Anders M.N. Niklasson
Abstract:An implicit purification scheme is proposed for calculation of the temperature-dependent, grand canonical single-particle density matrix, given as a Fermi operator expansion in terms of the Hamiltonian. The computational complexity is shown to scale with the logarithm of the polynomial order of the expansion, or equivalently, with the logarithm of the inverse temperature. The system of linear equations that arise in each implicit purification iteration is solved efficiently by a conjugate gradient solver. The scheme is particularly useful in connection with linear scaling electronic structure theory based on sparse matrix algebra. The efficiency of the implicit temperature expansion technique is analyzed and compared to some explicit purification methods for the zero temperature density matrix.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to efficiently calculate the temperature - dependent single - particle density matrix, especially in large systems. Specifically, the author proposes an implicit purification scheme for calculating the temperature - dependent density matrix in the form of Fermi - Dirac operator expansion for a given Hamiltonian.
### Main problems
1. **Computational complexity**: Traditional explicit purification methods have high computational complexity when dealing with large systems. In particular, as the temperature decreases, the computational cost increases significantly.
2. **Temperature - dependence**: Many existing methods mainly focus on the calculation of zero - temperature density matrices, and lack effective methods for dealing with density matrices at finite temperatures.
3. **Linear - scaling electron - structure theory**: In order to study complex large - scale material systems, a method that can be combined with linear - scaling electron - structure theory is required to achieve linear growth of computational complexity with system size.
### Solutions
The implicit purification scheme proposed by the author solves the above problems in the following ways:
- **Low logarithmic complexity**: The computational complexity of this method is proportional to the order of the polynomial or the logarithm of the inverse temperature, that is, \(O(\log n)\) or \(O(\log \beta)\), which greatly improves the computational efficiency.
- **High precision**: Use Padé approximation to approximate the Fermi - Dirac distribution function, so that high precision can be obtained at a lower order.
- **Iterative solution**: Efficiently solve the linear equations in each iteration step by the conjugate gradient solver, further improving the performance of the algorithm.
### Key formulas
1. **Fermi - Dirac distribution function**:
\[
\Phi_{\text{FD}}(\varepsilon, \mu, \beta) = \frac{1}{e^{\beta(\varepsilon - \mu)} + 1}
\]
2. **Temperature - dependent density matrix**:
\[
\rho(\beta) = \Phi_{\text{FD}}(H, \mu, \beta)
\]
3. **Padé approximation**:
\[
G_n(x) = \frac{x^n}{x^n+(1 - x)^n}
\]
\[
\Phi_{\text{FD}}(x, \frac{1}{2}, 4n)\approx 1 - G_n(x)
\]
4. **Implicit purification iteration formula**:
\[
X_1 = F(H, \mu)
\]
\[
X_{i + 1}=G_m(X_i),\quad i = 1, 2,\ldots,\log_m(n)
\]
\[
\rho(\beta)=I - X_{i + 1}
\]
Through these improvements, this method can not only achieve high - precision temperature - dependent density matrix calculations at a lower computational cost, but is also particularly suitable for the study of large - scale material systems in combination with linear - scaling electron - structure theory.