Basic Law V: Predicative Abstraction and Arithmetic

• Basic Law V is an abstraction principle that converts unary concepts into first-order objects by equating extensions of coextensive concepts.
• It employs restricted hyperarithmetic comprehension to skirt Russell’s paradox while defining an internal successor operator essential for arithmetic.
• This model of predicative arithmetic situates BLV’s interpretability strength above Hume’s Principle but below impredicative second-order systems.

Basic Law V is Frege’s abstraction principle for extensions: in a many-sorted second-order framework with first-order domain $M$, unary concepts $S_1 \subseteq P(M)$, higher-arity relations $S_n \subseteq P(M^n)$, and an extension operator $\partial:S_1\to M$, it asserts that $\partial$ identifies exactly coextensive unary concepts, namely

$$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$

where $X=Y$ abbreviates $\forall x\,(x\in X\leftrightarrow x\in Y)$. Equivalently, a structure $(M,S_1,S_2,\ldots,\partial)$ is a Basic Law V model iff $\partial$ is injective on $S_1 \subseteq P(M)$0. In the setting studied in "Comparing Hume’s Principle, Basic Law V and Peano Arithmetic" (Walsh, 2014), Basic Law V is examined not with full impredicative comprehension, where it is inconsistent by Russell’s paradox, but through restricted comprehension schemes that permit a precise comparison with Hume’s Principle and with canonical subsystems of second-order arithmetic.

1. Formal setting and the content of Basic Law V

The background framework is many-sorted and second-order. Structures have the form

$S_1 \subseteq P(M)$1

with $S_1 \subseteq P(M)$2 the domain of first-order objects, $S_1 \subseteq P(M)$3 the sort of $S_1 \subseteq P(M)$4-ary second-order relations on $S_1 \subseteq P(M)$5, and $S_1 \subseteq P(M)$6 either the number operator $S_1 \subseteq P(M)$7 in the Hume’s Principle signature or the extension operator $S_1 \subseteq P(M)$8 in the Basic Law V signature. Membership relations $S_1 \subseteq P(M)$9 between tuples and $S_n \subseteq P(M^n)$0-ary relations are interpreted absolutely and suppressed; every structure is isomorphic to one in which these relations are absolute (Walsh, 2014).

Within that setting, Basic Law V states that extensions are extensional in the strongest possible sense: two concepts have the same extension exactly when they are coextensive. The principle therefore turns unary concepts into first-order objects via $S_n \subseteq P(M^n)$1, and the equation $S_n \subseteq P(M^n)$2 becomes a first-order proxy for second-order extensional identity.

The central formal distinction from Hume’s Principle is that BLV is stated directly in terms of equality of concepts, whereas HP uses equinumerosity:

$S_n \subseteq P(M^n)$3

In BLV, the right-hand side contains only concept identity; in HP, it quantifies over higher-order objects, namely bijections. This structural difference is decisive in the later comparison of interpretability strength.

The classical inconsistency of BLV arises when it is combined with full impredicative second-order comprehension. From full comprehension one forms

$S_n \subseteq P(M^n)$4

and BLV then yields a contradiction by Russell’s paradox. The inconsistency is therefore not attributed to the abstraction principle in isolation, but to its interaction with unrestricted comprehension.

2. Restricted comprehension and predicative fragments

The paper studies BLV through a hierarchy of restricted comprehension schemes parameterized by the complexity of second-order formulas (Walsh, 2014). Arithmetical formulas are those with no bound relation variables, and the usual classes $S_n \subseteq P(M^n)$5 and $S_n \subseteq P(M^n)$6 are formed by alternating blocks of second-order quantifiers over relation variables. Full comprehension for a theory $S_n \subseteq P(M^n)$7 has the form

$S_n \subseteq P(M^n)$8

with parameters allowed and $S_n \subseteq P(M^n)$9 not free in $\partial:S_1\to M$0.

From this schema the paper isolates several subsystems. $\partial:S_1\to M$1 is $\partial:S_1\to M$2 with comprehension restricted to arithmetical formulas. $\partial:S_1\to M$3 replaces comprehension by the hyperarithmetic scheme

$\partial:S_1\to M$4

where $\partial:S_1\to M$5 is $\partial:S_1\to M$6 and $\partial:S_1\to M$7 is $\partial:S_1\to M$8. $\partial:S_1\to M$9 consists of arithmetical comprehension together with $\partial$0-Choice:

$\partial$1

for $\partial$2 formulas $\partial$3.

