Abstract
Purpose
The aims of the paper are to obtain necessary and sufficient conditions ofexistence and smoothness for non-polynomial spline spaces of fifth order, toestablish the uniqueness of the B φ -spline spaces in the class C4among mentioned spaces (under condition of fixed grid), and toprove the embedding of the B φ -spline spaces corresponding to embedded grids.
Methods
In the paper, the approximation relations with initial grid and with completechain of vectors are applied to obtain the minimal spline spaces. Usage oflocally orthogonal chain of vectors gives opportunity to construct specialapproximation relations from which the initial space of B φ splines is constructed.
Results
Deletion of a knot from initial grid gives a new grid, and as result, a newspace of B φ splines is embedded in the initial space mentioned above.
Conclusions
Consequent deletion of the knots (one by one) generates the sequence of theembedded spaces of B φ splines. Obtained results are successfully proved. They may be appliedto spline-wavelet decompositions.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
Introduction
Wide application of spline approximations in the theory and practise is well known [1–4]. Essential point for spline constructions is the usage of approximationrelations which firstly appeared in the finite element method. Approximationrelations were introduced in [5–7]. Approximation and interpolation properties of spline-waveletdecompositions are based on relations just mentioned [8].
The main problem of spline-wavelet decompositions is the construction of the chain ofembedded spaces. The problem is discussed in the case of the spaces ofnon-polynomial splines for irregular grids [9–12]; there, the chains are constructed for smooth spaces (which are calledspaces of B φ splines) of the first, second, and third orders. Spline spaces of arbitraryorder were constructed by [8], but the embedding was obtained only for special non-smooth splinespaces.
The aims of this paper are to obtain necessary and sufficient conditions of existenceand smoothness for non-polynomial spline spaces of fifth order, to establish theuniqueness of the B φ -spline spaces in the class C4among mentioned spaces (of fixed grid), and to prove the embedding ofthe B φ -spline spaces corresponding to embedded grids.
Methods
In the paper, the approximation relations with initial grid and with complete chainof vectors are applied to obtaining of the minimal spline spaces. Usage of locallyorthogonal chain of vectors gives opportunity to construct special approximationrelations from which the initial space of B φ splines is constructed.
Results and discussion
Spaces of (X,A,φ) splines of fifth order
Consider the grid X={x j }j∈Z on interval (α,β),
where x j ∈(α,β), , , ∀j∈Z, c1, and c2 are fixed.
Let φ(t) be column vector functionφ(t)∈R6,φ∈C5[α,β] such that
A set A={a j }j∈Z of column vectors a j ∈R6is called a chain of vectors. Differentenumerations are permitted for the same chain of vectors: two differentenumerations can be distinguished from each other with constant summand anddirection of numbering, for example, with j′=−j + j0 (where j0 is an integer constant) is another enumeration of the samechain.
A chain A={a i }i∈Z is locally orthogonal to a chain B={b j }j∈Zif there exist enumerations such that
where I5={1,2,3,4,5}.
Lemma 1
If a chain A is locally orthogonal to a chain B, then B islocally orthogonal to A.
Proof
Using Equation 1.2, we get
Now, we enumerate the chain {b j }j∈Z by definition b〈j〉=bj + 6.
Then,
This concludes the proof. □
By Lemma 1 the local orthogonality is symmetric, and therefore, the discussedchains A and B are called locally orthogonal chains.
A chain A={a j }j∈Z is called a non-generated chain ifa j ≠0∀j∈Z.
Consider a square matrix A j =(aj−5,aj−4,aj−3,aj−2,aj−1,a j ).
The chain A={a j }j∈Z is called a complete chain of vectors ifdetA j ≠0∀j∈Z. It is clear that the complete chain isnon-generated.
Let denote a set of all complete chains.
Lemma 2
Suppose A={a i }i∈Zand B={b j }j∈Zare non-generated locally orthogonal chains,then the following conditions are equivalent:
-
(1)
the chain A is a complete chain,
-
(2)
the chain B is a complete chain.
Proof
Let B be a complete chain. If A is not complete, then thereexists j∈Z such that a j =c1aj−1 + c2aj−2 + c3aj−3 + c4aj−4 + c5aj−5, where c1,c2,c3,c4,c5∈R1. Taking into account that A andB are locally orthogonal (see (1.2)), we obtain
From Equation 1.2, we also deduce that
Combining Equations 1.3 and 1.4 gives a j =0. This contradiction proves the first part of the lemma. Second partis proved analogously. □
Lemma 3
If A={a i }i∈Z and B={b j }j∈Z are complete chains and relation (1.2) isfulfilled, then
Proof
The proof is by reductio ad absurdum. If there existsj∈Z such that relation (1.3) holds, then by Equation 1.4, weget a j =0; this contradicts the completeness of the chain A. The firstrelation in Equation 1.5 is proved. Arguing as above, we find that thesecond relation in Equation 1.5 is true. □
Lemma 4
For an arbitrary complete chain A={a j }j∈Z, there exists a non-generated locallyorthogonal chain B={b j }j∈Zwhere the directions of vectors b j are uniquely defined (up to non-zero constant factor).
Proof
We fix integer k; let b k ∈R6be a vector satisfying
The conditions (1.6) are the linear system with respect to unknown vectorb k :
By assumption, the chain A is complete; hence, vectors ak−5,ak−4, ak−3,ak−2, and ak−1are linearly independent. System (1.7) has aunique (up to constant factor) non-trivial solution b k , which can be defined by the identity
This concludes the proof. □
Corollary 1
For any complete chain, there exists a locally orthogonal complete chaindefined uniquely up to non-zero constant factors.
Proof
Combining Lemmas 2 and 4, we prove the corollary.
By definition, put
For we define the functions ω j (t), t∈M, and j∈Z, byapproximation relations
If t is fixed in (x k ,xk + 1), then relation (1.8) contains at most six non-zerosummands:
Discuss relation (1.9) as the system of linear equations with respect tounknown values ω j (t).
Let j be an arbitrary fixed integer. Consider relations (1.9) withk=j, k=j + 1, k=j+ 2, k=j + 3, k=j + 4, andk=j + 5. By Cramer’s theorem, we obtain
Now, it follows from supp ω j ⊂[x j ,xj + 6] (see (1.8)) that the function ω j (t) is defined for all t∈M.
Equivalent record of formulas (1.10) to (1.15) is
here J k ={k−5,k−4,k−3,k−2,k−1,k},columns of determinants are ordered uniformly, symbol ||′jφ(t) indicates that column vectorφ(t) has to be placed instead of the columnvector a j .
By (1.1) the set of functions ω j ∀j∈Z is a linearly independent system onarbitrary subinterval (a,b) of the interval(α,β). Consider the linear space
where is the linear span, and C l p is the closure in the topology of point convergence. The spaceS5(X,A,φ) is called a spaceof minimal(X,A,φ)splines of fifthorder. □
Properties of minimal splines
Lemma 5
Suppose A is a complete chain and let k∈Z, t∗∈[x k ,xk + 1] be fixed, the following relations areequivalent:
Proof
The proof is a consequence of (1.16). □
Lemma 6
Let A be a complete chain. For the validity of the equalities
it is necessary (and if ak−6 and a k are not collinear, then it is also sufficient) that
Proof
Necessity. Replacement of k by k−1 in (1.9) gives
Hence, by passing to the limits as t→x k −0 and using the first condition (2.1), we obtain
Analogously by using (1.9) and the second condition (2.1) ast→x k + 0, we arrive at
In view of the linear independence of vectors ak−5, ak−4, ak−3, ak−2, and ak−1 by equalities (2.4) to (2.5), we obtainrelations (2.2). These completes the proof of necessity.
Sufficiency. Using (2.3) and (1.9), we have
and
respectively. It follows from (2.6), (2.7), and (2.2) that ak−6ωk−6(x k −0)=a k ω k (x k + 0); using linear independence of vectors ak−6and a k , we obtain equalities (2.1).
This completes the proof of Lemma 6. □
Theorem 1
Let A be a complete chain of vectors; then, the following conditionsare equivalent:
-
(1)
for any j∈Z functions ω j (t) are continuous on (α,β);
-
(2)
limit values at the boundary points of supp ω j are zero for all j∈Z.
Proof
Sufficiency. Since vector-function φ(t) is continuous,it will suffice to prove the continuity of the function ω j (t) in the knots of the grid X. If a knot x k is placed on the boundary of S j , the continuity of ω j (t) at x k follows from the hypotheses of the theorem. If a knot x k is placed inside S j , then conditions of Lemma 6 are fulfilled (in the part of necessity),and therefore, the relations (2.2) are true. Sufficiency is proved.
Necessity is obvious.
By definition, put φ k =φ(x k ), . □
Theorem 2
The following conditions are equivalent:
-
(1)
functions ω j (t)(∀j∈Z) can be extended to (α,β) continuously;
-
(2)
relations
(2.8)
are fulfilled.
