A Functorial View of Polynomial Irreducibility Test Rings, Lifting Obstructions, and Limit Diagrams

Factorization Functors and Irreducibility1. The Factorization Functor2. Irreducibility as Emptiness3. The Monic Version4. Representability5. Test Rings and Test Diagrams6. A Basic Example7. Eisenstein as a Square-Zero Test8. Fiber Products: Incompatible Factorizations9. Equalizers and Galois Descent10. Inverse Limits and Compatible Lifting

 

Factorization Functors and Irreducibility

A polynomial factorization can be viewed as a solution to a system of equations in the coefficients of its factors. Changing the coefficient ring transports these solutions. This gives a simple framework for irreducibility: construct a ring in which the relevant solution set is empty, and transfer this obstruction back to the original ring.

Throughout, rings are commutative with identity. Fix a UFD $R$$R$, its fraction field $K$$K$, and a primitive polynomial

$$f(x)=a_n{x}^n+\cdots +a_0\in R[x],a_n\ne 0,n\ge 2.$$

Here primitive means that the greatest common divisor of the coefficients is a unit. Test algebras need not be domains or UFDs. To avoid conventions about the degree of a polynomial over the zero ring, we work with nonzero $R$$R$-algebras and nonempty small diagrams.

1. The Factorization Functor

For an $R$$R$-algebra $A$$A$, let $f_A$$f_A$ denote the image of $f$$f$ in $A[x]$$A[x]$, and write

$$A[x{]}_{\le d}=\{\sum _{i=0}^d{b}_i{x}^i:b_i\in A\}.$$

This set includes the zero polynomial.

For $1\le e<n$$1\le e<n$, define

$${\tilde{D}}_{f,e}(A)=\{(g,h)\in A[x{]}_{\le e}\times A[x{]}_{\le n-e}:gh=f_A\}.$$

A point of this set is an ordered factorization with prescribed degree bounds.

If $\phi :A\to B$$\varphi:A\to B$ is an $R$$R$-algebra homomorphism, apply $\phi$$\varphi$ coefficientwise:

$${\tilde{D}}_{f,e}(\phi )(g,h)=({\phi }_{\ast }(g),{\phi }_{\ast }(h)).$$

This is well-defined because

$$gh=f_A⟹{\phi }_{\ast }(g){\phi }_{\ast }(h)=f_B,$$

and applying a ring homomorphism cannot increase degree. These maps respect identities and composition, so ${\tilde{D}}_{f,e}$$\widetilde D_{f,e}$ is a covariant functor from nonzero $R$$R$-algebras to sets.

Degree bounds are essential. A ring homomorphism may kill a leading coefficient, so exact degrees need not survive a change of rings. The bounded-degree definition retains these specialized factorizations.

2. Irreducibility as Emptiness

Although degrees may drop over a test algebra, they cannot drop at a point of ${\tilde{D}}_{f,e}(R)$$\widetilde D_{f,e}(R)$. Indeed, since $R$$R$ is a domain,

$$n=\mathrm{deg}f=\mathrm{deg}g+\mathrm{deg}h.$$

Together with the bounds $\mathrm{deg}g\le e$$\deg g\le e$ and $\mathrm{deg}h\le n-e$$\deg h\le n-e$, this forces

$$\mathrm{deg}g=e,\mathrm{deg}h=n-e.$$

Thus ${\tilde{D}}_{f,e}(R)$$\widetilde D_{f,e}(R)$ records precisely the factorizations of degrees $e$$e$ and $n-e$$n-e$.

Proposition. The following are equivalent:

1. $f$$f$ is irreducible in $K[x]$$K[x]$.

2. $f$$f$ is irreducible in $R[x]$$R[x]$.

3. ${\tilde{D}}_{f,e}(R)=\varnothing$$\widetilde D_{f,e}(R)=\varnothing$ for every $1\le e\le \lfloor n/2\rfloor$$1\le e\le\lfloor n/2\rfloor$.

Proof. The first two conditions are equivalent by Gauss's lemma. Since $f$$f$ is primitive, a constant factor in $R[x]$$R[x]$ must be a unit: otherwise it would be a nonunit common divisor of all coefficients. Hence reducibility is equivalent to a factorization into two positive-degree polynomials.

Exchanging the factors gives a natural bijection

$${\tilde{D}}_{f,e}(A)\cong {\tilde{D}}_{f,n-e}(A).$$

It therefore suffices to check degrees at most $\lfloor n/2\rfloor$$\lfloor n/2\rfloor$. $◻$$\square$

Degree zero is excluded deliberately. If the definition were extended to $e=0$$e=0$, the factorization $(1,f_A)$$(1,f_A)$ would always give a point, so emptiness would not express the desired obstruction.

3. The Monic Version

When $f$$f$ is monic, define

$$\begin{array}{c}{D}_{f,e}(A)=\{(g,h)\in A[x{]}^2:\begin{array}{l}g,h\text{ are monic},\\ \mathrm{deg}g=e,\mathrm{deg}h=n-e,\\ gh=f_A\end{array}\}.\end{array}$$

This is again a covariant functor. Monic leading coefficients remain equal to $1$$1$ under every change of rings, so exact degrees are preserved.

The two versions are related by a natural isomorphism

$${\tilde{D}}_{f,e}\cong {\mathbb{G}}_m\times D_{f,e},{\mathbb{G}}_m(A)=A^{\times }.$$

Explicitly,

$$(u,(g,h))⟼(ug,u^{-1}h).$$

To see this, the top coefficient equation for a bounded-degree factorization of a monic polynomial is

$$g_e{h}_{n-e}=1.$$

Both leading coefficients are therefore units, and the factors can be normalized to be monic. In particular,

$${\tilde{D}}_{f,e}(A)=\varnothing ⟺D_{f,e}(A)=\varnothing .$$

4. Representability

The factorization functors are represented by explicit coefficient algebras.

Introduce universal polynomials

$$U(x)=\sum _{i=0}^e{u}_i{x}^i,V(x)=\sum _{j=0}^{n-e}{v}_j{x}^j,$$

and set

$${\tilde{C}}_{f,e}=\frac{R[u_0,\dots ,u_e,v_0,\dots ,v_{n-e}]}{([x^k](UV-f):0\le k\le n)}.$$

Here $[x^k](UV-f)$$[x^k](UV-f)$ denotes the coefficient of $x^k$$x^k$. The ideal is generated by these coefficient equations.

Giving an $R$$R$-algebra homomorphism ${\tilde{C}}_{f,e}\to A$$\widetilde C_{f,e}\to A$ is exactly giving coefficients for $g$$g$ and $h$$h$ satisfying $gh=f_A$$gh=f_A$. Hence, naturally in $A$$A$,

$${\tilde{D}}_{f,e}(A)\cong {\mathrm{Hom}}_{R\text{-alg}}({\tilde{C}}_{f,e},A).$$

The representing algebra is nonzero: a factorization of the required degrees exists over an algebraic closure of $K$$K$.

For monic $f$$f$, the monic functor is represented by

$$C_{f,e}={\tilde{C}}_{f,e}/(u_e-1,v_{n-e}-1).$$

Equivalently, $\mathrm{Spec}{\tilde{C}}_{f,e}$$\operatorname{Spec}\widetilde C_{f,e}$ and $\mathrm{Spec}{C}_{f,e}$$\operatorname{Spec}C_{f,e}$ are affine schemes whose points over $A$$A$ are the corresponding factorizations.

5. Test Rings and Test Diagrams

Write $F_e$$F_e$ for either ${\tilde{D}}_{f,e}$$\widetilde D_{f,e}$ or, in the monic case, $D_{f,e}$$D_{f,e}$.

