8 Minimum Polynomial

Throughout this section let T : V → V be a linear transformation on a finite-dimensional

vector space V over a field K.
We have seen that a T satisfies its characteristic polynomial χT (X). However, this may
not be the ‘best’ result—there may be polynomials of smaller degree with the same property.
We want to investigate this in some detail.

8.1 The polynomials satisfied by T

Although we work from scratch much of what we say here can be deduced from theorems
proved in the parallel Rings and Arithmetic lecture course.
We begin with a piece of notation. Let I(T ) := {f | f ∈ K[X] such that f (T ) = 0}. In
words, I(T ) is the set of polynomials with coefficients in K which our linear transformation
T satisfies. We want to describe the properties of this set.

8.1.1. 0 ∈ I(T ).

This is trivial.

8.1.2. If f, g ∈ I(T ) then f + g ∈ I(T ) and −f ∈ I(T ).

As (f + g)(X) := f (X) + g(X), and (−f )(X) := −f (X) these are trivial.

8.1.3. If f ∈ I(T ) and g ∈ K[X] then f · g ∈ I(T ).

Again, (f · g)(X) := f (X)g(X) so the result is clear.

In other words, I(T ) is an ideal of the ring K[X].

8.1.4. For some monic polynomial m we have that I(T ) = {f m | f ∈ K[X]}.

If I(T ) = {0} then take m = 01 and we are done.

Otherwise let m∗ be chosen of least degree in I(T ) \ {0}; then its monic multiple m is
also in I(T ). By the above proposition all multiples of m are in I(T ). Let f be an arbitrary
member of I(T ). Then by the Division Algorithm we have that f = gm + r where r is either
0 or of lesser degree than m. Now r = f − gm ∈ I(T ), so by choice of m we must have r = 0,
and f = mg.

8.1.5. Let m1 , m2 be monic polynomials such that I(T ) = {f m1 | f ∈ K[X]} = {f m2 | f ∈

K[X]}; then m1 = m2 .

We have m1 = f1 m2 and m2 = f2 m1 . If either is 0 so is the other. Otherwise m1 =

f1 f2 m1 , and as m1 6= 0 we have f1 f2 = 1. So f1 and f2 must lie in K. As m1 and m2 are
monic they are equal.

Definition. The minimum polynomial of T , which we denote by mT , is the unique monic

polynomial of least degree satisfied by T .

Note. For this to make sense as a definition one needs to prove—as we have done, and as
you have to do in examinations—existence and uniqueness.
1
Conventionally 0 is an honorary member of the monic polynomials!

24
8.2 The minimum polynomial is non-zero
So far everything we have said is independent of the fact that V is finite dimensional. It
would be very dull, though, if mT were just 0. By the Cayley–Hamilton Theorem we know
that this is not so. Indeed the propositions above yield this reformulation of that theorem:

Theorem (Cayley–Hamilton again). Let T : V → V be a linear transformation on a finite-

dimensional vector space V over a field K. Then mT |χT .

However, as we’ve seen the Cayley–Hamilton Theorem is tricky and if we can it’s good
to avoid it. Here are two props which yield a different proof that I(T ) 6= {0} in the case
when dim(V ) = n < ∞.

8.2.1. Suppose v ∈ V . Then there exists a non-zero polynomial fv ∈ K[X] such that fv (T )v =
0.

If n = 0 then V = 0 and fv = 1 will serve. Otherwise consider the (n + 1) vectors

v, T v, T 2 v, . . . , T n v. There are too many to be LI, so there are constants a0 , a1 , . . . , an ∈ K
not all zero, such that a0 v + a1 T v + · · · + an T n v = 0. take fv := nr=0 ar X r and we have
P
the result.

8.2.2. There exists a polynomial of degree at most n2 in I(T ).

If V = 0 then 1 will suffice. Otherwise let {v1 , . . . , vn } be a basis of V . Then using the
previous proposition form f = fv1 fv2 . . . fvn . Then f (T ) = fv1 (T )fv2 (T ) . . . fvn (T ) is a linear
transformation which kills each basis vector, and so kills every vector in V . Hence this is the
zero transformation, and f ∈ I(T ).

8.3 The minimum polynomial and eigenvalues

We have seen that the eigenvalues are precisely the roots of χT . By the Cayley–Hamilton
Theorem we know that mT |χT . Can we say more? And can we avoid appealing to the
Cayley–Hamilton Theorem? The answers are yes, yes.

8.3.1. Every root of mT is an eigenvalue of T .

Let α ∈ K satisfy mT . Then by the Remainder Theorem (or by the Division Algorithm
if you prefer), mT (X) = (X − α)g(X) for some g ∈ K[X]. Now g has too small degree for
g(T ) = 0, so for some vector v ∈ V we have g(T )v 6= 0. However mT (T )v = 0; that is
(T − α)g(T )v = 0. Then u := g(T )v is an α-eigenvector of T .

