KEVOS
ArticlesServicesCase studiesAboutContact
ArticlesServicesCase studiesAboutContact
← ArticlesOrderings on Division RingsEngineering · Engineering MathematicsLesson 462/884← PrevNext →
ArticlePublished 8 Aug 202619 min readBy KEVOS®
On this page

Ask about this page

KEVOS AIOrderings on Division Rings

KEVOS knowledge first · trusted web sources when needed

Skip to content

Engineering Mathematics Advanced Ordered division rings

Ordered Division Rings

In a division ring an ordering is nothing more than an additively closed subgroup of index 2 in D∗, a preordering is an additively closed subgroup containing every square, and D is orderable exactly when −1 is not a sum of square-products.

Page ID
KEVOS-ENG-MATH-NCR-0134
Taxonomy
ENG / ENG-MATH
Collection
noncommutative-rings-core
Source
(18.1)–(18.2), §18 (pp. 285–286)
Reviewed
2026-08-08
Version
1.0.0

01Executive Summary

For a general ring the ordering axioms of §17 are awkward: the multiplicative axiom has to be stated with permuted products per(a12⋯am2t1⋯tn) because there is no way to move a factor past its neighbours. In a division ring every nonzero element is invertible, and that single fact collapses the whole apparatus.

An ordering P⊆D∗ becomes a normal subgroup of index 2; a preordering T becomes a normal subgroup containing every square and every commutator, with D∗/T elementary abelian. The orderability criterion becomes a single arithmetic statement: D can be ordered if and only if −1 is not a sum of square-products.

[D∗:P]=2An ordering as a subgroup
3 axiomsPreordering test (18.1)
Exponent 2The group D∗/T
−1∉T(D)Orderability (18.2)

02Overview

Throughout, D is a division ring and D∗=D∖{0} its multiplicative group. An ordering on D is a subset P⊆D∗ satisfying the three axioms inherited from the theory of ordered rings.

P+P⊆P,P⋅P⊆P,P∪(−P)=D∖{0}.
(17.1)–(17.3)

Setting a<biffb−a∈P recovers a total order compatible with addition and with multiplication by positive elements.

Two consequences from the general theory carry over at once. First, P∩(−P)=∅ and 1∈P, so D has characteristic 0 — an ordered division ring contains a copy of ℚ. Second, P is closed under multiplication, and because inverses are available it is closed under inversion as well: for a∈P we have a−1=a(a−1)2, and squares of nonzero elements always lie in P.

The one thing to remember

In a division ring, an ordering is an index-2 subgroup of D∗ that is closed under addition, and a preordering is a subgroup containing all squares that is closed under addition. Every result in this section is a consequence of that translation.

Because D∗=P⊔(−P), the cone P has index 2 and is therefore automatically normal. With the induced order P is itself a multiplicative ordered group, with positive cone {a∈P:a>1} — the link to the material of Ordered Groups and Group Rings and to the Mal'cev–Neumann construction that produces the standard examples.

03Learning Objectives

  • Translate the ordering axioms into group-theoretic language for a division ring.
  • State (18.1) and use its three closure conditions to test a candidate set.
  • Prove that any preordering contains [D∗,D∗], hence is normal with D∗/T of exponent 2.
  • Distinguish a square-product from a square, and explain why the distinction is invisible for fields.
  • State the Szele–Pickert theorem (18.2) and deduce the Artin–Schreier criterion for fields.
  • Decide whether ℚ, ℍ and a twisted Laurent series ring are orderable.

04Definitions

Definition(17.5)–(17.6)Preordering, specialised

A preordering in a ring R is a subset T⊆R∖{0} with T+T⊆T such that every permuted product per(a12⋯am2t1⋯tn) lies in T, for all ai∈R∖{0} and tj∈T. Here per denotes the product of the listed factors — each ai occurring twice — taken in any order.

Every ordering is a preordering, and any intersection of orderings is a preordering.

Definition—Square-product and the weak preordering

A square-product in D is an element of the form a12a22⋯am2 with all ai∈D∗; these form the subgroup Σ(D)≤D∗ generated by the squares. The weak preordering T(D) is the set of all finite sums of square-products. D is formally real when 0∉T(D).

