Elements of Spacetime Geometry: Hands-On Exercise: Manifolds and Coordinate Patches

3
Elements of spacetime geometry
Hands-on exercise: manifolds and coordinate patches

To complete this exercise you will need the following supplies:
• Two rulers or other rigid length-measuring sticks.
• Large flat surface like the top of a table or desk, similar to that used in the Chapter 1
hands-on exercise.
• Large spherical object (LSO), similar to that used in the Chapter 1 hands-on exercise.
• A device for measuring time.
Build a model of E2 (two-dimensional Euclidean space) with two rulers taped

together at right angles to make rectangular coordinate axes. Place the coordinate
axes on the surface of the table or desk that you used in the Chapter 1 hands-on
exercise so that the corners are touching the surface. Move them around to see
what it means for the surface to be both locally and globally like E2 . Could you
accurately measure the distance between any two points on this surface using only
these straight axes, if they were long enough and couldn’t bend?
Place the axes on the surface of the LSO that you used in the Chapter 1 hands-
on exercise so that they are tangent to the surface. Move them around from point
to point and ponder what it means for the surface to be locally but not globally
like E2 . Could you accurately measure the distance between any two points on this
surface using only these straight axes, if they were long enough and couldn’t bend?
How could you add a time axis to your model of space to make a model of
spacetime? With what combination of objects or devices could you measure the
distance between two events in this model spacetime?
3.1 Space and spacetime

We learned some surprising things about space and time in Chapter 2. Space and
time have failed to be absolute structures that can always be distinguished from
55
Downloaded from https://www.cambridge.org/core. Columbia University Libraries, on 21 Aug 2018 at 13:10:11, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511755811.004
56 Elements of spacetime geometry
one another. According to everything that we’ve learned about the propagation of
light, space and time must transform into one another, and so rather than existing
in space that evolves in time, with space and time distinct phenomena, we exist
in spacetime, with space and time mixing with one another, time able to turn into
space and space able to turn into time through the Lorentz transformation, which
in two spacetime dimensions looks like
τ̃ = γ τ − γβx
x̃ = −γβτ + γ x. (3.1)
It’s more than just a cliché to say that this relativity of the phenomena of space
and time has enormous implications. In this chapter, however, we will ignore the
enormous implications and delve into the nuts and bolts of spacetime, to look into
spacetime geometry from the dry exacting perspective of mathematics, to see what
makes spacetime tick, so to speak.
Coordinates in space
Before we get to spacetime, let’s first clarify what we mean by space. Let’s start
from the abstract concept of a manifold.
To make a manifold M, we start with a well-behaved set of points, that is, a set
of points fulfilling two requirements regarding how the points in M can be divided
into open subsets or neighborhoods. First, near any point p in M, we should be
able to find some open subset of points U ( p) that is near p. Second, any two points
p and q in M can be put into different open subsets U ( p) and U (q) of M that
don’t intersect.
For any point p in M, points in an open neighborhood U ( p) near p should
be describable by D real numbers (x 1 , x 2 , . . . , x D ) representing the values of D
coordinates in a Euclidean space E D in D dimensions. These coordinates define a
distance lAB between any two points pA and pB given by

lAB = (xA1 − xB1 )2 + (xA2 − xB2 )2 + · · · + (xAD − xBD )2 (3.2)
in this local neighborhood of p. However, this distance is a property of the particu-

lar choice of coordinates and might not be the distance function that we want on the
manifold. Later, we will discuss adding another mathematical structure (namely, a
metric) that gives an intrinsic meaning to distances independent of the choice of
local coordinates.
Near some other point q, points can be described by D real numbers
(w 1 , w 2 , . . . , w D ), with the distance lAB between any two points qA and qB
Space and spacetime 57
U(q)
U(p)
Fig. 3.1. A manifold M with two locally Euclidean coordinate patches U ( p) and
U (q). The intersection U ( p) ∩ U (q) is the shaded area. The local coordinates in
the two neighborhoods are related by a coordinate transformation.
the same as (3.2) but with x → w. In the intersection U ( p) ∩ U (q) of these two
neighborhoods, points can be described by either local coordinate system, with
a smooth coordinate transformation relating the two coordinate systems given by
transition functions
wi = wi (x 1 , . . . , x D ), i = 1, . . . , D. (3.3)
This mapping from one set of local coordinates to another is one-to-one if the
Jacobian
∂wi
J = det M, M ij = (3.4)
∂x j
is everywhere nonzero. If M can be covered by a union of such coordinates
patches, with the overlapping local patches of E D related to one another as above,
then M is a manifold of dimension D.1
The surface of the LSO in the exercise above is an example of a manifold, if we
idealize it so that it is smooth at all distances scales, even the subatomic scale. Any
point p on the surface of the LSO can serve as the origin of a local E2 coordinate
patch, let’s call it U ( p), with local coordinates (x, y) measured on the axes you
made with the two rulers, with their origins located at p. Any nearby point P on the
LSO could be described in terms of coordinates (x P , y P ) on the two axes located
at p. But if P is too far from p, the curvature of the LSO becomes important, and
the location of P can’t be described by coordinates on the axes located at p.
The way out of this problem is to define another local coordinate patch U (q) at
some other point q that is close to P, and then define transition functions that tell us
1 The name differentiable manifold is also used. The transition functions define a differentiable structure on the
manifold.
how the two coordinate patches at p and q are related in the regions U ( p) ∩ U (q)
on the LSO where they overlap. We can cover the whole LSO with coordinate
patches and transition functions, and then have a way to find the coordinates of
any point on the LSO.
On the LSO, the local set of axes twists and turns when it is moved from point
to point on the LSO, and there isn’t a simple way to patch them together and still
have a rectangular coordinate system with one fixed set of axes. This is because
the surface of the LSO is a curved surface. In differential geometry, the curvature
can be computed from the way the local Euclidean coordinate system twists and
turns from one patch to another. When there is no curvature, so that the local
Euclidean coordinate system doesn’t twist or turn from one patch to another, the
whole manifold can be covered globally by a single Euclidean coordinate patch,
and then we say that this manifold is flat.2
On the top of a desk or table, any point P can be described in terms of coordi-
nates along the axes located at some point p, if we imagine the axes extended in
either direction as far as necessary. We really only need one coordinate patch of
E2 to specify the coordinates of any point q on the desk. This is because the desk
is flat.
A flat space can be covered by Euclidean coordinates, but it’s still flat when
described in terms of any other coordinates. For example, flat space in three di-
mensions can be described in rectangular coordinates (x, y, z), or in spherical co-
ordinates (r, θ, φ), related by the coordinate transformation
x = r sin θ cos φ
y = r sin θ sin φ
z = r cos θ, (3.5)
with 0 ≤ θ ≤ π and 0 ≤ φ < 2π . In these coordinates, the metric
dl 2 = d x 2 + dy 2 + dz 2 (3.6)
becomes
dl 2 = dr 2 + r 2 dθ 2 + r 2 sin2 θdφ 2 , (3.7)

but the space is still flat – it’s just E3 in spherical coordinates. We can always invert
the coordinate transformation (3.5) to cover the space with rectangular coordinates
again, and that’s what makes the manifold flat.
2 Note that this definition of flatness assumes that the manifold is infinite, or is finite with a boundary. There
are flat manifolds that cannot be covered globally by one set of Euclidean coordinates because of topological
issues such as periodic boundary conditions. A torus T D in D dimensions is one such case.
Coordinates in spacetime
What happens if we add a time dimension to the flat space described above, and
get flat spacetime? All of the definitions above can be recycled if we just substitute
Minkowski space Md for Euclidean space E D . The local coordinate patches have d
spacetime coordinates (x 0 , x 1 , x 2 , . . . , x D ), where x 0 ≡ τ = ct, and any two
points pA and pB in the local neighborhood U ( p) are separated by the Minkowski
interval sAB , given by
sAB
2
= −(τA − τB )2 + (xA1 − xB1 )2 + · · · + (xAD − xBD )2 . (3.8)
A flat spacetime is a spacetime manifold where one patch of Minkowski coor-

