02/01/2008, 01:28 AM (This post was last modified: 02/01/2008, 01:29 AM by GFR.)
Hi, Henrik!
bo198214 Wrote:
GFR Wrote:Actually, the beautiful formula I propose is:
ssqrt(x) = ln(x) / W(ln(x)), which, for x = 1/2, gives:
ssqrt(1/2) = ln(1/2) / W(ln(1/2)) = 0.26289282802173525.. + 0.4996694356833174.. i
That reminds me to update the wikipedia tetration article as there is no formula for the square super root yet.
And indeed your formula is also a solution:
We first see that your formula \( \frac{\ln(x)}{W(\ln(x))} \) is equal to \( \frac{1}{h(1/x)} \) which is a solution to \( y^y=x \):
\( \left(\frac{1}{h(1/x)}\right)^{\frac{1}{h(1/x)}}=\frac{1}{1/x}=x \) as \( h \) is the inverse function of \( x^{1/x} \).
Please don't forget the "lower branch" of y = ssqrt(x), which can be obtained via the second (-1 level) branch of Lambert Function.
(see the attachment)
By calling "plog(x)" [Product Logarithm] the logical union of the two real branches of the Lambert Function, i.e. W(0) and (W-1), i.e.:
plog(z) = W(0,z) OR W(-1,z), we may write:
02/01/2008, 10:31 AM (This post was last modified: 02/01/2008, 10:33 AM by GFR.)
Well, the use of the Lambert Function, for finding back x in y = x^x, is an old affair. The point is that, the present Wikipedia formula seems to covers only the "upper branch" of ssqrt(x), if we only use the W(0) "classical" branch, as it seems to be indicated, afaik (as far as I know), in the the first wikiplot of the W article..
A more complete ssqrt formula should also cover its "lower branch". The definition of "plog(x)" as the logical union of the two W(0) and W(-1) product logarithm real branches will do the business. But, I am sure that Henryk will amend and upgrade the tetration article, accordingly.
02/01/2008, 11:48 AM (This post was last modified: 02/01/2008, 11:49 AM by GFR.)
Ops ... sorry! Once again!
the same situation is found in the definition of h(b), inverse of b = h^(1/h). In fact, we know that the "classical" solution is:
if: b = h^(1/h)
then: h = - W(-lnb)/lnb
We also know that, if we only take the W(0) branch of the Lambert Function, we only get the lower branch of h(b), which is not sufficient. Therefore, I propose to use the plog(z) operator, always meaning: plog(z) = W(0,z) OR W(-1,z), and write:
h(b) = plog(-lnb)/(-lnb) = b#(+oo)
See the attachment.
We can see (as we also know) that:
h(1/a) . ssqrt(a) = h(a) . ssqrt(1/a) = 1
I agree that it has a real limit in the Eta-Beta range [Eta = e^(1/e), Beta = e^(-e) !!]. Nevertheless, we must admit that there is also another real upper branch. Think of the famous example of b = sqrt(2). We have two heights h(sqrt(2)) = {2, 4}.
In fact, by applying b = h^(1/h), we have: 2^(1/2) = 4^(1/4) = sqrt(2).
The lower branch is reachable starting from points (-1,0) and (0,1) by applying b#(x+1) = b^(b#x). The upper branch is unreachablke (big mystery), but still it satisfies the same conditions. Always in the case of b = sqrt(1/2), if the upper branch is 4, we have that:
if: sqrt(2) # x = 4,
then: sqrt(2) ^ (sqrt(2) # x) = sqrt(2) ^ 4 = 4 , for any "x" !!
I mean that the upper branch defines (according to my point of view) a second constant real numerical value in y = b^x, in the 1-to-Eta domain of b.
I might be wrong. But, I still believe it. We always have the same problems with two-valued "functions", one of which, in this particular case, should have a "constant" value. I think we already discussed about the example y = x^2. A lot of orthodox mathematicians say that it is not inversible (!?!), and that sqrt(4) = 2, like all the pocket calculators indicate. Neverthaless, we know that sqrt(4) = {-2, +2}.
This time, I go and drink a "koke", or a "cola", as you say in Germany.
