1: \begin{abstract}

2: 

3: We present a modified  Lanczos algorithm  to diagonalize lattice Hamiltonians with dramatically reduced memory requirements, {\em without restricting to variational ansatzes}.

4: The lattice of size $N$ is partitioned  into two subclusters. At each iteration  the Lanczos vector is projected into two sets of $n_{{\rm svd}}$

5: smaller subcluster vectors using  singular value decomposition. For low entanglement entropy $S_{ee}$,  (satisfied by

6: short range Hamiltonians), the truncation error is expected to vanish as $\exp(-n_{{\rm svd}}^{1/S_{ee}})$.

7: Convergence is tested for the Heisenberg model on

8: Kagom\'e clusters of 24, 30 and 36 sites, with no lattice symmetries exploited, using less than 15GB of dynamical memory. Generalization of the Lanczos-SVD algorithm to multiple partitioning is discussed, and comparisons to other techniques are given.

9: \end{abstract}

10: