Irreducible polynomials and characteristic-zero separability

Component notice and licence. Adapted from the exact 09H0 statement and proof. The division/Bézout prerequisites and root consequence are explicit elementary completions. Fields are nonzero commutative rings in which every nonzero element has an inverse; an irreducible polynomial is a nonconstant nonunit with no factorization into two nonunits.

The statement

For a field FF and irreducible P∈F[x]P\in F[x], either PP and its formal derivative P′P' are relatively prime, or P′=0P'=0. The latter case occurs only in characteristic p>0p>0; then P(x)=Q(xpf)P(x)=Q(x^{p^f}) for some f≥1f\geq1, where QQ is irreducible and relatively prime to Q′Q'. In particular every irreducible polynomial over a characteristic-zero field is separable, and for every root α\alpha in any extension field, P′(α)≠0P'(\alpha)\ne0.

Elementary completion: polynomial division and Bézout

For nonzero D∈F[x]D\in F[x] and arbitrary A∈F[x]A\in F[x], repeatedly subtract the multiple of DD that cancels the highest-degree term of the current remainder. The leading coefficient of DD is invertible, and the degree strictly decreases. This terminates with A=QD+RA=QD+R, deg⁡R<deg⁡D\deg R<\deg D (or R=0R=0). Uniqueness follows because a nonzero multiple of DD has degree at least deg⁡D\deg D.

Every nonzero ideal of F[x]F[x] is principal: choose a nonzero element DD of least degree in the ideal and divide any other element by it. The remainder belongs to the same ideal and must vanish. Apply this to the ideal (P,P′)(P,P'). Its generator DD divides both polynomials, and being in that ideal has the form D=UP+VP′D=UP+VP' for polynomials U,VU,V. Relatively prime means that this ideal is the whole ring, equivalently that one can take D=1D=1.

Proof, including the positive-characteristic alternative

Suppose P′≠0P'\ne0. Then deg⁡P′<deg⁡P\deg P'<\deg P. If (P,P′)=(D)(P,P')=(D) were a proper ideal, DD would be a nonunit divisor of PP, so irreducibility makes DD associate to PP. Since it also divides the nonzero P′P', one would have deg⁡P≤deg⁡P′<deg⁡P\deg P\leq\deg P'<\deg P, a contradiction. Thus PP and P′P' are relatively prime.

If P′=0P'=0 and P=∑j=0dajxjP=\sum_{j=0}^d a_jx^j, all jajj a_j for j≥1j\geq1 vanish. In characteristic zero this forces PP to be constant, a contradiction. A field of positive characteristic has prime characteristic pp: if the least positive integer annihilating 11 factored into two smaller positive integers, their nonzero images would multiply to zero in a field. It follows that aj=0a_j=0 whenever pp does not divide jj. Thus P(x)=P1(xp)P(x)=P_1(x^p) with P1P_1 nonconstant. A factorization of P1P_1 would give one of PP, so P1P_1 is irreducible. Its degree is strictly smaller. Repeat while the derivative vanishes; natural-number induction gives P(x)=Q(xpf)P(x)=Q(x^{p^f}) with Q′≠0Q'\ne0. The first paragraph then makes Q,Q′Q,Q' relatively prime. This proves the full source alternative.

Root consequence needed in AN06-U008

In characteristic zero, UP+VP′=1UP+VP'=1. If P(α)=0P(\alpha)=0 in any extension field, evaluation gives V(α)P′(α)=1V(\alpha)P'(\alpha)=1, hence P′(α)≠0P'(\alpha)\ne0. To see that this is precisely the simple-root conclusion, divide by x−αx-\alpha in the extension field and write P=(x−α)RP=(x-\alpha)R. The product rule for the formal derivative, obtained term by term from the polynomial coefficients, gives P′(α)=R(α)P'(\alpha)=R(\alpha). Therefore P′(α)=0P'(\alpha)=0 holds exactly when (x−α)2(x-\alpha)^2 divides PP. No construction of an algebraic closure is needed: the assertion holds in every field where a root is being considered.

Elementary completion: algebraic and rational function fields

The ring and fraction-field construction supplies the ambient fields. If aa in a field extension is algebraic over EE, choose a nonzero polynomial of least degree vanishing at aa and divide by its leading coefficient. This gives a monic polynomial hh. A factorization of hh into positive-degree factors would make one factor vanish at aa, since the extension is a field, contradicting minimal degree. Thus hh is irreducible. Polynomial division shows that every polynomial vanishing at aa is a multiple of hh: the remainder also vanishes and has smaller degree. In particular the monic minimal polynomial is unique.

Evaluation therefore identifies E[a]E[a] with E[T]/(h)E[T]/(h). This quotient is a field. For a nonzero class represented by gg, the principal-ideal and Bézout proof above gives (g,h)=E[T](g,h)=E[T]: a nonunit common divisor would be associated to the irreducible hh, whereas hh does not divide gg. Thus ug+vh=1ug+vh=1 for some u,vu,v, and the class of uu is an inverse. Consequently E[a]=E(a)E[a]=E(a), with basis 1,a,…,ad−11,a,\ldots,a^{d-1} over EE, where d=deg⁡hd=\deg h. Division proves spanning; minimality of hh proves linear independence.

