The main object of this paper is to construct a new generating function of the (q-) Bernstein-type polynomials. We establish elementary properties of this function. By using this generating function, we derive recurrence relation and derivative of the (q-) Bernstein-type polynomials. We also give relations between the (q-) Bernstein-type polynomials, Hermite polynomials, Bernoulli polynomials of higher order, and the second-kind Stirling numbers. By applying Mellin transformation to this generating function, we define interpolation of the (q-) Bernstein-type polynomials. Moreover, we give some applications and questions
on approximations of (q-) Bernstein-type polynomials, moments of some distributions in
Statistics.
1. Introduction
In [1], Bernstein introduced the Bernstein polynomials. Since that time, many authors have studied these polynomials and other related subjects (cf, [1–25]), and see also the references cited in each of these earlier works. The Bernstein polynomials can also be defined in many different ways. Thus, recently, many applications of these polynomials have been looked for by many authors. These polynomials have been used not only for approximations of functions in various areas in Mathematics, but also for the other fields such as smoothing in statistics, numerical analysis and constructing Bezier curve which have many interesting applications in computer graphics (cf, [1, 5, 7, 13–20, 25] and see also the references cited in each of these earlier works).
The (q-) Bernstein polynomials have been investigated and studied by many authors without generating function. So far, we have not found any generating function of (q-) Bernstein polynomials in the literature. Therefore, we will consider the following question
How can one construct generating function of (q-) Bernstein-type polynomials?
The aim of this paper is to give answer this question and to construct generating function of the (q-) Bernstein-type polynomials which is given in Section 3. By using this generating function, we not only give recurrence relation and derivative of the (q-) Bernstein-type polynomials, but also find relations between higher-order Bernoulli polynomials, the Stirling numbers of the second-kind, and the Hermite polynomials. In Section 5, by applying Mellin transformation to the generating function of the (q-) Bernstein-type polynomials, we define interpolation function, which interpolates the (q-) Bernstein-type polynomials at negative integers.
2. Preliminary Results Related to the Classical Bernstein, Higher-Order Bernoulli, and Hermit Polynomials as well as the Stirling Numbers of the Second-Kind
The Bernstein polynomials play a crucial role in approximation theory and the other branches of Mathematics and Physics. Thus in this section we give definition and some properties of these polynomials.
Let f be a function on [0,1]. The classical Bernstein polynomials of degree n are defined by
𝔹nf(x)=∑j=0nf(jn)Bj,n(x),0≤x≤1,
where 𝔹nf is called the Bernstein operator and
Bj,n(x)=(nj)xj(1-x)n-j,j=0,1,…,n are called the Bernstein basis polynomials (or the Bernstein polynomials of degree n). There are n+1nth degree Bernstein polynomials. For mathematical convenience, we set Bj,n(x)=0 if j<0 or j>n (cf, [1, 5, 7, 9, 14, 18–20]).
If f:[0,1]→ℂ is a continuous function, then the sequence of Bernstein polynomials 𝔹nf(x) converges uniformly to f on [0,1] (cf, [10]).
A recursive definition of the kth nth Bernstein polynomials can be written as
Bk,n(x)=(1-x)Bk,n-1(x)+xBk-1,n-1(x).
For proof of the above relation see [9].
For 0≤k≤n, derivatives of the nth degree Bernstein polynomials are polynomials of degree n-1:
ddtBk,n(t)=n(Bk-1,n-1(t)-Bk,n-1(t)),
(cf, [1, 5, 7, 9, 14, 18, 19]). On the other hand, in Section 3, using our new generating function, we give the other proof of (2.4).
Observe that the Bernstein polynomial of degree n, 𝔹nf, uses only the sampled values of f at tnj=j/n, j=0, 1,…,n. For j=0, 1,…,n,
βj,n(x)≡(n+1)Bj,n(x),0≤x≤1,
is the density function of beta distribution beta(j+1,n+1-j).
Let yn(x) be a binomial b(n,x) random variable. Then
The classical higher-order Bernoulli polynomials ℬn(v)(z) are defined by means of the following generating function:
F(v)(z,t)=etx(tet-1)v=∑n=0∞ℬn(v)(z)tnn!.
The higher-order Bernoulli polynomials play an important role in the finite differences and in (analytic) number theory. So, the coefficients in all the usual central-difference formulae for interpolation, numerical differentiation, and integration and differences in terms of derivatives can be expressed in terms of these polynomials (cf, [2, 11, 12, 24]). These polynomials are related to the many branches of Mathematics. By substituting v=1 into the above, we have
F(t)=tetxet-1=∑n=1∞Bntnn!,
where Bn is usual Bernoulli polynomials (cf, [22]).
The usual Stirling numbers of the second-kind with parameters (n,k) are denoted by S(n,k), that is, the number of partitions of the set {1,2,…,n} into k nonempty set. For any t, it is well known that the Stirling numbers of the second-kind are defined by means of the generating function (cf, [3, 21, 23])
FS(t,k)=(-1)kk!(1-et)k=∑n=0∞S(n,k)tnn!.
These numbers play an important role in many branches of Mathematics, for example, combinatorics, number theory, discrete probability distributions for finding higher-order moments. In [8], Joarder and Mahmood demonstrated the application of the Stirling numbers of the second-kind in calculating moments of some discrete distributions, which are binomial distribution, geometric distribution, and negative binomial distribution.
The Hermite polynomials are defined by the following generating function.
For z, t∈ℂ,
e2zt-t2=∑n=0∞Hn(z)tnn!,
which gives the Cauchy-type integral
Hn(z)=n!2πi∫𝒞e2zt-t2dttn+1,
where 𝒞 is a circle around the origin and the integration is in positive direction (cf, [12]). The Hermite polynomials play a crucial role in certain limits of the classical orthogonal polynomials. These polynomials are related to the higher-order Bernoulli polynomials, Gegenbauer polynomials, Laguerre polynomials, the Tricomi-Carlitz polynomials and Buchholz polynomials, (cf, [12]). These polynomials also play a crucial role not only in Mathematics but also in Physics and in the other sciences. In Section 4 we give relation between the Hermite polynomials and (q-) Bernstein-type polynomials.
3. Generating Function of the Bernstein-Type Polynomials
Let {Bk,n(x)}0≤k≤n be a sequence of Bernstein polynomials. The aim of this section is to construct generating function of the sequence {Bk,n(x)}0≤k≤n. It is well known that most of generating functions are obtained from the recurrence formulae. However, we do not use the recurrence formula of the Bernstein polynomials for constructing generating function of them.
We now give the following notation:
[x]=[x:q]={1-qx1-q,q≠1,x,q=1.
If q∈ℂ, then we assume that |q|<1.
We define
Fk,q(t,x)=(-1)ktkexp([1-x]t)×∑m,l=0(k+l-1l)qlS(m,k)(xlogq)mm!,
where |q|<1, exp(x)=ex and S(m,k) denotes the second-kind Stirling numbers and
∑m,l=0f(m)g(l)=∑m=0∞f(m)∑l=0∞g(l).
By (3.2), we define the following new generating function of polynomial Yn(k;x;q) by
Fk,q(t,x)=∑n=k∞Yn(k;x;q)tnn!,
where t∈ℂ.
Observe that if q→1 in (3.4), we have
Yn(k;x;q)→Bk,n(x).
Hence
Fk(t,x)=∑n=k∞Bk,n(x)tnn!.
From (3.4), we obtain the following theorem.
Theorem 3.1.
Let n be a positive integer with k≤n. Then one has
Yn(k;x;q)=(nk)(-1)kk!(1-q)n-k×∑m,l=0∑j=0n-k(k+l-1l)(n-kk)(-1)jql+j(1-x)S(m,k)(xlogq)mm!.
By using (3.2) and (3.4), we obtain
Fk,q(t,x)=([x]t)kk!exp([1-x]t)=∑n=k∞Yn(k;x;q)tnn!.
The generating function Fk,q(t,x) depends on integer parameter k, real variable x, and complex variable q and t. Therefore the properties of this function are closely related to these variables and parameter. By using this function, we give many properties of the (q-) Bernstein-type polynomials and the other well-known special numbers and polynomials. By applying Mellin transformation to this function, in Section 5, we construct interpolation function of the (q-) Bernstein-type polynomials.
By the umbral calculus convention in (3.8), then we obtain
([x]t)kk!exp([1-x]t)=exp(Y(k;x;q)t).
