mm-303
Subject: Re: Source code for fft in C> I've been using
fftn for several weeks and find it pretty satisfactory> for 2D
complex FFTs upto 1024x1024. I've no idea how it compares with>
other implentations (it's lis as dsp79-singleton on the FFTW>
comparison page but doesn't appear on the comparison charts).
The code> looks tight anyway.There are two (very similar)
versions that we benchmark, lis as> singleton and
dsp79-singleton. It is in the charts (fftw.org/speed),> but
you have to scroll down to the single-precision graphs (that's
how> it is supplied in Fortran). We don't currently benchmark
any C> translation of it, but usually the performance of such
a thing is> roughly similar to the original.> Ah, I found the
comparison now for 1d and 3d - not 2d though which iswhat I
looked for before. It's pretty slow in both cases and I
doubt2d is different. Maybe I'll bite the bullet and try fftw
again.Thanks.
===
Subject: Re: Help: nonlinear optimization in
scheme language ??X-ID:
rqhRWuZvrehXY3p0joM8bGHSa2sdBnsAptp-F3lqDgjloAs5Jzye4B>Does
anybody know a source for scheme implementations of nonlinear
>optimization algorithms such as levenberg-marquardt
??>>Thanks>what 's that? > maybe >
http://plato.la.asu.edu/guide.html > helps?> peterThanks very
much, but the algorithms listet there are mostly Fortran,
C/C++, MatLab ... Sources for algorithms implemen in scheme
are very rare.Norbert
===
Subject: Re: Help: nonlinear
optimization in scheme language ??as a declarative language
SCHEME is not exactly ideal foroptimization. You may find a
pure constraint solver in it more likelythan a LS solver.
Aren't there more suitable newsgroups you could askin?Hans
Mittelmann
===
=================================================

===
=========================>Does anybody know a source for
scheme implementations of nonlinear >optimization algorithms
such as levenberg-marquardt ??>>Thanks>what 's that? > maybe >
http://plato.la.asu.edu/guide.html > helps?> peter> Thanks very
much, but the algorithms listet there are mostly Fortran, >
C/C++, MatLab ... Sources for algorithms implemen in scheme
are very > rare.> Norbert
===
Subject: Numerical quadratures for
irregular gridsI am looking for the algorithms of numerically
evaluatingintegrals of functions of one (x) and two (x,y)
variables, at theassumption that the function values are
available exclusively as acollection of discrete values, and
that the grids of x or (x,y) pointsat which the function
values are given are generally irregular(please note that
standard textbooks discuss almost exclusively the caseof
regular grids, with constant grid spacings).I am interes not
only in the theory of various approximationsto the integrals
in such cases, but also in practical algorithmsof navigating
through the sets of discrete function data.Any pointers to the
literature will be apprecia.Thanks in
advance.L.B.*-------------------------------------------------
------------------*| Dr. Leslaw Bieniasz, || Institute of
Physical Chemistry of the Polish Academy of Sciences,|| Molten
Salts Department, ul. Zagrody 13, 30-318 Cracow, Poland. ||
tel./fax: +48 (12) 266-03-41 || E-mail: nbbienia@cyf-kr.edu.pl
|*-------------------------------------------------------------
------*| Interes in Computational Electrochemistry? || Visit
my web site: http://www.cyf-kr.edu.pl/~nbbienia
|*------------------------------------------------------------
-------*
===
Subject: Re: Numerical quadratures for irregular
grids> I am looking for the algorithms of numerically
evaluating> integrals of functions of one (x) and two (x,y)
variables, at the> assumption that the function values are
available exclusively as a> collection of discrete values, and
that the grids of x or (x,y) points> at which the function
values are given are generally irregular> (please note that
standard textbooks discuss almost exclusively the case> of
regular grids, with constant grid spacings).> I am interes not
only in the theory of various approximations> to the integrals
in such cases, but also in practical algorithms> of navigating
through the sets of discrete function data.> Any pointers to
the literature will be apprecia.> Thanks in advance.For one
dimension how about using piecewise Lagrange interpolation
andintegrating the resulting polynomials? Or you could use
cubic splinesthrough all the points and integrate the
cubics.For two dimensions either 2D curve fits or assign a
local effective area toeach point based on its distance from
surrounding points and do a sum ofvalue times area.> L.B.>
*-------------------------------------------------------------
------*> | Dr. Leslaw Bieniasz, |> | Institute of Physical
Chemistry of the Polish Academy of Sciences,|> | Molten Salts
Department, ul. Zagrody 13, 30-318 Cracow, Poland. |> |
tel./fax: +48 (12) 266-03-41 |> | E-mail:
nbbienia@cyf-kr.edu.pl |>
*-------------------------------------------------------------
------*> | Interes in Computational Electrochemistry? |> |
Visit my web site: http://www.cyf-kr.edu.pl/~nbbienia |>
*-------------------------------------------------------------
------*
===
Subject: Re: Numerical quadratures for irregular
grids> I am looking for the algorithms of numerically
evaluating> integrals of functions of one (x) and two (x,y)
variables, at the> assumption that the function values are
available exclusively as a> collection of discrete values, and
that the grids of x or (x,y) points> at which the function
values are given are generally irregular> (please note that
standard textbooks discuss almost exclusively the case> of
regular grids, with constant grid spacings).> I am interes not
only in the theory of various approximations> to the integrals
in such cases, but also in practical algorithms> of navigating
through the sets of discrete function data.What about using
scattered data interpolation followed by astandard integration
routine on the interpolant?Arnold Neumaier
===
Subject: floating
point numberswe are planning to develop a portable
mathematical librarywhich will rely heavily on the proper
implementations ofthe floating point standard (ieee754).does
anybody have any experience with poor implementations orother
troubles on different machines or systems ?in particular we
would like to know on which systemswe can trust the special
numbers - like NaN etc - and onwhich we could experiece
surprises ...
