Numerical Methods, Computational Mathematics
Numerical Mathematics is fascinating: optimization methods,
linear and nonliear Equation Systems, least square methods, SVD,
orthogonal polynomials, approximation theory and methods, etc.
are my favorite areas of interest. Professors
Gene Golub,
Walter Murray,
Carlo Tomasi,
Paul Nevai,
Peter Borwein,
Vilmos Totik, and
Tamás Erdélyi
are my mentors and co-authors.
MR95j:41015,
41A17 (41A20)
Peter Borwein,
Tamás Erdélyi
and J. Zhang
Chebyshev polynomials and Markov-Bernstein type inequalities for rational spaces.
J. London Math. Soc. (2) 50 (1994), no. 3, 501--519.
Summary
MR94h:42041 ,
42C05 (33C45)
J. Zhang
Relative growth of linear iterations and orthogonal polynomials on several intervals.
Linear Algebra Appl. 186 (1993), 97--115.
Summary
MR94f:42026
42C05 (39A10 41A17)
Peter Borwein,
Tamás Erdélyi
and J. Zhang
Müntz systems and orthogonal Müntz-Legendre polynomials.
Trans. Amer. Math. Soc. 342 (1994), no. 2, 523--542.
Summary
MR94d:42028
42C05 (33C45)
Paul Nevai and
J. Zhang
Rate of relative growth of orthogonal polynomials.
J. Math. Anal. Appl. 175 (1993), no. 1, 10--24.
Summary
MR93a:42010
42C05
Tamás Erdélyi,
Paul Nevai,
J. Zhang and
Jeff Geronimo
A simple proof of "Favard's theorem" on the unit circle.
Atti Sem. Mat. Fis. Univ. Modena 39 (1991), no. 2, 551--556.
Summary
MR92m:42025
42C05
Tamás Erdélyi,
Paul Nevai,
J. Zhang and
Jeff Geronimo
A simple proof of "Favard's theorem" on the unit circle.
Trends in functional analysis and approximation theory (Acquafredda di Maratea, 1989),
41--46,
Univ. Modena, Modena, 1991.
Summary
MR92k:42034
42C05
Paul Nevai,
J. Zhang and
Vilmos Totik
Orthogonal polynomials: their growth relative to their sums.
J. Approx. Theory 67 (1991), no. 2, 215--234.
Summary
... MORE ...
Summaries
95j:41015
41A17 (41A20)
Borwein, Peter(3-SFR); Erdélyi, Tamás(3-SFR); Zhang, J.(1-STF-C)
Chebyshev polynomials and Markov-Bernstein type inequalities for rational spaces.
(English. English summary)
J. London Math. Soc. (2) 50 (1994), no. 3, 501--519.
Summary: "We consider the trigonometric rational system
$$\Big \{1,\frac {1±\sin t}{\cos t-a\sb 1},\frac {1±\sin t}{\cos
t-a\sb 2},\cdots \Big \}$$
on $\bold R \bmod 2\pi$ and the algebraic rational system
$$\Big \{1,\frac {1}{x-a\sb 1},\frac {1}{x-a\sb 2},\cdots \Big \}$$
on the interval $[-1,1]$ associated with a sequence of distinct poles
$(a\sb k)\sp \infty\sb {k=1}$ in $\bold R\sbs [-1,1]$. Chebyshev polynomials
for the rational trigonometric system are explicitly found. Chebyshev
polynomials of the first and second kinds for the algebraic rational
system are also studied, as well as orthogonal polynomials with
respect to the weight function $(1-x\sp 2)\sp {-1/2}$. Notice that in these
situations, the `polynomials' are in fact rational functions. Several
explicit expressions for these polynomials are obtained. For the span
of these rational systems, an exact Bernstein-Szego type inequality
is proved whose limiting case gives back the classical
Bernstein-Szego inequality for trigonometric and algebraic
polynomials. It gives, for example, the sharp Bernstein-type
inequality
$$\vert p'(x)\vert \leq \frac {1}{\sqrt {1-x\sp 2}}\sum\sp n\sb {k=1}\frac
{\sqrt
{a\sp 2\sb k-1}}{\vert a\sb k-x\vert } \max\sb {y\in [-1,1]}\vert p(y)\vert
\quad \text {for} x\in
[-1,1],$$
where $p$ is any real rational function of type $(n,n)$ with poles
$a\sb k\in \bold R\sbs [-1,1]$. An asymptotically sharp Markov-type
inequality is also established, which is at most a factor of
$2n/(2n-1)$ away from the best possible result. With proper
interpretation of $\sqrt {a\sp 2\sb k-1}$, most of the results are
established for $(a\sb k)\sp \infty\sb {k=1}$ in $\bold C\sbs [-1,1]$ in a more
general setting."
