A memory saving algorithm for eigenvalue computations of quantum systems
We compute the low lying spectrum of quantum many-body Hamiltonians with considerably less memory requirements than standard exact diagonalisation (ED) methods like Lanczos. A truncation of the singular value decomposition in each ED iteration reduces the memory cost. We demonstrate convergence using the Heisenberg model on frustrated lattices up to 36 spin-1/2 sites, performed using modest computational resources. Our current efforts to extend the reach of Contractor Renormalisation to larger building blocks will be presented.
[Reference : arXiv:1105.0007, Marvin Weinstein, Assa Auerbach, V. Ravi Chandra]