Formally Verifying Noir Zero Knowledge Programs with NAVe
Pedro Antonino, Namrata Jain · 2026
Read Formally Verifying Noir Zero Knowledge Programs with NAVe by Pedro Antonino, Namrata Jain (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-09372.txt.
Introduction
Pedro Antonino and Namrata Jain work on checking zero-knowledge programs for errors before they run. Their 2026 arXiv paper presents NAVe, a tool for formally verifying programs written in Noir. Zero-knowledge proofs let one side show that a claim holds without revealing the private input behind it.
Behind each proof sits a complex arithmetic circuit, a set of polynomial equations over a finite field. A small flaw in that circuit can break the proof's promise. NAVe aims to verify that the circuit matches the intended relation between public and private values. Such checks matter as zero-knowledge systems move into money and identity uses.
Catalog record
| CREATOR | Pedro Antonino, Namrata Jain |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 53 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-09372
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Formally Verifying Noir Zero Knowledge Programs with NAVe Pedro Antonino, Namrata Jain Abstract Zero-Knowledge (ZK) proof systems are cryptographic protocols that can (with overwhelming probability) demonstrate that the pair ( X , W ) (X,W) is in a relation R R without revealing information about the private input W W . This membership checking is captured by a complex arithmetic circuit: a set of polynomial equations over a finite field. ZK programming languages, like Noir, have been proposed to simplify the description of these circuits. A developer can write a Noir program using traditional high-level constructs that can be compiled into a lower-level ACIR (Abstract Circuit Intermediate Representation), which is essentially a high-level description of an arithmetic circuit. In this paper, we formalise some of the ACIR language using SMT-LIB and its extended theory of finite fields. We use this formalisation to create an open-source formal verifier for the Noir language using the SMT solver cvc5. Our verifier can be used to check whether Noir programs behave appropriately. For instance, it can be used to check whether a Noir program has been properly constrained, that is, the finite-field polynomial equations generated truly capture the intended relation. We evaluate our verifier over 4 distinct sets of Noir programs, demonstrating its practical applicability and identifying a hard-to-check constraint type that charts an improvement path for our verification framework. 1 Introduction Zero Knowledge (ZK) proof systems were initially introduced as interactive protocols [ 12 , 13 , 11 ] where a prover \mathcal{P} convinces a verifier \mathcal{V} of a statement of the form R(X,W) where R is a relation (known to \mathcal{P} and \mathcal{V} ), X is an input common to both \mathcal{P} and \mathcal{V} , and W is an input private to the prover (commonly referred to as a witness ); the protocol’s execution must reveal nothing about W apart from the fact that R(X,W) holds. These proof systems have been made non-interactive [ 5 , 14 ] and (their generated proof) succinct [ 10 , 28 , 15 , 20 ] . ZK proof systems have gained renewed interest with the advent of blockchains and the growth of the Web3 ecosystem. They have been promoted as a technology to deliver privacy [ 2 , 31 ] and, even, improve the efficiency of blockchains [ 29 , 19 ] . In these proof systems, the relation R is typically captured by an arithmetic circuit: a set of constraints (i.e. polynomial equations) over a finite field that capture the relationship between X and W . If one is not familiar with such systems, describing these circuits can be challenging. To simplify the use of these systems and increase their adoption, Domain-Specific Languages (DSLs) targeting arithmetic circuits have been developed. These ZK programming languages provide high-level constructs and an infrastructure to compile these constructs into arithmetic circuits used by ZK proof systems; Noir [ 22 ] , Cairo [ 6 ] , and (in a lower level of abstraction) Circom [ 8 ] are examples of such languages. Simply put, they can be used to create a high-level program P that captures R : if R(X,W) holds, the program P successfully runs with parameters (X,W) . The Noir language [ 22 ] was proposed by the Aztec community [ 2 ] . It can be used to both create independent ZK programs and private smart contracts for the Aztec network; smart contracts are a type of program that manages digital assets sometimes worth large amounts of money. Noir programs are compiled into an intermediate language called ACIR (Abstract Circuit Intermediate Representation), which is then used by a ZK proof system. This intermediate representation has opcodes capturing finite-field polynomial equations, but it also has higher-level opcodes, such as range check — a constraint that ensures that a finite-field value can be represented with a determined number of bits — and brillig call — an opcode that allows the execution of unconstrained code . Unconstrained code is a feature of these languages designed to help the prover produce a proof, but it does not contribute to the underlying constraints capturing the relation R . The programmer is expected to include separately the appropriate matching constraints using, for instance, the assert command in Noir. This new programming paradigm is prone to several subtle types of bugs [ 23 , 7 ] . For instance, a lack of appropriate constraining in the Noir program, say if it executes unconstrained code that was not appropriately constrained, can lead to a divergence between the program P and the underlying relation it is trying to capture R . This type of bug has already led to serious real-world vulnerabilities; see [ 33 ] , for instance. ZK programs can also suffer from other types of bugs, such as non-determinism — the underlying relation was meant to be a function, but it is not — and even compiler bugs — the constraints generated do not appropriately capture the behaviour of the high-level programming constructs. The subtlety of these bugs and the increased application of ZK programs in high-stakes contexts — protecting users’ private information or managing large sums of digital assets — make them a perfect target for formal verification. In this paper, we formalise a subset of ACIR using SMT-LIB [ 4 ] and propose NAVe: the Noir Acir (formal) Verifier; not to be confused with the underlying ZK proof system verifier. We encode the semantics of a subset of the ACIR opcodes into SMT-LIB. We propose two encodings: one based on the standard theory of integer arithmetic and another based on a non-standard theory of finite fields [ 25 , 26 ] that is implemented in the cvc5 solver [ 3 ] . We use this formalisation(s) and the convention of capturing verification goals/proof obligations using the Noir function verify_assert ( b : bool ) to create our open-source formal verifier NAVe. Our verifier focuses on identifying unconstrained bugs: a Noir developer can use verify_assert call at the program’s exit point to specify the desired relation R it is expected to capture. We have integrated our verifier into Nargo, Noir’s package manager, to ease developers into using it. We have evaluated our verifier in some test programs that are part of the Noir language infrastructure to assess its efficiency in practice. We analyse its performance using both of our encodings with encouraging results. This evaluation suggests that: (a) some types of constraints, induced by opcodes like range checks, can significantly increase verification time and that (b) our encodings are complementary in the sense that some programs are more quickly analysed using one and some others using the other. So, our evaluation charts a path for future work: we need to propose abstractions/approximations to address (a) and some form of analysis of a Noir program with both encodings concurrently to benefit from (b). There has been some work on the automated formal verification of ZK programs [ 24 , 27 , 32 ] , none of which targets the Noir language and its particularities. Noir has high-level constructs with complex semantics. For instance, unlike other languages, Noir/ACIR supports high-level memory operations, unconstrained functions, and bit-level operations like range checks, bitwise operators,