Christoffel transformations for multivariate orthogonal polynomials (Q1685957): Difference between revisions

The paper under review is an extension and generalization of an earlier work of the authors [Adv. Math. 302, 628--739 (2016; Zbl 1350.15002)] wherein they showed that the Gauss-Borel or LU factorization of a moment matrix paved the way for investigating the links between the ``multivariate orthogonal polynomials (MVOPR) on a multidimensional real space \({{\mathbb R}^D},D \geqslant 1\), and integrable systems of Toda and KP type''. In that study the authors exhibited that the LU decomposition led to a ``simple construction of the Christoffel-Darboux formula'' which technique was also found fruitful by them in their present study in which they construct ``new MVOPR, associated with a perturbed measure, from the MVOPR of a given nonperturbed measure'' which has straightway led to a method of construction of new solutions of ``the associated Toda-KP type integrable systems''. In the previous study [loc. cit.] the authors presented the `elementary Darboux transformations' ``as the multiplication of the nonperturbed measure by a degree one multivariate polynomial'' while in this paper they show that the \(m\)th iteration of these elementary Darboux transformations generates ``a perturbation by a multivariate polynomial of degree \(m\)''. Remarking that the Christoffel transformations presented by the authors in their previous study [loc. cit.] are only a small part of a still larger undiscovered dominion, the authors generalize the Christoffel transformations of their preceding study [loc. cit.] in this paper ``for a perturbation by a polynomial of any degree'' whether prime or non-prime. This has produced a refined ``quasi-determinantal expression for the new MVOPR'' which generalizes their 1D determinantal Christoffel formula. An in-depth analysis of the theory of multivariate orthogonal polynomials can be found by the interested reader in [\textit{C. F. Dunkl} and \textit{Y. Xu}, Orthogonal polynomials of several variables. 2nd ed. Cambridge: Cambridge University Press (2014; Zbl 1317.33001)]. The authors have used tools like `sample matrices, poised sets and the use of quasi-determinants' developed in multivariate interpolation theory by \textit{P. J. Olver} [Stud. Appl. Math. 116, No. 2, 201--240 (2006; Zbl 1145.41311)]. They have used the concepts of poised sets which have non-singular sample matrix. Here they begin their ``study of poised sets for general Darboux transformations''. Their method of analysis can be divided into two categories: the first one is a ``measure-independent part depending exclusively on the relative positions of nodes in the algebraic hypersurface of the generating polynomial'' while the second one is ``related to the nonperturbed measure and the corresponding Jacobi matrices''. For the geometrical part of their study the authors have utilized the concepts of multivariate Vandermonde matrices and the multivariate confluent Vandermonde matrices. By invoking the tools of algebraic geometry developed in [\textit{R. Hartshorne}, Algebraic geometry. New York-Heidelberg-Berlin: Springer-Verlag (1977; Zbl 0367.14001); \textit{I. R. Shafarevich}, Basic notions of algebra. Translated from the 1986 Russian original by Miles Reid. Reprint of the 1997 English translation. Berlin: Springer (2005; Zbl 1074.16501)] the authors show in Theorem 3.3 that for perturbing polynomials which are expressible as the product \(Q = {Q_1} \cdots {Q_N}\) of \(N\) prime factors ``that there exists, in the complex domain, poised set of nodes by forbidding its belonging to any further algebraic hypersurface, different