The structural homomorphism $R\to A$$R\to A$ induces

$$F_e(R)⟶F_e(A).$$

Since a nonempty set cannot map to the empty set,

$$\boxed{{\displaystyle F_e(A)=\varnothing ⟹F_e(R)=\varnothing .}}$$

Consequently, if for each $1\le e\le \lfloor n/2\rfloor$$1\le e\le\lfloor n/2\rfloor$ there is an $R$$R$-algebra $A_e$$A_e$ such that $F_e(A_e)=\varnothing$$F_e(A_e)=\varnothing$, then $f$$f$ is irreducible over $K$$K$.

Different degrees may use different test algebras. A nonempty test set simply means that this particular test has not ruled out the degree; its points need not come from $R$$R$.

This argument uses only functoriality. Representability gives an additional tool: preservation of limits.

Let $(A_i)$$(A_i)$ be a nonempty small diagram of nonzero $R$$R$-algebras, with all arrows respecting their $R$$R$-algebra structures. Then

$$F_e(\underset{i}{\underleftarrow{\lim }}{A}_i)\cong \underset{i}{\underleftarrow{\lim }}{F}_e(A_i).$$

The structure maps from $R$$R$ form a cone, so

$$F_e(R)⟶\underset{i}{\underleftarrow{\lim }}{F}_e(A_i).$$

Therefore,

$$\boxed{{\displaystyle \underset{i}{\underleftarrow{\lim }}{F}_e(A_i)=\varnothing ⟹F_e(R)=\varnothing .}}$$

For instance,

$$F_e(A\times B)\cong F_e(A)\times F_e(B),$$

whereas

$$F_e(A{\times }_{C}B)\cong F_e(A){\times }_{F_e(C)}{F}_e(B).$$

The second formula requires the two factorizations to agree after mapping to $C$$C$. Thus all the individual solution sets may be nonempty while their fiber product is empty.

Fixing the degree matters. For monic $f$$f$, if

$$D_f(A)=\coprod _{e=1}^{n-1}{D}_{f,e}(A),$$

then

$$D_f(A\times B)\cong \coprod _{e=1}^{n-1}(D_{f,e}(A)\times D_{f,e}(B)),$$

which is generally not $D_f(A)\times D_f(B)$$D_f(A)\times D_f(B)$: the two components must use the same degree.

6. A Basic Example

Consider the primitive polynomial

$$f(x)=5x^4+x^2+2x+3\in \mathbb{Z}[x].$$

Only degrees $e=1,2$$e=1,2$ need to be excluded.

Modulo $2$$2$,

$$f(x)\equiv (x^2+x+1{)}^2.$$

The quadratic $x^2+x+1$$x^2+x+1$ is irreducible over ${\mathbb{F}}_2$$\mathbb F_2$, so

$${\tilde{D}}_{f,1}({\mathbb{F}}_2)=\varnothing .$$

Modulo $3$$3$,

$$f(x)\equiv 2x(x^3+2x+1).$$

For every $a\in {\mathbb{F}}_3$$a\in\mathbb F_3$, one has $a^3+2a+1=1$$a^3+2a+1=1$. Thus the cubic has no root and is irreducible. The only proper factor degrees are $1$$1$ and $3$$3$, giving

$${\tilde{D}}_{f,2}({\mathbb{F}}_3)=\varnothing .$$

The leading coefficient survives both reductions, so the bounded-degree conditions force exact degrees in these test rings. Functoriality now excludes both possible degrees over $\mathbb{Z}$$\mathbb Z$, proving that $f$$f$ is irreducible over $\mathbb{Q}$$\mathbb Q$.

