Volume 2009, Article ID 528454,22pages doi:10.1155/2009/528454
Research Article
Derivatives of Integrating Functions for
Orthonormal Polynomials with Exponential-Type
Weights
Hee Sun Jung
1and Ryozi Sakai
21Department of Mathematics Education, Sungkyunkwan University, Seoul 110-745, South Korea
2Fukami 675, Ichiba-cho, Toyota-city, Aichi 470-0574, Japan
Correspondence should be addressed to Hee Sun Jung,[email protected]
Received 22 December 2008; Revised 26 February 2009; Accepted 22 April 2009
Recommended by Jozsef Szabados
Letwρx:|x|ρexp−Qx,ρ >−1/2, whereQ∈C2:−∞,∞ → 0,∞is an even function. In
2008 we have a relation of the orthonormal polynomialpnw2
ρ;xwith respect to the weightwρ2x; pnx Anxpn−1x−Bnxpnx−2ρnpnx/x, whereAnxandBnxare some integrating
functions for orthonormal polynomialspnw2
ρ;x. In this paper, we get estimates of the higher derivatives ofAnxandBnx, which are important for estimates of the higher derivatives of pnw2
ρ;x.
Copyrightq2009 H. S. Jung and R. Sakai. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
1. Introduction and Results
Let R −∞,∞. LetQ ∈ C2 : R → R 0,∞ be an even function, and letwx
exp−Qxbe such that∞0xnw2xdx <∞for alln0,1,2, . . . .Forρ >−1/2, we set
wρx:|x|ρwx, x∈R. 1.1
Then we can construct the orthonormal polynomials pn,ρx pnw2
ρ;xof degree nwith
respect tow2
ρx. That is,
∞
−∞pn,ρxpm,ρxw 2
ρxdxδmn Kronecker’s delta,
pn,ρx γnxn· · ·, γnγn,ρ >0.
A functionf :R → Ris said to be quasi-increasing if there existsC >0 such that
fx ≤ Cfyfor 0 < x < y. For any two sequences {bn}∞n1 and {cn}∞n1 of nonzero real numbersor functions, we writebn cnif there exists a constantC >0 independent ofnor
xsuch thatbn≤Ccnfornlarge enough. We writebn ∼cnifbncnandcn bn. We denote the class of polynomials of degree at mostnbyPn.
Throughout C, C1, C2, . . . denote positive constants independent of n, x, t, and polynomials of degree at mostn. The same symbol does not necessarily denote the same constant in different occurrences.
We will be interested in the following subclass of weights from1.
Definition 1.1. LetQ:R → Rbe even and satisfy the following properties. aQxis continuous inR, withQ0 0.
bQxexists and is positive inR\ {0}. c
lim
x→ ∞Qx ∞. 1.3
dThe function
Tx: xQ
x
Qx , x /0 1.4
is quasi-increasing in0,∞with
Tx≥Λ>1, x∈R\ {0}. 1.5
eThere existsC1 >0 such that
Qx |Qx| ≤C1
Qx
Qx , a.e. x∈R\ {0}. 1.6
Then we writew ∈ FC2. If there also exist a compact subintervalJ 0ofRandC2 > 0 such that
Qx |Qx|≥C2
Qx
Qx , a.e. x∈R\J, 1.7
then we writew∈ FC2.
In the following we introduce useful notations.
aMhaskar-Rahmanov-Saff MRSnumbersaxare defined as the positive roots of the following equations:
x 2
π
1
0
axuQaxu
bLet
ηx xTax−2/3, x >0. 1.9
cThe functionϕuxis defined as follows:
ϕux
⎧ ⎪ ⎪ ⎨ ⎪ ⎪ ⎩
a2 2u−x2
uauxauηuau−xauηu1/2 , |x| ≤au,
ϕuau, au<|x|.