dinates can cover the entire manifold (with a caveat regarding topological issues
similar to that raised in the case of flat space). We will call this manifold Md , for
Minkowski spacetime in d dimensions.
An important property of the Minkowski interval in (3.8) is the fact that it is
the same in all inertial frames. In other words, it is unchanged when one makes
a change of coordinates that corresponds to a Lorentz transformation. In saying
this we are using the name Lorentz transformation in its most general sense, which
includes both the possibility of a Lorentz boost by an arbitrary velocity (of mag-
nitude less than c) and a spatial rotation, as well as an arbitrary combination of
the two. In the special case of one space and one time dimension, there are no
rotations and the only Lorentz transformations are boosts. When there are two
or more spatial dimensions, both types of transformations are possible. The set
of all such transformations forms the Lorentz group, which will be discussed in
Chapter 8.
A point in spacetime has a location in space and a location in time, in other
words, a point in spacetime represents an event, something that happens at a par-
ticular place at a particular time. Points pA and pB refer to events that we shall
call E A and E B . The quantity sAB 2 cannot be the square of the distance between
two events, because it can be negative. But if it’s not the distance between events,
then what does this interval signify? It tells us about the causal structure of the
spacetime – which event can be the cause of another event, and which cannot.
Normally, without taking special relativity into account, we would just use time
to decide the question of causality, assuming that the time between two events is
absolute and the same for all observers. If event E A occurs at time tA and event E B
occurs at time tB , if tA < tB , then event E A could be the cause of, or have influence
on, event E B , but if tB < tA , then the causal relationship is reversed. But this view
doesn’t take into account the speed of light, which is finite and the same for all
observers. When we take into account the speed of light, we lose absolute time
τ Timelike future
E5 E2
Spacelike
E4
Spacelike E0 Null
x
E3
E1
Timelike past
Fig. 3.2. The sign of s 2 divides the spacetime around any event E 0 into regions
of timelike, null and spacelike separation. Events E 3 and E 4 take place on the
null light cone L of E 0 , shown above by dashed lines. Event E 2 takes place at a
timelike interval to the future of E 0 , and event E 1 takes place at a timelike interval
to the past of E 0 . Event E 5 happens at a spacelike interval from E 0 , too soon for
light to travel from E 0 to the location where E 5 happens. Event E 0 can only cause
or influence events to the future of E 0 with a timelike or null separation from E 0 .
The principle of causality is encoded in the geometry of spacetime in a Lorentz
invariant manner.
and learn that time is relative, that two inertial observers will not agree when two
events occur at the same time.
So what happens to causality when the passage of time is relative? The answer
is encoded into the spacetime geometry by the sign of the Minkowski interval,
which is unchanged by a Lorentz transformation, and hence the same for all iner-
tial observers. There are three possibilities for the sign of sAB2 , and these three
possibilities divide up the spacetime around each point into the three Lorentz in-
variant regions described below:
(i) sAB2 > 0: Events E and E are separated by a spacelike interval. There exists a
A B
Lorentz boost at some velocity β to a frame where events E A and E B happen at the
same time, but there exists no Lorentz transformation to a frame where the two events
happen at the same place.
(ii) sAB2 = 0: Events E and E are separated by a lightlike or null interval. This is the
A B
path a beam of light or a massless particle would take to get from event E A to event
E B . There exists no Lorentz transformation at any velocity β to a frame where events
E A and E B happen at the same time or the same place.
τ τ% x%
x
E0
Fig. 3.3. In this spacetime diagram, the coordinates (τ̃ , x̃) are related to the coor-
dinates (τ, x) by a Lorentz boost. The dashed lines represent curves of constant
s 2 from event E 0 = (τ0 , x0 ). These curves are hyperbolas, which degenerate to
straight lines through the origin in the limit s 2 → 0. All of the dashed lines on
this diagram represent Lorentz invariant submanifolds of Minkowski spacetime.
The invariant submanifold with s 2 = 0 is the light cone of event E 0 .
(iii) sAB2 < 0: Events E and E are separated by a timelike interval. There exists a
A B
Lorentz transformation at some velocity β to a frame where events E A and E B happen
at the same place, but there exists no Lorentz transformation to a frame where the two
events happen at the same time.
The set of spacetime events that satisfy s 2 = 0 divides the timelike region of the
spacetime of any event E 0 from the spacelike region of that event. In d spacetime
dimensions, this set forms a submanifold of Md whose coordinates satisfy the
condition
(τ − τ0 )2 = (x 1 − x01 )2 + (x 2 − x02 )2 + · · · + (x D − x0D )2 . (3.9)
This is called a null hypersurface of Md . Slices of constant τ on this null hypersur-

face are (D − 1) spheres representing the fronts of light waves either leaving from
(for τ > τ0 ) or converging at (for τ < τ0 ) the event E 0 . The total hypersurface
L is a Lorentz-invariant manifold called the light cone or null cone of event E 0 .
The half of L with τ < τ0 , denoted L− , is called the past of event E 0 and the half
of N with τ > τ0 , denoted L+ , is called the future light cone of event E 0 . Even
though the time coordinate τ is itself relative, the light cone L of any event in Md
τ τ%
τ1 E1
τ%0 = τ%1
E0 τ0
x%
x
Fig. 3.4. Event E 1 is separated by a spacelike interval from event E 0 , whose light
cone L is shown by the dashed lines. In the S frame with coordinates (τ, x), events
E 0 and E 1 occur at different times, with τ0 < τ1 . In the S̃ frame, with coordinates
(τ̃ , x̃) related to (τ, x) by a Lorentz boost at some velocity β = β0 , events E 0
and E 1 occur at the same time τ̃0 = τ̃1 . If β > β0 , then E 1 happens before E 0 .
The time ordering of two events separated by a spacelike interval depends on the
motion of the observer.
is invariant under a Lorentz transformation. We will show below that the light cone
of an event serves as a boundary between the past and future of that event that is
the same for all inertial observers.
The set of events in Md with a timelike separation from an event E 0 satisfy the
condition
λ2 = (τ − τ0 )2 − |
x − x0 |2 , (3.10)
where λ2 > 0. Each possible value of λ defines a Lorentz-invariant submanifold

of Md with (d − 1) dimensions. For d = 2, we get submanifolds of M2 that are
the hyperbolas that cross the τ axis at τ = λ, shown as dashed lines in the timelike
regions of Figure 3.3. For d > 2, the submanifolds are hyperboloids of revolution
in the (d − 1) space coordinates. Since λ is a continuous parameter, this means the
timelike region of any event can be seen as being filled by an infinite number of
these surfaces. Rearranging the equation to
(τ − τ0 )2 = λ2 + |
x − x0 |2 (3.11)
shows that each value of λ2 corresponds to a minimum possible value for

(τ − τ0 )2 , which occurs when | x − x0 |2 = 0. So the parameter λ is the proper
time of an observer at rest with respect to the (τ, x) coordinate system.
Since each value of λ2 corresponds to a minimum of (τ − τ0 )2 , then τ − τ0 can

never pass through zero as long as λ doesn’t pass through zero. In the limit λ → 0,
the Lorentz-invariant hyperboloid (3.10) degenerates to the light cone (3.9).
Passing through the light cone takes us out of the timelike region of E 0 to the
spacelike region. The past and the future timelike regions of an event E 0 are sets
with zero intersection. The past light cone of E 0 is the Lorentz-invariant bound-
ary of the timelike past of E 0 , and the future light cone is the Lorentz-invariant
boundary of the timelike future of E 0 . Even though time and space mix in spe-
cial relativity, every observer agrees about the boundaries of the timelike past and
future of every event.
In the spacelike region of some event E 0 , however, the picture is different.
Events within the spacelike region of E 0 satisfy the condition
ρ 2 = |
x − x0 |2 − (τ − τ0 )2 , (3.12)
where ρ 2 > 0. As with λ in the timelike region, every value of ρ defines a different
Lorentz-invariant submanifold of Md in the spacelike region of E 0 . For d = 2, we
get submanifolds of M2 that are the hyperbolas that cross the x axis at x = ρ,
shown as dashed lines in the spacelike regions of Figure 3.3.
In the spacelike region of E 0 , τ − τ0 can pass through zero for x = x0 , for any
value of ρ = 0. This is opposite from the situation in the timelike region. In the
spacelike region of E 0 , it makes no sense to say whether any event is to the future
or the past of event E 0 . There is always a Lorentz boost from a frame S where
τ > τ0 to some other frame S̃ where τ̃ ≤ τ̃0 , and vice versa.
Notice that for any event in the spacelike region of E 0 , | x − x0 |2 ≥ ρ 2 , with
equality only for τ = τ0 . This means that |ρ| measures the proper distance between
the event and E 0 . It also means that it is not possible in the spacelike region of E 0
to find a Lorentz boost to a frame where | x − x0 |2 vanishes. In the timelike region
of E 0 , it is always possible to find a frame where a given event in the region takes
place at x0 . The spacelike and timelike regions of an event are in a sense dual to one
another. Differences in location can be transformed away in the timelike region of
an event, and differences in time can be transformed away in the spacelike region
of an event.
The pole in the barn

The difference between spacetime and space can be appreciated by revisiting the
difference between a rotation in space and a Lorentz boost in spacetime. A rotation
of the rectangular coordinates in E D is like a rotation of a rigid object. The axes
all turn together in the same direction as much as needed. A Lorentz boost in some
particular direction, for example the x 1 direction, of the coordinates in Md looks