Alternatively, the Chinese remainder theorem combines the tests:

$$\mathbb{Z}/6\mathbb{Z}\cong {\mathbb{F}}_2\times {\mathbb{F}}_3,$$

and therefore

$${\tilde{D}}_{f,e}(\mathbb{Z}/6\mathbb{Z})\cong {\tilde{D}}_{f,e}({\mathbb{F}}_2)\times {\tilde{D}}_{f,e}({\mathbb{F}}_3)=\varnothing (e=1,2).$$

The arithmetic work lies in finding useful test algebras and proving that the relevant solution sets, or sets of compatible solutions, are empty. Functoriality transfers these obstructions to the original ring; representability identifies compatible solutions with solutions over a limit ring.

7. Eisenstein as a Square-Zero Test

Eisenstein's criterion is a special case of the test-ring principle: every candidate factorization is ruled out over a quotient by the square of a prime ideal.

Proposition (Eisenstein). Let $f=\sum _{i=0}^n{a}_i{x}^i\in R[x]$$f=\sum_{i=0}^{n}a_ix^i\in R[x]$ be primitive. Suppose that a prime ideal $\mathfrak{p}\subset R$$\mathfrak p\subset R$ satisfies

$$a_n\notin \mathfrak{p},a_i\in \mathfrak{p}(0\le i<n),a_0\notin {\mathfrak{p}}^2.$$

Then

$${\tilde{D}}_{f,e}(R/{\mathfrak{p}}^2)=\varnothing (1\le e<n),$$

and consequently $f$$f$ is irreducible over $K$$K$.

Proof. Consider

$$R⟶A:=R/{\mathfrak{p}}^2⟶B:=R/\mathfrak{p}.$$

The kernel $I=\mathfrak{p}/{\mathfrak{p}}^2$$I=\mathfrak p/\mathfrak p^2$ of $A\to B$$A\to B$ satisfies $I^2=0$$I^2=0$.

Suppose $(g,h)\in {\tilde{D}}_{f,e}(A)$$(g,h)\in\widetilde D_{f,e}(A)$. Reducing to the domain $B$$B$ gives

$$\bar{g}\bar{h}={\bar{a}}_n{x}^n.$$

Since ${\bar{a}}_n\ne 0$$\overline a_n\ne0$, the degree bounds force $\mathrm{deg}\bar{g}=e$$\deg\overline g=e$ and $\mathrm{deg}\bar{h}=n-e$$\deg\overline h=n-e$. A factor of a nonzero scalar multiple of $x^n$$x^n$ over a domain is itself a scalar multiple of a power of $x$$x$, as can be checked over its fraction field. Thus

$$\bar{g}=b{x}^e,\bar{h}=c{x}^{n-e}$$

for nonzero $b,c\in B$$b,c\in B$. Both constant terms vanish in $B$$B$, so $g(0),h(0)\in I$$g(0),h(0)\in I$. Hence

$$f_A(0)=g(0)h(0)=0$$

because $I^2=0$$I^2=0$. This contradicts $a_0\notin {\mathfrak{p}}^2$$a_0\notin\mathfrak p^2$. $◻$$\square$

For $R=\mathbb{Z}$$R=\mathbb Z$ and $\mathfrak{p}=(p)$$\mathfrak p=(p)$, these are precisely the usual Eisenstein conditions. The test ring is $\mathbb{Z}/p^2\mathbb{Z}$$\mathbb Z/p^2\mathbb Z$. For example, $x^6-2$$x^6-2$ has no bounded-degree proper factorization over $\mathbb{Z}/4\mathbb{Z}$$\mathbb Z/4\mathbb Z$.

The mechanism is a lifting obstruction: factorizations exist after reduction to $B$$B$, but none can lift to the specified polynomial over $A$$A$.

8. Fiber Products: Incompatible Factorizations

For a cospan of $R$$R$-algebras

$$A_1⟶B⟵A_2,$$

the limit is the fiber product

$$P=A_1{\times }_B{A}_2=\{(a_1,a_2):a_1\text{ and }{a}_2\text{ have the same image in }B\}.$$

It fits into the pullback square

$$\begin{array}{ccc}P& ⟶& A_1\\ \downarrow & & \downarrow \\ A_2& ⟶& B.\end{array}$$

Preservation of limits gives

$$F_e(P)\cong F_e(A_1){\times }_{F_e(B)}{F}_e(A_2).$$

Thus points on both branches are insufficient: their images in $F_e(B)$$F_e(B)$ must agree.

