Further development of the new algorithms for accelerating the convergence of functional-type sequences (Q876658): Difference between revisions
From MaRDI portal
Created a new Item |
Added link to MaRDI item. |
||
links / mardi / name | links / mardi / name | ||
Revision as of 16:45, 30 January 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Further development of the new algorithms for accelerating the convergence of functional-type sequences |
scientific article |
Statements
Further development of the new algorithms for accelerating the convergence of functional-type sequences (English)
0 references
26 April 2007
0 references
The mathematical literature dealing with convergence acceleration is replete with determinantal quotient expressions written out in full. Square kilometres of page space might have been saved by adopting the suggestion that these expressions may be cast in a different form which reveals both their meaning and their common origin. \(\mathbf A \) being an \(r \times (r+1)\) matrix over a suitable field, \(\mathbf x\) and \(\mathbf y\) being \((r+1)\) row vectors and \([\mathbf x: \mathbf A]\) being the \((r+1) \times (r+1)\) matrix formed from \(\mathbf x\) followed by the rows of \(\mathbf A \), the first component of the solution vector \(\mathbf z\) of the equation \(\mathbf z\cdot[\mathbf y: \mathbf A]=\mathbf x\) is \(\det[\mathbf x: \mathbf A]/[\mathbf y: \mathbf A]\). This determinantal quotient may therefore be expressed as \([\mathbf x/\mathbf y\|\mathbf A\| r]\). The numbers, often referenced by integer indices, resulting from various recursive schemes may be expressed in closed form by use of this notation (\(\mathbf x\), \(\mathbf y\) and \(\mathbf A \) then depend upon the indices). Such quotients appear in the paper under review, which is concerned with the transformation of a sequence of real functions taken to be integrable over an interval \([a,b]\). A sequence of \(r\)-th order transforms is defined for \(m=0, 1, \dots\); each transform of order \(r+1\) is defined in terms of integrals of certain \(r\)-th order transforms. The author considers six transformations which he denotes by \(\alpha\), \(\beta\), \(\gamma\), \(\delta\), \(\phi\) and \(\rho\). For all of them, \(\mathbf y\) is the vector \((1, 1, \dots,1)\). For the \(\alpha\)-transform \(\mathbf x\) is the vector \(\alpha (m+\nu, r-1)\), for the \(\beta\) transform the components of \(\mathbf x\) are \(\beta (m+\nu , r-1)\) and so on. For all but one of the transformations, \(\mathbf A\) has the form \(\mathbf A'\). \(\mathbf d\) where \(\mathbf d\) is an \((r+1) \times (r+1)\) diagonal matrix. \(\mathbf A'(0)\) has elements \({\mathbf A}'(\tau , \nu)=(m+\nu)^{\tau}\) for \(0 \leq \tau < r\), \(0 \leq \nu \leq r\); for \(\mathbf A'(1)\), \({\mathbf A}'(\tau, \nu)= (m+\nu)^{-\tau}\) for the same indices; \(\mathbf d(0)\) has diagonal elements of the form \[ d(\nu,\nu)= \biggl\{\int \Delta \xi (m+ \nu ,r-1)\,dx \mid a \leq x \leq b\biggr\} \quad (0 \leq \nu \leq r), \] while those of \(\mathbf d(1)\) have the form \[ d(\nu,\nu)= \biggl\{\int \Delta\xi (m+ \nu +1,r-1)\,dx\cdot \Delta \xi (m+ \nu ,r-1)\,dx \mid a \leq x \leq b\biggr\} \] (\(\Delta\) operates on \(m: \Delta \xi (m,r)= \xi (m+1,r)-\xi (m,r)\)). For the \(\alpha\) transformation, \(\mathbf A'\) and \(\mathbf d\) are \(\mathbf A'(0)\) and \(\mathbf d(0)\); for the \(\beta\), \(\gamma\) and \(\delta\) transforms they are \(\mathbf A'(1)\), \(\mathbf d(0)\) and \(\mathbf A'(0)\),\(\mathbf d(1)\) and \(\mathbf A'(1)\) and \(\mathbf d(1)\). For the \(\rho\) transform, \(\mathbf A'\) is \(\mathbf A'(1)\) while the elements of \(\mathbf d\) are those of \(\mathbf d(1)\) divided by \(\{\int \Delta^{2} \rho (m+ \nu ,r-1) \mid a \leq x \leq b\}\). For the \(\varphi\)-transform, \(\mathbf A'\) is a bordered version of a matrix having inverse factorials as elements, while \(d\) is \(d(1)\). (In the paper all of the determinantal quotients are written out in full.) Numerical illustrations of the transforms are provided. In the proofs of theorems suggesting that the quotients are well determined there is some confusion: Hankel determinants are formed from elements \(c(\tau,\nu)\) for which \(c(\tau,\nu)= \psi (\tau+\nu)\) for some \(\psi\); an \(r \times r\) Hankel determinant is formed from \(2r-1\) elements. The determinants involved here are not of this form. The convergence results are straightforward guesses: the theorems are stated without proof.
0 references
sequences of functions
0 references
convergence acceleration
0 references
improved functional epsilon algorithm
0 references