Sparkles

Appearance

Whitney's Axiomatic Quantity Structures ​

Hassler Whitney's two-part The Mathematics of Physical Quantities (American Mathematical Monthly, 1968) is the axiomatic high-water mark of the classical quantity-calculus lineage. Instead of modelling a quantity as "a real number with a unit attached", it takes the quantities themselves — masses, lengths, time intervals — as the primitive elements of one-dimensional measurement models called rays and birays, constructs the number systems ℕ → ℚ⁺ → ℝ⁺ → ℝ afterwards as operators on those models, and (in Part II) assembles finitely many base birays into a quantity structure in which every quantity factors uniquely as a numerical multiple of a product of powers of base units — the representation theorem that every later formalization either inherits, restricts, or reacts against. Raposo's algebraic fiber bundles, Jonsson's quantity spaces, and (structurally, though — as shown below — without citation) Kennedy's dimension types are all recognizable as refinements of exactly this picture.

NOTE

Provenance discipline is part of this page's content. Both Monthly papers are paywalled. Part I is quoted from a local two-page re-typeset excerpt of its introduction only (whitney-1968-physical-quantities-i-monthly-excerpt.pdf) — Whitney's own framing, but only the framing: the actual postulate lists for rays and birays are behind the paywall. Part II was not inspected at all; every Part II claim below is grounded in explicit restatements by Raposo 2018/2019 and Jonsson 2021 and is tagged "as restated by …". Where the restatements disagree (they do, on the exponent ring), the disagreement is reported, not resolved. Kennedy's 1996 thesis — which this survey initially expected to restate Whitney — in fact never cites him (checked against the full text of the local copy); it appears here as a structural parallel and for mechanization, not as a witness for Whitney's constructions. Local artifact paths below are file names under the survey's pinned corpus ($REPOS/papers/units-of-measure/, catalogued in the grounding ledger).

At a glance ​

DimensionWhitney's quantity structures
Primary objectsOne-kind measurement models: rays (positive quantities, "like a half line") and birays (signed, "like an oriented line with starting point"); Part II: quantity structures built from finitely many base birays
QuantityAn element of a ray/biray — the physical property itself, taken as primitive ("why not let this property itself be an element of the model?")
UnitAny element l₀ "being kept fixed for a period" — pure bookkeeping, no algebraic privilege; the models contain no distinguished 1
DimensionThe biray a quantity belongs to (what Drobot calls a "dimension", Whitney a "biray" — Jonsson App. B.1); in the ℝ × ℚⁿ reading, the exponent tuple (r₁, …, rₙ)
KindNo notion beyond biray membership; for counting, distinct kinds are kept apart by using "several isomorphic models" (bl ≠ ck)
NumbersNot presupposed: ℕ (iterated addition), ℚ⁺ (subdivision), ℝ⁺ (completeness), ℝ (negatives) arise as operators on the models; multiplication in ℝ is derived via the isomorphism theorem
Central theoremsPart I: isomorphism theorem — every ray homomorphism "is necessarily an isomorphism onto"; Part II (as restated): representation q = α · u₁^r₁ ⋯ uₙ^rₙ uniquely, i.e. Q ≅ ℝ × ℚⁿ
Exponent ringℚ per Raposo 2018, ℝ per Raposo 2019, "Q or R" per Jonsson's comparison table — unresolved here because Part II is uninspected
Cross-dimension additionInexpressible at carrier level (m + l has no home: different sets, no shared operation) — yet Whitney's own counting example manipulates the formal sum 2 bl + 3 ck distributively without naming its habitat
Dimensional analysisPart II's subtitle; per Raposo 2019 the coordinatized model (real exponents in his telling) "has served its main purpose: giving a proof of Buckingham's Pi Theorem"
ProvenancePart I inspected via 2-page excerpt (intro §§1–7 only); Part II via restatements in raposo-2018/raposo-2019/jonsson-2021; Kennedy 1996 contains no Whitney citation
Modern descendantsRaposo's fiber bundle (dimension-centred, ℤ exponents, one zero per fiber), Jonsson's quantity spaces (proved equivalent to Raposo's), Kennedy's dimension types (structural parallel)

The neighbouring theory pages carve up the same territory: Buckingham's Π theorem is the dimensional-analysis payoff Whitney's Part II re-founds; the free-abelian-group view of dimensions is the ℤ-exponent restriction of Whitney's biray algebra; Tao's tensor-of-lines construction and the torsor representation are two modern answers to the fiber structure Whitney's rays anticipate; the comparison capstone places all of them side by side.

