Steinitz Primitive Element Criterion¶

A finite field extension has a single generator exactly when it has finitely many intermediate fields.

Version: v1 · 2026-10-17 · History
Domain-specific #: 14315
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomain: Field Theory → Mathematics
Aliases: Steinitz's theorem (field theory)

Core Idea¶

For a finite field extension \(L/K\), Steinitz's criterion says that \(L=K(\alpha)\) for some \(\alpha\in L\) if and only if there are only finitely many fields \(M\) between \(K\) and \(L\). One side asks whether a single element generates the whole extension; the other asks about the size of its complete intermediate-field interval. The theorem equates them without assuming that the extension is separable or normal.^[1]^[2]

This is more precise than the familiar statement that every finite separable extension has a primitive element. A finite purely inseparable extension can also be simple, while another finite purely inseparable extension need not be. The criterion distinguishes them through their intermediate fields.^[1]^[3]^[2]

Structural Signature¶

Finite carrier: \([L:K]<\infty\) is the theorem's stated domain. Neither an arbitrary infinite extension nor an unqualified relation between algebraic structures is being classified.^[1]
One-generator predicate: \(L=K(\alpha)\) for some \(\alpha\in L\). “Simple” refers to generation as a field over \(K\), not to separability or to a claim that \(\alpha\) is unique.^[1]
Complete interval predicate: The set \(\{M:K\subseteq M\subseteq L\}\) is finite. Seeing a few named subfields does not establish that all intermediate fields have been counted.^[1]
Biconditional: Each predicate implies the other. A one-way sufficient condition, including finite separability, is not the full Steinitz criterion.^[1]^[2]

The proof exposes why the directions differ. If \(L=K(\alpha)\), an intermediate field is recovered from coefficients of a suitable monic factor of \(\alpha\)'s minimal polynomial; only finitely many such factors occur. Conversely, over an infinite \(K\), finitely many proper intermediate fields are finitely many proper \(K\)-vector subspaces of \(L\), which cannot cover \(L\). Choose \(\alpha\) outside their union. Over a finite \(K\), use the cyclic multiplicative group of the finite field \(L\) instead. The subspace argument is not a finite-field proof.^[1]^[2]

Recognition test: Has the extension been shown finite, and has either a generator for all of \(L\) or the finiteness of the entire intermediate interval been established? If the argument only exhibits separability, invokes Galois correspondence without normality, or lists a few subfields, it has not yet applied this full criterion.

What It Is Not¶

The criterion does not say that every finite extension is simple. In characteristic \(p\), \(\mathbb F_p(x,y)/\mathbb F_p(x^p,y^p)\) has degree \(p^2\), while every \(\alpha\) in the upper field satisfies \(\alpha^p\) in the base and thus generates a subextension of degree at most \(p\). No one element generates the full extension. By the criterion, its intermediate-field interval is infinite.^[2]^[1]

Nor does “simple” mean “separable.” The degree-\(p\) extension \(\mathbb F_p(t^{1/p})/\mathbb F_p(t)\) is generated by \(t^{1/p}\), even though it is purely inseparable. Conversely, the existence of the two-variable non-simple example does not make inseparability a sufficient test for non-simplicity.^[3]^[2]

The theorem is not the Galois correspondence. A finite separable extension need not be normal, and the proof of the criterion does not require treating every extension as Galois. It is also distinct from the graph-theory theorem bearing Steinitz's name; that title collision supplies no field-theoretic role.^[1]^[2]

Scope of Application¶

The equivalence applies to finite extensions over any base field. The base may be infinite or finite, and the extension may be separable or inseparable. What changes between finite and infinite base fields is the converse proof method, not the theorem statement.^[1]

For \(\mathbb Q(\sqrt2)/\mathbb Q\), \(X^2-2\) is irreducible, so the extension has degree two and the displayed square root generates it. Any intermediate \(M\) has \(2=[L:M][M:K]\), hence is either the base or the top. This gives a fully checked characteristic-zero positive case.^[4]^[1]

For \(K=\mathbb F_p(t)\) and \(L=K(t^{1/p})\), \(X^p-t\) is irreducible over \(K\), and \(L/K\) is purely inseparable of prime degree \(p\). The same tower-law argument leaves only \(K\) and \(L\) as intermediate fields. This positive case is unlike the separable quadratic example, even though both have a prime-degree interval.^[3]^[4]^[1]

Clarity¶

Keep degree, number of generators, and number of intermediate fields separate. Degree two or \(p\) makes an individual interval easy to exhaust by the tower law; finite degree in general does not force a single generator. The degree-\(p^2\) counterexample shows the difference.^[4]^[2]