8.3.2. Every eigenvalue of T is a root of mT .

Let λ be an eigenvalue; and let u be a λ-eigenvector. If λ is not a root of mT then

(X − λ) does not divide mT , so that in fact (X − λ) and mT are coprime. Hence by Euclid’s
Algorithm we have that 1 = f (X)(X − λ) + g(X)mT (X) for some f, g ∈ K[X]. Now

0 6= u = 1u = f (T )(T − λ)u + g(T )mT (T )u = f (T )0 + g(T )Ou = 0 + 0 = 0

a contradiction.

8.4 How to Work Examples

How do we find the minimal polynomial, then?

1. Choose a basis E of V .

25
 2. Work out the matrix A of T w.r.t. E.
(Often someone else has done these two steps for us, and given us the matrix of T .)

3. Compute χT (X) = χA (X) as efficiently as possible.

4. Factorise χT (X) as far as possible.

5. Exploit the facts that

(a) mT |χT
(b) every root of χT must be a root of mT

to produce a finite list of possibilities for mT , say {g1 , g2 , . . . gk }.

6. Check (in order increasing degree, say) whether each gi (A) = 0; if it does then you
have found mT .

8.4.1 An explicit example

Here’s a concrete example.
P Suppose that V := {f ∈ K[X] | deg(f ) < 4, and T is the (formal)
ak X 7→ kak X k−1 .
k
P
differentiation operator

1. Choose the obvious basis E := {1, X, X 2 , X 3 } of V .

2. Work out the matrix A of T w.r.t. E; well T · 1 = 0, T · X = 1, T · X 2 = 2X, and

T · X 3 = 3X 2 , so that  
0 1 0 0
 0 0 2 0 
A=  0 0
.
0 3 
0 0 0 0

3. Now χT (X) = det (X − A), which we did in the first year (it’s upper triangular) and
got χT (X) = X 4 .

4. Now χT (X) factors as the ‘irreducible’ X occurring 4 times.

5. We use that

(a) mT |X 4
(b) 0 must be a root of mT

to produce the list of possibilities for mT —{X, X 2 , X 3 , X 4 }.

6. This means we are going to have to check powers of A so let’s work them out:
       
0 1 0 0 0 0 2 0 0 0 0 6 0 0 0 0
 0 0 2 0 0 0 0 6
 , A2 =  3  0 0 0 0  , A4 =  0 0 0 0 
    
A=  0 0 0 3  0 0 0 0 , A =  0 0
 .
 0 0   0 0 0 0 
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

(This last we’d know from Cayley–Hamilton, but pragmatically it’s always worth
checking—and so avoiding the appeal to that tricky result.)
Now:

(a) Is A = O? No, the (1, 2) entry is 1 6= 0 by the axioms. So mT 6= 0.

26
 (b) Is A2 = O? Usually the answer is ‘no’; but be careful. In any field where 1 + 1 = 0
the answer is yes. If we are working with such a K then mT = X 2 . Otherwise, it
is not.
(c) Is A3 = O? Usually the answer is ‘no’; but if 1 + 1 + 1 + 1 + 1 + 1 = 0 then the
answer is ‘yes’. Note that (1 + 1 + 1 + 1 + 1 + 1) = (1 + 1)(1 + 1 + 1). So the
awkward case can only occur if 1 + 1 = 0 or 1 + 1 + 1 = 0. We have alreadt dealt
with the case 1 + 1 = 0. So we see that if 1 + 1 + 1 = 0 in K then mT = X 3 ,
otherwise it is not.
(d) Is A4 = O? Yes it is. So except in fields with 1 + 1 = 0 or 1 + 1 + 1 = 0 we have
that mT = X 4 .

8.5 The minimum polynomial of the dual transformation

Recall the definition of the dual map T ′ : V ′ → V ′ : for any f ∈ V ′ and v ∈ V we have
T ′ (f )(v) = f (T (v)).
P Recall also that (T + S)′ = T ′ + S ′ and also (T S)′ = S ′ T ′ .
r
Let mT (X) = ar X ∈ K[X]. Then
X ³X ´′
mT (T ′ ) = ar (T ′ )r = ar T r = O′ = O

so that mT , which kills T ′ , must be a multiple of mT ′ .

At once we can deduce that mT ′ is a multiple of mT ′′ , and since T = T ′′ we have this
result:

Theorem. Let T : V → V be a linear transformation on a finite-dimensional vector space

V over a field K, and let T ′ : V ′ → V ′ be the dual transformation. Then mT = mT ′ .

27