Equation xy = yx

From HandWiki
Short description: none


Graph of xy = yx. The line and curve intersect at (e, e).

In general, exponentiation fails to be commutative. However, the equation [math]\displaystyle{ x^y = y^x }[/math] has solutions, such as [math]\displaystyle{ x=2,\ y=4. }[/math][1]

History

The equation [math]\displaystyle{ x^y=y^x }[/math] is mentioned in a letter of Bernoulli to Goldbach (29 June 1728[2]). The letter contains a statement that when [math]\displaystyle{ x\ne y, }[/math] the only solutions in natural numbers are [math]\displaystyle{ (2, 4) }[/math] and [math]\displaystyle{ (4, 2), }[/math] although there are infinitely many solutions in rational numbers, such as [math]\displaystyle{ (\tfrac{27}{8}, \tfrac{9}{4}) }[/math] and [math]\displaystyle{ (\tfrac{9}{4}, \tfrac{27}{8}) }[/math].[3][4] The reply by Goldbach (31 January 1729[2]) contains a general solution of the equation, obtained by substituting [math]\displaystyle{ y=vx. }[/math][3] A similar solution was found by Euler.[4]

J. van Hengel pointed out that if [math]\displaystyle{ r, n }[/math] are positive integers with [math]\displaystyle{ r \geq 3 }[/math], then [math]\displaystyle{ r^{r+n} \gt (r+n)^r; }[/math] therefore it is enough to consider possibilities [math]\displaystyle{ x = 1 }[/math] and [math]\displaystyle{ x = 2 }[/math] in order to find solutions in natural numbers.[4][5]

The problem was discussed in a number of publications.[2][3][4] In 1960, the equation was among the questions on the William Lowell Putnam Competition,[6][7] which prompted Alvin Hausner to extend results to algebraic number fields.[3][8]

Positive real solutions

Main source:[1]

Explicit form

An infinite set of trivial solutions in positive real numbers is given by [math]\displaystyle{ x = y. }[/math] Nontrivial solutions can be written explicitly using the Lambert W function. The idea is to write the equation as [math]\displaystyle{ ae^b = c }[/math] and try to match [math]\displaystyle{ a }[/math] and [math]\displaystyle{ b }[/math] by multiplying and raising both sides by the same value. Then apply the definition of the Lambert W function [math]\displaystyle{ a'e^{a'} = c' \Rightarrow a' = W(c') }[/math] to isolate the desired variable.

[math]\displaystyle{ \begin{align} y^x &= x^y = \exp\left(y\ln x\right) & \\ y^x \exp\left(-y\ln x\right) &= 1 & \left(\mbox{multiply by } \exp\left(-y\ln x\right)\right) \\ y\exp\left(-y\frac{\ln x}{x}\right) &= 1 & \left(\mbox{raise by } 1/x\right) \\ -y\frac{\ln x}{x}\exp\left(-y\frac{\ln x}{x}\right) &= \frac{-\ln x}{x} & \left(\mbox{multiply by } \frac{-\ln x}{x}\right) \end{align} }[/math]
[math]\displaystyle{ \Rightarrow -y\frac{\ln x}{x} = W\left(\frac{-\ln x}{x}\right) }[/math]
[math]\displaystyle{ \Rightarrow y = \frac{-x}{\ln x}\cdot W\left(\frac{-\ln x}{x}\right) = \exp\left(-W\left(\frac{-\ln x}{x}\right)\right) }[/math]

Where in the last step we used the identity [math]\displaystyle{ W(x)/x = \exp(-W(x)) }[/math].

Here we split the solution into the two branches of the Lambert W function and focus on each interval of interest, applying the identities:

[math]\displaystyle{ \begin{align} W_0\left(\frac{-\ln x}{x}\right) &= -\ln x \quad&\text{for } &0 \lt x \le e, \\ W_{-1}\left(\frac{-\ln x}{x}\right) &= -\ln x \quad&\text{for } &x \ge e. \end{align} }[/math]
  • [math]\displaystyle{ 0 \lt x \le 1 }[/math]:
[math]\displaystyle{ \Rightarrow \frac{-\ln x}{x} \ge 0 }[/math]
[math]\displaystyle{ \begin{align}\Rightarrow y &= \exp\left(-W_0\left(\frac{-\ln x}{x}\right)\right) \\ &= \exp\left(-(-\ln x)\right) \\ &= x \end{align} }[/math]
  • [math]\displaystyle{ 1 \lt x \lt e }[/math]:
[math]\displaystyle{ \Rightarrow \frac{-1}{e} \lt \frac{-\ln x}{x} \lt 0 }[/math]
[math]\displaystyle{ \Rightarrow y = \begin{cases} \exp\left(-W_0\left(\frac{-\ln x}{x}\right)\right) = x \\ \exp\left(-W_{-1}\left(\frac{-\ln x}{x}\right)\right) \end{cases} }[/math]
  • [math]\displaystyle{ x = e }[/math]:
[math]\displaystyle{ \Rightarrow \frac{-\ln x}{x} = \frac{-1}{e} }[/math]
[math]\displaystyle{ \Rightarrow y = \begin{cases} \exp\left(-W_0\left(\frac{-\ln x}{x}\right)\right) = x \\ \exp\left(-W_{-1}\left(\frac{-\ln x}{x}\right)\right) = x \end{cases} }[/math]
  • [math]\displaystyle{ x \gt e }[/math]:
[math]\displaystyle{ \Rightarrow \frac{-1}{e} \lt \frac{-\ln x}{x} \lt 0 }[/math]
[math]\displaystyle{ \Rightarrow y = \begin{cases} \exp\left(-W_0\left(\frac{-\ln x}{x}\right)\right) \\ \exp\left(-W_{-1}\left(\frac{-\ln x}{x}\right)\right) = x \end{cases} }[/math]

