≤
C
Numer Algor (2012) 60:545–553
ORIGINAL PAPER DOI 10.1007/s11075-012-9578-6
Nearly optimal meshes in subanalytic sets
Wiesław Ples´niak
Received: 12 December 2011 / Accepted: 12 April 2012 / Published online: 26 April 2012
© The Author(s) 2012. This article is published with open access at Springerlink.com
AbstractWe prove that any fat, subanalytic compact subset ofRNpossesses a nearly optimal (polynomial) admissible mesh. It is related to particular results that have recently appeared in the literature for very special (globally semianalytic)setslikeN-dimensionalpolynomialoranalyticgraphdomainsor
polynomialandanalyticpolyhedrons.(Hereagoodsourceofreferencesisthe
recentpaper(PiazzonandVianello,EastJApprox16(4):389–398,2010).)We also show that an infinitely differentiable mapffrom a compact setQinRNonto a Markov compact setKinCl(lN)transforms a (weakly) admissible mesh inQonto a (weakly) admissible mesh inK,which extends a result of PiazzonandVianello(EastJApprox16(4):389–398,2010)foranalyticmaps in caseQis a subset ofRN.Versions forkmaps with sufficiently largekarealsogiven.
KeywordsAdmissible polynomial meshes·Optimal meshes·Subanalytic geometry·Hironaka rectilinearization theorem·Bernstein-Walsh-Siciak theorem·Jackson theorem
AMS 2000 Subject ClassificationsPrimary 41A10; Secondary 32B20·
32U35·41A17·41A63·65D05
LetKbe a compact subset of theN-dimensional complex spaceCN. LetPd= Pd(CN)be the set of all polynomials onCNof degree at mostdand letP=
W. Ples´niak (B)
WydziałMatematykiiInformatyki,UniwersytetJagiellon´ski,ul.StanisławaŁojasiewicza6,30- 348 Kraków, Poland
e-mail:Wieslaw.Plesniak@im.uj.edu.pl
{ } ∈
∈
=
d
N N+d N
∈ =
≡
∞
=
∈
⊂ ≥
⊂
= →∞
≥
∈
⊂ = + ⊂
≥
≥
∈ ∩
∞d
1Pd.Afamily(A(d))∞d 1of finite subsetsA(d)ofKis said to be aweakly adm=
issiblemeshifthecard=inalityofA(d)growspolynomiallywhend , i.e.#A(d)
O(dα),f o r s o m e α>0,a n d t h e re e x i s t s a p o l y n o m i a l l y g r o w in g sequence C(d)of positive constants such that foreachd Nand
forallP Pdonehas
IPIK≤C(A(d))IPIA(d). (1)
HereIhISstandsfortheuniformnormsup|h|(S).IfmoreoversupC(A(d))<
, then(A(d))is said to be anadmissible mesh. Suppose thatKisP- determining, i.e. for eachPP,P0 onKforcesP(z)0. Then by the mul- tivariateLangrangeinterpolationformula(seee.g.[13,15])thereisaweakly
admissible mesh(A(d))onK,whereA(d)is a set{t1,...,tmd}of Fekete- Leja type extremal points ofKof ordermd:=dimPd= =O(d). IfK
is a Markov compact subset ofCN, i.e. a compact set that admits a Markov inequality
I∇PIK≤MdrIPIKforallP∈Pd (2) with positive constantsMandrdepending only onK, then following [3] one can construct an admissible mesh(A(d))onKwith #A(d)=O(d2rN)(and
withO(drN)cardinality, ifKRN∼RNi0CN). Observe thatr1 if
KCNandr2 for any compact setKRN(cf also Example 7) and for computational reasons one would like to construct meshes with moremodest cardinalities.Ontheotherhand,foranydN ,A(d)mustbePd-determining, whence
#A(d)md.This leads to the notion ofoptimalpolynomial meshes:
anadmissiblemesh(A(d))issaidtobeoptimal,if#A(d)O(dN)asd . If#A(d)O((dlnd)N),itiscallednearlyoptimal.Themainpurposeofthis
noteistoshowthatnearlyoptimalmeshescanbeconstructedonfat,compact
subanalyticsubsetsofRNthatareknowntoadmitMarkovinequality(2)(see [8]). Let us first recall some basic notions of subanalytic geometry that was developedmainlybyŁojasiewicz,GabrielovandHironaka.
A subsetEofRNis said to be semianalytic if for each pointxRNone can find a neighbourhoodUofxand a finite number of real analytic functionsfijandgijdefined inU, such that
E∩U=1n
{fij>0,gij=0}.
