# On an optimal interpolation formula in K_2(P_2) space

The paper is devoted to the construction of an optimal interpolation formula in K_2(P_2) Hilbert space. Here the interpolation formula consists of a linear combination ∑_β=0^NC_β(z)φ(x_β) of given values of a function φ from the space K_2(P_2). The difference between functions and the interpolation formula is considered as a linear functional called the error functional. The error of the interpolation formula is estimated by the norm of the error functional. We obtain the optimal interpolation formula by minimizing the norm of the error functional by coefficients C_β(z) of the interpolation formula. The obtained optimal interpolation formula is exact for trigonometric functions sinω x and cosω x. At the end of the paper, we give some numerical results which confirm our theoretical results.

02/15/2021

### On an optimal quadrature formula for approximation of Fourier integrals in the space W_2^(1,0)

The present paper is devoted to construction of an optimal quadrature fo...
09/06/2019

### An analog of the Sinc approximation for periodic functions

In this paper, we propose an interpolation formula for periodic function...
12/28/2021

### A New First Order Taylor-like Theorem With An Optimized Reduced Remainder

This paper is devoted to a new first order Taylor-like formula where the...
05/19/2021

### Numerical differentiation on scattered data through multivariate polynomial interpolation

We discuss a pointwise numerical differentiation formula on multivariate...
04/01/2021

### The Gamma Function via Interpolation

The Lanczos formula for the Gamma function is used in many software libr...
02/23/2021

### Transfer function interpolation remainder formula of rational Krylov subspace methods

Rational Krylov subspace projection methods are one of successful method...
08/26/2021

### A New Interpolation Approach and Corresponding Instance-Based Learning

Starting from finding approximate value of a function, introduces the me...

## 1 Introduction and statement of the Problem

There are algebraic and variational approaches in the spline theory. In the algebraic approach splines are considered as some smooth piecewise polynomial functions. In the variational approach splines are elements of Hilbert or Banach spaces minimizing certain functionals. The first spline functions were constructed from pieces of cubic polynomials. After that, this construction was modified, the degree of polynomials increased. The theory of splines based on variational methods studied and developed, for example, by J.H.Ahlberg et al. [1], C. de Boor [3], A.R.Hayotov, G.V.Milovanivić and Kh.M.Shadimetov [5], I.J.Schoenberg [6], L.L.Schumaker [7], S.L.Sobolev [8], V.A.Vasilenko [12] and others.

The present work is also devoted to the variational method of construction of optimal interpolation formulas.

Assume, we are given the table of the values , of a function at the points . It is required to approximate the function by another more simple function , i.e.

 φ(x)≅Pφ(x)=N∑β=0Cβ(x)⋅φ(xβ) (1.1)

which satisfies the following interpolation conditions

 φ(xβ)=Pφ(xβ), β=0,1,...,N.

Here and () are the coefficients and the nodes of the interpolation formula (1.1), respectively. By we denote the class of all functions defined on [0,1] which posses an absolutely continuous first derivative on [0,1] and whose second derivative is in . The class under the pseudo-inner product

 ⟨φ,ψ⟩=1∫0(φ′′(x)+ω2φ(x))(ψ′′ℓ(x)+ω2ψℓ(x))dx.

is a Hilbert space if we identify functions that differ by and , where . Here we consider the norm

 ∥φ|K2(P2)∥=⎧⎪⎨⎪⎩1∫0(φ′′(x)+ω2φ(x))2dx⎫⎪⎬⎪⎭1/2.

For the fixed the error

 (ℓ,φ)=∞∫−∞ℓ(x)φ(x)dx=φ(z)−Pφ(z)=φ(z)−N∑β=0Cβ(z)⋅φ(xβ) (1.2)

of the interpolation formula (1.1) is a linear functional.
Here, in fixed ,

 ℓ(x,z)=δ(x−z)−N∑β=0Cβ(z)δ(x−xβ),

and is the error functional of the interpolation formula (1.1) belongs to the space . Here is the conjugate space to the space , is Dirac’s delta-function.

By the Cauchy-Schwarz inequality the absolute value of the error (1.2) is estimated as follows

 |(ℓ,φ)|≤∥φ|K2(P2)∥⋅∥ℓ|K∗2(P2)∥,

where

 ∥∥ℓ|K∗2(P2)∥∥=supφ, ∥φ∥≠0|(ℓ,φ)|∥φ|K2(P2)∥.

Therefore, in order to estimate the error of the interpolation formula (1.1) on functions of the space it is required to find the norm of the error functional in the conjugate space . That is we get the following problem.

