Some days ago I posted a question in MSE in order to correct a solution to the problem of Prove that $[\mathbb{Q}(\sqrt{4+\sqrt{5}},\sqrt{4-\sqrt{5}}):\mathbb{Q}]=8$.

After posting this another question, I found a general argument for this type of extensions. I think that the ideas at the solution of Bill Dubuque in this question could be used to solve the following problem:

Let $p$ and $q$ be distinct positive prime numbers such that $p+q$ is a perfect square. Then $[\mathbb{Q}(\sqrt{\sqrt{p+q}+\sqrt{q}},\sqrt{\sqrt{p+q}-\sqrt{q}}):\mathbb{Q}]=8.$

My attempt of solution:

Let $\alpha_1 = \sqrt{\sqrt{p+q}+\sqrt{q}}$ and $\alpha_2=\sqrt{\sqrt{p+q}-\sqrt{q}}).$ Let $\mathbb{K}=\mathbb{Q}(\alpha_1,\alpha_2)$.

First observe that $$\alpha_1^2 = \sqrt{p+q}+\sqrt{q},$$ and $$\alpha_1 \alpha_2 = \sqrt{p}.$$

Let $\mathbb{L}=\mathbb{Q}(\alpha_1^2,\alpha_1 \alpha_2)=\mathbb{Q}(\sqrt{q},\sqrt{p}).$ We have that $[\mathbb{L}:\mathbb{Q}]=4,$ hence $\mathbb{L}$ is a 2-dimensional vector space over $\mathbb{Q}(\sqrt{q}),$ with basis $\{1,\sqrt{p}\}$. We will prove now that $\alpha_1 \not\in \mathbb{L}:$

Suppose that $\alpha_1 \in \mathbb{L}$ (this imply directly that $\alpha_2 \in \mathbb{L}$ too), then exists unique $a,b \in \mathbb{Q}(\sqrt{q})$ with $$\alpha_1 = a + b\sqrt{p}.$$ Hence, $$\sqrt{p+q}+\sqrt{q} = a^2 + p b^2 + 2ab\sqrt{p},$$ or equivalently: $$2ab\sqrt{p} = \sqrt{p+q}+\sqrt{q} - a^2 - p a^2.$$

Since the right member of the equality is in $\mathbb{Q}(\sqrt{q}),$ must be $a=0$ or $b=0$.

  • If $a=0$ then $\alpha_1 = b\sqrt{p}=b\alpha_1 \alpha_2,$ hence $1=b\alpha_2$ and we conclude that $\alpha_2^{-1}=b \in \mathbb{Q}(\sqrt{q}).$

  • If $b=0$ then $\alpha_1=a \in \mathbb{Q}(\sqrt{q}).$

Both cases gets a contradiction since $\sqrt{\sqrt{p+q}\pm\sqrt{q}}\not\in\mathbb{Q}(\sqrt{q}).$ If we suppose that $$\sqrt{\sqrt{p+q}\pm\sqrt{q}}\in\mathbb{Q}(\sqrt{q}),$$ then exists unique $a,b \in \mathbb{Q}$ such that $$\sqrt{\sqrt{p+q}\pm\sqrt{q}}=a+b\sqrt{q}.$$ Hence $$\sqrt{p+q}\pm\sqrt{q} = a^2 + qb^2+2ab\sqrt{q},$$ and must be $ab=\pm1/2$ and $\sqrt{p+q} = a^2 + qb^2.$ Solving for $a$ we get that $a$ is a root of the polynomial $$4x^4-4\sqrt{p+q}x^2+q.$$ Hence $a$ have one of the following four values: $$\pm\sqrt{\frac{\sqrt{p+q}}{2}\pm\frac{\sqrt{p}}{2}},$$ but any of these values is a rational, if not, $$\bigg(\pm\sqrt{\frac{\sqrt{p+q}}{2}\pm\frac{\sqrt{p}}{2}}\bigg)^2=\frac{\sqrt{p+q}}{2}\pm\frac{\sqrt{p}}{2} \in \mathbb{Q}.$$

With this we conclude the proof and get the original claim.


The problem I posted some days ago is a special case with $p = 11$ and $q = 5$.

Is this approach correct? I'm interested in reading Galois-type solutions since I think they are more "beautiful". Which are the pair of distinct positive primes whose sum is a perfect square? I see the pairs $(11,5)$, $(23,2)$ and $(31,5)$ for example.

Thaks to everyone.

  • $\begingroup$ Notice that your last question is either (case p=2) equivalent to looking for primes of the form $n^2 -2$, see (math.stackexchange.com/questions/591333/…) or (case pq odd) related to Goldbach's conjecture. In particular Goldbach predicts that for every $n\geq 3$ there are diferent primes $p,q$ such that $4n^2 = p +q$. $\endgroup$ – eduard Jan 31 '19 at 23:50

But this is again a quick application of the "kummerian argument" which I used in my answer to your question of a few days ago. Introduce $k=\mathbf Q(\sqrt p, \sqrt q)$, which is a biquadratic field because $pq$ cannot be a square in $\mathbf Q$ (by unique factorization in $\mathbf Z$) . Consider then the extensions $k(\sqrt {\sqrt {p+q} \pm \sqrt q})$, where $p+q$ is a perfect square. Since $\sqrt {\sqrt {p+q}+\sqrt q} .\sqrt {\sqrt {p+q}-\sqrt q}=p$ is a square in $k^*$, the kummerian argument above $k$ shows that the extensions $k(\sqrt {\sqrt {p+q} \pm \sqrt q})$ are the same field, say $K$. Applying again Kummer over $\mathbf Q(\sqrt q)$ as base field, we see that $K=k=\mathbf Q(\sqrt q)(\sqrt p)$ iff $p(\sqrt {p+q}\pm \sqrt q)$ are squares in $\mathbf Q(\sqrt q)$; multiplying the two relations, we get that $p^3$ is a square in $\mathbf Q(\sqrt q)$ : impossible. Hence $[K:k]=2$ and $[K:\mathbf Q]=8$.

