Stanis̷law Stoi´ nski
On the rate of summability of Fourier series by means of the method of the Voronoi-N¨orlund
Abstract. In this note we present the theorem on the rate of pointwise summability of Fourier series by means of generalized Voronoi-N¨orlund sums.
2000 Mathematics Subject Classification: 42A10.
Key words and phrases: Fourier series, generalized sum, rate of summability, Lebes- gue point.
1. Preliminaries. Let us denote by 𝑀 = [𝑐𝑚𝑘], 𝑚 = 0, 1, 2, ...; 𝑘 = 0, 1, 2, ..., 𝑚, the infinite triangular matrix, where 𝑐𝑚0 > 0 for 𝑚 = 0, 1, 2, ..., 𝑐𝑚𝑘 ≥ 0 for 𝑚 = 1, 2, ...; 𝑘 = 1, 2, ..., 𝑚. Write 𝐶𝑚 = ∑𝑚
𝑘=0𝑐𝑚𝑘. For the num- ber series ∑∞
𝑛=0𝑎𝑛 with partial sums 𝐴𝑛 we make the sequence (𝑤𝑚)∞𝑚=0, where 𝑤𝑚= 1
𝐶𝑚
∑𝑚 𝑘=0
𝑐𝑚𝑚−𝑘𝐴𝑘
If 𝑤𝑚→ 𝐴 ∈ ℝ, then the number 𝐴 is called the generalized Voronoi-N¨orlund sum of the series ∑∞
𝑛=0𝑎𝑛. In the particular case, when 𝑐𝑚𝑘 = 𝑐𝑘 for 𝑚 = 0, 1, 2, ...;
𝑘 = 0, 1, 2, ..., 𝑚, we have the method of a generalized summation which is linear.
This summation method is regular if and only if 𝑐𝑚/𝐶𝑚→ 0 ([2], [6]).
Let 𝐿𝑝2𝜋, 𝑝≥ 1 and 𝐶2𝜋 be the spaces composed of all 2𝜋-periodic real-valued functions Lebesgue-integrable on the interval [−𝜋, 𝜋] with 𝑝-th power and all 2𝜋- periodic real-valued functions continuous on [−𝜋, 𝜋], respectively. In the case 𝑝 = 1 we shall write 𝐿2𝜋 instead of 𝐿12𝜋. Let us introduce in these spaces the usual norms
𝑓 𝐿𝑝 =( ∫ 𝜋
−𝜋∣𝑓(𝑡)∣𝑝𝑑𝑡)1𝑝 , 𝑓
𝐶= max{∣𝑓(𝑡)∣ : −𝜋 ≤ 𝑡 ≤ 𝜋}.
In [4] there were applicated the method of a generalized summation of the Voronoi-N¨orlund to the Fourier series of 𝑓 ∈ 𝐿2𝜋 putting 𝑐𝑚𝑘 = ( 𝑚
𝑚−𝑘
)𝑠𝑚−𝑘, 𝑚 = 0, 1, 2, ...; 𝑘 = 0, 1, 2, ..., 𝑚 for 𝑠 ∈ (0, 1]. Let 𝑆𝑛[𝑓 ](𝑥) denote the 𝑛-th partial sum of the Fourier series
𝑆[𝑓 ](𝑥) = 𝑎0
2 +
∑∞ 𝑚=1
(𝑎𝑚cos 𝑚𝑥 + 𝑏𝑚sin 𝑚𝑥)
of 𝑓 . Linearity and regularity of the method will be maintained if a sequence of mean arithmetic forms
(𝑉 𝑁 )𝑛[𝑓 ](𝑠, 𝑥) = 1 𝑛 + 1
∑𝑛 𝑚=0
1 (1 + 𝑠)𝑚
∑𝑚 𝑘=0
(𝑚 𝑘
)
𝑠𝑘𝑆𝑘[𝑓 ](𝑥) =
∫ 𝜋
−𝜋
𝑓 (𝑥 + 𝑡)𝒦𝑛(𝑠, 𝑡)𝑑𝑡,
where
𝒦𝑛(𝑠, 𝑡) = 1−(√
1+2𝑠 cos 𝑡+𝑠2 1+𝑠
)𝑛+1
cos(
(𝑛 + 1)arc tg1+𝑠 cos 𝑡𝑠 sin 𝑡 )
4𝜋(𝑛 + 1)1+𝑠𝑠 (sin2𝑡)2 for 𝑡∕= 0, is taken into consideration.