Keep the if and only if directions separate as well. The factor-and-coefficient argument starts with a generator and proves finitely many intermediate fields. The converse starts with a finite interval and finds a generator. For infinite base fields it avoids a finite union of proper subspaces; for finite base fields it invokes finite-field multiplicative cyclicity. Neither proof branch licenses an assertion that all extensions are separable.^[1]^[2]

Manages Complexity¶

The theorem offers two equivalent ways to recognize simplicity in a finite extension. An explicit primitive element may be easier to find in one problem. In another, a characterization of the entire interval of intermediate fields can establish that one exists without first guessing it. This is a logical alternative in a proof, not a promise that either route is always computationally easy.^[1]

The two-variable inseparable counterexample keeps the criterion honest: a finite vector-space dimension does not by itself compress the field into one generator. Its failure of single generation and its infinite intermediate interval are the linked outcomes of the theorem.^[2]^[1]

Abstract Reasoning¶

Fix the domain: verify that \(L/K\) is finite.^[1]
Test one-element generation: identify a proposed \(\alpha\) and show \(K(\alpha)=L\), or prove that each possible \(K(\alpha)\) is proper.^[1]^[2]
Test the whole interval: classify all \(M\) with \(K\subseteq M\subseteq L\), not just familiar examples. Prime degree lets the tower law close this step immediately.^[4]
Use the direction actually warranted: a generator gives a finite interval through minimal-polynomial factors. A finite interval gives a generator by the appropriate infinite-base or finite-base proof.^[1]
State the scope: separability supplies a common sufficient condition but is not part of the biconditional hypothesis.^[1]^[3]

Knowledge Transfer¶

The same role map works in the two unlike positive cases. In \(\mathbb Q(\sqrt2)/\mathbb Q\), irreducibility gives degree two, the square root is a generator, and the prime-degree tower law exhausts the interval. In \(\mathbb F_p(t^{1/p})/\mathbb F_p(t)\), irreducibility gives degree \(p\), the root is a generator, and the tower law again exhausts the interval despite inseparability.^[4]^[3]

It also diagnoses the negative case without changing the theorem. In \(\mathbb F_p(x,y)/\mathbb F_p(x^p,y^p)\), every one-element subextension is too small to equal the degree-\(p^2\) top field; the criterion then forces infinitely many intermediate fields. Do not transfer the first two cases' prime-degree shortcut to this extension merely because all three have finite degree.^[2]^[1]

Examples¶

Characteristic-zero positive case. Let \(K=\mathbb Q\), \(L=\mathbb Q(\sqrt2)\). Irreducibility of \(X^2-2\) gives \([L:K]=2\). The element \(\sqrt2\) generates \(L\); if \(K\subseteq M\subseteq L\), then \([L:M][M:K]=2\), so \(M=K\) or \(M=L\). Both sides of Steinitz's equivalence hold.^[4]^[1]

Inseparable positive case. Let \(K=\mathbb F_p(t)\), \(L=K(t^{1/p})\). The polynomial \(X^p-t\) is irreducible; its derivative vanishes, so this is a purely inseparable degree-\(p\) extension. The root generates \(L\), and any intermediate \(M\) has degree divisor of \(p\), leaving only the endpoints. Both sides hold without separability.^[3]^[4]^[1]

Finite but non-simple case. Let \(L=\mathbb F_p(x,y)\) and \(K=\mathbb F_p(x^p,y^p)\). The degree is \(p^2\), but each \(\alpha\in L\) has \(\alpha^p\in K\), hence \([K(\alpha):K]\le p\). No \(\alpha\) generates \(L\), and by the theorem the interval of intermediate fields is infinite. This is a consequence of the equivalence, not a claim that the cited notes print a list of all those fields.^[2]^[1]

Structural Tensions¶

The theorem does not itself require balancing two opposing pressures. It gives a fixed classification boundary. Finite degree alone is not enough to infer a primitive element, as the degree-\(p^2\) example shows; inseparability alone is not enough to deny one, as the degree-\(p\) example shows. Those are contrasting cases and diagnostic cautions, not a tradeoff within a single extension.^[2]^[3]^[1]

The proof likewise has a necessary case distinction, not a competing design choice. The infinite-base finite-union argument and finite-base cyclicity argument establish the same converse in different settings. Applying the former to a finite field, or treating the latter as the general explanation of inseparable examples over infinite bases, leaves part of the theorem unjustified.^[1]^[2]

Structural–Framed Character¶

Evaluative weight: The criterion is a descriptive mathematical equivalence; simplicity here is a technical property, not a value judgment. Human-practice dependence: Mathematicians choose notation and proofs, but the relation between a specified \(L/K\), its generators, and its intermediate fields does not depend on who names it. Institutional origin: The theorem carries a historical name and appears in mathematical literature, yet no institution creates its truth by rule.^[1]^[2]

Vocabulary travel: “Finite extension,” “generated field,” and “intermediate field” are field-theoretic terms; a looser metaphor about compact description is not the same theorem. Import versus recognition: Applying the criterion recognizes properties already true or false of \(L/K\); writing \(L=K(\alpha)\) does not make a nonexistent generator appear. Its character: predominantly structural within a field-theoretic proof frame, because its two-way equivalence holds across unlike finite extensions while its roles and warrants remain specific to fields.^[1]

Structural Core vs. Domain Accent¶

The transferable Motif is a relation between compact generation and a constrained interval of intermediate structures. Here the actual theorem needs fields, finite extension degree, generated subfields, intermediate fields, and the two proof directions. Remove those, and the named Steinitz criterion no longer applies.^[1]

A broader Prime would require the same full two-way signature and boundary behavior in unrelated carriers, rather than merely a resemblance to “one thing generates many.” Neither these field examples nor a generic logical biconditional proves that wider pattern. The current entry therefore remains domain-specific while leaving that future comparison open.

This entry is part of Necessity and Sufficiency and presupposes Field Extension.

The carrier is a field extension: without a specified embedding of \(K\) into \(L\), generation over \(K\), \([L:K]\), and intermediate \(K\)-fields are undefined. The theorem requires that carrier but is itself a claim about extensions, not another kind of extension. A Galois extension is a more restrictive type, and Galois theory may offer other tools for studying subfields, but neither is this criterion's identity. The criterion applies even to the simple and non-simple purely inseparable examples that a Galois-only account would miss.^[1]^[3]^[2]

The theorem also realizes a necessary-and-sufficient relation: for finite \(L/K\), having one generator is both required for and guaranteed by having finitely many intermediate fields.^[1] That logical relation is wider than field theory; the Steinitz theorem adds the field-specific predicates and proofs that make both directions true here. An equivalence relation on a set, which partitions the set into classes, is a different structure despite the shared word “equivalence.”

Relationships to Other Abstractions¶

Current abstraction Steinitz Primitive Element Criterion Domain-specific

Parents (2) — more general patterns this builds on

Steinitz Primitive Element Criterion presupposes Field Extension Domain-specific

The theorem requires a specified base-field embedding K into L before generation, degree and intermediate K-fields are defined.
Steinitz Primitive Element Criterion is part of Necessity and Sufficiency Prime

The Steinitz iff theorem contains a two-way necessary-and-sufficient relation between single generation and finite intermediate-field count.

Hierarchy paths (2) — routes to 2 parentless roots

Steinitz Primitive Element Criterion → Field Extension

Show alternative path (1)

Neighborhood in Abstraction Space¶

Steinitz Primitive Element Criterion sits in a sparse region of the domain-specific corpus (66^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2721 abstractions)

Nearest neighbors

Complete Heyting algebra — 0.85
Bounded arithmetic — 0.84
Local class field theory — 0.84
Finite extensions of local fields — 0.84
Algebraic number field — 0.83

Computed from structural-signature embeddings · 2026-10-17

Not to Be Confused With¶

The separable primitive-element corollary: finite separable extensions are simple, but Steinitz's iff criterion also classifies inseparable cases.^[1]
Galois correspondence: normality cannot be silently added to all finite separable extensions, and the criterion has a direct proof.^[1]
Every finite extension is simple: the two-variable characteristic-\(p\) extension is a counterexample.^[2]
Steinitz's graph theorem: a different namesake result, not a field-extension statement.

References¶

[1] The Stacks Project Authors, Fields, undated online edition, §9.19, Lemma 9.19.1, theorem and complete proof, https://stacks.math.columbia.edu/tag/09HZ. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w ↩x ↩y ↩z ↩27 ↩28 ↩29 ↩30 ↩31 ↩32

[2] Zheng, Weizhe (2025), Lectures on Algebra, course notes dated January 10, 2025, Theorem 1.7.4 (Steinitz) and Example 1.7.3, physical PDF pp. 24–25 (viewer zero-based P23–24; printed pp. 18–19), https://server.mcm.ac.cn/~zheng/algebra.pdf. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r

[3] The Stacks Project Authors, Fields, undated online edition, §9.14, Lemma 9.14.2, purely inseparable degree-\(p\) construction, https://stacks.math.columbia.edu/tag/09HD. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h

[4] The Stacks Project Authors, Fields, undated online edition, §9.7, Example 9.7.6 and Lemma 9.7.7, degree of a simple extension and tower law, https://stacks.math.columbia.edu/download/book.pdf. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g