Discrete least squares approximation of piecewise-linear functions by trigonometric polynomials Текст научной статьи по специальности «Математика»

Probl. Anal. Issues Anal. Vol. 6(24), No. 2, 2017, pp. 3-24 3

DOI: 10.15393/j3.art.2017.4070

UDC 517.521.2

G. G. Akniyev

DISCRETE LEAST SQUARES APPROXIMATION OF PIECEWISE-LINEAR FUNCTIONS BY TRIGONOMETRIC POLYNOMIALS

Abstract. Let N be a natural number greater than 1. Select N uniformly distributed points tk = 2nk/N (0 < k <

< N — 1) on [0, 2n]. Denote by L„,n(f) = L„,w(f,x) (1 <

< n < N/2) the trigonometric polynomial of order n possessing the least quadratic deviation from f with respect to the system {tk}N="01. In this article approximation of functions by the polynomials Ln,N(f,x) is considered. Special attention is paid to approximation of 2n-periodic functions fi and f2 by the polynomials Ln,N(f, x), where fi(x) = |x| and f2(x) = signx for x G G [—n, n]. For the first function fi we show that instead of the estimation |fi(x) — Ln,N(fi,x)| < clnn/n which follows from the well-known Lebesgue inequality for the polynomials Ln,N (f, x) we found an exact order estimation |fi(x) — Ln,N(fi,x)| < c/n (x G R) which is uniform with respect to 1 < n < N/2. Moreover, we found a local estimation |fi(x) — Ln,N(fi,x)| < c(e)/n2 (|x — nk| > e) which is also uniform with respect to 1 < n <

< N/2. For the second function f2 we found only a local estimation |f2(x) — Ln,N(f2,x)| < c(e)/n (|x — nk| > e) which is uniform with respect to 1 < n < N/2. The proofs of these estimations are based on comparing of approximating properties of discrete and continuous finite Fourier series.

Key words: function approximation, trigonometric polynomials, Fourier series

2010 Mathematical Subject Classification: 41A25

1. Introduction. Let N be a natural number greater then 1. Let {tk lyti—)1 be a set of points on [0, 2n] where tk = 2nk/N. Denote by Ln,N(f) = Ln,N(f, x) (1 < n < [N/2J) a trigonometric polynomial of

©Petrozavodsk State University, 2017

[MglHl

order n possessing the least quadratic deviation from the function f with respect to the system {tk }fc=0 . In other words, the minimum of the sums N=01 |f (tk) — Tn(tk)|2 on the set of trigonometric polynomials Tn of order n is attained when Tn = Ln,N (f). In particular, L|_n/2j,n (f, tk) = = f(tk). It is easy to show (see [12]) that for n < N/2 the polynomial Ln,N (f, x) can be represented as

n 1 N-1

Ln,N(f,x) = £ cVN)(f)e-, cVN)(f) = - £ f(tk)e-ivtk,

v=—n k=0

and for n = N/2 (for an even N)

Ln/2,N (f,x) = Ln/2 —1,N (f,x)+ «N/2 (f )cos N/2(x — u), (1)

where

N1

«n2n)(f) = «NN/2(f) = N E f (tk) cos N/2(tk — u). (2)

N k=0

To read more about function approximation by trigonometric polynomials see [1, 3, 5, 6], [8]-[11], [13]. In this article we obtain estimations for |Ln,N(fi, x) fi(x)| and |Ln,N(f2,x) —f2(x)| as n, N ^ to, where fi(x) = = |x|, f2(x) = signx, x G [—n, n]. The following theorems are proved:

Theorem 1. Let f1 (x) = |x|, x G [—n, n] and n < |_N/2J. The following estimations hold:

|Ln,N(fi,x) — fi(x)| < c/n, x G [—n,n],

|Ln,N(fi,x) — fi(x)| < c(e)/n2, x G AJ(e).

Theorem 2. Let f2(x) = signx, x G [—n, n] and n < |_N/2J. The following estimation holds:

|Ln,N(f, x) — f (x)| < c(e)/n, x G AJ(e).

