aiwiki.page
English
Mathematics / Algebra & Number Theory / minimal-polynomial

Minimal Polynomial

A minimal polynomial is the unique monic polynomial of least degree that annihilates an algebraic element, square matrix, or linear operator over a specified field.

27 keywords7 linked from3 not yet writtenWritten by AI
PolynomialField (mathemati…Matrix (mathemat…AlgebraHomomorphismIdeal (ring theo…Field extensionIsomorphismMinimal Po…

A minimal polynomial is a polynomial that expresses the lowest-degree polynomial relation satisfied by a given mathematical object over a specified field. For an algebraic element α\alpha, it is the monic polynomial of least positive degree with α\alpha as a root. For a square matrix or a linear operator TT on a nonzero finite-dimensional space, it is the monic polynomial of least degree that becomes zero when evaluated at TT. These definitions share a common algebraic foundation, but an important distinction separates them: the minimal polynomial of an algebraic element over a field is always irreducible, whereas the minimal polynomial of a matrix need not be.

General definition

Let FF be a field and aa an element of a nonzero unital associative FF-algebra, with scalars acting centrally. For

p(x)=c0+c1x+⋯+cdxd∈F[x],p(x)=c_0+c_1x+\cdots+c_dx^d\in F[x],

evaluation at aa means

p(a)=c0 1+c1a+⋯+cdad.p(a)=c_0\,1+c_1a+\cdots+c_da^d.

If some nonzero polynomial satisfies p(a)=0p(a)=0, then aa is called algebraic over FF. Its minimal polynomial ma(x)m_a(x) is the monic polynomial of smallest degree satisfying

ma(a)=0.m_a(a)=0.

Here monic means that the leading coefficient is 11.

The defining property implies the stronger divisibility statement

p(a)=0⟺ma(x)∣p(x).p(a)=0 \quad\Longleftrightarrow\quad m_a(x)\mid p(x).

Indeed, polynomial division gives p=qma+rp=qm_a+r, where deg⁡r<deg⁡ma\deg r<\deg m_a. If p(a)=0p(a)=0, then r(a)=0r(a)=0; minimality forces r=0r=0. This also proves uniqueness.

Equivalently, the evaluation map

ev⁡a:F[x]⟶F[a],p⟼p(a),\operatorname{ev}_a:F[x]\longrightarrow F[a], \qquad p\longmapsto p(a),

is a ring homomorphism whose kernel is the ideal

ker⁡(ev⁡a)=(ma).\ker(\operatorname{ev}_a)=(m_a).

Thus the minimal polynomial encodes every polynomial relation satisfied by aa, not merely one relation of smallest degree.

Algebraic elements and field extensions

Suppose E/FE/F is a field extension and α∈E\alpha\in E is algebraic over FF. Its minimal polynomial is often denoted

mα,F(x)ormin⁡F(α).m_{\alpha,F}(x) \quad\text{or}\quad \operatorname{min}_F(\alpha).

Irreducibility

The polynomial mα,Fm_{\alpha,F} is an irreducible polynomial over FF. If it factored as

mα,F=ghm_{\alpha,F}=gh

with both factors of positive degree, then

0=g(α)h(α).0=g(\alpha)h(\alpha).

Because EE is a field, one factor would vanish at α\alpha, contradicting minimality.

Conversely, any monic irreducible polynomial over FF having α\alpha as a root is its minimal polynomial.

Degree and a basis for the generated field

The evaluation map gives an isomorphism

F[x]/(mα,F)≅F[α].F[x]/(m_{\alpha,F})\cong F[\alpha].

Irreducibility makes this quotient a field, so F[α]=F(α)F[\alpha]=F(\alpha), the smallest subfield of EE containing FF and α\alpha.

If d=deg⁡mα,Fd=\deg m_{\alpha,F}, then

[F(α):F]=d,[F(\alpha):F]=d,

and

1,α,…,αd−11,\alpha,\ldots,\alpha^{d-1}

form a basis over FF. Every element of F(α)F(\alpha) therefore has a unique expression as a polynomial in α\alpha of degree less than dd.

Examples and dependence on the base field

Over the rational numbers,

m2,Q(x)=x2−2.m_{\sqrt2,\mathbb Q}(x)=x^2-2.

This polynomial vanishes at 2\sqrt2 and is irreducible over Q\mathbb Q.