Reviewed by Dinh Dung
94h:42041
42C05 (33C45)
Zhang, J.(1-OHS)
Relative growth of linear iterations and orthogonal polynomials on several intervals.
(English. English summary)
Linear Algebra Appl. 186 (1993), 97--115.
Summary: "Assume $A\in {C}\sp {2\times2}$ and $y\sb 0\in {C}\sp 2$.
Let $\{y\sb n=y\sb n(A,y\sb 0)\}$ be defined by the linear iteration
$y\sb n=Ay\sb {n-1}$, $n=1,2\cdots$. Then for every $p>0$ and for every norm
$\Vert ·\Vert $ in ${C}\sp 2$, there is a constant $C$, depending only
on
$p$ and $\Vert ·\Vert $, such that
$$\sup\sb {\rho(A)\leq 1,y\sb 0\not=0}{\Vert y\sb L(A,y\sb 0)\Vert \sp p\over
\sum\sb {k=0}\sp L\Vert y\sb k(A,y\sb 0)\Vert \sp p}\leq\frac CL,\quad
L=1,2,\cdots,$$
where $\rho(A)$ denotes the spectral radius of $A$. Applications to
orthogonal polynomials $\{p\sb n\}$ with respect to a positive Borel
measure $\alpha$ on several intervals are studied. In particular, the
growth of orthogonal polynomials relative to their sums satisfies
$$\lim\sb {n\to\infty}\sup\sb {x\in{\rm supp}\,(\alpha)}{\vert p\sb n(x)\vert
\sp p\over
\sum\sp {n-1}\sb {k=0}\vert p\sb k(x)\vert \sp p}=0,$$
provided that the three-term recurrence coefficients of $\{p\sb n\}$ are
asymptotically periodic."
Reviewed by G. Gasper
Cited in reviews: 96a:42032
94f:42026
42C05 (39A10 41A17)
Borwein, Peter(3-SFR); Erdélyi, Tamás(3-SFR); Zhang, J.(1-STF-C)
Müntz systems and orthogonal Müntz-Legendre polynomials.
(English. English summary)
Trans. Amer. Math. Soc. 342 (1994), no. 2, 523--542.
For distinct, complex $\lambda\sb i$ for which Re $\lambda\sb i >1/2$, let
$M\sb n (\Lambda)$ denote the span of $\{x\sp {\lambda\sb 0}, x\sp {\lambda\sb
1},\cdots, x\sp {\lambda\sb n}\}$. The authors define certain elements of
$M\sb n(\Lambda)$, Muntz-Legendre polynomials, on $0<x\leq 1$, via a
simple contour integral, show that they are, in fact, Laguerre
polynomials evaluated at $(-\log x)$, and develop both differential
and integral difference formulae. A reproducing kernel, similar to
that appearing for "ordinary" polynomials, is given for elements in
$M\sb n(\Lambda)$. One principal result, $L\sp 2$ Markov inequalities, is
given by Theorem 3.4: an upper bound (in terms of $\vert \lambda\sb i\vert $
and Re$(\lambda\sb i)$) is found for the supremum, over all $p$ in $M\sb n
(\Lambda)$, of $\Vert xp'(x)\Vert /\Vert p\Vert $. Sections 4 and 5 focus on
real
$\lambda\sb i$ and provide some results about bounding or interlacing of
zeros of Muntz polynomials in $M\sb n(\Lambda)$ vis-a-vis zeros of
those in $M\sb n(\Delta)$ when, e.g., $\lambda\sb i \leq \delta\sb i$ for all
$i$. Christoffel functions permit development of four interesting
equivalent statements, for the $\lambda\sb i >0$ situation, summarized in
Theorem 5.1---one pair expresses the fact that $M\sb n(\Lambda)$ is not
dense in $C[0,1]$ in the uniform norm iff $\sum\sp \infty\sb {k=1}
1/\lambda\sb k <\infty$.
