A Generalization of the Remainder in Multivariate … Case of minimal interpolation schemes A...

6
A Generalization of the Remainder in Multivariate Polynomial Interpolation Dana Simian Department of Informatics Faculty of Sciences The ”Lucian Blaga” University of Sibiu Sibiu 5-7 dr. I.Ratiu str. ROMANIA Abstract: The aim of this article is to introduce a generalization of the interpolation remainder in multivariate interpolation and to study it, in least interpolation schemes and in minimal interpolation ones. This generalization, we named λ-remainder, allows us a deeper analysis of the error in the interpolation process. Connected to the λ - remainder, we introduced the notion of λ - error order of interpolation. In the end we provide particular applications of the results we obtained. Interesting additionally theorems are also proved. Keywords: multivariate polynomial interpolation, λ - remainder, λ - error order of interpolation 1 Introduction Interpolation by polynomials in several variables is an active area of research. A survey of the main re- sults on multivariate polynomial interpolation in the last thirty years can be found in [?]. An important problem in the field of interpolation is the construc- tion of a polynomial interpolation space for arbitrary interpolation nodes and the estimation of the interpo- lation remainder. In 1990, de Boor and Ron, in [?], constructed, for a given set of arbitrary nodes, an in- teresting polynomial interpolation space, named ”least interpolation space”. This construction admits a ge- neralization for an arbitrary set of conditions. We re- call now, the general formulation of the interpolation problem, we present the construction of the ”least in- terpolation space” for an arbitrary set of functionals (introduced in [?]) and then we prove a theorem we need in the next section. Let Λ = {λ 1 ,...,λ n } be a set of linear functionals, linear independent, Π d the space of polynomials in d variables and F a space of functions which includes polynomials. The polynomial interpolation problem with respect to the set of conditions Λ is to construct a polynomial subspace P such that for an arbitrary function f ∈F there exists a unique polynomial p ∈P satisfying the conditions λ(p)= λ(f ), λ Λ. In this case, we say that the pair (Λ, P ) is correct, or, equiva- lently, P is an interpolation space with respect to Λ. We can associate to any functional λ, its formal power series or its generating function, λ ν = X αN d λ(m α ) α! x α , (1) with, m α (x)= x α , x R d , α N d . Obviously, for any p Π d , one has λ(p)=(p(D)λ ν )(0) (2) We define, for any power series f (or analytical function), its least term, f , which is its nonzero ho- mogeneous term of minimal degree. ”Least interpolation space”, with respect to the set of functionals Λ, is the polynomial subspace generated from the least terms of the generating functions of the functionals in Λ: H Λ = span{λ ν ↓| λ Λ} (3) It is proved in [?], that H Λ is a minimal interpola- tion space, or equivalent a degree reducing interpola-

Transcript of A Generalization of the Remainder in Multivariate … Case of minimal interpolation schemes A...

Page 1: A Generalization of the Remainder in Multivariate … Case of minimal interpolation schemes A finite set, Λ, of linear functionals is said to admit an ideal interpolation scheme

A Generalization of the Remainder in Multivariate Polynomial

Interpolation

Dana Simian

Department of InformaticsFaculty of Sciences

The ”Lucian Blaga” University of SibiuSibiu 5-7 dr. I.Ratiu str.

ROMANIA

Abstract: The aim of this article is to introduce a generalization of the interpolation remainder in multivariateinterpolation and to study it, in least interpolation schemes and in minimal interpolation ones. This generalization,we named λ-remainder, allows us a deeper analysis of the error in the interpolation process. Connected to theλ - remainder, we introduced the notion of λ - error order of interpolation. In the end we provide particularapplications of the results we obtained. Interesting additionally theorems are also proved.

Keywords: multivariate polynomial interpolation, λ - remainder, λ - error order of interpolation

1 Introduction

Interpolation by polynomials in several variables isan active area of research. A survey of the main re-sults on multivariate polynomial interpolation in thelast thirty years can be found in [?]. An importantproblem in the field of interpolation is the construc-tion of a polynomial interpolation space for arbitraryinterpolation nodes and the estimation of the interpo-lation remainder. In 1990, de Boor and Ron, in [?],constructed, for a given set of arbitrary nodes, an in-teresting polynomial interpolation space, named ”leastinterpolation space”. This construction admits a ge-neralization for an arbitrary set of conditions. We re-call now, the general formulation of the interpolationproblem, we present the construction of the ”least in-terpolation space” for an arbitrary set of functionals(introduced in [?]) and then we prove a theorem weneed in the next section.

