ROCZNIKI POLSKIEGO TOWARZYSTWA MATEMATYCZNEGO Séria I: PRACE MATEMATYCZNE X IX (1976)
J .
P . Si n g h(Kuruksehtra, India)
O n the degree oî approximation oî an entire function
1. Let f(æ) be a real-valued continuous function defined on [ —1,1], and let
E J f ) = i n f ||/ - p ||, n = 0 , 1 , 2 , . . . , penn
where the norm is the maximum norm on [ —1,1] and лп denotes the set of all polynomials with real coefficients of degree at most n. Bernstein [1], p. 118, proved th a t
lim = 0
71-5-00
if and only if f(æ) is the restriction to [ —1 ,1 ] of an entire function.
Further, let f(z) be an entire function, and M{r) max |/ ( 0 ) I,
\e\=r
lim S.U^ log log M{r)jlogr — ® ( 0 < A < £ <
oo),
1П1 A
Km sup log Ж (r) T
™ inf VQ ~ t
0 < Q < OO
0 < t < T < oof
where q , à and T , t denote the order, lower order and type, lower type, respectively, of an entire function f(z).
Bernstein [1], p. 114, has shown th a t there exist constants q (posi
tive) a, T (non-negative) such th a t
lim sup nlleElln(f) = a 71— >0O
if and only if f(x) is the restriction to [ —1 ,1 ] of an entire function of order q and type T.
Becently Varga [11], Theorem 1, has proved th at lim sup {n log w /log 1 /,£/п ( / )} = q ,
n~>oo
where q is the non-negative real number if and only if f(œ) is the restriction
to [ —1 ,1 ] of an entire function of order
q.The results of Bernstein and Varga give us the clue th at the ia te at which E]ln(f) tends to zero depends on the order and type of an entire function f(z).
The object of this paper is to study th e relationship of growth con
stants defined above with the rate of growth of E n{f). Some theorems of this nature have been given in [7]-[10] but we consider some further results which will improve the results of Beddy [7], [8].
2. T
h eo rem1. Let f(co) be a real-valued continuous function defined on [ —1, 1]. I f f(x) is the restriction to Г —1, 1] of an entire function f(z) of order q and lower order X, there exist two sequences {k(n)}, (e(w)}
(n = 0,1, 2, ...) such that one has
k(n) =
r1+0(1),
s(
r)- >0 as n оо, (*) Em (f)IE m + h(f) >
(n ^ n0J h = l ?2 , 3 , . . . ) r then
у = limsup \fogEn(f)jE n+l{f)}j\ogn.
We require the following lemma in the proof of Theorem 1.
L
emma1. I f {tqj is a sequence of real or complex numbers, then limsup log \l[an\[n\ogn < limsup log \anlan+1\llogn.
n —> oo n—>oo
P r o o f of L e m m a 1. See [9], p. 193.
P r o o f of T h e o r e m 1. First, assume th at f(x) has an analytic ex
tension f(z), which is an entire function of order q and lower order A.
Following Bernstein’s original proof we have (cf. [5], p. 78, [6], p. 84) for each n > 0,
2B(a)
(2-1) E n( f ) ^ ~ —----—- for any u > 1 , a (и- l )
where В (a) is the maximum of the absolute value of f(z) on E a, and E a with a > 1 denotes the closed interior of ellipse with foci at ±1 and half
/ ( 72 + l \ / ( 72 - l \
major axis --- and half minor a x i s ---. The closed discs E A a )
\ 2(7 / \ 2(7 /
and D 2(<7) bound the ellipse E a in the sense that
( (72 — 1 1 1 (72 -h 1 1
Z > i((7)
= |« | \z\ < - 2 ~ [ c Е в с в г{а) = |я| \z\ J.
(
2
.2
)2(7 J u ' { 2(7
From this inclusion, it follows by definition th at
(72 2(7
1 \ / ( 7“ + l
for all a > 1.
From this, it can be verified th a t
(2.3) Q = lim ^ log log M{a)j\oga = lim ^ IoglogB(cr)/log(r.
' y G—>00 CO
From (2.1), we have K B {a)
<2.4) E n( f ) ^ ---- ±-L where К is some positive constant a"
since a > 1, 2/(cr —1) < K .
Consequently from (2.4), we have for any e > 0
ОСУ OQ CO
V / .
V A Y ^ jSl5 (o, + s) Y H / о
(2.5) > E n(f)a n < > ~ Г 1 г [ п = ^ ( a + e) }
H 1—i (or + e)n l - l
n = 0 n = 0 ' ' n = 0 '
ILB(<7-j- f)
G -\- EWe note [11] th at
OO
(2.6) JS(«
t) < |Р 0( г ) | + 2 с г у £ ). ( / ) 0''.
j=0 Let us write
OO
(2.7) ’
J ( a )3= 0
which represents an entire function, then we have from (2.5) and (2.6) (2.8) B(a) ^ K aJ (a) ^ K"a(a + e) В (a + s) ,
where K ' and K " are some positive constants. From (2.2) and (2.8) it follows th at
(2.9) ? = = lim ™P loglog J(a)/log(J.
Я .-.о® m f l o g
a
„ c c l n lNow, applying Lemma 1 to {En(f)} we have
lim sup log 1 jEn (f)lnlogn < lin is u p lo g { F ^ (/)/^ +1(/)}/logw.
( 2 . 10 )
By known result [7], Theorem 2A, we have 1 < lim sup log {En(f)jE n+1{f)}llogn.
If X = 0 , the theorem is an obvious consequence of this inequality, while, if X > 0 it is sufficient to prove th at
(
2
.11
)- ^ lim su p log {En(f) /E n+1(f)} /log n .
Л n— *oo
Let us pu t for simplicity
En(f)IEn+ 1 (f) =V»(«).
The sequence y)(n) cannot be bounded (in fact, yj(n)< Tc for each value of n it follows th a t El[n(f) > 1/Jc, which is absurd). Therefore limsup^(w) = oo.
П-УОО
Then denoting the sequence of the indices by {nh} for which ip{nh)
> ip(n) for each n<. nh.
Allowing y)(n) < y>(nh) for nh ^ n < nh+1. One has
( 2 . 12 ) logy(n) lo gy>(nh)
lirn sup —--- — lim sup
logn «-oo lo gnh
OO
If 9(H) = I bkzk is an entire function of lower order A and v'(r) is
*=о
the rank of the maximum term, then logv'(r)
l = liminf -- - - -- - [12].
r^oo logr Applying this to J(a), we see th at
l — liminf cr— M X?
logr (<r) logo- [v(a) is the rank of maximum term of J (a).]
Choose X' with 0 < A'< Я for each a > o-0(A'); then we get
(2.13) logv( or) > A'logcr.
Let oh = {fp{nn)}1 and let us evaluate v(ah). For each value of n > &(%), we have according to condition (*)
Therefore,
E H„j(f)<#n* > K ( f X - Hence v(oh) < k(nh).
From (2.13), it can be derived th a t
log 4 n h) ^ logv(<rft) > A '{ l-e (w A)}logy>(%), which leads to
logv4% )/log% <
As h(nn) — wA+0(1) and e(nh)
1 A'{1 - s ( n h)}
->0 for h ->
lo gk(nh) lo gnh
oo. One then obtains
(2.14) lo g y ( n h) 1
limsup —---< —
&~>oo log
71A for each A' with 0 < A' < Я.
Eow the desired conclusion follows from (2.10), (2.11), (2.12) and
(2.14). ,
E e m a r k . For s(n) = 1 /n, k(n) — n, Theorem 1 gives a result of Eeddy [8] (Theorem 3, к = 1) as a particular case.
3. T
heorem2. Let f(x) be a real valued continuous function defined on [ —1,1], which is the restriction of an entire function f(z) of order q and lower order X. Further, i f there exists a séquence {e(n)} such that one has:
s(n)-+ 0 as n-^oo, E m(f) E m+i(f) then
(3.1)
> (Д Л Я /Д и-Л Я )1 £(n) ( n > n 0, m = n, n + 1, ...),
lim inf log {Fn(f) /E n+1(f)} /log n = —,
n->-co Q
(3.2) lim sup log {Fn(f) /En+x ( / )} /log n = —.
Г7—VOO /
P r o o f of T h e o r e m 2. We omit the proof since it is based on th e same lines as the proof of Theorem 1.
E e m a r k s . (i) If the sequence F n(f)fF n+1(f) is non-decreasing for n ^ n 0, condition (**) is satisfied with e(n) = 1/n, hence our Theorem 2 contains a result of Eeddy [8] (Theorem 3, к = 1 ) as a particular case.
(ii) Condition (*) is not sufficient to guarantee the sign of equality in (3.1).
4. T
h eo rem3. L etf(x) be a real valued continuous function on [ —1 ,1 ];
then
(4.1) у = m inlim suplog{F%_i ( /) /F nJ/)} /(w A- % _ 1)log% _1, л {nh)
where {%} is an increasing sequence of positive integers and X is a non
negative real number, i f f ( x ) is the restriction to [ —1,1] of an entire function f{z) of lower order X.
P r o o f of T h e o r e m 3. From (2.2) and (2.8) it follows th at X = lim inf
<r—>oo
loglogE(o-)
loger = lim inf
<J-> 00
log log J ( a) loger
oo
If д(г) = 2 ’ h zk is an entire function of lower order X (0 < X < oo), k=0
then
X = max lim inf (% - n ^ J l o g n ^ J l o g [31> P- 310*
{ n jf n-> oo
Applying this to we get the required result, i.e.
5. To deal with functions of infinite order, we have the following classification [7]. There exists a positive integer h > 2 for which
are finite and positive, where lk(x) = loglog.. . (&-times)æ (Jc = 1, 2, 3, ...) and lkx > 0 for all sufficiently large positive x. An entire function with Qk_ l =
ooand Qk <
oois called an entire function of index h. Thus, Qk and Àk extend the definitions of q and Я, which correspond to h — 1.