===
Subject: Chair in Statistics - VacancyVacancy
for a Chair in Statistics at Eindhoven University of
Technology,Eindhoven, The
Netherlands
===
================================================

===
===========================================================

===
==The Department of Mathematics and Computer Science has a
vacancy for afull professorship in StatisticsGeneral
Information:--------------------------------------------------
--------------------------------------------------------------
--------------------------------------------------------------
----------------------------The Department of Mathematics and
Computer Science provides undergraduate,MSc and PhD programs
in Industrial and Applied Mathematics and in
ComputerScience.The Department has research collaborations
with other Departments at theTechnical University Eindhoven,
as well as with a large number of otheruniversities and
companies,both at home and abroad.The Department has
approximately 300 employees and over 700 students. Thechair of
Statistics is one of the nine chairs in Mathematics, and one of
thefour chairs in the section Statistics and Operations
Research. The other two sections in Mathematicscover Analysis
and Discrete Mathematics. For the future, the Department
ofMathematics andComputer Science envisions important new
opportunities for research onbiological, biomedical,
industrial and engineering applications. Thestatistics group
is activelyinvolved with the activities at EURANDOM, the
European institute forresearch in stochastics.What are your
duties?-------------------------------------------------------
--------------------------------------------------------------
--------------------------------------------------------------
-----------------------------------------------You are expec
to stimulate and coordinate the fundamental and
appliedresearch of the group, to initiate new research
directions, to establishlinks with other researchprograms at
our university and to become actively involved with
EURANDOM.You give and coordinate courses in Statistics and
Probability, and you areresponsible for updating these
courses.You advise MSc and PhD students.You fulfill key
management functions in the section and the faculty.Your
skills
are:----------------------------------------------------------
--------------------------------------------------------------
--------------------------------------------------------------
------------------------------A deep and broad insight into
statistics, as is reflec in a stronginternational reputation
in the field and a large number of - also recent -publications
in the international literature.Much affinity with, and
knowledge of stochastics.Interest in application-orien
research, experience and expertise inacquisition and
consultancy.Excellent didactical qualities; leadership and
management qualities.What we have to
offer:--------------------------------------------------------
--------------------------------------------------------------
--------------------------------------------------------------
-----------------------------------------A prominent leading
position in a stimulating scientific environment, inwhich you
will work with a strong group in stochastics,
enthusiasticstudents,trainee design engineers, and PhD
students, as well as postdocs fromEURANDOM.A full-time
appointment in accordance with the Collective Labour
Agreementfor Dutch Universities.A maximum salary of euro
7.875,00 gross per month depending on yourexperience.An
extensive package of fringe
benefits.Inquiries:-------------------------------------------
--------------------------------------------------------------
--------------------------------------------------------------
----------------------------------------------------For more
information, please contact Prof.dr. W.T.F. den Hollander,
e-mail:denhollander@eurandom.tue.nl, tel. +31 40 2478100.For
information concerning job conditions, please contact Mr.
W.C.J.Verhoef, head human resources, e-mail:
w.c.j.verhoef@tue.nl, tel. +31 402472321.How to
respond:------------------------------------------------------
--------------------------------------------------------------
--------------------------------------------------------------
---------------------------------------Please submit a written
letter of application accompanied by a recentcurriculum vitae
to:Mrs. Drs. S. Udo,Managing Director of the Department of
Mathematics and Computer Science,Eindhoven University of
Technology,HG 6.22,P.O. Box 513,5600 MB Eindhoven,The
Netherlands,mentioning the number of the vacancy V32854 in
your letter and on theenvelope.