Let Λ = {λ1, . . . , λn} be a set of linear functionals,linear independent, Πd the space of polynomials in dvariables and F a space of functions which includespolynomials. The polynomial interpolation problemwith respect to the set of conditions Λ is to constructa polynomial subspace P such that for an arbitrary

function f ∈ F there exists a unique polynomial p ∈ Psatisfying the conditions λ(p) = λ(f), ∀λ ∈ Λ. In thiscase, we say that the pair (Λ,P) is correct, or, equiva-lently, P is an interpolation space with respect to Λ.

We can associate to any functional λ, its formalpower series or its generating function,

λν =∑

α∈Nd

λ(mα)α!

xα, (1)

with, mα(x) = xα, x ∈ Rd, α ∈ Nd.Obviously, for any p ∈ Πd, one has

λ(p) = (p(D)λν)(0) (2)

We define, for any power series f (or analyticalfunction), its least term, f ↓, which is its nonzero ho-mogeneous term of minimal degree.

”Least interpolation space”, with respect to the setof functionals Λ, is the polynomial subspace generatedfrom the least terms of the generating functions of thefunctionals in Λ:

HΛ↓= span{λν↓ | λ ∈ Λ} (3)

It is proved in [?], that HΛ ↓ is a minimal interpola-tion space, or equivalent a degree reducing interpola-

Page 2: A Generalization of the Remainder in Multivariate … Case of minimal interpolation schemes A finite set, Λ, of linear functionals is said to admit an ideal interpolation scheme

tion space.A particular choice of functionals leads us to many

interesting interpolation spaces.In order to get an expression of the λ-remainder

in least interpolation we need some additional results.First we introduce the pair between an analytical func-tion and a polynomial:

< f, p >= (p(D)f)(0) (4)

which is a veritable inner product on polynomial spa-ces.

Proposition 1 (de Boor, [?]) There always is a basisgi, i ∈ {1, . . . , #HΛ} of the space

HΛ = span{λν | λ ∈ Λ},which is orthogonal to HΛ↓ in the sense of pair (??),that is < gi, gk↓>6= 0 ⇔ k = i.

The basis gi can be constructed using aGramm - Schmidt type algorithm, (see [?]), or a Gausselimination by segments algorithm. The description ofthe idea of the Gauss elimination by segments algo-rithm can be found in [?]. Using this basis, we canextend the product from (??) in the following sense:let cj,i be the coordinates of gj in the basis λν of HΛ,that is

gj(x) =n∑

i=1

cj,i λνi (x), (5)

then

< gj , f >=n∑

i=1

cj,iλi(f), ∀f ∈ A0 (6)

A generalization of the results from [?] is given inthe following theorem

Theorem 1 The unique element LΛ(f) ∈ HΛ↓ whichinterpolates f ∈ A0 with respect to the conditions Λ is

LΛ(f) =n∑

j=1

gj↓ < gj , f >

< gj , gj↓>, n = #Λ (7)

Proof: Taking into account that

< gi, LΛ(f) >=< gi, f >, ∀ f ∈ A0,

we get λk(LΛ(f)) = λk(f), k ∈ {1, . . . , n}. The unicityof LΛ follows from the interpolation property of thespace HΛ↓. ♠Theorem 2 ([?], [?]) The operator

L∗Λ(f) =n∑

j=1

gj< f, gj↓>< gj , gj↓> ; f ∈ A0. (8)

has the duality property:

< L∗Λ(g), f >=< g, LΛ(f) >, g, f ∈ A0, (9)

Theorem 3 The operator LΛ satisfies the inequality:degLΛ(f) ≤ deg(f), ∀ f ∈ Πd and the inequality isstrict if and only if f↑ ⊥HΛ↓, with f↑ being the leadingterm of the polynomial f .

