In the following proof that $\sqrt2$ is irrational, I cannot make sense of why $\frac{q a_n + p b_n}{q} \geqslant \frac{1}{q}$.
The sums corresponding to $a_n$ and $b_n$ are alternating sums, so why must $q a_n + p b_n \geqslant 1$ ?


Suppose $\sqrt2 = \frac{p}{q} \in \mathbb{Q}$. On the one hand, $0 < (\sqrt{2} - 1)^{n} < 1$. Thus,

$$ \lim_{n \to \infty} (\sqrt{2} - 1)^{n} = 0 $$

On the other hand, the binomial expansion of $(\sqrt{2} - 1)^{n}$ is given by

$$ (\sqrt{2} - 1)^{n} = \sum_{k = 0}^{n} {n \choose k}(\sqrt{2})^{k}(-1)^{n-k} $$ $$ = \sum_{\substack{k = 0 \\ k \text{ even}}}^{n} {n \choose k}(2)^{\frac k2}(-1)^{n-k} + \sqrt2 \sum_{\substack{k = 0 \\ k \text{ odd}}}^{n} {n \choose k}(2)^{\frac{k-1}{2}}(-1)^{n-k} $$ $$ = a_n + b_n \sqrt2 $$

Finally, we see that

$$ (\sqrt{2} - 1)^{n} = a_n + b_n \sqrt2 = a_n + b_n \frac{p}{q} = \frac{q a_n + p b_n}{q} \geqslant \frac{1}{q} $$

Contradicting the fact that $(\sqrt{2} - 1)^{n}$ tends to $0$ as $n \to \infty$


  • $\begingroup$ Fun proof! Different than the usual one , even though I'd say not better $\endgroup$ – Ant Jul 4 '16 at 5:59

As $(\sqrt{2}-1)^n > 0$ we know that $\frac{q a_n + p b_n}{q} >0$ and therefore $q a_n + p b_n>0$.

As $q a_n + p b_n$ is an integer we have $q a_n + p b_n \geq 1$

  • $\begingroup$ Is a positive integer (which is clear, but perhaps it's better to state it explicitly). $\endgroup$ – egreg Jul 3 '16 at 15:57
  • $\begingroup$ So simple but so much headache. Thank you! $\endgroup$ – HoopaU Jul 3 '16 at 22:50

The proof given is unnecessarily complicated, but never mind it is OK. The idea it is using is frequently seen in irrationality proofs.

In order to prove that a number $\alpha$ is irrational, we construct some expression dependent on $\alpha$ and some other parameters (in most case an integer $n$) say $f(\alpha, n)$ such that the expression $f$ is non-zero but can me made to be close to $0$ by choosing $n$ large (i.e. limit of $f$ as $n \to \infty$ is $0$). Moreover this expression $f$ is chosen to be such that is it a rational number with bounded denominator if $\alpha$ is rational (in most cases the denominator of $f$ is same as that of $\alpha$). This ensures that $|f|$ is a non-zero rational number with denominator no greater than say $q$. Then the numerator of $|f|$ is positive and thus $|f| \geq 1/q$ so that limit of $f$ cannot be $0$ and thus we get contradiction.

The expression $f$ here is $f(n) = (\sqrt{2} - 1)^{n}$ and it tends to $0$ as $n \to \infty$. Now if $\sqrt{2} = p/q$ then $$f(n) = \frac{(p - q)^{n}}{q^{n}}$$ and it appears that denominator of $f(n)$ is unbounded. Some ingenuity is needed to show that $f(n)$ indeed can be expressed as a rational number with denominator $q$ instead of $q^{n}$. Here we use the idea that a polynomial in $\sqrt{2} = \alpha$ with integer coefficients can be expressed as $a + b\alpha$ where $a, b$ are integers. Thus $f(n) = (\alpha - 1)^{n}$ being a polynomial in $\alpha$ can be expressed as $$a + b\alpha = \frac{aq + bp}{q}$$ if $\alpha = p/q$.

The idea descibed above is very simple and the real challenge is to find / invent an appropriate function $f$ with the desired property based on the number $\alpha$ under consideration. You can find a lot of such ingenious proofs by Ivan M. Niven in his book Irrational Numbers. Also see application of this technique for irrationality of $\zeta(2)$ and $\zeta(3)$.


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.