By using the above, we obtain all recurrence formulae of Yn(k;x;q) as follows:
([x]t)kk!=∑n=0∞(Y(k;x;q)-[1-x])ntnn!,
where each occurrence of Yn(k;x;q)are given by Yn(k;x;q) (symbolically Yn(k;x;q)→Yn(k;x;q)).
By (3.9),
[u+v]=[u]+qu[v],[-u]=-qu[u],
we obtain the following corollary.
Corollary 3.2.
Let n be a positive integer with k≤n. Then one has
Yn+k(k;x;q)=(n+kk)∑j=0n(-1)jqj(1-x)[x]j+k.
Remark 3.3.
By Corollary 3.2, for all k with 0≤k≤n, we see that
Yn+k(k;x;q)=(n+kk)∑j=0n(-1)jqj(1-x)[x]j+k.
The polynomials Yn+k(k;x;q) are so-called q-Bernstein-type polynomials. It is easily seen that
limq→1Yn+k(k;x;q)=Bk,n+k(x)=(n+kk)xk(1-x)n,
which give us (2.2).
By using derivative operator
ddx(limq→1Yn+k(k;x;q))
in (3.2), we obtain
∑n=k∞ddx(Yn(k;x;1))tnn!=∑n=k∞nYn-1(k-1;x;1)tnn!-∑n=k∞nYn-1(k;x;1)tnn!.
Consequently, we have
ddx(Yn(k;x;1))=nYn-1(k-1;x;1)-nYn-1(k;x;1),
or
ddx(Bk,n(x))=nBk-1,n-1(x)-nBk,n-1(x).
Observe that by using our generating function we give different proof of (2.4).
Let f be a function on [0,1]. The (q-) Bernstein-type polynomial of degree n is defined by
𝕐nf(x)=∑j=0nf([j][n])Yn(j;x;q),
where 0≤x≤1. 𝕐n is called the (q-) Bernstein-type operator and Yn(j;x;q), j=0,…,n, defined in (3.7), are called the (q-) Bernstein-type (basis) polynomials.
4. New Identities on Bernstein-Type Polynomials, Hermite Polynomials, and the Stirling Numbers of the Second-KindTheorem 4.1.
Let n be a positive integer with k≤n. Then one has
Yn(k;x;q)=[x]k∑j=0n(nj)ℬj(k)([1-x])S(n-j,k),
where ℬj(k)(x) and S(n,k) denote the classical higher-order Bernoulli polynomials and the Stirling numbers of the second-kind, respectively.
Proof.
By using (2.7), (2.9), and (3.4), we obtain
∑n=k∞Yn(k;x;q)tnn!=[x]k∑n=0∞S(n,k)tnn!∑n=0∞ℬj(k)([1-x])tnn!.
By using Cauchy product in the above, we have
∑n=k∞Y(k,n;x;q)tnn!=[x]k∑n=0∞∑j=0nℬj(k)([1-x])S(n-j,k)tnj!(n-j)!.
From the above, we have
∑n=k∞Yn(k;x;q)tnn!=[x]k∑n=0k-1∑j=0nℬj(k)([1-x])S(n-j,k)tnj!(n-j)!+[x]k∑n=k∞∑j=0nℬj(k)([1-x])S(n-j,k)tnj!(n-j)!.
By comparing coefficients of tn in both sides of the above equation, we arrive at the desired result.
Remark 4.2.
In [18–20], Phillips gave many results concerning the q-integers, and an account of the properties of q-Bernstein polynomials. He gave many applications of these polynomials on approximation theory. In [6], Gould gave a different relation between the Bernstein polynomials, generalized Bernoulli polynomials, and the second-kind Stirling numbers. Oruç and Tuncer [15] gave relation between the q-Bernstein polynomials and the second-kind q-Stirling numbers. In [13], Nowak studied approximation properties for generalized q-Bernstein polynomials and also obtained Stancu operators or Phillips polynomials.
From (4.4), we get the following corollary.
Corollary 4.3.
Let n be a positive integer with k≤n. Then one has
[x]k∑n=0k-1∑j=0nℬj(k)([1-x])S(n-j,k)j!(n-j)!=0.
Theorem 4.4.
Let n be a positive integer with k≤n. Then one has
Hn(1-y)=k!yk∑n=0∞Yn+k(k;y;q)2n(n+k)!.
Proof.
By (2.10), we have
e2zt=∑n=0∞t2nn!∑n=0∞Hn(z)tnn!.
By Cauchy product in the above, we obtain
e2zt=∑n=0∞(∑j=0n(nj)Hj(z))t2n-jn!.
By substituting z=1-y into (4.8), we have
∑n=0∞(∑j=0n(nj)Hj(1-y))t2n-jn!=k!yk∑n=0∞(2nYn+k(k;y;q))tn(n+k)!.
By comparing coefficients of tn in the both sides of the above equation, we arrive at the desired result.
5. Interpolation Function of the (q-) Bernstein-Type Polynomials
The classical Bernoulli numbers interpolate by Riemann’ zeta function, which has a profound effect on number theory and complex analysis. Thus, we construct interpolation function of the (q-) Bernstein-type polynomials.
For z∈ℂ, and x≠1, by applying the Mellin transformation to (3.2), we get
Sq(z,k;x)=1Γ(s)∫0∞tz-k-1Fk,q(-t,x)dt.
By using the above equation, we defined interpolation function of the polynomials Yn(k;x;q) as follows.
Definition 5.1.
Let z∈ℂ and x≠1. We define
Sq(z,k;x)=(1-q)z-k∑m,l=0(z+l-1l)ql(1-x)S(m,k)(xlogq)mm!.
By using (5.2), we obtain
Sq(z,k;x)=(-1)kk![x]k[1-x]-z,
where z∈ℂ and x≠1.
By (5.2), we have Sq(z,k;x)→S(z,k;x) as q→1. Thus one has
S(z,k;x)=(-1)kk!xk(1-x)-z.
By substituting x=1 into the above, we have
S(z,k;1)=∞.
We now evaluate the mth z-derivatives of S(z,k;x) as follows:
∂m∂zmS(z,k;x)=logm(11-x)S(z,k;x),
where x≠1.
By substituting z=-n into (5.2), we obtain
Sq(-n,k;x)=1(1-q)n+k∑m,l=0(-n+l-1l)ql(1-x)S(m,k)(xlogq)mm!.
By substituting (3.7) into the above, we arrive at the following theorem, which relates the polynomials Yn+k(k;x;q) and the function Sq(z,k;x).
Theorem 5.2.
Let n be a positive integer with k≤n and 0<x<1. Then we have
Sq(-n,k;x)=(-1)kn!(n+k)!Yn+k(k;x;q).
Remark 5.3.
Consider the following.
limq→1Sq(-n,k;x)=S(-n,k;x)=(-1)kn!(n+k)!xk(1-x)n=(-1)kn!(n+k)!Bk,n+k(x).
Therefore, for 0<x<1, the function
S(z,k;x)=(-1)kk!xk(1-x)-z
interpolates the classical Bernstein polynomials of degree n at negative integers.
By substituting z=-n into (5.6), we obtain the following corollary.
Corollary 5.4.
Let n be a positive integer with k≤n and 0<x<1. Then one has
∂m∂zmS(-n,k;x)=(-1)kn!(n+k)!Bk,n+k(x)logm(11-x).
6. Further Remarks and Observation
The Bernstein polynomials are used for important applications in many branches of Mathematics and the other sciences, for instance, approximation theory, probability theory, statistic theory, number theory, the solution of the differential equations, numerical analysis, constructing Bezier curve, q-analysis, operator theory, and applications in computer graphics. Thus we look for the applications of our new functions and the (q-) Bernstein-type polynomials.
Due to Oruç and Tuncer [15], the q-Bernstein polynomials share the well-known shape-preserving properties of the classical Bernstein polynomials. When the function f is convex, then
βn-1(f,x)≥βn(f,x)forn>1,0<q≤1,
where
βn(f,x)=∑r=0nfr[nr]xr∏s=0n-r-1(1-qsx),[nr]=[n]⋯[n-r+1][r]!.
As a consequence of this one can show that the approximation to convex function by the q-Bernstein polynomials is one sided, with βnf≥f for all n. βnf behaves in very nice way when one varies the parameter q. In [2], the authors gave some applications on the approximation theory related to Bernoulli and Euler polynomials.