The projection of a semianalytic set need not be semianalytic (cf [2,7]).
The class of sets obtained by enlarging that of semianalytic sets to include images under the projections has been called the class of subanalytic sets.
More precisely, a subsetEofRNis said to be subanalytic if for each pointxRNthere exists an open neighbourhoodUofxsuch thatE Uis the projection of a bounded semianalytic setAinRN+M,whereM0. IfN3,
→∞
i j i j
Numer Algor (2012) 60:545–553
theclassofsubanalyticsetsisessentiallylargerthanthatofsemianalyticsets, theclassesbeingidenticalifN≤2.Theunionofalocallyfinitefamilyandthe
: i→
: i→
1
⊂
intersectionofafinitefamilyofsemianalytic(resp.subanalytic)setsissemi- analytic (resp. subanalytic). The closure, interior, boundary and complement ofasemianalytic(resp.subanalytic)setisstillsemianalytic(resp.subanalytic), the last property in the case of subanalytic sets being a (non-trivial) theorem of Gabrielov. For an excellent survey on subanalytic geometry, the reader is referred to [2]. In particular, one can find there an elegant proof of a crucial for this theory Hironaka Rectilinearization Theorem which (in a scalar space version)readsasfollows.
Theorem 1Let E be a subanalytic subset ofRN.Let K be a compact subsetofRN.Then there are f initely many real analytic mappingsϕjRNRNsuchthat:
(1) There is a compact subsetKjofRN,for eachj,s u c h that
jϕj(Kj)isaneighbourhood of K inRN. (2) ϕ−j1(E)isaunionofquadrantsinRN.
With the aid of the above theorem one can prove (see [8]) the following Theorem 2Let E be a bounded, subanalytic subset ofRNof pure dimension
N. Then there are f initely many real analytic maps fjRNRNsuch that foreach j,
fj(JN)⊂E and fj(IN)=E,
j
where JN:= {x∈RN: |xi|<1,i=1,...,N}, and IN:= {x∈RN: |xi|≤
1,i=1,...,N}.
Subanalytic geometry methods have appeared very useful in polynomial approximation, since they provide tools for investigating regularity of the pluricomplexGreen’sfunction(seee.g.[8,9,12,13]).Asanexample,werefer the reader to an important application of Hironaka’s theorem (in version of Theorem2)whichisthefollowing Corollary 3[8]If K is a fat(i .e. K
intK)compactsubanalyticsubsetofRN,thenitadmitsMarkov’sinequality(2).
Actually, in [8], it has been shown essentially more, namely that the setKoftheabovecorollaryisUPC,i.e.itisuniformlypolynomiallycuspidaland
consequently,itspluricomplexGreenfunctionisHöldercontinuousinCN.
We shall need a multidimensional version of the well-known Bernstein- Walsh theorem which is due to Siciak [15].
/
= :={∈ :| | ≤ II ∈}
| | ≤NIIK
≤
N
n
−
M
Theorem 4Let K be a compact subset of the spaceCN.Assume that K ispolynomiallyconvex,i.e.K Kˆ z CN p(z)
pKforallp P.If
f is a holomorphic function in an open neighbourhood of K then lim supndistK(f,Pn)<1.
n→∞
One can also easily prove the following
Lemma 5(cf [13])If K is a Markov compact set inCNthen for every polyno-mial P∈Pd(d=1,2... ),
P(z) e P if dist(z,K) 1
Mdr, (3)
whereMandraretheconstantsofinequality(2).Now w e c a n s t a t e t h e m a i n r e s u l t o f t h i s p a p e r .
Theorem 6Let K be a fat, compact subanalytic subset ofRN. Then one canconstruct an admissible mesh(A(d))on K such that#A(d)=O((dlnd)N)asd→ ∞.
