arXiv subjects

Viktor Kuncak

Publications and source records attributed to Viktor Kuncak.

At least 19 recordsLinked to original sources

arXiv2026

LeanFlow: A Case Study in Workflow-Driven Lean Autoformalization

We present and evaluate LeanFlow, an LLM agent system specialized for translating mathematical papers into buildable Lean projects. Recent verifier-in-the-loop systems show that large formal artifacts can be produced, but it remains unclear which runtime mechanisms affect completion, auditability, or efficiency in document-to-project formalization. We study this question through case studies on two previously unformalized mathematical papers in number theory and measure theory, using model, proof-workflow, and toolset ablations with Kimi2.6 and GPT5.5; we report task outcome, API calls, input tokens, and output tokens. With Kimi2.6, the full workflow completes both document-level projects within the 2000-call budget, while no-queue variants reach the budget limit; with GPT5.5, all document-level variants complete, and the full workflow has the lowest or tied-lowest input-token cost on both sources. As complementary calibration, LeanFlow reaches 75.7% BEq+ on the PFR slice of RLM25 and solves all five ICML 2026 AI for Math TCS challenge projects in our GPT5.5 runs.

cs.AI↗

arXiv2024

Could ChatGPT get an Engineering Degree? Evaluating Higher Education Vulnerability to AI Assistants

AI assistants are being increasingly used by students enrolled in higher education institutions. While these tools provide opportunities for improved teaching and education, they also pose significant challenges for assessment and learning outcomes. We conceptualize these challenges through the lens of vulnerability, the potential for university assessments and learning outcomes to be impacted by student use of generative AI. We investigate the potential scale of this vulnerability by measuring the degree to which AI assistants can complete assessment questions in standard university-level STEM courses. Specifically, we compile a novel dataset of textual assessment questions from 50 courses at EPFL and evaluate whether two AI assistants, GPT-3.5 and GPT-4 can adequately answer these questions. We use eight prompting strategies to produce responses and find that GPT-4 answers an average of 65.8% of questions correctly, and can even produce the correct answer across at least one prompting strategy for 85.1% of questions. When grouping courses in our dataset by degree program, these systems already pass non-project assessments of large numbers of core courses in various degree programs, posing risks to higher education accreditation that will be amplified as these models improve. Our results call for revising program-level assessment design in higher education in light of advances in generative AI.

cs.CY↗

arXiv2023

Orthologic with Axioms

We study the proof theory and algorithms for orthologic, a logical system based on ortholattices, which have shown practical relevance in simplification and normalization of verification conditions. Ortholattices weaken Boolean algebras while having polynomial-time equivalence checking that is sound with respect to Boolean algebra semantics. We generalize ortholattice reasoning and obtain an algorithm for proving a larger class of classically valid formulas. As the key result, we analyze a proof system for orthologic augmented with axioms. An important feature of the system is that it limits the number of formulas in a sequent to at most two, which makes the extension with axioms non-trivial. We show a generalized form of cut elimination for this system, which implies a sub-formula property. From there we derive a cubic-time algorithm for provability from axioms, or equivalently, for validity in finitely presented ortholattices. We further show that propositional resolution of width 5 proves all formulas provable in orthologic with axioms. We show that orthologic system subsumes resolution of width 2 and arbitrarily wide unit resolution and is complete for reasoning about generalizations of propositional Horn clauses. Moving beyond ground axioms, we introduce effectively propositional orthologic, presenting its semantics as well as a sound and complete proof system. Our proof system implies the decidability of effectively propositional orthologic, as well as its fixed-parameter tractability for a bounded maximal number of variables in each axiom. As a special case, we obtain a generalization of Datalog with negation and disjunction.

cs.LO↗

arXiv2017

On Repair with Probabilistic Attribute Grammars

Program synthesis and repair have emerged as an exciting area of research, driven by the potential for revolutionary advances in programmer productivity. Among most promising ideas emerging for synthesis are syntax-driven search, probabilistic models of code, and the use of input-output examples. Our paper shows how to combine these techniques and use them for program repair, which is among the most relevant applications of synthesis to general-purpose code. Our approach combines semantic specifications, in the form of pre- and post-conditions and input-output examples with syntactic specifications in the form of term grammars and AST-level statistics extracted from code corpora. We show that synthesis in this framework can be viewed as an instance of graph search, permitting the use of well-understood families of techniques such as A*. We implement our algorithm in a framework for verification, synthesis and repair of functional programs, demonstrating that our approach can repair programs that are beyond the reach of previous tools.

