha$_{-}, $\alpha$_{+})
such thatthe Hausdorff dimension of
H(C_{0}, $\alpha$)
ispositive
and varies realanalytically
(see
Theorem 1.2below).
The first main result of this papergives
adynamical
description
ofthepointwise
Hölderexponentsforanarbitrary
C\in \mathscr{C}.Wesay thatC=\displaystyle \sum_{\mathrm{n}\in \mathrm{N}_{0}^{s}}$\beta$_{\mathrm{n}}C_{\mathrm{n}}\in \mathscr{C}
is non‐trivial if there exists\mathrm{n}\in \mathrm{N}_{0}^{s}
with$\beta$_{\mathrm{n}}\neq 0
.Ittumsoutin Theorem 1.1 below that every non‐trivialC\in \mathscr{C}has thesame
pointwise
Hölderexponents.Tostate theresult,wedefine the skewproduct
map(associated
with(f_{i})_{i\in}
) (see [SumOO])
\tilde{f}:I^{\mathrm{N}}\times\hat{\mathbb{C}}\rightarrow I^{\mathrm{N}}\times\hat{\mathbb{C}}, \tilde{f}( $\omega$,z):=( $\sigma$( $\omega$),f_{0)\mathrm{l}}(z))
,where
$\sigma$:I^{\mathrm{N}}\rightarrow I^{\mathrm{N}}
denotes the shift mapgiven by
$\sigma$($\omega$_{1},
$\varpi$_{2},:=($\omega$_{2},
$\omega$_{3}, for$\omega$=(\mathrm{t}0_{1},
w,\in I^{\mathrm{N}}.
For every
$\omega$=($\omega$_{j})_{j\in \mathbb{N}}\in I^{\mathrm{N}}
andn\in \mathrm{N},letf_{w|_{n}}
:=f_{$\omega$_{n}}\circ
\circ f_{$\omega$_{1}}
andwedenoteby
F_{ $\varpi$}the maximal opensubset of
\hat{\mathbb{C}}
onwhich\{f_{c\mathrm{o}|_{n}}\}_{n\in \mathrm{N}}
isequicontinuous
withrespecttod.LetJ_{ $\omega$}
:=\hat{\mathbb{C}}\backslash F_{a\mathrm{J}}
.The Juliasetof\overline{f}
isgiven
by
J(f $\gamma$=\displaystyle \bigcup_{ $\omega$\in I^{\mathrm{N}}}\{ $\omega$\}\times J_{0)}
where the closure is taken inI^{\mathrm{N}}\times\hat{\mathbb{C}}
.Notethatdenoting
by
$\pi$:I^{\mathrm{N}}\times\hat{\mathbb{C}}\rightarrow\hat{\mathbb{C}}
the canonicalprojection,
$\pi$:J(\tilde{f})\rightarrow J(G)
isahomeomorphism
([Sumlla,
Lemma4.5], [Sum97,
Lemma1.1.4]
andassumption (2))
and$\pi$\circ\overline{f}= $\sigma$\circ $\pi$
.Weintroduce thepotentials \overline{ $\varphi$},
\overline{ $\psi$}:J(\tilde{f})\rightarrow \mathbb{R}
given
by
\tilde{ $\varphi$}( $\omega$,z):=-\log\Vert f_{0 $\eta$}'(z)\Vert
, $\iota$ỹ(oJ,
z) :=\log p_{$\omega$_{1}},
where
\Vert
.|
denotes thenormofthe derivative withrespecttothespherical
metricon\hat{\mathbb{C}}
.Notethat\tilde{f}^{-1}(J(\tilde{f}))=
J(\tilde{f})=\tilde{f}(J(\overline{f}))
([Sum00]).
We denoteby
S_{n}theergodic
sumof thedynamical
system(J(\hat{f}),f $\gamma$.
Theorem 1.1. Foreverynon‐trivial
C=\displaystyle \sum_{\mathrm{n}\in \mathrm{N}_{0}^{s}}$\beta$_{\mathrm{n}}C_{\mathrm{n}}\in \mathscr{C}
wehave(1.1)
Höl(C,z)
=\displaystyle \lim_{k\rightarrow}\inf_{\infty}\frac{S_{k}\tilde{ $\psi$}( $\omega$,z)}{S_{k}\overline{ $\varphi$}( $\omega$,z)}
,forall
( $\omega$,z)\in J(\tilde{f})
.Combining
Theorem 1.1 withourresults from[\mathrm{J}\mathrm{S}15\mathrm{a},Theorem1.2]
onthe multifractalformalism, weestabhsh the multifractal formalism for the
pointwise
Hölderexponentsofanarbitrary
non‐trivialC\in \mathscr{C}.Tostatetheresults,for any non‐trivialC\in(\mathscr{E}and $\alpha$\in \mathbb{R}wedenote
by
H(C, $\alpha$):= {
y\in\hat{\mathbb{C}}
:Höl(C,y)= $\alpha$}
the levelsetof
prescribed
Hölderexponent $\alpha$.The range of the multifractalspectrumisgiven
by
$\alpha$_{-}:=\displaystyle \inf\{ $\alpha$\in \mathbb{R}:H(C, $\alpha$)\neq\emptyset\}\in \mathbb{R}
and$\alpha$_{+}:=\displaystyle \sup\{ $\alpha$\in \mathbb{R}:H(C, $\alpha$)\neq\emptyset\}\in \mathbb{R}.
By
Theorem 1.1,thesetsH(C, $\alpha$)
coincide for all non‐trivialC\in \mathscr{C}. Thus, $\alpha$_{-}and $\alpha$_{+} donotdepend
on the choice ofanon‐trivialC\in \mathscr{C}.Also,$\alpha$_{-}>0([Sum98,
Theorem2.6],seealsoCorollary
1.11).
Theorem 1.2. Allof
thefollowing
hold.(1)
LetC\in \mathscr{C}be non‐trivial.If $\alpha$_{-}<$\alpha$_{+}then
theHausdolffdimensionfunction
$\alpha$\mapsto\dim_{H}(H(C, $\alpha$))
,$\alpha$\in($\alpha$_{-}, $\alpha$_{+})
,defines
arealanalytic
andstrictly
concavepositivefunction
on($\alpha$_{-}, $\alpha$_{+})
withmax‐imum value
\dim_{H}(J(G))
.If
$\alpha$_{-}=$\alpha$_{+},thenwehaveH(C, $\alpha$_{-})=J(G)
.(2)
Wehavea_{-}= $\alpha$+if
andonly if
there existanautomorphism
$\theta$\in \mathrm{A}\mathrm{u}\mathrm{t}(\hat{\mathbb{C}})
,complex
numbers(a_{i})_{i\in I}
and $\lambda$\in \mathbb{R}such
thatfor
all i\in I andz\in\hat{\mathbb{C}}_{J}
$\theta$\circ f\circ$\theta$^{-1}(z)=a_{i}z^{\pm\deg(f)}
andlogdeg
(f_{i})= $\lambda$\log p_{i}.
Theorem 1.3. Foreverynon‐trivialC\in \mathscr{C}
andfor
every $\alpha$<$\alpha$_{-},thefunction
Cis$\alpha$-Ho7der
continuouson
\hat{\mathbb{C}}
.Moreover,C_{0}
is$\alpha$_{-}‐Hölder continuouson\hat{\mathbb{C}}.
To prove Theorem 1.3we
develop
someideas from[KS08,
JKPS09]for interval maps. The relation between the Höldercontinuity
ofsingular
measuresand their multifractalspectrahas been first observed in[KS08],where itwasshown that the Hölder
continuity
ofthe Minkowskisquestion
mark function coincides with the bottom of theLyapunov
spectrumof theFarey
map. In[JKPS09]asimilar result hasbeen obtained forexpanding
interval maps.JJ] the
following
Theorem 1.4weprove that $\alpha$_{-}<1. This result allowsustogive
acomplete
answertotwo
important problems
raised in[Sum13],
whichgreatly improves
theprevious partial
results in[Sumlla,
Sum13,\mathrm{J}\mathrm{S}15\mathrm{a}]
.The firstimplication
isthat,
under theassumptions
ofourpaper, every non‐trivial C\in \mathscr{C} isnotdifferentiableatevery
point
ofaBorel dense subset A ofJ(G)
with\dim_{H}(A)>0
.Secondly,
weobtain in Theorem 1.5 that theaveraged
systemstillactschaotically
onthe spaceC^{ $\alpha$}(\hat{\mathbb{C}})
for any$\alpha$\in($\alpha$_{-}, 1)
,although
theaveraged
systemactsverymildly
onthe Banach spaceC(\hat{\mathbb{C}})
of\mathbb{C}‐valued continuous functionson
\hat{\mathbb{C}}
endowed with the supremumnormandonthe Banach spaceC^{ $\alpha$}(\hat{\mathbb{C}})
for small $\alpha$>0(see [Sum97,
Lemma
1.1.4],
[Sumlla,Theorem3.15]
and[Sum13,Theorem1.10]).
We recall that ifHöl(C,
z)
<1 thenCisnotdifferentiableatz.If
Höl(C,
z)
>1 then C is differentiableatzand the derivative of Cat \mathrm{z}iszero. Theorem 1.4. We have a_{-}<1. Moreover,for
every$\alpha$\in($\alpha$_{-}, 1)
there existsaBorel dense subsetAofJ(G)
with
\dim_{H}(A)>0
suchthatfor
everynon‐trivialC\in \mathscr{C}andfor
every z\in A, wehaveHöl(C,
z)
= $\alpha$<1and C isnot
differentiable
at \mathrm{z}.In the
proof,
wecombine the result thatC_{0}
is$\alpha$_{-}‐Hölder continuouson\hat{\mathbb{C}}
(Theorem 1.3),
the multifractalanalysis
onthepointwise
HölderexponentsofC_{0} (Theorems
I.2),
anargumentonLipschitz
functionson\mathbb{C} and the fact that
\dim_{H}(J(G))<2
,which follows fromourassumptions (1)
and(2)
([Sum98]).
TostateTheorem 1.5,let M:
C(\hat{\mathbb{C}})\rightarrow C(\hat{\mathbb{C}})
be the transitionoperatorofthe systemwhich is definedUy
M( $\phi$)(z)=\displaystyle \sum_{j=1}^{s+1}p_{j} $\phi$(f_{j}(z))
,where$\psi$\in C(\hat{\mathbb{C}}),z\in\hat{\mathbb{C}}
.Note thatM(C^{ $\alpha$}(\hat{\mathbb{C}}))\subset C^{ $\alpha$}(\hat{\mathbb{C}})
for any$\alpha$\in(0,1].
Theorem 1.5. Let$\alpha$\in($\alpha$_{-}, 1)
and let$\phi$\in C^{ $\alpha$}(\hat{\mathbb{C}})
such that$\phi$|_{L}=1
and$\phi$|_{L'}=0
for
everyminimalset L'of
G withL'\neq L
. Then\Vert M^{n}( $\phi$)\Vert_{ $\alpha$}\rightarrow\infty
as n\rightarrow\infty.Inparticular, for
every$\xi$\in C^{ $\alpha$}(\hat{\mathbb{C}})
andfor
everya\in \mathbb{C}\backslash \{0\}
,wehave\Vert M^{n}( $\xi$+a $\phi$)-M^{n}( $\xi$)\Vert_{ $\alpha$}\rightarrow\infty
as n\rightarrow\infty.Wenowpresentthe corollaries ofourmain results. The firstoneestablishes that every non‐trivialC\in \mathscr{C}
varies
precisely
onthe JuliasetJ(G)
. This followsimmediately
from Theorem 1.1 because theright‐hand
side of(1.1)
isalways
finite([Sum98,
Theorem2.6],
seealsoCorollary
1.11).
Thisgeneralises
aprevious
result from
[Sumlla]
forC_{0}=T_{\mathrm{p}}
andapartial
result for thehigher
orderpamal
denvatives fiiom[Sum13].
Corollary
1.6.Every
non‐trivialC\in \mathscr{C}variesprecisely
onJ(G)
, i.e.,J(G)
isequal
to thesetofpoints
Z0\in\hat{\mathbb{C}}
such that C isnot constantin anyneighborhood of
z\mathrm{o}in\hat{\mathbb{C}}
.Inparticular, thefunctions
C_{\mathrm{n}},\mathrm{n}\in \mathrm{N}_{0}^{s},
are
linearly independent
over\mathbb{C} and (g hasarepresentation
as adirectsumof
vectorspacesgiven by
\displaystyle \mathscr{C}=\bigoplus_{\mathrm{n}\in \mathrm{N}_{0}^{s}}\mathbb{C}C_{\mathrm{n}}.
Weremark
again
that0<\dim_{H}(J(G))<2
([Sum98]).
By combining
Theorem 1.1 with Birkhoffsergodic
theoremweobtain thefollowing
extensionof[Sum13,
Theorem 3.40(2)].
Recall thataBorelprobability
measurev onJ(f)
is called\overline{f}
‐invmant ifv(\overline{f}^{-1}(A))=
Corollary
1.7. Letvbean\overline{f}
‐invanantergodic
Borelprobability
measure onJ(\hat{f})
. Let$\pi$:I^{\mathrm{N}}\times\hat{\mathbb{C}}\rightarrow\hat{\mathbb{C}}
denote the canonicalprojection
onto\hat{\mathbb{C}}
Then there existsaBorel subset Aof
J(G)
with($\pi$_{*}(v))(A)=1
such that
for
everynon‐trivialC\in \mathscr{C}andfor
every z\in A,wehaveHöl
(C,z)=\displaystyle \frac{-\int\log p_{$\omega$_{1}}dv(0),x)}{\int\log\Vert f_{$\omega$_{1}}(x)||dv( $\omega$,x)}
, where$\omega$=(0)_{1},
$\Phi$,)
\in I^{\mathrm{N}}.
By
combining Corollary
1,7 with[Sumlla,
Theorem3.82]
in which thepotential
theory
wasused, weobtain the
following
result(Corollary
1.8)
onthepointwise
Hölderexponentsand thenon‐differentiauility
of elements of\mathscr{C}. Tostatetheresult,when G isa
polynomial semigroup,
wedenoteby
\tilde{ $\mu$}_{\mathrm{p}}
the maximal relative entropymeasure onJ(f)
for\tilde{f}
withrespectto( $\sigma$,\tilde{ $\rho$}_{\mathrm{p}})
(see [Sum00], [Sumlla,
Remark3.79]).
Notethat\tilde{ $\mu$}_{\mathrm{p}}
is\tilde{f}
‐invanant andergodic ([Sum00]).
