home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math
- Path: sparky!uunet!usc!zaphod.mps.ohio-state.edu!darwin.sura.net!spool.mu.edu!sgiblab!munnari.oz.au!bunyip.cc.uq.oz.au!marlin.jcu.edu.au!coral.cs.jcu.edu.au!spuler
- From: spuler@coral.cs.jcu.edu.au (David Spuler)
- Subject: Recurrence relations
- Message-ID: <spuler.722054177@coral>
- Sender: news@marlin.jcu.edu.au (USENET News System)
- Organization: James Cook University
- Date: 18 Nov 92 02:36:17 GMT
- Lines: 104
-
-
- Can anyone help me solve the following two recurrence relations?
- They are both derived from optimal binary trees and must be minimized.
-
- I'm a PhD student in Computer Science, at James Cook University, Australia,
- and any help on these issues would improve my thesis.
-
-
- RECURRENCE A:
- c[0] = 0
- Generally, c[n] is minimum of 3 cases:
- c[n] = c[n-1] + alpha * n ; Case 1
- c[n] = c[i] + c[n-i] + alpha * n (with 1<=i<n) ; Case 2
- c[n] = c[i - 1] + c[n-i] + beta * n (with 1<=i<n) ; Case 3
-
-
- RECURRENCE B:
- c[0] = 0
- Generally, c[n] is minimum of 2 cases:
- c[n] = c[n-1] + alpha * n ; Case 1
- c[n] = c[i] + c[n-i] + beta * n (with 1<=i<n) ; Case 2
-
-
- Both of these recurrences can be simplified to change the two parameters alpha
- and beta to use only one parameter:
-
- R = beta/alpha
-
- Dividing all terms gives the new equivalent recurrences:
-
- RECURRENCE A2:
- c[0] = 0
- Generally, c[n] is minimum of 3 cases:
- c[n] = c[n-1] + n ; Case 1
- c[n] = c[i] + c[n-i] + n (with 1<=i<n) ; Case 2
- c[n] = c[i - 1] + c[n-i] + R * n (with 1<=i<n) ; Case 3
-
-
- RECURRENCE B2:
- c[0] = 0
- Generally, c[n] is minimum of 2 cases:
- c[n] = c[n-1] + n ; Case 1
- c[n] = c[i] + c[n-i] + R * n (with 1<=i<n) ; Case 2
-
-
- CONJECTURE 1: cut-offs for Case 1
- --------------------------------------
-
- I have empirical evidence to support the conjecture that Case 1 is only
- chosen for small n, and for some cut-off point k, we have n>=k implying
- that only Case 2 (or 3 for recurrence A) is applicable.
- Can anyone help me prove this?
-
- Is there an analytical formula for the cut-off?
- For recurrence B I have a theoretical approximate value of
-
- k ~=~ 4R
-
- which I derived by noting that if my conjecture is true, then
-
- n <= k ====> c[n] = 1/R * n(n+1)/2
-
-
- CONJECTURE 2: Choice of i for case 2/3
- --------------------------------------
-
- I also have the conjecture backed up by empirical evidence that if Case 3
- is chosen in Recurrence A, or Case 2 in Recurrence B, then the best choice of
- i is ceil(n/2) (equivalently floor, doesn't matter which).
- However, Case 2 in Recurrence A seems not to always be i==n/2, although it
- is always reasonably close to that value.
- Can anyone help me prove this..
-
- ---------------------------------------------------------
-
- Perhaps this will help ......
- These recurrences are all generalizations of a simpler recurrence about which
- I posted a query a while ago. That recurrence was:
-
- c[0] = 0
- c[n] = c[n - 1] + n Case 1
- c[n] = c[i] + c[n-i] + n Case 2
-
- Thanks to a few great responses the exact solution was derived as:
-
- c[n] = a(b+4) + (3b+6) 2^b, where n is parameterized as: n= a + 3*2^b
- equivalently: c[n] = 2a + (b+2)n
-
- This also helped prove similar conjectures to above, i.e that for n>=5 the
- best choice was always case 2 (i.e. the cut-off was k=4), and also that the
- best choice of i for case 2 was i=floor(n/2).
- This recurrence is the same as Recurrence B with R=1, or Recurrence A without
- Case 3.
-
- ---------------------------------------------------------
-
- More details are available on request. I look forward to any useful responses
- with hints, tips, or even pointers to papers/books which will help me solve
- such minimization recurrences (e.g. I've looked at Greene/Knuth's book
- "Mathematics for the Analysis of Algorithms" but my recurrences are more
- difficult than the one's they examine).
-
- Thanks in advance,
- David Spuler
-