If 0 < Qk <
oo,then there exists a proximate order Qk(r) (may be named as &-th proximate order of f(z)) satisfying the following conditions:
(i) Qk{r) is a non-negative, continuous function of r for r > r0 (ii) lim ^ (r) = Qk.
(iii) lim r^ (r)lo g r = 0, where Qk(r) is either the right-hand or the
T k will be called Jc-th proximate type. In this section we study the relationship of &-th proximate order Qk{r) and its corresponding T k with the rate of growth of ЕЦп(/). To state and prove the theorem precisely, we introduce a function Ф(х) which is defined as an unique solution (when x > x Q) of the equation
ek< .r)
X — Г .
We prove the following
T
h e o r em4. Let f(x) be a real valued continuous function on [ —1,1]
which is the restriction to [ —1, 1] of an entire function f(z) of index h with order Qk ( 0 < Qk < oo) and h-th proximate order Qk{r)\ then
-are finite.
Before we start with the actual pi oof, we first consider the following lemmas which will be needed in the proof of the theorem.
limSUp li‘+'M{r'> = e*
r-*co logr \
left-hand derivative at the points where they are different, (iv) lim suplkM(r)lvQk{r ) = T*k
(0< T*k <
oo).(5.1)
and if h > 1 (5.2)
lim s u p { 0 (n )^ /n(/)}
lim sup {Ф{1к- гп)Е]1п (/)}
L
emma2. The type T* (proximate type) of an entire function
OO
m = T “•**”
n = 0
with the proximate order g(r) is given by the equation Имтгр{Ф(п) \an\1/n} — (T*eg)lle.
oo
, 4
P r o o f of L e m m a 2. See [4], Theorem 2 , p. 42.
OO
L
emma3. Let f(z) be an entire function f (z) = anzn of k-th proximate
n=0
order Qk(r) and k-th proximate type T k . Then Tk is given by the equation ]imsup{#(Zfc_1« ) |a J 1/n} = (T*k)lleK
П-+ 00
We omit the proof since it can be proved on the same lines as given by Jain [2] for entire Dirichlet Series.
P r o o f of T h e o r e m 4. (2.2) and (2.8) lead to 2 Q]~ lim sup
£7—> 0 0
1кЩ ° )
Qk(c) = lim sup
a—>oo
h B ( a )
ееь{а)(а) — lim sup
<7—> 0 0
hJ(a)
*
On applying Lemma 2 to J(a) we obtain (5.1). Similarly on applying Lemma 3 to J (a) we get (5.2). This completes the proof of the theorem.
R e m a r k . Taking ok(r) = gk in Theorem 4 we get the result of Reddy [7], Theorem 3, which in tu rn includes the result of Bernstein [1] on finiteness.
My thanks are due to Dr. S. H. Dwivedi for his constant help in the preparation of this paper and to Dr. Y. B. Goyal for his encouragement.
I also thank CSIR Yew Delhi for the award of a Senior Research fellowship.
References
[1] S. N. B ern ste in , Leçons sur les propriétés extrémales et la meilleure approxi
mation des fonctions analytiques d'une variable réelle, Gauthier-Villars, Paris 1926.
[2] P. K. Jain, Proximate order of an entire Dirichlet Series of order (P) infinity, Rev. Roumaine Math. Pures Appl. 15 (1970), p. 367-371.
[3] О. P. Ju ne) a, On the lower order of entire functions, J. London Math. Soc. (2) 5 (1972), p. 310-312.
[4] B. J a. L evin , Distribution of zeros o f entire functions, Vol. 5, Amer. Math. Soc.
Trans., Providence 1964.
[5] G. G. L oren tz, Approximation of functions, Holt, Rinehart and Winston, New York 1966.
10 — Boczniki PTM — Prace Matematyczne XIX
[6] Gi. M einardus, Approximation of functions, Theory and Numerical Methods, Springer-Verlag, New York 1967.
[7] A. R. R edd y, Approxim ation of an entire function, J. Approximation Theory 3 (1970), p. 128-137.
[8] — Best polynomial approximation to certain entire functions, ibidem 5 (1972),.
p. 97-112.
[9] D. R oux, Sul divario fra Vordine e Vordine inferiore délie funsioni intere, Riv.
Mat. Univ. Parma (2) 4 (1963), p. 191-210.
[10] J. P. Singh, Approxim ation of an entire function, Yokohama Math. J. 19 (1971), p. 105-108.
[11] R. S. Varga, On an extension o f a result of S. N . Bernstein, J. Approximation Theory 1 (1968), p. 176-179.
[12] J. M. W h itta k er, The lower order o f integral functions, J. London Math. Soc*
8 (1933), p. 20-27.
DEPARTMENT OF MATHEMATICS KURUKSHETRA UNIVERSITY KURUKSHETRA, INDIA and
43 BRAHMAN PURI ALIGARH, U. P.
INDIA