QuantumMatter.Infrastructure 1.8.0

QuantumMatter SDK

Version 1.8 includes the v1.7 work and extends the separate repository PQC lab with authenticated-session and adversarial validation. These remain educational tooling rather than an SDK security protocol. See the security lab guide and release record.

QuantumMatter is an MIT-licensed quantum simulator built from scratch in C# using double-precision System.Numerics.Complex. Libraries target .NET 9; package-only consumer checks also run on .NET 10. Starting with v1.1, the full xUnit suite runs on both runtimes using the same net9.0 library assets.

The v1.7 work, included in v1.8, adds repository-only educational cryptography experiments and a separate .NET 10 PQC provider lab. No cryptography API is added to these SDK packages. See the experiment guide and development record. The opt-in CPU acceleration controls are documented in the v1.8 API additions.

Install Application and Infrastructure at the same version; Core is transitive.

dotnet add package QuantumMatter.Application
dotnet add package QuantumMatter.Infrastructure

These commands select the latest stable version. For a local/prerelease candidate, specify its version and feed. See the repository release record for publication status and validation evidence.

using QuantumMatter.Application;
using QuantumMatter.Core;
using QuantumMatter.Infrastructure;

var circuit = new QuantumCircuit(2).H(0).CNot(0, 1);
var simulator = new StateVectorSimulator(new CpuStateVectorBackend());
var result = simulator.Run(circuit, shots: 1000, seed: 42);
foreach (var (bits, count) in result.Counts)
    Console.WriteLine($"{bits}: {count}");

Qubit zero is the least significant bit; printed kets show the highest bit first. Angles are radians, gates act on column vectors in insertion order, and phases are retained. Memory budgets apply before allocating state/estimator arrays.

CPU execution needs no GPU driver. CUDA is explicitly selected with CudaStateVectorBackend and requires a compatible NVIDIA driver/device; dispose the backend after use. v1.0 CUDA support is validated on Windows x64; Linux CUDA is experimental. SIMD is opt-in. Noise uses a separate trajectory API and CPU channel kernels; it is not represented by one final ensemble ket.

ParallelCpuOptions.EnableAvx512 and EnableGateBlocking are optional CPU tuning controls, disabled by default. AVX-512 requires EnableSimd and falls back on unsupported hardware; blocking reuses compatible gate tiles with bounded workspace. See the measured AVX-512/CUDA comparison before choosing a worker count or assuming that GPU execution is faster.

Version 1.4 adds a separate NoisyDynamicCircuit and runner combining those channels with measurement, classical feedback and reset. See the noisy dynamic guide and release validation and API additions. Existing circuit contracts are retained.

Version 1.5 adds Pauli Hamiltonians, parameterized Rx/Ry/Rz templates and pure-state energies/gradients. Shared parameters use occurrence-wise shifts; reductions remain on the host after unitary backend execution. See the variational guide, API additions and release validation.

Version 1.6 adds deterministic pure-state VQE with bounded backtracking, explicit execution/update limits, termination reasons and immutable results. See the VQE guide, API additions and release validation.

Version 1.2 adds partial-register measurement, normalized marginals and Pauli-string expectations on host states; see docs/observations.md. These preserve the existing unitary circuit and terminal sampling contracts.

Version 1.3 adds a separate dynamic-circuit workflow with mid-circuit measurements, classical conditions and reset; see docs/dynamic-circuits.md.

The package contains XML IntelliSense documentation, declared public API files, LICENSE, CHANGELOG.md, third-party notices, and the docs directory. Start with docs/api/1.3.0.md, docs/compatibility.md, docs/noise.md and docs/gpu-execution.md. See the repository for anonymized validation evidence, raw benchmarks, known limitations and the release checklist and repository.