Primary sources ​

  • H. Whitney, "The Mathematics of Physical Quantities. Part I: Mathematical Models for Measurement", The American Mathematical Monthly 75(2), February 1968, pp. 115–138. Inspected via excerpt only: the local artifact whitney-1968-physical-quantities-i-monthly-excerpt.pdf is a two-page re-typeset of the introduction (§§1–7) — it confirms title, part subtitle, venue, and date ("American Mathematical Monthly. February 1968"), and even garbles the byline to "Whitney Hassler". The page range 115–138 is from the publisher's bibliographic record (the DOI resolves to Monthly 75(2), first page 115); the range itself is otherwise [unverified] against a local artifact. All Part I quotes below are from this excerpt; note that a re-typeset excerpt carries transcription risk (it visibly confuses l/1 in places), flagged where it matters.
  • H. Whitney, "The Mathematics of Physical Quantities. Part II: Quantity Structures and Dimensional Analysis", The American Mathematical Monthly 75(3), March 1968, pp. 227–256. Not inspected — grounded via restatement. The full citation including the 227–256 page range is printed in both local restatements: Raposo 2018 reference [22] (raposo-2018-algebraic-structure-quantity-calculus-msr.pdf, p. 157) and Jonsson 2021 reference [31] (jonsson-2021-magnitudes-scalable-monoids-quantity-spaces-arxiv.pdf).
  • Á. P. Raposo, "The Algebraic Structure of Quantity Calculus", Measurement Science Review 18(4):147–157, 2018 (local: raposo-2018-algebraic-structure-quantity-calculus-msr.pdf). Inspected. Positions Whitney in the "unit-centred" lineage (with Drobot and Carlson), restates the lineage's structure as ℝ × ℚⁿ with the unique factorization q = α u₁^r₁ ⋯ uₙ^rₙ, criticizes it, and replaces it with a dimension-centred algebraic fiber bundle.
  • Á. P. Raposo, "The Algebraic Structure of Quantity Calculus II: Dimensional Analysis and Differential and Integral Calculus", Measurement Science Review 19(2):70–78, 2019 (local: raposo-2019-algebraic-structure-quantity-calculus-ii-msr.pdf). Inspected. Restates the Whitney-lineage model again (this time with real exponents and the family-of-rays-glued-at-zero picture) and credits it with the proof of the Π theorem; then proves a Π theorem with integer exponents in the fiber-bundle setting.
  • D. Jonsson, "Magnitudes, Scalable Monoids and Quantity Spaces", arXiv:2108.02106 (v6, January 2023) (local: jonsson-2021-magnitudes-scalable-monoids-quantity-spaces-arxiv.pdf). Inspected. Appendix B.1 is the most detailed local restatement of Whitney's Part II construction (the multiplicative vector space V_Q with embedded scalars, birays, derived addition, which results Whitney proves), plus a primitive-vs-derived comparison table across Drobot/Quade/Whitney/Carlson/ Jonsson/Raposo.
  • A. J. Kennedy, "Programming Languages and Dimensions", PhD thesis, University of Cambridge, Computer Laboratory TR-391, 1996 (local: kennedy-1996-programming-languages-dimensions-thesis.pdf). Inspected — negative finding: a full-text search of the local copy finds no citation of Whitney (nor of Drobot or Carlson); Kennedy's related work engages Wand & O'Keefe, Goubault, House, and Rittri instead. His ℤ-module treatment of dimensions (thesis Appendix B) and scaling-invariance semantics (ch. 6–7) parallel Whitney structurally but were developed on a separate track. Used below for Mechanization.
  • S. Drobot, "On the Foundations of Dimensional Analysis", Studia Mathematica 14:84–99, 1953. The local artifact (drobot-1953-foundations-dimensional-analysis-studia.pdf) is an image-only scan and was not text-inspected for this page; Drobot appears here only as restated by Raposo and Jonsson ("The approach introduced by Drobot [5] and developed by Whitney [31]…"). The Buckingham Π page treats Drobot directly.
  • D. Carlson, "A mathematical theory of physical units, dimensions and measures", Archive for Rational Mechanics and Analysis 70:289–304, 1979. [unverified] — no local artifact; cited from Raposo 2018 reference [23] and Jonsson's Appendix B.1 restatement.
  • J. de Boer, "On the history of quantity calculus and the international system", Metrologia 31:405–429, 1994. [unverified] — no local artifact; both Raposo and Jonsson lean on it as the standard history of the lineage.