1.10
In the rest of this paper we often denotepn,ρxsimply bypnx. Letρnρifnis odd,
ρn 0 otherwise and define the integrating functionsAnxandBnxwith respect topnx
as follows:
Anx:2bn
∞
−∞p 2
nuQx, uw2ρudu,
Bnx:2bn
∞
−∞pnupn−1uQx, uw 2
ρudu,
1.11
whereQx, u Qx−Qu/x−uandbn γn−1/γn. Then in2, Theorem 4.1we have a relation of the orthonormal polynomialpnxwith respect to the weightw2
ρx:
pnx Anxpn−1x−Bnxpnx−2ρn pnx
x , ρn
⎧ ⎨ ⎩
ρ, nis odd,
0, nis even, 1.12
and in 2, Theorem 4.2we already have the estimates of the integrating functions Anx
andBnxwith respect topnx. So, in this paper we will estimate the higher derivatives of
AnxandBnxfor the estimates of the higher derivatives ofpnwρ2;x, because the higher
derivatives ofpn,ρxplay an important role in approximation theory such as investigating convergence of Hermite-Fej´er and Hermite interpolation based on the zeros ofpnw2
ρ;x see
3,4.
To estimate of the higher derivatives ofAnxandBnxwe need further assumptions forQxas follows.
Definition 1.2. Letwx exp−Qx∈ FC2, and letνbe a positive integer. Assume that
Qxisν−times continuously differentiable onRand satisfies the followings. aQν1xexists andQix,i0,1, . . . , ν1 are nonnegative forx >0.
bThere exist positive constantsCi>0 such that forx∈R\ {0}
Qi1x≤CiQix
Qx
cThere exist constants 0≤δ <1 andc1>0 such that on0, c1
Qν1x≤C 1
x
δ
. 1.14
Then we writewx∈ FνC2.
Letνbe a positive integer. Define formα−ν >0,m≥0, l≥1, andα≥0,
Ql,α,mx:|x|mexpl|x|α−α∗expl0, 1.15
whereα∗0 ifα0, otherwiseα∗1 and define
Qαx: 1|x||x|α−1, α >1. 1.16
Here we let exp0x : xand for l ≥ 1, explx : expexp· · ·expx· · · denotes the
lth iterated exponential. In particular, explx expexpl−1x. Then exp−Ql,α,mxand exp−Qαxare typical examples ofFνC2 see5.
In the following we improve the inequality4.3in2, Theorem 4.2.
Theorem 1.3. Letρ >−1/2andwx exp−Qx∈ FC2. Additionally assume thatQx
is nondecreasing. Then for|x| ≤εanwith0< ε <1/2one has
|Bnx|< λε, nAnx, 1.17
where
lim
ε→0nlim→ ∞λε, n 0. 1.18
In this paper our main theorem is as follows.
Theorem 1.4. Letρ >−1/2andwx exp−Qx∈ FνC2for positive integerν≥2. Assume that12ρ−δ≥0forρ <0and
ann1/1ν−δ, 1.19
where0≤δ <1is defined in1.14.
a If Qx/Qxis quasi-increasing on c2,∞, then one has for |x| ≤ an1ηnand
j0, . . . , ν−1
AnjxAnx Tan an
j
, BnjxAnx Tan an
j
Moreover, for any0< ε <1/2there existsε∗ε, n>0such that for|x| ≤εanandj1, . . . , ν−1,
Anjx≤ε∗ε, nAnx n an
j
, Bnjx≤ε∗ε, nAnx n an
j
, 1.21
withε∗ε, n → 0asn → ∞.
bIfQν1xis non-decreasing onc
2,∞, then one has1.20and1.21for the respective
ranges ofx.
cIf there exists a constant0 ≤δ <1such thatQν1x≤C1/xδonc2,∞, then one
has1.20and1.21for the respective ranges ofx.
The examples satisfying the conditionsa,b, orcofTheorem 1.4are given in5.
Remark 1.5. Under the assumptions ofTheorem 1.4, we have from2, Theorem 4.2that there existsC, n0>0 such that forn≥n0and|x| ≤an1Lηn,
Anx
2bn ∼ϕnx
−1a2
n12Lηn2−x2
−1/2
, |Bnx|Anx, 1.22
becausewx exp−Qx∈ FνC2for positive integerν≥1 and 12ρ−δ≥0 forρ <0.
In addition, for our future work we estimateatandTatusingλC1in1.6for the weight classFC2.