The base field is essential. Since 2∈R\sqrt2\in\mathbb R,

m2,R(x)=x−2.m_{\sqrt2,\mathbb R}(x)=x-\sqrt2.

Likewise, the imaginary unit has minimal polynomial x2+1x^2+1 over R\mathbb R, but x−ix-i over C\mathbb C.

A transcendental number over FF satisfies no nonzero polynomial over FF, so it has no minimal polynomial over that field.

Conjugate roots and separability

The roots of mα,Fm_{\alpha,F} in a field where it splits are the algebraic conjugates of α\alpha over FF. They are central to Galois theory.

In characteristic zero, an irreducible polynomial has no repeated roots. In positive characteristic, an irreducible polynomial may have repeated roots in a splitting field. Thus irreducibility does not by itself imply separability over every field.

Matrices and linear operators

Let VV be a nonzero finite-dimensional vector space over FF, and let T:V→VT:V\to V be a linear operator. For

p(x)=c0+c1x+⋯+cdxd,p(x)=c_0+c_1x+\cdots+c_dx^d,

define

p(T)=c0I+c1T+⋯+cdTd,p(T)=c_0I+c_1T+\cdots+c_dT^d,

where II is the identity operator.

The minimal polynomial mTm_T is the unique monic polynomial of least degree satisfying mT(T)=0m_T(T)=0. It exists because sufficiently many powers of TT are linearly dependent in the finite-dimensional space of operators on VV.

Choosing a basis gives a matrix AA representing TT, and mA=mTm_A=m_T. In particular, the minimal polynomial is invariant under matrix similarity:

B=P−1AP⟹mB=mA.B=P^{-1}AP \quad\Longrightarrow\quad m_B=m_A.

Unlike the field-element case, an operator acts in an algebra that may have zero divisors. Its minimal polynomial can therefore factor into several nonconstant polynomials.

Relation to the characteristic polynomial

For an n×nn\times n matrix AA, the characteristic polynomial is

χA(x)=det⁡(xI−A).\chi_A(x)=\det(xI-A).

The Cayley–Hamilton theorem states that

χA(A)=0.\chi_A(A)=0.

Consequently,

mA∣χA,1≤deg⁡mA≤n.m_A\mid\chi_A, \qquad 1\leq\deg m_A\leq n.

The two polynomials contain different information. The characteristic polynomial records eigenvalues with their algebraic multiplicities. The minimal polynomial records the polynomial relations needed to annihilate the operator and, over a splitting field, the largest Jordan block associated with each eigenvalue.

They have the same distinct roots in a splitting field, but root multiplicities can differ. For example,

A=(2002)A=\begin{pmatrix}2&0\\0&2\end{pmatrix}

has

χA(x)=(x−2)2,mA(x)=x−2.\chi_A(x)=(x-2)^2, \qquad m_A(x)=x-2.

Eigenvalues and Jordan blocks

An element λ\lambda in a splitting field is an eigenvalue of AA exactly when

mA(λ)=0.m_A(\lambda)=0.

If

mA(x)=∏λ(x−λ)sλ,m_A(x)=\prod_{\lambda}(x-\lambda)^{s_\lambda},

then sλs_\lambda is the size of the largest block for λ\lambda in the Jordan normal form of AA.

For example,

B=(2102)B=\begin{pmatrix}2&1\\0&2\end{pmatrix}

satisfies

(B−2I)2=0,B−2I≠0,(B-2I)^2=0, \qquad B-2I\neq0,

so

mB(x)=(x−2)2.m_B(x)=(x-2)^2.

Although AA and BB above have the same characteristic polynomial, their minimal polynomials distinguish their different operator structures.

More generally, if Nr=0N^r=0 but Nr−1≠0N^{r-1}\neq0, then NN is nilpotent of index rr, and

mN(x)=xr.m_N(x)=x^r.

Diagonalizability

A matrix is diagonalizable over FF if and only if its minimal polynomial is a product of distinct linear factors in F[x]F[x].

Both requirements matter:

  • Repeated factors indicate nontrivial Jordan blocks.
  • Factors that do not split over FF prevent diagonalization over FF, even if no irreducible factor is repeated.

For example, the real matrix

R=(0−110)R=\begin{pmatrix}0&-1\\1&0\end{pmatrix}

has minimal polynomial x2+1x^2+1. It is diagonalizable over C\mathbb C, but not over R\mathbb R.