like
τ̃ = γ τ − γβx 1
x̃ 1 = −γβτ + γ x 1
x̃ i = x i , 2 ≤ i ≤ D. (3.13)
When drawn in the (τ, x 1 ) plane, the τ̃ axis makes an angle of φ = π/2 −
2 tan−1 β with the x̃ 1 axis, so the angle between the axes goes to zero for β → 1.
The Lorentz boost squeezes the time and space coordinate axes into one another,
as shown in Figure 3.3. That’s not how a rigid body transforms. The Lorentz
boost (3.13) could be said to shear the spacetime in the (τ, x 1 ) plane, which tells
us that spacetime behaves more like an elastic medium than a rigid one.
There is a thought experiment that exemplifies this difference, in a situation that
appears paradoxical according to the reasoning that we learn in rigid time and
space, but which is not paradoxical at all once we understand special relativity.
This experiment features a pole being moved through a barn. The pole and the
barn both have proper length L 0 . The pole is being carried on a rocket moving at
velocity β through the barn, which has doors on the front and rear. Before the pole
enters the barn, the front door is open but the rear door is closed. After the rear end
of the pole passes the front door, the front door closes. When the front end of the
pole is about to hit the rear door, the rear door opens.
According to observers in the rest frame S of the barn, the pole is Lorentz-
contracted
from its proper length L 0 to length L P = L 0 /γ , where γ =
1/ 1 − β as usual. The distance between the two barn doors is L B = L 0 . Since
2
L P < L B , the pole easily fits inside the barn with both doors closed.
When we look at this same sequence of events in the rest frame S̃ of the pole,
however, a problem arises. According to observers in the pole frame, the pole
and the rocket are at rest, and the barn comes rushing at them with velocity −β.
The barn is Lorentz-contracted to length L̃ B = L 0 /γ , while the pole has length
L̃ P = L 0 . According to observers in this frame, L̃ P > L̃ B , in other words,
the pole is longer than the barn, so the pole cannot possibly fit inside the barn
with both doors closed.
This appears to violate both common sense and the symmetry that is supposed
to be inherent in the principle of relativity. There shouldn’t be one frame where the
pole can be trapped inside the barn, and another frame where it crashes through
the doors.
The key to this mystery is that it is spacetime, not space or time individually,
that does the stretching and contracting. The length of each object is measured in
τ E4 τ%
x%
E3
E2
E1 x
pole
barn
Fig. 3.5. The pole moving through the barn, plotted in coordinates (τ, x) in the
rest frame of the barn.
the rest frame of the object, by comparing both ends of the object at the same time.
But in special relativity, simultaneity is relative, and time ordering can be relative,
if there is a spacelike separation between two events.
The sequence of events under examination is shown in Figure 3.5 in the rest
frame of the barn, and in Figure 3.6 in the rest frame of the pole and rocket. In
Figure 3.5, the coordinate axes (τ, x) represent the rest frame of the barn, and the
coordinate axes (τ̃ , x̃) represent the rest frame of the pole, moving at velocity β in
the +x direction. The sequence of events according to the time τ in the barn rest
frame is:
E1: The front end of the pole enters through the front door of the barn.
E2: The rear end of the pole enters through the front door of the barn.
E3: The front end of the pole leaves through the rear door of the barn.
E4: The rear end of the pole leaves through the rear door of the barn.
Between times τ2 and τ3 , the pole, of length L 0 /γ , is completely inside the barn,
which in this frame has length L 0 .
τ E4 τ%
x
E2
E3
x% E1
pole barn
Fig. 3.6. The pole moving through the barn, plotted in coordinates (τ̃ , x̃) in the
rest frame of the pole.
In Figure 3.6, the coordinate axes (τ̃ , x̃) represent the rest frame of the pole, and
the coordinate axes (τ, x) represent the rest frame of the barn, moving at velocity
−β in the +x direction. The sequence of events according to the time τ̃ in the pole
rest frame is:
E1: The front door of the barn passes the front end of the pole.
E3: The rear door of the barn passes the front end of the pole.
E2: The front door of the barn passes the rear end of the pole.
E4: The rear door of the barn passes the rear end of the pole.
In the pole frame, the sequence of events E 2 and E 3 is the opposite from what
they were in the barn frame. This is possible because these two events – the rear
end of the pole entering the front door of the barn, and the front end of the pole
leaving the rear end of the barn – occur at a spacelike separation. (Proof of this
fact will be left to the reader as an exercise.)
The pole can’t be contained within the barn between times τ̃2 and τ̃3 , because
τ̃2 > τ̃3 . According to observers for whom the pole is at rest and the barn is mov-
ing, the front end of the pole is already out of the rear door of the barn before the
rear end of the pole has entered through the front door of the barn. According to
Vectors on a manifold 67
the sequence of events as measured according to clocks in the rest frame of the
pole, the pole is never completely contained within the barn at any time. And that
is perfectly consistent with the pole, with length L 0 , being longer than the barn,
which in this frame has length L 0 /γ .
If you feel reassured by this, then don’t be! How do we normally define any
object that exists in nature, such as a pole or a barn? We normally define an object
to exist in space at distinct moments in time. A pole is supposed to have extent in
space, not extent in time. But in special relativity, objects have extent in space and
time. That’s what spacetime means. The measurement of space is connected to the
measurement of time. An object exists in spacetime, and the measurement of its
length depends on a measurement of both space and time.
In Figures 3.5 and 3.6, the pole and barn are represented by the areas they sweep
out as they move in time. These areas are called the world sheets of the pole and
barn, for the one-dimensional representation of the pole and barn we’re using here.
In real life, the pole and barn sweep out world volumes in spacetime.
The pole at any moment in time is represented on the diagram by a slice of the
world sheet of the pole at that time. But the time slices will be at different angles,
depending on the angle of the time axis of the observer relative to the time axis in
the rest frame of the pole. A slice of the pole world sheet at some time τ in the
rest frame of the pole reveals a pole that has length L 0 . A slice of the world sheet
of the pole at some time τ̃ in the barn rest frame reveals a pole that has length
L 0 /γ .
So what do we mean by a rigid object such as a pole or a barn? Is an object that
we see in space just a particular time slice of the world volume of that object in
spacetime? Is there really such a thing as a rigid object in relativity at all? We will
ponder this question again in later chapters.
3.2 Vectors on a manifold

In introductory physics courses, students learn to describe an object by its loca-
tion in space as a function in time, using a time-dependent vector representing the
displacement from the origin of the coordinate system, which in Euclidean coor-
dinates looks like
x(t) = x(t)êx + y(t)ê y + z(t)êz . (3.14)
The basis vectors (êx , ê y , êz ) in the set are mutually orthogonal and have unit mag-
nitude, everywhere in space, for all values of the parameter t. If we have N such
objects, each with mass m i and position vector xi (t), then we can talk about the
collective motion of the whole ensemble by looking at the trajectory of the center
of mass of the whole ensemble

N
mi
N
X cm (t) = xi (t), M= mi . (3.15)
i =1
M i =1
In introductory physics it is taken for granted that all of the above mathematical
operations make sense. We can define vectors by the displacements in a coordinate
basis, the basis vectors are the same everywhere, we can add and subtract vectors,
multiply them by numbers, and check whether they are orthogonal, anywhere in
this space with no problem. In other words, we assume that we are living in a
vector space where all of these operations can be defined.
Newtonian physics makes sense because the physical space we are employing as
a model for nature has the structure of a vector space, namely E3 , Euclidean space
in three dimensions. On a general manifold, a vector space can only be defined at
each point in the manifold. In flat space and spacetime, it’s possible to get away
with ignoring this fact, but in this section, we will not ignore it, and we will show
how vectors in spacetime are properly defined on a general manifold, before we
make use of the convenient fact that the manifold we’re dealing with is flat.
Properties of a vector space

Assume that v, u and w are vectors belonging to some vector space that we will
label V, and that a and b are real numbers that we will call scalars. There are three
operations defined in a vector space: addition, scalar multiplication, and vector
multiplication through the inner product. Each operation comes with a set of ax-
ioms that, when taken together, guarantee that we can do the things that were done
to define a vector like X cm in (3.15).
The set of assumptions that we make about addition is:
(i) The sum of two vectors is a vector: v + u ∈ V.

(ii) Addition is commutative: v + u = u + v.
(iii) Addition is associative: (v + u) + w = v + (u + w).
(iv) The vector 0 ∈ V is the identity element under addition: 0 + v = v.
(v) Every vector v ∈ V has an inverse −v under addition: v + (−v) = 0.
The set of assumptions that we make about scalar multiplication is:
(i) A vector multiplied by a scalar is a vector: av ∈ V.

(ii) Scalar multiplication is associative: a(bv) = (ab)v.
(iii) The scalar 1 is the identity element under scalar multiplication: 1v = v.
(iv) Scalar multiplication is distributive over vectors: a(v + u) = av + au.
(v) Scalar multiplication is distributive over scalars: (a + b)v = av + bv.
Any vector in a vector space can be written as a linear combination of other

vectors in the space. If we can find a set of vectors in the vector space such that
any other vectors in the space can be written as a linear combination of vectors in
that set, that set of vectors forms a basis for the vector space. If there are D such
linearly independent vectors, then we say the vector space has dimension D. If we
denote the set of D basis vectors by {êi }, where 1 ≤ i ≤ D, then any vector v ∈ V
can be written as
v = v i êi , (3.16)
where {v i } are called the components of v in that particular basis, and we are
employing, as usual, the summation convention where pairs of upper and lower
indices are summed over all values.
For example, (êx , ê y , êz ) are the usual basis vectors for E3 . Any vector in
Euclidean space in three dimensions can be written as a linear combination of
those three vectors.
The tangent space

Euclidean space in D dimensions has the structure of a vector space already. On
a generic manifold that is not flat, or is flat but topologically nontrivial, there is
no global vector space where vectors can be defined. There is only a local vector
space defined from point to point. This local vector space is called the tangent
space to the manifold. The tangent space to a manifold M at point p is commonly
written as T p (M). But how are the vectors in this space defined?
In basic calculus, we learn that the derivative of a function f (x) is the tangent
to the curve y = f (x) at point x, and this tangent line acts like a vector, because
it points in some direction in the (x, y) plane. What isn’t usually taught in basic
calculus is that the set of derivatives operating on a function can be equivalent to
a set of vectors in a vector space, and that such a set can be used to define the
tangent space T p (M) for any M. Let’s look at the space of derivative operators
X = a i (∂/∂ x i ), where the a i are constants. X operates on some function f (x),
where x refers to all of the coordinates {x i } in the local patch of E D in M. The
object
∂f
X[ f ] = a i (3.17)
∂xi
represents the action of X on f .
If we have two such operators, say
∂ ∂
X = ai , Y = bi , (3.18)
∂xi ∂xi
then their sum is

∂f
X + Y = W, W[ f ] = ci , (3.19)
∂xi
where ci = a i + bi . Addition and scalar multiplication of these operators is asso-
ciative, commutative and distributive, one can define a 0 operator and an additive
inverse. All of the properties of a vector space noted above apply to these deriva-
tive operators as well. The set of D operators {∂/∂ x i } form a basis for this vector
space, so the vector space has dimension D, the same as the manifold.
But what are the components {a i } that make X a vector in T p (M)? Consider
some curve γ in M, that is, a mapping from an interval of some real parameter
that we will call λ, to the manifold M. This curve can be represented in the local
coordinates of the neighborhood U ( p) as x i = x i (λ), which passes through point
p at λ = 0. The function f (x) evaluated on this curve becomes f (x(λ)). If we
associate the tangent to this curve evaluated at λ = 0 with the vector X through

d f (x(λ)) ∂ f d x i
= = X[ f ], (3.20)
dλ λ = 0 ∂ x i dλ λ = 0
then the components in question are

d x i
a =
i
. (3.21)
dλ λ = 0
Partial derivatives commute with one another, that is,