Formal core ​

Part I: measurement models before numbers ​

Whitney's opening move (Part I §1) is a critique of exactly the "reals with attached units" design that most programming-language libraries still ship:

"Commonly one takes R+ or R as a model for measurement. There are disadvantages in this, however. These models contain a specific number 1, and there is no natural way of putting this number in correspondence with a particular measurement; moreover, the models contain an operation of multiplication, with no natural physical counterpart." — Whitney, The Mathematics of Physical Quantities, Part I §1 (whitney-1968-physical-quantities-i-monthly-excerpt.pdf, p. 1)

The fix is to admit the physical property itself as a mathematical object:

"Let us consider the problem of choosing a model M for masses. An object A has a certain property which we call its "mass" ; why not let this property itself be an element of the model? As far as the structure of the model is concerned, we need not theorize on what "mass" really is; we need merely give it certain properties in the model." — Whitney, Part I §1 (excerpt, p. 1)

The model M gets an addition — because combining two objects A, B into one object C gives m_C = m_A + m_B, a real physical process — "and any further properties we choose". Nothing else is primitive: no 1, no multiplication, no numbers. Two model shapes cover most measurement:

"The two types of models that best fit in most situations we shall call "rays" and "birays." A ray (like a half line) is used for positive measurements, and a biray ( like an oriented line with starting point ), for measurements of quantities both positive and negative." — Whitney, Part I §1 (excerpt, p. 1)

Numbers then appear as operators on the model (§2), in a tower that recapitulates how measurement actually refines:

text
ℕ    iterated addition:      2l = l + l,   3l = l + l + l, …
     ("Thus N appears as a natural set of operators on our model.")
ℚ⁺   subdivision:            find l′ with 3l′ = l, set l′ = (1/3)l, then 2l′ = (2/3)l
ℝ⁺   completion:             "if our model has a certain completeness property,
                              we may enlarge Q+ to R+ as operator system"
ℝ    signed quantities:      "if we have negative quantities, we may enlarge R+ to R"

Equational reasoning happens inside the model (§3): Whitney's own examples are

text
5 cakes + 2 cakes = (5 + 2)cakes = 7 cakes
2 yd = 2(3ft) = (2 x3)ft = 6ft.

(transcribed exactly from the re-typeset excerpt, spacing artifacts included), with the punchline that model-elementhood replaces "measures the same as":

"The fact that "2 yd" and "6 ft" name the same element of the model enables us to say they are equal; there is no need for such mysterious phrases as "2 yd measures the same as 6 ft."" — Whitney, Part I §3 (excerpt, p. 1)

IMPORTANT

The excerpt ends at the introduction. Whitney says the models are introduced postulationally and that "The postulates used for rays and birays are few in number and simple in character, and correspond to simple experimental phenomena" (§5) — but the postulate lists themselves are in the paywalled body. This page can quote Whitney's design intent verbatim; it cannot quote his axioms.

The Part I isomorphism theorem ​

The excerpt does state Part I's central theorem, in Whitney's own words:

"A basic theorem in the subject is an isomorphism theorem; a homomorphism of one ray into another is necessarily an isomorphism onto, and has certain additional properties (and similarly for birays). This theorem is a great aid in setting up the theory; in particular, with its use, multiplication in R+ and in R is introduced and its properties derived with a minimal effort." — Whitney, Part I §5 (excerpt, p. 2)

Two consequences matter downstream. First, rigidity: rays admit no interesting quotients or sub-models — any structure-preserving comparison of two rays is a full identification, so a ray is "one-dimensional" in the strongest sense, and choosing a single element (l₀) coordinatizes the whole ray. Second, numbers gain their multiplication from measurement, not vice versa: composing the operator "×2" with the operator "×3" on any ray must again be an operator, and the isomorphism theorem makes that composite well-defined — this is how "multiplication in R+ and in R is introduced". The same rigidity resurfaces in Raposo's bundle as "All fibers are isomorphic as vector spaces, and isomorphic to the field F" (raposo-2018-…msr.pdf, §2.2) and in the torsor reading of fibers.

Part II: quantity structures (as restated) ​

Part I flags the plan: "in mechanics, one uses separate rays M, L, T for measurement of mass, length and time. (We study structures containing several rays in Part II.)" (excerpt, p. 2). For what Part II actually builds, the best local witness is Jonsson's Appendix B.1:

  • The set of quantities itself — "rather than a set of pre-units" — forms a vector space V_Q written multiplicatively over a field of exponents: the "scalar product" is exponentiation q^λ. V_Q is also assumed to contain a set R of scalars, identified with the dimensionless quantities, giving a second scalar action r·q (Jonsson, App. B.1: "the authors identify dimensionless quantities with scalars").
  • Quantities of the same kind form classes — "called 'dimensions' by Drobot and 'birays' by Whitney". Addition is derived, not primitive: for q = α·u and r = β·u with u non-zero, q + r = (α + β)·u — and, Jonsson notes pointedly, "although only Whitney proves that this definition is legitimate" (i.e. proves independence from the chosen u).
  • The biray algebra is likewise derived from the quantity algebra: [x][y] = [xy] and [x]^λ = [x^λ], where [q] is the biray containing q. Whitney additionally proves distributivity q(r + s) = qr + qs for [r] = [s], and cancellation: qs = rs with s non-zero implies q = r (all per Jonsson, App. B.1).

Jonsson's comparison table (App. B.1) records Whitney's choice of primitives: product of quantities and exponentiation of quantities are primitive; addition, scalar product, product of dimensions and exponent of dimensions are all derived; the "ring of exponents" is listed as "Q or R". Raposo 2018 gives the complementary bird's-eye view of the same lineage:

"this structure starts with a system of units {u₁, . . . , uₙ} and writes any quantity q in a unique way as q = α u₁^r₁ · · · uₙ^rₙ … where α is a real number and r₁, . . . , rₙ are rational numbers. Therefore, the algebraic structure depicted by this theory is R × Qⁿ, where the factor R hosts the numerical value of q relative to this system of units, while Qⁿ hosts the rational exponents of the units and exhibits a linear space structure." — Raposo, raposo-2018-algebraic-structure-quantity-calculus-msr.pdf, §1 (notation lightly flattened from the PDF's u₁^r1 ··· uₙ^rn)

and Raposo 2019 adds the geometric picture and the payoff:

"The set of quantities, thus, adopts the form of a family of rays, each identified and spanned by a unit as in (1) by letting α run through the reals. The rays coincide in a point, the zero of the algebraic structure. The numbers (r₁, . . . , rₙ) are referred to as the dimensions of the quantity q. This algebraic structure has served its main purpose: giving a proof of Buckingham's Pi Theorem." — Raposo, raposo-2019-algebraic-structure-quantity-calculus-ii-msr.pdf, §1

The representation theorem, with proof sketch ​

Assembling the two restatements, Part II's central result is a representation (and implicitly classification) theorem:

text
Theorem (representation; Part II as restated by Raposo 2018 §1, Raposo 2019 §1,
Jonsson 2021 App. B.1).  Let Q be a quantity structure with a fundamental system
of units u₁, …, uₙ.  Every quantity q ∈ Q has a unique expression

    q = α · u₁^r₁ ⋯ uₙ^rₙ          α ∈ ℝ;  rᵢ ∈ ℚ (2018) / ℝ (2019)

and the coordinate map q ↦ (α, (r₁, …, rₙ)) is an isomorphism Q ≅ ℝ × ℚⁿ with

    (α, r) · (β, s) = (αβ, r + s)         product of quantities
        λ · (α, r)  = (λα, r)             scalar action
    (α, r) + (β, r) = (α + β, r)          addition — same exponent tuple only
    (α, r) + (β, s)   undefined, r ≠ s

Any two quantity structures with the same number of fundamental units are therefore isomorphic — non-canonically, since the isomorphism threads through the chosen units. (Raposo proves the exact analogue for his fiber bundles as his Theorem 2 — "Two spaces of quantities over the same field, free of zero divisors, are isomorphic if and only if they have the same rank" — and notes of the coordinate isomorphism "the isomorphism is not canonical, for it depends on the system of units chosen", raposo-2018-…msr.pdf §4, Example 4.5.)

WARNING

The sketch below is a reconstruction from Part I's quoted machinery plus the Part II restatements; Whitney's own proof was not inspected. It is the argument the restated ingredients force, not a paraphrase of his text.

  1. Each biray is one-dimensional over the operators. Fix a non-zero u in a biray B. Part I's operator construction gives α·u ∈ B for every α ∈ ℝ, and the isomorphism theorem makes the assignment α ↦ α·u an isomorphism of the operator biray ℝ onto B: every q ∈ B is q = α·u for a unique α. (This uniqueness is also exactly what makes the derived addition α·u + β·u = (α+β)·u legitimate — the fact Jonsson highlights Whitney alone bothered to prove.)
  2. The birays form an exponent-vector space. The derived operations [x][y] = [xy] and [x]^λ = [x^λ] make the set of birays a vector space over the exponent field, written multiplicatively (for ℤ exponents this is a free abelian group). The fundamental birays [u₁], …, [uₙ] are independent generators, so every biray has a unique expansion [q] = [u₁]^r₁ ⋯ [uₙ]^rₙ — the exponent tuple (r₁, …, rₙ).
  3. Compose. Given q, step 2 produces the unique tuple with [q] = [u₁^r₁ ⋯ uₙ^rₙ]; then q and u₁^r₁ ⋯ uₙ^rₙ share a biray, so step 1 produces the unique α with q = α · (u₁^r₁ ⋯ uₙ^rₙ).
  4. Operations transport. Multiplicativity of the coordinate map follows from (α·x)(β·y) = αβ·xy (scalars commute out — in Whitney's setting because scalars are dimensionless quantities; compare Jonsson's Lemma 2.5, which proves the same identity from his scalable-monoid axioms); addition transports within a fiber by step 1; and dim-preservation is step 2. Hence Q ≅ ℝ × ℚⁿ with the componentwise operations above. ∎

Dimensional analysis then rides on the representation: a change of units rescales α and fixes (r₁, …, rₙ), so a law invariant under all unit changes can depend only on the dimensionless combinations — the Π-theorem route Raposo 2019 credits to this structure. (Raposo's own integer-exponent Π theorem, Theorem 4.4 of raposo-2019-…msr.pdf, is the modern descendant, with homogeneity defined as equivariance under the group G of scale-factor maps χ : D → F*.)