Proof: There comes out from (??), thatdeg(LΛ) ≤ max(deg(gj ↓)). If deg(gj ↓) > k then< gj , f >= 0. If deg(gj ↓) = k, then, the followingimplications hold:f↑⊥ HΛ↓⇔< p, f↑>= 0,∀ p ∈ HΛ↓⇒< gj↓, f↑>= 0 ⇒< gj , f >= 0, ∀ j ∈ {1, . . . , n}

Consequently deg(LΛ(f)) < deg(f), ∀ f ∈ Πd,with f↑⊥ HΛ↓.

Let’s suppose now that deg LΛ(f) < deg (f).Hence (f − LΛ(f))↑= f↑.

We use the fact that if p(D) annihilates HΛ, thenp↑ (D) annihilates HΛ↓ (see [?]) and the following im-plications:

λ(p) = 0, ∀ λ ∈ Λ ⇔ p ⊥ HΛ ⇒ p↑⊥ HΛ↓ .

Therefore:λ(f − LΛ(f)) = 0, ∀ λ ∈ Λ ⇔ (f − LΛ(f)) ⊥ HΛ

⇒ (f − LΛ(f))↑⊥ HΛ↓⇔ f↑⊥ HΛ↓ .♠

2 The λ -remainder and λ-error order of in-terpolation

Let’s consider the general polynomial interpolationproblem with conditions Λ and let LΛ be the correspon-ding interpolation operator and RΛ be the remainderoperator. The interpolation formula is:

f = LΛ(f) + RΛ(f) (10)

Definition 1 We name λ-remainder, the value

RΛ,λ(f) = λ[(1− LΛ)(f)]; f ∈ A0; λ ∈ Π′ (11)

Consequently, for any x ∈ Rd, the classical remain-der (RΛ(f))(x) is in fact the δx-remainder. For anyfunctional λ ∈ Λ we obtain RΛ,λ(f) = 0, ∀f ∈ A0.

Definition 2 (C. de Boor, [?]) Let be L : A0 → Πd apolynomial interpolation operator. The error order ofinterpolation is the greatest integer k such thatf(x)− (L(f))(x) = 0, ∀ f ∈ Πd

<k.

We generalize this definition.

Definition 3 We name λ-error order of interpolationthe greatest integer k such that

RΛ,λ = 0, ∀ f ∈ Πd<k, λ ∈ (Πd)′,

with RΛ,λ the λ - remainder, defined in (??).

If in definition ?? we take λ = δx, ∀x ∈ Rd, weobtain definition ??.

Page 3: A Generalization of the Remainder in Multivariate … Case of minimal interpolation schemes A finite set, Λ, of linear functionals is said to admit an ideal interpolation scheme

2.1 Case of least interpolation scheme

Proposition 2 The λ-remainder in ”least interpola-tion” scheme can be expressed such as:

RΛ,λ(f) =< ενΛ,λ, f >, (12)

withενΛ,λ = (1− L∗Λ)(λν). (13)

Proof:RΛ,λ(f) = λ[(1− LΛ)(f)] =< λν , (1− LΛ)f >==< λν , f > − < L∗Λ(λν), f >=< εν

Λ,λ, f >. ♠Corollary 1 ([?]) The expression of the classical in-terpolation remainder is:

(R(f))(x) =< ex − L∗Λ(ex), f >, (14)

with ex(t) = ext; x, t ∈ Rd.

Theorem 4 ενΛ,λ ⊥ HΛ↓ and every homogeneous com-

ponent of ενΛ,λ satisfies the same orthogonality proper-

ty.

Proof: Let be p ∈ HΛ↓. Then,

< ενΛ,λ, p >= RΛ,λ(p) = 0.

Consequently, ενΛ,λ ⊥ HΛ↓.

The polynomial subspace HΛ ↓ is generated byhomogeneous polynomials. Therefore (εν

Λ,λ)[k] ⊥ HΛ↓.For any analytical function, g ∈ A0, we had definedthe k - order homogeneous component, such as:

g[k] =∑

|α|=k

Dαg(0)(·)α/α!

