|
A
MATHEMATICAL CONJURING TRICK
(THEOREM ON RECURRENCE SEQUENCES)
THE STORY SO FAR
Suppose we have a polynomial
f(x) = xd +
a1xd-1 +
a2xd-2 +...+
ad,
(1)
whose roots are R1,
R2,
...Rd.
We can create from these roots a sequence (sn),
each of whose terms sn is
the sum of their nth powers (integer n). Thus
s1 =
R1 +
R2 +...+
Rd
s2 =
R12 +
R22 +...+
Rd2
...
sn =
R1n +
R2n +...+
Rdn
or in short
d
sn =
Σ Rkn
k=1
This is an integer for n
≥ 0,
and for all n
if ad =
±1.
The terms of any
(sn)
form a recurrence sequence
sn =
-a1sn-1 -
a2sn-2 -...-
adsn-d.
(2)
(A recurrence sequence is a sequence of numbers in which each term is
obtained from its predecessors according to some specified rule. The best
known is the Fibonacci sequence 1, 1, 2, 3, 5, 8, ... in which each term is
the sum of the previous two.)
The first
d
terms can be calculated, without knowing the values of
Rk,
by Newton's formula
n-1
sn =
Σ -aisn-i -
nan,
(3)
i=1
and the sequence can then be continued indefinitely from
(2).
[Note that
(2)
and
(3)
are equivalent when n > d
since then an =
0.]
Example 1
For f(x) = x2 -
x - 1
we have from (3)
a1 =
-1
a2 =
-1
s1 =
-a1 =
1
s2 =
-a1s1 -
2a2 =
1 + 2 = 3
after which from (2)
s3 =
s2 +
s1 = 4
sn =
-a1sn-1 -
a2sn-2 =
sn-1 +
sn-2
giving the Lucas sequence
1, 3, 4, 7, 11, 18, 29,... (4)
well known for its similarity to the Fibonacci sequence whose rule it
shares.
Example 2
For f(x) = x3 -
x - 1,
we have from (3)
a1 =
0
a2 =
-1
a3 =
-1
s1 =
-a1 =
0
s2 =
-a1s1 -
2a2 =
0 + 2 = 2
s3 =
-a1s2 -
a2s1 -
3a3 =
0 - 0 + 3 = 3
after which from
(2)
s4 =
-a1s3 -
a2s2 -
a3s1 =
0 + 2 + 0 = 2
sn =
-a1sn-1 -
a2sn-2 -
a3sn-3 =
sn-2 +
sn-3
giving the sequence
0, 2, 3, 2, 5, 5, 7, 10, 12, 17, 22,... (5)
THE CONJURING TRICK
Step 1:
Choose a polynomial f of the form given in
(1):
f(x) = xd +
a1xd-1 +
a2xd-2 +...+
ad
Step 2:
Obtain from f
a recurrence sequence (sn)
as explained above.
Step 3:
Reverse the order of the powers in f
and call the result g:
g(x) = 1 + a1x
+ a2x2 +...+
adxd
(6)
Step 4:
Differentiate g
to get g':
g'(x) = a1 +
2a2x
+...+ dadxd-1
Step 5:
Define the rational function
G(x) = -g'(x)/g(x)
whose terms may be generated by polynomial long division.
Flourish:
It will be found that the series
s1 +
s2x
+ s3x2 +...+
snxn-1 +...
so generated by G(x)
has the same coefficients as the sequence (sn)
obtained from the original function f
in Step 2.
Example 1
As noted above, the function
f(x) = x2 -
x - 1
yields the Lucas sequence (4)
of sn terms.
The corresponding function
g(x) = 1 - x - x2
has the derivative
g'(x) = -1 - 2x
The rational function
| |
G(x) |
= |
-g'(x) |
= |
1 + 2x |
| |
|
|
g(x) |
|
1
- x -
x2 |
= 1 + 3x + 4x2 +
7x3 +
11x4 +
18x5 +
29x6 +...
whose coefficients are also the Lucas sequence
(4).
Example 2
Similarly, as noted above,
f(x) = x3 -
x - 1
yields the recurrence sequence (5)
of sn terms.
The corresponding function
g(x) = 1 - x2 -
x3
has the derivative
g'(x) = -2x - 3x2
The rational function
| |
G(x) |
= |
-g'(x) |
= |
2x + 3x2 |
| |
|
|
g(x) |
|
1
- x2 -
x3 |
= 0 + 2x + 3x2 +
2x3 +
5x4 +
5x5 +
7x6 +
10x7 +
12x8 +...
whose coefficients are also the sequence (5).
PROOF, or HOW IT IS DONE
Let g(x) = 1 + a1x
+ a2x2 +...+
adxd
(6
bis)
H(x) = s1 +
s2x
+ s3x2 +...+
snxn-1 +...
We have to prove that H(x) =
G(x).
Now g(x)H(x)
= (1 + a1x
+ a2x2 +...+
adxd)(s1 +
s2x
+ s3x2 +...+
snxn-1 +...)
= s1 +
(s2 +
a1s1)x
+ (s3 +
a1s2 +
a2s1)x2 +...
(7)
Consider the cases 1
≤ n
≤ d.
The general term of (7)
here is
(sn +
a1sn-1 +
a2sn-2 +...+
an-1s1)xn-1
However from (3)
above, for 1
≤ n
≤ d
sn +
a1sn-1 +
a2sn-2 +...+
an-1s1 =
-nan
Hence the general term here is -nanxn-1.
Now consider the cases n > d.
Here the general term of (7)
is
(sn +
a1sn-1 +
a2sn-2 +...+
adsn-d)xn-1
while from (2)
above
sn +
a1sn-1 +
a2sn-2 +...+
adsn-d =
0
Hence the general term here is 0.xn-1 =
0.
Hence g(x)H(x)
in (7)
reduces to a terminating series with d
terms of the form
-nan:
g(x)H(x) = -a1 -
2a2x
- 3a3x2 -...-
dadxd-1
But
g'(x) = a1 +
2a2x +
3a3x2 +...+
dadxd-1
Hence
g(x)H(x) = -g'(x)
H(x) = -g'(x)/g(x) = G(x)
as required.
COMMENT
The rational function g'(x)/g(x)
is of particular mathematical interest since
óg'(x)
dx = ln(g(x))
õ g(x)
Its inverse also features in Newton-Raphson iteration.
Martin Mosse,
August 2011.

ACKNOWLEDGEMENTS
The stimulus for this paper came from Reference 1, of which a copy was
kindly sent to me by Professor George Spencer-Brown in June 1999.
I wish to acknowledge the help of Dr Watcyn Wynn in providing a rather
better proof than I had found myself.
I believe that the theorem, which I found in November 2009, is original and
would welcome confirmation of this, or correction if not. Email
martin@brainwaves.org.uk .
REFERENCE
1. 'The Identification of Prime Numbers Using Algebraic Numbers of Degree >
2', by J.C.P. Miller, G. Spencer-Brown and D.J. Spencer-Brown; unpublished
but distributed by G. Spencer-Brown in 1965. |