During the calculation of the Power curve fit, is the error minimized on a log scale, or a linear scale?

During the calculation of the Power curve fit, the error is minimized on a log scale, not a linear scale.