vbcsr 0.2.2

Variable Block Compressed Sparse Row Matrix with MPI support

---

Distributed variable-block sparse matrices with fast block kernels + thresholded SpMM. Python-first, MPI-scalable, designed for localized-orbital Hamiltonians and multi-DOF PDE systems.

Why VBCSR?

• Hardware-Accelerated Performance: Leveraging SIMD (AVX/AVX2) instructions, precaching and threading, VBCSR delivers state-of-the-art performance for block-sparse matrix operations.
• Easy Integration: Header-only C++ core for easy inclusion in both Python and C++ projects.
• Pythonic & Intuitive: Perform complex linear algebra using natural Python syntax (A * x, A + B, A @ B) and standard NumPy arrays.
• Scalable & Distributed: Built on MPI to handle massive datasets across distributed computing clusters.
• Seamless Integration: Drop-in compatibility with SciPy solvers (scipy.sparse.linalg) for easy integration into existing workflows.

Real-World Applications

VBCSR is designed for complex physics and engineering simulations where block-sparse structures are natural.

1. Linear-Scaling Quantum Chemistry (McWeeny Purification)

Compute the density matrix of a large-scale system using McWeeny Purification. VBCSR's filtered SpMM ensures the matrix remains sparse throughout the iteration. See /examples/ex07_graphene_purification.py for a full implementation.

# 1. Create a large-scale tight-binding Hamiltonian (e.g., 5000 blocks)
H = VBCSR.create_random(global_blocks=5000, block_size_min=4, block_size_max=8)

# 2. Simple spectral scaling
H.scale(0.1); H.shift(0.5)

# 3. McWeeny Purification: P_{n+1} = 3P_n^2 - 2P_n^3
P = H.copy()
for _ in range(10):
    P2 = P.spmm(P, threshold=1e-5)
    P = 3.0 * P2 - 2.0 * P2.spmm(P, threshold=1e-5)
# P is now the density matrix, distributed across your MPI cluster!

2. Stochastic Trace Estimation & DOS (KPM)

Compute the Density of States (DOS) using the Kernel Polynomial Method (KPM) with stochastic trace estimation. See examples/ex08_graphene_dos.py for a full implementation.

energies, dos = compute_kpm_dos(H, num_moments=200, num_random_vecs=10)

3. Quantum Dynamics & PDE Solving

• Quantum Dynamics: Propagate wavepackets in time using Chebyshev expansion (examples/ex09_graphene_dynamics.py).

• Variable-Block PDEs: Solve heterogeneous systems (e.g., mixed Truss/Beam elements) with variable DOFs per node (examples/ex10_variable_pde.py).

Installation

VBCSR requires a C++ compiler, MPI, and BLAS/LAPACK (OpenBLAS or MKL).

Quick Start (Recommended)

conda install -c conda-forge openblas/mkl openmpi mpi4py
pip install vbcsr

From Source

git clone https://github.com/yourusername/vbcsr.git
cd vbcsr
pip install .

For advanced installation options (MKL, OpenMP, etc.), see doc/advanced_installation.md.

Documentation

• User Guide: Concepts, usage, and examples.
• API Reference: Detailed documentation of classes and methods.
• Advanced Installation: BLAS/MKL and OpenMP tuning.

Performance at a Glance

VBCSR is designed for high-performance block-sparse operations, significantly outperforming standard CSR implementations for block-structured matrices.

Blocks	Approx Rows	VBCSR Time (s)	SciPy Time (s)	Speedup
500	~9000	0.0049	0.0209	4.21x
1000	~18000	0.0096	0.0392	4.07x
5000	~90000	0.0468	0.2029	4.33x
10000	~180000	0.0931	0.4151	4.46x
20000	~360000	0.1866	0.8377	4.49x

Pythonic Usage

VBCSR matrices behave like first-class Python objects, supporting standard arithmetic and seamless integration with the SciPy ecosystem.

1. Easy Initialization from SciPy

Convert any SciPy sparse matrix (BSR or CSR) to a distributed VBCSR matrix in one line.

import scipy.sparse as sp
from vbcsr import VBCSR

# Create a SciPy BSR matrix
A_scipy = sp.bsr_matrix(np.random.rand(100, 100), blocksize=(10, 10))

# Convert to VBCSR (automatically distributed if MPI is initialized)
A = VBCSR.from_scipy(A_scipy)

2. Natural Arithmetic & Operators

Use standard Python operators for matrix-vector and matrix-matrix operations.

# Matrix-Vector Multiplication (SpMV)
y = A @ x  # or A.mult(x)

# Filtered Sparse Matrix-Matrix Multiplication (SpMM)
C = A @ B  # or A.spmm(B, threshold=1e-4)

# Matrix Arithmetic
D = 2.0 * A + B - 0.5 * C

3. Distributed Computing Made Simple

VBCSR handles the complexity of MPI communication behind the scenes.

from mpi4py import MPI
comm = MPI.COMM_WORLD

# Create a distributed matrix with 100 blocks
A = VBCSR.create_random(global_blocks=100, block_size_min=4, block_size_max=8, comm=comm)

# Perform distributed SpMV
x = A.create_vector()
x.set_constant(1.0)
y = A @ x

4. Custom Distributed Creation

For full control, define your own block distribution and connectivity pattern.

# Create the distributed matrix structure
A = VBCSR.create_distributed(owned_indices, block_sizes, adjacency, comm=comm)

# Fill blocks (can add local or remote blocks; VBCSR handles the exchange)
A.add_block(rank, rank, np.eye(2))
A.add_block(rank, (rank + 1) % comm.size, np.ones((2, 2)))

# Finalize assembly to synchronize remote data
A.assemble()

5. SciPy Solver Integration

VBCSR implements the LinearOperator interface, making it compatible with all SciPy iterative solvers.

from scipy.sparse.linalg import cg

# Solve Ax = b using Conjugate Gradient
x, info = cg(A, b, rtol=1e-6)