Let$\mu$_{\mathrm{p}}=$\pi$_{*}(\overline{ $\mu$}_{\mathrm{p}})
.For any( $\omega$,z)\in I^{\mathrm{N}}\times\hat{\mathbb{C}}
,let
\mathscr{G}_{ $\varpi$}(z)
:=\mathrm{h}\mathrm{m}_{n\rightarrow\infty}(1/\deg(f_{ $\omega$|_{n}}))\log^{+}|f_{o\mathrm{J}|_{n}}(z)|
, where\log^{+}(a)
:=\displaystyle \max\{\log a,0\}
for every a>0.By
theargumentin[SesOl],
wehave that\mathscr{G}_{ $\omega$}(y)
exists for every( $\omega$,z)\in I^{\mathrm{N}}\times \mathbb{C}, ( $\omega$,z)\in I^{\mathrm{N}}\times \mathbb{C}\mapsto \mathscr{G}_{ $\varpi$}(z)
is continuousonI^{\mathrm{N}}\times
\mathbb{C},g_{ $\omega$}
is subharmonicon\mathbb{C} andy_{0\mathrm{J}}
restrictedtothe intersection of \mathbb{C} and the basinA_{\infty, $\omega$}
of\inftyfor\{f_{w|_{n}}\}_{n=1}^{\infty}
isthe Greens functionon
A_{\infty, $\omega$}
withpole
at\infty.Let$\Lambda$( $\omega$)=\displaystyle \sum_{c}\mathscr{G}_{ $\omega$}(c)
,wherecruns overall critical
points
of
f_{o\mathrm{J}_{1}}
inA_{\infty, $\omega$}
,counting multiplicities.
Note that$\mu$_{\mathrm{p}}=\displaystyle \int_{I^{\mathrm{N}}}dd^{c}\mathscr{G}_{0)}d\tilde{p}_{\mathrm{p}}(\mathrm{t}0)
whered^{C}=(\sqrt{-1}/2 $\pi$)(\overline{\partial}-\partial)
([Sumlla,
Lemma5.51]),
supppp=J(G)
and$\mu$_{\mathrm{p}}is non‐atomic([Sum00]).
Also,wehave\dim_{H}($\mu$_{\mathrm{p}})=
(
\displaystyle \sum_{i\in I}p_{i}\log\deg f-\sum_{i\in J}
pilog
p_{j}) / (
\displaystyle \sum_{i\in l}p_{i}
logdeg
f_{i}+\displaystyle \int_{I^{\mathrm{N}}} $\Lambda$( $\omega$)d\tilde{ $\rho$}_{\mathrm{p}}(0)
))
>0([Sumlla,
Proof of Theorem3.82]).
Here,\dim_{H}($\mu$_{\mathrm{p}})
:=\displaystyle \inf\{\dim_{H}(A)\}
where the infimum is takenoverall Borel subsets A ofJ(G)
with$\mu$_{\mathrm{p}}(A)=1.
Corollary
1.8.(1) Suppose
thatf_{1}
,\cdots,f_{s+1}
arepolynomials.
Then thereexistsaBorel dense subsetAof
J(G)
with$\mu$_{\mathrm{p}}(A)=l
and\dim_{H}(A)\geq (
$\Sigma$_{i\in I}p_{i}
logdeg
f_{i}-\displaystyle \sum_{i\in I}
pilog
p_{i}) / (\displaystyle \sum_{i\in I}p_{i}
logdeg
f_{i}+\displaystyle \int_{I^{\mathrm{N}}} $\Lambda$( $\omega$)d\tilde{ $\rho$}_{\mathrm{p}}( $\omega$)
)
>0 suchthatfor
every non‐trivialC\in \mathscr{C}andfor
everyz\in A,wehaveHöl
(C,z)=\displaystyle \frac{-$\Sigma$_{i\in I}p_{i}\log p_{i}}{$\Sigma$_{i\in I}p_{i}\log\deg f_{i}+\int_{I^{\mathrm{N}}} $\Lambda$( $\omega$)d\overline{p}_{\mathrm{p}}( $\omega$)}\prime.
(2) Suppose
thatf_{1}
,\cdots,f_{s+1}
arepolynomials satisfying
atleastoneof
thefollowing
conditions:(a)
\displaystyle \sum_{i\in IPi}\log(Pi\log f_{i})>0.
(b)
G=(f_{1},
\ldots,f_{s+1}\}
ispostcntically
bounded,i.e.P(G)\backslash \{\infty\}
isboundedin\mathbb{C}.\prime(c)s=1.
Then there existsaBorel dense subsetA
of
J(G)
with$\mu$_{\mathrm{p}}(A)=1
such thatfor
everynon‐trivialC\in \mathscr{C}andfor
everyz\in A, wehave Höl(C,z)<1
. Inparticular,
everynon‐trivialC\in \mathscr{C}isnon‐differentiable
$\mu$_{\mathrm{p}}‐almostey
erywhere
onJ(G)
.Note that ifwe assumethat every
f_{i}
isapolynomial
andP(G)\backslash \{\infty\}
isbounded in \mathbb{C},then$\Lambda$( $\omega$)=0
forevery0)
\in I^{\mathrm{N}}
,thusCorollary
1,8implies
that there existsaBorel dense subset A ofJ(G)
with$\mu$_{\mathrm{p}}(A)=1, \displaystyle \dim_{H}(A)\geq 1+\frac{-\sum_{i\in I}p_{i}\log p_{i}}{\sum_{i\in I}p_{i}\log\deg(f_{i})}>1
such that for every non‐trivialC\in \mathscr{C}and for every
point
z\in A,wehaveHöl(C,
z)
=\displaystyle \frac{-\sum_{i\in I}p_{i}\log p_{i}}{\sum_{i\in I}p_{i}\log\deg(f_{i})}<1.
The
following
is one of the otherimportant applications
ofCorollary
1.7. In order to state the res‐ult,
let $\delta$:=\dim_{H}(J(G))
and letH^{ $\delta$}
denote the $\delta$‐dimensional Hausdorffmeasure on\hat{\mathbb{C}}
. Notethatby
[Sum05],
wehave0<H^{ $\delta$}(J(G))<\infty
.LetC(J(G))
be the space of all continuous \mathbb{C}‐valued functions\displaystyle \sum_{i\in I}\sum_{f $\iota$(y)=z} $\phi$(y)\Vert f_{i}'(y)\Vert^{- $\delta$}
where$\phi$\in C(J(G)),z\in J(G)
.By [Sum05] again,
wehave that$\gamma$=\displaystyle \lim_{n\rightarrow\infty}L^{n}(1)
\in C(J(G))
exists,where 1 denotes theconstantfunctiononJ(G)
taking
its value1,the function$\gamma$ \mathrm{i}\mathrm{s}
pos‐itiveon
J(G)
, and there exists an\tilde{f}
‐invanantergodic probability
measure\tilde{v}onJ(f)
such that$\pi$_{*}(\overline{v})=
$\gamma$ H^{ $\delta$}/H^{ $\delta$}(J(G))
and\mathrm{s}\mathrm{u}\mathrm{p}\mathrm{p}$\pi$_{*}(v)=J(G)
.By Corollary
1.7 and[Sumlla,
Theorem 3.84(5)],
weobtain thefollowing,
Corollary
1.9. Under the abovenotations, there existsaBorel dense subset Aof
J(G)
withH^{ $\delta$}(A)=
H^{ $\delta$}(J(G))>0
suchthatfor
every non‐trivialC\in\subset \mathscr{E}andfor
everyz\in A,wehaveHöl(C,
z)
=\displaystyle \frac{-\sum_{i\in I}\log p_{i}\int_{f_{i}^{-1}(J(G))} $\gamma$(y)dH^{ $\delta$}(y)}{\sum_{i\in I}\int_{f_{i}^{-1}(J(G))} $\gamma$(y)\log||f_{i}(\mathrm{y})\Vert dH^{ $\delta$}(y)}.
Remark 1.10. We remark thata non‐trivialC\in \mathscr{C} may possess
points
ofdifferentiability.
In fact,by
choosing
oneoftheprobability
parameterssufficiently
small,we candeduce fromCorollary
1.9 that for every non‐trivialC\in \mathscr{C}and forH^{ $\delta$}
‐almost everyz\in J(G)
,wehave Höl(C,z)>1,
C is differentiableatzand the derivative of Cat ziszero. Note thatevenunder the abovecondition,Theorem 1.4
implies
thatthere existan $\alpha$<1 andadense subset A of
J(G)
with\dim_{H}(A)>0
such that for every non‐trivialC\in \mathbb{C}and for everyz\in A,wehave
Höl(C,
z)
= $\alpha$<1 and C isnotdifferentiableatz.Jnparticular,
in thiscase,wehave
$\alpha$_{-}<1<$\alpha$_{+}
andwehaveadifferent kind ofphenomenon regarding
the(complex)
analogues
of theTakagi
function,whereas theoriginal Takagi
function doesnothave thisproperty.We also have the
following
corollary
of Theorem 1.1. Tostatetheresult,by
[Sum98,Theorem2.6]
there existsk0\in \mathbb{N}
such that for everyk\geq k0
and for every$\omega$=($\omega$_{\mathrm{i}})_{i=1}^{k}\in l^{k}
,wehave\displaystyle \min_{z\in f_{\text{の}}^{-1}(J(G))}\Vert f_{ $\omega$}'(z)\Vert>1,
where
f\text{の}=f_{$\omega$_{k}}\circ\cdots\circ f_{o\mathrm{J}_{1}}
.Letp_{ $\omega$}:=p_{$\omega$_{k}}\cdots p_{$\omega$_{1}}for$\omega$=(0 $\lambda$)_{i=1}^{k}\in I^{k}.
Corollary
1.11. Foreveryk\geq k_{0}
,wehave0<\displaystyle \min_{ $\omega$\in J^{k}}\frac{-\log p_{ $\omega$}}{\log\max_{z\in f_{i\mathrm{J}}^{-1}(J(G))}\Vert f_{ $\omega$}'(z)\Vert}\leq$\alpha$_{-}\leq$\alpha$_{+}\leq\max_{0)\in J^{k}}\frac{-\log p_{(i\mathrm{J}}}{\log\min_{z\in f_{\overline{ $\varpi$}}^{1}(J(G))}\Vert f_{ $\omega$}'(z)\Vert}<\infty.
In
particular, if
p_{i}\displaystyle \min_{z\in f_{\mathrm{i}}^{-1}(J(G))}\Vert
fí(z)
||
>1for
everyi\in I,thenfor
every non‐trivialC\in \mathscr{C}andfor
everyz\in J(G)
,wehave thatHöl(C,
z)
\leq$\alpha$_{+}<1
and C.isnotdifferentiable
atz.Remark 1.12. Under
assumptions (1)(2)(3),
suppose that the mapsf_{i},i\in I
,arepolynomials.
ThenJ(G)\subset
\mathbb{C}. Since the
spherical
metric and the Euclidian metricareequivalent
onJ(G)
,it follows thatwe can
replace
\Vert\cdot\Vert
in the definition of $\varphi$,Corollaries1.7, 1.9,1.11by
the modulus|\cdot|.
Remark 1.13. The function
C_{0}=T_{\mathrm{p}}
is continuous(in
fact,it is Höldercontinuous)
on\hat{\mathbb{C}}
and variesprecisely
onthe JuliasetJ(G)
.Note thatby assumptions (1)(2)
and[Sum98],
wehave thatJ(G)
isafractalsetwith0<\dim_{H}\{J(G))<2
. The functionC_{0}
canbeinterpreted
as acomplex analogue
of the devils staircaseand
Lebesgues singular
functions([Sumlla]).
Infact,the devils staircase isequal
tothe restrictionto[0
,1]
of the function ofprobability
oftending
to+\inftywhenweconsider randomdynamical
systemon\mathbb{R}such thatateverystepwechoose
f_{1}(x)=3x
withprobability
1/2
and we choosef_{2}(x)=3x-2
withprobability
1/2.
Similarly, Lebesgues
singular
functionL_{p}
withrespecttotheparameterp\in(0,1),p\neq
1/2
isequal
tothe restrictionto[0
,1]
of the function ofprobability
oftending
to+\infty whenweconsiderrandom
dynamical
systemon\mathbb{R} such thatateverystepwechooseg_{1}(x)=2x
withprobability
pandwechoose
g_{2}(x)=2x-1
withprobability
1-p
.Notethat theseare newinterpretations
ofthe devils staircaseand
LeUesgues singular
functions obtained in[Sumlla] by
the second author of this paper.Similarly,
itwas
pointed
outby
him that the distributional functions of self‐similarmeasuresof FSs of orientation‐preserving contracting diffeomorphisms h_{i}
on\mathbb{R}canbeinterpreted
asthe functions ofprobability
oftending
to+\infty
regarding
the randomdynamical
systemsgenerated
by
(h_{i}^{-1})
([Sumlla]).
From the abovepoint
ofview,when G isa
polynomial semigroup
andL=\{\infty\}
,wecallC_{0}=T_{\mathrm{p}}
adevils coliseum([Sumlla]).
Itis well‐known
([YHK97])
that the function\displaystyle \frac{1}{2}\frac{\partial L_{p}(x)}{\partial p}|_{p=1/2}
on[0
,1]
isequal
totheTakagi
function$\Phi$(x)=
\displaystyle \sum_{n=0}^{\infty}\frac{1}{2^{n}}\min_{m\in \mathbb{Z}}|2^{n}x-m|
(also
referredtoas theBlancmange function),
which isafamousexample
ofacontinuous but nowheredifferentiablefunctionon
[0
,1]
. From thispoint
ofview, the first derivativesC\in \mathscr{C}canbe
interpreted
ascomplex
analogues
oftheTakagi
function. The devilsstaircase,Lebesgues
singular
functions,theTakagi
fmction and the similar functions have beeninvestigated
solong
in fractal geometryand the related fields. Infact,thegraphs
of these functions have certain kind of self‐similarities and these functions have manyinteresting
anddeep
properties.
Therearemanyinteresting
studies about theoriginal Takagi
function and its relatedtopics ([AKII]).
In [AK06], manyinteresting
results(e.g.
continuity
andnon‐differentiauility,
Hölderorder,the Hausdorff dimension of thegraph,
thesetofpoints
where the functions takeontheir absolute maximum and minimum
values)
ofthehigher
orderpartial
derivatives\displaystyle \frac{\partial^{n}L_{p}(x)}{\partial p^{n}}|_{p=1/2}
ofL_{p}(x)
withrespecttopareobtained. The firststudy
of thecomplex
analogues
of the
Takagi
functionwasgiven by
the second author in[Sum13].
Inparticular,
somepartial
resultsonthe
pointwise
Hölderexponentsofthemwereobtained([Sum13,
Theorem3.40]).
However,it had beenanopen
problem
whether thecomplex analogues
oftheTakagi
function varyprecisely
onthe Juliasetornot,until this paperwaswritten. The results of this paper
greatly improve
the above results from[Sum13]. Inthe
prbofs
of the results of this paper,we usecompletely
newideas andsystematic approaches
whichareexplained
below. For thefigures
of the JuliasetJ(G)
and thegraphs
ofC_{0}
andC_{1}
whichwedeal with in this paper whens=1,G isapolynomial
semigroup
andL=\{\infty\}
,see[Sumlla,
Sum13].Remark 1.14. The resultsonthe classical
Takagi
functionon[0, 1|
give
someevidence that the results statedin Theorem 1.3are
sharp.
Indeed,
letusconsider the functionL_{1/2}
and$\psi$_{n}(x)=\displaystyle \frac{\partial^{n}L_{p}(x)}{\partial p^{n}}|_{p=1/2}
forn\geq 1.Note that
\displaystyle \frac{1}{2}$\psi$_{1}
isequal
totheoriginal Takagi
function. SincewehaveL_{1/2}|_{[0,1]}(x)=x, L_{1/2}|_{(-\infty,0)}(x)=0
and
L_{1/2}|_{(1,\infty)}(x)=1
, the functionL_{1/2}
is 1‐Hölder(Lipschitz).
