We present a generalised, constructive, and machine-checked approach to Kolmogorov complexity in the constructive type theory underlying the Coq proof assistant. By proving that nonrandom numbers form a simple predicate, we obtain elegant proofs of undecidability for random and nonrandom numbers and a proof of uncomputability of Kolmogorov complexity. We use a general and abstract definition of Kolmogorov complexity and subsequently instantiate it to several definitions frequently found in the literature. Whereas textbook treatments of Kolmogorov complexity usually rely heavily on classical logic and the axiom of choice, we put emphasis on the constructiveness of all our arguments, however without blurring their essence. We first give a high-level proof idea using classical logic, which can be formalised with Markov’s principle via folklore techniques we subsequently explain. Lastly, we show a strategy how to eliminate Markov’s principle from a certain class of computability proofs, rendering all our results fully constructive. All our results are machine-checked by the Coq proof assistant, which is enabled by using a synthetic approach to computability: rather than formalising a model of computation, which is well known to introduce a considerable overhead, we abstractly assume a universal function, allowing the proofs to focus on the mathematical essence.

The weak call-by-value λ-calculus Łand Turing machines can simulate each other with a polynomial overhead in time. This time invariance thesis for L, where the number of β-reductions of a computation is taken as its time complexity, is the culmination of a 25-years line of research, combining work by Blelloch, Greiner, Dal Lago, Martini, Accattoli, Forster, Kunze, Roth, and Smolka. The present paper presents a mechanised proof of the time invariance thesis for L, constituting the first mechanised equivalence proof between two standard models of computation covering time complexity. The mechanisation builds on an existing framework for the extraction of Coq functions to L and contributes a novel Hoare logic framework for the verification of Turing machines. The mechanised proof of the time invariance thesis establishes Łas model for future developments of mechanised computational complexity theory regarding time. It can also be seen as a non-trivial but elementary case study of time-complexity-preserving translations between a functional language and a sequential machine model. As a by-product, we obtain a mechanised many-one equivalence proof of the halting problems for Łand Turing machines, which we contribute to the Coq Library of Undecidability Proofs.

We mechanise the Cook-Levin theorem, i.e. the NP-completeness of SAT, in the proof assistant Coq. We use the call-by-value λ-calculus L as the model of computation to formalise time complexity, the class NP, and polynomial-time reductions. The latter two notions agree with the usual characterisations via Turing machines (TMs), as L and TMs are polynomial-time equivalent. The use of L as the computational model, as opposed to TMs, significantly eases program verification and the derivation of resource bounds. However, for showing the NP-hardness of SAT, computations of L need to be encoded in SAT, which is complicated by L’s more complex computational structure. Thus, the polynomial-time reduction chain to SAT employs TMs as an intermediate problem, for which we neatly factor out a known textbook reduction from TMs to SAT. Still, all reduction functions are implemented and analysed in L. To the best of our knowledge, this is the first result in computational complexity theory that has been mechanised with respect to any concrete computational model. We discuss what makes this area of computer science hard to mechanise and highlight the design choices which enable our mechanisations.

We present a framework for the verified programming of multi-tape Turing machines in Coq. Improving on prior work by Asperti and Ricciotti in Matita, we implement multiple layers of abstraction. The highest layer allows a user to implement nontrivial algorithms as Turing machines and verify their correctness, as well as time and space complexity compositionally. The user can do so without ever mentioning states, symbols on tapes or transition functions: They write programs in an imperative language with registers containing values of encodable data types, and our framework constructs corresponding Turing machines.

We study the weak call-by-value $\lambda$-calculus as a model for computational complexity theory and establish the natural measures for time and space -- the number of beta-reductions and the size of the largest term in a computation -- as reasonable measures with respect to the invariance thesis of Slot and van Emde Boas [STOC~84]. More precisely, we show that, using those measures, Turing machines and the weak call-by-value $\lambda$-calculus can simulate each other within a polynomial overhead in time and a constant factor overhead in space for all computations that terminate in (encodings) of 'true' or 'false'. We consider this result as a solution to the long-standing open problem, explicitly posed by Accattoli [ENTCS~18], of whether the natural measures for time and space of the $\lambda$-calculus are reasonable, at least in case of weak call-by-value evaluation. Our proof relies on a hybrid of two simulation strategies of reductions in the weak call-by-value $\lambda$-calculus by Turing machines, both of which are insufficient if taken alone. The first strategy is the most naive one in the sense that a reduction sequence is simulated precisely as given by the reduction rules; in particular, all substitutions are executed immediately. This simulation runs within a constant overhead in space, but the overhead in time might be exponential. The second strategy is heap-based and relies on structure sharing, similar to existing compilers of eager functional languages. This strategy only has a polynomial overhead in time, but the space consumption might require an additional factor of $\log n$, which is essentially due to the size of the pointers required for this strategy. Our main contribution is the construction and verification of a space-aware interleaving of the two strategies, which is shown to yield both a constant overhead in space and a polynomial overhead in time.

We provide a plugin extracting Coq functions of simple polymorphic types to the (untyped) call-by-value lambda calculus L. The plugin is implemented in the MetaCoq framework and entirely written in Coq. We provide Ltac tactics to automatically verify the extracted terms w.r.t a logical relation connecting Coq functions with correct extractions and time bounds, essentially performing a certifying translation and running time validation. We provide three case studies: A universal L-term obtained as extraction from the Coq definition of a step-indexed self-interpreter for L, a many-reduction from solvability of Diophantine equations to the halting problem of L, and a polynomial-time simulation of Turing machines in L.