Example. Let $R=\mathbb{Q}[t]$$R=\mathbb Q[t]$ and $f=x^2-t$$f=x^2-t$. Specify the following $R$$R$-algebras:

Algebra	Definition	Image of $t$$t$
$A_1$$A_1$	$\mathbb{Q}[{\epsilon }_1]/({\epsilon }_1^3)$$\mathbb Q[\varepsilon_1]/(\varepsilon_1^3)$	${\epsilon }_1^2$$\varepsilon_1^2$
$A_2$$A_2$	$\mathbb{Q}[{\epsilon }_2]/({\epsilon }_2^3)$$\mathbb Q[\varepsilon_2]/(\varepsilon_2^3)$	$4{\epsilon }_2^2$$4\varepsilon_2^2$
$B$$B$	$\mathbb{Q}[\delta ]/({\delta }^2)$$\mathbb Q[\delta]/(\delta^2)$	$0$$0$

The maps $A_i\to B$$A_i\to B$ send ${\epsilon }_i$$\varepsilon_i$ to $\delta$$\delta$. They respect the $R$$R$-algebra structures because both images of $t$$t$ become zero.

For any $R$$R$-algebra $A$$A$,

$$D_{f,1}(A)\cong \{r\in A:r^2=t_A\},r⟼(x-r,x+r).$$

In $A_1$$A_1$, all roots are

$$r=\pm {\epsilon }_1+c{\epsilon }_1^2,c\in \mathbb{Q},$$

so their images in $B$$B$ are $\delta$$\delta$ and $-\delta$$-\delta$. In $A_2$$A_2$, all roots are

$$r=\pm 2{\epsilon }_2+c{\epsilon }_2^2,$$

whose images are $2\delta$$2\delta$ and $-2\delta$$-2\delta$. These two image sets are disjoint. Therefore

$$D_{f,1}(A_1){\times }_{D_{f,1}(B)}{D}_{f,1}(A_2)=\varnothing .$$

The structural map

$$R⟶P,t⟼({\epsilon }_1^2,4{\epsilon }_2^2),$$

then implies $D_{f,1}(R)=\varnothing$$D_{f,1}(R)=\varnothing$. Hence $x^2-t$$x^2-t$ is irreducible over $\mathbb{Q}(t)$$\mathbb Q(t)$.

Every node of the cospan has factorizations, but there is no compatible family of factorizations.

9. Equalizers and Galois Descent

Let $\alpha ,\beta :A\to B$$\alpha,\beta:A\to B$ be $R$$R$-algebra homomorphisms. Their equalizer is

$$E=\{a\in A:\alpha (a)=\beta (a)\},$$

with diagram

$$E⟶A\stackrel{\alpha }{\underset{\beta }{\rightrightarrows }}B.$$

The factorization functor gives

$$F_e(E)\cong \{z\in F_e(A):F_e(\alpha )(z)=F_e(\beta )(z)\}.$$

Thus a factorization over $A$$A$ may fail to descend because its two images disagree.

For a finite Galois extension $L/K$$L/K$ with group $G$$G$, the action diagram has limit $L^G=K$$L^G=K$. Consequently, for a monic polynomial $f\in K[x]$$f\in K[x]$,

$$D_{f,e}(K)\cong D_{f,e}(L{)}^G,$$

where $G$$G$ acts coefficientwise on both factors.

