atlas-quantum 0.5.0

GPU-accelerated quantum tensor network simulator with adaptive MPS, TDVP, VQE/QAOA, and custom Triton kernels

ATLAS-Q: GPU-Accelerated Quantum Tensor Network Simulator

Adaptive Tensor Learning And Simulation – Quantum

Version 0.5.0 | October 2025

High-performance quantum simulation using GPU-accelerated tensor networks with custom Triton kernels

---

⚡ Performance Highlights

• 77K+ ops/sec gate throughput (GPU-optimized)
• 626,000× memory compression vs full statevector (30 qubits)
• 20× speedup on Clifford circuits (Stabilizer backend)
• 1.5-3× speedup on gate operations (custom Triton kernels)
• All 7/7 benchmarks passing

---

🚀 Quick Start

Option 1: Interactive Notebook (No Install!)

Try ATLAS-Q instantly in Google Colab or Jupyter:

📓 Open ATLAS_Q_Demo.ipynb in Colab

Or download and run locally:

wget https://github.com/followthesapper/ATLAS-Q/raw/ATLAS-Q/ATLAS_Q_Demo.ipynb
jupyter notebook ATLAS_Q_Demo.ipynb

---

Option 2: Python Package (Recommended)

# Install from PyPI
pip install atlas-quantum

# With GPU support
pip install atlas-quantum[gpu]

# Verify installation
python -c "from atlas_q import get_quantum_sim; print('✅ ATLAS-Q installed!')"

First example:

from atlas_q import get_quantum_sim

QCH, _, _, _ = get_quantum_sim()
sim = QCH()
factors = sim.factor_number(221)
print(f"221 = {factors[0]} × {factors[1]}")  # 221 = 13 × 17

---

Option 3: Docker

GPU version (recommended):

docker pull ghcr.io/followthesapper/atlas-q:cuda
docker run --rm -it --gpus all ghcr.io/followthesapper/atlas-q:cuda python3

CPU version:

docker pull ghcr.io/followthesapper/atlas-q:cpu
docker run --rm -it ghcr.io/followthesapper/atlas-q:cpu python3

Run benchmarks in Docker:

docker run --rm --gpus all ghcr.io/followthesapper/atlas-q:cuda \
  python3 /opt/atlas-q/scripts/benchmarks/validate_all_features.py

---

Option 4: From Source

# Clone repository
git clone https://github.com/followthsapper/ATLAS-Q.git
cd ATLAS-Q

# Install ATLAS-Q
pip install -e .[gpu]

# Setup GPU acceleration (auto-detects your GPU)
./setup_triton.sh

# Run benchmarks
python scripts/benchmarks/validate_all_features.py

---

GPU Acceleration Setup

The setup_triton.sh script automatically detects your GPU and configures Triton kernels:

• Auto-detects: V100, A100, H100, GB100/GB200, and future architectures
• Configures: TORCH_CUDA_ARCH_LIST and TRITON_PTXAS_PATH
• Persists: Adds settings to ~/.bashrc

Performance gains: 1.5-3× faster gate operations, 100-1000× faster period-finding

---

💡 Examples

Tensor Network Simulation

from atlas_q.adaptive_mps import AdaptiveMPS
import torch

# Create 10-qubit system with adaptive bond dimensions
mps = AdaptiveMPS(10, bond_dim=8, device='cuda')

# Apply Hadamard gates
H = torch.tensor([[1,1],[1,-1]], dtype=torch.complex64)/torch.sqrt(torch.tensor(2.0))
for q in range(10):
    mps.apply_single_qubit_gate(q, H.to('cuda'))

# Apply CNOT gates
CNOT = torch.tensor([[1,0,0,0],[0,1,0,0],[0,0,0,1],[0,0,1,0]],
                     dtype=torch.complex64).reshape(4,4).to('cuda')
for q in range(0, 9, 2):
    mps.apply_two_site_gate(q, CNOT)

print(f"Max bond dimension: {mps.stats_summary()['max_chi']}")
print(f"Memory usage: {mps.memory_usage() / (1024**2):.2f} MB")

Period-Finding & Factorization

from atlas_q import get_quantum_sim

# Get quantum classical hybrid simulator
QuantumClassicalHybrid, _, _, _ = get_quantum_sim()
qc = QuantumClassicalHybrid()