Reviewed by A. G. Law
94d:42028
42C05 (33C45)
Nevai, Paul(1-OHS); Zhang, J.(1-OHS)
Rate of relative growth of orthogonal polynomials.
(English. English summary)
J. Math. Anal. Appl. 175 (1993), no. 1, 10--24.
Let a sequence of polynomials $\{p\sb n\}$ be defined by the recurrence
formula (1) $xp\sb n(x)=a\sb {n+1}p\sb {n+1}(x)+b\sb np\sb n(x)+a\sb np\sb
{n-1}(x)$, where
$p\sb {-1}=0$, $p\sb 0=1$, $a\sb n\not=0$ and $\{a\sb n\}$ and $\{b\sb n\}$ are
otherwise arbitrary complex sequences. The authors investigate the
rate of growth (with respect to $n$) of the expression (2)
$\vert p\sb n(x)\vert \sp p/\sum\sp n\sb {k=0}\vert p\sb k(x)\vert \sp p$. For
complex numbers $a$ and $b$,
the interval $[b-a,b+a]$ is defined to be the segment
$\{b+ta\colon\;-1\leq t\leq 1\}$ and the class ${\rm CM}(b,a)$ is the set
of all pairs of complex sequences $\{a\sb n\}\sp \infty\sb {n=0}$,
$\{b\sb n\}\sp \infty\sb {n=0}$ which converge to $a/2$, $b/2$, respectively
(with $a\sb n\not=0$). For $\{a\sb n,b\sb n\}\sp \infty\sb {n=0}\in {\rm
CM}(b,a)$, define
$\epsilon\sb n=\vert 2a\sb n-a\vert +2\vert b\sb n-b\vert +\vert 2a\sb
{n+1}-a\vert $. The main result of the
authors is a theorem which, for every positive integer $n$ and every
$L\leq \sqrt n$, places an upper bound on the maximum of the expression
in (2), the maximum being computed for $x\in [b-a,b+a]$. This maximum
is a somewhat complicated expression involving $p$, $n$, $L$, and
$\epsilon\sb k$. It was shown previously that the expression in (2)
converges to 0 uniformly for $x\in [b-a,b+a]$.
When the $b\sb n$ are real and $a\sb n>0$, the polynomials become real
orthogonal polynomials and the authors specialize to the case
$\epsilon\sb n=O(1/n\sp \delta)$ to obtain bounds which apply to the Jacobi,
sieved ultraspherical, and Pollaczek polynomials.
Reviewed by T. S. Chihara
93a:42010
42C05
Erdélyi, Tamás(1-OHS); Nevai, Paul(1-OHS); Zhang, J.(1-OHS); Geronimo, Jeffrey S.(1-GAIT)
A simple proof of "Favard's theorem" on the unit circle.
Atti Sem. Mat. Fis. Univ. Modena 39 (1991), no. 2, 551--556.
Favard's theorem on the unit circle states that a sequence of monic polynomials
satisfying $\Phi\sb n(z)=z\Phi\sb {n-1}(z)+a\sb n\Phi\sb {n-1}\sp *(z)$ is an
orthogonal
system on the unit circle with respect to some positive Borel measure if and
only if $\vert a\sb n\vert <1$. The result is well known but not in the
standard
books on orthogonal polynomials. Some proofs exist in the literature (see the
references), but the present proof is short and constructive and in fact
a nice exercise for graduate students interested in orthogonal polynomials.
Reviewed by Walter Van Assche
92m:42025
42C05
Erdélyi, Tamás(1-OHS); Nevai, Paul(1-OHS); Zhang, J.(1-OHS); Geronimo, Jeffrey S.(1-GAIT)
A simple proof of "Favard's theorem" on the unit circle.
Trends in functional analysis and approximation theory (Acquafredda di Maratea, 1989),
41--46,
Univ. Modena, Modena, 1991.