Theorem 1.6. Letwx exp−Qx∈ FC2, and we assume
Qx |Qx|≤λ
Qx
Qx , |x| ≥b >0, 1.23
whereb >0is large enough.
aAssume thatTxis unbounded. Then for anyη >0there existsCη> 0such that for t≥1,
at≤Cηtη. 1.24
bSuppose that there exist constantsη >0andC2>0such thatat≤C2tη. Then there exists
a constantCdepending only onλ,η, andC2such that forat≥1, ifλ >1
Tat≤Ct2ηλ−1/λ1, 1.25
and if0< λ≤1,
Remark 1.7. aLevin and Lubinsky showed the following1, Lemma 3.7: there existsC >0 such that for someε >0, and for large enought,
Tat≤Ct2−ε. 1.27
If from1.25and1.26we set for any 0< η <2
ε
⎧ ⎪ ⎪ ⎨ ⎪ ⎪ ⎩
2−η, 0< λ≤1,
22−η
λ1 , λ >1,
1.28
then we have1.27in Levin and Lubinsky’s lemma.
bIfTxis unbounded, then1.19is trivially satisfied by1.24.
2. Proof of Theorems
In this section we will prove the theorems ofSection 1.
Lemma 2.1. Letρ >−1/2and letwx∈ FC2. Then uniformly forn≥1, a
sup
x∈R
pn,ρxwx |x|an n
ρ
x2−a2n
1/4
∼1. 2.1
b
sup
x∈R
pn,ρxwx |x|an n
ρ
∼a−n1/2nTan1/6. 2.2
cMarkov inequality. Let0< p≤ ∞. For any polynomialP ∈ Pn
Pwx |x|an n
ρ
LpR
nTan1/2 an
P wx |x|an n
ρ
LpR
. 2.3
dLetβ∈R,0 < p ≤ ∞, andr >1. Then there exist positive constantsL,δ, andC2such
that for any polynomialP ∈ Pn
P w)x |x|ann
β
Lparn≤|x|
exp−C2nδP wx |x|an
n
β
LpLan/n≤|x|≤an1−Lηn .
Proof. afollows from2, Theorem 2.3.bfollows from2, Theorem 2.4.cfollows form 6, Theorem 2.1b.dfollows form6, Theorem 2.3.
Lemma 2.2. Letρ >−1/2and letwx∈ FC2. Then one has forc >0,
0≤u≤c
pnwρ2udu 1
an. 2.5
Proof. Forρ ≥0, the results are immediate fromLemma 2.1a. So we assume−1/2 < ρ <0. First we see
0≤u≤an/n
pnwρ2udu
0≤u≤an/n
pnw2u |u|an n
2ρ | u|2ρ
|u|an/n2ρdu
≤C 1 an
0≤u≤an/n
|u|2ρ
|u|an/n2ρdu
≤C 1 an
n an
2ρ
0≤u≤an/n |u|2ρdu
≤C 1 an
n an
2ρ an
n
12ρ
≤C1 n,
2.6
because we know thatanonfrom1, Lemma 3.5c. Next we see byLemma 2.1a
an/n≤u≤c
pnwρ2udu≤C 1
an. 2.7
Therefore, we have the result.
Lemma 2.3. Letρ >−1/2and letwx∈ FC2. Then aone has
0≤u≤∞
pnw2u |u|an n
2ρ
Qudu∼ n
an, 2.8
bforx∈0, an/2one has
Qx≤Cn an
x an
Λ−1
. 2.9
Proof ofTheorem 1.3. SinceBnxis an odd function, we prove only for 0≤ x≤ εan. Letθ :
εΛ−1/2Λ. Then we have the following two lemmas.
Lemma 2.4. Uniformly forθandn
|u|≤θan
pnupn−1uw2ρuQx, udu
1
nθ1
θΛ−1 n
a2
n
. 2.10
Proof. For|u| ≤θan, we have byLemma 2.1a
p2nuw2ρu
1
a2
n−θan2
|u|2ρ
|u|an/n2ρ
1
an
|u|2ρ
|u|an/n2ρ. 2.11
Since Qx is nondecreasing and 1 − 1/2Λ1/2Λ ≤ θ − ε/θ ≤ 1, we have using
Lemma 2.3b:
Qx, u≤ Q
θan−Qx
θan−x
Qθan
θ−εan θ
Λ−2 n
a2
n
. 2.12
Moreover we know that forρ >−1/2,
θan
0
|u|2ρ
|u|an/n2ρdx
|u|≤an/n
an/n≤|u|≤θan an
n θan. 2.13
Therefore, we have
|u|≤θan p2
nuwρ2uQx, udu
1
nθ 1
θΛ−1 n
a2
n
. 2.14
Consequently, we have the result using Cauchy-Schwartz inequality
|u|≤θan
pnupn−1uw2ρuQx, udu
1
nθ1
θΛ−1 n a2
n
. 2.15
Lemma 2.5. Uniformly forθεΛ−1/2Λand forn
θan≤|u|≤a2n
pnupn−1uwρ2uQx, udu
ε1−1/ΛΛ−1ε1/Λn a2
n
Proof. Forθan≤ |u| ≤a2n, we have similarly to2,4.6
Qx, u−Qx,−u2uQ
x−xQu x2−u2
an
QεanεanQu θan2
ε1−1/ΛΛ−1 n a2
n
ε1/Λ
an
Qu
2.17
seeLemma 2.3b. Therefore, we have byLemma 2.3a,
θan≤|u|≤a2n
pnupn−1uw2ρuQx, udu
≤
θan≤|u|≤a2n
pnupn−1uw2ρuQx, u−Qx,−udu
ε1−1/ΛΛ−1 n a2n
θan≤|u|≤a2n
pnupn−1uw2
ρudu
ε1/Λ
an
θan≤|u|≤a2n
pnupn−1uw2
ρuQ
udu
ε1−1/ΛΛ−1 n a2
n
ε1/Λ n a2
n .
2.18
Here we usedLemma 2.1b.
Since for a constantC >0
a2n≤|u|
pnupn−1uwρ2uQx, udu
O
e−nC
, 2.19
see2, page 233, there existsλn>0 such that
a2n≤|u|
pnupn−1uw2ρuQx, udu
λnan2
n
, 2.20
and λn → 0 asn → ∞. We know from2, Lemma 4.7that bn γn−1/γn ∼ an. From 1.22we haveAnx/bn∼n/a2
nfor|x| ≤εanand from the preceding considerations and the
definition ofBnxit follows that for|x| ≤εan
|Bnx|
bn
λε, nn a2n
∼ λε, nAnx
where for some positive constantC >0
λε, n:C·max 1
nθ1
θΛ−1, ε1−1/ΛΛ−1, ε1/Λ, λn
. 2.22
Consequently,1.17is proved, and we can obtain that limε→0limn→ ∞λε, n 0. Now, we have for|x| ≤εan
Anx∼ n
an, |Bnx|< λε, n n
an. 2.23
Proof ofTheorem 1.4. First, we see that for 1≤j≤ν−1
Anjx 2bn
∞
−∞pnwρ 2
u d j
dxjQx, udu. 2.24
We split proof of1.20into some lemmas as follows:
1Lemma 2.6is for 0≤x≤an1ηn,a4n≤u, and 1≤j ≤ν−1;
2Lemma 2.9is foran/2≤x≤an1ηn, 0≤u≤a4n, andjν−1;
3Lemma 2.10is for 0≤x≤an/2, 0≤u≤a4n, andjν−1;
4Lemma 2.11is for 0≤x≤an1ηn, 0≤u≤a4n, and 1≤j≤ν−2;
on the other hand,1.21will be proved by Lemmas2.13and2.6. For 1≤j≤ν−1 there existsηbetweenuandxsuch that
dj dxjQx, u
j! x−uj1
j
k0 −1kQ
j1−kx
j−k! x−u
j−k −1j1
Qu
Qj1x−Qj1
η
x−u .
2.25
Then forx≥0 andu≥0, sinceQj1uis increasing for 1≤j≤ν−1, we have
0≤ d
j
dxjQx, u≤
Qj1x−Qj1u
x−u . 2.26
Ifu <0 andx >0, then since|Qj1η| ≤Qj1−uforη <0,
d
j
dxjQx, u
Qj1x−Qj1η
x−u
≤ Qj1x Qj1−u
x −u
≤ Qj1x−Qj1−u
x−−u 2
Qj1−u−Qj10
−u−0 .
So, for this case we can prove the result similarly to the casex, u > 0. For the other cases, we can prove it by the symmetry of Q, similarly. Therefore, we assume that u and xare nonnegative, and we will prove this theorem only for nonnegativexandu. Moreover, for simplicity, we letc1c2without loss of generality, because we know by1.13thatQν1u is bounded for anyubetweenc1andc2.
On the other hand, ifQj2uis increasing, then
Qj1u−Qj1x
u−x 2.28
is also increasing forubecause there exists a pointξbetweenxandusuch that
d du
Qj1u−Qj1x
u−x
Qj2u−
Qj1u−Qj1x/u−x
u−x
Qj2u−Qj2ξ
u−x ≥0.
2.29
Moreover, ifQν1t≤C1/tδfor
tbetweenxandu, then we see
Qνu−Qνx
u−x
1
u−x
u
x
Qν1tdt≤ C u−x
u1−δ−x1−δ≤C 1 u
δ
. 2.30
To complete the proof ofTheorem 1.4we prove a series of lemmas.
Lemma 2.6. Let0≤x≤an1ηnand1≤j≤ν−1:
a4n<u
pnwρ2u d j
dxjQx, udu
Tan an
jAnx
an . 2.31
Proof. Since
Anjx
2bn
0≤u≤a4n
a4n<u
pnwρ2u d j
dxjQx, udu, 2.32
we have to estimate
a4n<u
pnwρ2u d j
dxjQx, udu:
a4n<u
. 2.33
First, we see forx >0 large enough,
is decreasing because
Qj1xe−QxQj2x−Qj1xQxe−Qx, 2.35
and so from our assumption,
Qj2x−Qj1xQx≤CQj1xQ
x Qx −Q
j1xQx
Qj1xQx C Qx−1
<0,
2.36
ifC < Qx. We use this fact. Let 2ρ βiwhereβ <0, and letibe a nonnegative integer, and letPu p2
nuui. Letu >0. Then since
Qj1a
4n Qa4n ≤C
Tan an
j
, 2.37
by1.13, we have for someξbetweenxandu
a4n<u
a4n<u
pnwρ2uQj2ξdu
≤
a4n<u
pnwρ2uQj2udu
≤CQ j1a
4nwa4n
Qa4n a
β
4n
a4n<u
PuwuQudu by2.34
≤ Tan an
j
wa4naβ4n
∞
a4n
−Pu d
duwudu,
∞
a4n Pu d
duwudu P wa4n−
∞
a4n
Ptwudu.
2.38
ApplyingLemma 2.1dwithL∞,L1-norm andLemma 2.1c,
|P wa4n| ≤exp−C2nαP wxL∞Lan/n≤|x|≤an1−Lηn, ∞
a4n
Puwudu≤exp−C2nα
Pwx
L1Lan/n≤|x|≤an1−Lηn
≤exp−C2nαnTan 1/2
an P wxL1Lan/n≤|x|≤an1−Lηn.
Therefore,
∞
a4n
PuwuQudu≤exp−C2nαP wxL∞Lan/n≤|x|≤an1−Lηn
exp−C2nαnTan 1/2
an P wxL1Lan/n≤|x|≤an1−Lηn.
2.40
Consequently we have
Tan an
j
wa4naβ4n
a4n<u
PuwuQudu
≤ Tan an
j
exp−C2nαp2nwρ2L∞La
n/n≤|x|≤an1−Lηn
Tan
an
j
nTan1/2
an exp−C2n αp2
nwρ2L
1Lan/n≤|x|≤an1−Lηn
≤Oe−nd3 Tan
an
j
Tan an
j Anx
an .
2.41
Lemma 2.7. If Qx/Qx is quasi-increasing on c1,∞ or if Qν1x is nondecreasing on c1,∞, then one has
Qνx−Qνu
x−u
⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩
1 Tan
an
ν−1 n
a2
n
, 0≤u≤c1, c1≤x≤ an2 ,
1 Tan
an
ν−1
n a2n
, c1≤u≤2c1, 0≤x≤c1,
Tan an
ν−1
n a2
n
, 2c1 ≤u≤ an3 , 0≤x≤c1,
Tan an
ν−1
n a2n
, c1≤u≤ an
3 , c1≤x≤
an
2 .
2.42
Proof. Casea-1. 0≤u≤c1andc1≤x≤an/2. Let
Qνu−Qνx
u−x ≤
Qνu−Qνc1
u−c1
Qνc1−Qνx
c1−x
Then we haveQ1u1 from2.30. Then ifQx/Qxis quasi-increasing onc1,∞, there exists a pointξ∈c1, xsuch that by1.13
Q2x Q
ν1ξ
Qξ
Q
x−Qc1 x−c1
Qξ Qξ
ν−1 Q
an/2−Qc1 an/2−c1
Qan/2
Qan/2 ν−1
Q
an/2
an
Tan an
ν−1
n a2
n .
2.44
IfQν1xis nondecreasing onc1,∞, there exists a pointξ∈c1, xsuch that by2.28and 1.13
Q2x≤ Q
νan/2−Qνc1
an/2−c1
Qνan/2 Qan/2
Qan/2−Qc1 an/2−c1
Qan/2
Qan/2 ν−1
Q
an/2
an
Tan an
ν−1 n
a2
n .
2.45
Casea-2. Forc1 ≤u≤2c1and 0≤x≤c1, we have similarly to Casea-1,
Q2x1, Q1u Tan
an
ν−1
n a2
n
. 2.46
Caseb. 2c1≤u≤an/3 and 0≤x≤c1. Using the method of Casea-1, and similarly to Casea-2,
Qνu−Qνx
u−x ∼
Qνu−Qνc1
u−c1
Tan an
ν−1
n a2n
Casec.c1 ≤u≤an/3 andc1≤x≤an/2. We can prove similarly toQ1uandQ2x of Casea-1. IfQx/Qxis quasi-increasing onc1,∞, there exists a pointξ∈c1, xsuch that by1.13
Qνx−Qνu
x−u
Q
ν1ξ
Qξ
Q
x−Qu x−u
Qan/2
Qan/2 ν−1
Q an/2 an
∼ Tan an
ν−1
n a2
n .
2.48
IfQν1xis nondecreasing onc
1,∞, there exists a pointξ∈c1, xsuch that by1.13
Qνx−Qνu
x−u ≤
Qνan/2−Qνu an/2−u
|Qan/2|
Qan/2 ν−1
Q an/2 an
∼ Tan an
ν−1 n
a2
n .
2.49
Lemma 2.8. One has
Qνx−Qνu
x−u
⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ 1
uδ, 0≤u≤c1, 0≤x≤c1,
Tan an
ν−1
Qx, u, 0≤u≤a4n, an
2 ≤x≤an
1ηn,
Tan an
ν−1
Qx, u an
3 ≤u≤a4n, 0≤x≤
an
2 .
2.50
Proof. Casea. 0≤u≤c1and 0≤x≤c1. From2.30and1.14
Qνu−Qνx
u−x ≤C
1
u
δ
. 2.51
Caseb-1. 0 ≤ u ≤ an/3 andan/2 ≤ x ≤ an1ηn. Since by1, page 64, Lemma 3.2a
Qan/2
Qan/3 ≥ 3 2
Λ−1
we have
Qx−Qu≥Qx
1−Q
an/3
Qan/2
≥Qx
1− 2
3 Λ−1
. 2.53
Therefore, since for this case
Qx−Qu∼Qx, 2.54
we have
Qνx−Qνu
x−u
Qνx−Qνu
Qx−Qu Qx, u
Qνx Qx
Qx, u
Tan an
ν−1
Qx, u.
2.55
Caseb-2.an/3≤u≤a4nandan/2≤x≤an1ηn. There exists a pointξbetweenx
andusuch that by1.13
Qνx−Qνu
x−u
Qνx−Qνu Qx−Qu Qx, u
Qν1ξ Qξ
Qx, u
Tξ ξ
ν−1
Qx, u
Tan an
ν−1
Qx, u.
2.56
Casec.an/3≤u≤a4nand 0≤x≤an/4. By the same method as Caseb, we have
Qνx−Qνu
x−u
Tan an
ν−1
Qx, u. 2.57
Lemma 2.9. Letan/2≤x≤an1ηn. Then
0≤u≤a4n
pnwρ2u d ν−1
dxν−1Qx, udu
Tan an
ν−1Anx
an . 2.58
Lemma 2.10. Let0≤x≤an/2.
aIf0≤x≤c1, then
0≤u≤c1
pnwρ2u d ν−1
dxν−1Qx, udu 1
an. 2.59
Moreover, one knows that
1
an
Tan an
ν−1Anx
an . 2.60
bIfQx/Qxis quasi-increasing onc1,∞, or ifQν1xis nondecreasing onc1,∞,
then
0≤u≤a4n
pnwρ2u d ν−1
dxν−1Qx, udu
Tan an
ν−1
Anx
an . 2.61
cIf there exists a constant0 ≤δ < 1such thatQν1x≤ C1/xδon0,∞, then one has2.61.
