Reverse mathematics
This article has multiple issues. Please help improve it or discuss these issues on the talk page. (Learn how and when to remove these template messages)
(Learn how and when to remove this template message)

Reverse mathematics is a program in mathematical logic that seeks to determine which axioms are required to prove theorems of mathematics. Its defining method can briefly be described as "going backwards from the theorems to the axioms", in contrast to the ordinary mathematical practice of deriving theorems from axioms. It can be conceptualized as sculpting out necessary conditions from sufficient ones.
The reverse mathematics program was foreshadowed by results in set theory such as the classical theorem that the axiom of choice and Zorn's lemma are equivalent over ZF set theory. The goal of reverse mathematics, however, is to study possible axioms of ordinary theorems of mathematics rather than possible axioms for set theory.
Reverse mathematics is usually carried out using subsystems of secondorder arithmetic, where many of its definitions and methods are inspired by previous work in constructive analysis and proof theory. The use of secondorder arithmetic also allows many techniques from recursion theory to be employed; many results in reverse mathematics have corresponding results in computable analysis. Recently, higherorder reverse mathematics has been introduced, in which the focus is on subsystems of higherorder arithmetic, and the associated richer language.
The program was founded by Harvey Friedman (1975, 1976) and brought forward by Steve Simpson. A standard reference for the subject is (Simpson 2009), while an introduction for nonspecialists is (Stillwell 2018). An introduction to higherorder reverse mathematics, and also the founding paper, is (Kohlenbach (2005)).
General principlesEdit
In reverse mathematics, one starts with a framework language and a base theory—a core axiom system—that is too weak to prove most of the theorems one might be interested in, but still powerful enough to develop the definitions necessary to state these theorems. For example, to study the theorem “Every bounded sequence of real numbers has a supremum” it is necessary to use a base system which can speak of real numbers and sequences of real numbers.
For each theorem that can be stated in the base system but is not provable in the base system, the goal is to determine the particular axiom system (stronger than the base system) that is necessary to prove that theorem. To show that a system S is required to prove a theorem T, two proofs are required. The first proof shows T is provable from S; this is an ordinary mathematical proof along with a justification that it can be carried out in the system S. The second proof, known as a reversal, shows that T itself implies S; this proof is carried out in the base system. The reversal establishes that no axiom system S′ that extends the base system can be weaker than S while still proving T.
Use of secondorder arithmeticEdit
Most reverse mathematics research focuses on subsystems of secondorder arithmetic. The body of research in reverse mathematics has established that weak subsystems of secondorder arithmetic suffice to formalize almost all undergraduatelevel mathematics. In secondorder arithmetic, all objects can be represented as either natural numbers or sets of natural numbers. For example, in order to prove theorems about real numbers, the real numbers can be represented as Cauchy sequences of rational numbers, each of which can be represented as a set of natural numbers.
The axiom systems most often considered in reverse mathematics are defined using axiom schemes called comprehension schemes. Such a scheme states that any set of natural numbers definable by a formula of a given complexity exists. In this context, the complexity of formulas is measured using the arithmetical hierarchy and analytical hierarchy.
The reason that reverse mathematics is not carried out using set theory as a base system is that the language of set theory is too expressive. Extremely complex sets of natural numbers can be defined by simple formulas in the language of set theory (which can quantify over arbitrary sets). In the context of secondorder arithmetic, results such as Post's theorem establish a close link between the complexity of a formula and the (non)computability of the set it defines.
Another effect of using secondorder arithmetic is the need to restrict general mathematical theorems to forms that can be expressed within arithmetic. For example, secondorder arithmetic can express the principle "Every countable vector space has a basis" but it cannot express the principle "Every vector space has a basis". In practical terms, this means that theorems of algebra and combinatorics are restricted to countable structures, while theorems of analysis and topology are restricted to separable spaces. Many principles that imply the axiom of choice in their general form (such as "Every vector space has a basis") become provable in weak subsystems of secondorder arithmetic when they are restricted. For example, "every field has an algebraic closure" is not provable in ZF set theory, but the restricted form "every countable field has an algebraic closure" is provable in RCA_{0}, the weakest system typically employed in reverse mathematics.
