◉ COOLJAPAN
← All posts
August 30, 2026 • 8 min read

QuantRS2 0.2.1 Released — A Correct Rz Gate, Phase Estimation From Hours to Milliseconds, and Pure-Rust CUDA

QuantRS2 0.2.1 fixes a real correctness bug — Rz/CRZ/ParametricRotationZ now follow the IBM/Qiskit/OpenQASM-3 sign convention, caught by independent symbolic verification — cuts Quantum Phase Estimation from hours to milliseconds, lands real D-Wave chain-break decoding and AWS Braket SigV4 signing, and adds a pure-Rust oxicuda backend.

release quantrs2 quantum-computing rust quantum-gates performance quantum-annealing cuda pure-rust

A sign error in a rotation gate is the kind of bug that doesn’t crash — it just quietly gives you the wrong physics. QuantRS2 0.2.1 fixes one, along with a phase-estimation routine that took hours when it should have taken milliseconds.

Today we released QuantRS2 0.2.1 — a correctness-and-performance release. The headline fix: RotationZ, CRZ, and ParametricRotationZ were using a reversed diagonal and disagreeing with the IBM/Qiskit/OpenQASM-3 standard on every circuit that used them. Alongside it: Quantum Phase Estimation goes from hours to milliseconds, D-Wave chain-break decoding and AWS Braket request signing move from placeholder to real, and a new pure-Rust oxicuda backend lands.

No C. No Fortran. No CUDA toolkit required at build time. QuantRS2 is a comprehensive, modular, pure-Rust quantum computing framework — a Rust-native alternative to Qiskit, Cirq, and PennyLane. SymEngine’s C++ symbolic core stays replaced by the 100%-Rust quantrs2-symengine-pure; OpenBLAS stays replaced by the pure-Rust OxiBLAS backend. Default features remain 100% Pure Rust, and the whole framework still compiles to Linux, macOS, Windows, and WASM as a single static binary.

Why QuantRS2 0.2.1 is a game changer

Two failure modes plague quantum SDKs, and they’re both worse than a crash:

QuantRS2 0.2.1 closes both gaps:

This lands on 1,025,832 total lines of code, 836,324 Rust lines across 2,759 files, with 5,762 tests passing (0 failures, 74 skipped).

Technical Deep Dive

(a) Correctness — core/src/gate/functions.rs, core/src/parametric.rs, core/src/decomposition.rs. The Rz-family sign fix is a two-line diagonal change with a wide blast radius: it also fixes decompose_u_gate and ZYZ single-qubit reconstruction, which had been silently disagreeing between the internal simulator and OpenQASM export. Regression coverage lives in core/src/gate/functions.rs, core/src/parametric.rs, core/src/decomposition.rs, and a new sim/tests/issue_32_rz_convention.rs.

(b) Performance — sim/src/quantum_algorithms/types.rs, core/src/batch/operations.rs, device/src/photonic/cv_gates.rs. EnhancedPhaseEstimation now materializes the system-register operator once and forms U^(2^i) by repeated squaring. A related bug is fixed alongside it: PhaseEstimationResult::precisions was always length 1 regardless of how many eigenvalues were reported — it now has one entry per eigenvalue. Separately, apply_gate_sequence_batch detects fixed (non-parameterized) gates by name and caches their compiled matrices per sequence, so repeated CNOT chains or Hadamard layers compile once instead of once per application. Photonic CV gate optimization now coalesces adjacent same-mode PhaseRotation and Displacement operations.

(c) Real integrations — anneal/src/dwave/functions.rs, anneal/src/braket.rs, anneal/src/csp_compiler.rs, anneal/src/universal_annealing_compiler/, anneal/src/solution_clustering/. Beyond chain-break decoding, SigV4 signing, and CSP linear constraints: a new CostOptimizer does cross-platform cost estimation and cheapest-platform recommendation, a new PerformancePredictor models per-platform performance and confidence from recorded results, and SolutionClusteringAnalyzer gained real k-means clustering with structural feature extraction. solution_clustering/analyzer.rs was split into an analyzer/ module directory (mod.rs + quality.rs) to stay under the workspace’s 2,000-line file policy.

(d) Numerical foundation. The stack rides SciRS2 0.6.5 (up from 0.5.0), PyO3 0.29.0, and wgpu 30.0.0. New this release: oxicuda 0.5.5 — a pure-Rust CUDA replacement (driver/memory/launch/ptx/webgpu features) that loads libcuda.so at runtime, so no CUDA Toolkit is required at build time — plus pollster for synchronous wgpu adapter/device queries.

Getting Started

cargo add quantrs2-core quantrs2-circuit quantrs2-sim

use quantrs2_circuit::builder::Circuit;
use quantrs2_sim::statevector::StateVectorSimulator;

fn main() {
    // Create a circuit with 2 qubits
    let mut circuit = Circuit::<2>::new();

    // Build a Bell state circuit: H(0) followed by CNOT(0, 1)
    circuit.h(0).unwrap()
           .cnot(0, 1).unwrap();

    // Run the circuit on the state vector simulator
    let simulator = StateVectorSimulator::new();
    let result = circuit.run(simulator).unwrap();

    // Print the resulting probabilities
    for (i, prob) in result.probabilities().iter().enumerate() {
        let bits = format!("{:02b}", i);
        println!("|{}⟩: {:.6}", bits, prob);
    }
}

What’s New in 0.2.1

Added

Changed

Fixed

Tips

This is the foundation

QuantRS2 0.2.1 builds on SciRS2 0.6.5 (arrays, linalg, FFT-via-OxiFFT, optimize), the pure-Rust OxiBLAS backend, the new pure-Rust oxicuda GPU backend, NumRS2 0.4.1 and PandRS 0.4.1 for data, and Oxicode 0.2.6 plus OxiARC (oxiarc-deflate / oxiarc-lz4) for pure-Rust serialization and compression. The result stays an end-to-end, C/C++/Fortran-free quantum stack — from circuit construction through simulation, annealing, and hardware integration.

Repository: https://github.com/cool-japan/quantrs

Star the repo if a bug-for-bug-correct, pure-Rust path to quantum computing is something you want to see grow — and file an issue if you spot another sign convention we got backwards.

The era of silent quantum-simulation bugs is over. Pure Rust quantum computing is here — fast, safe, and sovereign.

— KitaSan at COOLJAPAN OÜ August 30, 2026

↑ Back to all posts