Abstract
In order to approximate continuous functions on \([0,+\infty )\), we consider a Lagrange–Hermite polynomial, interpolating a finite section of the function at the zeros of some orthogonal polynomials and, with its first \((r-1)\) derivatives, at the point 0. We give necessary and sufficient conditions on the weights for the uniform boundedness of the related operator. Moreover, we prove optimal estimates for the error of this process in the weighted \(L^p\) and uniform metric.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In this paper we discuss the weighted polynomial interpolation of continuous functions on \([0,+\infty )\), which are \((r-1)\)-times differentiable at 0 and can increase with order \(\mathcal {O}\left( \mathrm {e}^{x^\beta /2}\right) , \beta >1/2\), for \(x\rightarrow \infty \).
The case \(\beta =1\) has been treated in [1, 5, 15, 20, 22, 24–26], where the Authors considered Lagrange polynomials based on Laguerre zeros (see also [8]). Here we choose as a tool the orthonormal system \(\{p_m(w)\}_m\) in \((0,+\infty )\) related to a generalized Laguerre weight of the form \(w(x)=x^\alpha \mathrm {e}^{-x^\beta }\). From the numerical point of view, we observe that the weight w is nonclassical in general and, for the construction of the orthonormal system in the case \(\beta \ne 1\), we can use the procedure introduced in [2] (see also [16]).
The presence of the derivatives of the function in 0 leads in a natural way to the construction of Lagrange–Hermite polynomial \(\mathcal {L}_{m,r}(w,f)\) based at the zeros of the polynomial \(p_m(w), 0\) as a multiple node and another additional node. Applying the operator \(\mathcal {L}_{m,r}(w)\) to a suitable finite section of the function f, we obtain a new interpolation process, that we will denote by \(\mathcal {L}^*_{m,r}(w)\). This new operator is not a projector into the set of all polynomials of degree at most \(m+r\), \(\mathbb {P}_{m+r}\), but on a special subset \(\mathcal {P}_{m,r}^*\subset \mathbb {P}_{m+r}\) that can replace the space \(\mathbb {P}_{m+r}\), in the sense we are going to show. Thus the projector \(\mathcal {L}^*_{m,r}(w)\) can be profitably used in quadrature rules and in the numerical treatment of Boundary Value Problems on \((0,\infty )\) (see, e.g., [6, 27]).
We estimate the error of the process in weighted \(L^p\) and uniform metric and compare it with the order of convergence of the best weighted polynomial approximation (that can be found in [21]). The error estimates are sharp for the considered classes of functions. All the results of this paper are new and cover the ones in the literature.
The paper is structured as follows. In Sect. 2 we recall some basic facts and give some preliminary results, in Sect. 3 we will state our main results and in Sect. 4 we will prove them.
2 Preliminary results
In the sequel \(\mathcal {C}\) will stand for a positive constant that can assume different values in each formula and we shall write \(\mathcal {C}\ne \mathcal {C}(a,b,\ldots )\) when \(\mathcal {C}\) is independent of \(a,b,\ldots \). Furthermore \(A\sim B\) will mean that if A and B are positive quantities depending on some parameters, then there exists a positive constant \(\mathcal {C}\) independent of these parameters such that \((A/B)^{\pm 1}\le \mathcal {C}\). Finally, we denote by \(\mathbb {P}_m\) the set of all algebraic polynomials of degree at most m.
In order to introduce some interpolation processes we consider the weight
with \( \alpha > -1, \, \beta > \frac{1}{2}\), and the corresponding sequence of orthonormal polynomials \(\{p_m(w)\}_m\), with positive leading coefficient \(\gamma _m\). The zeros \(x_k=x_{m,k}(w)\) of \(p_m(w)\), \(m\ge 1\), are located as follows
where
is the Mhaskar–Rakhmanov–Saff number related to the weight \(\sqrt{w}\) (see also [18, 19]). If \(\beta =1\) the nodes \(x_k\) are the Laguerre zeros. Nevertheless, in general w is a nonstandard weight and, for the computation of the zeros and the Christoffel numbers, we may use the Mathematica package “OrthogonalPolynomials” introduced in [2].