We begin with some notation. Denote by

ck(f ) = 2n/f (t)6-'''dt, k G Z,

the Fourier coefficients of a function f, and by

n

f(x) = £Ck(f)eikx, Sn(f,x) = £ Ck(f)eikx

keZ k=-n

the Fourier series of a function f and its partial sum of order n, respectively. Denote by AJ(e) the set [—n + e, —e] U [e, n — e], where 0 < e < n/2. By c and c(e) we denote some positive constants that depend only on specified parameters; these constants may be different in different places.

Lemma 1. [12] If the Fourier series of f converges at the points tk = = u + 2kn/N, then the representation

Ln,N (f, x) = Sn (f, x) + Rn,N (f, x),

where

2 ^ n

Rn,N(f, x) = - Dn(x — t) cos ^N(u — t)f (t)dt, (3)

^ — n

holds true when 2n < N.

The following estimation follows from this lemma:

|Ln,N(f,x) — f(x)|<|Sn(f,x) — f(x)| + |Rn,N(f,x)|, n < N/2. (4)

Let us consider the case 2n = N. From (1) and (4)

|Ln,2n(f, x) — f (x)| <

<|Sn—i(f,x) — f (x)| + |Rn—i,2n(f,x)| + |«n2n)(f )|, n = N/2. (5)

From (4) and (5) we can see that estimation of the values |Sn(f1, x) — —fi(x)|, |Sn(f2,x) — f2(x)|, |Rn,N(fi,x)|, |Rn,N(f2,x)|, |«n2n)(fi)|, and |an2n) (f2)| implies an estimation |Ln,N(f, x) — f(x)| for the functions fi and f2.

2. Estimations for |Sn(fi,x) — fi(x)| and |Sn(f2,x) — f2(x)|.

Lemma 2. For the value |Sn(fi,x) — fi(x)|, where fi(x) = |x|, x G G [—n, n], we have the following estimations:

|Sn(fi,x) — fi(x)| < c/n, x G [—n,n], (6)

(fi,x) - fi(x)| < c(e)/n2, x e AJ(e).

(7)

Proof. Using [2, p.443] or [4, p.690], we can get the following representation for f1 (x) on [—n, n]:

From the above we can obtain estimation (6) (using the fact that |cos(2k — 1)x| < 1). Also, if we apply the Abel transformation to the above equation, we get (7). □

Lemma 3. For the value |Sn(f2, x) — f2(x)|, where f2 (x) = sign x, x e e [—n, n] we have the following estimation:

The proof of this lemma is obtained in [7].

3. An estimation for Rn,N(fi,x). The following lemma takes place.

Lemma 4. For Rn,N (f1,x), n < N/2 the following estimations hold:

Note that the estimation for remainder (3) for the function f1 has the following form:

|Sn(f2,x) — f2(x)|< c(e)/n, x e AJ(e).

x e [—n, n],

where

n

Dn (x — t) = 1/2 cos k(x — t).

From above we have

|Rn,N(fi,x)| < |Rn,N(fi,x)| + |Rn,N(fi,x)|, (8)

where

1 ^ n

Rn,N(fi,x) = |t| cos^N(u — t)dt, (9)

, x) = —

n

2 ^ ^ n

Rn,N(fi, x) = — ^^ |t| ^^ cos k(x — t) cos^N(u — t)dt. (10)

, n ^=i—n k = i

Lemma 5. The value (fi,x)| has the following estimation:

(fi,x)| < c/n2, x G [—n, n]. Proof. Using the formula

cos(^Nu — ^Nt) = cos ^Nu cos ^Nt + sin ^Nu sin ^Nt (11) we can rewrite (9) as follows:

2 ^ n Rn,N(fi,x) = — ^^ cos^Nu t cos^Ntdt.

n ^=i 0 Then we calculate the integral using the previous equation:

n (—1)^N — 1 t cos ^Ntdt =-^—.

J (^N)2

0

From these equations we get

T>1 (f \ 2 ^ cos ^Nu ^ i^N

Rn,N (fi,x) = „2 ((—— ^

or

i 0, N = 21,

Rn,N (fi,x) = \ 4 ^ cos(2^- i)Nu a7" _ 07 i 1

' l — nN2 ^ (2^-1)2 , N = 21 + 1

\ ^=i

Now we can estimate | Rn n (fi, x) |:

|Rn,N(fi,x)| < -N2 E

nN2 ^ (2^ — 1)2'

1

It is known that

So, using (12) we obtain

E

n

1 (2^ - 1)2 8

(12)

n n

\Rn,N (/i,x)\ < WT^ <

2N2 " 8n2

□

Lemma 6. The value |Rn n(fi,x)| has the following estimation for x e e [—n, n] :

| Rn,N (fi,x)| <-, x e [—n,n].

n

Before proving Lemma 6, we prove another one: Lemma 7. The following estimations take place:

^cos(2i — 1)x

i=0

<

1

2 |sin x| '

y^ cos 2ix

i=0

<

|sin x|

y sin(2i + 1)x

i=i

<

|sinx| '

k-1

y sin 2ix

i=0

<

|sin x|

Proof. After some simple transforms we obtain the following:

E cos(2i — 1)

- 1)x =

i=i

sin x

y sin x cos(2i — 1)

- 1)x =

i=i

1

2 sin x

y (sin (x — (2i — 1)x) + sin(x + (2i — 1)x) =

i=i

1 \—^ / • n n / 1 \ \ sin 2kx

(sin 2ix — sin 2 (i — 1) x) =

2 sin x

i=i

2 sin x

Likewise, obtain

k

E

i=i

cos 2ix =-V sin x cos 2ix =

sin x

i=i

sin(2k + 1)x — sin x 2 sin x

1

1

1

1

1

k

v^ . cos 2x — cos(2k + 2)x

> sin(2i + 1)x =---—,

2 sin x

i=i

cos x — cos(2k + 1)x

y sin2ix =-.

2 sin x

i=i

Now the proof is complete. □ Proof. (Lemma 6) From (11) and

cos k(x — t) = cos kx cos kt + sin kx sin kt (13)

(10) can be rewritten as follows:

n

4 to n „

Rn,N (f i, x) = — ^^ cos ^Nu ^^ cos kx t cos kt cos ^Ntdt+ n ^=i k=i 0

n

4 to n „

+— ^^ sin ^Nu ^^ sin kx t sin kt sin ^Ntdt.

n ,, — 1 7--1 J

n

^=i k=i

From above and the formulas

cos kt cos ^Nt = 1/2 (cos (^N — k) t + cos (^N + k) t),

sin kt sin ^Nt = 1 /2 (cos (^N — k) t — cos (^N + k) t) we can obtain

2 to

Rn N (fi, x) = — / ^ cos ^Nux

Ai=i /

V

n I n n \

£ cos kx J t cos (^N — k) tdt + J t cos (^N + k) tdt J +

00

n

x

k=i

2 to

+— ^^ sin ^Nux

n / ? ?

x y^ sin kx / t cos (^N — k) tdt — t cos (^N + k) tdt

k=i 0

0

After calculating the integrals, we have 2 TO

Rn N (A,x) = — / cos

n N

X V ((— 1)^N+fc — 1) cos kx -1-2 + -1-2 +

' V(^N — k)2 (,N + k)2 1

2 TO

+— ^^ sin mNux X V ((— 1)^N+k — 1) sin kx(-1-2--1-2 ). (14)

¿T( ) J V (MN — k)2 (MN + k)2' ( )

We can rewrite this as follows:

■ TO n / .. ..

|Rn N (fi,x)| < - V V -2 + -2 ) +

| , (f , )|< n (MN — k)2 (mN + k)2 I

4 oo n

7T EE

2<

n i=ii=iV (mN — k)2 (mn + k) 1

a (MN <nN i=i M2 k=N (1 — lN)

„ to n ^ _ to ^ n . _

12 1 12 1 1 4n 2n

< - > > -o < - > - > -o < - < -.

< n (,,.N - k)2 < n^M^^A n < N < -

□

Lemma 8. The value |Rn N(fi,x)| has the following estimation for x e e AJ (e) : ,

|RU(fi,x)| < = #, x e A1 (e).

1 , 1 -2 sin e -2

Proof. To prove this lemma, we use (14). It is easy to show that for

even N

4 to 4 TO

R2n (fi,x) =--V cos (x) — 4 V sin (x), (15)

where

tn/21 / 1 1 An,N (x) = V cos(2k — 1)x -2 +-2

n ( ) k=i( ) V (MN — (2k — 1))2 (MN + (2k — 1))2

1

1

tn/21

(x) = £ sin(2k - 1)x

k=i

(juN - (2k - 1))2 (uN + (2k - 1))2

And, therefore

.to .to

Kn(fi,x)| = - £ |(x)| + 4 £ |Bn,N(x)| .

I I n 1 1 n 1 1

Consider the values |An,N (x)| and |Bn,N (x)|. Apply the Abel transformation for An,N (x):

(x) =

+

(^N - (2 [n/2] - 1))2 (^N + (2 [n/2] - 1))2

x

[n/2] tn/21-1

x £ cos(2i-1)x- £

i=i

k=i

+

(uN - (2k + 1))2 (^N + (2k + 1))2

1

+

(uN - (2k - 1))2 (^N + (2k - 1))2

^ cos(2i - 1)x.

i=i

From this we get the following estimation:

(x)| <

+

(UN - (2 [n/2] - 1))2 (uN + (2 [n/2] - 1))2

x

x

[n/21

£ cos(2i - 1)

i=i

[n/21-1 + £

k=i

(UN - (2k + 1))2 (uN + (2k + 1))2

1

+

(UN - (2k - 1))2 (uN + (2k - 1))2

y cos(2i - 1)x

i=0

Using Lemma 7, we can rewrite the obtained estimation:

|sin x| \ (uN - (2 [n/2] - 1))2

+

(UN + (2 [n/2] - 1))2 / < (uN)2 |sinx|

10

(16)

1

1

1

1

1

1

1

1

1

1

1

1

2

1

1

Using the similar approach, we get an estimation for |Bn,N (x)|:

8

|Bn,N(x)| < vN|5nx• (17)

Using (15), (16), and (17) we can write

a to a to

|Rn,N(fi,x)| < 4 £ |(x)| + 4 £ |Bn,N(x)| <

<4 V 10 4 ^ 8 < 26 N -21

< n ^ (^N)2 |sinx| + n ^ 3^3N2 |sinx| < N2 |sinx|, = •

" " (18) If N is odd, then we can rewrite (14) as follows:

Rn ,N (fi,x) =

oo

A ^^ A

4 V Cn,N(x) cos(2^ — 1)Nu — - V D^,n(x) sin(2^ — 1)Nu—

.to . to

— 4 V En,N (x) cos 2^Nu —

V En,N (x) cos 2^Nu — 4 V F£,N (x) sin 2^Nu,

where

Ln/2J

(x) = £ cos 2kx I -2 +

k=1 Ln/2J

(x) =

((2^ — 1)N — 2k)2 ((2^ — 1)N + 2k)2

l_n/2j /

D^,n (x) = £ sin 2kx k=1

((2^ — 1)N — 2k)2 ((2^ — 1)N + 2k)2

tn/21 . En,N(x) = £ cos(2k—1)x -2 +

k=1 fn/21

(2^N — (2k — 1))2 (2^N + (2k — 1))2

(x) = V sin(2k—1)x ( -1-2 — .

k= V(2^N — (2k — 1))2 (2^N + (2k — 1))2

1

1

1

1

.00 ^00

4 ^ , 4

And R,n(/i,x)\ <-E \Cn,N(x)\ + -E \Di,N(x)\ +

1 1 7T ' 1 1 7T ' 1 1

A to /1 to

+ - E\En,N(x)\ + - E\FTN(x)\ .

7T 1 1 7T 1 1

Using the Abel transformation, we can easily obtain these estimations: C,N (x)\<,. 10 2 A„ , \D^,N (x)\ < 8

|sin x| N2' 1 " 1 |sin x| N2' (*)|<,. 2 2 A7~2 , (x)|< 8

|sinx| ^2N2' ' ^ ' -9 |sinx| N2 ' From this we have

(/>-x)l< • N = 21 + 1 (19)

From (18) and (19) we have (for every N > 2)

1 2 1 14n 7n 7n i

N( f 1 ,x)l < , .-t < ^ 01 •-r < ^ o 1 .-r, x G A (e).

I ^ N2 |sinx| < 2n2 |sinx| < 2n2 |sine|' w

□

4. An estimation for ai2n)(/1). When 2n = N from (2) we have the following:

.. 2n-1 ^ 2n-1 ✓ , s

ai2n) (/) = ^ E /(tk) cos nk = - £ (-1)'/ u + n- . (20) k=0 k=0 ^ 7

Subtract n from u to get (provided that / is 2n periodic function)

h E1(-i)k / ((«- n)+nk)=in E1(-i)k / (u+^

k=0 v 7 k=0 v

=2nE (-i)k-1/ (u+?)=- bE (-i)k / (u+?) =

27 2^'(-1)'+ +(-D2"f (u + 2")

k=1 v 7

¿17 E^)* /(« + = -an2"' (/ ).

Therefore, if we subtract or add — from u, we get

<e> (/) = <—^ EM)k /((u+3 +

In other words, adding — (l e Z) to u does not change the value of

«2n)(f)

, so we can assume, without loss of generality, that 0 < u < E.

It holds for both

an2n) (fi) and an2n) (f2)

(2n)

Lemma 9. The following estimation takes place:

«n2n)(fi) < , x e [—n, n].

n2

Proof. It is easy to show that

«nf = 2n E (—D"

ni

"=-n

ni

k

--+ u

n

7T

, 0 < u < —.

n

«n2n) (fi) = 2n E (—1)"

" = n

k

--+ u

n

i

2n

(—1)

1

2n

k=-n

n

k

---u

n

ni

Em)

k

--

k=i

1 n k

n

u

+2n £W( ?+») = " =0 n1

+ 2nE ("DM T + -) =

" =0

n k

2n

E(—1)k(-—^ —

"=1

n

1 £ (_1)kY i)+ui =

2n

"=1

n

1 £ (—W £—^—=2n £ (—1)k (n—2u

2n

"=1

n

"=1

Get the following estimation:

«n2n)(fi)

<

1

2n

£ (^Hn — 2«

1 ||

< — --2u

< 2n n

<

7T

2n2

□

1

"

5. An estimation for Rn,N(f2, x). We prove the following lemma. Lemma 10. The following takes place:

|Rn,N(f2,x)|< ^, x e AJ(e).

n

To prove this lemma we estimate |Rn,N(fi,x)| as follows:

|Rn,N (f2,x)| < R,N (f2 ,x)| + |Rn,N (f2 ,x) | , (21)

where

1 to r

#nN (f2,x) = — 7 cos ^N (u — t)sign tdt, (22)

, Tn

n

2 to „ n

Rn,N (f2 ,x) = 2 ^^ X^cos k(x — t) cos ^N(u — t)sign tdt. n ^=i-7r fc=i

First we prove the following lemmas:

Lemma 11. The following takes place:

|Ri,N (f2, x) | < n Proof. Using (11) for (22) we have

1 to n

Ri n (f2,x) = — 7 cos ^N (u — t)sign tdt =

1 to n

y / (cos ^Nu cos ^Nt + sin ^Nu sin ^Nt) sign tdt

(

71 i

n

n

1 to

— > cos ^Nu cos ^Ntsign tdt + sin ^Nu sin ^Ntsign tdt ) =

i /

1 to n 2 to n

= — > sin ^Nu sin ^Ntsign tdt = — > sin ^Nu sin ^Ntdt.