Problem 1.1. Find the norm of the error functional of the interpolation formula (1.1) in the space .

It is clear that the norm of the error functional depends on the coefficients and the nodes . The problem of minimization of the quantity by coefficients is the linear problem and by nodes is, in general, nonlinear and complicated problem. We consider the problem of minimization of the quantity by coefficients when nodes are fixed.

The coefficients (if there exist) satisfying the equality

 ∥∥˚ℓ|K∗2(P2)∥∥=infCβ(z)∥∥ℓ|K∗2(P2)∥∥ (1.3)

are called the optimal coefficients and corresponding interpolation formula is called the optimal interpolation formula in the space . Therefore, for construction of the interpolation formula we should solve the next problem.

Problem 1.2. Find the coefficients which satisfy equality (1.3) when the nodes are fixed.

It should be noted that Problems 1.1 and 1.2 were solved in [2] in the Hilbert space . There the optimal interpolation formulas, which are exact for any polynomial of degree and for the function , were obtained.

The rest of the paper is organized as follows. In Section 2, using the extremal function, the norm of the error functional is found. Existence and uniqueness of the optimal interpolation formula of the form (1.1) is discussed in Section 3. Section 4 is devoted to construction of the optimal interpolation formula. Finally, in Section 5 some numerical results are presented.

## 2 The extremal function and the norm of the error functional ℓ

Here we find explicit form of the norm of the error functional .

For finding the explicit form of the norm of the error functional in the space we use its extremal function which was introduced by Sobolev [8, 9]. The function from space is called the extremal function for the error functional if the following equality is fulfilled

 (ℓ,ψℓ)=∥∥ℓ∣∣K∗2(P2)∥∥⋅∥ψℓ|K2(P2)∥.

According to the Riesz theorem any linear continuous functional in a Hilbert space is represented in the form of a inner product. So, in our case, for any function from space we have

 (ℓ,φ)=⟨ψℓ,φ⟩. (2.1)

Here is the function from is defined uniquely by functional and is the extremal function.

It is easy to see from (2.1) that the error functional , defined on the space , satisfies the following equalities

 (ℓ,sin(ωx))=0, (2.2) (ℓ,cos(ωx))=0. (2.3)

The equalities (2.2) and (2.3) mean that our interpolation formula is exact for the functions and .

The equation (2.1) was solved in [5] and for the extremal function was obtained the following expression

 ψℓ(x)=(ℓ∗G2)(x)+d1sin(ωx)+d2cos(ωx), (2.4)

where

 G2(x)=sgnx4ω3[sin(ωx)−ωxcos(ωx)], (2.5)

is the operation of convolution which for the functions and is defined as follows

 (f∗g)(x)=∞∫−∞f(x−y)g(y)dy=∞∫−∞f(y)g(x−y)dy. (2.6)

Now we obtain the norm of the error functional . Since the space is the Hilbert space then by the Riesz theorem we have

 (ℓ,ψℓ)=∥ℓ∥⋅∥ψℓ∥=∥ℓ∥2. (2.7)

Hence, using (2.4) and (2.5), taking into account (2.6) and (2.7), we get

 ∥ℓ∥2 = (ℓ,ψℓ)=∞∫−∞ℓ(x,z)ψℓ(x)dx = ∞∫−∞⎛⎝δ(x−z)−N∑β=0Cβ(z)δ(x−xβ)⎞⎠ ×⎛⎝G2(x−z)−N∑β=0Cβ(z)G2(x−xβ)⎞⎠dx.

Hence, keeping in mind that , defined by (2.5), is the even function, we have

 ∥ℓ∥2=(−1)m⎛⎝N∑β=0N∑γ=0Cβ(z)Cγ(z)G2(xβ−xγ)−2N∑β=0Cβ(z)G2(z−xβ)⎞⎠. (2.8)

Thus, Problem 1.1 is solved.

Further, we solve Problem 1.2.

## 3 Existence and uniqueness of the optimal interpolation formula

Assume that the nodes of the interpolation formula (1.1) are fixed. The error functional (1.2) satisfies the conditions (2.2) and (2.3). The norm of the error functional is a multivariable function with respect to the coefficients . For finding the point of the conditional minimum of the expression (2.8) under the conditions (2.2) and (2.3) we apply the Lagrange method.

Consider the function

 Ψ(C0(z),C1(z),...,CN(z),d1,d2) =∥ℓ∥2−2(d1(z)(ℓ,sin(ωx))+d2(z)(ℓ,cos(ωx))).

Equating to 0 the partial derivatives of the function by , and , we get the following system of linear equations of unknowns

 N∑γ=0Cγ(z)G2(xβ−xγ)+d1(z)sin(ωxβ)+d2(z)cos(ωxβ)=G2(z−xβ), β=0,1,...,N, (3.1) N∑γ=0Cγ(z)sin(ωxγ)=sin(ωz), (3.2) N∑γ=0Cγ(z)cos(ωxγ)=cos(ωz), (3.3)

where is defined by equality (2.5).

The system (3.1)-(3.3) has a unique solution and this solution gives the minimum to under the conditions (3.2) and (3.3).

The uniqueness of the solution of the system (3.1)–(3.3) is proved as the uniqueness of the solution of the system (24)–(26) of the work [11].

Therefore, in fixed values of the nodes the square of the norm of the error functional , being quadratic function of the coefficients , has a unique minimum in some concrete value .

###### Remark 3.1.

It should be noted that by integrating both sides of the system (3.1)-(3.3) by from 0 to 1 we get the system (3.1)-(3.1) of the work [5]. This means that by integrating the optimal interpolation formula (1.1) in the space we get the optimal quadrature formula of the form (1.1) in the same space (see [5]).

###### Remark 3.2.

It is clear from the system (3.1)-(3.3) that for the optimal coefficients the following are true

 ˚Cβ(hγ) = {1,γ=β,0,γ≠β,γ=0,1,...,N,  β=0,1,...,N.

Below for convenience the optimal coefficients we remain as .

## 4 The algorithm for computation of coefficients of the optimal interpolation formula

In the present section we give the algorithm for solution of the system (3.1)-(3.3). Below mainly is used the concept of discrete argument functions and operations on them. The theory of discrete argument functions is given, for instance, in [9, 11]. For completeness we give some definitions about functions of discrete argument.

Assume that the nodes are equal spaced, i.e. , .

###### Definition 4.1.

The function is a discrete argument function (or discrete function) if it is given on some set of integer values of .

###### Definition 4.2.

The inner product of two discrete functions and is given by

 [φ(hβ),ψ(hβ)]=∞∑β=−∞φ(hβ)⋅ψ(hβ),

if the series on the right hand side converges absolutely.

###### Definition 4.3.

The convolution of two functions and is the inner product

 φ(hβ)∗ψ(hβ)=[φ(hγ),ψ(hβ−hγ)]=∞∑γ=−∞φ(hγ)⋅ψ(hβ−hγ).

Now we turn on to our problem.

Suppose that when and . Thus we have the following problem.

Problem 4.1 Find the discrete functions , , and which satisfy the system (3.2)-(3.3).

Further we investigate Problem 4.1 which is equivalent to Problem 1.2. Instead of we introduce the following functions

 v2(hβ) = G2(hβ)∗Cβ(z), u2(hβ) = v2(hβ)+d1(z)sin(ωhβ)+d2cos(ωhβ). (4.1)

Now we should express the coefficients by the function . For this we use the operator which satisfies the equality

 D2(hβ)∗G2(hβ)=δ(hβ), (4.2)

where is equal to 0 when and is equal to 1 when , i.e. is the discrete delta-function.

In [4] the operator which satisfies equation (4.2) is constructed and its some properties are studied.

The following theorems are proved in [4].

###### Theorem 4.4.

The discrete analogue of the differential operator satisfying the equation (4.2) has the form

 D2(hβ)=p⎧⎪ ⎪⎨⎪ ⎪⎩Aλ|β|−1,|β|≥2,1+A,|β|=1,C+Aλ,β=0. (4.3)

where

 p=2ω3sin(ωh)−ωhcos(ωh),C=2hωcos(2hω)−sin(2hω)sin(hω)−hωcos(hω),A=(2hω)2sin4(hω)⋅λ2(λ2−1)(sin(hω)−hωcos(hω))2,λ=2hω−sin(2hω)−2sin(hω)⋅√(hω)2−sin2(hω)2(hωcos(hω)−sin(hω)), |λ|<1.
###### Theorem 4.5.

The discrete analogue of the differential operator satisfies the following equalities

1)

2)

3)

4)

Then taking into account (4.2), (4.3), using Theorems 4.4 and 4.5, for optimal coefficients we have

 Cβ(z)=D2(hβ)∗u2(hβ). (4.4)

Thus if we find the function then the optimal coefficients will be found from equality (4.4).

In order to calculate the convolution (4.4) it is required to find the representation of the function at all integer values of . From equality (4.1) we get that when . Now we find the representation of the function when and .

