Geometry dependence in linear interface growth

The effect of geometry in the statistics of nonlinear universality classes for interface growth has been widely investigated in recent years, and it is well known to yield a split of them into subclasses. In this work, we investigate this for the linear classes of Edwards-Wilkinson and of Mullins-Herring in one and two dimensions. From comparison of analytical results with extensive numerical simulations of several discrete models belonging to these classes, as well as numerical integrations of the growth equations on substrates of fixed size (flat geometry) or expanding linearly in time (radial geometry), we verify that the height distributions (HDs) and the spatial and the temporal covariances are universal but geometry-dependent. In fact, the HDs are always Gaussian, and, when defined in terms of the so-called "KPZ ansatz" [h≃v_{∞}t+(Γt)^{β}χ], their probability density functions P(χ) have mean null, so that all their cumulants are null, except by their variances, which assume different values in the flat and radial cases. The shape of the (rescaled) covariance curves is analyzed in detail and compared with some existing analytical results for them. Overall, these results demonstrate that the splitting of such university classes is quite general, being not restricted to the nonlinear ones.