# Proving $\sqrt{a+\sqrt{b}}=\sqrt{m}+\sqrt{n}\iff a^{2}-b$ is a square

This is an exercise for the book Abstract Algebra by Dummit and Foote (pg. 530):

Let $F$ be a field of characteristic $\neq2$ . Let $a,b\in F$ with $b$ not a square in $F$. Prove $\sqrt{a+\sqrt{b}}=\sqrt{m}+\sqrt{n}$ for some $m,n\in F$ iff $a^{2}-b$ is a square in $F$.

I am having problem proving this claim, I tried to assume $\sqrt{a+\sqrt{b}}=\sqrt{m}+\sqrt{n}$ and I naturally squared both sides, to try and get $a^{2}$ I squared both sides again and then reduced $2b$ from both sides and rearranged to get $$a^{2}-b=(m+n+2\sqrt{mn})^{2}-2\sqrt{b}(a+\sqrt{b})$$ but I don't see how I can use it.

-
You have "not a square" in the title but your question asks the opposite. Is this intentional? –  Alex R. Sep 9 '12 at 20:01
@Sam - No, that was a typo. thanks for pointing this out to me! –  Belgi Sep 9 '12 at 20:03
Not sure if this will help, but maybe use $$a^2-b = (a+\sqrt{b})(a-\sqrt{b}) = x(x-2\sqrt{b})$$ for $x = m+n+2\sqrt{mn}$? –  gt6989b Sep 9 '12 at 20:16

$$\sqrt{a+\sqrt b}=\sqrt m + \sqrt n \Rightarrow a+\sqrt b = m+n+2\sqrt{mn}$$ Since $\phi(\alpha+\beta\sqrt b)=\alpha-\beta\sqrt b$ for $\alpha,\beta\in F$ defines an automorphism $\phi\colon F[\sqrt b]\to F[\sqrt b]$ that leaves $F$ fixed, we have that $\phi(\sqrt{mn})= \pm\sqrt{mn}$ because $\phi$ maps the polynomial $X^2-mn$ to itself and can at most interchange its roots. Thus we additionally get $a-\sqrt b=\phi(a+\sqrt b)=\phi(m+n+2\sqrt{mn})$, i.e. $$a-\sqrt b=m+n\pm2\sqrt{mn}.$$ Since $\sqrt b\ne -\sqrt b$ (characteristic $\ne 2$), the left hand sides differ, hence so do the right hand soides, hence "$\pm$" is really "$-$". By adding and subtracting these equations we find that $a=m+n$ and $\sqrt b =2\sqrt{mn}$. Hence $m,n$ are roots of $0=X^2-(m+n)X+mn=X^2-a X+\frac b4$ and can be found as $$\frac{a\pm\sqrt{a^2-b}}2$$ More precisely:

• If $a^2-b$ is a square, this actually produces $m,n\in F$ with the property that $(\sqrt m +\sqrt n)^2=a+\sqrt b$, i.e. $\sqrt m + \sqrt n$ is a root of $X^2-(a+\sqrt b)$ as desired.
• If $a^2-b$ is not a square, no solutions for $m,n$ exist in $F$.
-
Can you please explain "It can only map $\sqrt{mn}\mapsto \pm\sqrt{mn}$." ? I don't understand why –  Belgi Sep 9 '12 at 20:50
$\sqrt{mn}$ is a root of $X^2-mn\in F[X]$. Every automorphism of $F[\sqrt b]$ over $F$ leaves $F$ and hence this polynomial fixed. It can at most interchange the roots, which means flip sign for square roots. –  Hagen von Eitzen Sep 9 '12 at 20:53
I don't think that every automorphism of $F[\sqrt{b}]$ leaves $F$. for example take $F=\mathbb{Q}(\sqrt{2})$ and $b=3$ then you know that there is an automorphism of $F[\sqrt{b}]$ that takes $\sqrt{2}$ to $-\sqrt{2}$ –  Belgi Sep 9 '12 at 20:56
Yes. But every automorphism of $F[\sqrt b]$ over $F$. In fact, in this simple case it can be written explicitly as $\phi(\alpha +\beta\sqrt b)=\alpha-\beta\sqrt b$ for $\alpha,\beta\in F$. –  Hagen von Eitzen Sep 9 '12 at 21:02
sorry I didn't see it's over $F$. How do we get the equality after "Thus we additionally get" ? I know that the LHS it mapped to the RHS but that does not mean their equal... –  Belgi Sep 9 '12 at 21:05

$$\sqrt{a+\sqrt{b}} = \sqrt{m} + \sqrt{n}$$ $$a+\sqrt{b} = m+n+2\sqrt{mn}$$ Thus $a = m+n$ and $b = 4mn$ as $b$ is not a square . Finally, $$a^2 -b = m^2 + 2mn + n^2 -4mn = (m-n)^2$$.