However,in[AK06]
it is shown that thefunctions
$\phi$_{n}
on[0
,1]
are a‐Hölder for every a<1,butnot1‐Hölder continuous. It would beinteresting
tofurther
investigate
thisphenomenon
for thecomplex
analogues
of theTakagi
function.Remark 1.15. We endow Rat with the
topology
induced from the distance\mathrm{d}\mathrm{i}\mathrm{s}\mathrm{t}_{\mathrm{R}\mathrm{a}\mathrm{t}}
which is definedby
\mathrm{d}\mathrm{i}\mathrm{s}\mathrm{t}_{\mathrm{R}\mathrm{a}\mathrm{t}}(f,g)
:=\displaystyle \sup_{z\in\hat{\mathbb{C}}}d(f(z),g(z))
. Thenby
[Sum97,
Theorem2.4.1],
the factJ(G)=\displaystyle \bigcup_{i\in I}f_{i}^{-1}(J(G))
([Sum97,
Lemma1.1.4]), [Sumlla,
Remark3.64],
and[Sum13,
Theorem3.24]),
wehave that theset{
(f_{i})_{i\in I}\in(\mathrm{R}\mathrm{a}\mathrm{t})^{I}
:\deg(f_{\mathrm{i}})\geq 2(i\in I)
and the conditions(1)(2)(3)
hold for(f_{i})_{i\in J}
}
is open in
(\mathrm{R}\mathrm{a}\mathrm{t})^{I}
.Also,wehaveplenty
ofexamples
towhichwe canapply
the main results of this paper.SëeSection 2.
Remark 1.16. We remark that
Uy
using
the method in this paper,we canshow similar resultstothose of this paper for randomdynamical
systemsofdiffeomorphisms
on\mathbb{R}(or
\mathbb{R}\cup\{\pm\infty\}
).
Note that thecaseof the classicalTakagi
function $\Phi$corresponds
tothedegenerated
case $\alpha$_{-}=$\alpha$_{+} in Theorem 1.2,though
in thecaseof $\Phi$wehave the opensetcondition but donothave theseparating
condition. Weemphasize
that in this paperwealso deal with thenon‐degenerated
case,whichseemsgeneric.
Remark 1.17. We remark that under
assumptions (1)(2)(3),
the iteration of the transitionoperator Monsome
C^{a}(\hat{\mathbb{C}})
is well‐Uehaved(e.g.,
there existsanM‐invanant finite‐dimensionalsubspace
U ofC^{a}(\hat{\mathbb{C}})
such that for every
h\in C^{a}(\hat{\mathbb{C}})
,M^{n}(h)
tendsto Uas n\rightarrow\inftyexponentially
fast)
and Mhasaspectral
gapon
C^{a}(\hat{\mathbb{C}})
([Sum97,
Lemma1.1.4(2)], [Sumlla, Propositions
3.63,3.65],
[Sum13,Theorems3.30,3.31]).
Note that this is.arandomness‐induced
phenomenon
(new phenomenon)
in randomdynamical
systemsfor every
f\in
Rat with\deg(f)\geq 2
,thedynamics,
off
onJ(f)
is chaotic.Combining
the abovespectral
gappropertyof Mon
C^{a}(\hat{\mathbb{C}})
and thepermrbation
theory
forlinearoperators([Kato80]) implies
that themap
\mathrm{x}=(x\mathrm{l}, \cdots,x_{s})\mapsto T_{(x_{1)}\ldots,\mathrm{x}_{s},1-$\Sigma$_{i=1}^{s}x_{j})}\in C^{a}(\hat{\mathbb{C}})
isreal‐analytic
inaneighbourhood
of \mathrm{p} in the spaceW
:=\displaystyle \{(q_{i})_{i=1}^{s}\in(0,1)^{s}:\sum_{i=1}^{s}q_{i}<1\}
([Sum13,
Theorem3.32]).
Thus it is very natural andimportant
forthe
study
of the randomdynamical
systemtoconsider thehigher
orderpartial
derivatives ofT_{\mathrm{p}}
withrespecttothe
probability
vectors. Moreover,it is veryinteresting
thatC_{\mathrm{n}}
isasolution of the functionalequation
(Id‐M)
(C_{\mathrm{n}})=F
, whereFisafunction associated with lower orderpartial
derivativesofT_{\mathrm{p}}
. Infact,by
using
thespectral
gapproperties
ofMonC^{a}(\hat{\mathbb{C}})
and theargumentsin theproof
of[Sum13,
Theorem3.32],
for any\mathrm{n}\in \mathrm{N}_{0}^{s}\backslash \{0\}
,we canshow that(I)
C_{\mathrm{n}}is theunique
continuous solution of the above functionalequation
under theboundary
conditionC_{\mathrm{n}}|s_{G}=0
and(II)
C_{\mathrm{n}}=\displaystyle \sum_{j=0}^{\infty}M^{j}(F)
inC(\hat{\mathbb{C}})
and inC^{ $\alpha$}(\hat{\mathbb{C}})
for small$\alpha$>0.Thus,wehaveasystemof functional
equations
for elementsC_{\mathrm{n}}
.Note that this is the first papertoinvestigate
thepointwise
Hölderexponentsand otherproperties
of thehigher
orderpartial
derivativesC_{\mathrm{n}}
of the functionsT_{\mathrm{p}}
ofprobability
oftending
tominimalsetswithrespecttotheprobability
parametersregarding
randomdynamical
systemswhich have several variables ofprobability
parameters. This isacompletely
newconcept.Infact,evenin the realline,there has beennostudy regarding
theobjects
similartothe above. Evenmore,inthis paperwedeal with the
complex
hnear combinations ofpartial
derivativesC_{\mathrm{n}},whichareofcourse
completely
newobjects
in mathematicscoming naturally
from thestudy
of randomdynamical
systemsand fractalgeometry. We also remark that theoriginal
Takagi
function is associatedwith
Lebesgues singular
functions,but there has beennostudy
about thehigher
orderpartial
derivatives ofthe distribution functions of
singular
measureswithrespecttotheparameters.The
key
in theproof
ofthe main results of this paper istoconsider thesystemof functionalequations
satis‐ fiedby
the elements of \mathscr{C}.Thecomposition
of theseequations along
oruits is best described intermsofan associated matrixcocycle
A( $\omega$,k)
.By using
combinatorialarguments,weshowaformula for the compon‐entsof the matrix
A(\mathrm{o}\mathrm{J},k)
,andwecarefully
estimatethepolynomial
growth
order of thesecomponents,ask tendsto
infinity. Combining
this withsomecalculations of the determinants of matrices whicharesimilartothe Vandermonde
detenninant,
wededuce the linearindependence
ofthevectors(C_{\mathrm{r}}(a)-C_{\mathrm{r}}(b))_{\mathrm{r}\leq \mathrm{n}}
for cenainpoints
a,b\in J(G)
whichareclosetoagiven point
x_{0}\in J(G)
. Here, \mathrm{r}\leq \mathrm{n}meansthatr_{i}\leq n_{i}for each i.From the‐linearindependence
of thesevectorswededuce thatacertain linear combination ofvec‐tors
(C_{\mathrm{r}}(a)-C_{\mathrm{r}}(b))_{\mathrm{r}\leq \mathrm{n}}
is bounded away fromzero. Thisgives
usthe upper bound of thepointwise
HölderexponentsofC\in \mathscr{C}.Note that thisargumentis the
key
toprove Theorem 1.1 and it is the crucialpoint
to derive that the elementsC\in \mathscr{C}arenotlocally
constantin anypoint
of the Juliaset(Corollary
1.6).
Weemphasize
that those ideasareverynewandthey
give
usstrongandsystematic
toolstoanalyze
randomdynamical
systems,singular
functions,fractal functions and other relatedtopics.
