# Fastest way to check if $x^y > y^x$?

What is the fastest way to check if $x^y > y^x$ if I were writing a computer program to do that?

The issue is that $x$ and $y$ can be very large.

• You can test if $\ln(y) + \ln \big( \ln(x) \big) > \ln(x) + \ln \big( \ln(y) \big)$. – jibounet Oct 7 '13 at 10:39
• You want fast or robustly correct? – lhf Oct 7 '13 at 10:40
• @lhf, Fast and in checking among n such pairs of numbers, n/10 could go wrong. Could you please help me deduce what accuracy I'm looking for? – learner Oct 7 '13 at 10:44
• Are you doing http://projecteuler.net/problem=99? – Mårten W Oct 7 '13 at 10:50
• @MårtenW: His problem is more specific. The link you posted is about x^a < y^b. He is considers the special case a=y, b=x – Christian Fries Oct 7 '13 at 12:14

If both $x$ and $y$ are positive then you can just check: $$\frac{\log(x)}{x} \gt \frac{\log(y)}{y}$$ so if both $x$ and $y$ are greater than $e \approx 2.7183$ then you can just check: $$x \lt y$$

• But how x < y? How can we omit the logs? Intuitively it seems right, but is it really so? – learner Oct 7 '13 at 11:18
• @learner, the point is that the function $log(x)/x$ has a single maximum at $x=e$ and is decreasing after that. – lhf Oct 7 '13 at 11:19
• I don't know why this answer is the accepted one, but multiplication is usually faster than division, so I assume $y log(x) > x log(y)$ is faster than this one. – Christian Fries Oct 7 '13 at 12:15
• @ChristianFries, this answer is the accepted one because of the massive optimization given in the second half of the answer. Even if you don't use the monotonicity of $\frac{\log{x}}{x}$, the fact that it's a single function can make repeated queries much faster. – jwg Oct 7 '13 at 12:29
• If $x$ and $y$ are both between $0$ and $e$ then the test becomes $x \gt y$. The hard part is when one is below and one above: for example $2^4 = 4^2$ and $\frac{\log 2}{2} = \frac{\log 4}{4}$ – Henry Oct 7 '13 at 17:16

You might get by testing whether $y \log x > x \log y$, especially if the numbers are only moderately large.

We reduce the question to: is the quotient (x^y) / (y^x) less or greater than 1.

· Noticing that x > 0, y > 0, we take the xy-th root of the quotient:

is the ratio (x^y)^(1/xy) / (y^x)^(1/xy) less or greater than 1^(1/xy).

We reduce the complicated powers, the power of base 1 is 1:

is the ratio x^(1/x) / y^(1/y) less or greater than 1.

is the function f(x) = x^(1/x) increasing or decreasing at the interval [x,y]?

· If already known, via the derivate, that the function f(x) = x^(1/x) has only one maximum, so an absolute maximum for x = e , the point (e, e^(1/e)),

Consider that a) for x > e and y > e , the function f is decreasing, which means

• if e < x < y => f(x) > f(y). [both x and y are greater than e, the greater exponent wins.]

and b) for x < e and y < e the function f is increasing, which means

• if 0 < x < y < e => f(x) < f(y). [both x and y are less than e, the greater base wins.]

• But if x < e < y, the comparison is not decided through the function’s behaviour. [e is between both.] We have to calculate both powers to compare the outcomes.

So there are pairs of x,y existing for which (x^y) = (y^x), with x < e < y.

For example, x = √(√5) and y = √(√(5^5)) = √(√3125)
both x^y and y^x come out on ≈ 20,25372598…

Many pairs are findable, e.g. by setting y = a·x and resolving the equation (x^y) = (y^x), or

f(x) = f(y).