Skip to main content
Log in

Positive Results and Counterexamples in Comonotone Approximation

  • Published:
Constructive Approximation Aims and scope

Abstract

We estimate the degree of comonotone polynomial approximation of continuous functions f, on [−1,1], that change monotonicity s≥1 times in the interval, when the degree of unconstrained polynomial approximation E n (f)≤n α, n≥1. We ask whether the degree of comonotone approximation is necessarily ≤c(α,s)n α, n≥1, and if not, what can be said. It turns out that for each s≥1, there is an exceptional set A s of α’s for which the above estimate cannot be achieved.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Similar content being viewed by others

References

  1. Babenko, K.I.: On some problems of approximation theory and numerical analysis. Usp. Mat. Nauk SSSR, Ser. Mat. 22, 3–28 (1985)

    Google Scholar 

  2. Ditzian, Z., Totik, V.: Moduli of Smoothness. Springer, New York (1987)

    Book  MATH  Google Scholar 

  3. Dzyadyk, V.K., Shevchuk, I.A.: Theory of Uniform Approximation of Functions by Polynomials. Walter de Gruyter, Berlin (2008)

    MATH  Google Scholar 

  4. Gilewicz, J., Shevchuk, I.A.: Comonotone approximation. Fundam. Prikl. Mat. 2, 319–363 (1996) (In Russian)

    MathSciNet  MATH  Google Scholar 

  5. Kopotun, K.A.: Uniform estimates of monotone and convex approximation of smooth functions. J. Approx. Theory 80, 76–107 (1995)

    Article  MathSciNet  MATH  Google Scholar 

  6. Kopotun, K.A., Leviatan, D., Shevchuk, I.A.: Convex polynomial approximation in the uniform norm: conclusion. Can. J. Math. 57, 1224–1248 (2005)

    Article  MathSciNet  MATH  Google Scholar 

  7. Kopotun, K.A., Leviatan, D., Shevchuk, I.A.: Are the degrees of best (co)convex and unconstrained polynomial approximation the same? Acta Math. Hung. 123, 273–290 (2009)

    Article  MathSciNet  MATH  Google Scholar 

  8. Leviatan, D.: Pointwise estimates for convex polynomial approximation. Proc. Am. Math. Soc. 98(3), 471–474 (1986)

    Article  MathSciNet  MATH  Google Scholar 

  9. Leviatan, D., Shevchuk, I.A.: Monotone approximation estimates involving the third modulus of smoothness. In: Chui, C.K., Schumaker, L.L. (eds.) Approx. Theory IX, pp. 223–230. Vanderbilt University Press, Nashville (1998)

    Google Scholar 

  10. Leviatan, D., Shevchuk, I.A.: Some positive results and conterexamples in comonotone approximation II. J. Approx. Theory 100, 113–143 (1999)

    Article  MathSciNet  MATH  Google Scholar 

  11. Leviatan, D., Shevchuk, I.A.: More on comonotone polynomial approximation. Constr. Approx. 16, 475–486 (2000)

    Article  MathSciNet  MATH  Google Scholar 

  12. Lorentz, G.G., Zeller, K.L.: Degree of approximation by monotone polynomials II. J. Approx. Theory 2, 265–269 (1969)

    Article  MathSciNet  MATH  Google Scholar 

Download references

Acknowledgements

The authors are grateful to the referees for improving the presentation of the paper.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to D. Leviatan.

Additional information

Communicated by: Pencho Petrushev.

Part of this work was done while the third author visited Tel Aviv University.

Appendix

Appendix

For the sake of completeness, we include the proof of the fact that the inequalities (5.1) imply (2.6).

Let \(\varphi(x):=\sqrt{1-x^{2}}\), and write \(C^{0}_{\varphi}:=C[-1,1]\); for r≥1, we say that \(f\in C^{r}_{\varphi}\) if fC (r)(−1,1) and lim x→±1 φ r(x)f (r)(x)=0.

Finally, for \(f\in C^{r}_{\varphi}\), we write

$$\omega^\varphi_{k,r}\bigl(f^{(r)},t\bigr):=\sup_{0\le h\le t}\sup_{x:|x|+\frac{kh}{2}\varphi(x)<1} K^r\biggl(x,\frac{kh}{2}\biggr)\bigl|\varDelta _{h\varphi(x)}^k\bigl(f^{(r)},x\bigr)\bigr|,$$

where K(x,μ):=φ(|x|+μφ(x)) and the symmetric difference \(\varDelta _{u}^{k}\) is defined in (3.1).

Note that for r=0,

$$\omega^\varphi_{k,0}(f,t)\equiv\omega_k^\varphi(f,t),$$

the kth Ditzian–Totik modulus of smoothness.

Recall the Chebyshev knots x j :=cos(/n),0≤jn, and denote I j :=[x j ,x j−1], 1≤jn. Note that J j =I j I j+1, 1≤jn−1. Let Σ k,n be the collection of all continuous piecewise polynomials of degree <k, on the Chebyshev partition \(\{ x_{j}\}_{j=0}^{n}\). Also, denote by Σ k,n (Y s )⊂Σ k,n , the set of all piecewise polynomials S with the following property. Let j(i), 1≤is, be such that y i I j(i); then SΣ k,n (Y s ) if for every 1≤is, the restriction of S to (x j(i)+1,x j(i)−2)=:O i is a polynomial, where x n+l :=−1, x l :=1. Finally, write \(O:=\bigcup_{i=1}^{s}O_{i}\).