These restrictions are calibrated against the standard subsystems of second-order arithmetic used by Simpson: $\partial$4, $\partial$5, $\partial$6, $\partial$7, and $\partial$8. A key general fact is that in all three signatures—$\partial$9, $$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$0, and $$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$1—$$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$2 implies $$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$3. In the paper’s usage, the expression “predicative” refers to this hyperarithmetic level, especially $$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$4-comprehension and $$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$5-choice rather than full $$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$6-comprehension.

For BLV, these predicative fragments are nontrivial because BLV itself survives the restriction of comprehension. The resulting question is not whether BLV is consistent under full comprehension—it is not—but how much arithmetic and second-order reasoning remain interpretable once comprehension is weakened.

3. The extension operator as a source of successor and arithmetic

A distinctive feature of BLV is that the extension operator directly induces a successor-like map on first-order objects:

$$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$7

In $$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$8, the graph of this function can be defined by equivalent $$\forall X\,\forall Y\,\big(\partial X=\partial Y\leftrightarrow X=Y\big),$$9 and $X=Y$0 definitions and then obtained by $X=Y$1-comprehension (Walsh, 2014). Moreover, any $X=Y$2-model contains an injective non-surjective unary function: the $X=Y$3-successor necessarily gives an injective non-surjective map $X=Y$4 with hyperarithmetically definable graph. This is one of the paper’s most important structural observations about restricted BLV.

To convert this structural feature into arithmetic, the paper isolates a finite axiom $X=Y$5. It asserts the existence of the $X=Y$6-successor, the existence of a least inductive set containing $X=Y$7 and closed under that successor, and the existence of addition, multiplication, and order making the resulting structure a model of Robinson’s $X=Y$8:

$X=Y$9

The significance of $\forall x\,(x\in X\leftrightarrow x\in Y)$0 is that it packages, in finitely axiomatized form, the exact additional ingredients needed to recover a predicative arithmetic core inside BLV. The successor is not imported from an external arithmetic language; it is generated internally by the extension operator itself. This suggests that the interpretability strength of predicative BLV depends not only on restricted comprehension but also on the fact that $\forall x\,(x\in X\leftrightarrow x\in Y)$1 converts unary concepts into a first-order progression that can be recursively iterated.

The paper also records lower-level comparisons: $\forall x\,(x\in X\leftrightarrow x\in Y)$2 is mutually interpretable with $\forall x\,(x\in X\leftrightarrow x\in Y)$3, with this point attributed to work of Heck, Ganea, and Visser. That result locates very weak BLV fragments near Robinson arithmetic, while the addition of hyperarithmetic comprehension and $\forall x\,(x\in X\leftrightarrow x\in Y)$4 raises the strength substantially.

4. Hyperarithmetic BLV and interpretability bounds

Interpretability is understood model-theoretically: $\forall x\,(x\in X\leftrightarrow x\in Y)$5 means that every model of $\forall x\,(x\in X\leftrightarrow x\in Y)$6 uniformly defines, without parameters, a model of $\forall x\,(x\in X\leftrightarrow x\in Y)$7, where uniformity requires that the same defining formulas work in every $\forall x\,(x\in X\leftrightarrow x\in Y)$8-model (Walsh, 2014). The relations

$\forall x\,(x\in X\leftrightarrow x\in Y)$9

are used throughout. Proposition 1.8 links interpretability and consistency strength when $(M,S_1,S_2,\ldots,\partial)$0 is finitely axiomatizable and satisfies $(M,S_1,S_2,\ldots,\partial)$1.

The central BLV theorem is Theorem 4.2. For any real $(M,S_1,S_2,\ldots,\partial)$2, there is a map

$(M,S_1,S_2,\ldots,\partial)$3

whose graph is $(M,S_1,S_2,\ldots,\partial)$4 such that:

1. $(M,S_1,S_2,\ldots,\partial)$5 is a model of $(M,S_1,S_2,\ldots,\partial)$6 and of $(M,S_1,S_2,\ldots,\partial)$7.
2. $(M,S_1,S_2,\ldots,\partial)$8 and $(M,S_1,S_2,\ldots,\partial)$9 are mutually interpretable, uniformly in $\partial$0.