Proposition 3 ενΛ,λ satisfies the equality

< ενΛ,λ, f >= λ(f),∀ f ∈ kerLΛ (15)

Proof: < ενΛ,λ, f >=< λ, f > − < λ, LΛ(f) >.

f ∈ kerLΛ ⇒ LΛ(f) = 0 ⇒< λ, LΛ(f) >= 0. ♠The following theorem gets the λ-error order of in-

terpolation in the particular case of least interpolationscheme.

Theorem 5 If HΛ↓6= Πm, then the λ-error order ofinterpolation is given by the deg(εν

Λ,λ↓), ενΛ,λ being de-

fined in (??).

Proof: Let k = deg(ενΛ,λ↓) and f ∈ Πd

<k. Obviouslydeg(f ↑) < k and < εν

Λ,λ, f >= 0, hence the λ-errororder of interpolation is greater than or equal to k.

First, let’s consider that there is not any m ∈ Nsuch that HΛ↓= Πd

m. Let’s suppose by contradictionthat RΛ,λ(f) = 0,∀ f ∈ Πd

k. Then < ενΛ,λ, f >= 0 and

taking into account theorem ?? we also get

< ενΛ,λ↓, f >= 0,∀ f ∈ Πd

k.

This is a contradiction, because ενΛ,λ↓ is a homogeneous

polynomial of degree k.Similarly, the supposition that RΛ,λ(f) = 0,

∀ f ∈ Πd<q, with q > k leads us to the contradiction

< ενΛ,λ↓, f [k] >= 0, ∀ f ∈ Πd

<q.If HΛ↓= Πd

m, the λ- error order of interpolation ism + 1, be cause εν

Λ,λ↓⊥ HΛ↓. ♠Theorem 6 Let be

P = {p ∈ Πd| λ(p) 6= 0, λ 6∈ Λ; p ∈ ker(LΛ)}Then the following equality holds:

deg ενΛ,λ↓= min{deg p|p ∈ P}.

Proof: Let denote by

k = deg ενΛ,λ↓ and k′ = min{deg p|p ∈ P}

Taking into account theorem ?? we get:p ∈ ker(LΛ) ⇒< εν

Λ,λ, p >= λ(p) 6= 0⇒ deg εν

Λ,λ↓≤ deg p↑= deg p ⇒ k ≤ k′.On the other hand let q = εν

Λ,λ↓ −LΛ(ενΛ,λ↓). Using

theorems ?? and ??, we obtain

deg q = deg ενΛ,λ↓= k.

Much more λ(q) = RΛ,λ(ενΛ,λ↓) =< εν

Λ,λ, ενΛ,λ↓> > 0

and LΛ(q) = 0. But, from q ∈ Πd; λ(q) 6= 0 andq ∈ ker(LΛ) we obtain deg q ≥ k′, that is k ≥ k′. ♠Corollary 2 If q ∈ Πd

≥k, then the expression of the λ-remainder is

RΛ,λ(q) =∑

α∈Nd,|α|≥k

DαενΛ,λ(0) ·Dαq(0)

α!

with k = deg ενΛ,λ↓.

Corollary 3 ενΛ,λ vanishes to order k = deg εν

Λ,λ↓ at0.

Proof: It easily results from the equality(q(D)εν

Λ,λ

)(0) = 0,∀ q ∈ Π<k

The next theorem allows us to study the classicalremainder and hence to obtain the error order of inter-polation in least interpolation.

Theorem 7 The operator LΛ given in (??) reproducesthe monomials xα, x ∈ Rd, α ∈ Nd, if and only ifn∑

i=1

ck,iλi(xα) = Dαgk(0), with coefficients ck,i given

in (??).

Proof:n∑

i=1

< gk, ϕiλi(xα) >=n∑

i=1

λi(xα)cki.

On the other hand,n∑

i=1

< gk, ϕiλi(xα) >=< gk, xα >= D(α)gk(0)

Page 4: A Generalization of the Remainder in Multivariate … Case of minimal interpolation schemes A finite set, Λ, of linear functionals is said to admit an ideal interpolation scheme

2.2 Case of minimal interpolationschemes

A finite set, Λ, of linear functionals is said to admitan ideal interpolation scheme if ker(Λ) is a polynomialideal.