2. EXAMPLES
In thissection,we
give
someexamples
which illustrate the main results of this paper.For
f\in \mathrm{R}\mathrm{a}\mathrm{t}
,wesetF(f)
:=F( \{f\rangle),J(f) :=J(\{f))
,andP(f)=P(\langle f) ).
Wedenoteby
\mathscr{P} thesetofpoly‐
nomials of
degree
twoor more. Forg\in \mathscr{P}
,wedenoteby
K(g)
the filled‐in Juliaset. If G is arationalsemigroup
and if K isanon‐empty compactsubset of\hat{\mathbb{C}}
such thatg(K)\subset K
for eachg\in G
,then Zornslemma
implies
that there existsaminimalsetLof G with L\subset K([Sumlla,
Remark3.9]).
The
following
propositions
showusseveral methodstoproduce
manyexamples
of(fl,
\cdots,f_{s+1}
)
\in(\mathrm{R}\mathrm{a}\mathrm{t})^{s+1}
(p_{i})_{i=1}^{s}\in(0,1)^{s}
with\displaystyle \sum_{i=1}^{s}p_{i}<1
, we canapply
the results Theorems 1.1, 1.2, 1.3, 1.4, 1.5 and Corol‐laries 1.6, 1.7,1.9 and 1.11 in Section 1.
Proposition
2.1. Laet(gl,
\cdots, g_{s+1})
\in(\mathrm{R}\mathrm{a}\mathrm{t})^{s+1}
with\deg(g_{i})\geq 2,i=1\ldots,s+1
.Suppose
that\langle g_{1},
g_{s+1}}
ishyperbolic,
J(g_{i})\cap J(g_{j})=\emptyset
for
every(i,j)
withi\neq j
,and that there existatleasttwodistinct minimalsets
of
\langle g_{1}
,\cdots, g_{s+1}}.
Then thereexistsm\in \mathrm{N} suchthatfor
everyn\in \mathrm{N}withn\geq m,
setting
f=g_{i}^{n},i=
1,
s+1,the element(fl,
\cdots,f_{s+1}
)
satisfies assumptions (1)(2)(3) of
thispaperProposition
2.2. Let(gl,
\cdots, g_{s+1})
\in(\mathrm{R}\mathrm{a}\mathrm{t})^{s+1}
with\deg(g_{i})\geq 2,
i=1,
s+1.Suppose
that\displaystyle \bigcup_{i=1}^{s+1}P(g_{i})\subset
\displaystyle \bigcap_{i=1}^{s+1}F(g_{i})
,thatJ(g_{i})\cap J(g_{j})=\emptyset
for
every(i,j)
withi\neq j
,and that thereexisttwocompact subsetsK_{1},K_{2}
of
\hat{\mathbb{C}}
withK_{1}\cap K_{2}=\emptyset
such thatg_{i}(K_{j})\subset K_{j}
for
every i=1,\cdots, s+1andfor
\acute{j}=1,2
. Then there existsm\in \mathbb{N} such that
for
everyn\in \mathbb{N} withn\geq m,setting
f_{i}=g_{i}^{n},
i=1,\cdots, s+1, the element
(fl,
\cdots,
f_{s+1}
)
satisfies assumptions (1)(2)(3) of
thispaper.Combimng
[Sumlla,Remark3.9]
with[Sumlla, Proposition 6.1],
wealsoobtain thefollowing.
Proposition
2.3. Letf_{1}\in \mathscr{P}
behyperbolic,
i.e.,P(f_{1})\subset F(f_{1})
.Suppose
that Int(K(f_{1}))\neq\emptyset
, where Intdenotes theset
of
interiorpoints.
Letb\in \mathrm{I}\mathrm{n}\mathrm{t}(K(f_{1}))
be apoint.
Letd\in \mathbb{N}withd\geq 2.Suppose
that(\deg(f_{1}), d)\neq(2,2)
. Then thereexistsanumber c>0 suchthatfor
each$\lambda$\in\{ $\lambda$\in \mathbb{C}:0<| $\lambda$|<c\}
,
setting
f_{2, $\lambda$}(z) := $\lambda$(z-b)^{d}+b
,wehavethefollowing.
(1)
(f_{1},f_{2, $\lambda$})
satisfies
assumpnons(1)(2)(3) of
thispaperwith s=1.(2) If
J(f_{1})
isconnected,thenP(\{f_{1},f_{2, $\lambda$}\})\backslash \{\infty\}
isbounded in \mathbb{C}.Thus
combining
the above with Remark1.15,weobtain that for any(f_{1},f_{2, $\lambda$})
intheabove,there existsaneighborhood
Vof(f_{1},f_{2, $\lambda$})
in(Rat)2
such thatforevery(g_{1},g_{2})\in V
,assumptions (1)(2)(3)
of this paperaresatisfied and Theorems 1.1, 1.2, 1.3, 1.4, 1.5 and Corollaries
1.6,
1.7,1.9and 1.11 in Section 1 hold.Also,
endowing
\mathscr{P} with the relativetopology
fromRat,wehave that there existsanopenneighborhood
W of(f_{1},f_{2, $\lambda$})
in\mathscr{P}^{2}
such that for every(g_{1},g_{2})\in W
and for every\mathrm{p}=p_{1}\in(0,1)
,Corollary
1.8holds.Example
2.4. Let(f_{1},f_{2})\in \mathscr{P}^{2}
beanelement such that\{f_{1},f_{2}\}
ishyperbolic,
P((f_{1},f_{2}))\backslash \{\infty\}
isbounded in\mathbb{C} andJ(\{f_{1},h\rangle
)
is disconnected. Note that there areplenty
ofexamples
ofsuch elements(f_{1},f_{2})
(Proposition
2.3,[Sumllb,
\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{l}5\mathrm{b} ThenUy
[Sum09,
Theorems 1.5,1.7],
wehave thatf_{1}^{-1}(J(G))\cap
f_{2}^{-1}(J(G))=\emptyset
whereG=(f_{1},f_{2}
}.
Thus(f_{1},h)
satisfiesassumptions (1)(2)(3)
of this paper with s=1and \mathrm{a}\mathrm{U} results in Section 1 hold for
(f_{1},f_{2})
and for every\mathrm{p}=p_{1}\in(0,1)
.Example
2.5. Letg_{1}(z)=z^{2}-1,g_{2}(z)=z^{2}/4
,and letf_{i}=g_{i^{\mathrm{O}}}g_{i}, i=1,2
.Let\mathrm{p}=p_{1}=1/2
. Let G=\langle f_{1},h\}
.Then(f_{1},h)
satisfies theassumptions
(1)(2)(3)
of this paper with s=1 andP(G)\backslash \{\infty\}
is boundedin\mathbb{C}
([Sumlla, Example 6.2],[Suml3, Example 6.2]).
Thus for this(f_{1} ,h)
,all results of Section 1 hold,In
particular,
everynontrivialC\in \mathscr{C}is Hölder continuouson\hat{\mathbb{C}}
and variesprecisely
onthe JuliasetJ(G)
(Corollary
1.6).
Moreover,Uy Corollary
1.8,there exists aBoreldense subset A ofJ(G)
with$\mu$_{\mathrm{p}}(A)=
1,
\displaystyle \dim_{H}(A)\geq\dim_{H}($\mu$_{\mathrm{p}})=\frac{3}{2}
such that for everynontrivialC\in \mathscr{C}and for every z\in A,wehave $\alpha$_{-}\leqHöl(C,
z)
=\displaystyle \frac{1}{2}\leq$\alpha$_{+}
and C isnotdifferentiableatz.For thefigures
ofJ(G)
and thegraphs
ofC_{0},C_{1}
withL=\{\infty\}
,see[Sum13,
Figures
2,3,4].
