When a real weld is started, the weld pool is not formed instantly. First, a Gaussian flux distribution heats the solid to the melting point, then a thin layer of molten metal forms. This grows in width and depth with time. With increasing depth, the weld pool surface is depressed and stirring is induced by electro-magnetic, buoyancy, arc pressure and surface tension gradient forces.

Modeling this starting transient is much more complex than modeling the steady-state weld pool. A simple technique is to hold the weld pool stationary for a few seconds. In practice, welders usually do this. The total energy input for the prescribed temperature heat source for time (0, t{) will equal the energy input for a constant power heat source for a time (0, t2), where (ti<t2). Using the shorter time, ti, with the prescribed temperature heat source reduces this error.

There is also a second order error in the starting transient temperature field because of time-dependent diffusion.

During the starting transient of the weld, the natural boundary condition models, i. e., thermal flux models develop the weld pool more realistically.

Current technology uses distribution functions with constant size and shape. It would be more realistic to interpolate the distribution function from a Gaussian flux distribution on starting to the steady state flux and power density distributions. The prescribed temperature distribution function is quite unrealistic on starting. In terms of an electrical circuit analogy, it applies a constant voltage for charging a capacitor. This results in a high power starting transient. However, the effect on the temperature field at later times is small and unless one is specifically interested in the starting transient, it can be neglected. It has been proposed that this problem could be resolved by applying the prescribed temperature heat source until the region near the weld pool reaches steady state. At that time compute the power density and flux distribution. Then restart the weld with this power density and flux distribution.