TOPICS IN COMPUTATIONAL NUMBER THEORY AND CRYPTANALYSIS - On Simultaneous Chinese Remaindering, Primes, the MiNTRU Assumption, and Functional Encryption

This thesis reports on four independent projects that lie in the intersection of mathematics, computer science, and cryptology:

Simultaneous Chinese Remaindering:

The classical Chinese Remainder Problem asks to find all integer solutions to a given system of congruences where each congruence is defined by one modulus and one remainder. The Simultaneous Chinese Remainder Problem is a direct generalization of its classical counterpart where for each modulus the single remainder is replaced by a non-empty set of remainders. The solutions of a Simultaneous Chinese Remainder Problem instance are completely defined by a set of minimal positive solutions, called primitive solutions, which are upper bounded by the lowest common multiple of the considered moduli. However, contrary to its classical counterpart, which has at most one primitive solution, the Simultaneous Chinese Remainder Problem may have an exponential number of primitive solutions, so that any general-purpose solving algorithm requires exponential time. Furthermore, through a direct reduction from the 3-SAT problem, we prove first that deciding whether a solution exists is NP-complete, and second that if the existence of solutions is guaranteed, then deciding whether a solution of a particular size exists is also NP-complete. Despite these discouraging results, we studied methods to find the minimal solution to Simultaneous Chinese Remainder Problem instances and we discovered some interesting statistical properties.

A Conjecture On Primes In Arithmetic Progressions And Geometric Intervals:

Dirichlet’s theorem on primes in arithmetic progressions states that for any positive integer q and any coprime integer a, there are infinitely many primes in the arithmetic progression a + nq (n ∈ N), however, it does not indicate where those primes can be found. Linnik’s theorem predicts that the first such prime p0 can be found in the interval [0;q^L] where L denotes an absolute and explicitly computable constant. Albeit only L = 5 has been proven, it is widely believed that L ≤ 2. We generalize Linnik’s theorem by conjecturing that for any integers q ≥ 2, 1 ≤ a ≤ q − 1 with gcd(q, a) = 1, and t ≥ 1, there exists a prime p such that p ∈ [q^t;q^(t+1)] and p ≡ a mod q. Subsequently, we prove the conjecture for all sufficiently large exponent t, we computationally verify it for all sufficiently small modulus q, and we investigate its relation to other mathematical results such as Carmichael’s totient function conjecture.

On The (M)iNTRU Assumption Over Finite Rings:

The inhomogeneous NTRU (iNTRU) assumption is a recent computational hardness assumption, which claims that first adding a random low norm error vector to a known gadget vector and then multiplying the result with a secret vector is sufficient to obfuscate the considered secret vector. The matrix inhomogeneous NTRU (MiNTRU) assumption essentially replaces vectors with matrices. Albeit those assumptions strongly remind the well-known learning-with-errors (LWE) assumption, their hardness has not been studied in full detail yet. We provide an elementary analysis of the corresponding decision assumptions and break them in their basis case using an elementary q-ary lattice reduction attack. Concretely, we restrict our study to vectors over finite integer rings, which leads to a problem that we call (M)iNTRU. Starting from a challenge vector, we construct a particular q-ary lattice that contains an unusually short vector whenever the challenge vector follows the (M)iNTRU distribution. Thereby, elementary lattice reduction allows us to distinguish a random challenge vector from a synthetically constructed one.

A Conditional Attack Against Functional Encryption Schemes:

Functional encryption emerged as an ambitious cryptographic paradigm supporting function evaluations over encrypted data revealing the result in plain. Therein, the result consists either in a valid output or a special error symbol. We develop a conditional selective chosen-plaintext attack against the indistinguishability security notion of functional encryption. Intuitively, indistinguishability in the public-key setting is based on the premise that no adversary can distinguish between the encryptions of two known plaintext messages. As functional encryption allows us to evaluate functions over encrypted messages, the adversary is restricted to evaluations resulting in the same output only. To ensure consistency with other primitives, the decryption procedure of a functional encryption scheme is allowed to fail and output an error. We observe that an adversary may exploit the special role of these errors to craft challenge messages that can be used to win the indistinguishability game. Indeed, the adversary can choose the messages such that their functional evaluation leads to the common error symbol, but their intermediate computation values differ. A formal decomposition of the underlying functionality into a mathematical function and an error trigger reveals this dichotomy. Finally, we outline the impact of this observation on multiple DDH-based inner-product functional encryption schemes when we restrict them to bounded-norm evaluations only.