7T i J 7T i J

From above and

sin ^Ntdt = —

cos ^Nt

^N

1 — (— ^N

we have

2 ^

(f — V(1 — (— 1)"N)

N

^=1

sin ^Nm

If N is even, then N(/2,x) = 0, otherwise

(f2,x) = — ^J] ^=1

4 ^ sin (2^ — 1) Nu

nN ^ 2^ — 1

where

Therefore

□

4 ^^ sin(2^ — 1)x

7F

- 2^ — 1

= |/2(x)|< 1-

11

Kn(f2,x)i < N <

Lemma 12. Denote Yn(x) = 5^(1 — (—1)k+m)ak cos kx, where all

In I n

k=1

> 0, the sequence m=1 is monotone, and m = 0,1. Then

lYn(x)| <

|sin x|

Proof. Assume that > ak+1. After applying the Abel transformation we get

Yn(x) = £(1 — (—1)k+m)afc cos kx

k=i

cos kx =

n1

= an

£(1 — (— 1)i+m) cos ix — £ (ak+1 — ak) £(1 — (—1)i+m)

cosix.

i=1

k=1

i=1

n

n

0

c

From the above equation and Lemma 7 we have

lYn(x)| < a

- (-l)i+m)

cos ix

i=1 n — 1

+

+ £ (ak+1 - ak)

k=i

£(1 - (-1)*+m)

cos ix

i=1

<

^ 1 f \ ^ 2an

< --- (an + an - ao) < —

|sin x|

|sin x| |sin x|

□

Lemma 13. If vn(x) = (1 — (— 1) m)ak sin kx, where all ak > 0,

k=i

the sequence ak is monotone, and m = 0,1. Then

|Vn (x)| <

|sin x|

The proof of this lemma is analogous to the proof of Lemma 12. Lemma 14.

y sin kx

k=1

<

1

lsin 2 I

-, Vn G N.

Proof. From

sin kx =-- y^ sin kx

Z—/ Qin x '

k=1

sin x k=1

x

sin —

we can easily obtain the equality

n

y sin kx =

k=1

sin x sin n x sin x '

which gives us the desired estimation ^n=1 sin kx| < , .1 x ,. □

|sin x |

Lemma 15. The following takes place:

c(£)

,n(f2,x)| , x G A1 (e).

1 1 n

c

Proof. From (11) and (13) we have

n

2 to n

R2 (/2,x) = -£ £cos k(x — t) cos aN (u — t)sign tdt

, x) = —

k=1

2 to

y sin ^Nu ^^ cos kx / (cos kt sin aNt) sign tdt+

7T

^=1 k = 1

2 to

— n n

+— ^^ cos ^Nu ^^ sin kx (sin kt cos aNt) sign tdt

7T

^=1 k = 1

— n

n

4 c^o n «

y sin ^Nu ^^ cos kx cos kt sin aNtdt+ n ^=1 k = 1 0

4 oo n «

+— ^^ cos ^Nu ^^ sin kx

n U = 1 k = 1 n

n

4 to n

— cos ^Nu sin kx I sin kt cos aNtdt. n ^=1 fc=1 0

Using the formula sin a cos ft = 1 (sin (a — + sin (a + ft)) and calculating the integrals from the above equation, we get |Rn N(/2,x)| < < | Rn,N (/2 ,x) | + | R^V (/2 ,x) | , where '

RnVv (/2, x) = - V sin ^Nu V (1 — (—1)^v+k) cos kx-^-,

n'N ^ ) ^ ( ) ; (^N )2 — k2,

.to n ,

R^v (/2, x) = - £ cos ^Nu £ (1 — (—1)^v+k) sin kx———.

n ^=1 k=1 (a )

We estimate values |Rn'V(/2, x)| and ^^V(/2,x)| separately. Begin with

|R?Jv (/2,x)|. ' '

Rn''V (/2 ,x) = ^ £ £(1 — (—1)^v+k )cos kx ---„

^ = 1 a k = 1 1 (^v)2— k2

nN ^ a ^ V 7 7 1 - —k2

2 £ sin^ £(1 —(—1)^v+k)cos k^ 1+ k2

m=1 a fc=1 V

nN a v 7 7 V (aN)2 — k2y

7T

We can rewrite it as follows:

T-.2 1 2 ^^ sin ^Nm

Rn,N (f2,x) = --- 2^cos kx +

nN ^ -

2 ^ (—sin -Nu ,

> -—----> (—1)k cos kx+

nN ^ - y

2 ^^ sin -Nu

+--— > - > cos kx-

N - (

k2

- (-N )2 — k2

+ V (—sin -Nu ^ (—1)kk2

^=1

-

k = 1

(-N )2 — k2

Now we have

2

<JV (f ,x)| < ^

E

sin -Nu

-

+

2

+

(—sin -Nu -

k=i

y cos kx +

n

£(—1)k cos kx

k=i

E

2 ^^ sin -Nu , k2

—— > - > cos kx

N -

+

(-N )2 — k2

+

+

2 ^ (—sin -Nu ^ , (—1)k k2 > - > cos kx-

N

-

k=i

(-N )2 — k2

From [4, p. 448] we get

2

7T

E

sin -Nu -

Also, from Lemma 7

< 1 and —

7T

E

^=1

(—sin -Nu

-

cos kx

k=i

<

| sin x|

£(—1)k cos kx

k=i

<

| sin x|

1

1

Using the above estimations we can write

R. N (/ ,x)| <

+

N |sinx| nN

to , T n

sin uNu

> - > cos kx

k2

p=1

u

k=i

(UN)2 - k2

+

+

2

nN

^ (-1)pn sin uNu ^ , (-1)kk2 > - > cos kx-

p=1

u

k=i

(UN)2 - k2

. (24)

To estimate the value

2

sin uNu y^- y^ cos kx

k2

p=1

u

k=i

(UN)2 - k2

make some transformations:

sin uNu

2

sin uNu

> - cos kx

k2

p=1

u

k=i

(UN)2 - k2

2

sin uNu

¿v-3— cos kx

p=1

U

k=1

n 2 - 4 p2

<

<

2 xTO 1

nN E U p=1

cos kx

k=1 -1 /

N2 - 4

A v —

nN ^ u3

n

N2 - ^ ■ !

p2 j=1

y cosjx—

£

k=1

(k + 1)2

z

N 2 - ^ N2 - £ 7=1 p2 p / j=1

cos jx

<

<

2 A 1

E

2n

2

nN u3 \ N2 - ^ p=1 \ p2

-1

+£

k=1

(k + 1)2

p k2

cos jx

j=1

+

N2 _ (W N2 -

cos jx

j=1

<

2 ^ 1

< ——^E

2n2

1

nN |sinx| ^ uM N2 - 4 N2 - -1

<

2

2

2

k

2

k

2

k

<

8 ^ < 4 3nN |sinx| u3 < nN |sinx|

1 1 u=1

Using the similar approach, prove

2

nN

to £

^=1

(-1)^N sin^Nu ^ , (-1)kk2 cos kx

U

k=i

(uN)2 - k2

<

4

N | sin x|

Now we can rewrite (24) as follows

1

|<N (/2 ,x)| < Now consider |Rn.'2v (/2, x) |.

,2 + 8

N | sin x|

N | sin x|

(25)

4

R2,2v(/2, x) = - £cosuNu£ (1 - (-1)^N+k) sin kx ^=1

k

k=1

nN ^ u ^=1

4 to 1

I**(/2-x)|s nN £U2

i toto at

4 £ cosuNn £(1 - (_1)^)sinkx

k2 - (uN)2 k

k=1

^ _ 1 2 N 1

£(1 - (-1)^N+k)sinkx-

N

k

k = 1

1 1

^2N

Using the approach we used for estimating |Rn'N(/2,x)|, prove that

RnN (/2, x) | <

N | sin x|

(26)

From the obtained estimations (25) and (26), and the inequalities n < N/2 and |sine| < |sinx| for x G AJ(e), we have (23). □

From (21) and Lemmas 11 and 15 we have

|Rn,N(/2, x)| < 1 (c + c(e)) = ^, x G AJ(e). n n

So, Lemma 10 is proved.