The construction uses Kondo’s Uniformization Theorem to choose, $\partial$1-uniformly, hyperarithmetical indices for sets in $\partial$2. Writing

$\partial$3

the relation $\partial$4 is uniformized to $\partial$5, and $\partial$6 is defined by

$\partial$7

Since $\partial$8 is injective and $\partial$9-definable, the resulting structure satisfies the BLV axioms at the $S_1 \subseteq P(M)$00-choice level. The successor $S_1 \subseteq P(M)$01 is then $S_1 \subseteq P(M)$02-definable, and from the recursion

$S_1 \subseteq P(M)$03

the range $S_1 \subseteq P(M)$04 is used to transport the ordinary arithmetic operations of $S_1 \subseteq P(M)$05 onto the BLV-internal inductive set.

Corollary 4.3 yields the interpretability bounds

$S_1 \subseteq P(M)$06

Hence a consistent extension of the hyperarithmetic fragment of BLV interprets the hyperarithmetic fragment of second-order Peano arithmetic. The strict upper bound comes from formalizing the construction in $S_1 \subseteq P(M)$07, which proves the consistency of $S_1 \subseteq P(M)$08 and therefore cannot be interpretable in it by Proposition 1.8.

5. Contrast with Hume’s Principle and the failure of a predicative Frege theorem

The paper places BLV and HP side by side to measure the extent to which Fregean abstraction can recover arithmetic under predicative restrictions (Walsh, 2014). With full second-order comprehension, HP mutually interprets second-order Peano arithmetic: Frege showed $S_1 \subseteq P(M)$09, Boolos showed the converse, and consequently $S_1 \subseteq P(M)$10. No analogous full-comprehension result is available for BLV, because BLV with full comprehension is inconsistent.

At the hyperarithmetic level, however, the pattern reverses. BLV succeeds in interpreting predicative arithmetic through $S_1 \subseteq P(M)$11, whereas HP fails. The main HP bound is

$S_1 \subseteq P(M)$12

stated as Corollary 5.9. The proof proceeds by constructing $S_1 \subseteq P(M)$13-models over recursively saturated o-minimal expansions of real-closed fields. In these structures one defines

$S_1 \subseteq P(M)$14

where $S_1 \subseteq P(M)$15 is o-minimal dimension and $S_1 \subseteq P(M)$16 is Euler characteristic. These invariants are uniformly, indeed computably, definable in the parameters and classify definable sets up to definable bijection. Since such structures have definable Skolem functions, they satisfy $S_1 \subseteq P(M)$17; formalization in $S_1 \subseteq P(M)$18 then yields the consistency statement needed for the strict inequality.

Because $S_1 \subseteq P(M)$19 strictly extends $S_1 \subseteq P(M)$20, transitivity of interpretability implies that neither $S_1 \subseteq P(M)$21 nor $S_1 \subseteq P(M)$22 is interpretable in $S_1 \subseteq P(M)$23. This is the precise sense in which there is no predicative version of Frege’s Theorem. The phrase does not deny the impredicative Frege theorem for HP; rather, it states that the hyperarithmetic fragments of HP do not recover the corresponding hyperarithmetic fragments of second-order arithmetic.

The paper’s explanatory diagnosis is structural. BLV’s $S_1 \subseteq P(M)$24 gives a direct successor on extensions, while HP’s $S_1 \subseteq P(M)$25 remains a higher-order cardinality abstraction governed by the existence of bijections. A plausible implication is that the ability to generate a first-order successor internally is the key reason BLV’s predicative fragments outstrip HP’s in interpretability strength, despite HP’s much stronger impredicative behavior.

6. Model-theoretic tools, field-based constructions, and limitations

A general metatheorem organizes many of the model constructions. Let

$S_1 \subseteq P(M)$26

be a structure with arithmetical comprehension, in either the BLV or HP signature. If $S_1 \subseteq P(M)$27 is uniformly definable from parameters and $S_1 \subseteq P(M)$28 is recursively saturated, then $S_1 \subseteq P(M)$29 or $S_1 \subseteq P(M)$30 as appropriate; and $S_1 \subseteq P(M)$31 or $S_1 \subseteq P(M)$32 iff $S_1 \subseteq P(M)$33 has definable Skolem functions (Walsh, 2014). The proof uses compactness properties of saturated structures to finite-ize quantifier alternations, converting $S_1 \subseteq P(M)$34 and $S_1 \subseteq P(M)$35 definitions into uniform definability over $S_1 \subseteq P(M)$36.

