What makes something a four-vector
The coordinates of an event, x^\mu=(ct,x,y,z) with \mu=0,1,2,3, transform under a boost by a definite matrix rule. A four-vector is any set of four numbers that transforms by that same rule. That is not a triviality: it is a strict membership test. If a quantity transforms like x^\mu, it is a four-vector and inherits all the geometry; if it does not, it is just four numbers travelling together and none of the theorems apply.
The transformation law, using the Einstein summation convention: a repeated index (here \nu) is summed over $0,1,2,3$. \Lambda^{\mu}{}_{\nu} is the Lorentz (boost/rotation) matrix.
This is a strict upgrade over Volume I's ordinary vectors. A 3-vector like velocity has three components and rotates under spatial rotations. A four-vector has four and transforms under the full Lorentz group — boosts included. The pattern to internalise: good relativistic physics is written as equations between four-vectors (and tensors), because such an equation, if true in one frame, is automatically true in all of them.
Indices up and down, and the invariant dot product
Four-vectors come with components carrying an upper index, A^\mu (called contravariant). The metric tensor \eta_{\mu\nu}=\mathrm{diag}(+1,-1,-1,-1) — the Minkowski metric — lets you lower an index to make the covariant components A_\mu. Lowering just flips the sign of the spatial parts.
Lowering an index with the metric. Raising uses the inverse metric \eta^{\mu\nu}, which here is numerically the same.
The payoff is the invariant dot product. Contract an upper with a lower index and the result is a scalar — the same number in every frame:
The Minkowski inner product. Setting B=A gives the squared `length` A\cdot A=(A^0)^2-|\vec A|^2, a Lorentz invariant. The interval s^2 of Guide 1 is just x\cdot x.
Four-velocity and proper time
To build a four-vector out of a particle's motion, we need to differentiate its position — but with respect to what? Coordinate time t is frame-dependent, so dx^\mu/dt is not a four-vector. The fix is to use the one clock everyone agrees on: the particle's own clock, ticking proper time \tau, defined by the interval along the worldline.
Proper time is the interval divided by c; it is the time elapsed on a clock carried along the worldline, and it is a Lorentz scalar.
Because \tau is a scalar and x^\mu is a four-vector, the derivative U^\mu=dx^\mu/d\tau is automatically a four-vector — the four-velocity. Working it out with d\tau=dt/\gamma:
The four-velocity. Its invariant magnitude is always c — every particle moves through spacetime at the speed of light, splitting that motion between time and space.