The current models predicting fault geometry and growth are limited to theoretical considerations mostly from fracture mechanics theories (Rice, 1968; Palmer and Rice, 1973, Rudnicki, 1980) and restricted datasets from geological investigations. Consistent with fracture mechanics theories, in structural geology, a fault cutting through heterogeneous layers is described with elliptical tip-lines and maximum displacement in its center (Barnett et al., 1987). Although, fracture mechanics can explain fault propagation at the fault tip through a process zone, it cannot properly predict the lateral fault growth into cross-fault damage zone and eventual localization of strain into the fault core (Figure 1a-b). In addition, the theories fail to consider the effect of fault segmentation and asperities, on displacement partitioning. We report that new advances in numerical modelling of faults and direct fault imaging and property extraction from seismic data utilizing Deep Learning reveals a realistic 3D structure of fault and its mechanical behavior, which in fact shows discrepancy to the previous fault mechanical models. Statistically, faults are normally considered as fractal phenomena, which is consistent with most of the fracture mechanics theories. This means that the fault should repeat the same mechanical behavior at different scales. However, new fault studies at different scales using different methods, suggest that faults can change their behavior at different scales and there is no scale-invariant relation between the fault attributes (Kolyukhin and Torabi, 2012, 2013; and Torabi et al., 2023).

Figure 1. a) An illustration of a fault plane and its surrounding fault core and damage zones from structural point of view. b) Fault concept from fracture mechanics view. c) A 3D structure of a fault with displacement distributions from seismic data. Modified after Torabi et al., 2023.