SOME EXAMPLES OF THE GALOIS CORRESPONDENCE

KEITH CONRAD

Example 1. The eld extension Q( 3 2, )/Q, where is a nontrivial cube root of unity, is Galois: it is a splitting eld over Q for X 3 2, which is separable since any irreducible in Q[X ] is separable. The number of eld automorphisms of Q( 3 2, )/Q is [Q( 3 2, ) : Q] = 6. (For comparison, the number of eld automorphisms of Q( 3 2)/Q is 1, even though the 3 3 2 to go in Q ( 2) except to itself.) eld extension has degree 3: there is just nowhere for 3 We will give two ways to think about Gal(Q( 2, )/Q). Q( 3 2, ) Q( 3 2)

Q( )
2

Q For the rst way, each in Gal(Q( 3 2 , )/Q) is determined by its eect on the 3 roots 2 3 2, since these roots generate the top eld over of X 3 2, which are 3 2, 3 2, and the bottom eld (note = 3 2/ 3 2 is a ratio of two cube roots of 2). There are at most 6 permutations of these 3 roots, and since we know there are 6 automorphisms every permutation of the roots comes from an automorphism of the eld extension. Therefore Gal(Q( 3 2, )/Q) = S3 with S3 thought of as the symmetric group on the set of 3 roots of 3 X 2. For another viewpoint, any in the Galois group is determined by the two values ( 3 2) { 3 2, 3 2, 2 3 2} and ( ) {, 2 }. Therefore there are at most 3 2 = 6 possibilities for . Since 6 is the number of automorphisms, all of these possibilities really work: any choice of a root of X 3 2 for ( 3 2) and a nontrivial cube root of unity for ( ) does come from an automorphism . Write ( ) = a where a (Z/(3)) and ( 3 2) = b 3 2 where b Z/(3). For two automorphisms and , ( ( )) = ( a ) = ( )a = a a and 3 3 3 3 3 ( ( 2)) = ( b 2) = ( )b ( 2) = a b b 2 = a b +b 2.

Looking at the exponents of on the right side of these two equations, composition of and b a b a b behaves like multiplication of matrices ( a 0 1 ) with entries in Z/(3), since ( 0 1 )( 0 1 ) =
1

KEITH CONRAD

ab +b ): Gal(Q( 3 2, )/Q) is isomorphic to the group of mod 3 invertible matrices ( a b ) ( aa 01 0 1 b ). by ( a0 1 That we found two dierent models for Gal(Q( 3 2, )/Q), as permutations and as matrices, is no surprise since both of those groups are nonabelian and any two nonabelian groups of size 6 are isomorphic. Example 2. The extension Q( 4 2, i)/Q is Galois by the same reasoning as in the previous example: the top eld is the splitting eld over Q for X 4 2, which is separable. The diagram below shows some intermediate elds, but these are not all the intermediate of the elds. For instance, Q( 2) Q( 4 2), but this is not the only missing subeld. Q( 4 2, i) Q( 4 2) Q(i 4 2)

Q(i)
2

Q Although any element of Gal(Q( 2, i)/Q) permutes the 4 roots of X 4 2, not all 24 permutations of the roots are realized by the Galois group. (This is a contrast to 3 4 4 Gal(Q( 2, )/Q)!) For example, 2 and 2 add to 0, so under a eld automorphism these two roots go to roots which are of also negatives each other. No eld automorphism of Q( 4 2, i)/Q could send 4 2 to i 4 2 and 4 2 to 4 2 because that doesnt respect the 4 4 algebraic relation x + y = 0 which holds for x = 2 and y = 2. 4 To gure out what Gal( Q ( 2 , i ) /Q) is concretely, we think about an automorphism by what it does to 4 2 and i, rather than what it does to all the fourth roots of 2. Since ( 4 2) has to be a root of X 4 2 (4 possible values) and (i) has to be a root of X 2 + 1 4 (2 possible values), there are at most 4 2 = 8 automorphisms of Q( 2)/Q. Because 4 4 [Q( 2, i) : Q] = 8, Gal(Q( 2, i)/Q) has size 8 and therefore all assignments of ( 4 2) and (i) to roots of X 4 2 and X 2 + 1, respectively, must be realized by eld automorphisms. Let r and s be the automorphisms of Q( 4 2, i)/Q determined by 4 4 4 4 r( 2) = i 2, r(i) = i, s( 2) = 2, s(i) = i. By taking powers and products (that is, composites) of automorphisms, we obtain the 4 following table of 8 dierent automorphisms of Q( 2, i)/Q. (They are dierent because they dont have the same eect on both 4 2 and i, which generate the eld extension). 4
2 3 2s id r r r s rs r r3 s 4 4 4 4 4 4 4 4 2 i 2 2 i 2 2 i 2 2 i 2 i i i i i i i i Table 1

