Tags

Like domains which can be used to make the same variable distinct across different parts of space (e.g. “surface-evaporation” is evaporation from the “surface” domain, while “snow-evaporation” is evaporation from the “snow” domain), it is useful to make a similar distinction for the same variable but at different times. Time tags are used to indicate multiple copies of variables which evolve in time. Tags are used to label different data, or copies of variables, with respect to where they sit on the reference timestep interval. Two tags are of particular importance – the so-called “global” tags “CURRENT” and “NEXT”. These two tags define the endpoints of the reference timestep. Individual PDEs may define other tags their own use; e.g. “flow_midpoint” (for a midpoint time integration scheme) or “transport_subcycled_next” (used to subcycle the transport PK).

Throughout, it can be assumed that the same variable at different tags always has the same structure, e.g. a field on all cells of a given mesh, or a single scalar value, etc. When a step is complete, we can always assign all variables from their global “NEXT” tag’s data to their global “CURRENT” tag’s data to advance the step, e.g.:

snow-evaporation@CURRENT = snow-evaporation@NEXT

Note the use of the “@” delimiter to separate variables from tags.

Tags become particularly important when one realizes that Amanzi-ATS integrates in time with an adaptive timestep; any PDE is allowed to fail at advancing to the next timestep. In a set of coupled PDEs, if only one fails, we must be able to back up and recover the initial state, then try again with a smaller timestep. Therefore, all variables which cannot be recomputed must have at least two copies – one for the “CURRENT” and one for the “NEXT” tags; if another PDE fails after this PDE was successful, we must not have thrown away the “CURRENT” values.