ALGEBRA 2 HONORS: GALOIS THEORY

DAVID SMYTH

1. Polynomial Equations: High School Approach

1.1. Solving polynomial equations. Most of modern algebra was constructed in order to
come to grips with the following problem: Given a polynomial
f (x) = a0 xn + a1 xn−1 + . . . + an−1 x + an ,
how can we write down a number α such that f (α) = 0. For concreteness, let’s think of
a0 , . . . , an as rational numbers.

Linear Equations (n = 1). I know that you know how to solve a linear equation, but humor
me. Consider the equation:
f (x) = a0 x + a1 = 0.

We can divide through by a0 , then subtract a1 /a0 from both sides.

a1
x+ =0
a0
a1
x=−
a0
Even though this is completely trivial, I would like to make two observations.

• If a0 and a1 are rational numbers, then our solution −a1 /a0 is still a rational num-
ber. Similarly, if a0 and a1 are real numbers, then our solution is still a real number.
You probably remember from high school that simple quadratic polynomials sometimes
have complex solutions, but nothing like that happens here. In modern terminology, we
can say that no field extension is necessary in order to find solutions of linear equations.

• Our first bit of algebra - dividing through by the leading coefficient a0 - is actually a
completely general recipe for reducing an arbitrary polynomial to a monic polynomial,
i.e. a polynomial with a0 = 1. If we can solve monic polynomials, we can solve all poly-
nomials. Thus, from now on, I’ll simply assume our polynomial is monic to begin with.

Quadratic Equations (n = 2).

f (x) = x2 + a1 x + a2 = 0
The basic trick here, probably known to most of you, is to make a substitution which cancels
the a1 -term:
a1
x=y−
2
1
2 DAVID SMYTH

If we simply plug this in, we get

 a1 2 a1
y− + a1 (y − ) + a2 = 0
2  2
2 2

a a
y 2 − a1 y + 1 + a1 y − 1 + a2 = 0
4 2
a21
 
2
y + a2 − =0
4
s
a21

y=± − a2
4

Now that we have found y, we can go back and find x. We get

s
a21

a1
x=− ± − a2
2 4

Once again, I would like to make two observations.

• It is no longer true that if our coefficients are rational, then we will necessarily have a
rational solution. On the otherrhand, we can see precisely what we need to “add” in
 2 
a1
order to get solutions, namely 4 − a2 . This is the only piece of the solution that
might not be defined over the original field of definition. In modern terminology, we’ll
say that you always have a solution after passing to a degree two extension.
• Once again, the key trick here generalizes in an obvious way. Using a linear substi-
tution of the form x = y − a1 /n, we can reduce an arbitrary degree-n equation to an
equation satisfying a1 = 0. We are making progress! Perhaps with enough tricks, we
can solve any polynomial.

Cubic Equations (n = 3). Applying our previous two tricks, we may assume that we have an
equation of the form:
f (x) = x3 + a2 x + a3 = 0
Any ideas? As with the quadratic equations, it’s easy to see that if we could cancel the a2 -
term we’d be in business - we could simply take cube roots. The new trick is to recognize the
possibility of a non-linear substitution:
a2
x=y−
3y
Let’s see what happens when we plug this in:

a2 3
   
a2
y− + a2 y − + a3 = 0
3y 3y
a2 a3 a22
   
y 3 − a2 y + 2 − 2 3 + a2 y − + a3 = 0.
3y 27y 3y
a3
y 3 + a3 − 2 3 = 0
27y
 3

Clearing denominators, we get:

a32
y 6 + a3 y 3 −
= 0.
27
In most cases, going from a degree 3 equation to a degree 6 equation would not be consid-
ered progress. But here (rather miraculously), we can recognize this equation as a quadratic
equation in y 3 , i.e. we can rewrite it as:
a32
(y 3 )2 + a3 (y 3 ) − = 0.
27
Thus, we can solve for y 3 using the quadratic formula:
r
3 a3 a23 a3
y =− ± + 2.
2 4 27
Taking cube roots, we get:
s r 
3 a3 a3 2  a2 3
y= − ± + .
2 2 3
Now we should be a little careful here - just as one needs to account for positive and minus
square roots, one needs to account for all possible cube roots. Thus, one really has the following
solutions: s r 
i 3 a3 a3 2  a2 3
y=ω − ± + , i = 0, 1, 2,
2 2 3
where ω = e2πi/3 is a third root of unity. (The fact that we are suddenly using complex
numbers here is a little unsettling - eventually we will understand roots of unity in a purely
algebraic way without any complex numbers entering the picture.) Plugging these solutions
a2
back into the formula x = y − 3y , we obtain solutions of our original equation.
Optional Aside. You might notice something a little fishy here: We apparently have six
possible solutions for y, each of which should give a solution for x. But this would give six
possible solutions for our original cubic! In fact, there is no contradiction here. With some
elementary but tedious algebra one can check that these six different value y, break into three
pairs, with each pair giving the same value for x. The final solutions for x are:
s r  s r 
3 a3 a3 2  a2  3 3 a3 a3 2  a2 3
− + + − − − +
2 2 3 2 2 3
s r  s r
3 a3 a3 2  a2 3 2 3 a3  a 2  a  3
3 2
ω − + + −ω − − +
2 2 3 2 2 3
s r  s r 
2 3 a3 a3 2  a2  3 3 a3 a3 2  a2 3
ω − + + −ω − − +
2 2 3 2 2 3

What lessons can we draw from the cubic? The good news is that we are beginning to
see what it means to “write down a number α such that f (α) = 0.” What we would like to
do is find some formula for α in terms of the given coefficients a0 , a1 , . . . , an . Based on the
pattern we’ve seen so far, we would expect this formula to involve nothing more than the usual
operations of addition/subtraction, multiplication/division, and also taking roots.
4 DAVID SMYTH

But there’s also some bad news:

• The algebraic complexity of this problem is rapidly ballooning.
• The trick we used with the cubic does not seem to generalize in the way that our first
two tricks did. The first two tricks serve to reduce us to the equation.
x4 + a2 x2 + a3 x + a4
But it is not at all clear whether there is a substitution of the form x = f (y, 1/y) which
puts this in a simpler form.
Evidently, there are problems with the high school approach - these algebraic tricks feel
unmotivated and to check that they work one has to do many unilluminating calculations. We
need a different perspective on this problem. The basic shift in emphasis is to recognize the
connection between finding roots and factoring polynomials.
1.2. From finding roots to factoring. To see the connection between finding roots and
factoring the polynomial, we begin with the following easy lemma. It says that finding a root
α of f (x) is the same as factoring f (x) into (x − α) and a lower factor.
Lemma 1.1 (Remainder Theorem). Let k be a field, and let f (x) ∈ k[x] be a polynomial. For
any α ∈ k, we can write
f (x) = (x − α)g(x) + f (α),
where g(x) is a polynomial of degree n − 1. In particular if f (α) = 0, then f (x) admits a
factorization as (x − α)g(x).
Proof. Using the usual division algorithm for polynomials, just divide f (x) by (x − α). We
will get
f (x) = (x − α)g(x) + c
where c ∈ k is some constant. By plugging α into both sides of this equation, we see that
c = f (α). 
If we use this factoring procedure inductively, we get two useful corollaries.
Corollary 1.2. If f (x) ∈ k[x] is a degree n polynomial, then f has at most n roots.
Proof. If f has no roots, then there is nothing to prove, so we may assume that f has a root
α. By the Remainder Theorem, we may factor f as
f (x) = (x − α)g(x).
By induction on the degree of f , we may assume that g has no more than n − 1 roots. Since
any root of f must be either a root of (x − α) (namely α) or a root of g(x), it follows that
f (x) has no more than n roots. 
Corollary 1.3. If f (x) ∈ k[x] is a degree n polynomial with n distinct roots α1 , . . . , αn ∈ k,
then f can be factored as:
n
Y
f (x) = (x − αi )
i=1

Proof. We can factor f (x) as:

f (x) = (x − α1 )g(x).
Since the roots are distinct (αi − α1 ) 6= 0 for all i = 2, . . . , n. Thus, αQ
2 , α3 , . . . , αn must be
roots of g(x). By induction on the degree of f , we may assume g(x) = ni=2 (x − αi ), and the
desired result follows. 
 5

Now let us assume for the time being that f (x) actually has n distinct roots, so that we can
factor
Yn
f (x) = (x − αi )
i=1

Then we can view the roots αi as variables, and the coefficients of the polynomial as giving
equations involving these variables. At least in the case n = 2, this idea should be familiar
from high school, i.e. if we write

(x − α1 )(x − α2 ) = x2 + a1 x + a2 ,

where a1 and a2 are given to begin with, then we see the the problem of finding the roots of
f is just the same as finding two numbers α1 , α2 such that

−(α1 + α2 ) = a1
α1 α2 = a2 .