Let us write for 𝑓 ∈ 𝐿2𝜋
𝜑𝑥(𝑡) = 𝑓 (𝑥 + 𝑡)− 2𝑓(𝑥) + 𝑓(𝑥 − 𝑡), Φ𝑥(ℎ) =
∫ ℎ
0 ∣𝜑𝑥(𝑡)∣𝑑𝑡 (ℎ≥ 0) and let us introduce the following pointwise characteristic
𝜔𝑥(𝛿) = 𝜔𝑥(𝛿; 𝑓 ) = sup
0<ℎ≤𝛿
{1 ℎΦ𝑥(ℎ)
}
(see e.g. [1] and [5]),
˜
𝜔𝑥(𝛿) =𝜔˜𝑥(𝛿; 𝑓 ) = 1 𝛿Φ𝑥(𝛿)
(see e.g. [3]). If lim𝛿→0+𝜔˜𝑥(𝛿; 𝑓 ) = 0, then 𝑥 is called a Lebesgue point (L-point) of the second order of 𝑓 . For every L-point 𝑥 of 𝑓 we have 𝜔𝑥(𝛿)→ 0 as 𝛿 → 0+. If, in particular 𝑓 ∈ 𝐶2𝜋, then
(1) 𝜔𝑥(𝛿; 𝑓 )≤ 𝜔2(𝛿; 𝑓 )𝐶.
uniformly in 𝑥 ∈ [−𝜋, 𝜋], where 𝜔2(𝛿; 𝑓 )𝑐 = sup∣ℎ∣≤𝛿 𝜑(ℎ) 𝐶 is its modulus of smoothness. If 𝑓 ∈ 𝐿𝑝2𝜋, 𝑝 ≥ 1, and 𝜔2(𝛿; 𝑓 )𝐿𝑝 = sup∣ℎ∣≤𝛿 𝜑(ℎ) 𝐿𝑝 is its integral modulus of smoothness, then
(2)
˜𝜔(𝛿; 𝑓) 𝐿𝑝≤ 𝜔2(𝛿; 𝑓 )𝐿𝑝.
2. Main results. Now, we present estimates for the rate of poitwise conver- gence of (𝑉 𝑁 )𝑛[𝑓 ](𝑠, 𝑥) at the L-points of 𝑓 .
Let us write
Δ(𝑠, 𝑡) = 1 + 2𝑠 cos 𝑡 + 𝑠2= (1 + 𝑠)2− 4𝑠 sin2 𝑡 2, 𝛿𝑛= 𝛿𝑛(𝑠) = 2(1 + 𝑠)
√
(𝑛 + 1)𝑠 +(
(𝑛 + 1)𝑠(1 + 𝑠))2 for 𝑠∈ (0, 1], 𝑛 = 1, 2, ...
Theorem 2.1 If 𝑓 ∈ 𝐿2𝜋 and 𝑠∈ (0, 1], then at every real 𝑥
∣(𝑉 𝑁)𝑛[𝑓 ](𝑠, 𝑥)− 𝑓(𝑥)∣ ≤ (1 + 𝑠) 2(𝑛 + 1)𝑠
(𝜔˜𝑥(𝜋) + 2𝜋
∫ 𝜋 𝛿𝑛(𝑠)
˜ 𝜔𝑥(𝑡)
𝑡2 𝑑𝑡) , where 𝑛 is such large that 𝛿𝑛(𝑠)≤ 𝜋2.
Proof We can write
∣(𝑉 𝑁)𝑛[𝑓 ](𝑠, 𝑥)− 𝑓(𝑥)∣ ≤( ∫ 𝛿𝑛(𝑠)
0
+
∫ 𝜋 𝛿𝑛(𝑠)
)∣𝜑𝑥(𝑡)∣𝒦𝑛(𝑠, 𝑡)𝑑𝑡 = 𝑃𝑛(𝑠, 𝑥) + 𝑄𝑛(𝑠, 𝑥).
