The difficulty in conceiving spherical or elliptical space arises mainly because \index{Space, a network of intervals}% we think of space as a continuum in which objects are \emph{located}. But it was explained in \SecRef{1} that location is not the primitive conception, and is of the nature of a computational result based on the more fundamental notion of extension or distance. In using the word ``space'' it is difficult to repress irrelevant ideas; therefore let us abandon the word and state explicitly that we are considering a \emph{network of intervals} (or distances, since at present we are not dealing with time). The relation of interval or distance between two points is of some transcendental character comparable, for example, with a difference of potential or with a chemical affinity; the reason why this particular relation is always associated with geometrical ideas must be sought in human psychology rather than in its intrinsic nature. We apply measure-numbers to the interval as we should apply them to any other relation of the two points; and we thus obtain a network with a number attached to every chord of the net. We could then make a string model of the network, the length of each string corresponding to the measure-number of the interval. \PageSep{159} Clearly the form of this model---the existence or non-existence of unexpected cross-connections---cannot be predicted \Foreign{a~priori}; it must be the subject of observation and experiment. It may turn out to correspond to a lattice drawn by the mathematician in a Euclidean space; or it may be cross-connected in a way which cannot be represented in a lattice of that kind. Graphical representation is serviceable as a tool but is dangerous as an obsession. If we can find a graphical representation which conforms to the actual character of the network, we may employ it; but we must not imagine that any considerations as to suitability for graphical representation have determined the design of the network. From experience we know that small portions of the network do admit of easy representation as a lattice in flat space, just as small portions of the earth's surface can be mapped on a flat sheet. It does not follow that the whole earth is flat, or that the whole network can be represented in a space without multiple connection. \Section{69.}{Law of gravitation for curved space-time}