Proposition 3

([10])

For every k≥1 and s≥0, there are constants c=c(k,s) and c =c (k,s) such that if \(Y_{s}\in \mathbb{Y}_{s}\) and SΣ k,n (Y s )∩Δ 1(Y s ), n≥1, then

$$E^{(1)}_{c_*n}(S,Y_s)\leq c\omega_k^\varphi(S,1/n).$$

Theorem 5

Given sN, let α≥1. Suppose that \(f\in \varDelta ^{1}_{s}\) is such that (2.4) holds for nr+1 and the inequalities (5.1) are satisfied. Then

$$ n^{\alpha}E_{n}^{1,s}(f)\le c_1(\alpha,s),\quad n\ge c_2(\alpha,s)N_0.$$
(A.1)

Proof

Let fΔ (1)(Y s ). Evidently, it suffices to prove (A.1) for \(E_{n}^{(1)}(f,Y_{s})\) instead of \(E_{n}^{1,s}(f)\) (recall that \(E^{(1)}_{n}(f,Y_{s})\) and \(E_{n}^{1,s}(f)\) are defined in (2.2) and (2.3)). First, let

$$\sigma_{r+1,n}^{(1)}(f):=\inf\bigl\{\|f-S\|: S\in \varSigma _{r+1,n}(Y_s)\cap \varDelta ^1(Y_s)\bigr\},$$

and assume that (the analog of (A.1)),

$$n^{\alpha}\sigma_{r+1,n}^{(1)}(f)\leq c_3(\alpha,s),\quad n\ge N_1.$$

Take SΣ r+1,n (Y s )∩Δ 1(Y s ), so that \(\|f-S\|\leq2\sigma_{r+1,n}^{(1)}(f,Y_{s})\). Then by Proposition 3, we conclude that

$$E_n^{(1)}(S,Y_s)\le c\omega_{r+1}^\varphi \biggl(S,\frac{1}{n}\biggr),\quad n\geq c_*,$$

which, in turn, implies

$$E_n^{(1)}(f,Y_s)\le E_n^{(1)}(S,Y_s)+2\sigma_{r+1,n}^{(1)}(f,Y_s)\le c\omega_{r+1}^\varphi \biggl(f,\frac{1}{n}\biggr)+c\sigma_{r+1,n}^{(1)}(f,Y_s)$$

for all nc . Since (2.4) implies that \(\omega_{r+1}^{\varphi }(f,1/n)\leq cn^{-\alpha}\), we only need to verify (A.1) for N 1≥max{c ,c 3(α,s)N 0}, c 3(α,s) to be prescribed.

Let nc 3(α,s)N 0, and let \(O_{n}=\bigcup_{i=1}^{s'}O_{i}'\) be a partition of O n into connected intervals. First we prove that there is an mn/c 2 such that:

  1. (i)

    for each 1≤is′, \(O_{i}'\subset J_{j(i),m}\), and

  2. (ii)

    for i 1i 2, we have that either j(i 1)=j(i 2) or {j(i 1),j(i 1)+1}∩{j(i 2),j(i 2)+1}=∅.

To this end, we proceed by induction on s′≤s. If s′=1, then there is just one block \(O_{1}'\) of length ≤2s+1, so that we may take \(m=\lceil\frac{n}{2s+1}\rceil\). If s′>1, let \(n_{1}=\lceil\frac{n}{2s+1}\rceil\), then property (i) holds for m=n 1. If property (ii) is also valid, then we are done. Otherwise,

$$O_n=\bigcup_{i=1}^{s'}O_i'\subset\bigcup_{i=1}^{s'}J_{j(i),m}=:\bigcup_{i=1}^{s''}O_i''$$

is a new partition into connected intervals, where s″<s′. We apply the induction step to s″ to get the desired value of m. It is readily seen that we may take c 3(α,s)≥(2s+1)s.

By virtue of (5.1), for the appropriate m, for every 1≤is′, there is a polynomial p i , of degree r, comonotone with f on J j(i),m , satisfying p i (x j(i)+1,m )=f(x j(i)+1,m ) and such that

$$\|f-p_i\|_{J_{j(i),m}}\le\frac{c(\alpha,s)}{m^{\alpha}}\le \frac{c_3(\alpha,s)}{n^{\alpha}}.$$

For jj(i),j(i)+1, 1≤is′, f is monotone on I j,m , so that, by (5.1), we have a polynomial \(\tilde{q}_{j}\), of degree r, comonotone with it there, such that

$$\|f-\tilde{q}_j\|_{I_{j,m}}\le\frac{c_4(\alpha,s)}{n^{\alpha}}.$$

Adding the linear

we obtain q j that is comonotone with f on I j,m , interpolates f at endpoints of I j,m , and satisfies

$$\|f-q_j\|_{I_{j,m}}\le\frac{3c_4(\alpha,s)}{n^{\alpha}}.$$

Now we take piecewise polynomial s n to be defined p i on J j(i),m and q j on I j,m for the other intervals. Then s n is comonotone with f and satisfies

$$\|f-s_n\|\leq\frac{c(\alpha,s)}{n^{\alpha}}.$$

It may have discontinuities at the points x j(i)−1,m , but there are no more than s such points, so if we make it continuous by moving from left to right and making s n (x j(i)−1,m +):=s n (x j(i)−1,m −), then our estimate is still valid with c 1(α,s):=sc(α,s). This completes the proof. □

Rights and permissions

Reprints and permissions

About this article

Cite this article

Leviatan, D., Radchenko, D.V. & Shevchuk, I.A. Positive Results and Counterexamples in Comonotone Approximation. Constr Approx 36, 243–266 (2012). https://doi.org/10.1007/s00365-012-9159-x

Download citation

  • Received:

  • Revised:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00365-012-9159-x

Keywords

Mathematics Subject Classification (2000)

Navigation