home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math
- Path: sparky!uunet!usc!cs.utexas.edu!sun-barr!ames!pasteur!bellini.berkeley.edu!luzeaux
- From: luzeaux@bellini.berkeley.edu (Dominique Luzeaux)
- Subject: Re: Integer Sequence Problem
- Message-ID: <1992Jul28.225605.18476@pasteur.Berkeley.EDU>
- Sender: nntp@pasteur.Berkeley.EDU (NNTP Poster)
- Nntp-Posting-Host: bellini.berkeley.edu
- Reply-To: luzeaux@bellini.berkeley.edu (Dominique Luzeaux)
- Organization: University of California, Berkeley
- References: <1992Jul28.173452.24364@news.tu-graz.ac.at>
- Date: Tue, 28 Jul 1992 22:56:05 GMT
- Lines: 49
-
- In article <1992Jul28.173452.24364@news.tu-graz.ac.at>,
- hhassler@iaik.tu-graz.ac.at (Hannes Hassler) writes:
- |> For S an infinite, increasing sequence of positive integers (1=a_1 < a_2 <
- |> a_3 < ... < a_i < a_{i+1} < ...), we define the VALUE val(S) of S
- |>
- |> val(S) := sup_{n>1} \sum_{i=1}^n a_i / a_{n-1}
- |>
- |> (in words: For every n, sum up the first n terms of S and divide this sum
- |> by the (n-1)st term. Then val(S) is the supremum of all these values).
- |>
- |> PROBLEM: Find the minimum (infimum) of val(S) over all possible sequences S.
-
- Let us call : s_n=\frac{\sum_{i=1}^n}a_i}{a_{n-1}}
- and : q_n=\frac{a_n}{a_{n-1}}
-
- Then : s_n - s_{n-1} = q_n - (1-1/q_{n-1}) s_{n-1}
- Let us suppose val(S) is finite for some sequence S; in other words val(S)
- is an adherence value of (s_n); we can then consider that s_n converges to
- val(S) (it is always possible to find a subsequence converging to val(S) and
- we consider then this particular subsequence).
-
- If s_n converges to some limit L, this difference converges to 0; this
- implies that:
- q_n
- ------------- --> L
- 1 - 1/q_{n-1}
- which can be rewritten: q_n = (L + e_n)(1 - 1/q_{n-1}) where e_n is a small
- perturbation.
- Draw the function x \mapsto (L+e)(1-1/q_{n-1}) where e is small; after some
- iterations, q_n will be negative which is impossible by definition of S; this
- holds even if e is allowed to change for each n, but remains in a bounded small
- interval. The only way to avoid that situation is to have q_n = q_{n-1}
- starting
- from some n; let q be this constant value. As q has to be integer and
- greater or
- equal than 2, let us try q=2. This yields L=4.
-
- In fact, the sequence (1,2,...,2^n,...) satisfies all the conditions and yields
- val(S)=4
-
- Furthermore, we can say that val(S) takes as only finite values q^2/(q-1)
- where q is an integer >=2, hence 4 is the minimum of val(S) over all
- sequences S.
-
-
- Dominique Luzeaux
-
-
-
-