Results 1 to 14 of 14

Math Help - Find derivative!

  1. #1
    Member
    Joined
    Jan 2006
    From
    Gdansk, Poland
    Posts
    117

    Find derivative!

    The task is defined the following way:
    "We have the function f(x) = x^{x^{x^{\ldots}}}
    Find it's derivative!"

    There is no more additional information about f(x) so, obviously we have to find its natural domain, and the set where it is differetiable.

    I am pretty sure that such set, where f(x) is differetiable is the interval (0, 1] (not sure if it is the maximal).

    I have found a formula, for the derivative, but it contains x^{x^{x^{\ldots}}} terms (founding the inverse function leads to that formula). The problem is: can you find the finite formula for the derivative (does it exist???). Or maybe you can find the derivative written as an infinite series?

    And in the end, one thing. Please, do not confuse the function with \left(\left(\left(x^x\right)^x\right)^x\right)^{..  .} like I did in the first look. It is the function defined that way:

    <br />
f_1 (x) = x <br />

    <br />
f_{n+1} (x) = x^{f_n (x)}<br />

    <br />
f (x) = \lim_{n \rightarrow \infty} f_n (x)<br />
    Follow Math Help Forum on Facebook and Google+

  2. #2
    TD!
    TD! is offline
    Senior Member
    Joined
    Jan 2006
    From
    Brussels, Belgium
    Posts
    405
    Thanks
    3
    Have you tried taking the natural logarithm (f(x) = exp(ln(f(x))) to use the power property of the logarithm? Try to do it for the first few cases, can you detect a pattern and extend it to the general case?
    Follow Math Help Forum on Facebook and Google+

  3. #3
    Global Moderator

    Joined
    Nov 2005
    From
    New York City
    Posts
    10,616
    Thanks
    9
    It seems to be connected to the Ackermann Function
    problem is that your function is an "infinite" ackermann function thus I do not know if this is what you are looking for.
    Follow Math Help Forum on Facebook and Google+

  4. #4
    Member
    Joined
    Jan 2006
    From
    Gdansk, Poland
    Posts
    117
    Quote Originally Posted by TD!
    Have you tried taking the natural logarithm (f(x) = exp(ln(f(x))) to use the power property of the logarithm? Try to do it for the first few cases, can you detect a pattern and extend it to the general case?
    Frankly speaking I haven't tried taking the natural logarithm of this function, and calculate it's derivative (I doubt if it will lead me to the solution, the formulas will go more and more complicated). Maybe I should.

    There is also another problem with this attitude.... Shouldn't be the sequence of functions uniformly convergent to have the property that the limit of the derivatives is the derivative of the limit? I can prove that it is convergent on the (0, 1] interval, but is it uniformly convergent (I belive yes!)?

    But there is another thing. The function can be described this way:

    <br />
y = x^y<br />

    And from that, you can determine the inverse function:
    <br />
x = e^{\frac{\ln y}{y}}<br />

    From that I can prove uniquiness (of course I should prove convergence first)...


    Quote Originally Posted by ThePerfectHacker
    It seems to be connected to the Ackermann Function
    problem is that your function is an "infinite" ackermann function thus I do not know if this is what you are looking for.
    Hm... Ackermann function is a discreete thus it has no derivative. But, maybe it is some trail...
    Follow Math Help Forum on Facebook and Google+

  5. #5
    MHF Contributor
    Joined
    Oct 2005
    From
    Earth
    Posts
    1,599
    Start out with what you wrote.

    y=x^y

    Now take the natural log of both sides, and differentiate.

    ln(y)=yln(x)

    Follow Math Help Forum on Facebook and Google+

  6. #6
    harold84
    Guest

    maybe next step:)

    so...
    I wonder if this method is valid , please write what do you think about it:
    <br />
lny= ylnx<br />
x=e^(lny)/y<br />
    <br />
dx/dy = (1/y^2 - (lny)/y^2)*e^(ln/y) =
    <br />
= ((1-lny)*e^(lny/y))/y^2<br />
    so, now let's find dy/dx
    <br />
dy/dx = y^2/((1-lny)*e^(lny/y)) = <br />

    finally =
    (x^x^x^x^....)^2/((1-lnx^x^x^x....)*x)<br /> <br />
    *). xxxxx what you can see means x power x power x power x.....