-
I don't understand your "Thus"...maybe $\sqrt{b}=n+2\sqrt{mn}$.. –  Belgi Sep 9 '12 at 20:47
Say $\sqrt{b} = c + 2\sqrt{mn}$. Note that $a,b,m,n, (c=m+n-a) \in F$ but $\sqrt{mn}, \sqrt{b} \notin F$. After squaring you get $b = c^2 + 4mn + 4c\sqrt{mn}$ where $\text{rhs} \notin F$. –  Karolis Juodelė Sep 9 '12 at 20:58
Why $\sqrt{mn}\not\in F$ ? what if $m=n=2$ ? –  Belgi Sep 9 '12 at 21:00
Then also $a+\sqrt{b} \in F$ (follows from my second equation), which we know not to be the case. –  Karolis Juodelė Sep 10 '12 at 3:58

Hint $\rm\,\ \left[\sqrt{a+\sqrt b}\,=\,\sqrt m + \sqrt n\right]^2\! \Rightarrow\: a+\sqrt b \,=\, m\!+\!n+2\sqrt{mn}\in F[\sqrt{b}].\:$ Thus, taking "norms" $\rm\:a^2 - b\, =\, (m+n)^2 - 4mn = (m-n)^2,\:$ where norm = constant term of minimal polynomial.

-
How did you calculate those norms ? –  Belgi Sep 9 '12 at 22:55
@Belgi Compare their monic minimal polynomials $\rm\: (x-\alpha)(x-\alpha') = x^2 - (\alpha+\alpha')\, x + \alpha\alpha'.$ –  Bill Dubuque Sep 9 '12 at 23:09
Should I find the minimal polynomial of $a+\sqrt{b}$ to calculate its norm ? –  Belgi Sep 9 '12 at 23:12
@Belgi The monic minimal polynomial is unique (else subtracting two different ones gives a smaller deg poly, contra min). Thus both sides have the same monic minimal polynomial, so the same constant coefficient - which is the (formal) norm. –  Bill Dubuque Sep 9 '12 at 23:17
I know that this polynomial is unique, but I don't understand what you calculated to get the quality after the word - " "norms" " –  Belgi Sep 9 '12 at 23:19

Let $F$ be a field of charcteristic different from 2. Let $a$ and $b$ be elements of the field $F$ with $b$ not a square in F. Prove that a necessary and sufficient condition for $\sqrt{a+\sqrt{b}}={\sqrt{m}+\sqrt{n}}$ for $m,n\in F$ is that $a^2-b$ is a square in $F.$

Solution. $\Rightarrow:$ Suppose that $a^2-b$ is a square in $F$. Then $\sqrt{a^2-b}\in F.$ Let $$m= \frac{a+\sqrt{a^2-b}}{2}$$ and $$n= \frac{a-\sqrt{a^2-b}}{2}.$$ Then $n,m\in F$ because $\textrm{char}\, F\neq 0.$

$\Leftarrow:$ Now $$m =\frac{a+\sqrt{a^2-b}}{2} = \frac{(a+\sqrt{b})+2\sqrt{a^2-b}+(a-\sqrt{b})}{4} = \left( \frac{\sqrt{a+\sqrt{b}}+\sqrt{a-\sqrt{b}}}{2} \right)^2,$$ this means that $$\sqrt{m}=\frac{\sqrt{a+\sqrt{b}}+\sqrt{a-\sqrt{b}}}{2}.$$

Also $$n =\frac{a-\sqrt{a^2-b}}{2} = \frac{(a+\sqrt{b})-2\sqrt{a^2-b}+(a-\sqrt{b})}{4} = \left( \frac{\sqrt{a+\sqrt{b}}-\sqrt{a-\sqrt{b}}}{2} \right)^2,$$ this means that $$\sqrt{n}=\frac{\sqrt{a+\sqrt{b}}-\sqrt{a-\sqrt{b}}}{2}.$$

Thus $$\sqrt{m}+\sqrt{n} =\frac{\sqrt{a+\sqrt{b}}+\sqrt{a-\sqrt{b}}}{2}+\frac{\sqrt{a+\sqrt{b}}-\sqrt{a-\sqrt{b}}}{2} = \sqrt{a+\sqrt{b}}.$$

-