We apply a mass reconstruction technique to large scale structure gravitati
onal distortion maps, simulated for different cosmological scenarii on scal
es from 2.5 arcmin to 10 degrees. The projected mass is reconstructed using
a nonparametric least square method involving the reduced shear on which n
oise due to intrinsic galaxy ellipticities has been added. The distortion o
f the galaxies is calculated using the full lens equation, without any hypo
thesis like the weak lensing approximation, or other linearization.
It is shown that the noise in the reconstructed maps is perfectly uncorrela
ted Poissonian, with no propagation from short to large scales. The measure
d power spectrum and first four moments of the convergence can be corrected
accurately for this source of noise. The cosmic variance of these quantiti
es is then analyzed with respect to the density of the background galaxies
using 60 realizations of each model. We show that a moderately deep weak le
nsing survey (5 x 5 degrees with a typical background population of 30 gal/
arcmin(2) at a redshift z(s) similar or equal to 1) is able to probe the am
plitude of the power spectrum with a few percent accuracy for models with s
igma(8) Ohm(0.8) = 0.6.
Remarkably, we have found that, using the third moment of the local converg
ence only such a survey would lead to a 6 sigma separation between open (Oh
m = 0.3) and flat (Ohm = 1) models. This separation does not require a very
deep survey, and it is shown to be robust against different hypothesis for
the normalization or the shape of the power spectrum.
Finally, the observational strategy for an optimal measurement of the power
spectrum and the moments of the convergence is discussed.