Definition 4 ([?],[?]) A polynomial subspaceV ⊂ Πd

n is a minimal interpolation space of order nwith respect to Λ if it is a degree reducing interpola-tion space and the set of conditions Λ admits an idealinterpolation scheme.

As pointed out by many authors (de Boor in [?], Sauerin [?]), ideal interpolation schemes can even be charac-terized as Hermite interpolation schemes with an ad-ditional closedness condition. Minimal interpolationspaces are deeply connected with the notion of New-ton basis.

Definition 5 We say that the polynomial spaceV ⊂ Πd

n admits a Newton basis of order n with res-pect to the set of functionals Λ, if the functionals inΛ may be reindexed in the blocks Λ(k) = {λα : λα ∈Λ; α ∈ Ik \ Ik−1}, using a grading set of multiindicesI = (I0, . . . , In), Ik \ Ik−1 ⊂ {α : |α| = k};k = 0, . . . , n, such that1. There is a basis pα ∈ Πd

|α|, α ∈ In of P(Λ) with

λβ(pα) = δα,β ; β ∈ In; |β| ≤ |α| (16)

2.There are the complementary polynomials

p⊥α ∈ Πd|α| ∩ker(Λ), α ∈ I ′n = {α ∈ Nd : |α| ≤ n} \ In,

satisfying

Πdn = span{pα : α ∈ In} ⊕ span{p⊥α : α ∈ I ′n}

The number of functionals in the block Λ(k) equalsthe dimension of homogeneous subspace of V , of degreek.

It is known ( see [?]) that a polynomial subspace isa minimal interpolation space of order n with respectto Λ if and only if it admits a Newton basis of order nwith respect to Λ.

In order to compute the λ-remainder in minimalinterpolation schemes, we need to construct the corres-ponding Newton basis of the interpolation space. Wecan do this, in an inductive way, using a generalizationof the method presented in [?], if we know a basis of theminimal interpolation space. We will denote the func-tionals in block Λ(k), by λ

[k]i , 1 ≤ i ≤ nk, Vk = V ∩Πd

k

and Qk = Q ∩Πdk.

V is a degree reducing interpolation space, henceΠd

n = V ⊕ Qn and Πdk = Vk ⊕ Qk. Therefore, there is

a grading basis of V , g1, . . . , gN , with N = dim V .

Consequently, we may define a system of set of mul-tiindices I = (I0, . . . , In) so that I0 ⊂ I1 ⊂ . . . ⊂ In

and #Jk = nk = dim V 0k , with V 0

k being the homoge-neous subspace of V , of k order, Jk = Ik \ Ik−1 and wemay rewrite the grading basis like {gα : α ∈ Jk; k =0, . . . , n}.

We find the blocks Λ(k), reindex the functionals inthese blocks and starting from them, we construct thespaces Vk and Qk.

We choose the right functionals λ[s]r ;

s ∈ {0, . . . , n}; r ∈ {1, . . . , ns} and construct, for everypair (j, k), k ∈ {0, . . . , n}, j ∈ {1, . . . , nk}, the poly-nomials p

[l]i ∈ Πd

l ; l ∈ {0, . . . , n}; i ∈ {1, . . . , nl} suchthat, for (r, s) ≤ (i, l) ≤ (j, k):

λ[s]r (p[l]

i ) = δl,s · δi,r, (17)

and the polynomials q[l]i , with i ∈ {j, . . . , rl}, such that

λ[s]r (q[l]

i ) = 0, for (r, s) ≤ (j, k) < (i, l) (18)

We will use the double induction, first on k and,for a certain k, induction on j: 0 ≤ k ≤ n; 1 ≤ j ≤ nk.

