Let $D_4$ be the dihedral group of order 8. I want to find a field $K/\mathbb Q$ such that $K\supset \mathbb Q(\sqrt 2,\sqrt 3)$ and $Gal(K/\mathbb Q)=D_4$. I know the Galois group of $\mathbb Q(\sqrt 2,\sqrt 3)$ is $C_2^2$ and $D_4$ has a normal subgroup $C_2$ whose quotient is $C_2^2$. So I should be able to find such a field $K$. It should be a quadratic extension of $\mathbb Q(\sqrt 2,\sqrt 3)$. But in practice I cannot find it. I tried to look at splitting field of $\sqrt{\sqrt2+\sqrt3}$, which has Galois group $C_2\times D_4$. But do not know how to find an intermediate field with Galois group $D_4$.
-
In your convention, how many elements are in $D_4$? – J. W. Tanner Jan 09 '25 at 05:08
-
1@J.W.Tanner From what OP has written, clearly $8$. – Ted Shifrin Jan 09 '25 at 05:24
-
@TedShifrin: yes, except that, when I commented, before OP edited, it said "Let $D_4$ be the dihedral group of order $\color{blue}4$" rather than "of order $8$" – J. W. Tanner Jan 09 '25 at 12:05
3 Answers
This is a problem well-suited to the L-functions and modular forms database (LMFDB). The LMFDB label of $\newcommand{\Q}{\mathbb{Q}} \Q(\sqrt{2}, \sqrt{3})$ is 4.4.2304.1, and the image of the embedding of $D_4 \hookrightarrow S_8$ has transitive group label 8T4. Searching the LMFDB for number fields with the given Galois group and intermediate field, we find 50 matches:
https://www.lmfdb.org/NumberField/?galois_group=8T4&subfield=4.4.2304.1
The first such example is the field 8.8.3057647616.1, the number field defined by the irreducible polynomial
$$
x^8 - 4 x^7 - 4 x^6 + 20 x^5 + 4 x^4 - 20 x^3 - 4 x^2 + 4 x + 1 \, .
$$
- 20,897
-
1Cool. An easier one is the splitting field of $x^4 - 6x^2 + 6 $, i.e., $\mathbb Q(\sqrt{3+\sqrt3},\sqrt{3-\sqrt3})$. Thanks! – JKDASF Jan 09 '25 at 06:44
-
@JKDASF In my opinion that example deserves to be fleshed out in the form of an answer. – Jyrki Lahtonen Jan 09 '25 at 06:47
-
1@JKDASF I believe that example (whose splitting field is given by the polynomial $x^8 - 60 x^6 + 1104 x^4 - 6120 x^2 + 36$) is isomorphic to the one given in my answer. – Viktor Vaughn Jan 09 '25 at 06:54
With the help of LMFDB as Viktor Vaughn suggested, here is an explicit example:
$\mathbb Q(\sqrt{3+\sqrt3},\sqrt{3-\sqrt3})\supset\mathbb Q(\sqrt2, \sqrt3)\supset\mathbb Q$
which is the splitting field of $x^4−6x^2+6$, and isomorphic to the field in Viktor Vaughn's answer.
- 662
-
1
-
1
-
Thanks. I see that $\sqrt6=\sqrt{3+\sqrt3}\sqrt{3-\sqrt3}$ and $\sqrt3=\sqrt{3+\sqrt3}^2-3$ are in the field you gave, so their ratio $\sqrt2$ is too – J. W. Tanner Jan 10 '25 at 02:58
-
My knowledge of Galois theory is rusty. Can you explain how the splitting field of a degree $4$ polynomial has degree $8$? I know the splitting field of $x^3−2$ has degree $6$ since it has a complex cube root of unity, but I don’t see it in the case at hand – J. W. Tanner Jan 10 '25 at 04:00
-
@J.W.Tanner The splitting field is generated by one element, that element has minimal polynomial of degree 8, and is not one of the root of the quartic polynomial – JKDASF Jan 10 '25 at 05:31
-
@J.W.Tanner You may want to look at this older thread characterizing the possible Galois groups of irreducible biquadratics. – Jyrki Lahtonen Jan 10 '25 at 05:39
-
Maybe that primitive element could be $\sqrt2+\sqrt{3+\sqrt3}$; thanks @JyrkiLahtonen – J. W. Tanner Jan 10 '25 at 05:42
First consider $$L = ℚ(\sqrt{2}, \sqrt{3}, \sqrt{\sqrt{2}+\sqrt{3}})$$
Over ℚ, the element $\sqrt{2}+\sqrt{3}$ has various Galois conjugates obtained by sending $\sqrt{2} \mapsto \pm\sqrt{2}, \sqrt{3} \mapsto \pm\sqrt{3}$
Once you adjoin $\sqrt{2}$ and $\sqrt{3}$, the polynomial $$x^2-(\sqrt{2}+\sqrt{3})$$ is irreducible over $ℚ(\sqrt{2}, \sqrt{3})$, so adjoining $\sqrt{\sqrt{2}+\sqrt{3}}$ gives a genuine quadratic extension of $ℚ(\sqrt{2}, \sqrt{3})$.
