Since the early seventies, there was a conjecture that the first collapse of a selfgravitating dust--like medium (appropriate approximation for nonbaryonic dark matter) results in the formation of a ``pancake" object, that is a thin surface. The conjecture has been based on the Zel'dovich approximate solution of the nonlinear gravitational instability of a generic smooth density perturbation. Recent works cast doubt on the Zel'dovich conjecture, suggesting that the first collapse might be point--like or filament--like rather than pancake--like. Our $N$--body simulations show first pancake collapse. We can reject with 97\% confidence the Bayesian prior that the other kinds of collapse are more or equally probable.

To explain the rich structure of voids, clusters, sheets, and filaments apparent in the Universe, we present evidence for the convergence of the two classic approaches to gravitational clustering, the ``pancake'' and ``hierarchical'' pictures. We compare these two models by looking at agreement between individual structures -- the ``pancakes'' which are characteristic of the Zel'dovich Approximation (ZA) and also appear in hierarchical N-body simulations. We find that we can predict the orientation and position of N-body simulation objects rather well, with decreasing accuracy for increasing large-$k$ (small scale) power in the initial conditions. We examined an N-body simulation with initial power spectrum $P(k) \propto k^3$, and found that a modified version of ZA based on the smoothed initial potential worked well in this extreme hierarchical case, implying that even here very low-amplitude long waves dominate over local clumps (although we can see the beginning of the breakdown expected for $k^4$). In this case the correlation length of the initial potential is extremely small initially, but grows considerably as the simulation evolves. We show that the nonlinear gravitational potential strongly resembles the smoothed initial potential. This explains why ZA with smoothed initial conditions reproduces large-scale structure so well, and probably why our Universe has a coherent large-scale structure.

In recent years there has been a developing realization that the interesting large--scale structure of voids, ``pancakes", and filaments in the Universe is a consequence of the efficacy of an approximation scheme for cosmological gravitation clustering proposal by Zel'dovich in 1970. However, this scheme was only supposed to apply to smoothed initial conditions, and its efficacy for hierarchical clustering has not been justified. We show that this is explained by the fact that the gravitationally evolved potential from N--body simulations closely resembles the smoothed potential of the initial conditions. The resulting ``hierarchical pancaking" picture combines features of the former Soviet and Western theoretical pictures for galaxy and large--scale structure formation.