Flow separation is a ubiquitous phenomenon present in several aeronautical applications, such as low-pressure turbines (LPT) and Micro Aerial Vehicle wings, and is associated with detrimental effects on aerodynamics and performance. These effects are ultimately dominated by the large-scale coherent turbulent structures formed in the separated shear layer. Understanding the physics underlying the formation of these large-scale structures can be exploited towards their efficient manipulation, with the ultimate objective of informing flow control strategies that reduce the negative impact of flow separation. Large Eddy Simulations (LES) are today capable of accurately predicting the flow dynamics for cases of industrial interest, but their computational cost prevents their use in fast-return design cycles or to inform flow control strategies. As an alternative, linear instability equations can obtain predictive information on the dominant coherent structures, with reasonable fidelity in terms of spatial structure and frequency content, at a small fraction of the computational cost associated with LES. Resolvent analysis [1] is one formulation of the linearized equations in the frequency domain, in which the impact of non-linear interactions and external excitation are recast as an endogenous forcing term. The analysis then consists in computing, for each frequency, the forcing term that leads to an optimal response in terms of energy amplification, allowed by the linearized equations over the mean flow. In this work, we employ resolvent analysis to model the most energetic flow structures formed on the separated flow over a wall-mounted bump geometry that reproduces the pressure gradient distributions present in a LPTblade [2]. The formation of coherent structures in this flow is governed by the convective amplification of the pre-existing flow disturbances by the intense mean flow shear [3], making it a good test case to assess different numerical approaches for conducting resolvent analysis. In particular, we compare matrix-forming approaches based on different manners of defining the linearized equations, i.e. linearized analytical equations and a numerical Jacobian typically used in implicit solvers, and alternatives based on the time-marching of the equations.