This is not Galois over $ℚ(\sqrt{2}, \sqrt{3})$, so we also adjoin $\sqrt{\sqrt{2}-\sqrt{3}}$, forming $L = ℚ(\sqrt{2}, \sqrt{3}, \sqrt{\sqrt{2}+\sqrt{3}}, \sqrt{\sqrt{2}-\sqrt{3}})$.
A standard calculation then shows [$L:ℚ$] = 16. The full Galois group $Gal(L/ℚ)$ is isomorphic to $C_2 \times D_4$ (a group of order 16).
Since $Gal(L/ℚ)$ $\cong C_2 \times D_4$, it contains a normal subgroup $$N \cong C_2$$ lying entirely in the extra $C_2$-factor. This 'extra' $C_2$ is often generated by an automorphism $\tau$ which does something like $$\tau: \sqrt{\sqrt{2}+\sqrt{3}} \mapsto -\sqrt{\sqrt{2}+\sqrt{3}}, \sqrt{\sqrt{2}-\sqrt{3}} \mapsto -\sqrt{\sqrt{2}-\sqrt{3}}$$ while fixing $\sqrt{2}$ and $\sqrt{3}$. Being a subgroup of index 2, the quotient group $Gal(L/ℚ)/N$ has order 8. One checks (from how all the other generators act) that this quotient of order 8 is the dihedral group $D_4$.
By the Galois correspondence, if $N \subset Gal(L/ℚ)$ is a normal subgroup of index 2, then its fixed field $K = L^N$ is an intermediate field between ℚ and $L$, and $Gal(K/ℚ) \cong Gal(L/ℚ)/N \cong D_4$.
Hence, the field you want is exactly $$K = L^{\langle \tau \rangle} = ℚ(\sqrt{2}, \sqrt{3}, \sqrt{\sqrt{2}+\sqrt{3}}, \sqrt{\sqrt{2}-\sqrt{3}})^{\langle \tau \rangle}$$
- 624
-
This idea will work, if you have the right $\tau$. I'm just trying to make $K$ more concrete in my head. Is it clear that your automorphism $\tau$ is in the center of $Gal(L/\Bbb{Q})$? – Jyrki Lahtonen Jan 09 '25 at 07:13
-
yes, $\tau$ is in the center because it doesn't touch $\sqrt{2}$ or $\sqrt{3}$. Conjugating $\tau$ by any element that does send $\sqrt{2} \mapsto -\sqrt{2}$ or $\sqrt{3} \mapsto -\sqrt{3}$ flips or unflips the same pair of roots, so $\tau$ commutes with all those actions. Consequently, it is in the center. – Lorenzo Jan 09 '25 at 07:25
-
Does it also commute with the two automorphisms that map $\sqrt{\sqrt2+\sqrt3}\mapsto \sqrt{\sqrt2-\sqrt3}$? The center is isomorphic to Klein four (because $D_4$ also has a non-trivial center), so that doesn't resolve my main doubt which is the need to see your statement this extra C_2 is often generated by... fleshed out a bit. Your mileage may be different, but I haven't seen this ever before, so often is meaningless to me here :-) – Jyrki Lahtonen Jan 09 '25 at 07:31
-
Yes, if you analyze the actions of $\tau$. It only flips the signs of $\sqrt{\sqrt{2} \pm \sqrt{3}}$, it does not alter $\sqrt{2}$ or $\sqrt{3}$. Any automorphism that sends $\sqrt{\sqrt{2} + \sqrt{3}} \mapsto \sqrt{\sqrt{2} - \sqrt{3}}$ necessarily does something nontrivial to $\sqrt{3}$. When you compose those two types of actions (regardless of order), the net effect on each of the four generators ($\sqrt{2}, \sqrt{3}, \sqrt{\sqrt{2} + \sqrt{3}}, \sqrt{\sqrt{2} - \sqrt{3}}$) is the same, so they commute. – Lorenzo Jan 09 '25 at 07:42
-
As for saying the extra $C_2$ is often generated by $\tau$, that's just a standard way of constructing the degree-16 splitting field. – Lorenzo Jan 09 '25 at 07:43
-
I'm afraid I'm still not totally satisfied. I came back to this problem, and it seems to me that with $u=\sqrt{\sqrt2+\sqrt3}$ and $v=\sqrt{\sqrt2-\sqrt3}$ we get $uv=i$ (with suitable choices of branches for the square roots). So if $\tau(u)=-u$ and $\tau(v)=-v$, it follows that $\tau(i)=i$, and $\sqrt2\pm\sqrt3$ are also fixed points of $\tau$. Therefore the fixed field of $\tau$ is $\Bbb{Q}(i,\sqrt2,\sqrt3)$ but this has an elementary abelian Galois group rather than $D_4$. – Jyrki Lahtonen Jan 09 '25 at 21:55
