Ekeland's variational principle in weak and strong systems of arithmetic (Q2003985)

The paper under review studies various versions of Ekeland's variational principle in the setting of reverse mathematic. Let us first recall that reverse mathematics is a research programme within mathematical logic. It aims to determine the minimal set existence principles that are needed to establish a given mathematical theorem (with a focus on countable sets coded by subsets of \(\mathbb N\)). We also recall that, for a complete metric space \((X,d)\), a map \(f:X\to[0,\infty]\) is lower semi-continuous if it satisfies the \(\varepsilon\)-\(\delta\)-definition of continuity with \(|f(x_0)-f(x)|<\varepsilon\) replaced by \(f(x)>f(x_0)-\varepsilon\). Ekeland's variational principle asserts that any such \(f\) has an \(\varepsilon\)-critical point \(x_*\) for each \(\varepsilon>0\), which means that \(f(x)>f(x_*)-\varepsilon\cdot d(x,x_*)\) holds for all \(x\in X\backslash\{x_*\}\). More specifically, the paper refers to this statement as the free variational principle (FVP); the localized variational principle (LVP) asserts that one can additionally satisfy \(f(x_*)\leq f(x_0)-\varepsilon\cdot d(x_*,x_0)\) for any given \(x_0\in X\). More specifically, then, the paper shows that important set existence principles from reverse mathematics are equivalent to versions of Ekeland's variational principle (over the usual base theory \(\mathsf{RCA_0}\)): weak Kőnig's lemma is equivalent to the FVP under the assumptions that both (1) \(f\) is continuous and (2) \(X\) is compact; still assuming (1) and (2), the LVP is equivalent to arithmetical comprehension; the latter is also equivalent to the FVP when we assume (1) or (2) but not both; without assuming (1) or (2), both the FVP and the LVP is equivalent to \(\Pi^1_1\)-comprehension, a much stronger set existence axiom; finally, \(\Pi^1_1\)-comprehension is also equivalent to the LVP for continuous functions. As the authors point out, the last equivalence is particularly remarkable, because theorems about continuous functions are typically much weaker. While the coding of continuous functions in reverse mathematics is standard, codes for semi-continuous functions are introduced in the paper under review. Roughly speaking, a code of a lower semi-continuous \(f\) enumerates tuples \((a,r,q)\) such that \(r>d(x,a)\) entails \(f(x)\geq q\), and such that \(f(x)\) is the supremum of all relevant \(q\). A code is called honest when it contains all tuples with the aforementioned property. While it is possible to work with codes over \(\mathsf{RCA_0}\), the statement that any code can be transformed into an honest code is itself equivalent to \(\Pi^1_1\)-comprehension, at least over \(\mathsf{ACA_0}\) (see Lemma 5.2(iv) of the paper for one direction, and combine Theorem 8.2(i) with Proposition 9.5 for the reversal). It would be interesting to explore connections between the paper under review and other foundational perspectives on Ekeland's principle: Within reverse mathematics itself, one can ask which consequences of Ekeland's principle -- possibly of lower logical complexity -- require strong set existence principles as well (the authors mention forthcoming work on Caristi's fixed point theorem). In the context of proof mining, one may wonder whether these strong principles translate into large quantitative bounds, or whether concrete applications of Ekeland's principle are ``tame''. The role of coding can be formally investigated in third-order arithmetic, where Ekeland's principle has different foundational properties, as pointed out by \textit{S. Sanders} [``Representations and the foundations of mathematics'', Preprint, \url{arXiv:1910.07913}]. Relations with the axiom of choice are, for example, discussed in a survey by \textit{W. A. Kirk} [Fixed Point Theory Appl. 2015, Paper No. 215, 17 p. (2015; Zbl 1347.54094)]. The paper is well-written and accessible, as it recalls relevant prerequisites from mathematical logic. At the same time, it contains a lot of substantial mathematics, and it is very interesting on a technical level (consider the use of pseudo-fibrations in Section 10). I warmly recommend it to anyone who is interested in the area!