What the restatements disagree on ​

The three local witnesses give three exponent rings for Part II: ℚ (Raposo 2018: "r₁, . . . , rₙ are rational numbers … R × Qⁿ"), ℝ (Raposo 2019: "where α, r₁, . . . , rₙ are real numbers"), and "Q or R" (Jonsson's table). With Part II paywalled this page cannot adjudicate; the honest reading is that the lineage as a whole worked with dense exponents (contrast Quade and Raposo's ℤ, Kennedy's ℤ), and that Whitney's text supports at least ℚ. The disagreement is substantive, not clerical — it decides which mechanization theory applies and whether q^0.2 denotes anything.

Structural anatomy ​

Answers to the survey's shared comparability protocol, in Whitney's own terms.

What structure is primary? ​

The one-kind measurement model — a ray or biray with addition as its only primitive operation, postulated to "correspond to simple experimental phenomena". Numbers are not part of the furniture: they arrive later as operators, and even their multiplication is derived (via the isomorphism theorem). In Part II the primary structure is the quantity structure: per Jonsson's table, its primitives are product of quantities and exponentiation of quantities; addition, scalar product, and the entire dimension algebra are derived. Morphisms are ray/biray homomorphisms — which the isomorphism theorem collapses to isomorphisms, so the category of rays is a groupoid, and "the same quantity structure" is only ever determined up to a unit-dependent isomorphism. Raposo files the whole construction under the unit-centred camp: "authors such as Drobot [21], Whitney [22], and Carlson [23] developed an algebraic structure which resembled that of quantity calculus, but was centered on the concept of unit over which the rest of the set of quantities was built" — the axis along which his own dimension-centred bundle differs.

What is a quantity, a unit, a dimension, a kind? ​

  • Quantity — an element of a model; the property itself ("why not let this property itself be an element of the model?"). A quantity is not a number-unit pair; the pair (α, unit) is a description that exists only after a unit is chosen.
  • Unit — any element temporarily held fixed, with Whitney's deliberately deflationary gloss: "If we choose a length l0 ∈ L, and compare other lengths with it, we may call, l0 our "unit"; this serves merely to remind us that l0 is being kept fixed for a period." (excerpt, p. 2 — the stray comma is the re-typeset's). Units have no algebraic status a non-unit element lacks.
  • Dimension — the biray itself: a dimension is the set of mutually comparable quantities, not a label attached to them ("called 'dimensions' by Drobot and 'birays' by Whitney" — Jonsson App. B.1). Only in the coordinatized ℝ × ℚⁿ picture does a dimension flatten to an exponent tuple (Raposo 2019: "The numbers (r₁, . . . , rₙ) are referred to as the dimensions of the quantity q").
  • Kind — no separate notion inside a quantity structure: kind = biray membership. But Part I's counting discussion shows Whitney had the missing distinction in hand at the model level: "if several such types of quantities are considered together, it is better to use several isomorphic models" — balloons and cookies get isomorphic but distinct models precisely so that they remain un-addable. The formalism distinguishes kinds by model choice, not by dimension formula; nothing in the Part II restatements says whether that trick survives inside a single multiplicatively closed quantity structure (recorded as a silence — see limits).

How is dimensional homogeneity of physical laws expressed? ​

By well-formedness, not by a side condition. Addition and equality exist only inside a model, so a heterogeneous equation is not false — it is unformulable: there is no set in which 2 yd and 6 kg both live to be equated. The "2 yd = 6 ft" quote above is the positive half: sameness of dimension is sameness of model, and homogeneous equations are ordinary equalities between elements. The quantitative half — that a law relating quantities can be reduced to a relation among dimensionless combinations — is Part II's dimensional-analysis payload (its subtitle), delivered per Raposo 2019 as the lineage's "main purpose: giving a proof of Buckingham's Pi Theorem". The modern sharpening of "homogeneity = invariance under unit change" into equivariance under a group action on fibers is Raposo 2019's Definition 4.1, treated on the Buckingham Π page.

