Uniqueness of Ansus' extended sum superfunction
#1
First let me recapitulate the method from my point of view.
We start with as usual with a function \( f \) and want to obtain a superfunction \( \sigma \) of it, i.e. a function that satisfies
(1) \( \sigma(z+1)=f(\sigma(z)) \).
Now by differentiating once we get
\( \sigma'(z+1)=\sigma'(z) f'(\sigma(z)) \)
expanding twice:
\( \sigma'(z+2)=\sigma'(z) f'(\sigma(z)) f'(\sigma(z+1)) \)
and generally by induction:
\( \sigma'(z+n)=\sigma'(z) \prod_{k=0}^{n-1} f'(\sigma(z+k)) \)
making \( z=n \) our new variable:
\( \frac{\sigma'(z_0+z)}{\sigma'(z_0)}= \prod_{k=0}^{z-1} f'(\sigma(z_0+k)) \)

To extend the product to non-integer \( z \) we have a look at the sum
(2) \( \ln\sigma'(z_0+z)-\ln\sigma'(z_0)= \sum_{k=0}^{z-1} \ln f'(\sigma(z_0+k)) \)

Extending the sum to non-integer boundaries

The sum can extended to non-integer values \( z \) via Faulhaber's formula which provides the sum for monomials:

\( \sum_{k=n_0}^{n} k^p = \frac{\varphi_{p+1}(n+1) - \varphi_{p+1}(n_0)}{p+1} \),
where \( \varphi_{p}(x) = \sum_{k=0}^{p} \left(p\\k\right) B_k x^{p-k} \) is the \( p \)-th Bernoulli polynomial with \( B_k \) the Bernoulli numbers.

The sum operator is linear for natural boundaries so one can extend Faulhaber's formula to linear combinations of monomials (also called polynomials Wink ), and finally to powerseries. So for non-integer \( z \) we define for
\( f(z)=\sum_{n=0}^\infty f_n z^n \) the extended sum \( \sum_{\zeta = z_0}^{z} f(\zeta) := \sum_{n=0}^\infty f_n \frac{\varphi_{n+1}(z+1)-\varphi_{n+1}(z_0)}{n+1} \).

To simplify the matter we can define the antidifference or indefinite sum:
\( \Delta^{-1}[f](z) :=\sum_{n=0}^\infty f_n \frac{\varphi_{n+1}(z)}{n+1} \)
and get
\( \sum_{\zeta = z_0}^{z} f(\zeta) = \Delta^{-1}[f](z+1) - \Delta^{-1}[f](z_0) \)

Extended sum and shifts

We hope that the extended sum behaves like a normal sum with integer boundaries, e.g. we want to have
(3) \( \sum_{\zeta=a}^b f(\zeta+c) = \sum_{\zeta=a+c}^{b+c} f(\zeta) \).

We show it for polynomials. Let \( p \) be a polynomial of degree \( N \), then we know that (3) is satisfied for all integers \( c \) and both sides are polynomials of maximal degree \( N+1 \) in \( c \). But equality at \( N+1 \) different points already implies the equality of both polynomials.
Hence not only for integer \( c \) but for every \( c \) the equality holds.
This than carries over to powerseries (as limit of a polynomial sequence) if no convergence issues arise.

This property lets us formulate (2) in a much nicer way by applying (3) with \( c=z_0 \) and afterwards setting \( z:=z+z_0 \):

(4) \( \ln\sigma'(z)- \ln\sigma'(z_0)=\sum_{\zeta=z_0}^{z-1} \ln f'(\sigma(\zeta)) \)

Uniqueness of extended sum superfunction
Sketch of proof:

Let \( \rho \) beside \( \sigma \) be another at \( z_0 \) analytic super-exponential (1) that satisfies (2). We further assume \( \sigma \) and \( \rho \) to be invertible at \( z_0 \) and \( \sigma(z_0)=\rho(z_0) \). We set \( \delta=\sigma^{-1}\circ \rho \) in a vicinity of \( z_0 \), so \( \rho(z)=\sigma(\delta(z)) \) there.

\( \delta(z_0)=z_0 \) and by (1) \( \delta(z+n)=\delta(z)+n \) so \( \delta(z_0+n)=z_0+n \) for integer \( n \).