from the algebraic hypersurface of \(Q\), of lower degree''. For the bivariate orthogonal polynomials they inquire into the prime perturbations of the Legendre product polynomials. They deduce the Multivariate Christoffel Formula in Theorem 2.1 and state that this transformation is not applicable ``for a perturbation of the measure by a polynomial of the form \(Q = {\mathcal R}^d\)'' due to the non-existence of the poised sets. To overcome this situation they use multi-Wronskian type matrices and multivariate confluent Vandermonde matrices to extend the result of Theorem 2.1 in Theorem B.3 for dealing with these types of cases ``for a perturbing polynomial of the form \(Q = Q_1^{{d_1}} \cdots Q_N^{{d_N}}\) where the polynomials \(Q_i\) are irreducible''. They discuss the poised sets of these cases in Theorem B.4 and point out that in such cases ``the set of nodes when poised cannot belong to any further algebraic hypersurfaces of certain type''. In the first section of the paper the authors reproduce the necessary background material for the reader from their earlier study [the authors, loc. cit.] and discuss the ``Christoffel transformation generated by a multivariate polynomial'' in the Section 2. The third section of the paper is devoted to the detailed discussion on poised sets where the authors formulate certain ``conditions for the nodes in order to constitute a poised set''. They also present three appendices in this paper. In Appendix A they apply ``the bivariate Christoffel formula to the product Legendre orthogonal polynomials on the unit square \([- 1, 1] \times [- 1, 1]\)'' by using \(x+y+3\) and \(4-x^2-y^2\) as the perturbing irreducible polynomials and calculating the perturbed bivariate orthogonal polynomials up to \(k = 2\) for the former and \(k=4\) for the latter. They say that they have checked the orthogonality relations in all these cases by using SAGE. Their Appendix B is a detailed analysis of ``the general case when the perturbing polynomial is not a product of prime factors'' and Theorem 2.1 fails. Finally in Appendix C the authors appeal to the concepts of Schur complements and quasi-determinants and by applying the `\textit{heredity principle of quasi-determinants}, i.e., ``quasi-determinants of quasi-determinants are quasi-determinants''' develop their results for the Christoffel formulas in the multivariate case. Talking about the significant results developed in this paper the reviewer would like to refer to the following results: The notable result in Section 2 on p.252 for the Darboux transformation of MVOPR, found by using a polynomial perturbation of the measure in which the perturbations can be written as a product of \(N\) distinct prime polynomials, is: Theorem 2.1 (Multivariate Christoffel Formula). For a poised set of nodes \({{\mathcal P}_{k,m}} \subset Z(Q)\) the Christoffel transformation of the orthogonal polynomials can be expressed in terms of the original ones as the following last quasi-determinantal expression \[ [\widehat P(\mathbf{x})]_k=\frac{[Q(\Lambda)]_{k,k + m}} {Q(\mathbf{x})}\Theta _*\left(\begin{matrix} & | & [P(\mathbf{x})]_k \\ \Sigma _k^m & | & \vdots\\ &|& [P(\mathbf{x})]_{k+m-1}\\ {- - - - - - -} & | & {- - - - - - -} \\ \Sigma _{[k,m]} & | & [P(\mathbf{x})]_{k+m} \end{matrix}\right).