ProofLet
fj=(fj,1,...,fj,N):RNi→RN(j=1,..., m)
be real analytic functions of Theorem 2 forE=K.LetP∈Pd.Choose a pointw∈Ksuchthat|P(w)|
=IPIK.Thenthereisj∈{1,...,m}suchthatw∈fj(I).Nowchoosex∈Isuchthat w=fj(x).Sinceanycompactsetin
Ris polynomially convex, by Theorem 4 there exist polynomialsPn,k∈Pn, n=1,2,..., and constantsL>0 anda∈(0,1)independent ofnsuch that
Ifj,k−Pn,kI ≤Lan=:
εn (4)
fork=1, . . . ,N.SetPn√=(Pn,1, . . . ,Pn,N).Letw=Pn(x).ThenIw−
wnI = Ifj(x)−Pn(x)I ≤ Nεn.Let(A(d))∞d1beanoptimaladmissiblemesh inthecubeI.(Itiswell-knownthatsuchme=shesexist;e.g.onecantakethe Cartesian product of a one dimensional meshY(d)on [ 1,1]
with#Y(d)O (d),constructedin[4],chap.3,sec.7,Lemma3.)Bythemeanvaluetheorem, Lemma5andMarkov’sinequality(2),wehave
n n N r
|P(w)−P(w)| ≤I∇PI[w,wn]Iw−wI ≤NeMdIPIKεn, provided√
Nεn≤1r. Hence, settingϕ(d,n):=NeNMdrεngives IPIK= |P(w)|≤|P(w)−P(wn)|+|P(wn)|
≤ϕ(d,n)IPIK+CIPIPn(A(dn)) (5)
( ) m j=
1
j ( )
) ε /
4
1
p
= − = p
= withC=C(A(d))≥1,as√
Nεn≤1/Mdr.Byasimilarway,weshallnowestimateIPIP
n(A(dn)). Letz∈Pn(A(dn))be such that|P(z)|=IPIPn(A(dn)). Choosey∈A(dn)sothatPn(y)=z.Wehave
|P(z)|≤|P(Pn(y))−P(fj(y))|+|P(fj(y))|
≤ϕ(d,n)IPIK+CIPIfj(A(dn)). Hence by (5),
2
IPIK≤ϕ(d,n)IPIK+Cϕ(d,n)IPIK+CIPIAt(dn), whereAtdn:=f A dn,provided√
N ≤1Mdr. Now, it iseasily seenthatthereisa√sequencen(d)=O(lnd)ofpositiveintegerssuchthat
ϕ(d,n(d))≤C and
Nεn≤1/Mdr.Then
IPIK≤2C2IPIAt(dn(d)). Onealsoverifiesthat#At(dn(d))=O((dlnd)N).
nu
In general, Theorem 6 gives better estimates of the cardinality of accessible meshes in subanalytic sets than those yielded by [3, Theorem 5]. This is seen by thefollowing
Example 7Consider the set
K=x=(x1,x2)∈R2:0≤x1≤1,0≤x2≤g(x1) ,
wheregis an analytic function in an open neighbourhood of [0,1] such that 0<g(x1)≤x for somep∈N. ThenKis a semianalytic set, whenceby Corollary 3 i s Markov. Its Markov exponentrhas to be greater thanM1 p
ti lnp
forpsufficiently large, which can be easily seen by considering the polynomials P(x1,x2)x2(1x1)p. (Actually, ifg(x1)x1, then by Goetgheluck [5]r
2p.) Thus Markov’s exponent ofKcould be as large as we want. By Theorem 6onecanconstructanadmissiblemesh(A(d))inKwith#A(d)=O((dlnd)2), asd→
∞, while by [3, Theorem 5] we know only that there exists an admissible mesh(A(d))inKwith #A(d)=O(d2r).
The idea of applying Markov’s inequality and the mean value theorem to constructingadmissiblemeshesgoesbacktoCheneyandithasbeendescribed
inhismonograph[4]inthecaseofunivariatepolynomialapproximation.Inthe proofoftheabovetheoremwealsoexploitthepossibilityofrapid(geometric)
approximationofanalyticmapsbypolynomials.Suchamethodhasalsobeen used by the authors of the recent interesting paper [10], where they provethe following
Theorem 8Let K be a Markov compact subset ofCNand let Q be aP- determiningcompactsetinCNsuchthatK=f(Q),wherefisananalyticmap inanopenneighbourhoodofthepolynomialhullQˆofQ.Let(A(d))be a
≤
=
=
≤
=
: i→
≤
=[ ] ≤ ≤ ≡ ≡ ≤ ≤ ≤
∈
(weakly)admissiblemeshforQ.Thenthereexistsasequencej(d)=O(lnd)ofna turalnumberssuchthat(At(d)):=((f(A(dj(d))))isa(weakly)admissible mesh for K with C(At(d)):::C(A(dj(d)))and#At(d)≤#A(dj(d)).
Observe that in the above theorem we are able to letfhave values in the spaceClwithlN.Let us also note that we cannot directly apply Theorem 8 in the proof of Theorem 6, since we do not know whether the setsfj(I)are Markov. We only know, by [1], that this is the case if det[fjt(x)]/=0 ateveryp o i n t x∈I.
