Papers, preprints, theses & talks

Publications

Work on making verification-aware languages, mostly Dafny, easier to use: LLM-assisted proofs, debugging failed verifications, and checking whether a specification is strong enough.

2026

  1. ASEConference

    ProofPulse: Interactive Proof Coverage Analysis for Dafny

    Álvaro F. Silva, Ruben Martins, Alexandra Mendes

    A three-valued proof-coverage model that shows which specification elements a proof depends on and which are auxiliary or irrelevant. On 252 programs it detects unnecessary preconditions and vacuous proofs well, especially when combined with unsat-core minimization.

  2. ISSREConference

    Improving Debugging in Verification-Aware Languages Through Automated Fault Localization: A Case Study in Dafny

    Álvaro Silva, Isabel Amaral, João Pascoal Faria, Alexandra Mendes

    A comparison of state-based and counterexample-based fault localization for failing Dafny verifications, evaluated on mutated DafnyBench programs. Counterexample-based localization, using structured ranking and aggregation over several traces, performs substantially better.

  3. ASEConference

    DSpec2Test: Specification-Driven Test Generation in Dafny

    Sofia Vieira Pinto, Álvaro F. Silva, João Pascoal Faria, Alexandra Mendes

    Black-box test generation from Dafny specifications. It partitions the spec into equivalence classes in disjunctive normal form, optionally adds boundary-value analysis, and uses Z3 to synthesize inputs and expected outputs. The generated tests kill 93.9% of mutants, more than an implementation-based baseline.

  4. APR @ ASEWorkshop

    DafnyFix: Single-Transformation Automated Repair in Dafny

    Isabel Amaral, Álvaro F. Silva, João F. Ferreira, Alexandra Mendes

    Automated repair of failing Dafny programs through single program transformations.

2025

  1. arXivPreprint

    Inferring Multiple Helper Dafny Assertions with LLMs

    Álvaro Silva, Alexandra Mendes, Ruben Martins

    Introduces DAISY (Dafny Assertion Inference SYstem), which combines LLM predictions with analysis of verifier error messages, and extends DafnyBench with programs whose assertions have been removed. DAISY repairs 63.4% of programs missing one assertion and 31.7% of programs missing several.

  2. SEFMConference

    Can Large Language Models Help Students Prove Software Correctness? An Experimental Study with Dafny

    Carolina Carreira, Álvaro Silva, Alexandre Abreu, Alexandra Mendes

    An experimental study of how students interact with an LLM while solving formal verification exercises in Dafny.

2024

  1. FormaliSEConference

    Leveraging Large Language Models to Boost Dafny's Developers Productivity

    Álvaro Silva, Alexandra Mendes, João F. Ferreira

    A research-idea paper on a Dafny plugin that uses LLMs to suggest lemmas the verifier cannot find on its own and, when a lemma cannot be proved automatically, to propose a calculational proof for it.

  2. M.Sc.Supervision

    Evaluating High-Level SystemC Formal Verification Compared to RTL Verification

    Renato Pedrosa de Jesus · Álvaro Silva as company supervisor (Synopsys)

2021

  1. M.Sc.Thesis

    Using Dimensional Aware Genetic Programming to Find Interpretable Dispatching Rules for the Job Shop Scheduling Problem

    Álvaro Silva · Faculty of Engineering, University of Porto

  2. IJUPTalk

    A Simple Approach to Lamport's State Machine Replication

    Álvaro Silva, Bruno Maurício · IJUP 2021 (14th edition), oral communication