Using the estimates
1−( √1 + 2𝑠 cos 𝑡 + 𝑠2 1 + 𝑠
)𝑛+1
≤ 𝑐 sin2( 𝑡 2
) for 𝑐 = 2(𝑛 + 1)𝑠 (1 + 𝑠)2 and Δ(𝑠, 𝑡)≤ (1 + 𝑠)2, we obtain
𝑃𝑛(𝑠, 𝑥) ≤ (1 + 𝑠)𝜋 16(𝑛 + 1)𝑠
∫ 𝛿𝑛(𝑠) 0 ∣𝜑𝑥(𝑡)∣
𝑐𝑡2+ 8 sin2 (
1
2(𝑛 + 1)arc tg1+𝑠 cos 𝑡𝑠 sin 𝑡 )
𝑡2 𝑑𝑡
≤ (1 + 𝑠)𝜋
16(𝑛 + 1)𝑠(𝑐 + 2(𝑛 + 1)2𝑠2)
∫ 𝛿𝑛(𝑠)
0 ∣𝜑𝑥(𝑡)∣𝑑𝑡 and
𝑄𝑛(𝑠, 𝑥)≤ 1 + 𝑠 2𝜋(𝑛 + 1)𝑠
∫ 𝜋 𝛿𝑛(𝑠)
∣𝜑𝑥(𝑡)∣
sin2(2𝑡)𝑑𝑡≤ 𝜋(1 + 𝑠) 2(𝑛 + 1)𝑠
∫ 𝜋 𝛿𝑛(𝑠)
𝑑Φ𝑥(𝑡) 𝑡2 . Integration by parts, gives
𝑄𝑛(𝑠, 𝑥)≤ 𝜋(1 + 𝑠) 2(𝑛 + 1)𝑠
( Φ𝑥(𝜋)
𝜋2 −Φ𝑥(𝛿𝑛) 𝛿𝑛2 + 2
∫ 𝜋 𝛿𝑛(𝑠)
Φ𝑥(𝑡) 𝑡3
)𝑑𝑡.
Hence we obtain
∣(𝑉 𝑁)𝑛[𝑓 ](𝑠, 𝑥)− 𝑓(𝑥)∣ ≤ Φ𝑥(𝛿𝑛)( (1 + 𝑠) 4(𝑛 + 1)𝑠𝜋( 𝑐
4+1
2(𝑛 + 1)2𝑠2)
− 𝜋(1 + 𝑠) 2(𝑛 + 1)𝑠𝛿2𝑛
)
+ (1 + 𝑠)
2𝜋(𝑛 + 1)𝑠Φ𝑥(𝜋) +𝜋(1 + 𝑠) (𝑛 + 1)𝑠
∫ 𝜋 𝛿𝑛(𝑠)
Φ𝑥(𝑡) 𝑡3 𝑑𝑡.
In consideration of 𝛿𝑛 we have our result. ■
Corollary 2.2 If 𝑓 ∈ 𝑋, where 𝑋 = 𝐶2𝜋 or 𝑋 = 𝐿𝑝2𝜋, 𝑝≥ 1, then for 𝑠 ∈ (0, 1]
(3) (𝑉 𝑁)𝑛[𝑓 ](𝑠,⋅) − 𝑓(⋅) 𝑋 ≤ 1 + 𝑠 2(𝑛 + 1)𝑠
(𝜔2(𝜋; 𝑓∣𝑋+ 2𝜋
∫ 𝜋 𝛿𝑛(𝑠)
𝜔2(𝑡; 𝑓 )𝑋
𝑡2 𝑑𝑡) , where 𝑛 is such large that 𝛿𝑛(𝑠)≤ 𝜋/2.
Proof By the condition (1) it follows the estimate (3) for 𝑋 = 𝐶2𝜋. Appling the Minkowski inequality, we have
(𝑉 𝑁)𝑛[𝑓 ](𝑠;⋅) − 𝑓(⋅)
𝐿𝑝≤ 1 + 𝑠 2(𝑛 + 1)𝑠
(
˜𝜔⋅(𝜋; 𝑓 )
𝐿𝑝+ 2𝜋
∫ 𝜋 𝛿𝑛(𝑠)
˜ 𝜔⋅(𝑡; 𝑓 )
𝑡2 𝑑𝑡
𝐿𝑝
) . Using the generalized Minkowski inequality and the condition (2), we obtain
∫ 𝜋 𝛿𝑛(𝑠)
˜ 𝜔⋅(𝑡; 𝑓 )
𝑡2 𝑑𝑡 𝐿𝑝
≤
∫ 𝜋 𝛿𝑛(𝑠)
𝜔2(𝑡; 𝑓 )𝐿𝑝
𝑡2 𝑑𝑡.