We initialize q[l]i = g

[l]αi ; l = 0, . . . , n; i ∈ {1, . . . , nl}

and complete the set of polynomials q[l]i for i =

nl + 1, . . . , rl = dim Πdl to a basis for Πd

n. The cor-rectness of the pair (Λ, V ) implies the existence of afunctional λ

[0]1 ∈ Λ so that λ

[0]1 (g

α[0]1

) 6= 0. Then,

p[0]1 =

1

λ[0]1 (g

α[0]1

)and q

[l]i = g

α[l]1− λ

[0]1 (g

α[0]1

);

(1, 0) < (i, l); i = 1, . . . rl

Let us suppose that for certain k and j, 0 < k < n,1 < j < nk we have already done the required con-struction. Let Λ = Λ \ {λ[l]

i : (i, l) ≤ (j, k)} the set offunctionals that have not yet put into blocks. Again,the correctness of pair (Λ, V ), implies the existence ofa functional λ

[k]j+1 ∈ Λ so that λ

[k]j+1(q

[k]j+1) 6= 0 ( if not

q[k]j+1 vanishes on all Λ).

We set

p[k]j+1 =

q[k]j+1

λ[k]j+1(q

[k]j+1)

which satisfies λ[k]r (p[k]

j+1) = δj+1,r, ∀r ≤ j + 1. Thepolynomials

p[k]i = p

[k]i − λ

[k]j+1(p

[k]i ) · p[k]

j+1; i = 1, . . . , j

satisfy λ[k]j+1(p

[k]i ) = 0 and λ

[k]r (p[k]

i ) = 0, ∀r < j, hence,

replacing p[k]i , i ∈ {1, . . . , j} with p

[k]i we obtain polyno-

mials and functionals which satisfy (??) with j → j+1.The polynomials

q[l]i = q

[l]i − λ

[k]j+1(q

[l]i ) · p[k]

j+1, (j + 1, k) < (i, l)

Page 5: A Generalization of the Remainder in Multivariate … Case of minimal interpolation schemes A finite set, Λ, of linear functionals is said to admit an ideal interpolation scheme

satisfy

λ[s]r (q[l]

i ) = 0, for (r, s) ≤ (j + 1, k) < (i, l)

Hence, replacing the polynomials q[l]i with q

[l]i for

(j + 1, k) < (i, l), we obtain polynomials which satisfy(??) for j → j + 1.

This finish induction on j. Similarly, it may bedone the induction on k.

Definition 6 Let (pα), α ∈ In be the Newton basisfor the minimal interpolation space of n order V andΛ(k) the proper blocks of functionals. The λ-divideddifference is defined recursively by:d0[λ; f ] = λ(f)dk+1[Λ(0), . . . , Λ(k), λ; f ] =

= dk[Λ(0), . . . , Λ(k−1), λ; f ]−−

α∈Jk

dk[Λ(0), . . . , Λ(k−1), λα; f ]λ(pα)

with Jk = Ik \ Ik−1.

Taking Λ = {δθ : θ ∈ Θ}, where δx is the eva-luation functional in x ∈ Rd, we obtain the divideddifference used by T. Sauer in [?].

With the notations in the definition ??, we get:

Theorem 8 The λ-remainder in interpolation from aminimal interpolation space of order n has the expres-sion:

RΛ,λ(f) = dn+1[Λ(0), . . . , Λ(n), λ; f ] (19)

The proof may be done using induction on n.

3 Application

Let Λ = δΘ,

Θ = {θi; i = 1, . . . , 4} = {(a, 0); (0, b); (−a, 0); (0,−b)}a, b ∈ R+. We generate the basis {gi; i = 1, . . . , 4} and{gi↓; i = 1, . . . , 4} of the spaces HΛ and HΛ↓, using aGramm-Schmidt type algorithm or Gauss eliminationby segments (see [?],[?]). We start with g1 = eθ1 andobtain:g1(x, y) =

1a4 + b4

[b4 cosh(ax) + a4 cosh(by)

]

g1↓ (x, y) = 1; < g1, g1↓>= 1g2(x, y) = sinh(by)− sinh(ax)g2↓ (x, y) = by − ax; < g2, g2↓>= a2 + b2

g3(x, y) =1

(a2 + b2)(a4 + b4)· E(x, y)

E(x, y) = [(a6 − a4 + a4b2 + a2b2 + 2b6) cosh(ax)−−2(a4b2 + b6)eax++(a6 + a4 + a4b2 − a2b2 + 2a2b4) cosh(by)−−2(a6 + a2b4)eby]

g3↓ (x, y) = − 2ab2

a2 + b2x− 2a2b

a2 + b2y;

< g3, g3↓>=4a2b2

a2 + b2

g4(x, y) = 2[cosh(by)− cosh(ax)]g4↓ (x, y) = −a2x2 + b2y2;< g4, g4↓>= 2(a4 + b4).