Proof
Let us show that the condition (2) in Theorem 1 is equivalent to relations(2.8). Indeed, by (1.10), we see that equalities (2.8) are equivalent to
Changing j to j + 6 in (2.8), we get
now using (1.15), we obtain
Thus, the equalities (2.8) are equivalent to the equalities (2.9) and (2.11);the last ones indicate that (2) is fulfilled. To conclude the proof, itremains to use Theorem 1. □
Theorem 3
Suppose φ(S)∈C(α,β),where S is a positive integer; then, the following conditions areequivalent:
-
(1)
the functions can be extended to (α,β) continuously,
-
(2)
relations
are fulfilled.
Proof
By differentiating Equation 1.9 and using the arguments applied for proofs ofTheorems 1 and 2, we arrive at the required result. □
B φ splines of fifth order
Discuss a special case of relations (1.8) choosing a specific chain {a j }j∈Z.
First, we define vectors d k ∈R6by identities
and introduce vectors with symbolic determinant
Theorem 4
Suppose that (1.1) is valid. Then, there exists δ>0 suchthat for h X <δ, the chain is complete.
Proof
Using (1.1), we apply Taylor’s expansion for x=x j elements of determinant (3.2); as a result, we get asymptotic behaviorof (as h X →0) uniformly with respect tox∈(α,β). Using theasymptotics, the reader will easily prove the theorem. □
Theorem 5
The chains of vectors and are locally orthogonal:
Proof
Using (3.2), we get
If p=1, then (3.1) implies that the first row of the determinant iszero; if p=2,3,4,5, then there are two equal rows in the discusseddeterminant. This completes the proof of Theorem 5.
Suppose the chain is complete. We replace a j by in (1.8). Then, we obtain specific minimal splines, whichare denoted by . □
Theorem 6
The relations are true.
Proof
Relations (3.3) can be rewritten as or, equivalently,
Using (3.1), we have
It follows from (3.4) and (3.5) that
therefore, for S=0,1,2,3,4. Thus,
To conclude the proof, we refer to Theorem 3. □
By definition, put
According to Theorem 6, we obtain
by S5(X,φ) denoting the set of all spaces of thesplines of fifth order (where grid X and function φ arefixed):
Theorem 7
There is a unique space in S5that is contained in C4(α,β).
Proof
By Theorem 3, the condition ω j ∈C4(α,β) ∀j∈Z isequivalent to relations (2.12) for all S∈{0,1,2,3,4}. Supposea chain {a j }j∈Zsatisfies
We wish to prove that vector a j differs from with nonzero constant factors.
By Lemma 4, there exists a non-generated chain B={b j }j∈Zlocally orthogonal to chain {a j }j∈Z; directions of vectors b j are defined uniquely (up to constant factor):
Using condition (3.6), we have
hence, b j ⊥φ j(S), S=0,1,2,3,4.
Thus,
By assumption (1.1), vectors φ j , φ j′, φ j ′′, φ j ′′′, and φ j IV are linearly independent; it follows that b j =c j d j (see (3.1)), where constants c j are non-zero (for all j∈Z). The chains {a j }j∈Z and , which are locally orthogonal to the chains {b j }j∈Z and {d j }j∈Z respectively, also differ by non-zero constantfactors c j⋆: . Taking into account Equations 1.10 to 1.15, we have ; this completes the proof. □
Consider the relations
where t∈(x k ,xk + 1). Multiply both sides (3.7) on the left by , wherej=k−5,k−4,k−3,k−2,k−1,k.By local orthogonality of the chains and , we obtain six scalar relations
In the same way, multiplying both sides of (3.7) by for j=k + 1,k + 2,k +3,k + 4,k + 5,k + 6 and using the localorthogonality of the chains and , we get the next system of relations
Discuss relations (3.8) and (3.9) as the systems of linear equations with respectto unknown values ω j (t), j∈J k . Triangular matrices of these systems are non-singular since theirdiagonal elements are nonzero (see Lemma 3).
Note. Indeed, the solutions of the systems (3.8) and (3.9) areidentical.
Let I0={0,1,2,3,4,5}, , , where p∈I0.
Lemma 7
The formulas (3.8) define the functions on the interval t∈(x k ,xk + 1) by values of the vector-functionφ(t) on this interval complemented with values ofmentioned vector function and its derivatives up to the fourth order in theknots .
Proof
From (3.1), it follows that vector d j is defined by vector-function φ(t) and itsderivatives up to the fourth order in the knot x j . Using (3.2), we see that the vector is defined by vector-function φ(t)and its derivatives up to the fourth order in the knots xs + i∀i∈I5. To conclude the proof, one should use the formula (3.8).□
Lemma 8
According to (3.9) the functions for t∈(x k ,xk + 1) are defined by the vector-functionφ(t) on this interval complemented with values ofthe mentioned vector function and its derivatives up to the fourth order inthe knots .