Hence the non-trivial solutions are:

[math]\displaystyle{ y = \begin{cases} \exp\left(-W_0\left(\frac{-\ln(x)}{x}\right)\right) \quad &\text{for } x \gt e,\\ \exp\left(-W_{-1}\left(\frac{-\ln x}{x}\right)\right) \quad &\text{for } 1 \lt x \lt e. \end{cases} }[/math]

Parametric form

Nontrivial solutions can be more easily found by assuming [math]\displaystyle{ x \ne y }[/math] and letting [math]\displaystyle{ y = vx. }[/math] Then

[math]\displaystyle{ (vx)^x = x^{vx} = (x^v)^x. }[/math]

Raising both sides to the power [math]\displaystyle{ \tfrac{1}{x} }[/math] and dividing by [math]\displaystyle{ x }[/math], we get

[math]\displaystyle{ v = x^{v-1}. }[/math]

Then nontrivial solutions in positive real numbers are expressed as the parametric equation

[math]\displaystyle{ \begin{align}x &= v^{1/(v-1)}, \\ y &= v^{v/(v-1)}.\end{align} }[/math]

The full solution thus is [math]\displaystyle{ (y=x) \cup \left(v^{1/(v-1)},v^{v/(v-1)}\right) \text{ for } v \gt 0, v \neq 1 . }[/math]

Based on the above solution, the derivative [math]\displaystyle{ dy/dx }[/math] is [math]\displaystyle{ 1 }[/math] for the [math]\displaystyle{ (x,y) }[/math] pairs on the line [math]\displaystyle{ y=x, }[/math] and for the other [math]\displaystyle{ (x,y) }[/math] pairs can be found by [math]\displaystyle{ (dy/dv)/(dx/dv), }[/math] which straightforward calculus gives as:

[math]\displaystyle{ \frac{dy}{dx} = v^2\left(\frac{v-1-\ln v}{v-1-v\ln v}\right) }[/math]

for [math]\displaystyle{ v \gt 0 }[/math] and [math]\displaystyle{ v \neq 1. }[/math]

Setting [math]\displaystyle{ v=2 }[/math] or [math]\displaystyle{ v=\tfrac{1}{2} }[/math] generates the nontrivial solution in positive integers, [math]\displaystyle{ 4^2=2^4. }[/math]

Other pairs consisting of algebraic numbers exist, such as [math]\displaystyle{ \sqrt 3 }[/math] and [math]\displaystyle{ 3\sqrt 3 }[/math], as well as [math]\displaystyle{ \sqrt[3]4 }[/math] and [math]\displaystyle{ 4\sqrt[3]4 }[/math].

The parameterization above leads to a geometric property of this curve. It can be shown that [math]\displaystyle{ x^y = y^x }[/math] describes the isocline curve where power functions of the form [math]\displaystyle{ x^v }[/math] have slope [math]\displaystyle{ v^2 }[/math] for some positive real choice of [math]\displaystyle{ v\neq 1 }[/math]. For example, [math]\displaystyle{ x^8=y }[/math] has a slope of [math]\displaystyle{ 8^2 }[/math] at [math]\displaystyle{ (\sqrt[7]{8}, \sqrt[7]{8}^8), }[/math] which is also a point on the curve [math]\displaystyle{ x^y=y^x. }[/math]

The trivial and non-trivial solutions intersect when [math]\displaystyle{ v = 1 }[/math]. The equations above cannot be evaluated directly at [math]\displaystyle{ v = 1 }[/math], but we can take the limit as [math]\displaystyle{ v\to 1 }[/math]. This is most conveniently done by substituting [math]\displaystyle{ v = 1 + 1/n }[/math] and letting [math]\displaystyle{ n\to\infty }[/math], so

[math]\displaystyle{ x = \lim_{v\to 1}v^{1/(v-1)} = \lim_{n\to\infty}\left(1+\frac 1n\right)^n = e. }[/math]

Thus, the line [math]\displaystyle{ y = x }[/math] and the curve for [math]\displaystyle{ x^y-y^x = 0,\,\, y \ne x }[/math] intersect at x = y = e.