Remark 9In a recent paper [6], Kroó constructs admissible meshes in graph domains inRNthat are sets of the type
Kg:= {(x1,...,xN)∈RN:fk(x1,...,xk−1)≤xk≤gk(x1,...,xk−1), (x1,...,xk−1)∈Ik−1,1≤k≤N},
whereIk 0,1k,1 k N,f10,g11and0 fk(x) gk(x)
1,xIk−1,2 k N.(Such domains are
also called “normal domains”i n textbookson multiple integrals.) He shows (Proposition 1) that in case the functionsfkandgkare algebraic polynomials the domainKgpossesses ano p ti ma l
polynomialmesh.Actually,itimmediatelyfollowsfromthefactthatanygraph setKgissimplytheimageofthecube[0,1]Nbythemap
F(t1,...,tN):=(t1, (1−t2)f2(t1)+t2g2(t1), ...,
(1−tN)fN(t1,...,tN−1)+tNgN(t1,...,tN−1)).
Indeed, if(A(d))is an optimal mesh inINandF=(F1,...,FN):RNi→RNis a polynomial map of degreesmax1≤k≤NdegFk, then for any polynomialPinRNof degreedoneh a s
IPIF(IN)=IP◦FIIN≤CIP◦FIA(sd)≤CIPIF(A(sd))
with #F(A(sd))#A(sd)MsNdN.The same holds true ifKis a finite union of the imagesFj(IN)of the unit cubeINby polynomial mapsFjRNRN, in particular ifKis a polytope.
If the functionsfkandgkare traces onIk−1of real analytic functions then the corresponding graph domainKgis clearly a (global) semiana- lytic set.
Then by Theorem 6 one can construct inKgan admissible mesh(A(d))with
#A(d) O((dlnd)N)which is better than the
estimate#A(d)O(dNlnN(N−1)d)yieldedinsuchacaseby[6,Theorem1].Letusaddthatinthe analytic case the cardinality result#At(d)
O((dlnd)N)forKgalsofollowsfromCorollary3andTheorem7.
Other typical sets fulfilling the assumptions of Theorem 6 are analytic polyhedrons, i.e. compact subsetsKof a domainQinRNof the type
K:= {x∈Q: |hj(x)|≤1,j=1,...,m}, wherehjare real analytic functions inQ.
j
C
C
≤
=
:=
| | = I I
I− I≤ ∈ ∈
=
n
d
4
=
Now we are going to show that in caseQis a subset ofRNTheorem 8 is also valid for∞maps and even forkmaps with sufficiently largekdepending on Markov’s exponentrof inequality (2) and the growth of thes e q u e n c e {C(A(d))}.
Theorem 10Let Q be a compact set inRNand let f(f1,...,fl)be a mapdef ined on Q, with values inCl(lN),whose components fjare traceso f
∞-functions onRN.Suppose that the set K f(Q)is Markov.
Let(A(d))bea(weakly)admissiblemeshinQ.Thenthereisapositiveintegerm suchthat(f(A(md2)))isa(weakly)admissiblemeshinK.
ProofBy the multivariate Jackson theorem (applied to a cubeI⊃QinRN),onecanfindpolynomialsPj,n∈Pnsuchthatthesequenceεj,n:=
Ifj−Pj,nIQis rapidly decreasing, i.e. for eachk>0,nkεj,n→0 asn→ ∞ forj=1,..., l(see[13,16]).LetPn=(P1,n,...,Pl,n)andεn=maxεj,n.W ehavef Pn Q
√lεn. Take a polynomialWPd(Cl)and choosewK f(Q) so thatW(w)WK. Then, by a similar argument to that of the proof of Theorem 6 (cf also the proof of Theorem 7 in [10]) we arrive at the estimate
IWIK≤ψ(d,n)IWIK+C(A(dn))ψ(d,n)IWIK
+C(A(dn))IWIf(A(dn))
withψ(d,n)Mleldrεn,provided√
lεn1/Mdr.Observethatforeachk>0 we have
kdr k dr dr
ψ(d
,n)=Const.nεn
nk≤Const.sup(nεn)
nk=C(k) nk. Consider now two cases.
1◦C:=supC(A(d)) <∞, that is the mesh(A(d))is admissible. We may assume thatC≥1. Then, settingk= [r]+1, where[r]denotes the entire
part ofr, one can find a positive integermsuch thatCψ(d,md)≤1and εm
d
4
≤1/Mdr.Consequently,
IWIK≤2CIWIf(A(md2),
and if#A(d) O(dα)for someα>0, we get #f(A(md2) O(d2α).T h u s (f(A(md2)))is an admissible mesh inK.