∂ ∂f ∂ ∂f
− = 0. (3.22)
∂ xi ∂ x j ∂ x j ∂ xi
When all of the basis vectors in a tangent space commute with one another, the
basis is called a coordinate basis for T p (M). For example, the Euclidean basis
vectors (êx , ê y , êz ), used in physics when we use the coordinates (x, y, z) to de-
scribe space, correspond to the basis (∂x , ∂ y , ∂z ). These derivatives all commute
with one another, so this basis is a coordinate basis for T p (E3 ). As an example of
a set of basis vectors that do not commute, consider the orthonormal basis

∂ 1 ∂ 1 ∂
(êr , êθ , êφ ) = , , (3.23)
∂r r ∂θ r sin θ ∂φ
used with spherical coordinates (r, θ, φ) in E3 . Commuting the first two of these,
we get

∂ 1 ∂f 1 ∂ ∂f 1 ∂f
− =− 2 = 0. (3.24)
∂r r ∂θ r ∂θ ∂r r ∂θ
The basis vectors (3.23) do not commute, so they form what is called a non-
coordinate basis for T p (E3 ).
In flat spacetime in four dimensions, with coordinates (τ, x, y, z), the coordi-
nate basis for T p (M4 ) is (∂τ , ∂x , ∂ y , ∂z ). But it’s not the act of adding an extra
coordinate and calling it time that makes spacetime different from space. The dif-
ference between geometry in space and in spacetime has to do with the metric. To
understand the metric, we also need to look at the other geometrical objects that
can be defined on a manifold in addition to vectors.
The inner product and the metric

In Newtonian physics we normally employ the inner product as an operation be-
tween two vectors that yields a real number: v · u ∈ R. This is not strictly correct.
In orthodox differential geometry, the inner product v, ω̄
∈ R takes as arguments
a vector v in the tangent space T p (M), and an element ω̄ of a dual space to the
tangent space called the cotangent space, or T p∗ (M).
A basis {ēi } for T p∗ (M) that is dual to a basis {êi } for T p (M) can be constructed
from the relation
ê j , ēi
= δji . (3.25)
If we’re using a coordinate basis for T p (M), then the basis vectors {êi } are the
derivatives {∂i } with respect to the local coordinates {x i }. The corresponding dual
basis for the cotangent space consists of differentials {d x i } of the local coordinates,
which we can see from the relation

∂ ∂xi
, d x i
= = δji . (3.26)
∂x j ∂x j
Here we are using d x i as a basis one form in the cotangent space. We also use the
notation d x i in this book to represent an infinitesimal change in the coordinate x i ,
so the reader should be mindful of the context in which the object d x i is being
employed.
The geometrical objects that live in the cotangent space are called forms, more
specifically, one forms. Given a basis {d x i } of one forms, one can expand any one
form ω̄ ∈ T p∗ (M) in components as
ω̄ = ωi d x i . (3.27)
Combining (3.27) with (3.26) gives us the inner product of a vector v = v i ∂i and
a form ω̄ in terms of their components
v, ω̄
= v i ωi . (3.28)
We want a geometrical object that maps two vectors in T p (M) into a real num-
ber. Such an object is called the metric tensor. A tensor T is a generalization of a
vector that we will discuss in greater detail later. The metric tensor g takes two vec-
tors as arguments, and produces a real number. So it is a map from two copies of
the tangent space g : T p (M) ⊗ T p (M) → R. In a coordinate basis, using (3.26)
and (3.27) we can write
g = gi j d x i ⊗ d x j , (3.29)
where here d x i means the one form and not the infinitesimal line element. The
components of the metric tensor are

∂ ∂
gi j = g , . (3.30)
∂xi ∂x j
The metric product of two vectors u and v can then be written
g(u, v) = gi j u i v j . (3.31)
It is common to call the metric product the inner product. We will call it the
scalar product, because its value is a scalar quantity. The scalar product is related
to the inner product because the metric gives us a way to associate components of
vectors in the tangent space with components of forms in the cotangent space. If
we define the operation of lowering an index on a component as
u i = gi j u j , (3.32)
then the component u i with the lowered index could be thought of as a component
of a form in the cotangent space. The scalar product of two vectors can then be
written in terms of components as
g(u, v) = gi j u i v j = u i v i . (3.33)
Note that this is identical to the result we get from the inner product
v, ω̄u
(3.34)
using the one form
ω̄u = u i d x i . (3.35)
This means we can make the assignment
g(u, ) = ω̄u = u i d x i , (3.36)
making ω̄u the one form associated with the vector u through the metric g.
The operation (3.32) is invertible using the inverse metric, which in a coordinate
basis has components that satisfy the equation
g i j g jk = δki . (3.37)
The inverse metric gives us a vector vω associated with a one form ω̄

∂
vω = ω i , ωi = g i j ω j . (3.38)
∂xi
How is the scalar product of two vectors in the tangent space related to the
infinitesimal line element introduced in Chapter 1? A vector δs,
∂
δs = d x i , (3.39)
∂xi
representing an infinitesimal change in coordinates {x i } (where in this case d x i
is not a basis one form, but an infinitesimal change in the x i coordinate) has the
scalar product
ds 2 = g(δs, δs) = gi j d x i d x j . (3.40)
So the metric operating on vectors in the tangent space also gives the line element
on the manifold.
Vectors and coordinate transformations

It was common in the past for physics books to define a vector by how it behaves
under a change of coordinates
x̃ i = x̃ i (x 1 , x 2 , . . . , x D ). (3.41)
In that view, a vector v is represented by v i , denoting the set of components of v

in a coordinate basis. If an object v i transforms under (3.41) as
∂ x̃ i j
ṽ i = v , (3.42)
∂x j
then v i is a vector.
In the modern view, the vector v, expanded in a coordinate basis as
∂
v = vi , (3.43)
∂xi
is a fundamental geometrical object that remains unchanged by a change in
coordinates. The components {v i } of the vector change according to (3.42), and
the basis vectors transform via the inverse

∂ ∂x j ∂
= , (3.44)
∂ x̃ i ∂ x̃ i ∂ x j
so that
i j
∂ ∂ x̃ k ∂x ∂
ṽ = ṽ i
= v
∂ x̃ i ∂xk ∂ x̃ i ∂ x j
i
∂ x̃ ∂ x j k ∂ j ∂
= v = δk v k
∂ x k ∂ x̃ i ∂x j ∂x j
∂
= vj = v. (3.45)
∂x j
Performing a similar operation on a one form ω̄ shows that the basis one forms
transform as
∂ x̃ i
d x̃ i = dx j, (3.46)
∂x j
while the components {ωi } transform as
∂x j
ω̃i = ωj, (3.47)
∂ x̃ i
and the object ω̄ stays the same.
The action of a coordinate transformation on the components of the metric ten-
sor can be deduced from
g̃i j d x̃ i ⊗ d x̃ j = gi j d x i ⊗ d x j

∂xi ∂x j
= gi j k l d x̃ k ⊗ d x̃ l . (3.48)
∂ x̃ ∂ x̃
The scalar product of two vectors is a coordinate invariant object, because the
coordinate transformation of the components of the metric cancel the transforma-
tion of the components of the two vectors, as
i
∂ xk ∂ xl ∂ x̃ ∂ x̃ j m n
ṽ2 = g̃i j ṽ i ṽ j = gkl v v
∂ x̃ i ∂ x̃ j ∂xm ∂xn
k
∂ x ∂ x̃ i ∂ x l ∂ x̃ j
= gkl v m v n
∂ x̃ i ∂ x m ∂ x̃ j ∂ x n
= δm δn gkl v m v n = gkl v k vl = v2 .
k l
(3.49)
This is what it means for the metric tensor g to be a map T p M ⊗ T p M → R.

Coordinate invariant combinations of vectors, forms and tensors are also called
Vectors in spacetime 75
scalars, because what they yield is just a real number, which is the same when
evaluated in any coordinate system.
Some classes of coordinate transformations leave the metric invariant because
they represent symmetries of the spacetime. For example, Euclidean space in any
dimension is the same in all directions, at every point. It is isotropic (same in all
directions around some point) and homogeneous (the same at every point in some
given direction). Isotropy means the Euclidean metric is invariant under rotations
of the coordinate system. In D = 2 this takes the form
x̃ = cos θ x − sin θ y
ỹ = − sin θ x + cos θ y, (3.50)
where θ is a constant. Homogeneity means the Euclidean metric is invariant under
translations of the coordinates by constants
x̃ i = x i + ci . (3.51)
In both cases, the line element is the same in the new coordinates as it is in the old
coordinates
d x̃ 2 + d ỹ 2 = d x 2 + dy 2 , (3.52)
so the metric components in this basis are the same, g̃i j = gi j .
A coordinate transformation that leaves the metric unchanged is called an isom-
etry. Isometries in spacetime give rise to conserved quantities in physics, as we
shall see later in this book.
3.3 Vectors in spacetime

In Newtonian physics, the motion of an object is described by a vector in T p (E3 ),
with time appearing as the parameter of evolution in the equations of motion. In
relativity, both special and general, time and space are unified into spacetime, so
that time has to be accounted for in the geometrical structures in which objects are
represented, not just as a parameter in the equations of motion for the objects. In
the geometry of spacetime, time is a direction in which a vector can point.
The standard way to represent this in terms of components and bases is to la-
bel the timelike direction by the index 0 with basis vector ∂0 = ∂τ , and retain the
index {i | 1 ≤ i ≤ D} for the standard coordinate basis for T p (E D ) with basis vec-
tors {∂i = ∂x i }. The dimension of spacetime is then d = D + 1. Using these con-
ventions, a vector is expanded in T p (Md ) as
∂ ∂ ∂
µ
v = vµ
= vτ + vi i , (3.53)
∂x ∂τ ∂x
where the index i is assumed to be summed over all D space dimensions.
A Lorentz boost L(β) with velocity component β i in the i direction of frame

