Available online at www.tjnsa.com J. Nonlinear Sci. Appl. 9 (2016), 3090–3097 Research Article On approximation properties of certain multidimensional nonlinear integrals Sevgi Esen Almalia,∗, Akif D. Gadjievb a Faculty of Sciences and Art, Depatment of Mathematics, Kirikkale University, Kirikkale,Turkey. b National Academy of Sciences of Azerbaijan, 9 F. Agaev Street, 1141, Baku, Azerbaijan. Communicated by J. Brzdek Abstract We prove theorems on convergence of multidimensional nonlinear integrals in Lebesgue points of generated function, and show that the main results are applicable to a wide class of exponentially nonlinear integral operators, which may be constructed by using well known positive kernels in approximation theory. c 2016 All rights reserved. Keywords: Nonlinear integral operators, positive kernels, Lebesgue points, approximation, exponentially nonlinear integrals. 2010 MSC: 42A35, 42A63. 1. Introduction It is well known that the solution of boundary value problems for elliptic and parabolic differential equations can be expressed as integral operators with positive kernels. The most famous of these are the kernels of Gauss–Weierstrass, Poisson, Abel–Poisson, etc. Sequences or families of integral operators with positive kernels play an important role in approximation of functions and serve as the main model in the creation of the so called theory of approximation of positive linear operators [2, 6]. Approximative properties of the sequences of linear integral operators with positive kernels generated interest to study approximation by nonlinear integrals [3]–[5], [8]–[12]. Related to that, presumably it would be interesting to find a connection between the approximation by nonlinear integrals and boundary value problems for certain nonlinear differential operators. ∗ Corresponding author Email addresses: sevgi_esen@hotmail.com (Sevgi Esen Almali), akif_gadjiev@mail.us (Akif D. Gadjiev) Received 2015-09-30 S. E. Almali, A. D. Gadjiev, J. Nonlinear Sci. Appl. 9 (2016), 3090–3097 3091 On the other hand, given that the classical theorems on the convergence of sequences of operators are only valid for linear operators, approximation by using nonlinear integrals is of interest in terms of the convergence of sequences (or families) of these integrals and founding its limit values. The results of the papers [1, 7], on limiting properties of multidimensional integrals, called Riesz and Bessel potentials, can be considered as regarding this. The main goal of our work is to find appropriate expressions of limits of a family of certain nonlinear integrals. We will study nonlinear multidimensional integrals of the form Z λn Lλ (u, x) = K(λ |t − x| , u(t))dt, wn−1 Rn 1 2 where x ∈ Rn , |t − x| = (t1 − x1 )2 + (t2 − x2 )2 + · · · + (tn − xn )2 and wn−1 is the area of unit sphere S n−1 in Rn . The kernel-function K(ξ, u), depending on two variables ξ ≥ 0 and u, |u| ≤ M satisfy some properties which will be given in Section 2. We want to find the limit of Lλ (u, x0 ) as λ → ∞ where x0 is the Lebesgue point of the function u(x), x ∈ Rn . In Section 3 we will show that the kernel-function K(ξ, u), satisfying all our conditions, may be easily obtained by using classical positive kernels of approximation theory, such as Gauss–Weierstrass, Abel–Poisson, Jacson and other kernels. The main theorem, proved in Part 2, allow us to give a definition of a general subclass of nonlinear integrals Lλ (u, x0 ) exponentially depending of the function u(x), x ∈ Rn . 