For any continuous function in \([0,+\infty )\), \(f \in C^0([0,+\infty ))\), and \((r-1)\)-times differentiable at 0, briefly \(f \in C_r^0\), \(r\ge 1\), we define the Lagrange–Hermite polynomial \(\mathcal {L}_{m,r}(w,f)\) as follows
with \(x_{m+1}:=a_m\), and
where \(f^{(0)}\equiv f\). Here we used also an idea introduced by Szabados for Lagrange and Hermite–Fejér interpolation, adding the additional node \(a_m\) (see [28]).
The polynomial \(\mathcal {L}_{m,r}(w,f)\) can be written as
where
and
are the fundamental Lagrange polynomials. It is easily seen that
and \(\mathcal {L}_{m,r}(w)\) is a projector from \(C^0_r\) into \(\mathbb {P}_{m+r}\).
We are now going to introduce another Lagrange–Hermite type operator \(\mathcal {L}_{m,r}^*(w)\), modifying the operator \(\mathcal {L}_{m,r}(w)\). To this end, for a fixed \(\theta \in (0,1)\) and for a sufficiently large m we define an index \(j=j(m,\theta )\) such that
Of course if m is small we let \(j=m\). Hence we define the new Lagrange–Hermite type polynomial as follows
By this definition,
and
Moreover, we observe that \(\mathcal {L}_{m,r}^*(w)\) does not preserve all the polynomials of degree at most \(m+r\), for example \(\mathcal {L}_{m,r}^*(w,1) \ne 1\).
Nevertheless, if we introduce the following set of polynomials
then it is easy to show that for any \(f \in C^0_r\), \(\mathcal {L}^*_{m,r}(w,f) \in \mathcal {P}_{m,r}^*\) and for any \(Q \in \mathcal {P}_{m,r}^*, \mathcal {L}^*_{m,r}(w,Q)=Q\), therefore \(\mathcal {L}^*_{m,r}(w,f)\) is a projector from \(C^0_r\) into \(\mathcal {P}_{m,r}^*\).
In order to show some approximation properties of the projectors defined above, we are going to define some function spaces.
2.1 Function spaces
Let us consider a weight of the form
with \(\beta > \frac{1}{2}\) and the following weighted function spaces associated to u.
For \(1 \le p < \infty \) and \(\gamma >-1/p\), by \(L_u^p\) we denote the set of all measurable functions f such that
For \(p=\infty \) and \(\gamma \ge 0\), by a slight abuse of notation, we set
and we equip this space with the norm
Note that the Weierstrass theorem implies the limit conditions in the definition of \(C_u\).
Subspaces of \(L_u^p, 1\le p\le \infty \), are the Sobolev spaces, given by
where \(AC(0,+\infty )\) denotes the set of all functions which are absolutely continuous on every closed subset of \((0,+\infty )\) and \(\varphi (x)=\sqrt{x}\). We equip these spaces with the norm
In order to define some further function spaces, we consider the s-th modulus of smoothness of \(f \in L^p_u, 1\le p \le \infty , s\ge 1\),
where
is the forward finite difference of order s with variable step \(h\varphi (x)\), \(\varphi (x)=\sqrt{x}\) and \(I_{s h}=[8s^2h^2,\mathcal {C}h^{-\frac{1}{(\beta -1/2)}}]\), \(h>0\).
Let \(f\in L_u^p\), \(1\le p\le \infty \). Then the following estimate
holds with \(\mathcal {C}\ne \mathcal {C}(f,t)\), provided the norm on the right-hand side is finite (see [21]).
Using these moduli of smoothness we can define the Zygmund spaces as
with the norm
Let us denote by
the error of best weighted polynomial approximation in \(L^p_u\), \(1\le p\le \infty \). Estimates for the error of best weighted approximation have been proved in [21]. In particular it has been shown that
For our aims it is sufficient to recall the following weak Jackson inequality
that holds for any function \(f \in L^p_u, 1\le p\le \infty \).
In particular, if \(f\in Z^p_\lambda \), by (2.8) we deduce
and by (2.6) we have
for any \(f\in W^p_s, s\ge 1\) and \(1\le p \le \infty \).
We are now able to show an important property of the polynomials in \(\mathcal {P}_{m,r}^*\). To this end we need some further notation. We say that \(P_M\) is a polynomial of quasi best approximation for \(f\in L^p_u\), \(1\le p \le \infty \), if, for some \(\mathcal {C}\ge 1\),
Moreover, we denote by
the error of best approximation by means of polynomials of \(\mathcal {P}_{m,r}^*\). The relation between \(E_m(f)_{u,p}\) and \(E_{m}^*(f)_{u,p}\) is established by the following Lemma.