The polynomials g2↓ and g3↓ are linear indepen-dent, hence, the interpolation space is:HΛ↓= Π1 + span{−a2x2 + b2y2}

The coefficients ci,j ; i, j ∈ {1 . . . 4}, in formula (??)are given below:

c1,1 =b4

2(a4 + b4); c2,1 = −1/2;

c1,2 =a4

2(a4 + b4); c2,2 = 1/2;

c1,3 =b4

2(a4 + b4); c2,3 = 1/2;

c1,4 =a4

2(a4 + b4); c2,4 = −1/2;

c3,1 = k(a6 − 3a4b2 − a4 + a2b2 − 2b6);c3,2 = k(−3a6 + a4b2 + a4 − a2b2 − 2a2b4);c3,3 = k(a6 + a4b2 − a4 + a2b2 + 2b6);c3,4 = k(a6 + a4b2 + a4 − a2b2 + 2a2b4);

c4,1 = −1; c4,2 = 1;c4,3 = −1; c4,4 = 1;

We used the notation k =1

2(a2 + b2)(a4 + b4).

We will calculate and analyze the λ-error order ofinterpolation for various choices of functional λ.Case 1. The case of evaluation functionals

We want to study the classical error order of inter-polation, using theorem ??. In this case λ = δ(t1,t2),(t1, t2) ∈ R2. The generating function is λν(x, y) =et1x+t2y. We calculate the homogeneous componentsof εν

Λ,λ and obtain:(εν

Λ,λ)[0] = 0;

(ενΛ,λ)[1] = b(b−a)

a2+b2 (t1 − t2)x++ 1

a2+b2 [t1(ab− a2)− t2(ab + b2)]yIf a 6= b then deg(εν

Λ,λ)↓= 1, ∀(t1, t2) ∈ R2 and theλ-error order equals 1, ∀λ = δ(t1,t2), that is the classi-cal error order of interpolation is equal to 1.

If a = b, then for t2 = 0, (ενΛ,λ)[1] = 0, that is

the λ-error order of interpolation is greater than 1, forλ = δ(t1,0).

The same result can be obtained using theorem ??.LΛ reproduces the constant functions, because

4∑i=1

ck,iδθi(1) = gk(0), ∀k ∈ {1, . . . , 4}, but it does not

reproduce the polynomials of degree 1, because

4∑

i=1

c3,iδθi(m(1,0)) 6= D(1,0)g3(0).

Consequently, the error order of interpolation is equalto 1.

We notice that the analysis of λ- error order ofinterpolation is deeper than the analysis of classicalerror order of interpolation.Case 2. The case of Birchoff type functionals

Page 6: A Generalization of the Remainder in Multivariate … Case of minimal interpolation schemes A finite set, Λ, of linear functionals is said to admit an ideal interpolation scheme

The functional λq,θ(f) = (q(D)f)(θ), q ∈ Πd, θ ∈R2 is a generalization of the derivative of f at the pointθ, that is why, it is important to study the λq,θ - errororder of interpolation.

Let’s choose q(x, y) = m(1,0)(x, y) = x and θ =(t1, t2) ∈ R2. The generating function is λν

q,θ(x, y) =xet1x+t2y.

Next, we will denote λ = λm(1,0),(t1,t2). The resultswe have obtained are:

(ενΛ,λ)[0] = 0; (εν

Λ,λ)[1] = 0;

(ενΛ,λ)[2] = x2

[b4

a4+b4 t1 + +a(a6−a4−a4b2+a2b2)4(a2+b2)(a4+b4)

]+

+t2xy + y2[

a2b2

a4+b4 t1 + b2(−a6+a4+a4b2−a2b2)4a(a2+b2)(a4+b4)

]

That means that deg(ενΛ,λ) = 2 and the λ-error or-

der of interpolation is equal 2. If t2 = 0 and a6−a4b2−b6 − b4 = 0 we have deg(εν

Λ,λ) > 2 and in this case theλ-error order is greater then 2.

For the set of functionals Λ, we considered in thissection, obviously ker(Λ) is a polynomial ideal, that isHΛ↓ is a minimal interpolation space of order 2. Hence,we can apply the results obtained in subsection ??. Wewill have three blocks of functionals:Λ(0) = {λ[0]

1 }; λ[0]1 = λ(0,0)

Λ(1) = {λ[1]1 , λ

[1]2 }; λ

[1]1 = λ(1,0), λ

[1]2 = λ(0,1)

Λ(2) = {λ[2]1 }; λ

[2]1 = λ(2,0)

Actually, the functionals λα are the evaluation func-tionals, δθ, θ ∈ Θ and it is their order in blocks thatwe determine using the constructive method of Newtonbasis.

The sets of index are :I0 = {(0, 0)};I1 = {(0, 0), (1, 0), (0, 1)};I2 = {(0, 0), (1, 0), (0, 1), (2, 0)}.We start with basis gi, i = 1, . . . , 4, and obtain theNewton basis:p[0]1 (x, y) = 1;

p[1]1 (x, y) =

y

b;

p[1]2 (x, y) = − y

2b− x− a

2a;

p[2]1 (x, y) = p

[2]1 (x, y) =

d1x2 + d2y

2

2(d2b2 − d1a2)− y

2b−

− d1a2

2(d2b2 − d1a2).

RΛ,λ(f) = d3[Λ(0), Λ(1), Λ(2), λ; f ]J0 = {(0, 0)}; J1 = {(1, 0), (0, 1)}; J2 = {(2, 0)}d0[λ; f ] = λ(f)d1[Λ(0), λ; f ] = λ(f)− f(a, 0) · λ(p[0]

1 )d2[Λ(0), Λ(1), λ; f ] == d1[Λ(0), λ; f ]− d1[Λ(0), δ(0,b); f ] · λ(p[1]

1 )−−d1[Λ(0), δ(−a,0); f ] · λ(p[1]

2 )d3[Λ(0), Λ(1),Λ(2), λ; f ] = d2[Λ(0),Λ(1), λ; f ]−

−d2[Λ(0),Λ(1), δ(0,−b); f ] · λ(p[2]1 )

References

[1] de Boor C., Ron A., On multivariate polyno-mial interpolation, Constr. Approx. 6, 1990,pag. 287-302.

[2] de Boor C., On the error in multivariate poly-nomial interpolation, Math. Z., 220, 1992,pag.221-230.

[3] de Boor C., Ron A., The least solution for thepolynomial interpolation problem, Math.Z.,220, 1992, pag. 347-378.

[4] de Boor C., Gauss elimination by segmentsand multivariate polynomial interpolation, Ap-proximation and Computation : A Festschriftin Honor of Walter Gautschi, Birkhauser Ver-lag, 1994, pag. 87-96.

[5] Gasca M., Sauer T., Polynomial interpolationin several variables, Advances in Computa-tional Mathematics, 1999.

[6] Sauer T., Polynomial interpolation of minimaldegree, Numer. Math., 78, 1997, pag. 59-85.

[7] Sauer T., Polynomial interpolation of minimaldegree and Grobner bases, Grobner Bases andApplications (Proc. of the Conf. 33 Year ofGrobner Bases), vol.251 of London Math. Soc,Lecture Notes, Cambridge University Press,1998, pag.483-494,.

[8] Sauer T, Xu Y., On multivariate Lagrange in-terpolation, Math. Comp., 64, 1995, pag. 1147-1170.

[9] Simian D., The dual of a polynomial interpola-tion operator, Proceedings of the annual Meet-ing of the Romanian Society of Mathemati-cal Sciences, Ed. Univ. Transilvania, Brasov,2001, pag.289-296.

[10] Simian D., On some particular examples forleast interpolation, Proceedings of the 6-th An-nual Conference of the Romanian Society ofMathematical Sciences, Sibiu, vol. I, 2003, pag.129 - 146.