We conclude this section by the following questions.
How can one demonstrate approximation by (q-) Bernstein-type polynomials Yn+k(k;x;q)?
Is it possible to define uniform expansions of the (q-) Bernstein-type polynomials Yn+k(k;x;q)?
Is it possible to give applications of the (q-) Bernstein-type polynomials in calculating moments of some distributions in Statistics Yn+k(k;x;q)?
How can one give relations between the (q-) Bernstein-type polynomials Yn+k(k;x;q) and the Milnor algebras.
Acknowledgments
The first author is supported by the research fund of Akdeniz University. The authors would like to thank the referee for his/her comments. The authors also would like to thank Professor George M. Phillips for his comments and his some references.
BernsteinS. N.Démonstration du théorème de Weierstrass fondée sur la calcul des probabilités1912-19131312AcikgozM.SimsekY.On multiple interpolation functions of the Nörlund-type q-Euler polynomials2009200914382574MR2516009CakićN. P.MilovanovićG. V.On generalized Stirling numbers and polynomials2004183-4241248MR2076189ZBL1177.05014BergS.Some properties and applications of a ratio of Stirling numbers of the second kind1975229194MR0386083ZBL0312.62080GoodmanT. N. T.OruçH.PhillipsG. M.Convexity and generalized Bernstein polynomials1999421179190MR166934910.1017/S0013091500020101ZBL0930.41010GouldH. W.A theorem concerning the Bernstein polynomials1958315259264GuanZ.Iterated Bernstein polynomial approximationshttp://arxiv.org/abs/0909.0684JoarderA. H.MahmoodM.An inductive derivation of Stirling numbers of the second kind and their applications in statistics199712151157MR160973010.1155/S1173912697000138ZBL0910.62015JoyK. I.Bernstein polynomials, On-Line Geometric Modeling Noteshttp://en.wikipedia.org/wiki/Bernstein_polynomialKowalskiE.Bernstein polynomials and Brownian motion200611310865886MR2271533ZBL1149.60051NörlundN. E.1924Berlin, GermanySpringerLópezJ. L.TemmeN. M.nicot@cwi.nlHermite polynomials in asymptotic representations of generalized Bernoulli, Euler, Bessel, and Buchholz polynomials199923924574772-s2.0-003155657010.1006/jmaa.1999.6584NowakG.Approximation properties for generalized q-Bernstein polynomials200935015055MR247689110.1016/j.jmaa.2008.09.003OruçH.PhillipsG. M.A generalization of the Bernstein polynomials1999422403413MR169740710.1017/S0013091500020332ZBL0930.41009OruçH.TuncerN.On the convergence and iterates of q-Bernstein polynomials20021172301313MR192465510.1006/jath.2002.3703OstrovskaS.The approximation by q-Bernstein polynomials in the case q↓12006863282288MR221531710.1007/s00013-005-1503-yOstrovskaS.On the q-Bernstein polynomials2005112193204MR2169894PhillipsG. M.Bernstein polynomials based on the q-integers, the heritage of P. L. Chebyshev: a
Festschrift in honor of the 70th birthday of T. J. Rivlin199741–4511518MR1422700PhillipsG. M.200314New York, NY, USASpringerxiv+312CMS Books in Mathematics/Ouvrages de Mathématiques de la SMCMR1975918PhillipsG. M.gmp@st-andrews.ac.ukA survey of results on the q-Bernstein polynomials20103012772882-s2.0-3394728690210.1093/imanum/drn088PintérÁ.On a Diophantine problem concerning Stirling numbers1994654361364MR128144510.1007/BF01876037ZBL0811.11017SimsekY.Twisted (h,q)-Bernoulli numbers and polynomials related to twisted (h,q)-zeta function and L-function20063242790804MR226508110.1016/j.jmaa.2005.12.057SimsekY.On q-deformed Stirling numbershttp://arxiv.org/abs/0711.0481SimsekY.KurtV.KimD.New approach to the complete sum of products of the twisted (h,q)-Bernoulli numbers and polynomials20071414456MR228783310.2991/jnmp.2007.14.1.5WuZ.The saturation of convergence on the interval [0,1] for the q-Bernstein polynomials in the case q>120093571137141MR252681310.1016/j.jmaa.2009.04.003