If the present result sounds self-contradictory, the fault lies with the name ``absolute change'' which we have tentatively applied to the thing under discussion. The name is illuminating in some respects, because it shows the continuity of covariant differentiation with the conceptions of elementary physics. For instance, no one would hesitate to call~\Eq{(33.1)} the absolute rate of change of momentum in contrast to the apparent rate of change~$dh_{1}/dt$. But having shown the continuity, we find it better to avoid the term in the more general case of non-Euclidean space. Following Levi-Civita and Weyl we use the term \emph{parallel displacement} for \index{Displacement!parallel}% \index{Parallel displacement}% what we have hitherto called displacement without ``absolute change.'' The condition for parallel displacement is that the covariant derivative vanishes. We have hitherto considered the absolute change necessary in order that the vector may return to its original value, and so be a single-valued function of position. If we do not permit any change en route, i.e.\ if we move the vector by parallel displacement, the same quantity will appear (with reversed sign) as a discrepancy~$\delta A_{\mu}$ between the final and initial vectors. Since these are at the same point the difference of the initial and final vectors can be measured immediately. We have then by~\Eq{(33.2)} \[ \delta A_{\mu} = (A_{\mu\nu\sigma} - A_{\mu\sigma\nu})\, dx_{\nu}\, dx_{\sigma}, \] which may also be written \[ \delta A_{\mu} = \tfrac{1}{2} \iint (A_{\mu\nu\sigma} - A_{\mu\sigma\nu})\, dS^{\nu\sigma}, \Tag{(33.3)} \] where the summation convention is now restored. We have only proved this for an infinitesimal circuit occupying a coordinate-mesh, for which $dS^{\mu\nu}$~has only two non-vanishing components $dx_{\nu}\, dx_{\sigma}$ and $-dx_{\nu}\, dx_{\sigma}$. But the equation is seen to be a tensor-equation, and therefore holds independently of the coordinate-system; thus it applies to circuits of any shape, since we can always choose coordinates for which the circuit becomes a coordinate-mesh. But \Eq{(33.3)}~is still restricted to infinitesimal circuits and there is no way of extending it to finite circuits---unlike Stokes's theorem. The reason for this restriction is as follows---