2. Main result First, we recall the well known definition of Lebesgue points in Rn . Let S n−1 be the unit sphere in Rn and let for x ∈ Rn , Z f (r, u(x)) = |u(x + rθ) − u(x)| dθ, 0 < r < ∞. S n−1 Definition 2.1. The point x ∈ Rn is called the Lebesgue point of function u ∈ L1 (Rn ) if Z 1 lim |u(x − t) − u(x)| dt = 0. r→0 r n |t|≤r From this definition, we can find an δ > 0 such that Z 1 |u(x − t) − u(x)| dt < δ rn |t|≤r provided r ≤ δ. Let us define a new function by Z F (ρ) = ρ rn−1 f (r, u(x0 ))dr. (2.1) 0 Definition 2.1 gives that for any ε > 0 there exist a positive number δ = δ(ε) such that F (ρ) < ερn , for all ρ ≤ δ, (2.2) dF (ρ) = ρn−1 f (ρ, u(x0 ))dρ. (2.3) (see [11]). Moreover, from (2.1) we have S. E. Almali, A. D. Gadjiev, J. Nonlinear Sci. Appl. 9 (2016), 3090–3097 Consider now the following nonlinear multidimensional integral Z λn Lλ (u, x) = K(λ |t − x| , u(t))dt, wn−1 Rn 3092 (2.4) where, as we noted above, wn−1 is the area of unit sphere S n−1 in Rn and λ is a positive parameter. Note that for partial derivatives of the function K(ξ, u) with respect to the variable u, we will write for brevity ∂ m K(ξ, u) Kum (ξ, 0) = |u=0 , m = 1, 2, 3, . . . . ∂um We assume that the kernel function K(ξ, u) satisfy the following conditions: (a) For any ξ ≥ 0, K(ξ, u) is an infinitely differentiable function of variables u and ξ and K(ξ, 0) = 0. (m) (b) For any natural m, Ku (ξ, 0) is nonnegative monotonically decreasing function of ξ on semiaxis [0, ∞). (c) Kum (ξ, 0) ≤ Ku(m−1) (ξ, 0), (d) Z m = 1, 2, 3, . . . . ∞ Kum (ξ, 0)ξ n−1 dξ = cm < ∞, m = 1, 2, . . . . (2.5) 0 From the conditions(a)-(d), it follows that the kernel K(ξ, u) may be expanded in Maclaurin series by the powers of function u. Moreover, (c) gives that the first partial derivative Ku1 (ξ, 0) is a majorant function for other derivatives, that is Ku(m.) (ξ, 0) ≤ Ku1 (ξ, 0), m = 1, 2, . . . . From (d) we can infer the inequality Z p Z (m) n−1 (m) Ku (ξ, 0)ξ dξ ≥ Ku (p, 0) p 2 = p ξ n−1 dξ p 2 Ku(m) (p, 0)pn 1 1 1− n , m = 1, 2, . . . , 2 n for any positive number. Using (d) we obtain lim pn Ku(m) (p, 0) = 0, p→∞ and from this it follows that for any m = 1, 2, . . . , lim Ku(m) (p, 0) = 0. p→∞ The properties (d) gives also Z lim p→∞ p (2.6) ∞ Ku(m) (ξ, 0)ξ n−1 dξ = 0. (2.7) We will prove now the main theorem on convergence of the family Lλ (u, x) as λ → ∞ at a fixed point x0 , being the Lebesgue point of the function u(x). Theorem 2.2. Let u(x) be a bounded function in Rn , belonging to L1 (Rn ) and let the kernel K(ξ, u) satisfy conditions (a)–(d). Then at each Lebesgue point x0 of the function u(x) we have ∞ X cm m lim Lλ (u, x0 ) = u (x0 ), λ→∞ m! m=1 where cm are defined in (2.5). S. E. Almali, A. D. Gadjiev, J. Nonlinear Sci. Appl. 9 (2016), 3090–3097 3093 Proof. First we write the Maclurin expansion of the function K(ξ, u) by the powers of u. We have [1] ∞ X 1 (m) K (λ |t − x0 | , 0)um (t). m! u K(λ |t − x| , u(t)) = m=1 By condition (c) and boundedness of the function u(t), the series in right-hand side is majorating and therefore Z ∞ λn X 1 K (m) (λ |t − x0 | , 0)um (t)dt. Lλ (u, x0 ) = wn−1 m! Rn u m=1 Firstly, let us transform t − x0 = p and then take p = t Z ∞ λn X 1 Lλ (u, x0 ) = K (m) (λ |t| , 0)um (t + x0 )dt. wn−1 m! Rn u m=1 From binomial expansion of um (x0 + t) we can write Z ∞ m X λn X 1 k Lλ (u, x0 ) = Ku(m) (λ |t| , 0) Cm (u (x0 + t) − u(x0 ))m−k uk (x0 )dt, wn−1 m! Rn m=1 k=0 k are binomial coefficients. From this we obtain where Cm Z ∞ m−1 ∞ X X λn X 1 cm m k (m) Cm (u (x0 + t) − u(x0 ))m−k uk (x0 )dt. u (x0 ) = Ku (λ |t| , 0) Lλ (u, x0 ) − m! wn−1 m! Rn m=1 m=1 k=0 Since u(x) is bounded, say, |u(x)| ≤ M < ∞, we have m−1 m−1 X X k m−k k k Cm (2M )m−1−k M k C (u (x + t) − u(x )) u (x ) ≤ |u (x + t) − u(x )| 0 0 0 0 0 m k=0 k=0 ≤ (4M )m |u (x0 + t) − u(x0 )| . Therefore, using the property (c), we can write Z ∞ ∞ X cm m λn X (4M )m u (x0 ) ≤ |u (x0 + t) − u(x0 )| Ku1 (λ |t| , 0)dt. Lλ (u, x0 ) − wn−1 m m! n R m=1 m=1 So, for any fixed δ, which will be chosen below, we have Z Z ∞ X cm m λn u (x0 ) ≤ + |u (x0 + t) − u(x0 )| Ku1 (λ |t| , 0)dt Lλ (u, x0 ) − wn−1 m t<δ t≥δ m=1 = 1 wn−1 e4M (2.8) − 1 {λn Iλı + λn Iλıı } . From (b), for λn Iλıı , we immediately obtain Z ∞Z Z n ıı n 1 n−1 n λ Iλ ≤ λ Ku (λδ, 0) |u(x0 + ρθ| ρ dθdρ + |u(x0 )| λ Ku1 (λ |t| , 0)dt δ S n−1 t≥δ Z ∞ ≤ λn kukL1 (Rn ) Ku1 (λδ, 0) + λn u(x0 )wn−1 Ku1 (λρ, 0) ρn−1 dρ. δ The first term in right-hand side tends to zero as λ → ∞ by (2.6), and the second term by (2.7). Therefore lim λn Iλıı = 0. λ→∞ (2.9) S. E. Almali, A. D. Gadjiev, J. Nonlinear Sci. Appl. 9 (2016), 3090–3097 3094 Consider λn Iλı . As we noted above, assuming that x0 is a Lebesgue point of the function u, we obtain, by Definition 2.1, that for ε > 0 there exist a number δ > 0 such that the function F (ρ), defined in (2.1) satisfies the inequality (2.2). Fixing this δ > 0, we can write λn Iλı =λ n Z δ Z S n−1 0 |u(x0 + ρθ) − u(x0 )| Ku1 (λρ, 0) ρn−1 dθdρ, or, using (2.3) and integrating by parts λn Iλı =λ n Z δ Ku1 (λρ, 0) dF (ρ) 0 = λn Ku1 (λρ, 0) F (δ) + λn δ Z F (ρ)d −Ku1 (λρ, 0) . 0 Using (2.2), we have λn Iλı δ ρn d −Ku1 (λρ, 0) < ε (λδ) + ελ 0 Z δ Ku1 (λρ, 0) ρn−1 dρ = nελn 0 Z δ = nε Ku1 (ρ, 0) ρn−1 dρ. n Ku1 (λρ, 0) n Z 0 By (2.5), the limit of the integral in the right-hand side is finite and therefore lim λn Iλı = 0, λ→∞ (2.10) since ε is an arbitrary small number. Using (2.9) and (2.10) in (2.8), proves the theorem. We give the following simple example of an application of Theorem 2.2. Lets consider the kernel-function i 1 h 2 exp e−ξ u − 1 . K(ξ, u) = √ 2π (2.11) Obviously K(ξ, 0) = 0. We have 1 2 Ku(m) (ξ, 0) = √ e−mξ , 2π and all the conditions of Theorem 2.2 are satisfied. Therefore we use the following well known formula, Z ∞ Γ n2 1 n−1 −mρ2 dρ = ρ e . 2 m n+1 2 0 3. Exponentially nonlinear integrals The example of kernel K(ξ, u), given in formula (2.11), shows that Theorem 2.2 covers the subclass of nonlinear integrals which may be described by the following definition. Definition 3.1. We will say that a nonlinear integral operator contains the exponential nonlinearity if its kernel-function K(ξ, u) has the form K(ξ, u) = eA(ξ)u − 1, (3.1) where A(ξ) is a given function, defined for ξ ≥ 0. S. E. Almali, A. D. Gadjiev, J. Nonlinear Sci. Appl. 9 (2016), 3090–3097 3095 Definition 3.2. Nonlinear integral, having exponential nonlinearity, will be called exponentially nonlinear integral. As an example, consider the exponentially nonlinear integral Z h i λn Bλ (u, x) = eA(λ|t−x|)u(t) − 1 dt. wn−1 Rn (3.2) Following Theorem 2.2, we can see that the following statement holds. Theorem 3.3. Let a nonnegative function A(ξ) satisfy the following conditions: (a1 ) A(ξ) < 1 for all ξ ≥ 0, (b1 ) A(ξ) is monotonically decreasing function in semiaxis [0, ∞), R∞ (c1 ) 0 A(ξ)ξ n−1 dξ < ∞. Then at each Lebesgue point x0 of a bounded function u ∈ L1 (Rn ), we have lim Bλ (u, x0 ) = λ→∞ where dm = R∞ 0 ∞ X dm m u (x0 ), m! (3.3) m=1 Am (ξ)ξ n−1 dξ. (m) Proof. It is clear that for function (3.1) we have K(ξ, 0) = 0, Ku (ξ, 0) = Am (ξ), m = 1, 2, . . . , and by (a1 ) Ku(m) (ξ, 0) < Ku(m−1) (ξ, 0), and (b1 ) implies (b). Therefore, the conditions (a)–(c) of Theorem 2.2 are satisfied. Moreover, using (c1 ), we see that all conditions (a)–(d) of Theorem 2.2 are satisfied, which gives the proof. Remark 3.4. It is clear that Theorem 3.3 is a corollary of Theorem 2.2. We give this statement in the form