Proof. aFor 0≤x≤c1we have from Lemmas2.8,2.1a, and2.2
0≤u≤c1
pnwρ2u d ν−1
dxν−1Qx, udu
0≤u≤c1
pnwρ2uu−δdu
⎧ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎩
1
an, ρ≥0,
1
an n an
2ρ
0≤u≤c1
u2ρ−δ 1 an
n an
2ρ
, ρ <0
1 an,
2.62
because 12ρ−δ≥0 forρ <0. On the other hand, from1.19we seeaν
n≤nν/1ν−δ≤n, and
from1.22we seeAnx∼n/anfor 0≤x≤c1. So we have
1
an
n aνn1
n a2
n
Tan an
ν−1
∼ Anx an
Tan an
ν−1
bFor 0≤x≤c1, we have froma, Lemmas2.7, and2.8
0≤u≤a4n
pnwρ2u d ν−1
dxν−1Qx, udu
0≤u≤c1
c1≤u≤2c1
2c1≤u≤an/3
an/3≤u≤a4n
Tan an
ν−1Anx
an .
2.64
Similarly, forc1≤x≤an/2 we have from Lemmas2.7and2.8
0≤u≤a4n
pnwρ2u d ν−1
dxν−1Qx, udu
0≤u≤c1
c1≤u≤an/3
an/3≤u≤a4n
Anx an
Tan an
ν−1
.
2.65
cThen by2.26andLemma 2.1a
0≤u≤a4n
pnwρ2u d ν−1
dxν−1Qx, udu
0≤u≤a4n
pnwρ2uu−δdu
0≤u≤a4n
u2ρ−δ
uan/n2ρ
u2−a2
n du
0<u<an/n
an/n≤u≤an/2
an/2≤u≤a4n
1 aδ
n
n a2n
1
an
ν−1
Anx an
Tan an
ν−1
.
2.66
Here, we use the fact 1/aδ
n< n/aνn1from1.19for the last inequality.
Lemma 2.11. Let0≤x≤an1ηn. Then for1≤j≤ν−2,
0≤u≤a4n
pnwρ2u d j
dxjQx, udu Anx
an
Tan an
j
. 2.67
To prove1.21we need some lemmas.
Lemma 2.12. Let0< ε <1and|x| ≤εan.
aFor someC >0one has
Qεan Qεan ≤C
εΛ−1
Qεan n
an. 2.68
bFor any0< ε <1, there existsε1ε, n>0such that for2εan≤u
dj
dxjQx, u≤ε1ε, n Anx
an n an
j
Qx, u
εanj , 2.69
andε1ε, n → 0asn → ∞.
Proof. aIt follows fromLemma 2.3b.bBy2.25,Lemma 2.3b, anda, we have
dj
dxjQx, u≤C j−1
k0
Qj1−kx
εank1
Qx, u
εanj
≤C j−1
k0
Qεan
εank1
εΛ−1
Qεan n an
j−k
Qx, u εanj
≤C j−1
k0
εΛ−1 εank1
n an
εΛ−1
Qεan n an
j−k
Qx, u εanj
≤Cn a2
n n an
j j−1
k0
εΛ−k−2
εΛ−1
Qεan
j−k
1
n
k
Qx, u εanj
≤ε1ε, nAnx an
n an
j
Qx, u εanj ,
2.70
where we let
ε1ε, n:C j−1
k0
εΛ−k−2
εΛ−1
Qεan
j−k
1
n
k
−→0 2.71
asn → ∞. Therefore, this lemma is proved.
Lemma 2.13. Suppose that the one of the three conditions (a), (b), and (c) inTheorem 1.4is satisfied. Then for any0< ε <1/2, there existsε2ε, n>0such that for|x| ≤εanandj1, . . . , ν−1,
0≤u≤a4n
pnwρ2u d j
dxjQx, uduε2ε, n Anx
an n an
j
, 2.72
Proof. First, we consider the case of which cin Theorem 1.4is satisfied. Then the lemma follows from2.66withε2ε, n: 1/nν−1. Now, we consider the other cases. If we consider only for |x| ≤ εan and |u| ≤ 2εan in proving Lemmas 2.7and 2.8, then we know that for
|x| ≤εanandj1, . . . , ν−1
dj
dxjQx, u
⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩
1u−δ Qεan Qεan
j
Qεan
εan , 0≤u≤2c1,
Qεan Qεan
j
Qεan
εan , 2c1 ≤u≤
ε
2an,
Qεan Qεan
jQεan
εan ,
ε
2an≤u≤2εan.