6. An estimation for ai2n) (/2). We prove a lemma:

c

Lemma 16. The following estimation takes place:

42n) (/2)

c

< -. n

Proof. Using formula (20) we get (for a 2n-periodic function f)

1 2n-1 1 2n-1

ai2n)(f) = 2n E f (tk)cosnk = - £ (-1)kf (u + nk/n). k=0 k=0

As has been mentioned earlier, we can safely assume that 0 < u < n/n; so we write

n — 1

a(?n) (/2) = 1/2n £ (-1)k/2 (nk/n + u), 0 < u < n/n.

n

1 / -1 n-1

ann)(/2) = 2n E (-i)k+1 + E(-i)k+

Vk=-n+1 k=1

+ (-1)nf2 (U - n) + f2 (u) j = 2- (f2(u) - ( l)nf2 (U)) . From this we get

aL2n)(f2)| < 1/n.

□

7. Proofs of Theorems 1 and 2. From (4) and (5) we have |Ln,N (fl,x) - fl (x)| < |Sn (fl,x) - fi(x)| + |Rn,N (fl ,x)|, n < N/2,

lLn,2n (f1,x) - f1 (x)| <

< |Sn-1 (f1,x) - f1 (x)| + |Rn-1,2n(f1,x)| + |«n2n)(f1 )|, n = N/2. From Lemmas 2, 4, 9 we easily get

|Ln,N(f1,x) - f1 (x)| < c/n, x G [-n,n], |Ln,N(f1,x) - f1(x)| < c(e)/n2, x G AJ(e).

Theorem 1 is proved.

Using the same technique, inequalities (4) and (5), and Lemmas 3, 10

and 16, we get

|Ln,N(f,x) - f (x)| < c(e)/n, x G AJ(e).

So, theorem 2 is also proved.

References

[1] Bernshtein S. N. On trigonometric interpolation by the method of least squares. Dokl. Akad. Nauk USSR, 1934, vol. 4, pp. 1-5. (in Russian)

[2] Courant R. Differential and Integral Calculus. New Jersey. : Wiley-Interscience, 1988, vol. 1, 704 p.

[3] Erdos P. Some theorems and remarks on interpolation. Acta Sci. Math. (Szeged), 1950, vol. 12, pp. 11-17.

[4] Fikhtengol'ts G. M. Course of differential and integral calculus. Moscow. : FIZMATLIT, 1969, vol. 1, 656 p. (in Russian)

[5] Kalashnikov M. D. On polynomials of best (quadratic) approximation on a given system of points. Dokl. Akad. Nauk USSR, 1955, vol. 105, pp. 634636. (in Russian)

[6] Krilov V. I. Convergence of algebraic interpolation with respect to the roots of a Chebyshev polynomial for absolutely continuous functions and functions with bounded variation. Dokl. Akad. Nauk USSR, 1956, vol. 107, pp. 362-365. (in Russian)

[7] Magomed-Kasumov M. G. Approximation properties of de la Valle-Poussin means for piecewise smooth functions. Mat. Zametki, 2016, vol. 100, is. 2, pp. 229-244. D0I:10.1134/S000143461607018X

[8] Marcinkiewicz J. Quelques remarques sur l'interpolation. Acta Sci. Math. (Szeged), 1936, vol. 8, pp. 127-130. (in French)

[9] Marcinkiewicz J. Sur la divergence des polynômes d'interpolation. Acta Sci. Math. (Szeged), 1936, vol. 8, pp. 131-135. (in French)

[10] Natanson I. P. On the Convergence of Trigonometrical Interpolation at Equi-Distant Knots. Annals of Mathematics, Second Series, 1944, vol. 45, no. 3, pp. 457-471. D0I:10.2307/1969188.

[11] Nikol'skii S. M. On some methods of approximation by trigonometric sums. Mathematics of the USSR - Izvestiya, 1940, vol. 4, pp. 509-520. (in Russian)

[12] Sarapudinov I. I. On the best approximation and polynomials of the least quadratic deviation. Anal. Math., vol. 9, is. 3. pp. 223-234.

[13] Zygmund A. Trigonometric Series. Cambridge. : Cambridge University Press, 1959, vol. 1, 747 p.

Received October 11, 2017. In revised form, December 13, 2017. Accepted December 15, 2017. Published online December 27, 2017.

Dagestan scientific center of RAS

45, Gadzhieva st., Makhachkala 367025, Russia

E-mail: hasan.akniyev@gmail.com