Proof
We again recall (3.1) and conclude that d j is defined by φ j and for q∈{1,2,3,4}; besides, according toformula (3.2), the vector a s is defined by φs + iand , where q∈{1,2,3,4},i∈I5. Now, the result follows from (3.9). □
Lemma 9
The functions for p∈I0and t∈(x k ,xk + 1) are defined by complemented with φk + iand for q∈{1,2,3,4}, .
Proof
Since the values of vector function and its derivatives in the knots areunrestricted, the desired result follows from Lemmas 7 and 8. □
Theorem 8
The function is completely defined by .
Proof
Using Lemma 9, we put k=j,j + 1,j +2,j + 3,j + 4,j + 5. This completes the proofof Theorem 8. □
Calibration relations
We enlarge the origin grid X by deletion of knot xk + 1, i.e., we put
and discuss a new grid ,
We put , and introduce the vectors by identities
Now, we define vectors by symbolic determinant:
As before, we discuss approximation relations
and put
Theorem 9
If the grid is so fine that the chain is complete, then
Proof
From the definition of the grid and (4.2) to (4.3), we deduce that
With Equations 1.8 and 4.3, we have
and (after annihilation of identical terms, defined by relations (4.5) to(4.6)) we obtain
Discuss relations (4.7) as a system of linear equations with respect tounknown values , i∈I0. Because of the completeness of the chain , the matrix of the system is nonsingular. Now, we obtainmentioned functions , i∈I0, as linear combinations of basic functions of the space ; these representations are calibration relations indiscussed case [8]. The inclusion (4.4) is proved. □
Conclusions
Consequent deletion of the knots (one by one) generates the sequence of the embeddedspaces of B φ splines. Obtained results are successfully proved. They may be applied tospline-wavelet decompositions.
Author’s information
YKD is the Head of the Department of Parallel Algorithms. He is also a professor anda doctor of science.
References
Stechkin SV, Subbotin YN: Splines in calculate mathematics. Nayka Press, Moscow; 1976. (in Russian) (in Russian)
Malozemov VN, Pevnyj AV: Polynomial splines. University Press, Leningrad; 1986. (in Russian) (in Russian)
Schumaker LL: On super splines and finite elements. SIAM. J. Numer. Anal 1989, 26: 997–1005.
Burova IG, Dem’yanovich YK: Theory of minimal splines. University Press, St. Petersburg; 2000. (in Russian) (in Russian)
Goel JJ: Construction of basis functions for numerical utilization of Ritz’smethod. Numer. Math 1968, 12: 435–447.
Michlin SG: Variation-net approximation. Zap Nauchn Seminarov LOMI AN SSSR 1974, 48: 32–188. (in Russian) (in Russian)
Dem’yanovich YK, Michlin SG: On grid approximation of Sobolev’s spaces. Zap Nauchn Seminarov LOMI AN SSSR 1973, 35: 6–11. (in Russian) (in Russian)
Dem’yanovich YK: Minimal splines and wavelets. Mathematics 2008, 41(2):88–101. (in Russian) (in Russian)
Buhmann MD: Multiquadric prewavelets on nonequally spaced knots in one dimension. Math. Comput 1995, 64(212):1611–1625.
Dem’yanovich YK: Wavelet decompositions on nonuniform grid. In Proceedings of the St. Petersburg Mathematical Society XIII Edited by: Uraltseva NN. 2008. [Please insert venue and complete date of the proceeding]. AmericanMathematical Society Translations, series 2, vol. 222, pp. 23–43. AmericanMathematical Society, RI, USA [Please insert venue and complete date of the proceeding]. AmericanMathematical Society Translations, series 2, vol. 222, pp. 23–43. AmericanMathematical Society, RI, USA
Dem’yanovich YK, Kosogorov OM: Calibration relations for nonpolynomial splines. J. Math. Sci 2010, 164(3):364–382.
Dem’yanovich YK: Embedded spaces of trigonometric splines and their wavelet expansion. Math Notes 2005, 78(5):615–630.
Acknowledgements
The author is grateful to Professor S.V. Poborchi for the useful discussions.This research was partially supported by RFBR grant numbers 10-01-00245 and10-01-00297.
Author information
Authors and Affiliations
Corresponding author
Additional information
Competing interests
The author declares that he has no competing interests.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 2.0 International License (https://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
About this article
Cite this article
Dem’yanovich, Y.K. Embedding of non-polynomial spline spaces. Math Sci 6, 28 (2012). https://doi.org/10.1186/2251-7456-6-28
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/2251-7456-6-28