As [math]\displaystyle{ x \to \infty }[/math], the nontrivial solution asymptotes to the line [math]\displaystyle{ y = 1 }[/math]. A more complete asymptotic form is

[math]\displaystyle{ y = 1 + \frac{\ln x}{x} + \frac{3}{2} \frac{(\ln x)^2}{x^2} + \cdots. }[/math]

Other real solutions

An infinite set of discrete real solutions with at least one of [math]\displaystyle{ x }[/math] and [math]\displaystyle{ y }[/math] negative also exist. These are provided by the above parameterization when the values generated are real. For example, [math]\displaystyle{ x=\frac{1}{\sqrt[3]{-2}} }[/math], [math]\displaystyle{ y=\frac{-2}{\sqrt[3]{-2}} }[/math] is a solution (using the real cube root of [math]\displaystyle{ -2 }[/math]). Similarly an infinite set of discrete solutions is given by the trivial solution [math]\displaystyle{ y=x }[/math] for [math]\displaystyle{ x\lt 0 }[/math] when [math]\displaystyle{ x^x }[/math] is real; for example [math]\displaystyle{ x=y=-1 }[/math].

Similar graphs

Equation xy = yx

The equation [math]\displaystyle{ \sqrt[x]y = \sqrt[y]x }[/math] produces a graph where the line and curve intersect at [math]\displaystyle{ 1/e }[/math]. The curve also terminates at (0, 1) and (1, 0), instead of continuing on to infinity.

The curved section can be written explicitly as

[math]\displaystyle{ y=e^{W_0(\ln(x^x))} \quad \mathrm{for} \quad 0\lt x\lt 1/e, }[/math]

[math]\displaystyle{ y=e^{W_{-1}(\ln(x^x))} \quad \mathrm{for} \quad 1/e\lt x\lt 1. }[/math]

This equation describes the isocline curve where power functions have slope 1, analogous to the geometric property of [math]\displaystyle{ x^y = y^x }[/math] described above.

The equation is equivalent to [math]\displaystyle{ y^y=x^x, }[/math] as can be seen by raising both sides to the power [math]\displaystyle{ xy. }[/math] Equivalently, this can also be shown to demonstrate that the equation [math]\displaystyle{ \sqrt[y]{y}=\sqrt[x]{x} }[/math] is equivalent to [math]\displaystyle{ x^y = y^x }[/math].

Equation logx(y) = logy(x)

The equation [math]\displaystyle{ \log_x(y) = \log_y(x) }[/math] produces a graph where the curve and line intersect at (1, 1). The curve becomes asymptotic to 0, as opposed to 1; it is, in fact, the positive section of y = 1/x.

References

  1. 1.0 1.1 Lóczi, Lajos. "On commutative and associative powers". KöMaL. http://www.komal.hu/cikkek/loczy/powers/commpower.e.shtml.  Translation of: "Mikor kommutatív, illetve asszociatív a hatványozás?" (in hu). http://db.komal.hu/KomalHU/cikk.phtml?id=200047. 
  2. 2.0 2.1 2.2 Singmaster, David. "Sources in recreational mathematics: an annotated bibliography. 8th preliminary edition". http://www.gotham-corp.com/sources.htm#_Toc69534169. 
  3. 3.0 3.1 3.2 3.3 "On the Rational Solutions of xy = yx". Mathematics Magazine 63: 30–33. 1990. doi:10.1080/0025570X.1990.11977480. http://www.maa.org/sites/default/files/Sved50816668.pdf. 
  4. 4.0 4.1 4.2 4.3 "Rational solutions of xy = yx", History of the Theory of Numbers, II, Washington, 1920, pp. 687, https://books.google.com/books?id=dO7C02z4LlcC&pg=PA687 
  5. van Hengel, Johann (1888). Beweis des Satzes, dass unter allen reellen positiven ganzen Zahlen nur das Zahlenpaar 4 und 2 für a und b der Gleichung ab = ba genügt. http://digital.ub.uni-duesseldorf.de/ulbdsp/periodical/titleinfo/4315444. 
  6. "The twenty-first William Lowell Putnam mathematical competition (December 3, 1960), afternoon session, problem 1", The William Lowell Putnam mathematical competition problems and solutions: 1938-1964, MAA, 1980, pp. 59, ISBN 0-88385-428-7, https://books.google.com/books?id=7D0PAQAAMAAJ&q=%22prove+that+you+have+obtained+all+of+them%22 
  7. "21st Putnam 1960. Problem B1". 20 Oct 1999. http://www.kalva.demon.co.uk/putnam/putn60.html. 
  8. Hausner, Alvin (November 1961). "Algebraic Number Fields and the Diophantine Equation mn = nm". The American Mathematical Monthly 68 (9): 856–861. doi:10.1080/00029890.1961.11989781. ISSN 0002-9890. 

External links