Three hyperarithmetic tools are central on the BLV side. Kleene’s Theorem on Restricted Quantification states that if $S_1 \subseteq P(M)$37 is $S_1 \subseteq P(M)$38, then $S_1 \subseteq P(M)$39 is still $S_1 \subseteq P(M)$40, provably in $S_1 \subseteq P(M)$41. The Spector–Gandy Theorem states that every $S_1 \subseteq P(M)$42 predicate $S_1 \subseteq P(M)$43 is equivalent to $S_1 \subseteq P(M)$44 with arithmetic $S_1 \subseteq P(M)$45, and Corollary 4.6 shows that graphs of $S_1 \subseteq P(M)$46-uniform functions on $S_1 \subseteq P(M)$47 are $S_1 \subseteq P(M)$48-definable in $S_1 \subseteq P(M)$49. Kondo’s Uniformization Theorem turns $S_1 \subseteq P(M)$50 relations into functional ones and is used to obtain the injective map $S_1 \subseteq P(M)$51.

Field theory supplies contrasting model bases. In algebraically closed fields, Ax’s theorem implies that definable injective endofunctions are surjective. This yields models of $S_1 \subseteq P(M)$52 with $S_1 \subseteq P(M)$53 defined as finite cardinality for finite $S_1 \subseteq P(M)$54 and the negative of cofinite codimension for cofinite $S_1 \subseteq P(M)$55. The same Ax phenomenon prevents a $S_1 \subseteq P(M)$56 satisfying BLV at the $S_1 \subseteq P(M)$57 level, because $S_1 \subseteq P(M)$58 would be a definable injective non-surjective map. By contrast, o-minimal expansions of real-closed fields support the HP constructions because dimension and Euler characteristic classify definable sets up to definable bijection and definable Skolem functions are available. On the BLV side, separably closed fields of finite imperfection degree, together with uniform elimination of imaginaries and a definable pairing function, yield $S_1 \subseteq P(M)$59-models.

The paper also records several additional consequences. In $S_1 \subseteq P(M)$60, injective non-surjective definable maps always occur via the $S_1 \subseteq P(M)$61-successor; in $S_1 \subseteq P(M)$62, they need not occur. Linnebo’s successor axiom SA fails in a $S_1 \subseteq P(M)$63 model over an algebraically closed field with $S_1 \subseteq P(M)$64 interpreted as “finite size / negative cofinite deficit”: the pseudo-numbers are precisely $S_1 \subseteq P(M)$65, yet $S_1 \subseteq P(M)$66 has no $S_1 \subseteq P(M)$67 with $S_1 \subseteq P(M)$68 witnessing a successor by adding a point.

The open questions concern both proof-theoretic strength and formalization. The paper asks whether $S_1 \subseteq P(M)$69 implies $S_1 \subseteq P(M)$70, whether $S_1 \subseteq P(M)$71 implies $S_1 \subseteq P(M)$72, whether Ax’s theorem is provable in $S_1 \subseteq P(M)$73, whether uniform elimination of imaginaries for separably closed fields is provable in $S_1 \subseteq P(M)$74, and whether $S_1 \subseteq P(M)$75 or $S_1 \subseteq P(M)$76 are interpretable in $S_1 \subseteq P(M)$77. It also asks whether enriching the language with a binary-to-object function $S_1 \subseteq P(M)$78—whose graph is $S_1 \subseteq P(M)$79-definable but whose existence is not provable in $S_1 \subseteq P(M)$80 or $S_1 \subseteq P(M)$81 in general—would change interpretability strength.

Taken together, these results position Basic Law V as a principle that is inconsistent under full impredicative comprehension but robustly informative under hyperarithmetic restrictions. In that restricted environment, its extension operator supports an internal successor, a least inductive set, and an interpretation of hyperarithmetic second-order arithmetic, placing predicative BLV above predicative HP in interpretability strength while still below impredicative subsystems such as $S_1 \subseteq P(M)$82 (Walsh, 2014).