Abstract:Electron repulsion integral tensor has ubiquitous applications in quantum chemistry calculations. In this work, we propose an algorithm which compresses the electron repulsion tensor into the tensor hypercontraction format with $\mathcal{O}(n N^2 \log N)$ computational cost, where $N$ is the number of orbital functions and $n$ is the number of spatial grid points that the discretization of each orbital function has. The algorithm is based on a novel strategy of density fitting using a selection of a subset of spatial grid points to approximate the pair products of orbital functions on the whole domain.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to compress the electron repulsion integral tensor efficiently in order to reduce the computational cost of ab initio calculations in quantum chemistry and materials science. Specifically, the authors propose a new algorithm that can compress the electron repulsion integral tensor into the tensor hypercontraction format, and the computational complexity is \(O(nN^2 \log N)\), where \(N\) is the number of orbital functions and \(n\) is the number of spatial grid points used in the discretization of each orbital function.
### Main Problems and Solutions
1. **Problem Background**:
- The electron repulsion integral tensor is widely used in many electron - structure theories, such as Hartree - Fock, density functional theory (DFT), RPA, MP2, CCSD, and GW.
- Calculating these integrals is one of the key steps in accelerating ab initio calculations in quantum chemistry and materials science.
2. **Limitations of Existing Methods**:
- Although the traditional density fitting method is effective, its computational complexity is relatively high, usually \(O(N^4)\) or higher.
- Other methods such as multipole moment approaches and pseudospectral representation also have their own limitations.
3. **Advantages of the New Algorithm**:
- The proposed algorithm is based on a novel strategy, that is, by selecting a part of the spatial grid points to approximate the pair products of orbital functions over the entire domain.
- This algorithm uses the randomized column selection method for low - rank matrix approximation, thereby achieving efficient compression.
- The computational complexity of the new algorithm is \(O(nN^2 \log N)\), which significantly reduces the computational cost.
### Algorithm Overview
- **Core Idea**: Select specific grid points \(x_\mu\) through interpolative decomposition to approximate the orbital function pair product \(\rho_{ij}(x)\).
- **Formula Representation**: The approximate form is
\[
\rho_{ij}(x)\approx\sum_\mu\rho_{ij}(x_\mu)P_\mu(x)=\sum_\mu\psi_i(x_\mu)\psi_j(x_\mu)P_\mu(x)
\]
where \(P_\mu(x)\) is an auxiliary basis function determined by least - squares fitting.
- **Computational Steps**:
1. Reshape \(\rho_{ij}(x)\) into a \((N^2)\times n\) matrix.
2. Perform a random Fourier projection on this matrix.
3. Use the pivoted QR decomposition to select appropriate columns.
4. Calculate the auxiliary basis function \(P_\mu(x)\) according to the selected columns.
### Experimental Results
- In 1D and 3D numerical tests, this algorithm shows good performance and controllable errors.
- For a fixed error threshold \(\epsilon\), the number of auxiliary basis functions \(N_{\text{aux}}\) grows linearly with \(N\), verifying that the complexity of the algorithm is \(O(nN^2 \log N)\).
### Conclusion
The algorithm proposed in this paper not only realizes the effective compression of the electron repulsion integral tensor but also significantly reduces the computational complexity, providing new tools and methods for large - scale electron - structure calculations.