    What do you think?????
    Follow Math Help Forum on Facebook and Google+

  7. #7
    Member
    Joined
    Jan 2006
    From
    Gdansk, Poland
    Posts
    117
    Quote Originally Posted by Jameson
    Start out with what you wrote.

    y=x^y

    Now take the natural log of both sides, and differentiate.

    ln(y)=yln(x)

    Yes, and you will get what harold84 did.

    <br />
f'(x) = \frac{(x^{x^{x^{\ldots}}})^2}{x \left(1-\ln \left( (x^{x^{x^{\ldots}}} \right) \right) }<br />

    But, look at my first post. The problem is not if you can find SOME formula for the f(x) derivative. You are to find FINITE formula for the derivative, or prove that there is no FINITE formula for that (if you will find it as a infinite series, or infinite product I will also be happy ).
    Follow Math Help Forum on Facebook and Google+

  8. #8
    Member
    Joined
    Jan 2006
    From
    Gdansk, Poland
    Posts
    117

    Hmm...

    Yes, as I can see noone is able to solve it. If I find a solution I will post it here.
    Follow Math Help Forum on Facebook and Google+

  9. #9
    Newbie Arch_Stanton's Avatar
    Joined
    Nov 2008
    Posts
    21
    I have this thought: isn't f(x) = x^{x^{x^{\ldots}}} function equal to
    f\begin{cases}  \rightarrow \infty \mbox{ for x}>1 \\ \mbox{1     for x}\in (0,1> \\ \mbox{for x}<0 \mbox{ does not exist} \end{cases}
    Function is differentiable in interval x\in(0,1>, where f'(x)=0
    Well, that's my point, what do you think of it?

    Do you think, in this case the function is differentable in interval x\in (1,\infty)(it's somehow constant here).
    Last edited by Arch_Stanton; November 12th 2008 at 05:38 AM.
    Follow Math Help Forum on Facebook and Google+

  10. #10
    Member
    Joined
    Jan 2006
    From
    Gdansk, Poland
    Posts
    117
    How did you find this ooold topic :P ... um.... nevermind

    I guess you are wrong because

    <br />
\left (\left (\left( x^x \right)^x \right)^x \right)^{...} \neq  x^{x^{x^{...}}}<br />

     x^{x^{x^{...}}} is a function f(x) which satisfies equation

    x^{f(x)} = f(x)

    Hence for 0 < x < 1 we cannot have f(x) = 1
    Last edited by ThePerfectHacker; November 12th 2008 at 12:18 PM.
    Follow Math Help Forum on Facebook and Google+

  11. #11
    Newbie Arch_Stanton's Avatar
    Joined
    Nov 2008
    Posts
    21
    I wonder where I've found this old topic myself now...

    \left (\left (\left( x^x \right)^x \right)^x \right)^{...} \neq  x^{x^{x^{...}}}explaines everything.
    Follow Math Help Forum on Facebook and Google+

  12. #12
    Super Member
    Joined
    Aug 2008
    Posts
    903
    That's a power tower: h(z)=z^{z^{\ddots}}

    My dots are going in the wrong direction but you know what I mean. Note that:

    z^h=h

    e^{h\ln(z)}=h

    \ln(z)e^{h\ln(z)}=h\ln(z)

    \ln(z)=h\ln(z)e^{-h\ln(z)}

    -\ln(z)=-h\ln(z)e^{-h\ln(z)}

    At this point it's in the form ge^g which allows us to take the Lambert-W function of both sides:

    h(z)=-\frac{W\big[-ln(z)\big]}{ln(z)};\quad e^{-e}\leq z\leq e^{1/e}

    We can now take it's derivative although I don't know how to derive the derivative for the Lambert-W function:

    \frac{d}{dz} W(z)=\frac{W[z]}{z(1+W[z])}

    Thus using the chain rule:

    h'(z)=\frac{W\big[-\ln(z)\big]}{z\ln^2(z)}-\frac{W\big[-\ln(z)\big]}{z\ln^2(z)\bigg(1+W\big[-\ln(z)\big]\bigg)}

    Check out Mathworld on power towers.
    Last edited by shawsend; November 13th 2008 at 04:21 AM. Reason: added domain of convergence for h(z)
    Follow Math Help Forum on Facebook and Google+

  13. #13
    Member
    Joined
    Jan 2006
    From
    Gdansk, Poland
    Posts
    117
    Quote Originally Posted by shawsend View Post
    At this point it's in the form ge^g which allows us to take the Lambert-W function of both sides:

    h(z)=-\frac{W\big[-ln(z)\big]}{ln(z)};\quad e^{-e}\leq z\leq e^{1/e}
    exp(-exp(1)) = 0.0660 (approx.)

    But it seems to me, that z can be every number greater than 0 less than 1.

    It is obvious (convergence) that 0 < z < 1, but do the conditions should be more strict?

    Starting with equation
    <br />
-h \ln(z) e^{-h \ln z} = - \ln z<br />

    We see that it has to be -h \ln z \in [-1, \infty) to get the correct value on the LHS (when we use Lambert W function). However we can see that -h \ln z > 0 since h > 0 and \ln z < 0 (assumptions on z)

    And the second problem is on RHS, - \ln z must be in the domain of W. Hence - \ln z \in \left[-\frac{1}{e}, \infty \right) (which is obviously true).

    Thus condition on z is: z \in (0, 1)

    Quote Originally Posted by shawsend View Post
    We can now take it's derivative although I don't know how to derive the derivative for the Lambert-W function:

    \frac{d}{dz} W(z)=\frac{W[z]}{z(1+W[z])}
    It's the consequence of inverse function derivative theorem
     (f^{-1})'(x_0) = \frac{1}{f'(f^{-1}(x_0))}

    Having f(t) = t e^t, we get W = f^{-1}

     f'(t) = (t+1)e^t

    Thus
     W'(z) = \frac{1}{f'(W(z))} = \frac{1}{(W(z)+1) e^{W(z)}} = \frac{W(z)}{(W(z)+1)W(z) e^{W(z)}}<br />

    We get
    W'(z) = \frac{W(z)}{z(W(z)+1)}

    Since W(z) e^{W(z)} = z

    -----------------

    To sum up... It looks like we finally have some solution, at least we could express this derivative by some special function (which is satisfying I suppose). Hence probably there is no finite solution for this derivative (its existence would imply finite formula for Lambert W function, which I suppose does not exist).
    Last edited by albi; November 13th 2008 at 10:40 AM.
    Follow Math Help Forum on Facebook and Google+

  14. #14
    Super Member
    Joined
    Aug 2008
    Posts
    903
    Quote Originally Posted by albi View Post
    (which is satisfying I suppose).
    End special function discrimination: equal rights for special functions.

    Thanks also for that derivation above for the derivative!
    Follow Math Help Forum on Facebook and Google+

Similar Math Help Forum Discussions

  1. Find derivative.
    Posted in the Algebra Forum
    Replies: 1
    Last Post: October 11th 2010, 07:46 AM
  2. Find the derivative of: (1-x)*ln(1-x)
    Posted in the Calculus Forum
    Replies: 2
    Last Post: December 29th 2009, 08:35 AM
  3. find derivative
    Posted in the Calculus Forum
    Replies: 8
    Last Post: August 10th 2009, 10:36 AM
  4. find the second derivative of...
    Posted in the Calculus Forum
    Replies: 1
    Last Post: February 16th 2009, 07:06 PM
  5. find the Derivative
    Posted in the Calculus Forum
    Replies: 1
    Last Post: February 26th 2008, 05:35 PM

Search Tags


/mathhelpforum @mathhelpforum