S with coordinates (x 0 , x 1 , . . . , x D ) to frame S̃ with coordinates (x̃ 0 , x̃ 1 , . . . , x̃ D )
has components
L 0̃0 = γ
L i0̃ = −γβ i
L ĩ0 = −γβ i
βi β j
L ĩj = (γ − 1) + δi j , (3.54)
β2
with the inverse transformation obtained by sending β i → −β i . The components

of v in the new basis are then
ṽ µ = L µ̃ ν
ν v . (3.55)
The metric tensor g of flat spacetime has coordinate basis components gµν =
ηµν , where
η00 = −1
η0i = ηi0 = 0
ηi j = δi j . (3.56)
The metric serves as a map between the tangent space T p (Md ) and the cotan-
gent space T p∗ (Md ). In a coordinate basis this means index raising and lowering
via
vµ = ηµν v ν
ωµ = ηµν ων . (3.57)
The scalar product u · v = g(u, v) of two vectors u and v is

D
µ ν
u · v = ηµν v u = −v u + 0 0
vi u i . (3.58)
i =1
Because a Lorentz transformation represents an isometry of the spacetime, a

Lorentz transformation of this product yields

D
D
ηµν v µ u ν = −v 0 u 0 + v i u i = −ṽ 0 ũ 0 + ṽ i ũ i . (3.59)
i =1 i =1
The minus sign in the metric presents us with three options for the metric prod-
uct of a vector with itself:

D < 0 timelike
µ ν
v = ηµν v v = −(v ) +
2 0 2
(v )
i 2
= 0 null (3.60)

i =1 > 0 spacelike.
Since v2 is a scalar and hence the same for all observers, a timelike vector is
timelike, a null vector is null, and a spacelike vector is spacelike, in all coordinate
systems and in all inertial frames.
Timelike vectors
A rotation in space of the Euclidean coordinate axes can change the direction in
space in which a vector points, so that a vector pointing in the +x direction be-
comes a vector pointing in the −x direction. But a Lorentz transformation in space-
time cannot change the direction in time in which a timelike vector points, so that
a vector pointing into the future becomes a vector pointing into the past.
Let’s assume that the direction of increasing coordinate τ is the future. With that
convention, a timelike vector
∂ ∂
v = v0 + vi i , v2 < 0 (3.61)
∂τ ∂x
we will call future-pointing if v 0 > 0, and past-pointing if v 0 < 0. A future-
pointing timelike vector cannot be transformed by a continuous Lorentz transfor-
mation into a past-pointing timelike vector. Consider the action of a Lorentz boost
with spatial velocity β of the time component of v
ṽ 0 = L 0̃0 v 0 + L i0̃ v i
= γ v 0 − γ δi j β i v j
= γ v 0 − γ β · v. (3.62)
The condition for ṽ 0 ≤ 0 is
v 0 ≤ β · v, (3.63)
which when squared gives us the inequality
(v 0 )2 ≤ (β · v)2 ≤ |β|

2 |
v |2 , (3.64)
where the last term comes from the Schwarz inequality in flat space
(
v · u)2 ≤ (
v · v) (
u · u). (3.65)
The timelike condition v2 < 0 tells us that

(v 0 )2 > |
v |2 . (3.66)
< 1,
Since |β|
2 |
|β| v |2 < |
v |2 , (3.67)
and so the inequality (3.63) cannot be satisfied if v is a timelike vector.
So there is no inertial frame in which a future-pointing timelike vector points
to the past, or a past-pointing timelike vector points to the future. The integrated
mathematical structure of Minkowski spacetime conspires to keep the past and
future separated in a way that preserves the notion of a direction in time for a
timelike vector.
The space components of a timelike vector are another matter. A timelike vector
can be transformed to point only in the time direction. The space components {v i }
of a timelike vector v can always be transformed away by a Lorentz boost at some
Proof of this will be left for the reader.
velocity β.
Given a timelike vector v, let’s study the set of vectors u ∈ T p (Md ) that are
orthogonal to v, so that u · v = 0. The coordinate basis components of this vector
satisfies
u 0 v 0 = u · v, (3.68)
which squares to
(u 0 )2 (v 0 )2 = (
u · v)2 , (3.69)
which by the Schwarz inequality satisfies
(u 0 )2 (v 0 )2 = (
u · v)2 ≤ |
u |2 |
v |2 , (3.70)
or
(u 0 )2 (v 0 )2
≤ 1. (3.71)
|
u |2 |
v |2
The vector v is timelike, so (v 0 )2 > |
v |2 . The inequality (3.71) can only be satisfied
if (u ) < |
0 2 u | , in other words, only if u is spacelike. Only a spacelike vector can
2
be orthogonal to a timelike vector. A timelike or null vector cannot be orthogonal

to a timelike vector.
A corollary to the above result is that the scalar product u · v of two timelike
vectors satisfies u · v < 0 if the vectors are pointing the same direction in time
(either both to the future or both to the past) and satisfies u · v > 0 if the two
vectors are pointing the opposite direction in time. Proof of this will be left for the
reader.
Velocity and momentum in spacetime

The tangent space T p (M) is constructed using tangent vectors to curves on the
manifold M passing through the point p. So every vector v ∈ T p (M) must be
tangent to some curve (more precisely, some family of curves) on the manifold. If
M is flat spacetime in d dimensions, then there exist vectors u that are timelike,
with g(u, u) < 0, and are hence tangent to curves representing the world lines of
objects traveling through time.
The world line of an object traveling through time in Md can be represented by
the curve C (λ) = (τ (λ), x 1 (λ), . . . , x D (λ)), with D = d − 1, where the parameter
λ is the proper time along world line. At each point on this curve the Lorentz-
invariant line element is

−dλ2 = −dτ 2 + (d x i )2 . (3.72)
Dividing both sides by dλ gives
2 i 2
dτ dx
− + = −1. (3.73)
dλ dλ
We can also write this as
u2 = u µ u µ = ηµν u µ u ν = −1, (3.74)
where
∂ dxµ
u = uµ
, u µ
= (3.75)
∂xµ dλ
is the spacetime vector tangent to the curve C (λ).
For d = 4, which matches the world we live in as far as we can detect experi-
mentally, u is called the four-velocity. For general d, we will call it the spacetime
velocity. The space components {u i } of the spacetime velocity u are related to the
usual Newtonian space velocity d x i /dt by
dxi d x i dτ
= . (3.76)
dλ dτ dλ
Since λ is the proper time along the curve, then according to relativistic time dila-
tion it must be true that dτ = γ dλ. The spacetime velocity vector is then revealed
to be
∂ ∂
u=γ + γβ i i , (3.77)
∂τ ∂x
where
dxi 1 dxi
βi = = . (3.78)
dτ c dt
If the curve C (λ) is a path of a particle or object in spacetime, then the tangent
vector must represent a spacetime generalization of velocity. Normally in Newto-
nian physics the momentum is p = m v. If we generalize this to flat spacetime in
d dimensions, then we should write
p = mu, (3.79)
in which case (3.74) tells us that
p2 = ηµν p µ p µ = −( p 0 )2 + | p |2 = −m 2 . (3.80)
Technically speaking, momentum is a one form, and properly lives in the cotangent
space T p∗ (Md ). However, because the spacetime metric provides an isomorphism
between T p (Md ) and T p∗ (Md ), it’s usually okay to treat momentum as a vector.
Note that we have absorbed the speed of light into the coordinate system by
using τ = ct as a time coordinate with units of length, so that we can write the
time–time component of the Minkowski metric as η00 = ητ τ = −1 rather than
η00 = ηtt = −c2 , with the inverse being η00 = ητ τ = −1 rather than η00 = ηtt =
−1/c2 . With this choice of coordinates, spacetime velocity is dimensionless, be-
cause the proper time λ comes in units of length as well. However, this convenient
choice for the metric components introduces an issue for the units of other physical
quantities, as we shall see below.
The time and space components of p are
p0 = γ m

p = γ m β. (3.81)
Expanding γ for small β gives

1 2 1 1 2
p ∼ m 1 + β + · · · = 2 mc + mv + · · ·
0 2
2 c 2
v
p ∼ m + · · · (3.82)
c
The time component p 0 of the spacetime momentum looks like the kinetic energy
of the object whose world line is C (λ), but there is the extra term mc2 to account
for. This term is present even if the world line is the world line of an object at
rest, with v = 0. This term is called the rest energy of the object in question. The
time component of the momentum is therefore the relativistic energy of the object,
with a contribution from the kinetic energy and a contribution from the mass of the
object at rest. This is what lies behind Einstein’s famous equation
E = mc2 , (3.83)
which is what we get in the limit β → 0 if we make the assignment p 0 = E/c2 =

γ m.
The problem is that getting the factors of c out of the metric components ends
up putting factors of c into the definitions of energy and space momentum. Rather
than deal with all the factors of c that arise, physicists usually adjust the unit system
as a whole by setting c = 1, in which case the condition p2 = −m 2 can be written
E 2 − | p |2 = m 2 . (3.84)
This equation is a relationship between the momentum and energy of a massive
object called the mass hyperboloid or mass shell. You will learn more about the
mass hyperboloid in Chapter 4 when we discuss particle scattering.
Lorentz boost of velocity

Consider a particle or object traveling in frame S with spacetime velocity u with
Minkowski coordinate components {γu , γu βu }, where
1
γu = . (3.85)

1 − |βu | 2
The components in some frame S̃ moving at velocity β (as measured in S) relative

to S are related to the original components through a Lorentz boost
ũ µ = L µ̃ ν
ν u . (3.86)
The time component of u transforms like
ũ 0 = L 0̃0 u 0 + L i0̃ u i
= γ u 0 − γ β · u
= γ γu (1 − β · βu ). (3.87)
Since ũ 0 = γ̃u , we see that the Lorentz boost rule for γu is
γ̃u = ũ 0 = γ γu (1 − β · βu )
(1 − β · βu )
= . (3.88)