-
I looked at this one more time now that I woke up. I cannot make this work. The non-trivial automorphisms of the center are $\tau_1:u\mapsto -u, i\mapsto i$, $\tau_2:u\mapsto -vi, i\mapsto -i$ and $\tau_3:u\mapsto vi, i\mapsto -i$. Of these only $\tau_1$ fixes $u^2$ and hence $\sqrt2$ and $\sqrt3$. But the fixed field of $\tau_1$ is the elementary 2-abelian extension $\Bbb{Q}(i,\sqrt2,\sqrt3)$. Can you explain this in detail, please? – Jyrki Lahtonen Jan 10 '25 at 05:30
-
So I suspect that your choice of $\tau$ is the non-trivial element in the center of the $D_4$ part. Simply because it is the square of an element of order four. On the other hand, the subgroup $N$ is not generated by a square of an automorphism. Such automorphisms exist, but won't have $\sqrt2\pm\sqrt3$ as fixed points. – Jyrki Lahtonen Jan 10 '25 at 06:01
-
You are seeing a 'wrong' central element because you labeled $uv = i$. If you choose your square root branches differently, the same map $\tau$ will send $i \mapsto -i$ Then it is no longer central in $D_4$, and its fixed field is the nonabelian (dihedral) extension you want. – Lorenzo Jan 11 '25 at 06:12
-
1Excuse me! Your definition of $\tau$ maps both $\sqrt{\sqrt2+\sqrt3}$ and $\sqrt{\sqrt2-\sqrt3}$ to their negatives. Yet you claim that their product is not fixed by $\tau$. Can you please make the answer explicit rather than using vague claims. List generating elements of the field. – Jyrki Lahtonen Jan 11 '25 at 06:49
-
Let $x := \sqrt{\sqrt{2}+\sqrt{3}}, y := \sqrt{\sqrt{2}-\sqrt{3}}$. Then L = $ℚ(\sqrt{2}, \sqrt{3}, x, y)$ is the degree-16 extension of ℚ. $x^2 = \sqrt{2}+\sqrt{3}$ and $y^2 = \sqrt{2} - \sqrt{3}$, so $x^2$ and $y^2$ both lie in $ℚ(\sqrt{2}, \sqrt{3})$. Moreover, $(xy)^2 = -1$, so $xy$ is an imaginary unit $i$, but the sign is just a labeling convention. – Lorenzo Jan 11 '25 at 06:58
-
Yes, but irrespective of choice of sign the product $xy$ is a fixed point of $\tau$, so $i$ is in the fixed field regardless. The other elements in the center of the Galois mix $x$ and $y$ (up to a power of $i$), so they won't have $x^2$ in the fixed field. – Jyrki Lahtonen Jan 11 '25 at 07:05
-
What if we define the automorphism as $\tau: \sqrt{2} \mapsto \sqrt{2}, \sqrt{3} \mapsto \sqrt{3}, x \mapsto x, y \mapsto -y$ and label $xy = -y$, so that $xy$ is not fixed by $\tau$? – Lorenzo Jan 11 '25 at 07:12
-
So the automorphism is not central in $D_4$, and its fixed field is nonabelian (dihedral). – Lorenzo Jan 11 '25 at 07:13
-
I listed the automorphisms in the center a few comments up. The fixed field is not Galois unless $\langle\tau\rangle$ is a normal subgroup. Given that $\tau$ has order two, that can be the case only when $\tau$ is in the center of $C_2\times D_4$. Mind you, you only need to specify the images of $x$ and $i$ as the splitting field is $\Bbb{Q}(x,i)$. The possible images of $x$ are $xi^k$ and $yi^k$, $k=0,1,2,3$. Sixteen total (accounting for two choices for the image of $i$), so all combinations occur. – Jyrki Lahtonen Jan 11 '25 at 07:18
-
You may be right -- I don't see a point to extensively discussing semantics. There is no canonically correct sign here, so it really comes down to the way the branches are chosen and labeled. – Lorenzo Jan 11 '25 at 07:26
-