The linear algorithm of the Wiener filter and constrained realizations (CRs) of Gaussian random fields is extended here to perform non-linear CRs. The procedure consists of: (1) Using low resolution data to constrain a high resolution realization of the underlying field, as if the linear theory is valid; (2) Taking the linear CR backwards in time, by the linear theory, to set initial conditions for N-body simulations; (3) Forwarding the field in time by an N-body code. An intermediate step is introduced to `linearize' the low resolution data. The non-linear CR can be applied to any observational data set that is quasi-linearly related to the underlying field. Here it is applied to the IRAS 1.2Jy catalog using 846 data points within a sphere of 6000 km/s, to reconstruct the full non-linear large scale structure of our `local' universe. The method is tested against mock IRAS surveys, taken from random non-linear realizations. A detaile analysis of the reconstructed non-linear structure is presented.

The linear algorithm of the the full non-linear large scale structure of Gaussian random fields is extended here to to perform non-linear CRs. The procedure consists of: (1) Using linear CR of low resolution data to construct a high resolution realization of the underlying field, as if the linear theory is valid; (2) Taking the linear CR backwards in time, by the linear theory, to set initial conditions for an N-body simulation; (3) Forwarding the field in time by an N-body code. An intermediate step might be introduced to `linearize' the low resolution data. The non-linear CR can be applied to any observational data set that is linearly related to the underlying field. Here it is applied to the IRAS 1.2Jy catalog using 843 data points within a sphere of $6000 Km/s$, to reconstruct the full non-linear large scale structure of our `local' universe.