Lemma 1
Let w, u be the weights defined in (2.1) and (2.5) with \(\alpha > -1, \beta > \frac{1}{2}\) and \(u\in L^p\), \(1\le p \le \infty \). Let \(f\in L^p_u\) and \(P_M\in \mathbb {P}_M\) one of its polynomial of quasi best approximation, with
for a fixed \(\theta \in (0,1)\). Then we have
where \(\mathcal {C}, c\) are independent of m, r, f. Moreover, for \(s\ge 1\), we get
Using Lemma 1 and (2.7) it follows that the order of convergence of \(E_{m,r}^*(f)_{u,p}\) is the same as that of \(E_{m,r}(f)_{u,p}\). Therefore, \(\bigcup _m \mathcal {P}_{m,r}^*\) is dense in \(L^p_u\). In the next Section we will show the behaviour of the operators \(\mathcal {L}_{m,r}(w)\) and \(\mathcal {L}_{m,r}^*(w)\) in different function spaces.
3 Main results
The following results hold true in weighted uniform norm. In order to state it we need a further definition. We say that a function is quasi increasing on I if there exists a constant \(\mathcal {C}\) such that \(f(x)\le \mathcal {C}f(y)\) for \(x<y\), \(x,y\in I\).
Theorem 2
Let w and u be the previously defined weights, with \(\gamma \ge 0\) and \(\alpha >-1\). For every \(f\in C_r^0 \cap C_u\), \(r\ge 1\), we have
where \(\mathcal {C}\ne \mathcal {C}(m,f)\), if and only if
Moreover, under the assumptions (3.2), if \(\left| f^{(i)}\right| \) is quasi increasing on \([0,\alpha _m]\), with \(\alpha _m=a_m/m^2\) and \(i=0,1,\ldots , r-1\), we get
where \(\mathcal {C}\ne \mathcal {C}(m,f)\) and the constant in “\(\mathcal {O}\)” is independent of m.
Therefore, in weighted spaces of continuous functions, the behaviour of the operators \(\left\{ \mathcal {L}_{m,r}\left( w\right) \right\} _m\) is comparable with similar interpolation processes based on Jacobi zeros on bounded intervals (see, e.g., [15, 17]).
We emphasize that if r and \(\gamma \) are given, then, in order to approximate a function f, we can always choose some \(\alpha \) such that the Lagrange–Hermite polynomial converges with the order of the best polynomial approximation times the extra factor \(\log m\). In fact, we can rewrite (3.2) as follows
Moreover, since the Lebesgue constants related to the processes \(\{\mathcal {L}_{m,r}^*(w)\}_m\) are bounded by the ones related to \(\{\mathcal {L}_{m,r}(w)\}_m\), by Theorem 2 and Lemma 1, we deduce the following error estimate for the “truncated” process.
Corollary 3
Let \(\theta \in (0,1)\). Under the assumptions of Theorem 2, for every \(f\in C_r^0, r\ge 1\), for m sufficiently large (say \(m>m_0)\) we have
where M given by (2.11), \(\mathcal {C}\ne \mathcal {C}(m,f)\) and the constant in “\(\mathcal {O}\)” is independent of m.
In weighted \(L^p\)-spaces the behaviour of \(\left\{ \mathcal {L}_{m,r}\left( w\right) \right\} _m\) is not optimal, while for the new process \(\left\{ \mathcal {L}_{m,r}^*\left( w\right) \right\} _m\) we can state the following
Theorem 4
Let \(u\in L^p, 1< p<\infty \) and \(\theta \in (0,1)\) be fixed. Then, for every function \(f\in C_r^0\), \(r\ge 1\), we have
if and only if
where the constants in “\(\sim \)” depend on \(\theta \) and are independent of m and f.
If the function \(f\in L^p_u\) fulfills the additional assumption
for some \(s\ge 1\), then f is continuous in \((0,+\infty )\) (see [21]). Moreover, using the same arguments as in [22], we obtain
Hence, if \(f\in L^p_u\) fulfills (3.6) and is \((r-1)\)-times differentiable at 0, the bound (3.4) becomes
where \(\mathcal {C}\) depends on \(\theta \in (0,1)\) (fixed) and is independent of m and f. Note that if \(f \in \mathcal {P}_{m,r}^*\), (3.7) becomes an equivalence.
Theorem 5
Under the assumptions (3.5), let \(f\in L^p_u\, \cap \, C^0_r\) satisfy (3.6) for some \(s\ge r\) and \(\left| f^{(i)}\right| \) be quasi increasing on \([0,\alpha _m]\), with \(\alpha _m=a_m/m^2\) and \(i=0,1,\ldots , r-1\). Then, for \(m>m_0\), we have
where \(\mathcal {C}\) depends on \(\theta \) and is independent of m and f, the constant in “\(\mathcal {O}\)” is independent of m.
In particular, under the assumptions of Theorem 5, if \(f\in Z^p_\lambda (u), 1/p < \lambda \le r-1\), the error estimate (3.8) becomes
where the constant in \(\mathcal {O}\) is independent of m and f, that is the order of the best approximation in \(Z^p_\lambda (u)\) [see (2.9)].
Analogously, if \(f\in W^p_s(u), s\le r-1\), from Theorem 5 we get
where the constant in \(\mathcal {O}\) is independent of m and f [see (2.10)]. While, for \(f\in W^p_r(u)\), the following corollary holds.
Corollary 6
Under the assumptions (3.5), let \(f\in W^p_r(u)\cap C^0_r, 1<p<\infty \), such that \(\left| f^{(i)}\right| \) is quasi increasing on \([0,\alpha _m], \alpha _m=a_m/m^2\), for \(i=0,1,\ldots , r-1\) and m sufficiently large (say \(m>m_0\)). Then we have
and
where \(\mathcal {C},c\) depend on \(\theta \) and are independent of m and f.
Finally, the following remark is of some interest.
Remark 7
The behaviours of the two interpolation processes \(\{\mathcal {L}_{m,r}(w)\}_m\) and \(\{\mathcal {L}_{m,r}^*(w)\}_m\) are essentially equivalent in weighted uniform norm. Nevertheless, the “truncated” process \(\{\mathcal {L}_{m,r}^*(w)\}_m\) has the advantage of dropping \(c\,m\) terms, \(c<1\). This turns out to be useful in the numerical treatment of functional equations, since the dimension of the matrices obtained from the discretization of operators are strongly reduced with an evident computational saving.
On the other hand, the two operators behave in a completely different way in weighted \(L^p\)-norm. In fact, the equivalence (3.4) of Theorem 4 is not true for the “nontruncated” operator \(\{\mathcal {L}_{m,r}(w)\}_m\).
Moreover, all the constants in Corollary 3, Theorems 4 and 5 are independent of m and f, but depend on \(\theta \). For instance, the constant in the upper bound of (3.4) is \(\mathcal {O}\left( (1-\theta )^{-3/4}\right) \) for \(\theta \rightarrow 1\) and so \(\theta \) cannot assume value 1.
We refer to Sect. 4 for more details.
4 Proofs
First of all we recall some weighted polynomial inequalities with the weight u defined in (2.5) which will be used in the sequel (see [21]).
Let \(1\le p \le \infty \) and \(a_m=a_m(u)\sim m^{1/\beta }\). For any \(P_m\in \mathbb {P}_m\), the restricted range inequalities
and
hold with \(\mathcal {C}\ne \mathcal {C}(m,P_m)\) and \(c \ne c(m,P_m)\). Moreover, we recall the Bernstein and Markov inequalities
and
where \(\mathcal {C}\ne \mathcal {C}(m,P_m)\) in both cases. Finally, for \(1\le p<\infty \), we will need the following Nikolskii inequality
We also need some estimates for the polynomials of the orthonormal system \(\{p_m(w)\}_{m\in \mathbb {N}}\), where w is the weight defined in (2.1). The generalized Laguerre weight w is a nonstandard weight and the estimates for the related orthogonal polynomials can be deduced by [7, 11, 12, 21, 22].
The estimate (which can be deduced from [9], see also [19, 21])
holds with \(x \in [0, +\infty )\) and \(x_d\) a zero closest to x. From (4.6) it follows that
Moreover, the following proposition will be useful.
Proposition 8
Let w be the weight defined by (2.1) and \(\{p_m(w)\}_m\) its related orthonormal system. Then we have
and
where \(\mathcal {C}\) and the constant in \(``\sim ''\) are independent of m.
Proof
Equivalence (4.8) easily follows from (4.6) for \(x=0\).
In order to prove (4.9) we set
Then, using (4.4) with u replaced by g, by (4.6), we obtain
Hence, for \(x=0\), by (4.8), we get
Finally, let us show (4.10) by induction. For \(k=1\), by (4.9) we have
For \(k>1\), from the identity
it follows that
Then, using the induction hypothesis and (4.9), we obtain
which completes the proof. \(\square \)
Furthermore, the following relation will be useful (see [10, 21])
By (4.6) and (4.11) if \(x\in \left[ \left( \frac{\sqrt{a_m}}{m}\right) ^2, a_m\right] \) we have
Moreover, if \(x_d\) is a node closest to x, we get (see [4])
Proof of Lemma 1
We are going to prove (2.12) only for \(p<\infty \), since the case \(p=\infty \) is simpler.
Let \(P_M \in \mathbb {P}_M\), with \(M=\left\lfloor \frac{\theta m}{\theta +1}\right\rfloor \), be a polynomial of quasi best approximation for \(f\in L^p_u\). So we can write
Since
then, letting \(v^r(x)=x^r\), we get
By (4.1) the second summand on the right-hand side is dominated by
and, using (4.12) and (4.13), we have
for some \(\tau >0\). Moreover, by (4.2) and (4.5), we obtain
Let us now prove (2.13). By (4.14) we have
For the first term on the right-hand side we use the following estimate (see [21])
While, for the second term on the right-hand side, using s times the Bernstein inequality (4.3) and (4.16), we obtain
Hence we get (2.13). \(\square \)
We recall the following estimate, concerning the distance \({\varDelta } x_k\) between two consecutive zeros (see [10, 21]):
where the constants in “\(\sim \)” are independent of m and k. We remark that from (4.17), with j given by (2.3), it follows that
Now, letting
and
we can state the following proposition.
Proposition 9
With the notation (4.18) and (4.19), we have
if and only if
Moreover, for \(p\ge 1\), we get
if and only if
Proof
We first prove (4.20). Recalling (4.19), we may write
where
and
Then we have
On the other hand, by (4.1), (3.2) and (4.8), we get
Combining (4.25) and (4.26) in (4.24), we obtain
since \(\displaystyle \sum \nolimits _{i=\eta +1}^{r-1}\left( {\begin{array}{c}i\\ \eta \end{array}}\right) \frac{1}{i!} \sim 1\).
Let us now show (4.22). Recalling (4.24), we have
where we have applied (4.1), (4.6), (4.8), taking also into account the assumption (4.23). Then recalling (4.25), we deduce (4.22).
The necessity of the conditions (4.21) and (4.23) can be proved using standard arguments (see [22]). We omit the details. \(\square \)
Proof of Theorem 2
Let us first prove that conditions (3.2) imply inequality (3.1). Recalling (2.2) and Proposition 9, it remains to prove that
Let \(x \in I_m=\left[ c\frac{a_m}{m^2}, a_m\right] \) and \(x_d\) be a zero closest to x. By using (4.7), (4.11) and (4.13), we get
By (4.6) the last summand on the right-hand side is bounded by \(\left( \frac{x}{a_m}\right) ^{r+\gamma -\frac{\alpha }{2}-\frac{1}{4}}\) and so the whole right-hand side is dominated by
Therefore, if the parameters \(r, \gamma \) and \(\alpha \) fulfill the assumptions (3.2), the sum on the right-hand side is bounded by \(\mathcal {C}(\log m)\) and the last term is bounded by 1.
Let us now prove that inequality (3.1) implies conditions (3.2). To this aim, for any \(f\in C_u\), consider the piecewise linear function \(f_1\) defined as follows
By (3.1) we have
where, for \(x\in [0,1]\) and \(x_k\in [1,2]\),
since \(x-x_k<0\). Now, if \(x\in [0,1]\) and \(x_k\in [1,2]\), then \(|x-x_k|\le 2\) and \(\frac{a_m-x}{a_m-x_k} \sim 1\). Hence, setting
and using (4.11), we get
Combining (4.28) and (4.29), we obtain
with \(\mathcal {C}\ne \mathcal {C}(m,f)\), whence letting \(b_i=|fu|(x_i)\) and \(\Vert \underline{b}\,\Vert ^*_\infty =\displaystyle \max _{1\le x_i\le 2}b_i\), we have
and then
Hence we get
Obviously the sum on the left-hand side is \(\sim \! 1\). Moreover, by (4.6), we have
So, from
it follows that \(r+\gamma \ge \frac{\alpha }{2}+\frac{1}{4}\).
In order to prove the necessity of the other inequality in (3.2), we introduce the piecewise linear function \(f_2\) defined as
Using similar arguments we obtain
Now, the sum on the left-hand side is \(\sim \int _{x_1}^1 x^{-r-\gamma +\frac{\alpha }{2}+\frac{1}{4}}\,\text {d} x\), while the norm is bounded. So, from
we deduce \(r+\gamma \le \frac{\alpha }{2} + \frac{5}{4}\) and the first part of the theorem follows.
Let us now prove (3.3). Since \(\left| f^{(i)}\right| \) is quasi increasing on \([0,\alpha _m]\), with \(i\in \{0,1,\ldots , r-1\}\) and \(\alpha _m=a_m/m^2\), we have
Recalling a formula in [3, p.15], it follows that
Hence (3.1) becomes
Letting \(P_{m+r} \in \mathbb {P}_{m+r}\) be the polynomial of best approximation for \(f\in C_u\), it follows that
i.e. (3.3). \(\square \)
In order to prove Theorem 4 we recall some known results concerning the Hilbert transform H(f, t). The Hilbert transform related to the interval (0, a) is defined as follows
where the integral is understood in the Cauchy principal value sense. Letting \(v^\rho (x)=x^\rho \) and \(1<p<\infty \), the bound
holds for any f such that \(fv^\rho \in L^p(0,a)\) if and only if \(-\frac{1}{p} < \rho < 1 - \frac{1}{p}\) (see [23]).
The following lemma, that can be found in [21] (see also [8]), will be also useful in the proof of Theorem 4.
Lemma 10
Let \(0<\theta <1,\, j\) be given by (2.3), \(u \in L^p\), \(1\le p < \infty \) and \({\varDelta } x_k= x_{k+1}- x_k\). Then, for an arbitrary polynomial \(P\in \mathbb {P}_{l m}\) (where l is a fixed integer), there exists \(\bar{\theta }\in (\theta ,1\)) such that
with \(\mathcal {C}\ne \mathcal {C}(m,f)\).
We want to emphasize that the first Marcinkiewicz inequality (4.32) does not hold if the sum on the left-hand side is extended to all the zeros of \(p_m(w)\), i.e. \(k=1,\ldots , m\) (see [13, 14]).
Proof of Theorem 4
Taking into account Proposition 9 it remains to estimate the \(L_u^p\)-norm of the first term in (2.4). Letting \(v^r(x)=x^r\) and
this term can be rewritten as \(v^r L_{m}^*\left( w,\frac{f}{v^r}\right) \). So, using (4.1), we have
where \(I_m=\left[ c\,a_m m^{-2},a_m\right] \) and
By using (4.11), we have
Denoting by \(G(x_k)\) the integral on the right-hand side and using the Hölder inequality, we get
For any polynomial \(0<Q\in \mathbb {P}_{lm}\), with l a fixed integer, we can write
and G(t) is a polynomial of degree \(m+lm\). Thus, using Lemma 10, we have
where \(0<\theta <\bar{\theta }<1\) and \(I_{\bar{\theta }m}=[c\,a_m m^{-2},\bar{\theta }a_m]\). Then we obtain
Let us estimate the term \(B_1\). By (4.7) we have
Moreover, under the assumptions (3.5), \(t^{-r-\gamma +\frac{\alpha }{2}+\frac{1}{4}}\in L^q\) and \(t^{r+\gamma -\frac{\alpha }{2}-\frac{1}{4}}\in L^p\), we can apply (4.31), obtaining
In order to estimate \(B_2\) we can construct a polynomial \(Q\in \mathbb {P}_{lm}\) such that \(Q(x) \sim x^r u(x)\) for \(x \in I_m\). So, using (4.7), we have
Then, by (4.31), we get
So, taking into account (4.36), (4.37) and (4.35), it follows that
and combining this last inequality with (4.34) and (4.33), the first term in (2.4) is bounded by
Therefore, recalling (4.22), we have that (3.5) implies (3.4).
