Reductions in PPP
Introduction
Total search problems [7], [9] still constitute an exotic domain of complexity theory — exotic in both senses “strange, unusual” and “remote, unexplored.” Take the class PPP, for example. It is known to include all of PPAD, and is defined in terms of the generic complete problem
Pigeon: Given a Boolean circuit C with n inputs and n outputs, find such that either or .
The class PPP consists of all search problems reducible to Pigeon. As for other problems known to be in PPP, [9] only mentions
Equal sums: Given positive integers such that , find two subsets such that .
and these two problems, Pigeon and Equal sums, are to our knowledge1 the only problems discussed in the literature that are known to be in PPP and not known to be in included classes, e.g. Nash.
Over the past decade, the class PPAD and its complete problems have substantially informed and advanced our algorithmic understanding of Game Theory [8]. Can PPP serve the same role for Cryptography? The generic problem Pigeon is loosely about collisions of hash functions; furthermore, it was recently pointed out that Factoring belongs to PWPP (the class of problems whose totality is proved through a weak pigeonhole principle), a subclass of PPP, via randomized reductions [6]. The two new problems that we introduce here, Minkowski and Dirichlet, are also motivated by Cryptography, since they are both the complexity renderings of mathematical results which have been used in the foundations of lattice-based crypto systems [1]. There may be important dividends in understanding better this class and its complete problems.
Incidentally, other classes of total problems from [9] have recently been connected to Cryptography: Factoring was shown to also belong in the class PPA, again via randomized reductions, and also in the class PWPP [6]. Note that the only other problems known to be in PPP ∩ PPA are the problems in PPAD, such as Nash [3], [4]. Determining whether Factoring is in PPAD (via randomized reductions) is a most important open question. Recall also the recent insight that PPAD is intractable under standard cryptographic assumptions [2] (standard in the sense that they are adopted in parts of the mainstream literature).
As we mentioned, Pigeon and Equal sums are, as far as we know, the only problems in PPP discussed in the literature.2 In other words, the picture of PPP prior to this note has been as shown in Fig. 1(a). In this note we introduce two new problems:
Minkowski: Given an matrix A with , find a nontrivial integer combination of its rows with norm less than one.
Dirichlet: Given n rational numbers and an integer N, find integers such that for all i, and .
We show that these two problems are in PPP. In fact, we show that Minkowski is remarkably expressive, in the sense that the two currently known non-generic problems in PPP (Equal sums and Dirichlet, leaving out problems, such as Nash, which belong to subclasses) are reducible to Minkowski.
Theorem 1 Minkowski and Dirichlet are reducible to Pigeon.
Theorem 2 Equal sums and Dirichlet are reducible to Minkowski.
None of these results is trivial, but the reduction from Minkowski to Pigeon is, surprisingly, the hardest to prove. In other words, the new picture of problems in PPP is as shown in Fig. 1(b). These results suggest that Minkowski is a natural candidate for a non-generic complete problem for PPP. Thus, the following important open problem is the main message of this work:
Conjecture 1 Pigeon is reducible to Minkowski.
When writing this note the authors were not aware of the following two, very relevant works (or the works didn't exist in the case of the second paper by Sotiraki et al. [10]).
Hoberg et al. [5]
In a recent work, Hoberg et al. [5] show (independently of this work) that the problem Number balancing is equivalent to polynomial approximations of Minkowski's theorem. The result most relevant to this note is the following. Given an algorithm that takes as input a lattice with and finds a non-zero vector such that , there exists a δ-approximation algorithm for the number balancing problem, where . Their proof is similar to the reduction of Equal sums to Minkowski presented in this note.
Sotiraki et al. [10]
In a recent unpublished manuscript, Sotiraki et al. [10] identify two new problems in PPP: a computational problem associated with Blichfeldt's fundamental theorem, and a generalized version of the Short Integer Solution problem from lattice based cryptography. They prove that both problems are in PPP, and furthermore that both problems are PPP-hard. This breakthrough therefore gives us the first natural PPP-complete problems. The authors provide a new proof that Minkowski is in PPP, via a reduction to Blichfeldt. Their reduction of Blichfeldt to Pigeon is similar in spirit to the reduction of Minkowski to Pigeon presented in this note.
Section snippets
Preliminaries
We are interested in search problems, that is, problems in which we are given the input to a problem in NP, and we are asked for a witness of this input (if none exists, we return “no”). This class of problems is often called FNP (for function problems in NP).
Within FNP, we are interested in total search problems, that is, problems in which “no” is never a legitimate answer, for all inputs. The class of all total search problems is denoted TFNP. Evidently, every total search problem in TFNP
Reducing to Pigeon
Lemma 1 Minkowski is reducible to Pigeon.
Proof Given an n by n matrix A with determinant strictly smaller than 1, let P be the fundamental parallepiped of the vectors spanned by the rows of A. We will construct a circuit C that computes a “modulo lattice” function similar to the function f in the proof of Minkowski's theorem, mapping the hypercube to P. A collision in this circuit (two inputs mapped to the same output) will give us a short lattice vector, i.e. an integer combination of rows of A with
Reducing to Minkowski
Lemma 3 Equal sums reduces to Minkowski
Proof We will construct an lower diagonal matrix, with determinant strictly less than 1. The first column of the matrix is . The diagonal is , and all other entries are zero: The determinant of the matrix is . By Minkowski's theorem, there must be a non-trivial integer combination of rows so that each coordinate is less than 1 in magnitude. Moreover, this integer combination
Acknowledgements
We are grateful to anonymous reviewers for a major simplification of the proof of Lemma 1 and other helpful comments.
References (10)
- et al.
Settling the complexity of two-player nash equilibrium
- et al.
On total functions, existence theorems and computational complexity
Theor. Comput. Sci.
(1991) On the complexity of the parity argument and other inefficient proofs of existence
J. Comput. Syst. Sci.
(1994)Generating hard instances of lattice problems
- et al.
On the cryptographic hardness of finding a nash equilibrium
Cited by (9)
A Faster Algorithm for Pigeonhole Equal Sums
2024, arXivPPP-Completeness and Extremal Combinatorics
2023, Leibniz International Proceedings in Informatics, LIPIcsPPP-Completeness and Extremal Combinatorics
2022, arXivOn search complexity of discrete logarithm
2021, arXivTotal functions in the polynomial hierarchy
2021, Leibniz International Proceedings in Informatics, LIPIcsEqual-subset-sum faster than the meet-in-the-middle
2019, Leibniz International Proceedings in Informatics, LIPIcs