\tag{1} \] In Section 3 by utilizing the various concepts from algebraic geometry like poised sets, multivariate Vandermonde matrices, Hilbert's Nullstellensatz, etc. the authors state and prove on p.257: Theorem 3.3: (Algebraic Hypersurfaces and Poised Sets). Let the perturbing polynomial be of the form \(Q(z) = {Q_1}(z) \cdots {Q_N}(z)\), a product of \(s\) different irreducible polynomials with \(\deg {Q_a}(z) = {m_a},a \in \left\{ {1, \ldots ,N} \right\}\), and \(\deg Q(z) = m = \sum_{a = 1}^N {{m_a}}. \) Then, the set of nodes \({{\mathcal {P}}_{k,m}} \subset Z(Q) = \bigcup_{a = 1}^N {Z({{Q_a}})} \) is poised if the nodes do not belong to any further complex algebraic hypersurface of degree smaller than \(k + m\) and different from \(Z(Q)\). The reviewer feels that the Appendix B is the most impressive of all the three appendices of the paper, where the authors give a Christoffel formula for a general perturbation by generalizing the construction of nodes, sample matrices and poised sets by considering the multi-Wronskian type matrices and the multivariate confluent Vandermonde matrices and identifying the multivariate polynomials with symmetric tensors. Stating various definitions, theorems and propositions they state and prove on pp.277--278: Theorem B.3 (General Multivariate Christoffel Formula). For a poised set of nodes \({{\mathcal P}_{k,{n_1}{r_1} + \cdots + {n_s}{r_s}}}\) in the algebraic hypersurface \(\bigcup_{a = 1}^s {Z({{Q_a}})}\) the Christoffel perturbed orthogonal polynomials can be expressed in terms of the original ones as the following last quasi-determinantal expression \[ \begin{multlined} {[{\widehat P}]_k}(\mathbf{x}) = \frac{{{{[{\prod_{a = 1}^s {{{({{R_a}(\mathbf \Lambda)})}^{{n_a}}}} }]}_{k,k + {n_1}{r_1} + \cdots + {n_s}{r_s}}}}} {{{{({{R_1}(\mathbf{x})})}^{{n_1}}} \cdots {{({{R_s}(\mathbf{x})})}^{{n_s}}}}}{\Theta _*}\\ \times\left(\begin{matrix} & | & {{{[{P(\mathbf{x})}]}_k}}\\ {\Sigma _k^{{n_1}{r_1} + \cdots + {n_s}{r_s}}} & | & \vdots \\ & | & {{{[{P(\mathbf{x})}]}_{k + {n_1}{r_1} + \cdots + {n_s}{r_s} - 1}}} \\ {- - - - - - - - -} & | & {- - - - - - - - - - -} \\ {{\Sigma _{[{k,{n_1}{r_1} + \cdots + {n_s}{r_s}}]}}} & | & {{{[{P(\mathbf{x})}]}_{k + {n_1}{r_1} + \cdots + {n_s}{r_s}}}} \end{matrix}\right)\end{multlined}\tag{2} \] and in the sequel the following result on p.278: Theorem B.4 (Poised Sets for Permutations in General Position). The node set \({{\mathcal P}_{k,{n_1}{r_1} + \cdots + {n_s}{r_s}}} \subset {{\mathbb C}^D}\) is poised if there does not exist a polynomial \(V(\mathbf z) \neq \prod_{a = 1}^N {{{({{R_a}(\mathbf z)})}^{{n_a}}}} ,\deg V \leqslant k + {n_1}{r_1} + \cdots + {n_s}{r_s} - 1\), such that \({\mathcal P}_i^{({a,j})} \subset Z({\frac{{{\partial ^j}V}} {{\partial n_i^{({a,j})}}}})\), for \(a = 1, \ldots ,N,j \in \left\{ {1, \ldots ,{\delta _a} - 1} \right\}\) and \(i \in \left\{ {1, \ldots ,{\rho _{a,j}}} \right\}\). The reviewer has noticed that the paper needs minor linguistic corrections at certain places and also some minor mathematical typesetting errors have crept in inadvertently. Now we point out the linguistic corrections first: On p.245 in lines 10--11 instead of the words ``\dots in order to extend to the multivariate scenario the classical interpolation theory are precisely \dots'' , it would be better to read as ``\dots in order to extend the classical interpolation theory to the multivariate scenario are precisely \dots''. On the same page in the line 15 the question should read ``In what sense do we extend the Christoffel formula?'' instead of ``In what sense we do extend the Christoffel formula?''