[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference rusure::math

Title:Mathematics at DEC
Moderator:RUSURE::EDP
Created:Mon Feb 03 1986
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:2083
Total number of notes:14613

219.0. "Got a headache from this one" by TAV02::NITSAN () Mon Feb 04 1985 02:12

       / unknown non-negative integers  for n<0
       |
       |
 U  =  | 1                              for n=0
  n    |
       |   m
       | SIGMA B U                      for n>0
       \  k=1   k n-k


 B  , B  , ... , B   are positive integers.
  1    2          m

 Prove that  U  / U    <=  1 + max B   for every n>1
              n    n-1              k

 (Trivial examples: Geometric sequences, Fibonachi sequence, ...)
T.RTitleUserPersonal
Name
DateLines
219.1TAV02::NITSANMon Feb 11 1985 05:021
Woops: Unknown non-negative FIXED integer.
219.2TRACE::GILBERTOwnership ObligatesThu Feb 15 1990 10:171
    I assume the summation should be from 1 to n (not m).
219.3AITG::DERAMODan D&#039;Eramo, nice personThu Feb 15 1990 14:334
	I think m is correct; for example, for Fibonacci numbers
	m = 2.

	Dan
219.4KOBAL::GILBERTOwnership ObligatesWed Feb 21 1990 12:0179
Well, I'm been making some (very) meager progress on this for m=2.
The case for m=1 is trivial, since in that case:

	       n
	U  = B  , and so  U  / U    = B  <=  1 + max B
	 n    1            n    n-1    1              k

When m=2, we have:
	          n       n
	U  =  c a   + c a  ,  where
	 n     0 0     1 1

	c  = (U  (D-B ) + 2) / (2 D),	c  = (U  (D+B ) - 2) / (2 D),
	 0     -1    0			 1     -1    0

	a  = (D+B ) / 2,		a  = (B -D) / 2,  and
	 0       0			 1     0
	            2
	D = sqrt( B   + 4 B  ).		(Note that D > B )
	           0       1				0

Thus, we see that the ratio U  / U    takes the form:
                             n    n-1

	  K U   + K
	   a -1    b
	--------------,  where the K are independent of U  .
	  K U   + K					 -1
	   c -1    d

The derivative of this ratio with respect to U   is
					      -1
	( K K  - K K  ) / (K U   + K )^2,
	   a d    b c       c -1    d

and this expands/simplifies to 

	                                   n       n
	                  - 16 B  D (B - D)  (D+B )
	                        0     0          0
	-------------------------------------------------------------------
	                    n               n             n         n   2
	{ 2 U  [(D-B )(D+B )  + (D+B )(B -D) ] + 4 [(D+B )  + (B -D) ] }
	     -1     0     0         0   0               0       0

This is positive if n is odd, and negative if n is even.  Thus, to maximize

the ratio U  / U    by choosing U  , we should let U   be a very large number
           n    n-1              -1                 -1
for odd n, and a very small number for even n (recall that U  is constrained
							    -1
to be an integer >= 0).


If n is odd and we choose a large U   to maximize U  / U   , then the ratio
				   -1		   n    n-1
becomes simply K  / K , or:
		a    c

	                         n+1               n+1
	  U          (D-B )(D+B )    + (D+B )(B -D)
	   n             0     0           0   0
	------  =  -------------------------------------
	 U                          n               n
	  n-1       2 [ (D-B )(D+B )  + (D+B )(B -D)  ]
	                    0     0         0   0

	                                            n
	                              D (D+B )(B -D)
	                                    0   0
	        =  (D+B )/2  -  -------------------------------
	               0                     n               n
	                         (D-B )(D+B )  + (D+B )(B -D)
	                             0     0         0   0

Which is somewhat larger than (D+B )/2.
				  0

(That's it so far)
219.5AITG::DERAMODan D&#039;Eramo, nice personWed Feb 21 1990 15:4123
        An example for m=2 is the Fibonacci sequence:
        
        Let U   = 0 (restricted to non-negative integers in general)
             -1
        
        and U  = 1 and U  = U    + U    for n > 0 
             0          n    n-1    n-2
        
        i.e., B  = B  = 1 (in general they are positive integers)
               1    2
        
        Then the claim in .0 is that for all n>1, U  / U    <= 2.
                                                   n    n-1
        
        And for the Fibonacci sequence, this is true--successive
        terms don't more than double.
        
        For the general case I would expect that a proof by
        induction would be possible and that it wouldn't be
        necessary to actually solve for U  in closed form.
                                         n
        
        Dan