2.73
Then we have byLemma 2.12a
0≤u≤2εan
pnwρ2u d j
dxjQx, udu
0≤u≤2c1
2c1≤u≤2εan
1 aδn
Qεan Qεan
j
Qεan εan
⎛ ⎝a2nj−δ
n1j ε
Λ−2
εΛ−1
Qεan
j⎞
⎠Anx
an n an j , 2.74
and we can see that
ε3ε, n: a 2j−δ n n1j ε
Λ−2
εΛ−1
Qεan
j
−→0 as n−→ ∞. 2.75
Finally, we estimate2εa
n≤u≤a4n. ByLemma 2.12bwe have
2εan≤u≤a4n
pnwρ2u d j
dxjQx, udu
≤
2εan≤u≤a4n
ε1ε, nAnx an
n an
j
1
εanjQx, u
pnwρ2udu
≤
ε1ε, n 1
εnj
Therefore, if we letε2ε, n:ε3ε, n ε1ε, n 1/εnj,then
0≤u≤a4n
pnwρ2u d j
dxjQx, uduε2ε, n Anx
an n an
j
, 2.77
andε2ε, n → 0 asn → ∞by2.75and2.76.
From the proof ofLemma 2.6, we have the following. There existsε4n>0 satisfying
ε4n → 0 asn → ∞such that
u≥a4n
pnwρ2u d j
dxjQx, udu≤ε4n Anx
an n an
j
. 2.78
Therefore, from Lemmas2.6,2.9,2.10, and2.11we obtain the estimate forAnjxin1.20,
and fromLemma 2.13and2.78we have the estimate for Anjxin1.21. Using
Cauchy-Schwarz Inequality we also have the estimate forBnjxin1.20and1.21. Consequently,
we provedTheorem 1.4, completely.
Proof ofTheorem 1.6. a 1.24follows from1,3.45easily.
bSuppose that1.23is satisfied on|x| ≥Dfor someD >0 large enough. Letx > D. From1.23we have for largex > D
ln
Qx QD
≤ln Qx
QD
λ
, 2.79
and we have for largex > D
Qx QD ≤
Qx QD
λ
. 2.80
Caseλ >1. Then we can see by1, Lemma 3.43.18and2.80
Tat atQ
at Qat ≤
QD
QDλatQat λ−1≤
Cat
t
Tat
λ−1
. 2.81
Therefore from the assumptionat≤C2tηwe have for anyη >0
Tat≤Cλ, ηt2ηλ−1/λ1. 2.82
Case 0< λ≤1. Then we have by2.80
Tx xQ
x Qx ≤x
QD QDλQx
λ−1≤
xQ
D
Therefore, from the assumptionat≤C2tηwe have for anyη >0
Tat≤Cλ, ηtη. 2.84
Acknowledgments
The authors thank the referees for many kind suggestions and comments. The first author was supported by the Korea Research Foundation Grant funded by the Korean Government KRF-2008-C00017.
References
1 E. Levin and D. S. Lubinsky,Orthogonal Polynomials for Exponential Weights, vol. 4 ofCMS Books in Mathematics/Ouvrages de Math´ematiques de la SMC, Springer, New York, NY, USA, 2001.
2 H. S. Jung and R. Sakai, “Orthonormal polynomials with exponential-type weights,” Journal of Approximation Theory, vol. 152, no. 2, pp. 215–238, 2008.
3 Y. Kanjin and R. Sakai, “Pointwise convergence of Hermite-Fej´er interpolation of higher order for Freud weights,”Tohoku Mathematical Journal, vol. 46, no. 2, pp. 181–206, 1994.
4 P. V´ertesi, “Hermite-Fej´er interpolations of higher order. I,”Acta Mathematica Hungarica, vol. 54, no. 1-2, pp. 135–152, 1989.
5 H. S. Jung and R. Sakai, “Specific examples of exponential weights,” Communications of Korean Mathematical Society, vol. 24, no. 2, pp. 303–319, 2009.