Hence it follows the estimate (3) for 𝑋 = 𝐿𝑝2𝜋. ■ Corollary 2.3 If 𝑓 ∈ 𝐿2𝜋, then for 𝑠∈ (0, 1]
𝑛→∞lim(𝑉 𝑁 )𝑛[𝑓 ](𝑠; 𝑥) = 𝑓 (𝑥) at every L-point 𝑥 of 𝑓 .
Proof It is known ([5]) that almost all real numbers are the L-points of 𝑓 ∈ 𝐿2𝜋. Let 𝑥∈ ℝ be the L-point of 𝑓. For an arbitrary 𝜀 > 0 there exists Δ ∈ (0, 𝜋/2] such that 𝜔𝑥(𝛿; 𝑓 ) < 𝑠𝜀/(√
5𝜋) for 0 < 𝛿 < Δ. Since for 𝑠∈ (0, 1] we have 𝛿𝑛(𝑠)→ 0 with 𝑛→ ∞, so there exists 𝑛0∈ ℕ such that 𝛿𝑛0 = 𝛿𝑛0(𝑠) < Δ. Let us write
∫ 𝜋 𝛿𝑛0
˜ 𝜔𝑥(𝑡)
𝑡2 𝑑𝑡 =
∫ Δ
𝛿𝑛0
+
∫ 𝜋
Δ. The following estimations are true
∫ Δ
𝛿𝑛0
˜ 𝜔𝑥(𝑡)
𝑡2 𝑑𝑡≤ 𝑠𝜀
√5𝜋 ( 1
𝛿𝑛0
− 1 Δ
)
and ∫ 𝜋
Δ
˜ 𝜔𝑥(𝑡)
𝑡2 𝑑𝑡≤ 𝜔𝑥(𝜋) Δ . Hence we have
∣(𝑉 𝑁)𝑛0[𝑓 ](𝑠, 𝑥)− 𝑓(𝑥)∣ < 𝑊 (𝑥, 𝑛0, 𝑠) + 𝜀 2, where
𝑊 (𝑥, 𝑛0, 𝑠) = 1 + 𝑠 2(𝑛0+ 1)𝑠
((2𝜋 Δ + 1
)
𝜔𝑥(𝜋)− 2 𝑠𝜀
√5Δ )
.
Taking the natural number 𝑛1 ≥ 𝑛0 such that 𝑊 (𝑥, 𝑛1, 𝑠) < 𝜀2, we obtain our
result. ■
Acknowledgement. I would like to express my gratitude to the Reviewer for his invaluable critical remarks and suggestions.
References
[1] S. Aljanˇci´c, R. Bojanic and M. Tomi´c, On the degree of convergence of Fej´er-Lebesgue sums, L’Enseignement Mathematique, Geneve, Tome XV (1969), 21–28.
[2] G. H. Hardy, Divergent series, Oxford 1949.
[3] W. ̷Lenski, On the rate of pointwise strong (𝐶, 𝛼) summability of Fourier series, Colloquia Math. Soc. J´anos Bolyai, 58 Approx. Theory, Kecskem´et (Hungary) (1990), 453–486.
[4] S. Stoi´nski, Approximation of periodic functions by means of certain linear operators with regard to Hausdorff metric, Funct. Approximatio Comment. Math. 1 (1974), 123–131.
[5] R. Taberski, Aproksymacja funkcji wielomianami trygonometrycznymi , Wyd. Nauk. UAM, Pozna´n 1979.
[6] K. Zeller, Theorie der Limitierungsverfahren , Springer Verlag, Berlin-G¨ottingen-Heidelberg 1958.
Stanis̷law Stoi´nski
Faculty of Mathematics and Computer Science, Adam Mickiewicz University Umultowska 87, 61-614 Pozna´n, Poland
E-mail: stoi@amu.edu.pl
(Received: 17.12.2008)