In the next lecture, we will generalize this system of equations to higher n.

2. Symmetric Functions
Definition 2.1 (Elementary Symmetric Functions). Let k[x1 , . . . , xn ] be a polynomial ring in
n variables. For i = 1, . . . , n, we define the following special polynomials si ∈ k[x1 , . . . , xn ]:

s 1 = x1 + x2 + · · · + xn
s2 = x1 x2 + x1 x3 + . . . + xn−1 xn
..
.
X
sk = xi1 xi2 . . . xik
1≤i1 ≤i2 ≤...≤ik ≤n
..
.
sn = x1 x2 · · · xn

In words, we can say that the k th symmetric function is simply the sum of all degree k
monomials with no repeated variables.
The point of this definition is that the functions si precisely encode the relationship between
the roots of a polynomial and its coefficients. By some straight-forward high school algebra,
you can check:
n
Y
(x − αi ) = xn − s1 (α1 , . . . , αn )xn−1 + s2 (α1 , . . . , αn )xn−2 − . . . + (−1)n sn (α1 , . . . , αn ),
i=1

This means that finding the roots of a given polynomial f (x) = xn + a1 xn−1 + . . . + an (at
least under the assumption that f (x) has n distinct roots - later we’ll see this assumption is
6 DAVID SMYTH

unnecessary) is precisely equivalent to finding α1 , . . . , αn which satisfy the following equations.

α1 + . . . + αn = a1
α1 α2 + . . . + αn−1 αn = −a2
..
.
α1 α2 · · · αn = (−1)n an
In these equations, you should think of ai as given, and αi as being unknown numbers that
you are trying to find.
We started off with a single equation in one variable, and now we have n equations in n
variables. How on earth could this be any easier than the original problem? The point is
that these are not just any random old equations; the elementary symmetric functions have
very special properties that will make them easier to work with than arbitrary functions. For
starters, they are symmetric. What exactly does that mean?
Definition 2.2. Let Sn act on k[x1 , . . . , xn ] by permuting the variables, i.e. σ(f (x1 , . . . , xn )) =
f (xσ (1), . . . , xσ(n) ). We say that a function is symmetric if σ(f ) = f .
Example 2.3 (n = 3). If σ = (123) is the cyclic permutation of 3 variables, then σ(x31 x22 x3 ) =
x32 x23 x1 . Evidently, x31 x22 x3 is not a symmetric function. On the other hand, x31 + x32 + x33 is a
symmetric function.
The elementary symmetric functions si are all symmetric. While there are many symmetric
functions besides the elementary ones, it turns out that they are all generated as polynomial
combinations of the elementary symmetric functions. This is an astounding fact!
Theorem 2.4 (Fundamental Theorem of Symmetric Functions). Let f (x1 , . . . , xn ) be any
symmetric polynomial. Then, f can be expressed as polynomial in the symmetric function, i.e
f = g(s1 , . . . , sn ) for some polynomial g.
Example 2.5. The theorem says that one can express x31 +x32 +x33 as a polynomial in s1 , s2 , s3 .
One can easily check that
x31 + x32 + x33 = s31 − 3s1 s2 + 3s3
Is there a way to derive this formula systematically? Yes, there is - we shall spell out the
algorithm in full detail when we prove the fundamental theorem, but it may be useful to sketch
the idea informally in the context of this example. To begin with, it’s easy to see that s31 , s1 s2 , s3
are the only monomials in s1 , s2 , s3 that give rise to degree 3 monomials in x1 , x2 , x3 , so these
are the only monomials that can appear in our formula. In other words, we must have a
formula like
x31 + x32 + x33 = as31 + bs1 s2 + cs3 ,
for some coefficients a, b, c, and the question is how to figure out these coefficients.
First, focus attention on the x31 term. On the left, it occurs with coefficient 1. On the right,
it’s easy to see that only s31 contains an x31 term and it occurs with coefficient one. Thus, we
must have a = 1.
Next, let’s subtract s31 from both sides, to get:
x31 + x32 + x33 − (x1 + x2 + x3 )3 = bs1 s2 + cs3 ,
If we expand out the left hand side, the x31 term cancels, so let’s examine the next lowest order
term, i.e. x21 x2 . On the left, it occurs with coefficient −3. On the right, it’s easy to see that
only s1 s2 contains an x21 x2 term and it occurs with coefficient 1. Thus, we must have b = −3.
 7