Notethat Theorem 1.2implies
that$\alpha$_{-}<$\alpha$_{+}for everyprobability
vector
(parameter)
\mathrm{p}'\in(0,1)
.Example
2.6. Let $\lambda$\in \mathbb{C}with0<| $\lambda$|\leq 0.01
and letf_{1}(z)=z^{2}-1,f_{2}(z)= $\lambda$ z^{3}
.Thenby
[\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{i}\mathrm{l}5\mathrm{a}
,Exam‐ple 5.4],
the element(f_{1},f_{2})
satisfiesassumptions (1)(2)(3)
of this paper with s=1 andP((f_{1},f_{2}\rangle)\{\infty\}
isbounded in \mathbb{C}.Thus all results in Section 1 hold for(f_{1},f_{2})
and for everyprobability
vector(parameter)
\mathrm{p}=p_{1}\in(0,1)
.Thus, semngp_{1}=\displaystyle \frac{1}{2}
,G=\langle f_{1},h)
andL=\{\infty\}
,every non‐tnvialC\in \mathscr{C}is Hölder contin‐uous on
\hat{\mathbb{C}}
andvmesprecisely
onJ(G)
,andCorollary
1.8implies
that there existsaBoreldense subset Aof
J(G)
with$\mu$_{\mathrm{p}}(A)=1
and\displaystyle \dim_{H}(A)'\geq 1+\frac{2\log 2}{\log 2+\log 3}
1.7737 such that for every non‐mvialC\in \mathscr{C}and for everyz\in A,wehave $\alpha$‐\leqHöl(C,
z)
=\displaystyle \frac{2\log 2}{\log 2+\log 3}(=.0.7737)\leq$\alpha$_{+}
and C isnotdifferentiableatz.Also,by
Theorem1.2,wehave $\alpha$_{-}<$\alpha$_{+} for every\mathrm{p}'.
\in(0,1)
.Example
2.7. Letg_{1},g_{2}\in \mathscr{P}
behyperbolic. Suppose
that(J(g_{1})\cup J(g_{2}))\cap(P(g_{1})\cup P(g_{2}))=\emptyset, K(g_{1})\subset
\mathrm{I}\mathrm{n}\mathrm{t}(K(g_{2}))
, and the union ofattracting
cycles
ofg_{2} in\mathbb{C} is included in Int(K(g_{1}))
. Thenby [Sumlla,
Proposition 6.3],
there existsanm\in \mathbb{N} such that for each n\in \mathbb{N} withn\geq m,setting
f_{1}=\cdot g_{1}^{n},f_{2}=g_{2}^{n}
,wehave that
(f_{1},h)
satisfiesassumptions (1)(2)(3)
of this paper with s=1.Thus allstatementsofthe results in Section 1 holdfbr(f_{1},h)
and for every\mathrm{p}=p_{1}\in(0,1)
.The
following proposition provides
us amethodto constructexamples
of(fl,
\cdots,f_{s+1}
)
\in \mathscr{P}^{s+1}
for which(1)(2)(3)
hold andP((f_{1}, \ldots,f_{s+1}))\backslash \{\infty\}
is bounded in\mathbb{C}. For such elements(fl,
\cdots,f_{s+1}
)
andfor every\mathrm{p}\in(0,1)^{S}
with\displaystyle \sum_{i=1}^{s}p_{i}<1
,we canapply
all the results in Section 1.Proposition
2.8. Let g_{1},g_{s+1}\in \mathscr{P}
behyperbolic
andsuppose thatJ(f_{i})
isconnectedfor
everyi= 1,\cdots, s+1.Suppose
thatJ(f_{i})\subset Int(K(f_{i+1}))
for
everyi=1,\cdots, s.Suppose
also that\displaystyle \bigcup_{i=2}^{s+1}P(g_{i})\backslash \{\infty\}\subset
Int(K(f_{1}))
.Then there existsanm\in \mathrm{N}suchthatfor
everyn\in \mathbb{N} withn\geq m,
setting
f=g_{\mathrm{i}}^{n},i=1
,\cdots, s+1,the element
(fl,
\cdots,f_{s+1}
)
satisfies assumptions (1)(2)(3)
andP(\{f_{1}, \ldots,f_{s+1}\rangle)
\backslash \{\infty\}
isbounded in\mathbb{C}.Example
2.9. Letg_{1}(z)=z^{2}-1
and letg_{i}(z)=\displaystyle \frac{1}{10i}z^{2},
i=2,\cdots, s+1. Then(g_{1}, \cdots,g_{s+1})
satisfies theassumptions
ofProposition
2.8. Note thatz^{2}-1
canbereplaced by
anyhyperbolic
elementf\in \mathscr{P}
with connected Juliasetsuch thatJ(f)\subset\{z\in \mathbb{C}:|z|<10\}
and0\in \mathrm{I}\mathrm{n}\mathrm{t}(K(f))
.Fromoneelement
(gl,
. g_{m}) \in(\mathrm{R}\mathrm{a}\mathrm{t})^{m}
which satisfiesassumptions (1)(2)(3) (with
s+1=m),
weobtain many elements whichsatisfy assumptions (1)(2)(3)
ofourpaperasfollows.Proposition
2.10. Let(gl,
. g_{m}) \in(\mathrm{R}\mathrm{a}\mathrm{t})^{m}
with\deg(g_{i})\geq 2,i=1
,. m, andsupposethat(gl,
\cdots, g_{m})
satisfies
assumptions
(1)(2)(3) of
thispaper Letn\in \mathrm{N} withn\geq 2and letf_{1}
,\cdots,f_{s+1}
bemutually
distinctelements
of
\{g_{0\}_{l}}\mathrm{o}\cdots \mathrm{o}g_{a11}| ($\omega$_{1}, \cdots, w)\in\{1, m\}^{n}\}
wheres\geq 1. Thenwehave thefollowing.
(I)
(fl,
\cdots,
f_{s+1}
)
satisfies assumptions (1)(2)(3) of
thispaper Thus allstatementsin Theorems1.1,1.2, 1.3, 1.4, 1.5 and Corollaries1.6, 1.7, 1.9 and 1.1l in Section 1
holdfor
(fl,
\cdots,f_{s+1}
),
for
every minimalsetLof
\{f_{1}
, ,f_{s+1}\rangle
andfor
every\mathrm{p}=(p_{1}, \ldots,p_{s})\in(0,1)^{s}
with\displaystyle \sum_{i=1}^{s}p_{i}<1.
(II) If,
in additiontotheassumption,
(fl,
\cdots,f_{s+1}
)
\in \mathscr{P}^{s+1}
,thenstatement(1)
inCorollary
1.8 holdsfor
(fl,
\cdots,f_{s+1}
)
andfor
every\mathrm{p}, andstatement(2)
inCorollary
1.8 holdsfor
(fl,
\cdots,f_{s+1}
)
andfor
every\mathrm{p}provided
thatoneof (a)(b)(c)
intheassumption of Corollary
1.8(2)
holds.(m) If
inadditiontotheassumption ofourproposition,
(gl,
\cdots, g_{m})
\in \mathscr{P}^{m}andP((g_{1}, \ldots,g_{m}\rangle)\backslash \{\infty\}
isbounded in\mathbb{C}thenP(\langle f_{1},\ldots,f_{s+1}\})\backslash \{\infty\}
isbounded in \mathbb{C}.Thus,statement(2)
inCorollary
1.8holdsfor
(fl,
\cdots,f_{+1}
)
andfor
every\mathrm{p}.Regarding
Remark1.15,wealso have thefollowing.
Lemma 2.11. Lets\geq 1and
letl=\{1, s+1\}
. Then theset{
(f_{i})_{i\in I}\in \mathscr{P}^{I}
:(f_{i})_{i\in I}
satisfies assumptions (1)(2)(3)
andP( (
f_{1},
\ldots,f_{s+1}\})\backslash \{\infty\}
isboundedin\mathbb{C}}
We remark that the above
examples, propositions
and lemma in this section and Remark 1.15imply
thatwehave
plenty
ofexamples
towhichwe canapply
the results in Section 1. Wegive examples
towhichwe canapply Corollary
1.11.Lemma 2.12. Let
(gl,
\cdots, g_{s+1})
beanelement whichsatisfies assumptions (1)(2)(3).