What acts as a change of units, and what is invariant? ​

In Whitney's models a change of units is not a map on quantities at all — it is a change of description, and the quantities are untouched:

"Suppose we wish to "change units," say from ft to in. Then since, for any a ∈ R+, a ft = a(12 in) =12 a in, we would replace "the length a" by "the length 12a." If any problems about units arise, they are at once resolved by going back to the explicit phrase "a ft."" — Whitney, Part I §4 (excerpt, p. 2; "=12 a in" spacing is the re-typeset's)

Invariant: every element of every model, and every in-model equation — a ft and 12a in name the same element. Variant: the numerals, i.e. the coordinates induced by the choice of l₀. Choosing l₀ is what "replaces L by R+", and Whitney treats that replacement as a notational convenience with an explicit escape hatch back to the invariant phrase "a ft". In the descendants this becomes a theorem about non-canonicity (Raposo: Q ≅ F × D "but the isomorphism is not canonical, for it depends on the system of units chosen") and, in Kennedy's world, the scaling-invariance ("dimensional invariance") theorem for well-typed programs — same idea, restated as parametricity.

Addition across different dimensions: forbidden, partial, undefined, meaningless? ​

Inexpressible — which is stronger than forbidden. m + l for m ∈ M, l ∈ L is not an error the axioms rule out; it is a string with no denotation, because the two summands inhabit disjoint carriers and no operation spans them. Whitney's reason is operationalist and stated up front: a model gets exactly the operations with physical counterparts. Addition of masses earns its place from a physical process (aggregating objects); no process aggregates a mass with a length; therefore no such operation is postulated. Two further Whitney-specific twists sharpen the survey's central question:

  1. The multiplication asymmetry is inverted at the base. The standard puzzle asks why quantities multiply freely across dimensions but add only within one. In Part I, multiplication does not exist at all — it is dismissed in §1 as having "no natural physical counterpart" for a single kind. Cross-kind multiplication is a Part II superstructure erected over the one-kind additive models (and even there, per Jonsson's table, it is a primitive of the aggregate, not something inherited from the models). So for Whitney the deep operation is addition-within-a-kind; multiplication is algebraic scaffolding added later to relate kinds. The free-abelian-group page and Tao's tensor construction give the two modern justifications of that scaffolding.

  2. Whitney himself computes with a cross-kind sum. Part I §6, on counting:

    "However, if several such types of quantities are considered together, it is better to use several isomorphic models. For a plebeian illustration, suppose there will be six children at a party. We wish each to have two balloons and three cookies. What is the total supply needed? The answer is: 6(2 bl + 3 ck) = 6(2 bl) + 6(3 ck) = 12 bl + 18 ck." — Whitney, Part I §6 (excerpt, p. 2)

    The expression 2 bl + 3 ck is manipulated distributively — operators act on it, nothing collapses — yet the excerpt never says what it is an element of. The natural modern home is a direct sum of the two models, i.e. exactly the dimensioned vector spaces of Hart's multidimensional analysis; Whitney's text (as available here) leaves the sum formal. So the finding is: cross-kind addition is undefined within any one model, but not treated as meaningless as notation — a subtlety most of the lineage's descendants legislate away.

One more delta worth recording: in the coordinatized reading restated by Raposo 2019 (with real exponents there), "the rays coincide in a point, the zero of the algebraic structure" — a single shared zero for all dimensions. Raposo's replacement structure takes the opposite choice (a zero per fiber: "0 ms−1 is a different quantity than 0 kg", raposo-2018-…msr.pdf §2.3), and his 2018 critique of the unit-centred lineage is precisely that its addability relation is not intrinsic: "we can find quantities which, dependending on the unit system of choice, can be compared or cannot, can be added or cannot" [sic] (raposo-2018-…msr.pdf, §1). Whether Whitney's original axioms are guilty of that charge cannot be checked against the excerpt; the charge is Raposo's, aimed at the lineage's ℝ × ℚⁿ skeleton.

Expressive power & limits ​

What it handles that "reals with attached units" cannot ​

  • No privileged 1, no phantom multiplication. The two defects Whitney indicts in ℝ-as-model (§1 quote above) are absent by construction: a ray has no distinguished element and no internal product. Libraries that store a quantity as a bare double reintroduce both defects and then police them with types; the type-system-mechanisms page catalogues how.
  • Affine quantities are first-class. Part I §6 supplies the exact structure that modern libraries bolt on for temperatures and timestamps: "A model of a somewhat different nature is an oriented affine one-dimensional space T*; this is the natural model for instance for moments in time (or positions on a line). There is a corresponding biray T of translations of T*; this is the natural model for intervals of time (or directed lengths)." (excerpt, p. 2). Point-vs-translation is the torsor story avant la lettre — 1968, in the introduction.
  • Zero is a design decision, not an accident. A ray deliberately excludes zero ("in measuring masses, one wishes to allow the mass zero (not present in a ray). This extra element may be introduced and related to the remaining elements in the obvious manner", §6) — so positivity, signedness, and the existence of a zero are per-kind modelling choices rather than global consequences of the number type.
  • Counting, and even non-cancellative measurement. Progressions get ℕ-models; distinct countable kinds get distinct isomorphic models (balloons ≠ cookies); and §7's finite model — where a′ + b′ = (a+b)′ saturates at n′ — deliberately breaks the embedding into ℝ: "these operations are commutative and associative, and the distributive laws hold. However, the cancellation laws fail" (Helen's spoon-drawer bookkeeping: 8s + 4s = 8s + 2s). Measurement models need not be sub-structures of the reals at all — a degree of freedom no descendant in this survey retains.
  • The reals are an output, not an input. Part I constructs ℝ from the models ("the real number system is constructed along with the models in a natural manner", §1) — the formalization does not presuppose the very number system whose role in measurement it is trying to explain.

Fractional and irrational exponents ​

Whitney's exponent field (ℚ or ℝ, per the restatements) buys closure under √ — dimensionally [x]^λ = [x^λ] is total — at two documented costs. First, over-expressiveness: Raposo argues physical equations reach quantities only through product, scalar product, and same-kind addition, "Therefore, only integer exponents of quantities, and thus of units, should be expected … An algebraic structure for quantity calculus, which allows fractional exponents, is oversized" (raposo-2018-…msr.pdf, §1 — his pendulum and standard-deviation examples, and the v = √(2T/m) formula spelled out in raposo-2019-…msr.pdf §1, show every physical square root acting on a quantity that is already a square). Second, incoherence at the scalars: Jonsson's verdict on the Drobot–Whitney design is that the two assumptions — V_Q is a vector space over the exponent field via (λ, q) ↦ q^λ, and V_Q contains the scalars — "are not fully compatible": closure forces q^λ to be a real number for scalar q, which fails unless the embedded scalars are restricted to ℝ_{>0}, "then all quantities must be positive". And interpretability is open regardless:

"while integral powers of quantities make sense in physics, it is not clear how to interpret q^0.2 or q^π, where q is a "dimensionful" quantity rather than a number." — Jonsson, jonsson-2021-magnitudes-scalable-monoids-quantity-spaces-arxiv.pdf, App. B.1 (superscripts restored from the PDF's layout)

Same dimension, different kind (torque vs energy, Hz vs Bq) ​

Inside one quantity structure there is no mechanism: the biray of a product is determined by [x][y] = [xy], so once torque and energy are both M L² T⁻²-birays they are the same biray and freely addable. Whitney's isomorphic-copies trick from counting does not obviously survive multiplicative closure — duplicating a biray breaks the uniqueness of the exponent expansion that the representation theorem needs — and neither the excerpt nor any restatement addresses the question (recorded as a silence). The descendants inherit the gap knowingly: Raposo, citing the VIM, concedes "quantities of the same kind belong to the same fiber, while the opposite is not necessarily true. However, the algebraic structure cannot distinguish this detail" (raposo-2018-…msr.pdf, §2.2). The systems that do distinguish torque from energy (notably mp-units' quantity_spec hierarchy) had to add structure the Whitney lineage never had.

Logarithmic quantities, angles — silences ​

  • Logarithmic scales (dB, pH): nothing in the excerpt or in any restatement. A logarithm of a dimensionful quantity is exactly a non-integer/transcendental functional image, which the previous subsection shows the lineage cannot interpret; the silence is therefore consistent but total.
  • Angles: no treatment. Worse, the lineage's identification of dimensionless quantities with scalars (Jonsson: "the authors identify dimensionless quantities with scalars") forecloses the modern move of giving angle its own dimension — a dimensionless rad collapses into the bare number 1. (Raposo's fiber bundle keeps the dimension-one fiber as a distinguished object — "naturally isomorphic with the field" but still a fiber — which at least leaves the question visible.)
  • Quantized quantities: Raposo 2019 §1 notes that one-dimensional vector-space fibers "cannot be the ultimate model … for it does not fit properly to quantized quantities" — a limit the fibers inherit directly from Whitney's rays.

Mechanization ​

No proof assistant or library in this survey mechanizes rays, birays, or Whitney's quantity structures as such — a negative finding (checked: the local corpus's Lean artifacts formalize dimensions as groups for Buckingham-Π purposes, and mathlib4 has nothing dedicated to physical quantities at all; see the Lean page). What is mechanized, pervasively, is the decision problem the representation theorem induces: once Q ≅ ℝ × ℚⁿ (or ℝ × ℤⁿ), checking dimensional consistency is linear algebra over the exponent ring, and inference is unification in the corresponding equational theory. The exponent-ring fork in the restatements maps exactly onto the mechanization landscape, per Kennedy's thesis (§3.4's related-work survey and §3.5's summary, quoted from kennedy-1996-programming-languages-dimensions-thesis.pdf):