Since when then

 Cβ(z)=D2(hβ)∗u2(hβ)=0,hβ∉[0,1].

Now we calculate the convolution when .

Suppose and then taking into account equalities (2.5), (3.2) and (3.3), we have

 u2(hβ)=⎧⎪ ⎪ ⎪ ⎪⎨⎪ ⎪ ⎪ ⎪⎩d−1(z)sin(ωhβ)+d−2(z)cos(ωhβ)+hβ4ω2cos(ω(hβ−z)),β≤0,G2(z−hβ),0≤β≤N,d+1(z)sin(ωhβ)+d+2(z)cos(ωhβ)−hβ4ω2cos(ω(hβ−z)),β≥N, (4.5)

where and are unknowns.

From (4.5) when and we get

 d−2(z)=G2(z), (4.6) d+2(z)=1cos(ω)[G2(z−1)+cos(ω(1−z))4ω2−d+1(z)sin(ω)]. (4.7)

Thus, putting (4.6) and (4.7) to (4.5) we have the following explicit form of the function :

 u2(hβ)=⎧⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪⎨⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪⎩d−1(z)sin(ωhβ)+G2(z)cos(ωhβ)+hβ4ω2cos(ω(hβ−z)),β≤0,G2(z−hβ),0≤β≤N,d+1(z)sin(ω(hβ−1))cos(ω)+cos(ωhβ)cos(ω)[G2(z−1)+cos(ω(1−z))4ω2]−hβ4ω2cos(ω(hβ−z)),β≥N. (4.8)

In the last expression of the function we have only two unknowns and .

Hence using (4.3) and (4.8) we get the following problem.

Problem 4.2. Find the solution of the equation

 D2(hβ)∗u2(hβ)=0,hβ∉[0,1],

having the form (4.8). Here and are unknowns.

Unknowns and we find from the equation

 D2(hβ)∗u2(hβ)=0 (4.9)

when and . From the last equation

From the system (4.9) in the case and after some simplifications we have the following system of equations

 A11d−1(z)+A12d+1(z)=S1 (4.10)

where

 A11=−sin(2ωh)−Csin(ωh)−Asin(ωh)λ(λ2−2λcos(ωh)+1),A12=AλN+1sin(ωh)cos(ω)(λ2−2λcos(ωh)+1),S1=14ω2(Chcos(ω(h+z)+2hcos(ω(2h+z))−AλNcos(ω(1−z))cos(ω))K1  +Ahλ2K2+AhλNK3)−AK4−AλNG2(z−1)cos(ω)K1  −G2(z)(Ccos(ωh)+1+cos(2ωh)+Aλ2K5),K1=λ(cos(ω(1+h))−λcos(ω))λ2−2λcos(ωh)+1,K2=λ(λ2cos(ω(h−z))−2λcos(ωz)+cos(ω(h+z)))(λ2−2λcos(ωh)+1)2,
 K3=∞∑γ=1λγ(N+γ)cos(ω(h(N+γ)−z)),K4=N∑γ=0λγG2(z−hγ),K5=λ(cos(ωh)−λ)λ2−2λcos(ωh)+1.

Now, from (4.9) in the case doing some calculations we get the next equation

 A21d−1(z)+A22d+1(z)=S2, (4.11)

where

 A21=−AλN+1sin(ωh)λ2−2λcos(ωh)+1,A22=1cos(ω)[Csin(ωh)+sin(2ωh)+Asin(ωh)λ(λ2−2λcos(ωh)+1)],S2=14ω2(AλNhK2+cos(ω(1−z))+Ch(N+1)cos(ω(h(N+1)−z))  +h(N+2)cos(ω(h(N+2)−z))+AhK3λ2−cos(w(1−z))K6cos(ω))  −AλNK7−G2(z)AλNK5−G2(z−1)cos(ω)K6,K6=cos(ω)+Ccos(ωh(N+1))+cos(ωh(N+2))+AK1λ2,K7=N∑γ=0λ−γG2(z−hγ),K8=λ(sin(ω(h+1))−λsin(ω))λ2−2λcos(ωh)+1.

Then solving the system (4.10), (4.11) of equations we get and . Finally, from (4.4) for we get the explicit formulas for optimal coefficients as claimed in the following theorem.