In order to prove that (3.4) implies (3.5) we use arguments similar to those in the proof of Theorem 4. Therefore, we are going to show only the main steps. Considering the function \(f_1\) as in (4.27) we can write
where \([c\,a_m m^{-2},1]=[0,1]\cap I_m\). Recalling the expression of \(\mathcal {L}_{m,r}^*(w,f_1,x)\) for \(x\in [c\,a_m m^{-2},1]\), with
we obtain
Now, to simplify the notation, we set
Hence (4.38) becomes
Taking the supremum on \(\underline{b}\), we get
which implies \(r+\gamma - \frac{\alpha }{2} - \frac{1}{4} > -\frac{1}{p}\), taking into account that, by (4.7), \(\left| \widetilde{A}(x)\right| \le \mathcal {C}, x\in [c\,a_m m^{-2},1]\).
Analogously, considering the function \(f_2\) defined in (4.30), we get
i.e.
whence \(-r-\gamma + \frac{\alpha }{2} +\frac{1}{4} > -\frac{1}{q}\) that is \(r+\gamma - \frac{\alpha }{2} - \frac{1}{4} < 1 -\frac{1}{p}\), which completes the proof. \(\square \)
Proof of Theorem 5
Since \(\left| f^{(i)}\right| \) is quasi increasing on \([0,\alpha _m]\), with \(i\in \{0,1,\ldots , r-1\}\) and \(\alpha _m=a_m/m^2\), using the Hölder inequality, we have
Recalling a formula in [3, p.15], it follows that
So inequality (3.7) becomes
Let \(P_M \in \mathbb {P}_M\), with \(M=\left\lfloor \frac{\theta m}{\theta +1}\right\rfloor \), be a polynomial of quasi best approximation for \(f\in L^p_u\), and set \(Q=\mathcal {L}_{m,r}^*(w,P_M)\). Hence, by (4.39), we get
Now, to estimate the first summand at the right-hand side we can use (2.12) and (2.8). For the second summand we note that
So, by (2.13), taking into account that
inequality (3.8) follows. \(\square \)
Proof of Corollary 6
Proceeding as in the proof of Theorem 5, since \(\left| f^{(i)}\right| \) is quasi increasing on \([0,\alpha _m]\), with \(i\in \{0,1,\ldots , r-1\}\) and \(\alpha _m=a_m/m^2\), and \(f\in W_r^p(u)\), we get
Hence, by (3.7), (2.6) and (4.40), we obtain (3.9).
Now, in order to prove (3.10), let \(Q=\mathcal {L}_{m,r}^*(w,P_M)\), where \(P_M \in \mathbb {P}_M\), with \(M=\left\lfloor \frac{\theta m}{\theta +1}\right\rfloor \), is a polynomial of quasi best approximation for \(f\in L^p_u\). By (3.9) we have
Using Lemma 1, (2.10) and (2.6), we obtain (3.10). \(\square \)
References
Balázs, K.: Lagrange and Hermite interpolation processes on the positive real line. J. Approx. Theory 50(1), 18–24 (1987)
Cvetkovic, A.S., Milovanovic, G.V.: The mathematica package “Orthogonal Polynomials”. Facta Univ. Ser. Math. Inform. 19, 17–36 (2004)
Ditzian, Z., Totik, V.: Moduli of Smoothness, Springer Series in Computational Mathematics 9. Springer, New York (1987)
Erdős, P., Turán, P.: On interpolation. III. Interpolatory theory of polynomials. Ann. Math. (2) 41, 510–553 (1940)
Freud, G.: Convergence of the Lagrange interpolation on the infinite interval. Mat. Lapok 18, 289–292 (1967)
Frammartino, C.: A Nyström method for solving a boundary value. In: Gautchi, W., Mastroianni, G., Rassias, T. (eds.) Approximation and Computation. Springer Optim. Appl. 42, pp. 311–325. Springer, New York (2011)
Kasuga, T., Sakai, R.: Orthonormal polynomials with generalized Freud-type weights. J. Approx. Theory 121, 13–53 (2003)
Laurita, C., Mastroianni, G.: \(L^p\)-Convergence of lagrange interpolation on the semiaxis. Acta Math. Hungar. 120(3), 249–273 (2008)
Levin, A.L., Lubinsky, D.S.: Christoffel functions, orthogonal polynomials, and Nevai’s conjecture for Freud weights. Constr. Approx. 8(4), 463–535 (1992)
Levin, A.L., Lubinsky, D.S.: Orthogonal Polynomials for Exponential Weights, CSM Books in Mathematics/Ouvrages de Mathématiques de la SMC, 4. Springer, New York (2001)
Levin, A.L., Lubinsky, D.S.: Orthogonal polynomials for weights \(x^{2\rho } e^{-2Q(x)}\) on \([0, d)\). J. Approx. Theory 134(2), 199–256 (2005)