If b1,…,bjb_1,\ldots,b_j is a basis of a field LL over EE, and c1,…,ckc_1,\ldots,c_k a basis of FF over LL, then the products biclb_i c_l form a basis of FF over EE. Expand coefficients in the two bases to prove spanning. In a relation among the products, independence of the clc_l first makes each LL coefficient zero, and independence of the bib_i makes every EE coefficient zero. It follows by induction that adjoining finitely many algebraic elements produces a finite-dimensional extension. Every element vv of a finite-dimensional extension is algebraic, since sufficiently many powers 1,v,v2,…1,v,v^2,\ldots are linearly dependent. Explicitly, write these powers in a basis of size dd and perform row elimination on the resulting dd-by-(d+1)(d+1) coefficient matrix. There are at most dd pivot columns; assigning a nonzero value to one free variable and solving for the pivots gives a nontrivial relation. Elimination uses only division by nonzero field elements.

Elements u1,…,uru_1,\ldots,u_r are algebraically independent over kk if evaluating polynomials in them is injective on k[T1,…,Tr]k[T_1,\ldots,T_r]. Their generated field is consequently its fraction field, denoted k(u1,…,ur)k(u_1,\ldots,u_r). Suppose F=k(x1,…,xn)F=k(x_1,\ldots,x_n), and begin with any finite algebraically independent list already in FF. Examine the xjx_j in order, adding xjx_j precisely when independence is preserved. If it is not added, a nonzero polynomial relation can be collected in powers of xjx_j; at least one coefficient remains nonzero on the independent list. Thus xjx_j is algebraic over the field generated by that list. It remains algebraic as the independent list is enlarged, because its nonzero coefficients remain nonzero under field inclusions. The final independent list is finite; all the generators are algebraic over its rational function field EE. Since the initial list was in FF, adjoining the xjx_j to EE recovers FF. The preceding finite-tower argument proves that F/EF/E is finite. This proves the precise transcendence-basis assertion used in AN06-U008, including a prescribed initial transcendental element.

Elementary completion: extending derivations

A derivation DD is an additive map satisfying D(ab)=aD(b)+bD(a)D(ab)=aD(b)+bD(a). It has D(1)=0D(1)=0, by applying this identity to 1⋅11\cdot1. If it is zero on a subfield kk, the product rule makes it kk-linear. On a polynomial ring over kk, prescribing the derivatives of its variables and differentiating each monomial defines a derivation: expansion verifies the product rule on two monomials and finite distributivity extends it to arbitrary polynomials. Algebraic independence therefore defines the usual partial derivation on the polynomial algebra generated by an independent list.

A derivation on a domain AA, with values in its fraction field, extends uniquely to that field by

D(a/s)=D(a)s−aD(s)s2,s≠0. D(a/s)=\frac{D(a)}s-\frac{aD(s)}{s^2},\qquad s\ne0.

For well-definedness, if a/s=b/ta/s=b/t, then at=bsat=bs. Differentiate that equality and multiply out the difference of the proposed formulas with denominator s2t2s^2t^2; using at=bsat=bs reduces the numerator to zero. Expansion with a common denominator proves additivity and the product rule. Conversely the product rule applied to s⋅s−1=1s\cdot s^{-1}=1 forces this formula, proving uniqueness. This supplies a derivation on each rational function field, without choosing representations of its elements.

Let now EE have characteristic zero, D:E→ED:E\to E be a derivation, and aa be algebraic over EE with monic minimal polynomial h(T)=∑hjTjh(T)=\sum h_jT^j. Separability, proved above, gives h′(a)≠0h'(a)\ne0. Define

b=−∑jD(hj)ajh′(a). b=-\frac{\sum_jD(h_j)a^j}{h'(a)}.

For g(T)=∑gjTjg(T)=\sum g_jT^j define δg=∑D(gj)aj+g′(a)b\delta g=\sum D(g_j)a^j+g'(a)b. Expanding polynomial products shows δ(gq)=g(a)δq+q(a)δg\delta(gq)=g(a)\delta q+q(a)\delta g. The choice of bb gives δh=0\delta h=0, and hence δ(qh)=0\delta(qh)=0 for every qq. The evaluation kernel is exactly (h)(h), by the minimal-polynomial proof. Thus δ\delta descends to a derivation of E[a]=E(a)E[a]=E(a) extending DD and sending aa to bb. Any extension must have that value, by differentiating h(a)=0h(a)=0, and the polynomial and quotient rules then give uniqueness. Adjoining finitely many algebraic generators successively proves existence and uniqueness over every finite algebraic extension in characteristic zero. Such an extension is generated as a field by any finite vector-space basis, since a field containing the basis contains all its linear combinations. If the original derivation vanishes on kk, each extension still does. These arguments justify both the rational-function derivative and its algebraic extension in the critical-values proof.