Use of higherorder arithmeticEdit
A recent strand of higherorder reverse mathematics research, initiated by Ulrich Kohlenbach, focuses on subsystems of higherorder arithmetic (Kohlenbach (2005)). Due to the richer language of higherorder arithmetic, the use of representations (aka 'codes') common in secondorder arithmetic, is greatly reduced. For example, a continuous function on the Cantor space is just a function that maps binary sequences to binary sequences, and that also satisfies the usual 'epsilondelta'definition of continuity.
Higherorder reverse mathematics includes higherorder versions of (secondorder) comprehension schemes. Such a higherorder axiom states the existence of a functional that decides the truth or falsity of formulas of a given complexity. In this context, the complexity of formulas is also measured using the arithmetical hierarchy and analytical hierarchy. The higherorder counterparts of the major subsystems of secondorder arithmetic generally prove the same secondorder sentences (or a large subset) as the original secondorder systems (see Kohlenbach (2005) and Hunter (2008)). For instance, the base theory of higherorder reverse mathematics, called RCA^{$\omega$ }
_{0}, proves the same sentences as RCA_{0}, up to language.
As noted in the previous paragraph, secondorder comprehension axioms easily generalize to the higherorder framework. However, theorems expressing the compactness of basic spaces behave quite differently in second and higherorder arithmetic: on one hand, when restricted to countable covers/the language of secondorder arithmetic, the compactness of the unit interval is provable in WKL_{0} from the next section. On the other hand, given uncountable covers/the language of higherorder arithmetic, the compactness of the unit interval is only provable from (full) secondorder arithmetic (NormannSanders (2018)). Other covering lemmas (e.g. due to Lindelöf, Vitali, Besicovitch, etc.) exhibit the same behavior, and many basic properties of the gauge integral are equivalent to the compactness of the underlying space.
The big five subsystems of secondorder arithmeticEdit
Secondorder arithmetic is a formal theory of the natural numbers and sets of natural numbers. Many mathematical objects, such as countable rings, groups, and fields, as well as points in effective Polish spaces, can be represented as sets of natural numbers, and modulo this representation can be studied in secondorder arithmetic.
Reverse mathematics makes use of several subsystems of secondorder arithmetic. A typical reverse mathematics theorem shows that a particular mathematical theorem T is equivalent to a particular subsystem S of secondorder arithmetic over a weaker subsystem B. This weaker system B is known as the base system for the result; in order for the reverse mathematics result to have meaning, this system must not itself be able to prove the mathematical theorem T.^{[citation needed]}
Simpson (2009) describes five particular subsystems of secondorder arithmetic, which he calls the Big Five, that occur frequently in reverse mathematics. In order of increasing strength, these systems are named by the initialisms RCA_{0}, WKL_{0}, ACA_{0}, ATR_{0}, and Π^{1}
_{1}CA_{0}.
The following table summarizes the "big five" systems (Simpson (2009, p.42)) and lists the counterpart systems in higherorder arithmetic (Kohlenbach (2008)). The latter generally prove the same secondorder sentences (or a large subset) as the original secondorder systems (see Kohlenbach (2005) and Hunter (2008)).
Subsystem  Stands for  Ordinal  Corresponds roughly to  Comments  Higherorder counterpart 

RCA_{0}  Recursive comprehension axiom  ω^{ω}  Constructive mathematics (Bishop)  The base theory  RCA^{ω} _{0}; proves the same secondorder sentences as RCA_{0} 
WKL_{0}  Weak Kőnig's lemma  ω^{ω}  Finitistic reductionism (Hilbert)  Conservative over PRA (resp. RCA_{0}) for Π^{0} _{2} (resp. Π^{1} _{1}) sentences 
Fan functional; computes modulus of uniform continuity on for continuous functions 
ACA_{0}  Arithmetical comprehension axiom  ε_{0}  Predicativism (Weyl, Feferman)  Conservative over Peano arithmetic for arithmetical sentences  The 'Turing jump' functional expresses the existence of a discontinuous function on 
ATR_{0}  Arithmetical transfinite recursion  Γ_{0}  Predicative reductionism (Friedman, Simpson)  Conservative over Feferman's system IR for Π^{1} _{1} sentences 
The 'transfinite recursion' functional outputs the set claimed to exist by ATR_{0}. 
