A cross-machine comparison of global parameters that determine the Runaway Electron (RE) generation and loss process during tokamak start-up was carried out with the aim to extrapolate these to ITER. The study found that all considered discharges, also those that do not show signs of RE, are non-thermal at the start, i.e. have a streaming parameter larger than 0.1. During the current ramp-up the electric field, E, remains above the critical value, Ec, that allows RE in the plasma. The distinction to be made is not if RE can form but, if sufficient RE can form fast enough such that they are detected or start to dominate the dynamics of the tokamak discharge. The dynamics of the value of E, density and temperature during tokamak are key to the formation of RE. It was found that larger devices operate with E closer to Ec, due to their higher temperatures, hence the RE generation is relatively slower. The slower time scales for the formation of RE, estimated to be of the order of 100s of ms in ITER simplifies the development of avoidance schemes. The RE confinement time is also an important determinant of the entire process and is found to increase with the device size. The study also revealed that drift orbit losses, a mechanism often attributed as the main RE loss mechanism during the early tokamak discharge, are actually more difficult to achieve. RE losses might be more likely attributed to RE diffusion due to magnetic turbulence.

Understanding formation of start-up Runaway Electrons (REs) is essential to ensure successful plasma initiation in ITER. The design of ITER start-up scenarios requires not only predictive simulations but also validation of assumptions. There are two generation mechanisms of REs: the Dreicer generation and runaway avalanche. Steady-state models for the Dreicer generation are likely to be valid due to the slower evolution of the plasma parameters during the plasma start-up. This is confirmed through the kinetic simulation, which demonstrate that the Dreicer generation can be extracted from the distribution function in dynamic scenarios. However, the runaway avalanche growth rate can differ from its steady state. There are two stages of evolution in the runaway avalanche growth rate before it reachs the steady state. In stage I, REs incapable of participating in the RE amplification reduce the runaway avalanche growth rate. In stage II, REs with finite energy can increase the runaway avalanche growth rate, but the effect is not significant. An error in the RE density due to the runaway avalanche is usually insignificant in most start-up RE modelling. The kinetic modelling of the KSTAR standard ohmic scenario confirms that the RE density can be modelled using the analytic formulas if it is the runaway-free discharge. However, the runaway-dominant discharge may demand the kinetic modelling. The scan of the Coulomb logarithm for the nonstandard ITER plasma start-up suggests that the timing of runaway current takeover predicted by the fluid simulation may change depending on the choice of the Coulomb logarithm.