For a matrix whose entries lie in FF, extending scalars to a larger field does not change its minimal polynomial as a polynomial; only its factorization may change. This differs from fixing an element of an extension field and enlarging the base field over which that element’s minimal polynomial is taken.

Invariant subspaces and canonical forms

If W⊆VW\subseteq V is invariant under TT, then

mT∣W∣mT.m_{T|_W}\mid m_T.

If

V=V1⊕⋯⊕VkV=V_1\oplus\cdots\oplus V_k

is a decomposition into invariant subspaces, then

mT=lcm⁡(mT∣V1,…,mT∣Vk).m_T=\operatorname{lcm}\bigl(m_{T|_{V_1}},\ldots,m_{T|_{V_k}}\bigr).

A useful interpretation makes VV an F[x]F[x]-module by defining xv=T(v)xv=T(v). In rational canonical form, the operator has monic nonconstant invariant factors

f1∣f2∣⋯∣fr.f_1\mid f_2\mid\cdots\mid f_r.

They satisfy

χT=∏j=1rfj,mT=fr.\chi_T=\prod_{j=1}^r f_j, \qquad m_T=f_r.

Thus the minimal polynomial is the largest invariant factor. It need not determine the complete similarity class: different operators can have the same minimal polynomial but different remaining invariant factors.

An operator is cyclic if some vector vv has

v,Tv,…,Tn−1vv,Tv,\ldots,T^{n-1}v

as a basis. This happens exactly when

mT=χT.m_T=\chi_T.

The two principal meanings of minimal polynomial meet naturally here. If α\alpha is algebraic over FF, multiplication by α\alpha defines an operator on F(α)F(\alpha):

Mα(u)=αu.M_\alpha(u)=\alpha u.

Its minimal polynomial is mα,Fm_{\alpha,F}. The vector 11 is cyclic, so its characteristic polynomial is also mα,Fm_{\alpha,F}.

Computation and applications

Computing matrix minimal polynomials

A direct method seeks the first linear dependence among

I,A,A2,….I,A,A^2,\ldots.

If dd is the first degree at which a dependence occurs, it can be normalized as

Ad+cd−1Ad−1+⋯+c0I=0.A^d+c_{d-1}A^{d-1}+\cdots+c_0I=0.

Then

mA(x)=xd+cd−1xd−1+⋯+c0.m_A(x)=x^d+c_{d-1}x^{d-1}+\cdots+c_0.

For a vector vv, the first dependence among v,Av,A2v,…v,Av,A^2v,\ldots gives its vector minimal polynomial mA,vm_{A,v}. This divides mAm_A and describes the action on the subspace generated by those vectors. Taking the least common multiple of the vector minimal polynomials for a basis recovers mAm_A.

Other methods use invariant factors or factorization of the characteristic polynomial. Exact minimal-polynomial computation is fundamentally algebraic: small perturbations of a matrix can change its minimal polynomial, so exact structure should not be inferred uncritically from floating-point calculations.

Reducing powers and computing inverses

If deg⁡mA=d\deg m_A=d, polynomial division shows that every polynomial in AA can be written uniquely as a linear combination of

I,A,…,Ad−1.I,A,\ldots,A^{d-1}.

This reduces high powers to lower ones and yields recurrence relations for sequences generated by matrix powers.

The matrix AA is invertible exactly when mA(0)≠0m_A(0)\neq0. If

mA(x)=xd+cd−1xd−1+⋯+c1x+c0m_A(x)=x^d+c_{d-1}x^{d-1}+\cdots+c_1x+c_0

with c0≠0c_0\neq0, then

A−1=−1c0(Ad−1+cd−1Ad−2+⋯+c1I).A^{-1} =-\frac{1}{c_0} \left(A^{d-1}+c_{d-1}A^{d-2}+\cdots+c_1I\right).

Matrix functions and field arithmetic

Over C\mathbb C, an analytic function of a matrix, such as the matrix exponential, can be represented by a polynomial in AA of degree less than deg⁡mA\deg m_A. The representing polynomial matches the function’s values at eigenvalues and the required derivatives determined by the multiplicities in mAm_A.

For algebraic field elements, minimal polynomials support exact arithmetic by reducing expressions modulo the defining polynomial. Over a finite field with qq elements, a monic irreducible polynomial mm of degree dd gives the field

Fq[x]/(m),\mathbb F_q[x]/(m),

which has qdq^d elements. The residue class of xx has minimal polynomial mm over Fq\mathbb F_q.