Π^{1} _{1}CA_{0} 
Π^{1} _{1} comprehension axiom 
Ψ_{0}(Ω_{ω})  Impredicativism  The Suslin functional decides Π^{1} _{1}formulas (restricted to secondorder parameters). 
The subscript _{0} in these names means that the induction scheme has been restricted from the full secondorder induction scheme (Simpson 2009, p. 6). For example, ACA_{0} includes the induction axiom (0 ∈ X ∧ ∀n(n ∈ X → n + 1 ∈ X)) → ∀n n ∈ X. This together with the full comprehension axiom of secondorder arithmetic implies the full secondorder induction scheme given by the universal closure of (φ(0) ∧ ∀n(φ(n) → φ(n+1))) → ∀n φ(n) for any secondorder formula φ. However ACA_{0} does not have the full comprehension axiom, and the subscript _{0} is a reminder that it does not have the full secondorder induction scheme either. This restriction is important: systems with restricted induction have significantly lower prooftheoretical ordinals than systems with the full secondorder induction scheme.
The base system RCA_{0}Edit
RCA_{0} is the fragment of secondorder arithmetic whose axioms are the axioms of Robinson arithmetic, induction for Σ^{0}
_{1} formulas, and comprehension for Δ^{0}
_{1} formulas.
The subsystem RCA_{0} is the one most commonly used as a base system for reverse mathematics. The initials "RCA" stand for "recursive comprehension axiom", where "recursive" means "computable", as in recursive function. This name is used because RCA_{0} corresponds informally to "computable mathematics". In particular, any set of natural numbers that can be proven to exist in RCA_{0} is computable, and thus any theorem which implies that noncomputable sets exist is not provable in RCA_{0}. To this extent, RCA_{0} is a constructive system, although it does not meet the requirements of the program of constructivism because it is a theory in classical logic including the law of excluded middle.
Despite its seeming weakness (of not proving any noncomputable sets exist), RCA_{0} is sufficient to prove a number of classical theorems which, therefore, require only minimal logical strength. These theorems are, in a sense, below the reach of the reverse mathematics enterprise because they are already provable in the base system. The classical theorems provable in RCA_{0} include:
 Basic properties of the natural numbers, integers, and rational numbers (for example, that the latter form an ordered field).
 Basic properties of the real numbers (the real numbers are an Archimedean ordered field; any nested sequence of closed intervals whose lengths tend to zero has a single point in its intersection; the real numbers are not countable).
 The Baire category theorem for a complete separable metric space (the separability condition is necessary to even state the theorem in the language of secondorder arithmetic).
 The intermediate value theorem on continuous real functions.
 The Banach–Steinhaus theorem for a sequence of continuous linear operators on separable Banach spaces.
 A weak version of Gödel's completeness theorem (for a set of sentences, in a countable language, that is already closed under consequence).
 The existence of an algebraic closure for a countable field (but not its uniqueness).
 The existence and uniqueness of the real closure of a countable ordered field.
The firstorder part of RCA_{0} (the theorems of the system that do not involve any set variables) is the set of theorems of firstorder Peano arithmetic with induction limited to Σ^{0}
_{1} formulas. It is provably consistent, as is RCA_{0}, in full firstorder Peano arithmetic.
Weak Kőnig's lemma WKL_{0}Edit
The subsystem WKL_{0} consists of RCA_{0} plus a weak form of Kőnig's lemma, namely the statement that every infinite subtree of the full binary tree (the tree of all finite sequences of 0's and 1's) has an infinite path. This proposition, which is known as weak Kőnig's lemma, is easy to state in the language of secondorder arithmetic. WKL_{0} can also be defined as the principle of Σ^{0}
_{1} separation (given two Σ^{0}
_{1} formulas of a free variable n which are exclusive, there is a class containing all n satisfying the one and no n satisfying the other).