Remark: In the kind of questions you are dealing with, the kummerian approach is more natural in the sense that it appeals only to the multiplicative structure of the fields involved, whereas a blunt direct approach mixes the multiplicative and additive structures.

| cite | improve this answer | |
  • $\begingroup$ Your remark is really useful, it gave me a good intuition for these extensions. Thanks! $\endgroup$ – DrinkingDonuts Feb 5 '19 at 13:26

We have $\sqrt {p+q}=n\in\Bbb N$, so i will use this $n$ below.

Let us consider the tower of fields: $\require{AMScd}$ \begin{CD} {} @. L=\Bbb Q\left(\ \sqrt {n\pm\sqrt q}\ \right)\\ @. @AAA\\ {} @. K=\Bbb Q(\ \sqrt p, \sqrt q\ )\\ @.\nearrow @.\nwarrow\\ \Bbb Q(\sqrt p) @. {} @. \Bbb Q(\sqrt q )\\ @.\nwarrow @.\nearrow\\ {} @. \Bbb Q @.{} \end{CD} Some remarks first:

  • The extension $K=\Bbb Q(\sqrt p,\sqrt q):\Bbb Q$ has degree four, else $\sqrt p$, $\sqrt q$ would differ by a rational factor, but $p\ne q$.

  • The vertical arrow is an extension of fields. First $\sqrt q\in L$, since $n\pm\sqrt q\in L$. Also, because the product of the two numbers $\sqrt{n\pm\sqrt q}$ is $\sqrt{n^2-q}=\sqrt{(p+q)-q}=\sqrt p$, we also have $\sqrt p\in L$.

  • For short, $L= \Bbb Q\left(\ \sqrt {n+\sqrt q},\ \sqrt p,\ \sqrt q\ \right) = \Bbb Q\left(\ \sqrt {n-\sqrt q},\ \sqrt p,\ \sqrt q\ \right) =K(\sqrt {n+\sqrt q}) =K(\sqrt {n-\sqrt q}) $.

It remains to show that the extension $L:K$ has degree two.

If not, then we would have a linear relation over $\Bbb Q$ of the shape: $$ \sqrt{n+\sqrt q}=A+B\sqrt q+\sqrt p(C+D\sqrt q)\in K\ . $$ Apply now the Galois morphism $\sqrt p\to -\sqrt p$, $\sqrt q\to+\sqrt q$ of $K=\Bbb Q(\sqrt p,\sqrt q)$, to get parallely $$ \begin{aligned} \sqrt{n+\sqrt q} &=A+B\sqrt q+\sqrt p(C+D\sqrt q)\in K\ ,\\ \pm \sqrt{n+\sqrt q} &=A+B\sqrt q-\sqrt p(C+D\sqrt q)\in K\ . \end{aligned} $$ (To be pedant and avoid any questions that i may put myself, i added that $\pm$ in the last relation, imposed by the minimal polynomial condition over $K$, the L.H.S being a root of $X^2 -n-\sqrt q\in \Bbb Q(\sqrt q)\ [X]$.)

The representation is unique, so we have either $$ \begin{aligned} \sqrt{n+\sqrt q} &=A+B\sqrt q\ ,\text{ or}\\ \sqrt{n+\sqrt q} &=\sqrt p(C+D\sqrt q)\ . \end{aligned} $$ We use now the other Galois morphism, $\sqrt p\to \sqrt p$, $\sqrt q\to-\sqrt q$, getting either $$ \begin{aligned} \pm\sqrt{n-\sqrt q} &=A-B\sqrt q\ ,\text{ or}\\ \pm\sqrt{n-\sqrt q} &=\sqrt p(C-D\sqrt q)\ . \end{aligned} $$ We multiply, so $\pm \sqrt{n+\sqrt q} \cdot \sqrt{n-\sqrt q}=\pm \sqrt{n^2-q}=\pm\sqrt p$ is either $A^2-qB^2\in\Bbb Q$ or $p(C^2-qD^2)\in \Bbb Q$, thus a contradiction.

The linear relation cannot hold. So the degree of the vertial field extension is two.


| cite | improve this answer | |
  • $\begingroup$ I'm bothered by your Q-linear relation, which involves only $\sqrt p$ and $\sqrt q$, whereas a basis of K/Q is {$1, \sqrt p, \sqrt q, \sqrt {pq}$}. $\endgroup$ – nguyen quang do Feb 1 '19 at 11:42
  • $\begingroup$ yes, thanks, a good point! have to rearrange things! $\endgroup$ – dan_fulea Feb 1 '19 at 14:45

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.