Sparse Full Configuration Interaction

Lijun Wang

2024-01-05

Abstract:We propose an efficient deterministic method to calculate the full configuration interaction (FCI) ground state energy. This method leverages the sparseness of the Lanczos basis vectors that span the Krylov subspace associated with the Hamiltonian to rapidly calculate the lowest eigenvalue of the effective Hamiltonian formed in this basis. By iteratively performing the spanning and diagonalization steps, this method is capable of rapidly reaching chemical accuracy for a variety of strongly correlated molecules, including stretched N$_2$ and C$_2$, in merely several tens of iterations. To accelerate our algorithm, we implement parallelized sparse matrix-sparse vector multiplication, which features a novel hashing method that establishes a one-to-one correspondence between the set of full configuration Slater determinants and a set of consecutive integers. A near-linear acceleration was observed.

Chemical Physics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to calculate the full configuration interaction (FCI) ground - state energy efficiently and accurately in computational chemistry. Specifically, the author proposes a new deterministic method - Sparse Full Configuration Interaction (SFCI), aiming to use the sparsity of Lanczos basis vectors to quickly calculate the minimum eigenvalue of the effective Hamiltonian in this basis. By iteratively performing the basis vector generation and diagonalization steps, this method can reach chemical accuracy within dozens of iterations and is applicable to a variety of strongly correlated molecules, such as stretched nitrogen (N₂) and carbon gas (C₂). In addition, to accelerate the algorithm, the author also implements parallel sparse matrix - sparse vector multiplication and introduces a novel hashing method to establish a one - to - one correspondence between the full - configuration Slater determinants and the continuous integer set, thereby achieving a nearly linear acceleration effect.

### Core problems of the paper
1. **Improving computational efficiency**: Due to the combinatorial explosion of computational complexity in the traditional FCI method, it can only be applied to relatively small chemical systems. The SFCI method significantly reduces the computational cost by using the sparsity of Lanczos basis vectors, making it possible to handle larger - scale systems.
2. **Improving computational accuracy**: The SFCI method can reach chemical accuracy within dozens of iterations, which is particularly important for the accurate calculation of strongly correlated systems.
3. **Parallel acceleration**: Through parallel sparse matrix - sparse vector multiplication, the SFCI method achieves efficient computational acceleration and shows a nearly linear acceleration effect especially on multi - core processors.

### Main contributions
- **Utilization of sparsity**: By truncating unimportant components in Lanczos basis vectors, the exponentially growing computational complexity is avoided.
- **Parallel algorithm**: A parallel sparse matrix - sparse vector multiplication algorithm is proposed, which significantly improves computational efficiency.
- **Hashing method**: An efficient hashing method is introduced to establish the correspondence between the full - configuration Slater determinants and the continuous integer set, solving the indexing problem in large - scale systems.

### Application scenarios
- **Strongly correlated molecules**: Such as stretched N₂ and C₂. The electronic structures of these molecules are complex and difficult to handle by traditional methods.
- **Quantum materials**: Predicting the properties of quantum materials requires an accurate solution to the Schrödinger equation.

### Experimental results
- **Convergence**: The SFCI method can reach chemical accuracy within dozens of iterations, and the accuracy is further improved as the number of included Slater determinants increases.
- **Parallel acceleration**: On multi - core processors, the SFCI method achieves a nearly linear acceleration effect, significantly improving computational efficiency.

In conclusion, this paper solves the limitations of the traditional FCI method in terms of computational complexity and accuracy by proposing the SFCI method, providing a new solution for the accurate calculation of strongly correlated systems.