The following remark on terminology is in order. The term “weak Kőnig's lemma” refers to the sentence which says that any infinite subtree of the binary tree has an infinite path. When this axiom is added to RCA_{0}, the resulting subsystem is called WKL_{0}. A similar distinction between particular axioms, on the one hand, and subsystems including the basic axioms and induction, on the other hand, is made for the stronger subsystems described below.
In a sense, weak Kőnig's lemma is a form of the axiom of choice (although, as stated, it can be proven in classical Zermelo–Fraenkel set theory without the axiom of choice). It is not constructively valid in some senses of the word constructive.
To show that WKL_{0} is actually stronger than (not provable in) RCA_{0}, it is sufficient to exhibit a theorem of WKL_{0} which implies that noncomputable sets exist. This is not difficult; WKL_{0} implies the existence of separating sets for effectively inseparable recursively enumerable sets.
It turns out that RCA_{0} and WKL_{0} have the same firstorder part, meaning that they prove the same firstorder sentences. WKL_{0} can prove a good number of classical mathematical results which do not follow from RCA_{0}, however. These results are not expressible as firstorder statements but can be expressed as secondorder statements.
The following results are equivalent to weak Kőnig's lemma and thus to WKL_{0} over RCA_{0}:
 The Heine–Borel theorem for the closed unit real interval, in the following sense: every covering by a sequence of open intervals has a finite subcovering.
 The Heine–Borel theorem for complete totally bounded separable metric spaces (where covering is by a sequence of open balls).
 A continuous real function on the closed unit interval (or on any compact separable metric space, as above) is bounded (or: bounded and reaches its bounds).
 A continuous real function on the closed unit interval can be uniformly approximated by polynomials (with rational coefficients).
 A continuous real function on the closed unit interval is uniformly continuous.
 A continuous real function on the closed unit interval is Riemann integrable.
 The Brouwer fixed point theorem (for continuous functions on a finite product of copies of the closed unit interval).
 The separable Hahn–Banach theorem in the form: a bounded linear form on a subspace of a separable Banach space extends to a bounded linear form on the whole space.
 The Jordan curve theorem
 Gödel's completeness theorem (for a countable language).
 Determinacy for open (or even clopen) games on {0,1} of length ω.
 Every countable commutative ring has a prime ideal.
 Every countable formally real field is orderable.
 Uniqueness of algebraic closure (for a countable field).
Arithmetical comprehension ACA_{0}Edit
ACA_{0} is RCA_{0} plus the comprehension scheme for arithmetical formulas (which is sometimes called the "arithmetical comprehension axiom"). That is, ACA_{0} allows us to form the set of natural numbers satisfying an arbitrary arithmetical formula (one with no bound set variables, although possibly containing set parameters). Actually, it suffices to add to RCA_{0} the comprehension scheme for Σ_{1} formulas in order to obtain full arithmetical comprehension.
The firstorder part of ACA_{0} is exactly firstorder Peano arithmetic; ACA_{0} is a conservative extension of firstorder Peano arithmetic. The two systems are provably (in a weak system) equiconsistent. ACA_{0} can be thought of as a framework of predicative mathematics, although there are predicatively provable theorems that are not provable in ACA_{0}. Most of the fundamental results about the natural numbers, and many other mathematical theorems, can be proven in this system.
One way of seeing that ACA_{0} is stronger than WKL_{0} is to exhibit a model of WKL_{0} that doesn't contain all arithmetical sets. In fact, it is possible to build a model of WKL_{0} consisting entirely of low sets using the low basis theorem, since low sets relative to low sets are low.
The following assertions are equivalent to ACA_{0} over RCA_{0}:
 The sequential completeness of the real numbers (every bounded increasing sequence of real numbers has a limit).
 The Bolzano–Weierstrass theorem.
 Ascoli's theorem: every bounded equicontinuous sequence of real functions on the unit interval has a uniformly convergent subsequence.
 Every countable commutative ring has a maximal ideal.
 Every countable vector space over the rationals (or over any countable field) has a basis.
 Every countable field has a transcendence basis.
 Kőnig's lemma (for arbitrary finitely branching trees, as opposed to the weak version described above).