( 2) (i)

SOME EXAMPLES OF THE GALOIS CORRESPONDENCE

A calculation at 4 2 and i shows r4 = id, s2 = id, and rs = sr1 , so Gal(Q( 4 2, i)/Q) is isomorphic (not equal, just isomorphic!) to D4 , where D4 can be viewed as the 8 symmetries of the square whose vertices are the four complex roots of X 4 2: r is rotation by 90 degrees counterclockwise and s is complex conjugation, which is a reection across one diagonal of 4 this square. (Strictly speaking, r and s as automorphisms are only dened on Q( 2, i), not on all complex numbers. While r looks like a rotation by 90 degrees on the four roots of X 4 2, it is not really a rotation on most elements of Q( 4 2), since r is not multiplication by i everywhere. For example, r(1) is 1 rather than i, and r(i) is i rather than 1. The 4 function s, however, does coincide with complex conjugation on all of Q( 2, i).) 4 Since Q ( 2 , i ) is a Galois extension of Q, we can compute the degree of a number in Q( 4 2, i) over Q by counting the size of its Galois orbit. For example, let 4

2+

2 + 1.

Applying Gal( Q( 4 2, i)/Q) to and seeing what dierent numbers come out amounts to replacing 4 2 in the expression for by the four dierent fourth roots of 2 and replacing 2 2 = 4 2 in the expression for by the squares of those respective fourth roots of 2. We obtain the list 4 4 4 4 2 + 1, i 2 2 + 1, 2 + 2 + 1, i 2 2 + 1.

2+

of only has size 4. Therefore Although Gal(Q( 4 2, i)/Q) has size 8, the Galois orbit 4 4 2), so Q ( ) Q ( 2), a degree the eld extension Q()/Q has degree 4. Since Q ( 4 comparison implies Q() = Q( 2). It is easy to see why the Galois orbit has fewer than 8 numbers in it: complex conjugation s does not change , so every and s have the same value at . Example 3. The extension Q( 3 2, )/Q has Galois group isomorphic to S3 (Example 1). This group has 3 subgroups of order 2 and one subgroup (just A3 ) of order 3. In the diagram we have indicated the indices in S3 of subgroups.

S3
2 3 3

A3

(12)

(13)

(23)

{(1)}

KEITH CONRAD

Lets ip this upside down, so larger groups are on the bottom. {(1)}

(12)

(13)
3

(23)

A3
2

S3 By the Galois correspondence, the arrangement of subelds of Q( 3 2, ) looks the same, with indices of a subgroup in the Galois group turning into degrees of a subeld over Q. Q( 3 2, )

?
3 3

?
2

Q So there is one quadratic subeld and three cubic subelds. It is easy to write down enough such elds by inspection: Q( ) is quadratic and Q( 3 2), Q( 3 2), and Q( 2 3 2) are all cubic. (These three cubic elds are distinct since two dierent cube roots of 2 cant lie in the same cubic eld.) So these are the only (proper) intermediate elds, and the eld diagram looks like this: Q( 3 2, ) Q( 3 2) Q( 3 2) Q( 2 3 2)

Q( )
2

Q We were somewhat cavalier about the way we just wrote down the cubic elds without really paying attention to which ones should correspond to which subgroups of index 3 (order 2) in the Galois group. But we cant be more careful at this stage (beyond keeping track of indices of subgroups and degrees of subelds) because we didnt really keep track here of how Gal(Q( 3 2, )/Q) is isomorphic to S3 . We simply used the subgroup structure

SOME EXAMPLES OF THE GALOIS CORRESPONDENCE

of S3 to gure out the subeld structure of Q( 3 2, ). If we want to match specic subgroups with specic subelds through the Galois correspondence, we have to think about S3 as the Galois group in a denite way. There are three roots of X 3 2 being permuted by the Galois group (in all 6 possible ways), so if we label these roots abstractly as 1, 2, and 3 then 3 3 3 2 2 as 3. we can see what the correspondence should be. Label 2 as 1, 2 as 2, and 3 Then (12) xes 2 3 2, and therefore Q( 2 3 2) is contained in the xed eld Q ( 2, ) (12) . 3 3 2 2 The subgroup (12) has index 3 and Q( 2)/Q has degree 3, so Q( 2) is the full 3 xed eld of (12) . In a similar way, (13) has xed eld Q( 2) and (23) has xed eld Q( 3 2). So the subgroup and subeld diagrams are aligned if we draw them as follows:

{(1)}

(12)

(13)
3

(23)

A3
2

S3

Q( 3 2, ) Q( 2 3 2) Q( 3 2) Q( 3 2)

Q( )
2

Q Example 4. The extension Q( 4 2, i)/Q has Galois group isomorphic to D4 according to the permutations which the Galois group induces on the fourth roots of 2. Generators are r 4 4 4 and s where r( 2) = i 2, r(i) = i and s( 2) = 4 2, s(i) = i (s is complex conjugation). See Table 1 in Example 2. Below is the diagram of all subgroups of D4 , written upside down.

KEITH CONRAD

{id}

r2 s

r2

rs

r3 s

r2 , s

r2 , rs

D4 All indices of successive subgroups here are 2, so we dont include that information in the diagram. The lattice of intermediate elds in Q( 4 2, i)/Q looks the same: Q( 4 2, i) Q( 4 2) Q(i 4 2) Q( 2) Q( 2, i)

? Q(i 2)

Q(i)

Q To check the elds have been placed correctly according to the Galois correspondence H ; Q( 4 2, i)H , verify in each case that each eld in the eld diagram is xed by the subgroup in the same relative position in the subgroup diagram, and the degree of the eld over Q 4 equals the index of the subgroup over Q: if F Q( 2, i)H and [F : Q] = [D4 : H ] then 4 F = Q( 2, i)H . As an example, the subextension Q(i)/Q has degree 2, so its corresponding subgroup H in D4 has index 2. Since r(i) = i, r is a subgroup xing i with index 8/4 = 2, so H = r . Thus Q(i) corresponds to r . We have left two elds undetermined in the eld diagram. They correspond to the subgroups rs and r3 s , The smallest subgroup properly containing either of these is 4 2 r , rs , so we can gure out what the undetermined elds are by looking for Q( 2, i) of degree 4 over Q that is xed by rs and not by r2 , and likewise nd of degree 4 over Q that is xed by r3 s and not by r2 . Then the two missing elds are Q() and Q( ). To nd , rather than blind guessing we simply write out a general element of Q( 4 2, i) in a basis over Q and see what the condition rs() = means about the coecients. Writing 4 4 2 4 4 3 = a + b 2 + c 2 + d 2 + ei + f i 2 + gi 2 + hi 2 , with rational coecients a, b, c, d, e, f, g, h, applying rs to all terms gives 4 4 2 4 4 3 rs() = a + bi 2 c 2 di 2 ei + f 2 + gi 2 h 2 ,

SOME EXAMPLES OF THE GALOIS CORRESPONDENCE

so b = f, c = c, e = e, d = h. Therefore 4 4 3 4 3 4 = a + b( 2 + i 2) + d( 2 i 2 ) + gi 2. The coecients a, b, d, g can be any rational numbers. To pick something simple of degree 4, we try b = 1 and the other coecients equal to 0: 4 4 4 = 2 + i 2 = (1 + i) 2. Easily r2 () = , so is xed by rs but not by r2 , which means the eld Q() is inside the xed eld of rs but not inside the xed eld of r2 , so Q() must be the xed eld of rs . The dierence = 4 2 i 4 2 is xed by r3 s and not by r2 , so the xed eld 4 3 of r s is (1 i) 2. Now we have a complete eld diagram. Q( 4 2, i) Q( 4 2) Q(i 4 2) Q( 2) Q( 2, i) Q((1 + i) 4 2) Q(i 2) Q((1 i) 4 2)

Q(i)

Q Example 5. The polynomial X 4 X 2 1 is irreducible over Q since it is irreducible mod 3. Lets nd its splitting eld over Q and all of its subelds. The roots of X 4 X 2 1 are (1 + 5)/2 and (1 5)/2. Let = (1 + 5)/2, so (1 5)/2 = i/. Therefore the splitting eld of X 4 X 2 1 over Q is Q(, i). Since is real, i Q(), so as the diagram below illustrates [Q(, i) : Q] = 8. Q(, i)
2

Q()

4 2

Q(i)

Q Any Gal(Q(, i)/Q) is determined by () and (i). Since () has four possible values ( and i/) and (i) has two possible values (i), there are at most eight pairs ( (), (i)) and hence at most 8 possibilities for . The group Gal(Q(, i)/Q) has order 8, so all 8 possible choices for ( (), (i)) really do arise. See Table 2. The fth column is complex conjugation on Q(, i).