Next, let’s add 3s1 s2 to both sides to get:

x31 + x32 + x33 − (x1 + x2 + x3 )3 + 3(x1 + x2 + x3 )(x1 x2 + x1 x3 + x2 x3 ) = cs3 ,
If we expand out the left hand side, we see that x31 and x21 x2 terms cancel - in fact everything
cancels except the x1 x2 x3 term which occurs with coefficient 3. Since s3 = x1 x2 x3 , we must
have c = 3, and we are done.
In the example, I made use of the idea of the “lowest” monomial without actually explaining
what I meant. The key technical tool in proving the general theorem is the introduction of an
ordering on monomials which makes this concept precise.
Definition 2.6 (Lexicographic Ordering). The lexicographic ordering is a total ordering on
all degree m monomials in n variables, which can be defined as follows. Given any monomial,
we can write it as xi1 xi2 . . . xim with i1 ≤ i2 ≤ . . . ≤ im . In other words, we can write it as a
product of m variables whose subscripts go from lowest to highest. To compare two monomials,
we then just look at the first subscript were two monomials differ. More formally, we say that
xi1 xi2 . . . xim < xj1 xj2 . . . xjm
if i1 = j1 , i2 = j2 , . . . , ik−1 = jk−1 and ik < jk for some k ∈ {1, . . . , m}.
Example 2.7. The lexicographic ordering for degree 3 monomials in 3 variables goes like this:
x31 < x21 x2 < x21 x3 < x1 x22 < x1 x2 x3 < x1 x23 < x32 < x22 x3 < x2 x23 < x33
Definition 2.8. If f is a homogeneous polynomial of degree m (homogeneous means that
every monomial in f has the same degree), we let L(f ) be the “lowest” monomial of f , i.e.
the monomial of f which is least with respect to the lexicographic ordering.
Example 2.9. If f = 2x21 x2 + x1 x2 x3 + 3x33 , then L(f ) = 2x21 x2 , because x21 x2 < x1 x2 x3 < x33
in the lexicographic ordering.
If you think about it, you will see that certain monomials cannot occur as L(f ) for a
symmetric function f . For example, x1 x22 could never be the lowest monomial of a symmetric
function. Why not? Because if f contains the monomial x1 x22 , then it must also (by symmetry)
contain the monomial x21 x2 and x21 x2 < x1 x22 . More generally, we have the following lemma.
Lemma 2.10. If f is a symmetric function, and L(f ) = cxk11 xk21 . . . xknn , then k1 ≥ k2 ≥ . . . ≥
kn .
Proof. Let f be a symmetric function with L(f ) = xk11 xk21 . . . xknn , and suppose the statement
of the lemma fails, i.e. suppose that the ki ’s are not ordered from largest to smallest. Let
σ ∈ Sn be a permutation that orders the ki ’s correctly, i.e. such that
kσ(1) ≥ kσ(2) ≥ . . . ≥ kσ(n) .
k k k
Since f is symmetric, f must contain the monomial x1σ(1) x2σ(2) . . . xnσ(n) . By the definition
k k k
of the lexicographic ordering, we have x1σ(1) x2σ(2) . . . xnσ(n) < xk11 xk21 . . . xknn . But this is a
contradiction, since we started by assuming that xk11 xk21 . . . xknn was the lowest monomial in
f. 
Now we are ready to prove the fundamental theorem of symmetric functions. The idea, as
demonstrated in Example 2.5, is to focus on the lowest monomial of our symmetric function,
and then find a monomial in the elementary symmetric functions which matches it. By suc-
cessively subtracting off appropriate multiples of monomials in the symmetric functions, we
8 DAVID SMYTH

can work our way up the lexicographic ordering until there are monomials left! At that point,
we have expressed f as a polynomial in the symmetric functions.
Proof. First, we reduce to the case that f is homogenous. We claim that if we know the
fundamental theorem for homogenous symmetric functions, then we know the fundamental
theorem for all symmetric functions. To see this, let f be any symmetric function and write
f = f1 + . . . + fm , where each fi is homogenous of degree i. If f is symmetric, then each fi
must be as well (because the action of Sn preserves the degree of each monomial of f ). If we
know the fundamental theorem for homogenous symmetric functions, then we can write each
fi as a polynomial in elementary symmetric function. But then we clearly get a representation
of f as a polynomial in elementary symmetric functions as desired.
Now let f be a homogeneous symmetric function and let L(f ) = cxk1 . . . xkn . By Lemma
2.10, we know that k1 ≥ k2 ≥ . . . ≥ kn . We claim that there exists a monomial in the
symmetric functions, say csl11 sl22 . . . slnn such that
L(f ) = L(csl11 sl22 . . . slnn ).
To check this, we need to investigate the lowest monomials of the elementary symmetric
functions. By the definition of the elementary symmetric functions, one easily checks that:
L(si ) = x1 x2 . . . xi .
It follows that
l +l2 +...+ln−1
L(csl11 sl22 . . . slnn ) = cxl11 +l2 +...+ln x21 . . . xlnn .
Thus, in order to get L(f ) = L(csl11 sl22 . . . slnn ), we simply need to find non-negative integers
l1 , l2 , . . . , ln such that
l1 + . . . + ln = k1
l1 + . . . + ln−1 = k2
..
.
ln = kn .
Happily, the condition k1 ≥ k2 ≥ . . . ≥ kn guarantees that we can do this. Indeed, we
simply set ln = kn and li = ki − ki+1 for i = 1, . . . , n − 1. With this choice of li , we have
L(f ) = L(csl11 sl22 . . . slnn ) as desired.
Now we are basically done. If we let f 0 := f − csl11 sl22 . . . slnn , then f 0 is a symmetric function
with L(f 0 ) > L(f ). Thus, we can simply replace f by f 0 and repeat this procedure. As we do
this, we will subtract off multiples of monomials of the symmetric functions to get a sequence
of functions f, f 0 , f 00 , . . . with higher and higher lowest monomials. The only way this process
can terminate is to have f k = 0 for some k. At that point, we have an equation expressing f
as a sum of monomials of elementary symmetric functions, i.e. a polynomial in the elementary
symmetric functions. 

3. Lagrange’s Solution to the Quartic

Lagrange actually proved the fundamental theorem on symmetric functions in the course of
developing a more systematic approach to solving polynomial equations. In this lecture, we’ll
see how Lagrange made use of the theorem to give a solution to the general quartic equation.
Lagrange’s solution begins with the following observation, which tells you how to turn an
arbitrary function into a symmetric function.
 9

Lemma 3.1. Let f1 ∈ k[α1 , . . . , αn ] be any polynomial, and let f1 , . . . , fk be the orbit of f1
under the action of Sn , i.e. the set of all functions you get by acting on f1 with elements of
Sn . If s(x1 , . . . , xk ) is any symmetric function in k variables, then s(f1 , . . . , fk ) is a symmetric
function in α1 , . . . , αn .
Example 3.2. Let f1 = (α1 + α2 )(α3 + α4 ) ∈ k[α1 , α2 , α3 , α4 ]. One can easily check that the
orbit of f1 is:
f1 = (α1 + α2 )(α3 + α4 )
f2 = (α1 + α3 )(α2 + α4 )
f3 = (α1 + α4 )(α2 + α3 )

According to the theorem, any symmetric function of f1 , f2 , f3 must give a symmetric function
of the αi . In particular, one can verify by inspection that f1 + f2 + f3 , f1 f2 + f1 f3 + f2 f3 , f1 f2 f3
are all symmetric in α1 , . . . , α4 . It follows of course that they can be expressed as polynomials
in the elementary symmetric functions, and indeed one can easily check:
f1 + f2 + f3 = 2s2 (α1 , . . . , α4 )
f1 f2 + f1 f3 + f2 f3 = (Exercise!)
f1 f2 f3 = (Exercise!)

How does Lagrange use this to solve the quartic polynomial? Given an equation
f (x) = x4 + a2 x + a3 + a4 = 0,
Lagrange starts by assuming that f (x) has 4 distinct roots α1 , α2 , α3 , α4 . (It is not true of
course that all degree 4 polynomials have 4 distinct roots - this extra assumption is a weakness
of Lagrange’s method, and later in the course we will have the technical machinery to get
around it.) As we discussed in the last lecture, these roots must satisfy the four equations:

s1 (α1 , α2 , α3 , α4 ) = α1 + α2 + α3 + α4 = −a1 = 0.
s2 (α1 , α2 , α3 , α4 ) = a2
s3 (α1 , α2 , α3 , α4 ) = −a3
s4 (α1 , α2 , α3 , α4 ) = a4