 Various theorems in combinatorics, such as certain forms of Ramsey's theorem.
Arithmetical transfinite recursion ATR_{0}Edit
The system ATR_{0} adds to ACA_{0} an axiom which states, informally, that any arithmetical functional (meaning any arithmetical formula with a free number variable n and a free class variable X, seen as the operator taking X to the set of n satisfying the formula) can be iterated transfinitely along any countable well ordering starting with any set. ATR_{0} is equivalent over ACA_{0} to the principle of Σ^{1}
_{1} separation. ATR_{0} is impredicative, and has the prooftheoretic ordinal , the supremum of that of predicative systems.
ATR_{0} proves the consistency of ACA_{0}, and thus by Gödel's theorem it is strictly stronger.
The following assertions are equivalent to ATR_{0} over RCA_{0}:
 Any two countable well orderings are comparable. That is, they are isomorphic or one is isomorphic to a proper initial segment of the other.
 Ulm's theorem for countable reduced Abelian groups.
 The perfect set theorem, which states that every uncountable closed subset of a complete separable metric space contains a perfect closed set.
 Lusin's separation theorem (essentially Σ^{1}
_{1} separation).  Determinacy for open sets in the Baire space.
Π^{1}
_{1} comprehension Π^{1}
_{1}CA_{0}Edit
Π^{1}
_{1}CA_{0} is stronger than arithmetical transfinite recursion and is fully impredicative. It consists of RCA_{0} plus the comprehension scheme for Π^{1}
_{1} formulas.
In a sense, Π^{1}
_{1}CA_{0} comprehension is to arithmetical transfinite recursion (Σ^{1}
_{1} separation) as ACA_{0} is to weak Kőnig's lemma (Σ^{0}
_{1} separation). It is equivalent to several statements of descriptive set theory whose proofs make use of strongly impredicative arguments; this equivalence shows that these impredicative arguments cannot be removed.
The following theorems are equivalent to Π^{1}
_{1}CA_{0} over RCA_{0}:
 The Cantor–Bendixson theorem (every closed set of reals is the union of a perfect set and a countable set).
 Every countable abelian group is the direct sum of a divisible group and a reduced group.
Additional systemsEdit
 Weaker systems than recursive comprehension can be defined. The weak system RCA^{*}
_{0} consists of elementary function arithmetic EFA (the basic axioms plus Δ^{0}
_{0} induction in the enriched language with an exponential operation) plus Δ^{0}
_{1} comprehension. Over RCA^{*}
_{0}, recursive comprehension as defined earlier (that is, with Σ^{0}
_{1} induction) is equivalent to the statement that a polynomial (over a countable field) has only finitely many roots and to the classification theorem for finitely generated Abelian groups. The system RCA^{*}
_{0} has the same proof theoretic ordinal ω^{3} as EFA and is conservative over EFA for Π^{0}
_{2} sentences.  Weak Weak Kőnig's Lemma is the statement that a subtree of the infinite binary tree having no infinite paths has an asymptotically vanishing proportion of the leaves at length n (with a uniform estimate as to how many leaves of length n exist). An equivalent formulation is that any subset of Cantor space that has positive measure is nonempty (this is not provable in RCA_{0}). WWKL_{0} is obtained by adjoining this axiom to RCA_{0}. It is equivalent to the statement that if the unit real interval is covered by a sequence of intervals then the sum of their lengths is at least one. The model theory of WWKL_{0} is closely connected to the theory of algorithmically random sequences. In particular, an ωmodel of RCA_{0} satisfies weak weak Kőnig's lemma if and only if for every set X there is a set Y which is 1random relative to X.
 DNR (short for "diagonally nonrecursive") adds to RCA_{0} an axiom asserting the existence of a diagonally nonrecursive function relative to every set. That is, DNR states that, for any set A, there exists a total function f such that for all e the eth partial recursive function with oracle A is not equal to f. DNR is strictly weaker than WWKL (Lempp et al., 2004).