cs.PL↗

arXiv2017

Polynomial-Time Proactive Synthesis of Tree-to-String Functions from Examples

Synthesis from examples enables non-expert users to generate programs by specifying examples of their behavior. A domain-specific form of such synthesis has been recently deployed in a widely used spreadsheet software product. In this paper we contribute to foundations of such techniques and present a complete algorithm for synthesis of a class of recursive functions defined by structural recursion over a given algebraic data type definition. The functions we consider map an algebraic data type to a string; they are useful for, e.g., pretty printing and serialization of programs and data. We formalize our problem as learning deterministic sequential top-down tree-to-string transducers with a single state. The first problem we consider is learning a tree-to-string transducer from any set of input/output examples provided by the user. We show that this problem is NP-complete in general, but can be solved in polynomial time under a (practically useful) closure condition that each subtree of a tree in the input/output example set is also part of the input/output examples. Because coming up with relevant input/output examples may be difficult for the user while creating hard constraint problems for the synthesizer, we also study a more automated active learning scenario in which the algorithm chooses the inputs for which the user provides the outputs. Our algorithm asks a worst-case linear number of queries as a function of the size of the algebraic data type definition to determine a unique transducer.

cs.FL↗

arXiv2016

An Update on Deductive Synthesis and Repair in the Leon Tool

We report our progress in scaling deductive synthesis and repair of recursive functional Scala programs in the Leon tool. We describe new techniques, including a more precise mechanism for encoding the space of meaningful candidate programs. Our techniques increase the scope of synthesis by expanding the space of programs we can synthesize and by reducing the synthesis time in many cases. As a new example, we present a run-length encoding function for a list of values, which Leon can now automatically synthesize from specification consisting of the decoding function and the local minimality property of the encoded value.

cs.LO↗

arXiv2016

Translating Scala Programs to Isabelle/HOL

We present a trustworthy connection between the Leon verification system and the Isabelle proof assistant. Leon is a system for verifying functional Scala programs. It uses a variety of automated theorem provers (ATPs) to check verification conditions (VCs) stemming from the input program. Isabelle, on the other hand, is an interactive theorem prover used to verify mathematical specifications using its own input language Isabelle/Isar. Users specify (inductive) definitions and write proofs about them manually, albeit with the help of semi-automated tactics. The integration of these two systems allows us to exploit Isabelle's rich standard library and give greater confidence guarantees in the correctness of analysed programs.

cs.LO↗

arXiv2016

Towards a Compiler for Reals

Numerical software, common in scientific computing or embedded systems, inevitably uses an approximation of the real arithmetic in which most algorithms are designed. In many domains, roundoff errors are not the only source of inaccuracy and measurement and other input errors further increase the uncertainty of the computed results. Adequate tools are needed to help users select suitable approximations, especially for safety-critical applications. We present the source-to-source compiler Rosa which takes as input a real-valued program with error specifications and synthesizes code over an appropriate floating-point or fixed-point data type. The main challenge of such a compiler is a fully automated, sound and yet accurate enough numerical error estimation. We present a unified technique for floating-point and fixed-point arithmetic of various precisions which can handle nonlinear arithmetic, determine closed- form symbolic invariants for unbounded loops and quantify the effects of discontinuities on numerical errors. We evaluate Rosa on a number of benchmarks from scientific computing and embedded systems and, comparing it to state-of-the-art in automated error estimation, show it presents an interesting trade-off between accuracy and performance.

cs.PL↗

arXiv2016

An Instantiation-Based Approach for Solving Quantified Linear Arithmetic

This paper presents a framework to derive instantiation-based decision procedures for satisfiability of quantified formulas in first-order theories, including its correctness, implementation, and evaluation. Using this framework we derive decision procedures for linear real arithmetic (LRA) and linear integer arithmetic (LIA) formulas with one quantifier alternation. Our procedure can be integrated into the solving architecture used by typical SMT solvers. Experimental results on standardized benchmarks from model checking, static analysis, and synthesis show that our implementation of the procedure in the SMT solver CVC4 outperforms existing tools for quantified linear arithmetic.