===
Subject: floating point
numberswe are planning to develop a portable mathematical
librarywhich will rely heavily on the proper implementations
ofthe floating point standard (ieee754).does anybody have any
experience with poor implementations orother troubles on
different machines or systems ?in particular we would like to
know on which systemswe can trust the special numbers - like
NaN etc - and onwhich we could experiece surprises
...
===
Subject: Good intro. book?Dear readers,I would like to
learn howto solve ODE's and PDE's using numericalanalysis. I
have the book A First Course in the Numerical Analysis of
Differential
Equationshttp://titles.cambridge.org/catalogue.asp?ISBN=
0521556554but the problems in the book are of the type
Prove... or Show that...where I prefure problems of the type
Solve....Can anyone recommand a introduction book that have
applications, examples,and perhaps solutions to the problems?I
am using Matlab if that makes a difference.Lots of
love,Janni
===
Subject: estimate tail decayHi allI have some
data f(n), n=1,2...,m ~ 80 (singular valuesof a mildly
ill-posed problem) which I expect to decayas n^alpha for small
negative alpha. I'd like to geta numerical estimate of alpha,
and have tried the obviousweigh least-squares fit of the
log-log transformeddata. But this seems to me unsatisfactory,
as it does not take into account the full force of the
assumption that the decay is a power law for large n. Could
any readers of the group point me towards a more sophistica
approach?Thanks in anticipation.-j-- J. J. Green, Department
of Applied Mathematics, Hicks Bd.,Hounsfield Rd., University
of Sheffield, Sheffield, UK. +44 (0114) 222 3742,
http://www.vindaloo.uklinux.net/jjg 
===
Subject: Diagonalising
large matricesHiI wonder if anyone can help me. I have to
diagonalise quite largematrices - 1000 x 1000 for example. I
am using the Matrix TemplateLibrary for C++ which uses LAPACK
algorithms to do the actual work. Iget results just fine but
would like to speed up my calculationssomewhat so can anyone
answer the following questions for me please?1) Is there a
faster way of doing bog standard diagonalisation thenthe
method I am currently using?2) I only actually need the lowest
10 eigenvales (and later I willneed the corresponding
eigenvectors) - Is there a way of only gettingthese - and
would such a method be faster.I apologise if these are stupid
questions. I am afraid that mycurrent numerics knowledge goes
about as far as knowing how to pass mymatrix to a black box
routine that spits out the required eigenvaluesand
eigenvectors.Thanking people in advanceMike
===
Subject: Re:
Diagonalising large matricesHiI wonder if anyone can help me.
I have to diagonalise quite large> matrices - 1000 x 1000 for
example. I am using the Matrix Template> Library for C++ which
uses LAPACK algorithms to do the actual work. I> get results
just fine but would like to speed up my calculations> somewhat
so can anyone answer the following questions for me please?1)