1 − |βu | 1 − |β|
2 2
The space components transform in a more complicated manner, with
ũ i = L ĩ0 u 0 + L ĩj u j
β · u i
= u i − γβ i u 0 + (γ − 1) β. (3.89)
β2
In this form it’s hard to see that this is a Lorentz boost. For simplicity let’s work in
d = 3 with β = β êx , so that β · u = βu x . We then get
β · u
ũ x = u x − γβu 0 + (γ − 1) β
β2
= γ u x − γβu 0 = γ γu ((βu )x − β)
ũ y = u y = γu (βu ) y , (3.90)
which is the usual formula for a Lorentz boost in one dimension (here in the
x direction). The components of the transformed velocity become
ũ x (βu )x − β
(β̃u )x = =
ũ 0 1 − β · βu
ũ y (βu ) y
(β̃u ) y = 0 = . (3.91)
ũ γ (1 − β · βu )
Notice that although the relative motion between frames S and S̃ is constrained
to the x direction, the object’s velocity in the y direction is changed by the trans-
formation. In Galilean relativity, the components of velocity orthogonal to the rel-
ative motion between the frames are not changed, but in special relativity they are.
This is necessary for the speed of light to be preserved by the Lorentz transfor-
mation. The equations in (3.91) tell us that in the limit β → 1, (β̃u )x → −1 and
(β̃u ) y → 0. The component in the y direction ought to vanish if we’re boosting the
x direction by the speed of light, and the Lorentz transformation guarantees that it
does.
Null vectors
The limit m → 0 of (3.84) gives a null momentum vector with time component
p 0 = E = ±| p |. Null vectors are traditionally labeled by letters from the middle
of the alphabet, so let’s call this null vector k. A massless object has null momen-
tum, and so travels at the speed of light. A null vector is tangent to the world line of
an object traveling at the speed of light. But that world line can only be a straight
line, as will be left for the reader to prove as an exercise. So a null vector is a very
constrained object, unlike a timelike or a spacelike vector.
A null vector k is orthogonal to itself, because k · k = 0. Suppose there is some
other null vector l orthogonal to k. If k · l = 0, then
|l| = k · l.
|k| (3.92)
Tensors and forms 83
This is true if and only if l = α k, then since

where α is a constant. But if l = α k,

l = |l | and k = |k|, it is also true that l = αk. Therefore, two null vectors l and
0 0
k are orthogonal if and only if l is a constant multiple of k.

We have proven that a timelike vector v cannot be orthogonal to a null vector,
because the inequality (3.71) cannot be satisfied if (v 0 )2 > |v |2 and (u 0 )2 = |
u |2 .
But this inequality is satisfied automatically if v is a spacelike vector, so that
(v 0 )2 < |
v |2 . Therefore a null vector can be orthogonal to a spacelike vector.
Spacelike vectors
Spacelike vectors are tangent to curves that are not the world lines of objects trav-
eling in time. A timelike vector points in a definite direction in time, past or future,
but a spacelike vector can point to the past or future depending on the Lorentz
frame of the observer. The time component of a spacelike vector v can be got-
ten rid of entirely by a Lorentz boost at some velocity β. To get rid of the time
component, we need to satisfy the equation
v 0 = β · v. (3.93)
Squaring this equation and applying the Schwarz inequality tells us that
(v 0 )2
2 < 1,
≤ |β| (3.94)
|
v |2
which can be satisfied if and only if v is a spacelike vector.
3.4 Tensors and forms

From one forms to p forms
In (3.2) we learned that in addition to the tangent space T p (M) in which vectors
are defined on M there is also a dual space called the cotangent space T p∗ (M)
µ
in which objects called forms are defined. The inner product êν , ēµ
= δν shows
how to construct the dual basis for T p∗ (M) given the basis for T p (M). This dual
relationship means that a vector v can be seen as an object that operates on a one
form ω̄ ∈ T p∗ (M) and produces a real number v, ω̄
= v µ ωµ , and a one form ω̄
can be seen as an object that operates on a vector v ∈ T p (M) and produces a real
number v, ω̄
= v µ ωµ .
Because of this dual relationship, and the fact that a vector has a direction like an
arrow, a one form can be viewed as a kind of surface. The action of a one form on
a vector through the inner product v, ω̄
gives a real number that could be thought
of as the number of surfaces of ω̄ pierced by the arrow v. This provides a heuristic
explanation of why momentum is properly treated as a one form, rather than a

vector. The phase of a wave is given by the inner product x, k̄
, which yields a
number that we can think of as the number of surfaces of equal de Broglie wave
momentum k̄ pierced by the vector x.
There is an operation defined on one forms that is not defined on vectors, and
that is the exterior product (or wedge product). The exterior product is an antisym-
metric direct product. The exterior product of two one forms ᾱ and β̄ yields a two
form ω̄ defined as
ω̄ = ᾱ ∧ β̄ ≡ ᾱ ⊗ β̄ − β̄ ⊗ ᾱ. (3.95)
By this definition we can see that
ᾱ ∧ β̄ = −β̄ ∧ ᾱ. (3.96)
Therefore if β̄ = cᾱ, where c is a constant, then β̄ ∧ ᾱ = cᾱ ∧ ᾱ = 0.

In a spacetime coordinate basis, if ᾱ = αµ d x µ and β̄ = βµ d x µ , then the com-
ponents of the resulting two form ω̄ are given by
ω̄ = αµ βν (d x µ ⊗ d x ν − d x ν ⊗ d x µ )
= α µ βν d x µ ∧ d x ν
1
= (αµ βν − αν βµ ) d x µ ∧ d x ν
2
1
= ωµν d x µ ∧ d x ν . (3.97)
2
The components of a two form are antisymmetric, ωµν = −ωνµ . Note that the
most general two form is not a product of two one forms, but any two form ω̄ can
be expanded in a coordinate basis as
1
ω̄ = ωµν d x µ ∧ d x ν , (3.98)
2
with ωµν = −ωνµ . A two form that you will become deeply acquainted with in
Chapter 5 is the electromagnetic field strength F̄, known in spacetime component
notation by Fµν . Electric and magnetic field vectors in T p (E3 ) do not give rise
to electric and magnetic field vectors in T p (M4 ). The electric and magnetic fields
are instead components of the two form field strength F̄. This will be discussed in
much greater detail in Chapter 5.
We can keep using the antisymmetric direct product on the coordinate basis one
forms {d x µ } until we run out of coordinates. If we take an antisymmetric product
of p basis forms, then we get the basis for a p form, also called a form of degree
p, which can be expanded in this basis as

1
ω̄ = ωµ ... µ d x µ1 ∧ . . . ∧ d x µ p , (3.99)
p! 1 p
where the components ωµ1 ... µ p are antisymmetric under the exchange of any two
indices. The exterior product of a p form ᾱ p and a q form β̄q obeys the rule
ᾱ p ∧ β̄q = (−1) pq β̄q ∧ ᾱ p . (3.100)
Let’s count how many linearly independent p forms exist in d dimensions. Each
of the indices can take d values, but they must all be different in order that the dif-
ferential form not vanish. Moreover, changing the order of the indices can (at most)
give an overall sign change. Putting these facts together, it is clear that the number
of independent p forms in d dimensions is given by the binomial coefficient

d d!
= . (3.101)
p p!(d − p)!
As this formula indicates, we run out of coordinates in d dimensions when p > d,
because
dxµ ∧ dx0 ∧ dx1 ∧ . . . ∧ dx D = 0 (3.102)
for any value of µ. Forms of degree p do not exist on a manifold of dimension d
for p > d.
At the limit p = d, there is just one possible combination of basis forms,
¯ = d x 0 ∧ d x 1 ∧ . . . ∧ d x D , (3.103)
and this is called the spacetime volume form. The components are

1 (µ0 µ1 . . . µ D ) = even perm. of (0 1 . . . D)
µ0 µ1 ... µ D = −1 (µ0 µ1 . . . µ D ) = odd perm. of (0 1 . . . D)

0 otherwise. (3.104)
This d form is also called the Levi-Civita permutation symbol, or the Levi-Civita
tensor. A tensor is a generalization of vectors and forms that we will get to shortly.
The mathematical theory of permutations is discussed in Chapter 8.
Lorentz transformation of forms

A general coordinate transformation x̃ µ = x̃ µ (x ν ) acts on coordinate basis one
forms d x µ as
∂ x̃ µ ν
d x̃ µ = dx . (3.105)
∂xν
As a geometric object, a one form ω̄ = ωµ d x µ exists independently of any basis