cs.LO↗

arXiv2016

Proceedings Fourth Workshop on Synthesis

The SYNT workshop aims to bring together researchers interested in the broad area of synthesis of computing systems. The goal is to foster the development of frontier techniques in automating the development of computing system. Contributions of interest include algorithms, complexity and decidability analysis, as well as reproducible heuristics, implemented tools, and experimental evaluation. Application domains include software, hardware, embedded, and cyberphysical systems. Computation models include functional, reactive, hybrid and timed systems. Identifying, formalizing, and evaluating synthesis in particular application domains is encouraged. The fourth iteration of the workshop took place in San Francisco, CA, USA. It was co-located with the 27th International Conference on Computer Aided Verification. The workshop included five contributed talks and two invited talks. In addition, it featured a special session about the Syntax-Guided Synthesis Competition (SyGuS) and the SyntComp Synthesis competition.

cs.PL↗

arXiv2015

On Counterexample Guided Quantifier Instantiation for Synthesis in CVC4

We introduce the first program synthesis engine implemented inside an SMT solver. We present an approach that extracts solution functions from unsatisfiability proofs of the negated form of synthesis conjectures. We also discuss novel counterexample-guided techniques for quantifier instantiation that we use to make finding such proofs practically feasible. A particularly important class of specifications are single-invocation properties, for which we present a dedicated algorithm. To support syntax restrictions on generated solutions, our approach can transform a solution found without restrictions into the desired syntactic form. As an alternative, we show how to use evaluation function axioms to embed syntactic restrictions into constraints over algebraic datatypes, and then use an algebraic datatype decision procedure to drive synthesis. Our experimental evaluation on syntax-guided synthesis benchmarks shows that our implementation in the CVC4 SMT solver is competitive with state-of-the-art tools for synthesis.

cs.LO↗

arXiv2013

On Sound Compilation of Reals

Writing accurate numerical software is hard because of many sources of unavoidable uncertainties, including finite numerical precision of implementations. We present a programming model where the user writes a program in a real-valued implementation and specification language that explicitly includes different types of uncertainties. We then present a compilation algorithm that generates a conventional implementation that is guaranteed to meet the desired precision with respect to real numbers. Our verification step generates verification conditions that treat different uncertainties in a unified way and encode reasoning about floating-point roundoff errors into reasoning about real numbers. Such verification conditions can be used as a standardized format for verifying the precision and the correctness of numerical programs. Due to their often non-linear nature, precise reasoning about such verification conditions remains difficult. We show that current state-of-the art SMT solvers do not scale well to solving such verification conditions. We propose a new procedure that combines exact SMT solving over reals with approximate and sound affine and interval arithmetic. We show that this approach overcomes scalability limitations of SMT solvers while providing improved precision over affine and interval arithmetic. Using our initial implementation we show the usefullness and effectiveness of our approach on several examples, including those containing non-linear computation.

cs.PL↗

arXiv2013

On Integrating Deductive Synthesis and Verification Systems

We describe techniques for synthesis and verification of recursive functional programs over unbounded domains. Our techniques build on top of an algorithm for satisfiability modulo recursive functions, a framework for deductive synthesis, and complete synthesis procedures for algebraic data types. We present new counterexample-guided algorithms for constructing verified programs. We have implemented these algorithms in an integrated environment for interactive verification and synthesis from relational specifications. Our system was able to synthesize a number of useful recursive functions that manipulate unbounded numbers and data structures.

cs.PL↗

arXiv2013

The Relationship between Craig Interpolation and Recursion-Free Horn Clauses

Despite decades of research, there are still a number of concepts commonly found in software programs that are considered challenging for verification: among others, such concepts include concurrency, and the compositional analysis of programs with procedures. As a promising direction to overcome such difficulties, recently the use of Horn constraints as intermediate representation of software programs has been proposed. Horn constraints are related to Craig interpolation, which is one of the main techniques used to construct and refine abstractions in verification, and to synthesise inductive loop invariants. We give a survey of the different forms of Craig interpolation found in literature, and show that all of them correspond to natural fragments of (recursion-free) Horn constraints. We also discuss techniques for solving systems of recursion-free Horn constraints.

