Given the usual field and ordering axioms for the real numbers, it isn't difficult to prove that $x<y$, without any further restrictions on the signs of $x$ and $y$, implies $x^n<y^n$, with $n$ an odd natural number. However, all the proofs I've seen and come up with myself seem to rely on distinguishing between the different possible combinations of signs of $x$ and $y$. I find such "case-by-case" proofs, especially of elementary facts, rather unappealing.

Now, my question is: does there exist an elegant one-line proof of the fact that $x<y\Rightarrow x^n<y^n$ (for $n$ an odd natural number) which doesn't take into account the signs of $x$ and $y$?

Thanks for reading and for any comments!

  • $\begingroup$ Are you looking for a deduction from the axioms? $\endgroup$ – Marco Disce Feb 23 '16 at 21:58
  • $\begingroup$ @Marco Disce. All valid proofs of this are deductions from the axioms, one way or another. I'm looking for a fairly elementary proof, vague as that concept may be. $\endgroup$ – Kim Fierens Feb 23 '16 at 22:03
  • $\begingroup$ This is a good question, since $n$ has to be odd I don't think you will get a one liner. $\endgroup$ – Brennan.Tobias Feb 24 '16 at 0:30
  • $\begingroup$ $0 <x <y => x^{n+1}=x^nx <y^nx <y^n*y=y^{n+1} $ via induction for all (odd and even) natural numbers. $\endgroup$ – fleablood Feb 24 '16 at 1:48

The "One Liner": $$ x^n - y^n = \frac{(X-Y)^n - (X+Y)^n}{2^{n/2}} = \sum_{k=0}^n \binom{n}{k} \frac{\left[ X^{n-k}Y^k \left( (-1)^k - 1 \right) \right]}{2^{n/2}} = N Y = P (x - y) $$ where $N$ is a negative and $P$ is a positive number (both dependent on the particular $x$, $y$).

Explanation: \begin{align} x^n - y^n &= \frac{1}{2^{n/2}} \left[ (X-Y)^n - (X+Y)^n \right] \\ &= \frac{1}{2^{n/2}} \sum_{k=0}^n \binom{n}{k} \left[ X^{n-k}(-1)^k Y^k - X^{n-k} Y^k \right] \\ &= \frac{1}{2^{n/2}} \sum_{k=0}^n \binom{n}{k} \left[ X^{n-k} Y^k ((-1)^k - 1) \right] \quad (*) \\ &= N Y \\ &= \frac{N}{\sqrt{2}} (y - x) \\ &= -\frac{N}{\sqrt{2}}(x - y) \\ &= P (x - y) \end{align} This one uses a coordinate transformation $(x,y) \to (X,Y)$, a rotation by $-45^\circ$, to move all points with $x < y$ into the upper half plane such that the expression simplifies.

Here are plots (3D, isolines) for $x^3 - y^3$ before and after the rotation:

x^3-y^3, 3D plot $\quad$ x^3-y^3, isolines $\quad \begin{matrix}\to \\ \\ \\ \\ \\ \end{matrix} \quad$ transformed function, 3D plot $\quad$ transformed function, isolines

We see that $N$ consists mainly of the even powers of the binomial expansion, and then we transform back.

The necessary fact that $n$ is odd is used at $(*)$ to note that the differences $n-k$ are even, so $X$ shows up with even powers, while $Y$ stays with odd powers, so we can pull $Y$ out and are left with even powers of $X$ and $Y$.

  • $\begingroup$ I don't mean to be critical but you could make any proof a "one-liner" by making the characters sufficiently small. The key word is "elegant". Binomial expansions are not elegant. $\endgroup$ – MathematicsStudent1122 Feb 23 '16 at 23:46
  • $\begingroup$ I massaged it a bit. The only elegant idea here is the coordinate transformation. I thought about just putting the four images out as one liner, but I was not sure how many folks would get it. $\endgroup$ – mvw Feb 23 '16 at 23:56
  • $\begingroup$ @MathematicsStudent1122: you have a strange notion of elegance in mathematics if you find the binomial theorem inelegant. You could argue that this particular use of the binomial theorem is inelegant. $\endgroup$ – Rob Arthan Feb 24 '16 at 0:07
  • $\begingroup$ I happen to think this proof is nice and beautiful. Exactly the kind of thing I've been looking for but couldn't find myself. Well done! $\endgroup$ – Kim Fierens Feb 24 '16 at 1:03

$f(t)=t^{2k+1}$ is a strictly monotonic function, as $f'(t)>0$ (except at $t=0$).

  • $\begingroup$ Yes, that certainly fits the description "one-line" proof. $\endgroup$ – Kim Fierens Feb 23 '16 at 21:57

$$\begin{align}y^n-x^n=(y-x)\sum_{k=0}^{n-1}x^ky^{n-1-k}=y^{n-1}(y-x)\underset{\text{$>1+x/y>0$ if} \ \lvert x/y\rvert<1}{\underbrace{\sum_{k=0}^{n-1}(x/y)^k}}&\ge y^{n-1}(y-x)\underset{>0\ \text{if}\ \lvert x/y\rvert>1}{\underbrace{\sum_{k=0}^{(n-1)/2}\lvert x/y\rvert^{2k}(1-\lvert x/y\rvert^{-1})}}\end{align}$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.