used for the cotangent space in which it lives, so the transformation of the compo-
nents has to cancel the transformation of the basis. Therefore, as discussed earlier,
the one form components ωµ transform as
∂xν
ω̃µ = ων . (3.106)
∂ x̃ µ
We have seen that one forms and vectors transform differently under coordinate
transformations. This is true, in particular, for a coordinate transformation that
is a Lorentz transformation from frame S to frame S̃ moving at velocity β with
µ̃
respect to the S frame. Instead of the Lorentz components L ν , which appear in the
transformation law of a vector, such as the velocity vector in Eq. (3.86), we need
to use the inverse components L νµ̃ ,
ω̃µ = L νµ̃ ων . (3.107)
The inverse components of a Lorentz boost L νµ̃ are
L 00̃ = γ
L i = γβ i
0̃
0
L ĩ = γβ i
βi β j
L ij̃ = (γ − 1) + δi j , (3.108)
β2
as can be verified by matrix multiplication. The components ωµ1 µ2 ···µ p of a p form
ω̄ transform as
ν
ω̃µ1 µ2 ...µ p = L νµ˜11 L νµ˜22 . . . L µ˜pp ων1 ν2 ...ν p , (3.109)
where, as usual, all pairs of repeated upper and lower indices are to be summed
over all spacetime dimensions.
What is a tensor?
We can take as many copies as we want of the tangent space T p (M) and
the cotangent space T p∗ (M), take the direct product T p (M) ⊗ . . . ⊗ T p (M) ⊗
T p∗ (M) . . . ⊗ T p∗ (M), and use this as a space for defining the bases of geometrical
objects of the manifold M. Such objects are called tensors. If we take a direct
product of m copies of the tangent space T p (M) and n copies of the cotangent
space T p∗ (M), then we have a space for defining what is called a rank (m, n)
tensor. A vector is a rank (1, 0) tensor and a one form is a rank (0, 1) tensor. A
scalar, that is, a number, can be considered to be a rank (0, 0) tensor.
A rank (m, n) tensor T can be expanded in a coordinate basis as
∂ ∂
T = Tνµ1 ···ν
1 ···µm
µ
⊗ · · · ⊗ µ ⊗ d x ν1 ⊗ · · · ⊗ d x νn . (3.110)
n
∂x 1 ∂x m
We learned previously that a one form ω̄ = ωµ d x µ ∈ T p∗ (M) can be thought of
as a map that operates on a vector v = v µ ∂µ ∈ T p (M) to produce a coordinate
invariant scalar v µ ωµ ∈ R. A rank (m, n) tensor T can be thought of as a map
(T p∗ (M))m ⊗ (T p (M))n → R, taking as arguments m one forms and n vectors to
produce a coordinate invariant scalar
T(ω̄1 , . . . , ω̄m , v1 , . . . , vn ) = Tνµ1 ...ν
1 ...µm ω
n
ν1 νn
µ1 . . . ωµm v . . . v . (3.111)
When there is a metric tensor defined on the manifold, there is an isomorphism
between the tangent space and the cotangent space at each point, which is ex-
pressed in a coordinate basis through the operation of index raising and lower-
ing. This operation extends naturally from vectors and forms to tensors of any
rank. The metric tensor used in this way can turn a rank (m, n) tensor into a rank
(m − 1, n + 1) tensor by lowering one of the lower indices. For example,
T µν κ = gκλ T µνλ . (3.112)
The inverse metric operates on a rank (m, n) tensor to produce a rank (m + 1,
n − 1) tensor by raising an index, for example
T µνκ = g κλ T µν λ . (3.113)
The metric tensor operates on two vectors to give a coordinate invariant scalar,
so it is a map T p (M) ⊗ T p (M) → R. Therefore the metric tensor can operate on
a rank (m, n) tensor to produce a rank (m − 2, n) tensor, for example
T κ = gµν T µνκ . (3.114)
The inverse metric yields a rank (m, n − 2) tensor, for example
T κ = g µν T κ µν . (3.115)
The study of tensors in full generality is a big subject. In physics we are usually
only concerned with certain types of tensors. In spacetime physics, we’re con-
cerned with the behavior under Lorentz transformations. A general tensor can be
reduced into parts, each of which transforms into itself under a Lorentz transfor-
mation. When we can no longer reduce the tensor any further, we say that the
individual parts are irreducible tensors. Part of this story involves symmetry and
antisymmetry, as will be explained below.
When we discussed p forms, we defined the exterior product of two one forms
as their antisymmetric direct product. This is meaningful as a definition because
the direct product T p∗ (M) ⊗ T p∗ (M) can be divided into antisymmetric and sym-
metric subspaces, and this division is coordinate-independent. The antisymmetric
subspace of T p∗ (M) ⊗ T p∗ (M) is spanned in a coordinate basis by
d x µ ∧ d x ν = d x µ ⊗ d x ν − d x ν ⊗ d x µ. (3.116)
The components of a two form, or antisymmetric tensor of rank (0, 2) are anti-
symmetric under exchange of indices, so that Tµν = −Tνµ . Under a coordinate
transformation, an antisymmetric tensor remains antisymmetric, so this subspace
of the direct product space transforms into itself.
There is also a symmetric subspace of T p∗ (M) ⊗ T p∗ (M), spanned in a coordi-
nate basis by
d x µ ⊗ d x ν + d x ν ⊗ d x µ. (3.117)
The components of a tensor defined in this subspace are then symmetric under
exchange of indices so that Tµν = Tνµ . Under a Lorentz transformation, a sym-
metric tensor remains symmetric, so this subspace of the direct product space also
transforms into itself.
If we look at tensors as represented by their components in a coordinate basis,
then any general rank (0, 2) tensor Tµν is the sum of its symmetric and antisym-
metric parts.
Tµν = T(µν) + T[µν]
1
T(µν) ≡ (Tµν + Tνµ ) symmetric
2
1
T[µν] ≡ (Tµν − Tνµ ) antisymmetric. (3.118)
2
But this is not yet the full reduction of the tensor into its irreducible parts. The
trace T = Tµν g µν of a rank (0, 2) tensor is a real number, a coordinate invariant,
and hence trivially transforms into itself under a coordinate transformation. The
symmetric part of a rank (0, 2) tensor is the sum
1
T(µν) = T{µν} + gµν T
d
1 1
T{µν} ≡ (Tµν + Tνµ ) − gµν T traceless symmetric
2 d
κλ
T ≡ g Tκλ trace (3.119)
of the traceless symmetric part and the trace.
A symmetric tensor has zero antisymmetric part, and an antisymmetric tensor
has zero symmetric part. The electromagnetic field strength tensor F is the primary
The Principle of Relativity as a geometric principle 89
example of an antisymmetric tensor of rank (0, 2) used in physics, while the metric
tensor g is the symmetric (0, 2) tensor with which physicists and mathematicians
tend to be the most familiar.
The symmetrization and anti-symmetrization process can be extended to ten-
sors of any rank (m, n). A tensor can be symmetric in some pairs of indices and
antisymmetric in others, or symmetric or antisymmetric in all pairs of indices.3
Examples will be left to the reader as an exercise.
Lorentz transformation of tensors

Once you learn how to make a Lorentz transformation of the components of a
vector
ṽ µ = L µ̃ ν
ν v , (3.120)
and a one form,
ω̃µ = L νµ̃ ων , (3.121)
transforming the components of a general tensor of rank (m, n) is simple. You just
apply vector transformations on the m upper indices, and apply one form transfor-
mations on the n lower indices to get
1 ... µm = L µ˜1 . . . L µ˜m L λ1 . . . L λn T κ1 ... κm .
T̃νµ1 ... (3.122)
νn κ1 κm ν˜1 ν˜n λ1 ... λn
3.5 The Principle of Relativity as a geometric principle

Einstein’s two postulates don’t say anything about a unified spacetime, manifolds,
tangent spaces, metrics or tensors. Einstein made two simple, but powerful, pro-
posals:
(i) All physical laws valid in one frame of reference are equally valid in any other frame
of reference moving uniformly relative to the first.
(ii) The speed of light (in a vacuum) is the same in all inertial frames of reference, regard-
less of the motion of the light source.
The rich geometric structure of flat spacetime grows out of those two simple
postulates if we follow them to their logical conclusions. In order for all physical
laws to be equally valid in frames of reference moving relative to one another, the
physical laws have to be expressed in a form that allows such a transformation to
3 There are also more subtle permutation symmetries, which are neither symmetric nor antisymmetric. They are
best analyzed using the mathematical theory of the symmetric group.
be defined, and that structure is a manifold with a tangent space and a cotangent
space.
In order for the speed of light to be the same in all inertial frames of reference,
space and time cannot be absolute and independent. Relativity of simultaneity,
time dilation and length contraction are what we get when we follow the sec-
ond postulate to its logical conclusions. The invariance of the speed of light leads
us to a unified picture of spacetime. The spacetime coordinate transformations
that leave invariant the speed of light are Lorentz transformations and spacetime
translations.
In this chapter we surveyed the fundamentals of geometry in flat spacetime. In
Chapter 4 we will put this geometry to work when we examine relativistic me-
chanics in flat spacetime.
Exercises
3.1 The following pairs of numbers represent the (τ, x) coordinates of events
in a spacetime of two dimensions. Using grid paper or your favorite
plotting software, plot these events on a spacetime diagram similar to
Figure 3.2. Using the invariant interval between each set of events, de-
termine which events have timelike, null or spacelike separations from the
other events.
E 0 = (0, 0)
E 1 = (1, 3)
E 2 = (−2, 5)
E 3 = (3, 0)
E 4 = (1, −3).
On the same grid, draw a pair of straight lines with slope ±1 that intersect
at event E 0 . In what way does this pair of lines relate to the sign of the
invariant interval between some other event and E 0 ? What does this pair
of lines represent?
3.2 Suppose that the S̃ frame with coordinates (τ̃ , x̃) is moving at velocity
β relative to the S frame with coordinates (τ, x), with the events (0, 0)
coinciding in both frames. On the same grid as in the previous exercise,
draw the (τ̃ , x̃) axes for β = 1/5, 1/2, 4/5.
3.3 On a (τ, x) coordinate grid, plot the curves
−τ 2 + x 2 = n 2 (E3.1)
Exercises 91
for values n = 0, 1, 2, 3, over the region between τ = ±10 and x = ±10.

Are these null, timelike or spacelike curves? How would these curves look
in coordinates (τ̃ , x̃), where (τ̃ , x̃) and (τ, x) are related by a Lorentz
transformation at velocity β ?
3.4 On a (τ, x) coordinate grid, plot the curves
−τ 2 + x 2 = −n 2 (E3.2)
for values n = 0, 1, 2, 3, over the region between τ = ±10 and x = ±10.

