A high-performance quantum simulation toolkit Statevector/MPS engines with compilation, noise, VQE, QEC, chemistry, and agent-native tooling.
Table of Contents
▍ What It Is
Run up to 28 qubits in about 3 seconds, without crashing. Dense Evolution JIT-compiles statevector circuits through JAX XLA, automatically chunks and — past even that RAM ceiling — spills to disk when memory fills up, so a real simulation stays alive instead of OOM-ing.
📖 Full documentation, API reference, and worked examples →
A local Streamlit dashboard and web-based Composer editor are also included — see Composer & MCP Server below.
▍ Install
pip install dense-evolution # JAX is a core dependency, installed by default
# full stack: GPU · dashboard · Qiskit/PennyLane interop
pip install dense-evolution[full]
# just the interop bridge
pip install dense-evolution[qiskit]
pip install dense-evolution[pennylane]
# Composer's local kernel (see "Composer" below)
pip install dense-evolution[composer]
# MCP server for the Composer kernel (see "MCP Server" below)
pip install dense-evolution[mcp]
# development
git clone https://github.com/tatopenn-cell/Dense-Evolution.git
cd Dense-Evolution && pip install -e .[full]
⚠️ macOS + dense-evolution[qiskit] users
Qiskit's own QuantumCircuit.__init__ is known to segfault the whole process on macOS/arm64 (an upstream Qiskit bug, not something Dense-Evolution can fix from its side — see release v8.1.43 for the full reproduction). dense_evolution/interop.py now warns (RuntimeWarning, once per process) the first time you touch the Qiskit bridge on sys.platform == 'darwin', but it does not block — some Qiskit/macOS combinations may work fine. If you hit a crash, pip install dense-evolution[pennylane] gives the same circuit-interop functionality without constructing any Qiskit object.
Google Colab (3 lines):
!git clone https://github.com/tatopenn-cell/Dense-Evolution.git
%cd Dense-Evolution
!pip install -e .
▍ Quick Start
from dense_evolution import DenseSVSimulator, QASMParser
# parse any OpenQASM 2.0 / 3.0 string -- single-qubit rotations, a barrier
# (a real OpenQASM synchronization marker: parsed like hardware would, no
# effect on the simulated state), then an entangling layer
qasm = """
OPENQASM 2.0;
include "qelib1.inc";
qreg q[3];
rx(pi/3) q[0];
ry(pi/4) q[1];
h q[2];
barrier q;
cx q[0], q[1];
cx q[1], q[2];
rz(pi/6) q[2];
"""
parser = QASMParser()
circuit = parser.parse(qasm)
sim = DenseSVSimulator(n_qubits=3)
sim.run_circuit_jit(circuit.to_tuples())
probs = sim.get_probabilities()
sv = sim.get_statevector()
# probs = [0.3201 0.3201 0.0549 0.0549 0.0183 0.0183 0.1067 0.1067]
Noise:
import numpy as np
from dense_evolution import DenseSVSimulator, QASMParser, NoiseModel
qasm = 'OPENQASM 2.0; include "qelib1.inc"; qreg q[2]; h q[0]; cx q[0],q[1];'
circuit = QASMParser().parse(qasm)
sim = DenseSVSimulator(n_qubits=2)
sim.run_circuit_jit(circuit.to_tuples())
noisy_sv = NoiseModel.apply_to_sv(np.asarray(sim.sv), n=2, model='depolarizing', p=0.05, rng=np.random.default_rng(0))
np.abs(noisy_sv) ** 2
# [0. 0.5 0.5 0. ]
VQE:
import jax
import jax.numpy as jnp
from dense_evolution import QASMParser, circuit_to_energy_fn
qasm = 'OPENQASM 2.0; include "qelib1.inc"; qreg q[1]; ry(0.0) q[0];'
circuit = QASMParser().parse(qasm)
energy_fn, n_params = circuit_to_energy_fn(circuit, n_qubits=1)
h = jnp.array([[1.0, 0.0], [0.0, -1.0]], dtype=jnp.complex128) # Pauli Z
theta = jnp.array([0.1])
grad_fn = jax.value_and_grad(energy_fn, argnums=0, has_aux=True)
for _ in range(40):
(energy, sv), grad = grad_fn(theta, h)
theta = theta - 0.5 * grad
# energy = -1.0, theta = [3.14159265]
Mitigation (ZNE):
import dense_evolution as de
e1, e2, e3 = 1.234, 0.876, 0.611 # values at 1x, 2x, 3x noise
de.zero_noise_extrapolation([e1, e2, e3], [1.0, 2.0, 3.0])
# 1.622
Dashboard (local, Streamlit):
pip install "dense-evolution[dashboard]" # JAX already included by default
streamlit run tools/dashboard/app.py
Anti-OOM for large circuits:
from dense_evolution import Chunk, QASMParser
qasm = 'OPENQASM 2.0; include "qelib1.inc"; qreg q[27]; h q[0:27];'
circuit = QASMParser().parse(qasm)
sim = Chunk(27)
sim.run_chunk(circuit.to_tuples(), chunk_size_gates=500)
▍ Key Features
- JIT-fused statevector engine. No Kronecker-product overhead — stride-sliced linear kernel fusion compiled through JAX XLA, real GPU/TPU dispatch with no code change. Simulator · MPS backend for low-entanglement circuits at scale.