# Factor semiprimes
factors = qc.factor_number(143)  # Returns [11, 13]
print(f"143 = {factors[0]} × {factors[1]}")

# Verified against canonical benchmarks:
# - IBM 2001 (N=15): ✅ Pass
# - Photonic 2012 (N=21): ✅ Pass
# - NMR 2012 (N=143): ✅ Pass

---

📊 Performance vs Competition

Feature	ATLAS-Q	Qiskit Aer	Cirq	Winner
Memory (30q)	0.03 MB	16 GB	16 GB	ATLAS-Q (626k×)
GPU Support	✅ Triton	✅ cuQuantum	❌	ATLAS-Q
Stabilizer	20× speedup	Standard	Standard	ATLAS-Q
Tensor Networks	✅ Native	❌	❌	ATLAS-Q
Ease of Use	Good	Excellent	Excellent	Qiskit/Cirq

Note: Run python scripts/benchmarks/compare_with_competitors.py for detailed performance comparisons

---

🎯 What is ATLAS-Q?

ATLAS-Q is a GPU-accelerated quantum simulator with two complementary capabilities:

Tensor Network Simulation

1. Adaptive MPS: Memory-efficient quantum state representation (O(n·χ²) vs O(2ⁿ))
2. NISQ Algorithms: VQE, QAOA with noise models
3. Time Evolution: TDVP for Hamiltonian dynamics
4. Specialized Backends: Stabilizer for Clifford circuits, MPO for observables
5. GPU Acceleration: Custom Triton kernels + cuBLAS tensor cores

Period-Finding & Factorization

1. Shor's Algorithm: Integer factorization via quantum period-finding
2. Compressed States: Periodic states (O(1) memory), product states (O(n) memory)
3. Verified Results: Matches canonical benchmarks (N=15, 21, 143)

Key Innovations

• ✅ Custom Triton Kernels: Fused gate operations for 1.5-3× speedup
• ✅ Adaptive Bond Dimensions: Dynamic memory management based on entanglement
• ✅ Hybrid Stabilizer/MPS: 20× faster Clifford circuits with automatic switching
• ✅ GPU-Optimized Einsums: cuBLAS + tensor cores for tensor contractions
• ✅ Specialized Representations: O(1) memory for periodic states, O(n) for product states

---

📚 Documentation

Interactive Tutorial

• 📓 Jupyter Notebook - Complete interactive demo (works in Colab!)

Online Documentation

• 📖 Documentation Site - Browse all docs online

Guides & References

• Complete Guide - Installation, tutorials, API reference (start here!)
• Feature Status - What's actually implemented
• Research Paper - Mathematical foundations and algorithms
• Whitepaper - Technical architecture and implementation
• Overview - High-level explanation for all audiences

---

🏗️ Architecture

Core Components

ATLAS-Q/
├── src/atlas_q/
│   ├── adaptive_mps.py             # Adaptive MPS with GPU support
│   ├── quantum_hybrid_system.py   # Period-finding & factorization
│   ├── mpo_ops.py                  # MPO operations (Hamiltonians)
│   ├── tdvp.py                     # Time evolution (TDVP)
│   ├── vqe_qaoa.py                 # Variational algorithms
│   ├── stabilizer_backend.py      # Fast Clifford simulation
│   ├── noise_models.py             # NISQ noise models
│   ├── peps.py                     # 2D tensor networks
│   └── tools_qih/                  # Quantum-inspired ML
├── triton_kernels/
│   ├── mps_complex.py              # Custom Triton kernels (1.5-3× faster)
│   ├── mps_ops.py                  # MPS tensor operations
│   └── modpow.py                   # Modular exponentiation
├── benchmarks/
│   ├── comprehensive_benchmark.py  # 7/7 tensor network benchmarks
│   └── competitive_comparison.py   # vs Qiskit/Cirq/ITensor
└── tests/                          # 75+ unit tests

Technology Stack

• PyTorch 2.10+ (CUDA backend)
• Triton (custom GPU kernels)
• cuBLAS/CUTLASS (tensor cores)
• NumPy/SciPy (linear algebra)

---

🎓 Use Cases

✅ BEST FOR:

• Tensor Networks: 20-50 qubits with moderate entanglement
• VQE/QAOA: Optimization on NISQ devices with noise
• Time Evolution: Hamiltonian dynamics via TDVP
• Period-Finding: Shor's algorithm for integer factorization
• Memory-Constrained: 626,000× compression vs statevector
• GPU Workloads: Custom Triton kernels + cuBLAS

⚠️ NOT IDEAL FOR:

• Highly entangled states (use full statevector)
• Arbitrary connectivity (MPS assumes 1D/2D structure)
• CPU-only environments

---

📈 Benchmark Results

Internal Benchmarks (All Passing)

✅ Benchmark 1: Noise Models          - 3/3 passing
✅ Benchmark 2: Stabilizer Backend    - 3/3 passing (20× speedup)
✅ Benchmark 3: MPO Operations        - 3/3 passing
✅ Benchmark 4: TDVP Time Evolution   - 2/2 passing
✅ Benchmark 5: VQE/QAOA             - 2/2 passing
✅ Benchmark 6: 2D Circuits          - 2/2 passing
✅ Benchmark 7: Integration Tests    - 2/2 passing

Key Metrics

Metric	Value	Notes
Gate throughput	77,304 ops/sec	GPU-optimized
Stabilizer speedup	20.4×	vs generic MPS
MPO evaluations	1,372/sec	Hamiltonian expectations
VQE time (6q)	1.68s	50 iterations
Memory (30q)	0.03 MB	vs 16 GB statevector

---

🔬 Example Applications

VQE for Quantum Chemistry

from atlas_q.vqe_qaoa import VQE, VQEConfig
from atlas_q.mpo_ops import MPOBuilder

# Build Heisenberg Hamiltonian
H = MPOBuilder.heisenberg_hamiltonian(n_sites=6, device='cuda')

# Configure VQE
config = VQEConfig(n_layers=3, max_iter=50)
vqe = VQE(H, config)

# Run optimization
energy, params = vqe.run()
print(f"Ground state energy: {energy:.6f}")

TDVP Time Evolution

from atlas_q.tdvp import TDVP1Site, TDVPConfig
from atlas_q.mpo_ops import MPOBuilder
from atlas_q.adaptive_mps import AdaptiveMPS

# Create Hamiltonian and initial state
H = MPOBuilder.ising_hamiltonian(n_sites=10, J=1.0, h=0.5, device='cuda')
mps = AdaptiveMPS(10, bond_dim=8, device='cuda')

# Configure TDVP
config = TDVPConfig(dt=0.01, t_final=1.0, use_gpu_optimized=True)
tdvp = TDVP1Site(H, mps, config)

# Run time evolution
times, energies = tdvp.run()

---

🚧 Roadmap

Current Status (v0.5.0)

• ✅ GPU-accelerated tensor networks with custom Triton kernels
• ✅ Adaptive MPS with error tracking
• ✅ Stabilizer backend (20× speedup)
• ✅ TDVP, VQE/QAOA implementations
• ✅ All 7/7 benchmark suites passing

Planned Features

• Multi-GPU distributed MPS
• Enhanced PEPS implementation
• Integration adapters for Qiskit/Cirq circuits
• Extended cuQuantum backend support
• Additional tutorial notebooks
• PyPI package distribution

---

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

Development Setup

# Clone with submodules
git clone --recursive https://github.com/followthsapper/ATLAS-Q.git

# Install dev dependencies
pip install -r requirements.txt
pip install pytest pytest-cov black isort

# Run tests
pytest tests/ -v

# Run benchmarks
python scripts/benchmarks/validate_all_features.py

---

📝 Citation

If you use ATLAS-Q in your research, please cite:

@software{atlasq2025,
  title={ATLAS-Q: Adaptive Tensor Learning And Simulation – Quantum},
  author={ATLAS-Q Development Team},
  year={2025},
  url={https://github.com/followthsapper/ATLAS-Q},
  version={0.5.0}
}

---

📄 License

MIT License - see LICENSE for details

---

🙏 Acknowledgments

• PyTorch team for GPU infrastructure
• Triton team for custom kernel framework
• ITensor/TeNPy for tensor network inspiration
• Qiskit/Cirq for quantum computing ecosystem

---

📞 Contact

• Issues: GitHub Issues
• Discussions: GitHub Discussions

---

ATLAS-Q: GPU-accelerated tensor network simulator achieving 626,000× memory compression through adaptive MPS, custom Triton kernels, and specialized quantum state representations.