###### Theorem 4.6.

Coefficients of the optimal interpolation formula (1.1) with equal spaced nodes in the space have the following form

 C0(z)=p[−d−1(z)sin(ωh)+d−2(z)cos(ωh)−14ω2hcos(ω(h+z))+CG2(z)+G2(z−h)+Aλ[N∑γ=0λγG2(z−hγ)+M1+λNN1]],Cβ(z)=p[G2(z−h(β−1))+CG2(z−hβ)+G2(z−h(β+1))
 +Aλ[N∑γ=0λ|β−γ|G2(z−hγ)+λβM1+λN−βN1]],β=1,2,..,N−1.
 CN(z)=p[d+1(z)sin(ωh(N+1))+d+2(z)cos(ωh(N+1))−14ω2h(N+1)cos(ω(h(N+1)−z))+CG2(z−1)+G2(z−h(N−1))+Aλ[N∑γ=0λN−γG2(z−hγ)+λNM1+N1]],

here

 M1=−d−1(z)λsin(ωh)λ2−2λcos(ωh)+1+d−2(z)K5−h4ω2K2,N1=d+1(z)K8+d+2(z)K1−h4ω2K3,d−1(z)=S1⋅A22−S2⋅A12A11⋅A22−A21⋅A12,d+1(z)=S2⋅A11−S1⋅A21A11⋅A22−A21⋅A12,d−2(z)=G2(z),d+2(z)=1cos(ω)[G2(z−1)+cos(ω(1−z))4ω2−d+1(z)sin(ω)].

## 5 Numerical results

In this section we give some numerical results.

First, when using Theorem 4.6, we get the graphs of the coefficients of the optimal interpolation formulas

 φ(z)≅˚Pφ(z)=5∑β=0˚Cβ(z)φ(hβ), z∈[0,1].

They are presented in Fig 1. These graphical results confirm Remark 3.2 for the case , i.e. for the optimal coefficients the following hold

 ˚Cβ(hγ)=δβγ,  β,γ=0,1,...,5,

where is the Kronecker symbol.

Now, in numerical examples, we interpolate the functions

 φ1(z)=z2, φ2(z)=ez and φ3(z)=sinz

by optimal interpolation formulas of the form (1.1) in the cases , using Theorem 4.6. For the functions , the graphs of absolute errors , , are given in Fig 2, Fig 3, Fig 4. In these Figures one can see that by increasing value of absolute errors between optimal interpolation formulas and given functions are decreasing.

Fig. 1. Graphs of coefficients of the optimal interpolation formulas (1.1) in the case .

Fig. 2. Graphs of absolute errors for and : .

Fig. 3. Graphs of absolute errors for and : .

Fig. 4. Graphs of absolute errors for and : .

The Figure 4 shows exactness our optimal interpolation formula for the function .

References

## Список литературы

• [1] Ahlberg.J.H, Nilson.E.N, Walsh.J.L, The theory of splines and their applications, Mathematics in Science and Engineering, New York: Academic Press, 1967.
• [2] Babaev.S.S, Hayotov.A.R, Optimal interpolation formulas in space. arXiv:1802.00562.
• [3]

de Boor.C, Best approximation properties of spline functions of odd degree, J. Math. Mech. 12, (1963), pp.747-749.

• [4] Hayotov.A.R, The discrete analogue of a differential operator and its applications. Lithuanian Mathematical Journal. 2014.-vol54. No2, pp.290-307.
• [5] Hayotov.A.R , Milovanovic.G.V , Shadimetov.Kh.M, Optimal quadrature formulas and interpolation splines minimizing the semi-norm in space // Milovanovic.G.V and Rassias.M.Th (eds.), Analytic Number Theory, Approximation Theory, and Special Functions, Springer, 2014, -pp.573-611.
• [6] Schoenberg.I.J, On trigonometric spline interpolation, J. Math. Mech. 13, (1964), pp.795-825.
• [7] Schumaker.L, Spline functions: basic theory, Cambridge university press, 2007, 600 p.
• [8] Sobolev.S.L, On Interpolation of Functions of Variables, in: Selected Works of Sobolev.S.L, Springer, 2006, pp. 451-456.
• [9] Sobolev.S.L, Introduction to the Theory of Cubature Formulas, Nauka, Moscow, 1974, 808 p.
• [10] Sobolev.S.L, The coefficients of optimal quadrature formulas, in: Selected Works of Sobolev.S.L. Springer, 2006, pp.561-566.
• [11] Sobolev.S.L, Vaskevich.V.L, The Theory of Cubature Formulas. Kluwer Academic Publishers Group, Dordrecht (1997).
• [12] Vasilenko.V.A, Spline-fucntions: Theory, Algorithms, Programs, Nauka, Novosibirsk, 1983, 216 p. (in Russian)