A short constructive proof of the following "Favard theorem" on the
unit circle is given: Let $\{\epsilon\sb n\colon\;n=1,2,\cdots\}$ be a
sequence of complex numbers with $\vert \epsilon\sb n\vert <1$ for all $n$, and
let
the sequence of polynomials $\{\Phi\sb n\colon\;n=0,1,2,\cdots\}$
satisfy the Szego recursion
$\Phi\sb n(z)=z\Phi\sb {n-1}(z)+\epsilon\sb n\Phi\sp *\sb {n-1}(z)$, $\Phi\sb
0(z)=1$,
where $\Phi\sp *\sb {n-1}(z)=z\sp {n-1}\overline{\Phi\sb {n-1}(1/\overline
z)}$.
Then there exists a unique finite positive Borel measure $µ$ on the
unit circle $T$ with infinite support such that $\{\Phi\sb n\colon\;
n=0,1,2,\cdots\}$ is orthogonal with respect to $µ$. The orthonormal
polynomials $\phi\sb n$ are given by
$$\phi\sb n(z)=\Phi\sb n(z)\Big[\prod\sp n\sb {k=1}(1-\vert \epsilon\sb k\vert
\sp 2)\sp {-1/2}\Big].$$
The construction of $µ$ is as follows: The distribution function
$µ\sb n$ is defined by
$µ\sb n(\tau)=\frac1{2\pi}\int\sp \tau\sb 0\vert \phi\sb n(e\sp
{i\theta})\vert \sp {-2}\,d\theta$.
It follows by properties obtained from the recurrence relation that the
system $\{\phi\sb n\colon\;0,1,\cdots,n\}$ is orthonormal with respect
to $µ\sb n$. By Helly's selection and convergence theorems there eixsts
a subsequence $\{n\sb k\}$ such that $µ\sb {n\sb k}(\theta)$ tends to a
distribution function $µ(\theta)$ for which $\{\phi\sb n\colon\;
n=0,1,2,\cdots\}$ is an orthonormal system. The uniqueness follows by
the unique representation of bounded linear functionals on $C(T)$.
The authors point out that, unknown to them and to many other workers
in the field, similar methods were used by P. Delsarte, Y. V. Genin
and Y. G. Kamp [IEEE Trans. Circuits and Systems CAS-25
(1978), no. 3, 149--160; MR 58 #1981].
\{For the entire collection see MR 92f:00028\}.
Reviewed by Olav Njastad
92k:42034
42C05
Nevai, Paul(1-OHS); Zhang, J.(1-OHS); Totik, Vilmos(H-SZEG-B)
Orthogonal polynomials: their growth relative to their sums.
J. Approx. Theory 67 (1991), no. 2, 215--234.
The uniform convergence
$ \lim\sb {n \to \infty} { \vert p\sb n(x)\vert \sp p /\sum\sb {k=0}\sp {n-1}
\vert p\sb k(x)\vert \sp p }
= 0$,
for $x \in [b-a,b+a]$ and $0 < p< \infty$, is considered for solutions of a
three-term recurrence relation
$xp\sb n(x) = a\sb {n+1}p\sb {n+1}(x) + b\sb np\sb n(x) + a\sb np\sb {n-1}(x)$,
with the property that $a\sb n$ $(\neq 0)$, $b\sb n \in {\bold C}$ with $a\sb n
\to a/2$
and $b\sb n \to b$. Almost all initial conditions $p\sb 0$ and $p\sb {-1}$
are allowed. The special case $p\sb {-1}=0, p\sb 0=1$ gives orthogonal
polynomials whenever $a\sb n>0$ and $b \in
{\bold R}$. This paper thus treats three
generalizations of an important result in the theory of orthogonal polynomials:
the relative growth is considered for every $p \in (0,\infty)$ and not for
$p=2$ alone; complex coefficients $a\sb n \neq 0$ and $b\sb n$ are allowed and
a great variety of initial conditions $p\sb {-1},p\sb 0$ lead to the same
uniform
convergence.
Reviewed by Walter Van Assche
Cited in reviews: 96a:42032 94b:39005