Is there a faster way of doing bog standard diagonalisation
then> the method I am currently using?2) I only actually need
the lowest 10 eigenvales (and later I will> need the
corresponding eigenvectors) - Is there a way of only getting>
these - and would such a method be faster.I apologise if these
are stupid questions. I am afraid that my> current numerics
knowledge goes about as far as knowing how to pass my> matrix
to a black box routine that spits out the required
eigenvalues> and eigenvectors.Thanking people in advance>
MikeTo get the first few eigenvalues/eigenvectors, you can use
the powermethod. It is iterative, but lots faster than schemes
that get all Nof the eigenvalues/vectors.Any decent numerical
analysis book will tell you how.--
^^^^^^^^^^^^^^^^^^http://galileo.phys.virginia.edu/~jvn/ God
is not willing to do everything and thereby take away our free
will and that share of glory that rightfully belongs to us. --
N. Machiavelli, The Prince.
===
Subject: Re: Diagonalising large
matrices> To get the first few eigenvalues/eigenvectors, you
can use the power> method. IThat's the basic idea. The Lanczos
method greatly improves on thestability of that idea by
generating an orthogonal basis, and theimplicitly restar
Arnoldi method (the one in Arpack) is animprovement on
that.V.-- email: lastname at cs utk eduhomepage: cs utk edu
tilde lastname
===
Subject: Re: Diagonalising large matrices >Hi
>I wonder if anyone can help me. I have to diagonalise quite
large >matrices - 1000 x 1000 for example. I am using the
Matrix Template >Library for C++ which uses LAPACK algorithms
to do the actual work. I >get results just fine but would like
to speed up my calculations >somewhat so can anyone answer the
following questions for me please? >1) Is there a faster way
of doing bog standard diagonalisation then >the method I am
currently using? >2) I only actually need the lowest 10
eigenvales (and later I will >need the corresponding
eigenvectors) - Is there a way of only getting >these - and
would such a method be faster. >I apologise if these are
stupid questions. I am afraid that my >current numerics
knowledge goes about as far as knowing how to pass my >matrix
to a black box routine that spits out the required eigenvalues
>and eigenvectors. >Thanking people in advance >Mikethe 10
smallest eigenvalues of a 1000 by 1000 matrix: this should be
not hard job. you said nothing about symmetry. but for both
cases, symmetric or notthere exist variants of the
simultaneous vector iteration (inverse iteration with a system
a vectors instead of one plus a device forkeeeping thme
linearly independent.) search for ritzit. also ready to use
software like ARPACK comes into
mind.http://www.caam.rice.edu/software/ARPACKhthpeter
===

Subject: Re: Diagonalising large matrices> 2) I only actually
need the lowest 10 eigenvales (and later I will> need the
corresponding eigenvectors) - Is there a way of only getting>
these - and would such a method be faster.Lanczos-type methods
are good for this, and they may be a lot faster.Check out
Arpack.V.-- email: lastname at cs utk eduhomepage: cs utk edu
tilde lastname
===
Subject: Re: Diagonalising large matrices>
HiI wonder if anyone can help me. I have to diagonalise quite
large> matrices - 1000 x 1000 for example. I am using the
Matrix Template> Library for C++ which uses LAPACK algorithms
to do the actual work. I> get results just fine but would like
to speed up my calculations> somewhat so can anyone answer the
following questions for me please?1) Is there a faster way of
doing bog standard diagonalisation then> the method I am
currently using?2) I only actually need the lowest 10
eigenvales (and later I will> need the corresponding
eigenvectors) - Is there a way of only getting> these - and
would such a method be faster.Subspace iteration works well
for this, at least if your matirx issymmetric. Look at the
book by Parlett.Arnold Neumaier
===
Subject: Source code for
finding first 3 eigenvectorI need code in C, C++ or Java for
finding the eigenvectorscorresponding to the three largest
eigenvalues of a given matrix.Could someone help
me?
===
Subject: Re: function interpolation by using
normal distributions>is there an algorithm which permits to
abtain a function interpolation>by using a set of normal
distributions?>Mdo you mean sum {i=1,...,n}
a_i*exp(-(x-m(i))^2/s(i)^2) ?with the m(i) and s(i) given this
amounts to a linear system of equationsand is a special case of
interpolation by radial functions. this is known also for
several independent variables and often used in scattered data
interpolation. be warned that the linear systems can be quite
illconditioned. if m(i) and s(i) are unknown, this becomes a
nasty nonlinear problem and then problems with n>=4 already
pose severe problems. even worse if you arein hihger
dimensions and have are more general model involving
covariancematrices.hthpeter
===
Subject: Re: Multivariate skew
normal distribution in C>> Dear All,>> I really wonder if
there is somwhere an inplementation of the>> simulation
algorithm (random numbers) of the multivariate skew normal>>
distribution. If yes, would you like to give me the source.>>
Thank you very much>> Przem>After a search of the web, I found
that Azzalini assigned the name skew>normal to the
density:>f(x) = 2.phi(x).PHI(alpha.x)>where phi(x) is the
probability density for the normal distribution, and>PHI() is
the cumulative normal distribution function. phi is the
std.>normal (mean = 0, std. devn. = 1), but x can easily be
scaled and shif to>give any other mean and std. devn.>I have
no experience with this distribution, but I guess that you
can>generate a random variable with this distribution simply
by generating>random normals (x), and then generating another
random normal (y) and>accepting the first one if a uniform
random variable (z) is less than>PHI(alpha.y). I have not
thought about how you extend it to the>multivariate case.It
might be just as easy to generate another normal
randomvariable and see if it is less than alpha.y. If this
isthe multivariate version, it will go over easily.There are
CHEAP methods of generating normal randomvariables, never
computing transcendental functions,or rarely computing them.--
This address is for information only. I do not claim that these
viewsare those of the Statistics Department or of Purdue
University.Herman Rubin, Department of Statistics, Purdue
Universityhrubin@stat.purdue.edu Phone: (765)494-6054 FAX:
(765)494-0558