. In Section 2 on p.250 the word ``Latter'' appearing in line 16 should be replaced by the word ``Later''. On p.254 in the statement of the Definition 3.2 the words ``\dots of degree less or equal than \(k + m - 1\) \dots'' in the line 8 should be replaced by ``\dots of degree less than or equal to \(k + m - 1\) \dots''. On p.259 in the Acknowledgments in the line 3 the word ``improve'' should better read as ``improved''. On p.276 the word ``remove'' appearing in the line 15 should be replaced by the word ``removed''. On p.278 in the line 18 in the statement ``Finally, the complete confluent Vandermonde matrix in \(\in {{\mathbb C}^{{D_{k + {n_1}{r_1} + \cdots + {n_s}{r_s} - 1}} \times {d_{k,{n_1}{r_1} + \cdots + {n_s}{r_s}}}}}\)'' the ``\(\in\)'' needs to be removed. On p.279 in the line 3 the words ``\dots considered at the light of \dots'' should better read as ``\dots considered in the light of \dots''. On p.280 the word ``disguise'' appearing in the line 10 should read as ``disguised''. On p.281 in the line 21 the words ``\dots the quasi-determinant associated to \dots'' should better read as ``\dots the quasi-determinant associated with \dots''. Next we point out some minor mathematical typesetting errors in this paper as noticed by this reviewer: On p.258 the statement of Proposition 3.9 should read as ``We have \(\sum_{a = 1}^N {{d_{k + m - {m_a},{m_a}}}} \geqslant {d_{k,m}}\)'' instead of ``We have \(\sum_{i = 1}^N {{d_{k + m - {m_a},{m_a}}}} \geqslant {d_{k,m}}\)''. On p.263 the entry at the position \((3,3)\) in the matrix \({\Sigma _{[{2,1}]}}\) should read as \({x_3}({y_3^2 - \frac{3} {5}})\) instead of \(x({y_3^2 - \frac{3} {5}})\) while on the same page the entry at the \((2,3)\) position in the matrix \(\Sigma _2^1\) should read \({x_3}{y_3}\) in place of \({x_2}{y_3}\). On p.267 the entry in the matrix \(\Sigma _1^2\) at the \((5,2)\) position should read as \(y_2^2 - \frac{1} {3}\) instead of \(2_3^2 - \frac{1} {3}\). On p.268 the entry at the \((5,2)\) position in the matrix \({\Sigma _{[{2,2}]}}\) should read as \(y_2^4 - \frac{6} {7}y_2^2 + \frac{3} {{35}}\) in lieu of \(y_2^4 - \frac{6} {7}y_{''}^2 + \frac{3} {{35}}\), further on the same page the entry at \((7,6)\) position in the matrix \(\Sigma _2^2\) should read \(({y_6^3 - \frac{3} {5}{y_6}})\) rather than \(({y_7^3 - \frac{3} {5}{y_7}})\). On p.270 the entry at \((4,1)\) position in the matrix \({[{Q({\Lambda})}]_{3,5}}{[{P({x,y})}]_{5}}\) should accurately read as \(({{x^2} + {y^2}})({{y^3} - \frac{3} {5}y}) - \frac{{38}} {{45}}({{y^3} - \frac{3} {5}y}) - \frac{{12}} {{175}}y\) instead of \(({{x^2} + {y^2}})({{y^3} - \frac{3} {5}y}) - \frac{{38}} {{45}}({{y^3} - \frac{3} {5}y}) + - \frac{{12}} {{175}}y\). This very same correction needs to be incorporated on p.271 at the \((4,1)\) position in the first matrix on the right side of the expression \(({4 - {x^2} - {y^2}}){[{\widehat P({x,y})}]_3}\). On p.274 the entry at the last position in the matrix \(({\Sigma _k^{nd}})_i^{(j)}\) needs to be read as \({[{\frac{{{\partial ^j}P}} {{\partial \mathbf {n}_i^{(j)}}}({\mathbf{p}_{i,v_i^{(j)}}^{(j)}})}]_{k + nr - 1}}\) in place of \({[{\frac{{{\partial ^j}P}} {{{\partial ^j}\mathbf{n}_i^{(j)}}}({\mathbf{p}_{i,v_i^{(j)}}^{(j)}})}]_{k + nr - 1}}\). Similarly, on p.277 the last entry of the matrix \(({\Sigma _k^{{n_1}{r_1} + \cdots + {n_s}{r_s}}})_i^{({a,j})}\) should read \({[{\frac{{{\partial ^j}P}} {{\partial {\mathbf n}_i^{({a,j})}}}({\mathbf {p}_{i,v_i^{({a,j})}}^{({a,j})}})}]_{k + {n_1}{r_1} + \cdots + {n_s}{r_s} - 1}}\) in place of \({[{\frac{{{\partial ^j}P}} {{{\partial ^j}\mathbf{n}_i^{({a,j})}}}({\mathbf{p}_{i,v_i^{({a,j})}}^{({a,j})}})}]_{k + {n_1}{r_1} + \cdots + {n_s}{r_s} - 1}.}\) On the whole the reviewer feels that this interesting paper advances the existing knowledge in the field of MVOPR and he agrees with the authors' remarks in the lines 29--31 on p. 245 ``To our knowledge this result (i.e., (2) above) is new. Moreover, to `the best of our' knowledge there is no similar result getting `this much closer' to the essence of the original Christoffel formula''.