Triangle
Gottfried Helms
helms at uni-kassel.de
Thu Apr 17 15:10:47 CEST 2008
Dear seqfans,
a short review and resume.
Assume a matrix-function rowshift(M) which
computes M1 = rowshift(M) in the following way:
M = [a,b,c,...]
[k,l,m,...]
[r,s,t,...]
[... ]
M1 = [a,b,c, ... ]
[0,k,l,m, ... ]
[0,0,r,s,t,...]
[ ... ]
Assume the lower-trianguler matrix of Stirling-numbers 2'nd kind
S = [1 0 0 0 ...]
[1 1 0 0 ...]
[1 3 1 0 ...]
[1 7 6 1 ...]
[ ... ]
then with
H0 = [1]
[1]
[1]
[1]
...
we have the iterative Mathar-products
H1 = S * rowshift(H0) \\ = S
H2 = S * rowshift(H1)
H3 = S * rowshift(H2)
...
H1 =
1 . . . .
1 1 . . .
1 3 1 . .
1 7 6 1 .
1 15 25 10 1
H2=
1 . . . . . . . .
1 1 1 . . . . . .
1 3 4 3 1 . . . .
1 7 13 19 13 6 1 . .
1 15 40 85 96 75 35 10 1
H3=
1 . . . . . . . . . . . .
1 1 1 1 . . . . . . . . .
1 3 4 6 4 3 1 . . . . . .
1 7 13 26 31 31 25 13 6 1 . . .
1 15 40 100 171 220 255 215 156 85 35 10 1
(construction based on a interpretation of
the Maple-implementation of R.Mathar)
----------------------------------
In my basic problem-description I also had the vector D,
which I also should index now.
It contains the coefficients of the polynomials in u:
Dm = polcoeffs(prod(k=1,m-1,u^k-1))
Say
D3 = columnvector([1 -1 -1 1])
Then define the matrix MD3 as the concatenation of shifted D3
MD3 =
1
-1 1
-1 -1 1
1 -1 -1
1 -1 ...
1
...
up to the required dimension for the following matrix-multiplication
to obtain the k'th coefficient in the original powerseries
Ut(x,h) = a_1(t,h)/1!*x + a_2(t,h)/2!*x^2 + ...
Then the coefficients for the bivariate polynomial in u=log(t) and
v=u^h of the k'th coefficient a_k(t,h) occur in the vector Vk
Vk = MDk * transpose(Hj[k,])
where [k,] denotes the k'th row and the index j at Hj indicates the
1/2*k*(k^2-4k+5) 'th iterate/row-shift
------------------------------------
This is an eminent simple recursive scheme to obtain the coefficients
for the integer (and fractional!) iteration of x -> t^x - 1.
However it is again enormous consumtive: the required iterations of
the Mathar-products is cubic with the index!
So for index k=20 I need already 8000 iterations with always growing
matrices... surely some shortcuts may be implemented, but this may
well be the same amount of time and memory as I would need for
the symbolic eigensystem-analysis. Hmmm....
However - the fact, that this simple scheme is able to *mimic* the
symbolic eigendecomposition of a matrix-operator is a very astonishing
aspect.
Gottfried Helms
More information about the SeqFan
mailing list