1: \begin{abstract}

2: We use a recently developed technique, which allows to perform few Lanczos steps

3: on a given wavefunction even for large system sizes,

4: to investigate the $t{-}J$ model in the physical parameter region

5: and to check the stability of the BCS d-wave variational

6: wavefunction \cite{gros}.

7: Our statistical Lanczos algorithm, which extends and improves the

8: one Lanczos step proposed in Ref.~\cite{heeb}, has been extensively

9: tested on the small $L=18$ sites cluster where many  Lanczos iterations can be performed exactly.

10: In this case, at doping $\delta\sim 10\%$ the BCS wavefunction represents

11: a very good initial state to achieve extremely accurate energies and 

12: correlation functions with few Lanczos iterations.

13: For large sizes ($L\le 98$) the behavior is  similar: the

14: low-energy d-wave order parameter $P_d$ is weakly affected by a couple 

15: of Lanczos iterations in the low doping $\delta\sim 10\%$ region,

16: whereas the energy is considerably lowered.

17: As a further test of our calculation we have computed the variance of

18: the Hamiltonian  

19: $\Delta E_p=(\langle \hat{H}^2 \rangle-\langle \hat{H} \rangle^2)/L^2$  

20: on the BCS wavefunction with $p=0,1,2$ Lanczos steps. 

21: For large $p$, when the Lanczos algorithm converges to the exact

22: ground state, the variance vanishes exponentially with increasing $p$.

23: The remarkable reduction of the variance, observed for $p=1,2$ Lanczos steps

24: even for the largest lattice size considered, suggests

25: a smooth and rapid convergence to the exact ground state.

26: These results support the existence of off-diagonal long-range d-wave 

27: superconducting order in the two-dimensional $t{-}J$ model.

28: \end{abstract}

29: