The long-term behavior of rock salt is commonly predicted using creep laws fitted on laboratory data at medium-high deviatoric stresses ( \(\ge\) 5 MPa), which do not fully capture the relevant stress range for underground applications. Indeed, it is a common practice in numerical modeling to extrapolate high-stress experimental data to estimate material behavior at lower stress levels. Additionally, the analysis of the simulation results is often combined with specific design criteria, with particular focus on dilatancy. While some dilatancy criteria are embedded at the constitutive level, others are used only during post-processing, or are of empirical nature. This study simulates hydrogen storage in salt caverns using two in-house finite element codes and the RTL constitutive model. This model accounts for deviatoric and volumetric viscoplastic strains, incorporates dilatancy at the constitutive level and has been calibrated with recent multi-stage confined creep tests over a differential stress range of 0.5–20 MPa and temperatures of 20–50 \(^\circ\) C. The results of the thermomechanical simulations show that extrapolating high-stress data to low-stress regimes can lead to significant underestimations of cavern volume loss and to misrepresentation of stress and strain magnitudes and profiles. Likewise, incorporating dilatancy at the constitutive level offers more accurate predictions of the onset and extent of dilatancy zones. Such zones stabilize over time, remaining smaller than the region influenced by very low deviatoric stresses. As emphasized in this study, accurate assessments of stability require that laboratory experiments and the calibration of constitutive models be conducted as much as possible within the ranges of stress, temperature, strain and strain rates relevant for the underground application.