Exponent ringEquational theory of dimensionsDecision procedure
ℤ (Kennedy, Quade, Raposo, Jonsson)free abelian groupAG-unification — decidable, unitary; the basis of Kennedy's dimension types and F#'s UnifyMeasures solver; with polymorphic recursion, AG semi-unification is solved for one inequation (Rittri), general case open per Kennedy
ℚ (Whitney per Raposo 2018; Wand–O'Keefe, Goubault)vector space over ℚGaussian elimination for solving (Kennedy on Wand & O'Keefe: "equations are not necessarily integral, so Gaussian elimination is used to solve them"; their types can be dimensionally nonsensical, e.g. exponent-swapping); Kennedy: inference with dimension-polymorphic recursion "can be reduced to semi-unification over vector spaces. That problem is decidable and admits a straightforward algorithm" (Rittri's result)
ℝ (Whitney per Raposo 2019)vector space over ℝThe same Gaussian elimination — rank and kernel are field-independent for the rational data that arise in practice; no checker in this survey actually infers over ℝ-exponent dimensions

The curiosity is that the ℚ-exponent theory — Whitney's, on the 2018 restatement — is in one respect better-behaved for type inference than the ℤ theory the programming-language tradition standardized on: over a field, semi-unification is decidable in general, while over ℤ its general case was still open at thesis time (Kennedy §3.5). The ℤ choice won for the physical reason Raposo articulates (integer exponents are all that physics produces) rather than for a computational one. Runtime-checked systems, unburdened by unification, largely keep dense exponents — see python-pint — while the statically-checked systems in this tree are ℤ-exponent descendants; the mechanisms page tracks the split. Whitney's Part I program itself — numbers as operators, quantities opaque, units as mere bookkeeping — survives most visibly in the semantics used to prove such systems sound: Kennedy's dimensional-invariance theorem interprets real δ over models with a positive-reals scaling action and derives "scaling theorems" from types, reinventing (independently — no citation, as established above) Whitney's operator picture as relational parametricity.

Open problems & frontier ​

  1. The exponent ring. ℤ vs ℚ vs ℝ is still the live fault line: Raposo 2018/2019 argue ℤ suffices and that dense exponents are "oversized"; Whitney's Part II (per the restatements) chose ℚ or ℝ; Jonsson leaves the interpretability of q^0.2/q^π explicitly unresolved. Any mechanization must pick a side before it can state the representation theorem.
  2. Unit-centred vs dimension-centred axiomatics. Raposo's critique — the operations' rules "need to be set in the axioms", and addability can depend on the unit system — is aimed at the whole Drobot–Whitney–Carlson line; whether Whitney's actual postulates (uninspected) are vulnerable to it, or already anticipate the fiber-bundle reading, is a reading problem blocked on the paywall, and worth settling given how much of the modern literature cites Part II through secondaries.
  3. Scalars: embedded or acting? Jonsson's incompatibility argument (embedded scalars + exponentiation-as-scalar-product force positivity) identifies a real defect in the Part II design. Kock's short-exact-sequence repair (ℚ_{>0} → P → D, restated by Jonsson) accepts the positivity restriction; Raposo and Jonsson instead make scalars act externally. Which repair preserves more of Whitney's "numbers as operators" program is unexamined.
  4. One zero or many? The lineage's rays "coincide in a point" (one global zero); Raposo's bundle has a zero per fiber (0 ms⁻¹ ≠ 0 kg) and must then manage zero divisors and non-comparable zeros (his 2019 order analysis shows zeros of different fibers are never comparable). Neither choice dominates; the torsor page shows a third option (no distinguished zero at all).
  5. Kinds beyond dimension. Nothing in the lineage separates torque from energy; Raposo concedes the algebra "cannot distinguish this detail". Whitney's several-isomorphic-models device for counting is the germ of a kind system, but whether it can coexist with multiplicative closure (and a representation theorem) is open — the mp-units quantity_spec design is the engineering answer awaiting a Whitney-style algebraic one.
  6. The habitat of 2 bl + 3 ck. Whitney computes distributively with formal cross-kind sums but (in the available text) never assigns them a structure. Hart's dimensioned vector spaces are the natural completion; connecting them back to quantity structures — direct sums of birays with a compatible product — appears in none of the local artifacts.
  7. Convergence of the descendants. Jonsson reports his quantity spaces and Raposo's fiber bundles "have been shown to be completely equivalent", citing a 2021 Raposo manuscript — i.e. the two independent modern re-axiomatizations of the Whitney lineage agree, essentially on "free-abelian-group of dimensions + one-dimensional fibers + per-fiber zeros". Whether that equivalence extends to a faithful embedding of Whitney's original (dense-exponent, single-zero) structures is not addressed in the manuscript's published surroundings.

Sources ​