Levin, A.L., Lubinsky, D.S.: Orthogonal polynomials for weights \(x^{2\rho } e^{-2Q(x)}\) on \([0, d)\), II. J. Approx. Theory 139(1), 107–143 (2006)
Lubinsky, D.S., Mastroianni, G.: Mean convergence of extended Lagrange interpolation with Freud weights. Acta Math. Hungar. 84(1–2), 47–63 (1999)
Lubinsky, D.S., Mastroianni, G.: Converse quadrature sum inequalities for Freud weights. II. Acta Math. Hungar. 96(1–2), 147–168 (2002)
Mastroianni, G., Milovanović, G.V.: Interpolation Processes. Basic Theory and Applications, Springer Monographs in Mathematics. Springer, Berlin (2008)
Mastroianni, G., Milovanovic, G.V.: Some numerical methods for second kind Fredholm integral equation on the real semiaxis. IMA J. Numer. Anal. 29, 1046–1066 (2009)
Mastroianni, G., Milovanović, G.V., Notarangelo, I.: On an interpolation process of Lagrange-Hermite type. Publication de l’Institut Mathématique Nouvelle série tome 91(105), 163–175 (2012)
Mastroianni, G., Notarangelo, I.: A Lagrange-type projector on the real line. Math. Comput. 79(269), 327–352 (2010)
Mastroianni, G., Notarangelo, I.: Some Fourier-type operators for functions on unbounded intervals. Acta Math. Hungar. 127(4), 347–375 (2010)
Mastroianni, G., Occorsio, D.: Lagrange interpolation at Laguerre zeros in some weighted uniform spaces. Acta Math. Hungar. 91(1–2), 27–52 (2001)
Mastroianni, G., Szabados, J.: Polynomial approximation on the real semiaxis with generalized Laguerre weights. Stud. Univ. Babes-Bolyai LII(Number 4), 89–103 (2007)
Mastroianni, G., Vértesi, P.: Fourier sums and Lagrange interpolation on \((0,+\infty )\) and \((-\infty,+\infty )\). In: Govil, N.K., Mhaskar, H.N., Mohapatra, R.N., Nashed, Z., Szabados, J. (eds.) Frontiers in Interpolation and Approximation, Dedicated to the Memory of A. Sharma, pp. 307–344. Taylor & Francis Books, Boca Raton, Florida (2006)
Muckenhoupt, B., Wheeden, L.: Two weight function norm inequalities for the Hardy-Littlewood maximal function and the Hilbert transform. Stud. Math. 55(3), 279–294 (1976)
Nevai, P.: On Lagrange interpolation based on Laguerre roots [Hungarian]. Mat. Lapok 22, 149–164 (1971)
Nevai, P.: On the convergence of Lagrange interpolation based on Laguerre roots, [Russian]. Publ. Math. 20, 235–239 (1973)
Nevai, P.: Lagrange interpolation at zeros of orthogonal polynomials. In: Lorentz, G.G. (ed.) Approximation Theory, pp. 163–201. Academic Press, New York (1976)
Shen, J., Wang, L.-L.: Some recent advances on spectral methods for unbounded domains. Commun. Comput. Phys. 5(2–4), 195–241 (2009)
Szabados, J.: Weighted Lagrange and Hermite-Fejér interpolation on the real line. J. Inequal. Appl. 1(2), 99–123 (1997)
Acknowledgments
The authors would like to thank the referees for carefully reading the manuscript and for their suggestions which contributed in improving the presentation of the paper.
Author information
Authors and Affiliations
Corresponding author
Additional information
The first author and the third authors were partially supported by University of Basilicata (local funds). The second author was supported by University of Basilicata (local funds) and by National Group of Computing Science GNCS–INDAM.
Rights and permissions
About this article
Cite this article
Mastroianni, G., Notarangelo, I. & Pastore, P. Lagrange–Hermite interpolation on the real semiaxis. Calcolo 53, 235–261 (2016). https://doi.org/10.1007/s10092-015-0147-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10092-015-0147-y
Keywords
- Hermite–Lagrange interpolation
- Approximation by algebraic polynomials
- Orthogonal polynomials
- Generalized Laguerre weights
- Real semiaxis