We should review how we are weighting and normalizing our objective functions. Questions to consider and check the implementation of:

• For "local" errors, do we want the sum of squares (scalar) error to be proportional to the error integrated over the volume?
• Should weights/normalizations be applied in the compute method, or the compute_scalar method?
• Should all objective values be independent of the grid resolution, even if the user specifies normalize=False? (This is not currently the case, but I think it should be)
• What terms should be considered "weights" and which terms should be considered "normalizations"? (I propose normalizations should only be making the output units dimensionless, not making them independent of resolution)