Thus, we are given three complex numbers a2 , a3 , a4 , and we want to find four complex numbers
α1 , α2 , α3 , α4 satisfying the above equations. Though we don’t know what α1 , α2 , α3 , α4 are,
Lagrange begins by defining f1 , f2 , f3 to be the expressions in Example 3.2. In other words, we
have used four unknown complex numbers to define three more unknown complex numbers.
The key point, however, is that we have a way to find formulas for f1 , f2 , f3 in terms of
a2 , a3 , a4 . Here comes Lagrange’s stroke of genius - consider the cubic polynomial:
(x − f1 )(x − f2 )(x − f3 ) = x3 − (f1 + f2 + f3 )x2 + (f1 f2 + f1 f3 + f2 f3 )x − f1 f1 f3
As we argued in Example 3.2, the coefficients of this polynomial are symmetric functions in
α1 , α2 , α3 , α4 . By the fundamental theorem of symmetric functions, we can therefore write
the coefficients of this function as polynomials in a2 , a3 , a4 . But this means that we can use
10 DAVID SMYTH

our solution for the cubic to solve for f1 , f2 , f3 in terms of a2 , a3 , a4 ! Needless to say, actually
writing down the formula would be a rather tedious affair, but the key point is that we know
one exists.
Once we have solved for f1 , f2 , f3 , it is easy to see that we can solve for α1 , α2 , α3 , α4 .
Indeed, using the equation
α1 + α2 + α3 + α4 = 0,
we see that
f1 = (α1 + α2 )(α3 + α4 ) = −(α1 + α2 )2 ,
and we conclude that
p
α1 + α2 = −f1 ,
p
α3 + α4 = − −f1 .
Similarly, we have
p
α1 + α3 = −f2 ,
p
α2 + α4 = − −f2 ,
p
α1 + α4 = −f3 ,
p
α2 + α3 = − −f3 .
From here, one can solve for each of the αi individually. For example,
√ √ √
(α1 + α2 ) + (α1 + α3 ) − (α2 + α3 ) −f1 + −f2 + −f3
α1 = = .
2 2
We have solved the quartic!
Now, let us step back from the particulars of this example, and consider Lagrange’s overall
strategy. If we are trying to solve a polynomial of degree n, Lagrange’s strategy is to find a
function of the roots f1 (α1 , . . . , αn ) whose orbit f1 , . . . , fk under Sn is less than n. Assuming
we know how to solve equations of degree less than n, we can solve for f1 , . . . , fk by considering
the polynomial
Yk
(x − fi ).
i=1
As in the example of the quartic, the fundamental theorem of symmetric functions implies that
the coefficients of this polynomial will be polynomials in the original coefficients a1 , . . . , an , so
we can solve.
Of course, as someone pointed out in class, if we take the orbit of f1 to be too small, e.g.
if we simply take f1 to be a symmetric function, then knowing the value of f1 won’t help us
much in our quest to solve for the αi ’s. The idea, therefore, is to find a function which has
a small enough orbit that you can solve for it, but is also a useful bridge to solving for the
αi ’s. In fact, Lagrange spent the last years of his life looking for a function f (α1 , α2 , α3 , α4 , α5 )
which would allow him to solve the quintic. Alas, we now know that solving the quintic is
impossible.
Let us now step back and consider the overall idea of Galois Theory. The problem with
polynomials is that they are really not very transparent. When we solved the cubic in Lecture
1, we found that we could essentially reduce the cubic equation to a quadratic equation. In
some sense, therefore, there was a quadratic equation “hidden” inside the cubic equation.
Similarly, Lagrange found a cubic equation “hidden” inside a quartic equation. We need
 11

to switch to a kind of mathematical structure in which this hidden structure becomes more
transparent.
In the following weeks, we will see how to associate to any polynomial f (x) ∈ K[x] a certain
algebraic object L/K, called a field extension. While the field extension contains, in some
sense, all the relevant information about f , it is much easier to deal with. For example,
the fact that solving the cubic involves a hidden quadratic equation, will appear as the fact
the associated field extension L/K has an intermediate extension associated to a quadratic
equation, i.e. we have a picture like this:
L

x3 +a2 x+a3 K0

x2 +a3 x−(a32 )/27

K
Thus, the problem of solving polynomials will be reduced to understanding the structure of
field extensions, especially the problem of understanding all the intermediate field extensions
of a given field extension. This problem, in turn, will turn out to be solvable in terms of pure
group theory. Associated to a field extension L/K, we will define a certain group G(L/K)
called the Galois group of the extension. Subfields will correspond to subgroups/quotients
of the Galois group, just as Lagrange’s intermediate cubic equation was found by finding a
homomorphism S4 → S3 . This will give us a very pretty answer to our original problem!