 Δ^{1}
_{1}comprehension is in certain ways analogous to arithmetical transfinite recursion as recursive comprehension is to weak Kőnig's lemma. It has the hyperarithmetical sets as minimal ωmodel. Arithmetical transfinite recursion proves Δ^{1}
_{1}comprehension but not the other way around.  Σ^{1}
_{1}choice is the statement that if η(n,X) is a Σ^{1}
_{1} formula such that for each n there exists an X satisfying η then there is a sequence of sets X_{n} such that η(n,X_{n}) holds for each n. Σ^{1}
_{1}choice also has the hyperarithmetical sets as minimal ωmodel. Arithmetical transfinite recursion proves Σ^{1}
_{1}choice but not the other way around.  HBU (short for "uncountable HeineBorel") expresses the (opencover) compactness of the unit interval, involving uncountable covers. The latter aspect of HBU makes it only expressible in the language of thirdorder arithmetic. Cousin's theorem (1895) implies HBU, and these theorems use the same notion of cover due to Cousin and Lindelöf. HBU is hard to prove: in terms of the usual hierarchy of comprehension axioms, a proof of HBU requires full secondorder arithmetic (NormannSanders (2018)).
ωmodels and βmodelsEdit
The ω in ωmodel stands for the set of nonnegative integers (or finite ordinals). An ωmodel is a model for a fragment of secondorder arithmetic whose firstorder part is the standard model of Peano arithmetic, but whose secondorder part may be nonstandard. More precisely, an ωmodel is given by a choice S⊆2^{ω} of subsets of ω. The firstorder variables are interpreted in the usual way as elements of ω, and +, × have their usual meanings, while secondorder variables are interpreted as elements of S. There is a standard ω model where one just takes S to consist of all subsets of the integers. However, there are also other ωmodels; for example, RCA_{0} has a minimal ωmodel where S consists of the recursive subsets of ω.
A β model is an ω model that is equivalent to the standard ωmodel for Π^{1}
_{1} and Σ^{1}
_{1} sentences (with parameters).
Nonω models are also useful, especially in the proofs of conservation theorems.
See alsoEdit
ReferencesEdit
 AmbosSpies, K.; KjosHanssen, B.; Lempp, S.; Slaman, T.A. (2004), "Comparing DNR and WWKL", Journal of Symbolic Logic, 69 (4): 1089, arXiv:1408.2281, doi:10.2178/jsl/1102022212.
 Friedman, Harvey (1975), "Some systems of secondorder arithmetic and their use", Proceedings of the International Congress of Mathematicians (Vancouver, B. C., 1974), Vol. 1, Canad. Math. Congress, Montreal, Que., pp. 235–242, MR 0429508
 Friedman, Harvey; Martin, D. A.; Soare, R. I.; Tait, W. W. (1976), "Meeting of the Association for Symbolic Logic: Systems of secondorder arithmetic with restricted induction, I, II", The Journal of Symbolic Logic, 41 (2): 557–559, doi:10.2307/2272259, JSTOR 2272259
 Hunter, James (2008), Reverse Topology (PDF), PhD, University of WisconsinMadison
 Kohlenbach, Ulrich (2005), "Higher order reversemathematics", Higher Order Reverse Mathematics, Reverse Mathematics 2001 (PDF), Lecture notes in Logic, Cambridge University Press, pp. 281–295, doi:10.1017/9781316755846.018, ISBN 9781316755846
 Normann, Dag; Sanders, Sam (2018), "On the mathematical and foundational significance of the uncountable", Journal of Mathematical Logic, 19: 1950001, arXiv:1711.08939, doi:10.1142/S0219061319500016
 Simpson, Stephen G. (2009), Subsystems of secondorder arithmetic, Perspectives in Logic (2nd ed.), Cambridge University Press, doi:10.1017/CBO9780511581007, ISBN 9780521884396, MR 2517689
 Stillwell, John (2018), Reverse Mathematics, proofs from the inside out, Princeton University Press, ISBN 9780691177175
 Solomon, Reed (1999), "Ordered groups: a case study in reverse mathematics", The Bulletin of Symbolic Logic, 5 (1): 45–58, CiteSeerX 10.1.1.364.9553, doi:10.2307/421140, ISSN 10798986, JSTOR 421140, MR 1681895