So put \( \rho \) into (4), with \( \rho'(z)=\sigma'(\delta(z)) \delta'(z) \):
\( \ln\sigma'(\delta(z))- \ln\sigma'(z_0)+\ln\delta'(z) =\sum_{\zeta=z_0}^{z-1} \ln f'(\sigma(\delta(\zeta))) \)
the left difference can be replaced with (4), knowing \( \delta(z_0)=z_0 \):
(5) \( \sum_{\zeta=\delta(z_0)}^{\delta(z)-1} \ln f'(\sigma(\zeta))+\ln\delta'(z) =\sum_{\zeta=z_0}^{z-1} \ln f'(\sigma(\delta(\zeta))) \)

Now consider the forward difference
\( \Delta (h\circ \delta) (z) = h(\delta(z+1))-h(\delta(z)) = h(\delta(z)+1)-h(\delta(z)) = \Delta(h)\circ \delta(z) \).
Hence
\( \Delta^{-1} (h\circ \delta) = \Delta^{-1}(h) \circ \delta \) for any \( h \) up to perhaps a constant, so
\( \sum_{\zeta=z_0}^{z-1} h(\delta(\zeta)) = \sum_{\zeta=\delta(z_0)}^{\delta(z)-1} h(\zeta) \).

Now we apply this to (5):
\( \ln\delta'(z) =0 \)
\( \delta'(z)=1 \)
\( \delta(z)=a z + b \)
together with \( \delta(z_0)=z_0 \) and \( \delta(z_0+1)=z_0+1 \) this gives \( a=1 \) and \( b=0 \), so \( \delta(z)=z \) and \( \sigma(z)=\rho(z) \).
Reply
#2
nice proof!

Does this prove the uniqueness for all tetration? i.e. that all tetrations are equal where they are invertible, since they are supposed to be analytic, equal to 1 at 0, and is a superfunction of exponentiation, ex hypothesis.

bo198214 Wrote:if no convergence issues arise.

I am confused on exactly what the problem you tetration guys are trying to solve with tetration.
I am guessing that it may be convergence or holomorphicity.
Reply
#3
(09/06/2009, 04:56 AM)Tetratophile Wrote: Does this prove the uniqueness for all tetration? i.e. that all tetrations are equal where they are invertible, since they are supposed to be analytic, equal to 1 at 0, and is a superfunction of exponentiation, ex hypothesis.

No, this is particularly about Ansus' construction, i.e. all tetrations that satisfy (4) are equal. There are so far no numerical testing on the other tetrations, I will see what I can do about this in the next time.

Quote:
bo198214 Wrote:if no convergence issues arise.

I am confused on exactly what the problem you tetration guys are trying to solve with tetration.
I am guessing that it may be convergence or holomorphicity.

Most methods compute coefficients of powerseries as limits of some computations.
To be on solid ground we need to prove that:
1. The limit for each coefficient exists.
2. The resulting series has non-zero convergence radius.
Nothing of that is proved neither for the matrix power method, the intuitive method nor Ansus' extended sum method.
Though numerically everything looks good.
3. Also it would be nice to prove that the corresponding super-exponential is holomorphic on C without (-oo,-2]. Which also looks good numerically.
Reply
#4
(09/06/2009, 06:54 AM)bo198214 Wrote: 3. Also it would be nice to prove that the corresponding super-exponential is holomorphic on C without (-oo,-2]. Which also looks good numerically.

but I dont think it is.

i will give arguments when i find the time...
Reply
#5
@Bo

In post 1 for your proof I doubt the statements concerning the forward difference.

Could you elaborate on that ? I prefer no use of triangles and clear brackets to avoid confusion.
Afterall this is continuum summation and the hard subject called tetration.

I doubt the uniqueness.

regards

tommy1729
Reply


Possibly Related Threads…
Thread Author Replies Views Last Post
  Extended Schroeder equation Shanghai46 2 5,910 03/23/2025, 11:24 PM
Last Post: MphLee
  Some "Theorem" on the generalized superfunction Leo.W 59 84,091 09/18/2022, 11:05 PM
Last Post: tommy1729
  Uniqueness of fractionally iterated functions Daniel 7 10,997 07/05/2022, 01:21 AM
Last Post: JmsNxn
  Universal uniqueness criterion? bo198214 57 173,961 06/28/2022, 12:00 AM
Last Post: JmsNxn
  A question concerning uniqueness JmsNxn 4 16,930 06/10/2022, 08:45 AM
Last Post: Catullus
  Generalized Kneser superfunction trick (the iterated limit definition) MphLee 25 43,191 05/26/2021, 11:55 PM
Last Post: MphLee
  [Exercise] A deal of Uniqueness-critrion:Gamma-functionas iteration Gottfried 6 15,561 03/19/2021, 01:25 PM
Last Post: tommy1729
  Extended xor to Tetrion space Xorter 10 30,772 08/18/2018, 02:54 AM
Last Post: 11Keith22
  A conjectured uniqueness criteria for analytic tetration Vladimir Reshetnikov 13 40,498 02/17/2017, 05:21 AM
Last Post: JmsNxn
  Uniqueness of half-iterate of exp(x) ? tommy1729 14 53,162 01/09/2017, 02:41 AM
Last Post: Gottfried



Users browsing this thread: 1 Guest(s)