cs.LO↗

arXiv2013

Disjunctive Interpolants for Horn-Clause Verification (Extended Technical Report)

One of the main challenges in software verification is efficient and precise compositional analysis of programs with procedures and loops. Interpolation methods remain one of the most promising techniques for such verification, and are closely related to solving Horn clause constraints. We introduce a new notion of interpolation, disjunctive interpolation, which solve a more general class of problems in one step compared to previous notions of interpolants, such as tree interpolants or inductive sequences of interpolants. We present algorithms and complexity for construction of disjunctive interpolants, as well as their use within an abstraction-refinement loop. We have implemented Horn clause verification algorithms that use disjunctive interpolants and evaluate them on benchmarks expressed as Horn clauses over the theory of integer linear arithmetic.

cs.LO↗

arXiv2012

Software Verification and Graph Similarity for Automated Evaluation of Students' Assignments

In this paper we promote introducing software verification and control flow graph similarity measurement in automated evaluation of students' programs. We present a new grading framework that merges results obtained by combination of these two approaches with results obtained by automated testing, leading to improved quality and precision of automated grading. These two approaches are also useful in providing a comprehensible feedback that can help students to improve the quality of their programs We also present our corresponding tools that are publicly available and open source. The tools are based on LLVM low-level intermediate code representation, so they could be applied to a number of programming languages. Experimental evaluation of the proposed grading framework is performed on a corpus of university students' programs written in programming language C. Results of the experiments show that automatically generated grades are highly correlated with manually determined grades suggesting that the presented tools can find real-world applications in studying and grading.

cs.AI↗

arXiv2006

On Verifying Complex Properties using Symbolic Shape Analysis

One of the main challenges in the verification of software systems is the analysis of unbounded data structures with dynamic memory allocation, such as linked data structures and arrays. We describe Bohne, a new analysis for verifying data structures. Bohne verifies data structure operations and shows that 1) the operations preserve data structure invariants and 2) the operations satisfy their specifications expressed in terms of changes to the set of objects stored in the data structure. During the analysis, Bohne infers loop invariants in the form of disjunctions of universally quantified Boolean combinations of formulas. To synthesize loop invariants of this form, Bohne uses a combination of decision procedures for Monadic Second-Order Logic over trees, SMT-LIB decision procedures (currently CVC Lite), and an automated reasoner within the Isabelle interactive theorem prover. This architecture shows that synthesized loop invariants can serve as a useful communication mechanism between different decision procedures. Using Bohne, we have verified operations on data structures such as linked lists with iterators and back pointers, trees with and without parent pointers, two-level skip lists, array data structures, and sorted lists. We have deployed Bohne in the Hob and Jahob data structure analysis systems, enabling us to combine Bohne with analyses of data structure clients and apply it in the context of larger programs. This report describes the Bohne algorithm as well as techniques that Bohne uses to reduce the ammount of annotations and the running time of the analysis.

cs.PL↗

arXiv2005

On Algorithms and Complexity for Sets with Cardinality Constraints

Typestate systems ensure many desirable properties of imperative programs, including initialization of object fields and correct use of stateful library interfaces. Abstract sets with cardinality constraints naturally generalize typestate properties: relationships between the typestates of objects can be expressed as subset and disjointness relations on sets, and elements of sets can be represented as sets of cardinality one. Motivated by these applications, this paper presents new algorithms and new complexity results for constraints on sets and their cardinalities. We study several classes of constraints and demonstrate a trade-off between their expressive power and their complexity. Our first result concerns a quantifier-free fragment of Boolean Algebra with Presburger Arithmetic. We give a nondeterministic polynomial-time algorithm for reducing the satisfiability of sets with symbolic cardinalities to constraints on constant cardinalities, and give a polynomial-space algorithm for the resulting problem. In a quest for more efficient fragments, we identify several subclasses of sets with cardinality constraints whose satisfiability is NP-hard. Finally, we identify a class of constraints that has polynomial-time satisfiability and entailment problems and can serve as a foundation for efficient program analysis.

cs.PL↗