GFR Wrote:A lot of orthodox mathematicians say that it is not inversible (!?!),
"inver-t-ible"
Quote:This time, I go and drink a "koke", or a "cola", as you say in Germany.
"coke" or "latte macchiato" ...
If tetration is defined as iterated exponentiation, with
y = a_n^a_(n-1)^...^a_2^a_1^a_0
and a_1=a_2=a_3=...=a_n =b, a_0=a , thus y = b^...^b^b^a
then the partial-expression argument, from which the multivaluedness of tetration is denied, is removed by a clean definition.
We should initiate an international campaign on this subject
(Are NGO's allowed as consultants at MAA?)
I think I have to intervene and to clarify the basics.
1. There is the concept of a function in mathematics: We assign to every argument of the domain exactly one value of the codomain of the function. An example is \( f(x)=x^2 \) with the domain of the real numbers and also (!) \( f(x)=\sqrt{x} \) with the domain (and codomain!) being the positive real numbers. In the same way is \( f(x)=\ln(x) \) a function defined on the positive real numbers and especially \( f(x)={^\infty}x \) is a function defined on \( e^{-e}<x<e^{1/e} \).
Those above functions can be (uniquely) developed into power series (at points of their domain) and hence being continued somewhat outside the real line on the complex plane. This continuation is performed along a path and usually the continuation along two different paths gives the same value at the common end point, however in certain cases - if you start at point \( z \) wind the path around a singularity and finish again at \( z \) - the continuation value is different from the starting value, and these different values are to be said the branches of the function \( f \).
For example if you take the function (single valued!) \( f(x)=\sqrt{x} \), you can develop it at 1 by
\( \sqrt{z+1}=\sum_{n=0}^\infty \left(0.5\\n\right) z^n \).
Surely there is a (/the only) singularity at 0.
If you continue the function to -1 along a path above 0 you get \( +i \) and if you continue the function to -1 along a path below 0 you get \( -i \).
To avoid this scenario you slice the complex plane at the negative real axis. This means no path for continuation is allowed to cross the negative real axis, more precisely the argument \( \alpha(t) \) of such a path must be continuous and completely lie in \( -\pi < \alpha(t) \le \pi \).
With this sliced complex plane by continuation you get exactly one value of \( \sqrt{z} \) for each point on the complex plane, particularely you get \( \sqrt{-1}=i \).
From there you can get to the branches. You can continue the function around 0 and get the values \( e^{2\pi I k/2} \sqrt{z} = (-1)^k \sqrt{z} \) where \( k \) is the winding number of the path, i.e. how often it does anticlockwise cross the negative real axis (minus the clockwise crossings). This is the same as the number of discontinuities in \( \alpha(t) \) going from \( \pi \) to \( -\pi \) (minus the discontinuities going from \( -\pi \) to \( \pi \)).
For the logarithm you use also this sliced complex plane (which guaranties a unique value everywhere except 0) and for the branches you get \( \ln(z)+2\pi I k \) where \( k \) is again the winding number.
Now back to \( h(x)={^\infty}x \) this is a function (single valued!) on \( e^{-e}<x<e^{1/e} \). It can be developed there into a power series (did Ioannis derive a powerseries?) and hence continued to the complex plane.
It has (the only?) singularity at \( e^{1/e} \) and hence it would make sense to define it uniquely on the at \( [e^{1/e},\infty) \) sliced complex plane, more precisely a path for continuation \( p(t) \) must satisfy \( 0\le \text{arg}\left(p(t)-e^{1/e}\right)<2\pi \).
Starting fromt this definition \( h_k(z) \) would be the continuation for a path with winding number \( k \) in the above sliced plane.
However I am not completely sure if that defintion of sliced plane is compatible with the definition of the branches of the lambert function, such that we could write:
\( h_k(z)=\frac{W_k(-\ln(z))}{-\ln(z)} \)
or if we even should take into account the branches of the logarithm.
Perhaps someone takes pity on that somewhat tedious task.
Are the "things" covered by those theories invertible? (... thank you, Gottfried!).
"Latte makkiato" is very good! An international campaign will certainly coming soon into the picture. We are in the appropriate times. Which NGO, Gottfried? By the way, MAA is the name of a car insurance company, in France.