4

Is it possible to solve equations in the form $k^n=n$ for n and if so, How? I am new to logarithms and so would be glad if someone could explain even if there is an obvious answer.

Also What about $k^{a+b}=a$ for a? Or $k^{ab}=a$?

  • You can solve it in term of the Lambert W function, look here https://en.wikipedia.org/wiki/Lambert_W_function#Applications – Ant May 10 '16 at 19:08
  • 1
    $2^n=n$ has no solution for $n \in Z$ because $2^n > n$. – lhf May 10 '16 at 19:09
  • @Ant ... He asks about three equations, and you tell him about "it". Yes, all three may be solved using W. $2^n=n$ has no real solution but it does have complex solutions. – GEdgar May 10 '16 at 19:12
  • At least two or three times a week we get questions of this sort, equating polynomials of degree $n\geq 1$ with exponential functions which have no solutions without Lambert W. Surely these are all duplicates of each other? – MathematicsStudent1122 May 10 '16 at 21:10

2 Answers2

3

All these equations can be standardized to the form

$$xe^x=y$$ for which the general solution has been studied in depth and is denoted as the function $x=W(y)$ known as Lambert's.

  • $k^n=n$

Write $k^n=e^{-x}$, i.e. $x=-n\log(k)$, and $e^{-x}=-\dfrac x{\log(k)}$ or $xe^x=-\log(k)$.

  • $k^{a+b}=a=k^ak^b$

Like before, $xe^x=-\log(k)k^b$, with $x=-a\log(k)$.

  • $k^{ab}=a$.

Can be rewritten $bk^{ab}=ab$, and $xe^x=-\log(k)b$, with $x=-ab\log(k)$.


In general, there is no simpler approach, but you can create equations with a known solution by working in reverse.

For instance, taking $x=-\log(2)$ so that $\log(k)=\dfrac{\log(2)}2$ yields the equation

$$(\sqrt2)^n=n$$ with the solution $n=2$.


For those not willing to take Lambert's $W$ for granted we can discuss the real roots of $xe^x=y$.

The derivative is $(x+1)e^x$ so that the function is decreasing from an horizontal asymptote $(-\infty,0)$ to the minimum at $(-1,-1/e)$, then increasing exponentially to $(\infty,\infty)$ after crossing the origin $(0,0)$.

So there are no solutions in $y$ for $x<\frac1e$, two negative solutions on both sides of $x=-1$ for $1/e<y<0$ and a single solution for $y>0$.

  • 1
    Actually, according to Corless, Gonnet, Hare, Jeffrey and Knuth, the equation Lambert studied was $x = q + x^m$; Euler further developed this, eventually getting an equation equivalent to $x e^x = y$ in a limiting case. But Euler has enough things named after him, so Corless, Gonnet, Hare, Jeffrey and Knuth named their function after Lambert. – Robert Israel May 10 '16 at 20:13
  • @RobertIsrael: interesting, thanks. I will rephrase to comply. –  May 10 '16 at 20:17
2

I think that using Lambert's $W$ is like cheating. ;-) To get a more general approach, let's study the equation $k^x=x$, with $k>0$.

Consider the function $f(x)=k^x-x$. We have $$ \lim_{x\to-\infty}f(x)=\infty $$ and $$ \lim_{x\to\infty}f(x)= \begin{cases} -\infty & \text{if $0<k\le 1$}\\[4px] \infty & \text{if $k>1$} \end{cases} $$ Moreover $$ f'(x)=k^x\log k-1 $$ If $0<k\le1$ we have $f'(x)<0$ for all $x$, so the function crosses the $x$-axis exactly once and the equation has a single solution. Since $f(0)=1$ and $f(1)=k-1<0$, we know the solution is in the interval $(0,1)$.

The more interesting case is $k>1$. The derivative vanishes for $$ k^x=\frac{1}{\log k} $$ so for $x\log k=-\log\log k$ or $$ x=-\frac{\log\log k}{\log k} $$ where the function has an absolute minimum.

Now $$ f\left(-\frac{\log\log k}{\log k}\right)= \frac{1}{\log k}+\frac{\log\log k}{\log k}= \frac{1+\log\log k}{\log k} $$ The equation has

  • two solutions if $1+\log\log k<0$, that is, $\log k<e^{-1}$ or $1<k<e^{1/e}$;

  • one solution if $k=e^{1/e}$;

  • no solution if $k>e^{1/e}$.

The equations of the form $k^{x+b}=x$ can be studied similarly; more generally, you can study $qk^x=x$, taking into account that $k^{x+b}=k^b\cdot k^x$.

Equations like $k^{xb}=x$ are just the same as before, because they can be rewritten as $(k^b)^x=x$.

egreg
  • 238,574
  • It does feel a bit like cheating, I agree, but it is a convenient vehicle for describing how solutions to similar equations are related, so it's at least somewhat worthwhile. – Brian Tung May 10 '16 at 21:17
  • @BrianTung No doubt about it. But one should always keep in mind that Lambert's $W$ has two branches. – egreg May 10 '16 at 21:25