KEITH CONRAD

() (i)

i/ i/ i/ i i i i i i i i i Table 2

To help us recognize Gal(Q(, i)/Q), the last two automorphisms in Table 2 have order 4 and the other nonidentity automorphisms in the table have order 2 (check!). The extension Q()/Q is not Galois (after all, has Q-conjugate i/, which is not in Q() since i/ is not real), so Gal(Q(, i)/Q) has a non-normal subgroup and in particular is not abelian. This is enough information to pin down the Galois group up to isomorphism: the two nonabelian groups of order 8 are D4 and Q8 , and every subgroup of Q8 is normal, so Gal(Q(, i)/Q) = D4 . To make this isomorphism concrete, let r be the automorphism with the eect in the second to last column of Table 2 (it has order 4) and let s be complex conjugation on Q(, i). Then we can list the automorphisms described in Table 2 as in Table 3. As an exercise, check from Table 3 that sr = r3 s.

() (i)

id r2 rs r3 s s r2 s r r3 i/ i/ i/ i i i i i i i i i Table 3

Below is the lattice of subgroups of D4 , upside down. {id} Q(, i)

r2 s

r2

rs

r3 s

r2 , s

r2 , rs

D4

The eld xed by s is a real subeld of Q(, i) whose degree over Q is 8/2 = 4. This eld must be Q(), since it has degree 4 and is a real eld. From Table 3, i is xed by {1, r2 , rs, r2 s} = r2 , rs , so the eld xed by r2 , rs , which must be quadratic, is Q(i). From the diagram of subgroups of D4 , there is a unique quadratic subeld of Q() on account of there being a unique subgroup of D4 containing s with index 2, namely r2 , s . An obvious quadratic subeld of Q() is Q(2 ) = Q((1 + 5)/2) = Q( 5), so this is the xed eld of r2 , s .

SOME EXAMPLES OF THE GALOIS CORRESPONDENCE

Here is a diagram of subelds of Q(, i) so far. Q(, i)

Q()

Q(2 )

Q(i)

Q Using Table 3, i is xed by r2 s, and i has degree 4 over Q (its a root of X 4 + X 2 1, which is irreducible mod 3 and thus irreducible over Q). Here is a more lled-in subeld diagram. Check for each number listed in the diagram that its xed group is the corresponding subgroup in the subgroup diagram for D4 . Q(, i)

Q()

Q(i)

Q(2 , i)

Q(2 )

Q(i2 )

Q(i)

Q To complete the eld diagram we seek elements of degree 4 over Q that are xed by rs and r3 s. Since both of these automorphisms have order 2, its natural to consider + (rs)() = + i/ and + (r3 s)() = i/. To prove + i/ generates the xed eld of rs, lets use the eld diagram: Q( + i/) is inside the xed eld of rs, so if it does not have degree 4 over Q then this eld is inside Q(i) and thus is xed by r2 . Since r2 ( + i/) = i/ = ( + i/), the only way + i/ can be xed by r2 is if it is 0, but this would be absurd since is a real number. So the rst question mark in the above diagram is Q( + i/). In a similar way, the eld xed by r3 s is Q( i/). We can make the generator for the eld Q(+i/) more explicit. Since = (1 + 5)/2, by direct calculation i +
2

1 1+ 5 51 = + 2i 2 = + 2i = 1 + 2i, 2 2
2

2 and likewise ( i/) = 1 2i. Therefore Q( + i/) = Q( 1 + 2i) and Q( i/) = Q( 1 2i). Here is the eld diagram with more explicit generators of the elds.

10

KEITH CONRAD

Q(, i) Q( 5, i) Q( 5) Q( 1 + 2i) Q( 1 2i)

Q()

Q(i) Q( 5)

Q(i)

Q Galois theory tells us that Q( 1 + 2i) = Q( 1 2i) because these elds correspond to dierent subgroups of Gal( Q ( , i ) / Q ). Since s ( + i/ ) = i/ , the eld Q ( 1 + 2 i ) is carried over to Q( 1 2i) by complex conjugation.