Let\mathrm{p}=(p_{i})_{i=1}^{s}\in
(0,1)^{S}
with\displaystyle \sum_{i=1}^{s+1}p_{i}<1
.Letp_{s+1}=1-\displaystyle \sum_{i=\mathrm{i}}^{s}p_{i}
.Then there existsanm\in \mathrm{N} such thatfor
every n\in \mathbb{N} withn\geq m,
setting
f=g_{i}^{n},i=1\ldots,s+1
,andsetting
G:=\{f_{1}
,\cdots,f_{s+1}
),
wehave that(f_{1}, f_{s+1})
satisfies
assumptions
(1)(2)(3)
andp_{i}\displaystyle \min_{z\in f_{i}^{-1}(J(G))}\Vert
fí
(z)\Vert>1
for
everyi=1,
s+1. Thus,for
everyminimalsetL
of
\langle f_{1}
,\cdots,f_{s+1}\rangle
,andfor
everyz\in J(G)
,wehave that every non‐trivialC\in \mathscr{C}satisfies
Höl(C,
z)
\leq$\alpha$_{+}<1
and C isnot
differentiable
atz.REFERENCES
[AK06] P. Allaartand K.Kawamura,Extreme valuesofsomecontinuous nowheredifferentiablefunctions,Math. Proc.Cambridge Philos. Soc. 140(2(\mathrm{K})6),no.2,269‐295.
[AKII] P. AJlaart and KKawamura,TheTakagifunction:asurvey, Real Anal.Exchange37(2011‐12),no.1,1‐54. MR 3016850
[FS91] J. E. Fomzess and N.Sibony,Random iterationsof rationalfunctions, Ergodic Theory Dynam. Systems11(1991),no.4,
687‐708. MR 1145616(93c:58l73)
[GR96] Z.Gongand F.Ren,Arandomdynamicalsystemformed by infinitely manyfunctions,J.Fudan Univ. Nat. Sci. 35(1996),
no.4,387‐392. MR 1435167(98k:30032)
[HM96] A. Hinkkanen and G. J.Martin,Thedynamics ofsemigroups ofrationalfunctions.LProc. London Math. Soc.(3)73(1996),
no.2,358‐384. MR 1397693(97e:58l98)
[JKPS09]T.Jordan,M.Kesseb6hmer,M.Pollicott,and B. O.Stratmann,Setsofnondifferentiabilityfor conjugaciesbetweenexpand‐
inginterval maps, Fund. Math. 206(2009),161‐183. MR 2576266
[JS15a] J.Jaerisch and H.Sumi,Multifractalformalismfor expandingrationalsemigroupsand randomcomplex dynamicalsystems, Nonlinearity28(2015),2913‐2938.
[JS15b] J. Jaerisch and H.Sumi,Dynamics ofinfinitely generated nicely.expandingrationalsemigroupsand theinducing method,to
appear in Trans. Amer. Math.Soc.,http://arxiv.oryabsi1501.06772.
[JS16] J. Jaerisch and H.Sumi,Pointwise HölderExponents oftheComplex Analogues oftheTakagiFunction in RandomComplex
Dynamics, preprint2016,https://arxiv.oryabs/1603.08744.
[KatoSO] T.Kato,PerturbationTheoryforLinearOperators,Classics inMathematics,Springer,1980.
[KS08] M. Kesseböhmer and B. O.Stratmann,Fractalanalysisforsetsofnondifferentiability ofminkowskisquestion markfunc‐ tion,Joumal of NumberTheory128(2008),2663‐2686.
[Roc70] R. T.Rockafellar,Convexanalysis,Princeton MathematicalSeries,No.28, PrincetonUniversityPress,Princeton,N.J.,
1970. MR MR0274683(43#445)
[SesOl] O.Sester,Combinatonalconfigurations offiberedpolynomials, Ergodic Theory Dynam. Systems21(2001),915‐955.
[Sta12] R.Stankewitz, Density of repellingfxed pointsinthe Juliasetofarationalorentiresemigroup,Il,Discrete Contin.Dyn.
Syst.32(2012),no.7,2583‐2589. MR 2900562
[SSII] R.Stankewitz and H.Sumi,Dynamical propertiesandstructureofJuliasetsofpostcriticauyboundedpolynomialsemi‐
groups, Trans. Amer. Math. Soc. 363(2011)5293‐5319.
[Sum97] H.Sumi,Ondynamics ofhyperbolicrationalsemigroups,J.Math.Kyoto.Univ. Vol.37,No.4,(1997),717‐733.
[Sum98] H.Sumi,OnHausdorffdimension ofJuliasetsofhyperbolicrationalsemigroups,Kodai Math. J. 21(1998),no.1,1\mathrm{f}\succ 28.
MR 1625124(99h:30029)
[SumOO] H.Sumi,Skewproductmops relatedtofinitely generatedrationalsemigroups, Nonlinearîty13(2000),no.4,995−1019..
[SumOl] H.Sumi,Dynamics ofsub‐hyperbolicandsemi‐hyperUolicrationalsemigroupsand skewproducts,ErgodicTheory Dynam. Systems21(2001),no.2,563‐603. MR 1827119(2002d:37073a)
[Sum05]H.Sumi,DimensionsofJuliasetsofexpandingrationalsemigroups,Kodai Math. J. 28(2005),no.2,390‐422. MR 2153926
(2006c:37050)
[Sum09] H.Sumi,Interactioncohomology offorwardorbackwardself‐similarsystems, Adv.Math.,222(2009),no.3,729‐781.
[SumlO] H.Sumi,Dynamics ofpostciticallyboundedpolynomial semigroupsm..classification ofsemi‐kyperbolic semigroupsand
random JuliasetswhichareJordancurvesbutnotquasicircles, Ergodic Theory Dynam. Systems30(6) (2010)1869‐1902.
[Sumlla] H.Sumi,Randomcomplex dynamicsandsemigroupsofholomorphicmaps, Proc. London Math. Soc.(1) (2011),no.102,
[Sumllb]H.Sumi,Dynamics ofpostcnticallyboundedpolynomial semigroupsJ.. connected componentsofthe Julia sets, Discrete
and ContinuousDynamical SystemsSeriesA,Vol.29,No.3, 2011,1205‐1244.
[Sum13] H.Sumi,Cooperation principle, stabilityandbifurcationinrandomcomplex dynamics,Adv. Math.245(2013),137‐181.
[Sumi15a] H.Sumi,Randomcomplex dynamicsand devilscoliseums,Nonlinearity28(2015)1135‐1161.
[Sum15b] H.Sumi,The spaceof 2‐generator postcriticallyboundedpolynomial semigroupsand randomcomplex dynamics,Adv.
Math.,290(2016),809‐859.
[SU12] H. Sumi and M.Urbafiski,Bowen paroeoeter and Hausdo/ffdimensionfor expandingrationalsemigroups,Discrete Contin.
Dyn. Syst.32(2012),no.7,2591‐2606. MR 2900563
[SU13] H. Sumi and M.Urbański, Transversalityfamily ofexpandingrationalsemigroups,Adv. Math. 234
, (2013)697‐734. [Wa182] P.Walters,An introductiontoergodic theory,Graduate Texts inMathematics,vol.79.Springer‐Verlag,NewYork,1982.
MR MR648108(84e:280l7)
[YHK97] M.Yamaguti,M.Hata,and J.Kigami,Mathematicsoffractals,Translations of MathematicalMonographs167(American