D∗
The multiplicative group D∖{0} of a division ring D.
Σ(D)
The subgroup of D∗ generated by {a2:a∈D∗}; its elements are exactly the square-products.
T(D)
Sums of square-products. It is contained in every preordering of D, which is why it is called weak.
Totally positive
Positive with respect to every ordering of D. Characterised in Preorderings in Division Rings.
[D∗,D∗]
The commutator subgroup of D∗, generated by all aba−1b−1 with a,b∈D∗.

Orderings and preorderings are sets of nonzero elements throughout: 0 never belongs to a cone, and the sign convention is that P is the set of strictly positive elements.

05Core Concepts

Inverses do the work

The general definition of a preordering is complicated only because the two copies of each ai may be separated by arbitrary other factors, and there is no commutativity to bring them together. In a division ring the separating block u is invertible, and one identity closes the gap.

aua=(au)2(u−1)2u(a,u∈D∗),
(C.1)

Expand the right side: auauu−1u−1u=aua. Two squares are produced and the pair a,a disappears.

So a block aua inside a long permuted product may be traded for a single element (au)2(u−1)2 — a product of two squares — followed by the shortened block u. Iterating removes every matched pair. That is the whole content of (18.1).

Commutators are square-products

The same identity, applied with the roles rearranged, exhibits every multiplicative commutator as a product of three squares.

aba−1b−1=a2(a−1b)2(b−1)2(a,b∈D∗).
(C.2)

Expanding: aaa−1ba−1bb−1b−1=aba−1b−1.

Hence [D∗,D∗]⊆Σ(D)⊆T for every preordering T. This is the structural reason a preordering can never be a lopsided subset: it is forced to be normal, and D∗/T is an abelian group killed by squaring.

[D∗,D∗]⊆Σ(D)⊆any preordering T⊆an ordering P⊆D∗

Square-product versus square

In a field, a2b2=(ab)2, so Σ(F)=F∗2 and square-products add nothing. In a genuinely noncommutative D the product of two squares need not be a square, and Σ(D) is strictly larger than the set of squares. Formally Real Division Rings constructs an explicit D in which −1 is a square-product but not even a sum of squares.

06Key Results

Proposition(18.1)Preorderings in a division ring

Let D be a division ring. A subset T⊆D∗ is a preordering of D if and only if

  1. T+T⊆T;
  2. T⋅T⊆T;
  3. a2∈T for every a∈D∗.

If T is a preordering then t∈T⇒t−1∈T, and T⊇[D∗,D∗]. In particular T is a normal subgroup of D∗, and D∗/T is an abelian group of exponent 2.

Proof

Necessity. Condition (1) is the axiom (17.5). Taking m=0 in (17.6) gives t1t2∈T, which is (2); taking m=1, n=0 gives per(a2)=a⋅a=a2∈T, which is (3).

Sufficiency. Assume (1)–(3). Only (17.6) needs checking. Let W be any arrangement of the list a1,a1,…,am,am,t1,…,tn with ai∈D∗ and tj∈T; we show the value of W lies in T, by induction on m.

If m=0 then W is a product of elements of T, so W∈T by (2) (and W=1=12∈T if n=0 as well). If m≥1, choose an index i and the two positions of ai in W, and write W=w1aiuaiw3 where u is the (possibly empty) block between the two occurrences. Every letter is nonzero and D is a division ring, so u∈D∗. By (C.1),

W=w1[(aiu)2(u−1)2]uw3,(aiu)2(u−1)2∈T⋅T⊆T
(18.1a)

using (3) and (2). The right-hand side is an arrangement of a1,a1,…,am−1,am−1 together with t1,…,tn and the one extra element (aiu)2(u−1)2 of T. The inductive hypothesis applies and gives W∈T.

The remaining assertions. 1=12∈T; for t∈T, t−1=t(t−1)2∈T⋅T⊆T, so T is a subgroup of D∗. By (C.2) each commutator aba−1b−1 is a product of three squares, hence lies in T; therefore [D∗,D∗]⊆T, so T⊴D∗ and D∗/T is abelian. Finally a2∈T for all a, so every element of D∗/T squares to the identity.

Corollary—Orderings are the index-two preorderings

For a subset P⊆D∗ the following are equivalent: (i) P is an ordering of D; (ii) P is a preordering of D with [D∗:P]=2.

