More about HKUST
From Intent to Constraints: The Software Stack for ZK Computation
PhD Qualifying Examination
Title: "From Intent to Constraints: The Software Stack for ZK
Computation"
by
Mr. Zhantong XUE
Abstract:
Zero-knowledge (ZK) proof systems are usually presented through their
cryptographic backends: proving schemes, polynomial commitments, and
protocol-level efficiency. The deployed record locates the risk elsewhere.
Over two thirds of catalogued real-world SNARK vulnerabilities arise not in
the cryptography but in the software that translates high-level intent into
circuits, constraints, and execution traces. This survey studies that
translation layer: how programs and other semantic sources are represented,
lowered, optimized, and checked before they reach a ZK backend. Proving
systems themselves are outside its scope.
The survey organizes the ZK software stack into three frontend layers above a
common backend interface. Layer 1 covers program-text frontends: circuit
DSLs, higher-level ZK languages, and shared compiler infrastructures. Layer 2
covers non-program frontends, whose source is a tensor graph, a relational
query, or an automaton. Layer 3 covers zkVMs, which replace bespoke circuits
with proofs of execution traces over fixed machine models such as RISC-V,
WASM, the EVM, and Cairo. Across these layers it compares source
abstractions, intermediate representations, and lowering pipelines, and
catalogues the cross-cutting optimization methods, from constraint reduction
and lookup-based encodings to quantization and proof-cost-aware compiler
passes, together with the trust each demands.
Throughout, ZK compilation is treated as a correctness and security problem
as much as a performance problem. Every frontend emits two coupled artifacts,
a witness generator and a constraint system, and every gap between them is a
potential failure of soundness, completeness, or zero knowledge. The survey
derives a failure-mode taxonomy from these three guarantees:
under-constrained circuits, over-constrained circuits, and leakage, produced
by recurring causes such as witness--constraint inconsistency, missing range
checks, unsafe hints, and compiler defects. Crossing these modes against the
assurance techniques that claim to catch them (static analysis, SMT-based
verification, refinement types, theorem proving, fuzzing, and metamorphic
testing) yields a coverage matrix that shows where the stack is maturing,
where reliability remains fragile, and which gaps must close before
zero-knowledge computation can be trusted as ordinary software
infrastructure.
Date: Tuesday, 28 July 2026
Time: 3:00pm - 5:00pm
Venue: Room 3494
Lift 25/26
Committee Members: Dr. Shuai Wang (Supervisor)
Dr. Dongdong She (Chairperson)
Dr. Lionel Parreaux