of special Theorem since it covers many applicable cases. Now we consider the case of exponential nonlinear integrals where the function A(ξ) does not satisfy the condition of monotony. Theorem 3.5. Let A(ξ) be a nonnegative function such that there exist a monotonically decreasing majorant function D (ξ) on [0, ∞) such that Am (ξ) ≤ D (ξ) , m = 1, 2, . . . , and Z (3.4) ∞ ξ n−1 D(ξ)dξ < ∞. (3.5) 0 Then (3.3) holds at every Lebesgue point x0 of bounded function u ∈ L1 (Rn ). Proof. As in the proof of Theorem 2.2 we can write the inequality (2.8) for integral (3.2) in the following form ∞ X dm m λn B (u, x ) − u (x ) e4M − 1 {λn Iλı + λn Iλıı } , λ 0 0 ≤ wn−1 m! m=1 where λn Iλı Z |u (x0 + t) − u(x0 )| D(λ |t|)dt, = t<δ S. E. Almali, A. D. Gadjiev, J. Nonlinear Sci. Appl. 9 (2016), 3090–3097 and λn Iλıı = 3096 Z |u (x0 + t) − u(x0 )| D(λ |t|)dt. t≥δ Since D (ξ) is monotonically decreasing, we obtain that lim λn Iλı = lim λn Iλıı = 0. λ→∞ λ→∞ By the same way as above (2.9) and (2.10). The proof is complete. Consider the following Fejer-type kernel F (t) = sin t t 2n , and corresponding exponentially nonlinear integral (3.2) Z h i λn Fλ (u, x) = eF (λ|t−x|)u(t) − 1 dt. wn−1 Rn We have λn Fλ (u, x) = wn−1 ∞ X sin (λ |t − x|) 2nm Z Rn λ |t − x| m=1 um (t)dt. Obviously for any natural number n and m sin (λ |t − x|) λ |t − x| 2nm < sin (λ |t − x|) λ |t − x| Therefore, we have F (λ |t − x|) < 2n < 2n 1 + λ2 |t − x|2 2n 1 + λ2 |t − x|2 n . n , which means that the majorant function D (ξ) has the form D (ξ) = 2n , (1 + ξ 2 )n ξ ∈ [0, ∞) . It is easy to see that D (ξ) is a monotonically decreasing function on [0, ∞) and Z ∞ ξ 0 n−1 Z D (ξ) dξ ≤ 1 n 2 dξ + 2 0 n Z ∞ 1 dξ ξ n+1 < ∞. This show that all the conditions of Theorem 3.5 are satisfied. Acknowledgements The authors would like to thank the referee for valuable comments. References [1] I. A. Aliiev, A. D. Gadjiev, A. Aral, On approximation properties of a family of linear operators at critical value of parameter, J. Approx. Theory, 38 (2006), 242–253. 1, 2 [2] F. Altomare, M. Campiti, Korovkin-Type Approximation Theory And Its Applications, Walter de Gruyter, Berlin, (1994). 1 S. E. Almali, A. D. Gadjiev, J. Nonlinear Sci. Appl. 9 (2016), 3090–3097 3097 [3] L. Angeloni, G. Vinti, Convergence in variation and rate of approximation for nonlinear integral operators of convolution type, Results Math., 49 (2006), 1–23. 1 [4] C. Bardaro, J. Musielak, G. Vinti, Nonlinear integral operators and applications, Walter de Gruyter, Berlin, (2003). [5] C. Bardaro, G. Vinti, H. Karsli, Nonlinear integral operators with homogeneous kernels: pointwise approximation theorems, Appl. Anal., 90 (2011), 463–474. 1 [6] P. L. Butzer, R. J. Nessel, Fourier Analysis And Approximation, Academic Press, New York-London, (1971). 1 [7] A. D. Gadjiev, A. Aral, I. A. Aliev, On behaviour of the Riesz and generalized Riesz potentials as order tends to zero, Math. Ineq. Appl., 10 (2007), 875–888. 1 [8] H. Karsli, On approximation properties of non-convolution type nonlinear integral operators, Anal. Theory Appl., 26 (2010), 140–152. 1 [9] J. Musielak, Approximation by nonlinear singular integral operators in generalized Orlicz spaces, Comment. Math. Prace Mat., 31 (1991), 79–88. [10] L. Rempulska, A. Thiel, Approximation of functions by certain nonlinear integral operators, Lith. Math. J., 48 (2008), 451–462. [11] L. M. Stein, G. Weiss, Introduction To Fourier Analysis On Euclidean Spaces, Princeton University Press, New Jersey, (1971). 2 [12] T. Swiderski, E. Wachnicki, Nonlinear singular integral depending on two parameters, Comment. Math. Prace Mat., 40 (2000), 181–189. 1