Proof

(i) ⇒ (ii): P+P⊆P and P⋅P⊆P are axioms, and for a∈D∗ one of ±a lies in P, so a2=(±a)(±a)∈P; thus P is a preordering by (18.1). Since D∗=P∪(−P) and P∩(−P)=∅, the coset decomposition D∗=P⊔(−1)P shows [D∗:P]=2.

(ii) ⇒ (i): −1∉P, since otherwise 0=1+(−1)∈P+P⊆P, contradicting 0∉P. So (−1)P≠P, and as [D∗:P]=2 these are the only two cosets: D∗=P⊔(−P), which is axiom (17.3). Axioms (17.1) and (17.2) hold because P is a preordering.

Lemma—Formal reality tested at −1

For a division ring D: 0∈T(D) if and only if −1∈T(D). Hence D is formally real if and only if −1 is not a sum of square-products.

Proof

If −1∈T(D) then 0=1+(−1) is a sum of square-products, since 1=12 is one. Conversely suppose 0=s1+⋯+sn with each si a square-product. As square-products are nonzero, n≥2; then −s1=s2+⋯+sn, and multiplying on the right by s1−1 gives −1=s2s1−1+⋯+sns1−1. Each sis1−1 lies in the group Σ(D), so −1∈T(D).

Theorem(18.2)Szele–Pickert orderability criterion

A division ring D admits an ordering if and only if −1 is not a sum of square-products in D.

Proof

By the Lemma, the stated condition is exactly formal reality of D. R. E. Johnson's theorem (17.11) says that for any nonzero ring, formal reality, the existence of a preordering, and the existence of an ordering are equivalent: T(D) is a preordering as soon as 0∉T(D), and Zorn's Lemma enlarges any preordering to a maximal one, which by (17.10) is an ordering. Conversely an ordering P contains T(D) and misses 0.

Corollary—Artin–Schreier for fields

A field F admits an ordering if and only if −1 is not a sum of squares in F. Indeed in a commutative ring a12⋯am2=(a1⋯am)2, so square-products are squares and (18.2) specialises to the 1927 criterion of Artin and Schreier.

07Proof Techniques and Method

How these proofs work, and which move to reuse.

Three techniques carry the section; all three reappear in the later results on formally real division rings.

Move 1

Sandwich elimination

A block aua becomes (au)2(u−1)2u. Whenever a hypothesis is stated for permuted products, this identity reduces it to ordinary products of squares.

Move 2

Right-translate to normalise

From 0=s1+⋯+sn, multiply through by s1−1 to reach a statement about −1. Only available because Σ(D) is a group — this is the step that fails in a general ring.

Move 3

Zorn plus maximality

Preorderings are closed under unions of chains, so a maximal one exists; (17.10) identifies maximal preorderings with orderings. Existence proofs never construct an ordering explicitly.

Move 2 explains why (18.2) can be phrased with −1 rather than with 0. In a general ring the weak preordering T(R) is closed under addition and permuted multiplication but not under division, and the passage from *0 is a sum* to *−1 is a sum* is exactly the step that needs an inverse.

08Worked Example

The rationals: one ordering, and T(ℚ) is already it

Take D=ℚ. Square-products are squares. Which positive rationals are sums of squares of rationals? Write a/b with a,b positive integers; then a/b=ab/b2, and by Lagrange's four-square theorem ab=n12+n22+n32+n42, so

ab=(n1b)2+(n2b)2+(n3b)2+(n4b)2.
(E.1)

Every positive rational is a sum of at most four rational squares; no negative rational is, since sums of squares are positive.

Hence T(ℚ)=ℚ>0, which is already an index-2 subgroup of ℚ∗ and therefore, by the corollary above, an ordering. The weak preordering is the unique ordering, which is the group-theoretic form of the statement that ℚ is orderable in exactly one way. Note ℚ∗/T(ℚ)={±1}, of exponent 2 as (18.1) demands, even though ℚ∗/ℚ∗2 is infinite.

The real quaternions: no ordering at all

Take D=ℍ, the division ring of real quaternions. Then i2=−1, so −1 is a square, in particular a sum of square-products, and (18.2) rules out any ordering of ℍ.

−1=i2∈Σ(ℍ)⊆T(ℍ)⟹0=1+i2∈T(ℍ).
(E.2)

This is instructive because ℍ passes every necessary condition from the general theory: it is a domain of characteristic 0 with no zero divisors. Orderability is strictly stronger than those conditions, and the obstruction is arithmetic, not structural.