2◦SupposeC(A(d))=O(dβ)for someβ >0. Then again, settingk= [β+
r]+1, we can find a positive integermtsuch thatC(A(mtd2))ψ(d,mtd)≤
1andεmd≤1/Mdr.Thisyieldstheinequality
552 Numer Algor (2012) 60:545–553
IWIK≤2C(A(mtd2))IWIf(a(mtd2)).
n
{ }
C
Moreover,if#A(d)=O(d forsomeγ> 0,then#f(A(md ))=O(d .Thismea nsthatthemesh(f(A(mtd2)))isweaklyadmissible. nu Remark11By aversionofthemultivariateJacksontheoremin[16],ifamapf=(f1,..., fl)defined onQextends to aCk+1map fromRNtoCl,then for eachj∈{1,...,l},
supnkεn≤C(k)IDαfjII≤D(k,f),
|α|≤k+1
whereIis a compact cube inRNcontaining the setQ. Then, if the mesh(A(d))is admissible, Theorem 10 holds iffis aC[r]+2map, and ifC(A(d))=O(dβ)(β >0), then Theorem 10 is valid for anyC[β+r]+2mapf.
Remark 12By a non-trivial result of [11], bounded, fat and definable sets in some polynomially bounded o-minimal structures generated by special classes ofC∞functions inRNare uniformly polynomially cuspidal, whence by [8] they are Markov. This is e.g. the case of the Rolin-Speissegger-Wilkie structure (cf [14]) generated by the Denjoy-Carleman classes of quasianalytic functions with partial derivatives tempered by a strongly logarithmically convex sequenceMp. In [11], Pierzchała has proved a version of Theorem 2 for such a structure. Thus it should be possible to extend Theorem 6 to the case of definable sets in the Rolin-Speissegger-Wilkie o-minimals truc tu re.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.
References
1. Baran, M.,Ples´niak,W.: Markov’s exponent of compact sets inCn.Proc. Am. Math.
Soc.123,2785–2791(1995)
2. Bierstone, E., Milman, P.D.: Semianalytic and subanalytic sets. Publ. Math. Inst. Hautes.
ÉtudesSci.67,5–42(1988)
3. Calvi, J.-P., Levenberg, N.: Uniform approximation by discrete least squares polynomials. J.
Approx. Theory152, 82–100(2008)
4. Cheney,E.W.:IntroductiontoApproximationTheory.AMSChelseaPublishing,Providence, RhodeIsland(1982) 5. Goetgheluck,P.:InégalitédeMarkovdanslesensembleseffilés.J.Approx.Theory30,149–154(1980) 6. Kroó, A.: On optimal polynomial meshes. J. Approx. Theory163, 1107–1124(2011) 7. Łojasiewicz,S.:Ensemblessemi-analytiques.Inst.HautesÉtudesSci.,Bures-sur-Yvette(1964) 8. Pawłucki, W.,Ples´niak,W.: Markov’s inequality and∞functions on sets with polynomial
cusps.Math.Ann.275,467–480(1986)
9. Pawłucki, W.,Ples´niak,W.: Extension ofC∞functions from sets with polynomial cusps. Studia Math.88(3),279–287(1988)
10. Piazzon, F., Vianello, M.: Analytic transformations of admissible meshes. East J.A p p r o x . 16(4), 389–398 (2010)
C 11. Pierzchała, R.: UPC condition in polynomially bounded o-minimal structures. J. Approx.
Theory132,25–33(2005)
12. Ples´niak,W.:L-regularity of subanalytic sets inRn.Bull. Acad. Polon. Sci. Sér. Sci. Math.3 2, 647–651 (1984)
13. Ples´niak,W.:Markov’sinequalityandtheexistenceofanextensionoperatorfor∞functions. J.Approx.Theory61,106–
117(1990)
14. Rolin, J.-P., Speissegger, P., Wilkie, A.J.: Quasianalutic Denjoy-Carleman classes and o- minimality.J.Am.Math.Soc.16(4),751–777(2003)
15. Siciak,J.:Onsomeextremalfunctionsandtheirapplicationsinthetheoryofanalyticfunctions ofseveralcomplexvariables.Trans.Am.Math.Soc.105,322–357(1962)
16. Timan, A.F.: Theory of Approximation of Functions of a Real Variable. Oxford PergamonPress(1963)