- Anti-OOM
Chunkengine. Circuits too large for one array, split dynamically and sized off the real compute device's own free memory — with disk-backed overflow past even that ceiling. Chunk guide. - Real noise, real mitigation. Stochastic Kraus channels, real-device noise imported from Qiskit backends, and Zero-Noise Extrapolation to correct for it. Noise · Mitigation · what noise/mitigation/healing each mean.
- Differentiable VQE, from scratch.
circuit_to_energy_fnis the same JAX-differentiable engine real molecular VQE runs on — real Hartree-Fock Hamiltonians, UCCSD/hardware-efficient ansätze, Adam optimization. Autodiff. - OpenQASM 2.0/3.0, both directions. A real parser, plus Qiskit/PennyLane interop bridges. QASM Parser · Interop.
- Code-agnostic QEC decoding, Majorana/Jordan-Wigner fermion mapping, from-scratch Hartree-Fock for elements outside PennyLane's own basis set, and a traversable-wormhole-inspired teleportation protocol — see the full API reference for all of it.
▍ Benchmarks
Measured on Windows, CPU only, 8 GB RAM — PennyLane's default.qubit allocates the full statevector; Dense Evolution's Chunk holds constant ~2 GB regardless of qubit count.
| Qubits | Hilbert Space | PennyLane | Dense Evolution | Chunk Geometry |
|---|---|---|---|---|
| 26 | 67,108,864 | ✅ 1,074 MB | ✅ 2,050 MB | 1× (2²⁷) |
| 28 | 268,435,456 | ❌ OOM | ✅ 2,050 MB | 2× (2²⁷) |
| 32 | 4,294,967,296 | ❌ OOM | ✅ 2,048 MB | 32× (2²⁷) |
On Google Colab (12 GB RAM), n=28 runs in ~3s end-to-end (JIT-compiled, num_chunks=2) — the multi-chunk path is fast, not just OOM-safe. Past the real RAM ceiling (Chunk alone raises MemoryPressureError cleanly rather than crashing the process), allow_disk_overflow=True falls back to a slower disk-backed path instead of failing — correctness-first, not benchmarked for speed yet. Full measured table: docs/api/chunk.md.
▍ Composer & MCP Server
A real circuit editor (graphical or OpenQASM) running on your own machine, plus an MCP server exposing 22 tools so an agent (Claude Code, Claude Desktop, ...) can drive it directly. Composer · MCP Server.
▍ Key Resources
▍ Changelog
📜 Full Changelog & Releases — every version, latest first.
▍ License
Business Source License 1.1 — converts automatically to Apache 2.0 on 1 June 2029.
- Non-commercial use: unrestricted
- Commercial use: ≤ 24 allocated qubits · ≤ 1,000 circuits/day · ≤ 10,000 shots/circuit
- Attribution required:
© 2026 Salvatore Pennacchio <jtatopenn@libero.it> — Dense Evolution
Full text: LICENSE.md
▍ Cite This
If Dense-Evolution is useful in academic work, please cite it via the metadata in CITATION.cff (recognized by GitHub's own "Cite this repository" button, and by reference managers that support the Citation File Format).
Archived on Zenodo:
- Concept DOI (always resolves to the latest version): 10.5281/zenodo.21855643
- This release (v8.1.61): 10.5281/zenodo.22009005
1 maintainer
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file dense_evolution-8.1.73.tar.gz.
File metadata
- Download URL: dense_evolution-8.1.73.tar.gz
- Upload date:
- Size: 311.0 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via:
twine/6.2.0 CPython/3.12.10
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
67abeed5211599ead80ac1c01d40de8c77cfd88044d403ebedab2fb6c55e7757
|
|
| MD5 |
af2fee2b6b822d7da5b307fef42e762f
|
|
| BLAKE2b-256 |
0c8dc37a8c820ffcbea0b986bb474b5bc94671f6c8409931615450c488ecd6bf
|
File details
Details for the file dense_evolution-8.1.73-py3-none-any.whl.
File metadata
- Download URL: dense_evolution-8.1.73-py3-none-any.whl
- Upload date:
- Size: 360.4 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via:
twine/6.2.0 CPython/3.12.10
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
ce5723c64eff49f8cbc05d63cdccb8013e4e367ce8aa96719fee98caa63f8e9e
|
|
| MD5 |
a7ad83c0edccd1d1a2ef01c1b449be2e
|
|
| BLAKE2b-256 |
d9479053feff2a878d90bc1996af16b3dff3ca545855358de83d933831352915
|