If $L$$L$ splits $f$$f$ and $f$$f$ is separable, with root set $\mathrm{\Omega }$$\Omega$, then

$$D_{f,e}(L)\cong \{S\subseteq \mathrm{\Omega }:|S|=e\},$$

by sending $S$$S$ to the ordered pair

$$(\prod _{\alpha \in S}(x-\alpha ),\prod _{\alpha \in \mathrm{\Omega }\setminus S}(x-\alpha )).$$

This correspondence is $G$$G$-equivariant. Hence factorizations over $K$$K$ correspond to $G$$G$-invariant subsets, recovering

$$f\text{ is irreducible over }K⟺G\text{ acts transitively on }\mathrm{\Omega }.$$

Example. Let $f=x^2-2$$f=x^2-2$, $L=\mathbb{Q}(\sqrt{2})$$L=\mathbb Q(\sqrt2)$, and let $\sigma (\sqrt{2})=-\sqrt{2}$$\sigma(\sqrt2)=-\sqrt2$. Then

$$\mathbb{Q}=\mathrm{Eq}\left(L\stackrel{\mathrm{id}}{\underset{\sigma }{\rightrightarrows }}L\right).$$

The two points of $D_{f,1}(L)$$D_{f,1}(L)$ are

$$(x-\sqrt{2},x+\sqrt{2}),(x+\sqrt{2},x-\sqrt{2}).$$

They are exchanged by $\sigma$$\sigma$, so the equalizer of $\mathrm{id}$$\mathrm{id}$ and $D_{f,1}(\sigma )$$D_{f,1}(\sigma)$ is empty. This gives $D_{f,1}(\mathbb{Q})=\varnothing$$D_{f,1}(\mathbb Q)=\varnothing$.

10. Inverse Limits and Compatible Lifting

For a monic polynomial $f\in \mathbb{Z}[x]$$f\in\mathbb Z[x]$, consider the inverse system

$$\cdots ⟶\mathbb{Z}/p^3\mathbb{Z}⟶\mathbb{Z}/p^2\mathbb{Z}⟶\mathbb{Z}/p\mathbb{Z}.$$

Its limit is ${\mathbb{Z}}_p$$\mathbb Z_p$, so

$$D_{f,e}({\mathbb{Z}}_p)\cong \underset{k\ge 1}{\underleftarrow{\lim }}{D}_{f,e}(\mathbb{Z}/p^k\mathbb{Z}).$$

A point on the right is a sequence of factorizations whose coefficients agree under every reduction map.

Each finite-level solution set is finite. An inverse system of finite nonempty sets has a nonempty limit; surjectivity of the transition maps is not required. This follows from compactness, or from the finite branching argument for compatible partial sequences. Therefore,

$$\boxed{{\displaystyle D_{f,e}({\mathbb{Z}}_p)=\varnothing ⟺D_{f,e}(\mathbb{Z}/p^k\mathbb{Z})=\varnothing \text{ for some }k.}}$$

For a fixed degree, the absence of a $p$$p$-adic factorization is thus detectable at finite precision.

Example. Take $f=x^2-5$$f=x^2-5$ and $p=2$$p=2$. There are roots modulo $2$$2$ and modulo $4$$4$, but none modulo $8$$8$: every square modulo $8$$8$ belongs to $\{0,1,4\}$$\{0,1,4\}$. Hence

$$D_{f,1}(\mathbb{Z}/2\mathbb{Z})\ne \varnothing ,D_{f,1}(\mathbb{Z}/4\mathbb{Z})\ne \varnothing ,D_{f,1}(\mathbb{Z}/8\mathbb{Z})=\varnothing .$$

The third level obstructs every possible compatible sequence, so $D_{f,1}({\mathbb{Z}}_2)=\varnothing$$D_{f,1}(\mathbb Z_2)=\varnothing$.

A limit diagram can always be combined into its limit ring. Its usefulness is that it expresses one test through simpler calculations: simultaneous existence for products, agreement for fiber products, invariance for equalizers, and compatible lifting for inverse limits.

The arithmetic work lies in finding useful test algebras and proving that the relevant solution sets, or sets of compatible solutions, are empty. Functoriality transfers these obstructions to the original ring; representability identifies compatible solutions with solutions over a limit ring.