home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!pipex!warwick!uknet!qmw-dcs!arodgers
- From: arodgers@dcs.qmw.ac.uk (Angus H Rodgers)
- Newsgroups: sci.math
- Subject: Re: Closed forms for sums
- Message-ID: <1993Jan9.163732.8881@dcs.qmw.ac.uk>
- Date: 9 Jan 93 16:37:32 GMT
- References: <964.165.uupcb@hal9k.ann-arbor.mi.us>
- Sender: usenet@dcs.qmw.ac.uk (Usenet News System)
- Organization: Computer Science Dept, QMW, University of London
- Lines: 66
- Nntp-Posting-Host: theoryc.dcs.qmw.ac.uk
-
- In <964.165.uupcb@hal9k.ann-arbor.mi.us>
- james.jones@hal9k.ann-arbor.mi.us (James Jones) writes:
-
- >Some things I remember:
-
- >1. All closed forms had n and (n + 1) as a factors.
-
- > 2 2
- >2. All odd j > 1 had n and (n + 1) as a factors.
-
-
- >3. Add even j > 2 had interesting coefficients on the terms in
- > the last factor.
- > (at least the last factor the way Derive factored it)
-
- > [...] my question is, how does one go about determining a closed
- > form for a sum? [...]
-
- >Any explanation or references (explanation preferred) would be
- >appreciated.
-
- Perhaps this should be an FAQ?
-
- I got interested in the problem when I was at school in the 60s,
- and wrote it up then. Unfortunately I later destroyed my notes,
- but I still kept copies of the general formulae for these
- factorisations, in terms of the Bernoulli numbers; and when the
- topic came up in sci.math recently, I reconstructed some of the
- proofs, and additionally reassured myself that the simplicity of
- the coefficients was at least partly a consequence of von Staudt's
- theorem on the Bernoulli numbers (see Hardy & Wright).
-
- I've got a LaTeX file with most of this stuff in it, but it's a
- bit long to post to sci.math, and it still needs some work.
-
- I've also got some pari-gp code which you can use to calculate
- enormous coefficients (expressed as products of powers of primes)
- to your heart's content.
-
- Briefly, you need to prove, first, that the power sum, satisfying
- a 1st-order difference equation, is a polynomial in n, and second,
- that it is also an even or odd polynomial in n + (1/2). You can then
- get a pair of recurrence relations for the coefficients in the
- factorisation -- which is more neatly expressed in terms of n(n + 1)
- than in terms of n itself. (Something that Derive didn't notice.)
-
- (I can't face trying to work through the messy details of this part
- of the work again; and in any case there may be some more elegant
- approach than the pedestrian route which I took, using only school
- algebra.)
-
- The recurrence relations can be solved, in two ways. One way seems
- useful for estimating the numerical size of the coefficients (which
- can be proved to alternate in sign, a fact which you didn't mention),
- but I haven't followed this up, and it may not be easy. The other
- solution is in terms of the Bernoulli numbers, and is useful for
- getting at the more number-theoretic properties of the coefficients.
-
- I'll post the document to you when I've edited it a bit. I suppose
- it's not absolutely too long to post to sci.math, if there's enough
- interest.
- --
- Gus Rodgers, Dept. of Computer Science, | Cinema's worst romantic dialogue? --
- Queen Mary & Westfield College, Mile End | "Mechagodzilla's brain is installed
- Road, London, England. +44 71 975 5241 | in my stomach!" "I don't care if you
- E-mail (JANET): arodgers@dcs.qmw.ac.uk | are a cyborg -- I still love you!"
-