Sanity check on the exponent-2 claim

In ℍ the commutator subgroup of ℍ∗ is the group of unit quaternions, and ℍ∗/[ℍ∗,ℍ∗]≅ℝ>0, a divisible group. So the only subgroup T with D∗/T of exponent 2 and Σ(D)⊆T is T=ℍ∗ itself — which is not additively closed away from 0. Consistent with there being no preordering, hence no ordering.

09Comparison and Classification

The general theory of §17 versus its division-ring form
NotionGeneral ring RDivision ring D
Preordering axiomper(a12⋯am2t1⋯tn)∈TT⋅T⊆T and a2∈T
Algebraic type of Tadditively closed subsetnormal subgroup of D∗
Quotientno group structure availableD∗/T abelian of exponent 2
Orderingmaximal preordering (17.10)preordering of index 2
Division closure T¯can be strictly larger than Talways T¯=T
Orderability test0∉T(R)−1∉T(D)
Basic positive elementssums of per(a12⋯am2)sums of square-products
Formal reality of some standard division rings
Formally realOrderableCommutative−1 a square
ℚ, ℝ●yes●yes●yes○no
ℂ○no○no●yes●yes
𝔽p○no○no●yes◐partial
ℍ○no○no○no●yes
ℚ((y))((x;σ)), σ(y)=2y●yes●yes○no○no
ℝ((y))((x;σ)), σ(y)=−y○no○no○no○no

Formal reality of some standard division rings

The last row is the point of the whole section: a division ring in which −1 is a square-product but is not a square, and not even a sum of squares.

10Relationship Map

The logical layout of the section is a single chain of equivalences supported by one structural proposition.

−1∉T(D)⟺D formally real⟺D has a preordering⟺D has an ordering
  • (18.1) Preordering T≤D∗ — additively closed subgroup containing all squares
    • contains
      • every square a2
      • every square-product, i.e. Σ(D)
      • every commutator aba−1b−1
      • every inverse t−1 of its own elements
    • is contained in
      • every ordering P⊇T
      • D∗, with elementary abelian quotient
    • never contains
      • 0
      • −1 (else 0=1+(−1)∈T)

Downstream, (18.1) is the reason the intersection theorem of Preorderings in Division Rings is so much simpler than its general-ring ancestor, and the normality of orderings is exactly what makes Albert's theorem work in Formally Real Division Rings.

11Standards and Notation

Standards here covers notation, symbol and markup standards, and reference implementations, rather than material or design codes.

Cone notationP for an ordering, T for a preordering (Lam, Serre); ΣF2 is common for the field case
Multiplicative groupD∗ here; D× in much of the arithmetic literature
Weak preorderingT(D) (Lam); ΣD2 or ΣD∗2 elsewhere — the latter is unsafe in the noncommutative case
Square-productsNo universal symbol; Σ(D) is used on this page for the subgroup generated by squares
Order relationa<biffb−a∈P; ISO 80000-2 symbols for <, ≤
SoftwareRealField/ordering machinery in Sage and Magma covers fields only; no CAS implements orderings of noncommutative division rings

Notation hazard

Writing ΣD2 for the set of sums of squares invites the false reading that the totally positive elements of a division ring are sums of squares. In the noncommutative case the right object is sums of square-products, and the two differ — see (18.7).

12Failure Modes and Common Mistakes

A square-product is not a square

a2b2=(ab)2 needs ab=ba. In a noncommutative division ring the set of squares need not even be closed under multiplication, which is why (18.2) is stated with square-products and why the Artin–Schreier criterion cannot simply be transcribed.

Being a domain of characteristic 0 is not enough

(17.4) says an ordered ring is a domain of characteristic 0; the converse fails badly. ℍ and ℂ are characteristic-0 division rings with no ordering whatever.

Do not assume T(D) is an ordering

T(ℚ)=ℚ>0 happens to be an ordering, but that is a coincidence of ℚ. In ℚ(2) the sums of squares form a preordering of index 4 in the multiplicative group: the two real embeddings give a surjection onto {±1}×{±1} whose kernel is exactly the set of sums of squares. There are two distinct orderings above it, and the preordering is their intersection.

  • Do not read P∪(−P)=D∖{0} as allowing 0∈P; the cone consists of strictly positive elements and P∩(−P)=∅ is forced.
  • Do not assume an ordering makes D into an ordered group under addition only; the multiplicative compatibility P⋅P⊆P is a separate axiom and is what fails in most attempted constructions.
  • Do not confuse formally real with real closed: the first is the orderability criterion, the second requires in addition that every positive element is a square and every odd-degree polynomial has a root — a condition with no useful noncommutative analogue.
  • Do not expect uniqueness of the ordering. A formally real division ring generally has many, and the intersection of them all is the division closure of T(D).

13Historical Notes and Lessons Learned

  • 1903Hilbert's twisted Laurent seriesIn the second edition of the Grundlagen der Geometrie, Hilbert produces the first noncommutative ordered division ring, to separate the axioms of geometry — arithmetic of ends over a twisted series field.
  • 1927Artin–SchreierA field is orderable exactly when −1 is not a sum of squares. The theory of formally real fields is launched and, with it, Artin's solution of Hilbert's 17th problem.
  • 1940Albert on ordered algebrasAlbert proves that the centre of an ordered division ring is algebraically closed in it, the first genuinely noncommutative theorem of the subject.
  • 1948–49Mal'cev and NeumannThe Laurent series construction with well-ordered support generalises Hilbert's example to an arbitrary ordered group, producing ordered division rings in abundance.
  • early 1950sSzele, Pickert, R. E. JohnsonThe orderability criterion is extended from fields to division rings, and Johnson gives the version for arbitrary rings via preorderings — the form used here.
  • 1983Scharlau–TschimmelThe level of a non-formally-real division ring is shown to take every positive integer value, in sharp contrast with Pfister's powers-of-two theorem for fields.

The methodological lesson is that the correct noncommutative generalisation was not found by weakening the field statement but by strengthening the object: replacing square by square-product, the smallest multiplicatively closed set that a cone is forced to contain. Once that substitution is made, the Artin–Schreier proof runs almost verbatim.

14Quick Reference

OrderingP⊆D∗ with P+P⊆P, P⋅P⊆P, P∪(−P)=D∖{0}
Ordering, group formAdditively closed subgroup of index 2 in D∗
Preordering (18.1)T+T⊆T, T⋅T⊆T, D∗2⊆T
AutomaticT⊴D∗, [D∗,D∗]⊆T, D∗/T of exponent 2
Square-producta12⋯am2; the group Σ(D) generated by squares
Weak preorderingT(D)= sums of square-products; smallest preordering when D is formally real
Formally real0∉T(D)iff−1∉T(D)
Szele–Pickert (18.2)D orderable iff −1 is not a sum of square-products
Which statement to quote
You wantUseReference
To verify a candidate conethe three closure conditions(18.1)
To promote a preordering to an orderingmaximality, via Zorn(17.10)
To rule out any orderingexhibit −1 as a sum of square-products(18.2)
To order a fieldcheck −1 is not a sum of squaresArtin–Schreier
To build an ordered division ringMal'cev–Neumann series over an ordered group(18.5)

15Frequently Asked Questions

Why does the definition of a preordering need permuted products at all?

Because in a noncommutative ring you cannot bring the two copies of a together, and the axiom must be strong enough to force the cone to be closed under conjugation. In a division ring the identity aua=(au)2(u−1)2u does the bringing-together for you, so the permutations can be dropped — that is exactly (18.1). In a general ring they cannot.

Is every subgroup of D∗ of index 2 an ordering?

No. Index 2 gives axiom (17.3) and multiplicative closure, but additive closure P+P⊆P is an extra condition and usually fails. For example ℚ∗ has many index-2 subgroups — one for each subgroup of ℚ∗/ℚ∗2 of index 2 — and only ℚ>0 is additively closed.

Can a division ring of characteristic p be ordered?

Never. An ordering contains 1 and is closed under addition, so it contains every n⋅1 with n≥1; if p⋅1=0 this would put 0 in the cone. So (17.4) forces characteristic 0, and orderable division rings all contain ℚ in their centre.

How many orderings can a division ring have?

As many as there are maximal preorderings above T(D), and for formally real D there is at least one. For ℚ there is exactly one; for ℚ(2) there are two; for the rational function field ℚ(t) there are infinitely many. The set of all orderings carries a natural topology in the field case (the real spectrum), but no comparable theory exists for division rings.

Does (18.2) give an algorithm for deciding orderability?

Not in general. It converts orderability into a single arithmetic question, but deciding whether −1 is a sum of square-products requires understanding the subgroup generated by squares, which for a presented division ring is not effectively computable. For a finitely generated field extension of ℚ the question is decidable by real-algebraic methods.

Why is the quotient D∗/T of exponent 2 so useful?

It converts questions about orderings into questions about index-2 subgroups of an elementary abelian 2-group, that is, into linear algebra over 𝔽2. Orderings containing a given preordering T correspond to hyperplanes in D∗/T that remain additively closed, which is how one counts orderings in practice.

16Related KEVOS Topics

Ordered RingsA compatible total order on a ring is the same data as a positive cone P obeying three axioms — and the mere existenPreorderings in Division RingsBecause a preordering of a division ring is a subgroup of D^*, it is automatically division-closed — so every preorderinConstructing Ordered Division RingsOrder a Mal'cev–Neumann series ring by the sign of the coefficient at the least element of its support: if every twist aFormally Real Division RingsA twisted series construction separates sum of squares from sum of square-products and realises every integer as a lEquations over Ordered Division RingsIn a formally real division ring, if a nonconstant polynomial g(a) over the centre commutes with b, then a itself commut

17References

  1. T. Y. Lam, A First Course in Noncommutative Rings, Graduate Texts in Mathematics 131, Springer-Verlag, 1991, &#167;18, pp. 285&#8211;286.
  2. E. Artin and O. Schreier, &#8220;Algebraische Konstruktion reeller K&#246;rper&#8221;, Abhandlungen aus dem Mathematischen Seminar der Universit&#228;t Hamburg 5 (1927), 85&#8211;99.
  3. T. Szele, &#8220;On ordered skew fields&#8221;, Proceedings of the American Mathematical Society 3 (1952).
  4. G. Pickert, Einf&#252;hrung in die h&#246;here Algebra, Vandenhoeck &amp; Ruprecht, G&#246;ttingen, 1951.
  5. T. Y. Lam, The Algebraic Theory of Quadratic Forms, W. A. Benjamin, Reading, Massachusetts, 1973, Chapters 8 and 10.
  6. N. Jacobson, Structure of Rings, American Mathematical Society Colloquium Publications 37, revised edition, 1964.

18AI Suggested Questions

  • Give a full proof that a maximal preordering in a division ring is an ordering, following (17.10).
  • Which subgroups of index 2 in ℚ(2)∗ are additively closed, and how do they correspond to the two orderings?
  • Construct a division ring in which the set of squares is not closed under multiplication.
  • How does the space of orderings of a formally real field relate to the real spectrum of its coordinate ring?
  • What replaces the Artin–Schreier theory of real closures for noncommutative division rings, if anything?
  • Show that the group Σ(D) of square-products equals the smallest normal subgroup containing all squares, and compute it for the real quaternions.
  • Compare the orderability criterion (18.2) with the criterion for a domain to be orderable in (17.11).
Page
KEVOS-ENG-MATH-NCR-0134
Path
Engineering / Mathematics
Template
kevos-knowledge-article-v2
KEVOS® Knowledge Library — reviewed 2026-08-08

On this page

  1. Executive Summary
  2. Overview
  3. Learning Objectives
  4. Definitions
  5. Core Concepts
  6. Key Results
  7. Proof Techniques and Method
  8. Worked Example
  9. Comparison and Classification
  10. Relationship Map
  11. Standards and Notation
  12. Failure Modes and Common Mistakes
  13. Historical Notes and Lessons Learned
  14. Quick Reference
  15. Frequently Asked Questions
  16. Related KEVOS Topics
  17. References
  18. AI Suggested Questions

Continue learning

Archimedean Ordered Rings Are CommutativeArticle · Engineering MathematicsNEXT LESSON →Preorderings in Division Rings and the Intersection TheoremArticle · Engineering MathematicsExtending Orderings to Larger RingsArticle · Engineering MathematicsConstructing Ordered Division RingsArticle · Engineering Mathematics
KEVOS · Engineering, manufacturing and project improvement
ArticlesServicesCase studiesAboutContact
© 2026 KEVOS®