Are these null, timelike or spacelike curves? How would these curves look
in coordinates (τ̃ , x̃), where (τ̃ , x̃) and (τ, x) are related by a Lorentz
transformation at velocity β ?
3.5 In three spacetime dimensions (d = 3) with coordinates (t, x, y), the
Minkowski metric can be written
ds 2 = −dτ 2 + d x 2 + dy 2 . (E3.3)
(a) Rewrite this metric using coordinates (u, v, y), where u = τ + x and v =
τ − x.
(b) Forgetting about the y direction, on a piece of grid paper or using your favorite
computer software, plot the (u, v) axes in (τ, x) coordinates.
(c) What type of world line is represented by a line of constant u or v?
Coordinates such as (u, v) are known as null coordinates, or light-cone
coordinates. If one wants to learn string theory, it is a good idea to become
familiar with light cone coordinates.
3.6 Consider the pole in the barn scenario discussed in this chapter. Prove that
in the case where the pole and barn have the same proper length L 0 , the
events E 2 and E 3 always have a spacelike separation. Suppose the barn
and pole have different proper lengths. Under what conditions, if any, can
the separation between events E 2 and E 3 be timelike? In such a case, is
there any contradiction in the time ordering between the two events in the
pole frame and the barn frame?
3.7 Consider a thin pole of proper length L 0 at rest in the S frame, with one
end at x = 0, y = 0 and the other end at x = L 0 , y = 0. Suppose we are
looking at this pole in a universe with three spacetime dimensions with
coordinates (τ, x, y).
(a) Find the equation for the world line of each end of the pole in frame S.
(b) Consider an observer in frame S̃ with coordinates (τ̃ , x̃, ỹ) moving at velocity
β x = 0, β y = β relative to frame S. Using (3.54), compute the equations for
the world lines of the ends of the pole in terms of (τ̃ , x̃, ỹ). What is the length
of the pole according to the observer in frame S̃?
√
(c) Suppose frame S̃ moves instead with velocity β x = β y = β/ 2. Us-
ing (3.54), compute the equations for the world lines of the ends of the pole
in terms of (τ̃ , x̃, ỹ) in this case, and compare with your answer above. What
is the length of the pole according to the observer in frame S̃?
3.8 Using (3.54), write the Lorentz transformation for four spacetime dimen-
sions (d = 4) as a 4 × 4 matrix, and calculate the determinant.
3.9 Consider the pole in the barn scenario discussed in this chapter. Imagine
a pole with proper length L 0 = 10 m heading towards a barn with proper
length L 0 = 10 m at speed β = 4/5. Let’s call the frame in which the barn
is at rest S with coordinates (τ, x), and label the frame in which the pole
is at rest by S̃ with coordinates (τ̃ , x̃). Suppose that the leading edge of
the pole passes the front door of the barn at time τ = τ̃ = 0. Find the total
amount of time τi that the pole is completely inside the barn. What do
you learn when you try to calculate the corresponding interval τ̃i ?
3.10 Picture a rigid pole at rest according to an observer in frame S with space-
time coordinates (τ, x). One end is at x = 0 and the other is at x = L 0 .
(a) Draw the world sheet of the pole between τ = 0 and τ = L.
(b) On the same plot, draw a line representing the path of a flash of light at τ =
x = 0 aimed along the pole in the +x direction.
(c) Suppose the end of the pole at x = 0 is sharply tapped at τ = 0. On your plot,
identify the set of events on the world sheet of the pole that could possibly
be influenced by the tap at the end of the pole, according to special relativity.
Identify the set of events on the world sheet of the pole that could not possibly
be influenced by the tap at the end of the pole, according to special relativity.
(d) Suppose the tap at the end x = 0 is forceful enough to make the pole move at
speed β = 1/5. On a new plot, again using the (τ, x) coordinate system, draw
a possible world sheet for the accelerating pole from time τ = 0 to τ = 2L 0 .
Make sure that this world sheet is consistent with special relativity.
(e) What conclusion would you draw from this exercise about the nature of a
rigid body in spacetime?
3.11 The metric for flat space in three dimensions can be written using Eu-
clidean coordinates (x, y, z) as
ds 2 = d x 2 + dy 2 + dz 2 . (E3.4)
Rewrite this metric in (u, v, w) coordinates and describe the curves of

constant u, v and w in each case given below
(a) Cylindrical coordinates
x = u cos w
y = u sin w
z = v, (E3.5)
Exercises 93
(b) Spherical coordinates
x = u cos w sin v
y = u sin w sin v
z = u cos v. (E3.6)
3.12 Consider two flat space dimensions, with Euclidean coordinates (x, y).
Let’s write the metric in tensor form as
g = d x ⊗ d x + dy ⊗ dy. (E3.7)
(a) Rewrite this metric in polar coordinates x = r cos φ and y = r sin φ.
(b) Write the basis vectors (êr , êφ ) = (∂/∂r , ∂/∂φ) in terms of the basis vectors
(êx , ê y ) = (∂/∂ x, ∂/∂ y). Is the basis (∂/∂r , ∂/∂φ) a coordinate basis?
(c) If some set of basis vectors êi is an orthonormal basis, then g(êi , ê j ) = δi j . Is
the basis (∂/∂r , ∂/∂φ) an orthonormal basis?
∂
(d) Is the basis (∂/∂r , r1 ∂φ ) an orthonormal basis? Is it a coordinate basis?
(e) Is the basis (∂/∂ x, ∂/∂ y) an orthonormal basis? Is it a coordinate basis?
3.13 Consider four flat spacetime dimensions, with Minkowski coordinates
(τ, x, y, z). Let’s write the metric in tensor form as
g = −dτ ⊗ dτ + d x ⊗ d x + dy ⊗ dy + dz ⊗ dz. (E3.8)
In Minkowski spacetime we define an orthonormal set of basis vectors êµ

by the requirement g(êµ , êν ) = ηµν .
(a) Rewrite this metric using null coordinates (u, v, y, z), where u = τ + x, v =
τ − x.
(b) Write the basis vectors (∂/∂u, ∂/∂v) in terms of the basis vectors
(∂/∂τ , ∂/∂ x). Is the basis (∂/∂u, ∂/∂v, ∂/∂ y, ∂/∂z) a coordinate basis? Is
(∂/∂u, ∂/∂v, ∂/∂ y, ∂/∂z) an orthonormal basis? (In other words, do all of
these basis vectors have unit norm and are they all orthogonal to each other?)
(c) Is the basis (∂/∂τ , ∂/∂ x, ∂/∂ y, ∂/∂z) a coordinate basis? Is it an orthonormal
basis?
3.14 Show that in flat spacetime, for any number of dimensions, the only curve
whose tangent vector is everywhere lightlike is a straight line.
3.15 Given the following one forms
α = αµ d x µ β = βµ d x µ
γ = γµ d x µ δ = δµ d x µ (E3.9)
where, as usual, the repeated Greek index implies a sum over all d
spacetime dimensions, compute α ∧ β, α ∧ β ∧ γ and α ∧ β ∧ γ ∧ δ for
d = 2, 3, 4.
3.16 Find the coordinate basis components of a rank ( 30 ) tensor symmetrized

over all pairs of indices.
3.17 Find the coordinate basis components of a rank ( 30 ) tensor antisym-
metrized over all pairs of indices.
3.18 Consider an antisymmetric rank ( 02 ) tensor Fµν in flat spacetime in d = 4
with coordinate basis components
F0i = −E i i = x, y, z
Fi j = i jk Bk , (E3.10)
where i jk is antisymmetric on all three indices, with x yz = 1. Calculate
the components of this tensor under a Lorentz transformation at speed β
in the x direction. This tensor is the electromagnetic field strength tensor,
also known as the Faraday tensor. The Lorentz transform of this tensor
gives the correct Lorentz transform of the electric and magnetic fields E
and B for the Maxwell equations to be the same in all inertial frames. You
will be seeing more of this tensor in Chapter 5.

Elements of Spacetime Geometry: Hands-On Exercise: Manifolds and Coordinate Patches

Uploaded by

Copyright:

Available Formats

Elements of Spacetime Geometry: Hands-On Exercise: Manifolds and Coordinate Patches

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Elements of Spacetime Geometry: Hands-On Exercise: Manifolds and Coordinate Patches

Uploaded by

Copyright:

Available Formats

3

Elements of spacetime geometry

Hands-on exercise: manifolds and coordinate patches

Build a model of E2 (two-dimensional Euclidean space) with two rulers taped

3.1 Space and spacetime

in this local neighborhood of p. However, this distance is a property of the particu-

dl 2 = dr 2 + r 2 dθ 2 + r 2 sin2 θdφ 2 , (3.7)

A flat spacetime is a spacetime manifold where one patch of Minkowski coor-

(τ − τ0 )2 = (x 1 − x01 )2 + (x 2 − x02 )2 + · · · + (x D − x0D )2 . (3.9)

This is called a null hypersurface of Md . Slices of constant τ on this null hypersur-

where λ2 > 0. Each possible value of λ defines a Lorentz-invariant submanifold

shows that each value of λ2 corresponds to a minimum possible value for

Since each value of λ2 corresponds to a minimum of (τ − τ0 )2 , then τ − τ0 can

The pole in the barn

particular direction, for example the x 1 direction, of the coordinates in Md looks

3.2 Vectors on a manifold

x(t) = x(t)êx + y(t)ê y + z(t)êz . (3.14)

of mass of the whole ensemble

Properties of a vector space

(i) The sum of two vectors is a vector: v + u ∈ V.

The set of assumptions that we make about scalar multiplication is:

(i) A vector multiplied by a scalar is a vector: av ∈ V.

Any vector in a vector space can be written as a linear combination of other

The tangent space

then their sum is

The inner product and the metric

using the one form

This means we can make the assignment

g(u, ) = ω̄u = u i d x i , (3.36)

The inverse metric gives us a vector vω associated with a one form ω̄

ds 2 = g(δs, δs) = gi j d x i d x j . (3.40)

Vectors and coordinate transformations

In that view, a vector v is represented by v i , denoting the set of components of v

the basis vectors transform via the inverse

This is what it means for the metric tensor g to be a map T p M ⊗ T p M → R.

3.3 Vectors in spacetime

A Lorentz boost L(β) with velocity component β i in the i direction of frame

with the inverse transformation obtained by sending β i → −β i . The components

The scalar product u · v = g(u, v) of two vectors u and v is

Because a Lorentz transformation represents an isometry of the spacetime, a

The condition for ṽ 0 ≤ 0 is

which when squared gives us the inequality

(v 0 )2 ≤ (β · v)2 ≤ |β|

The timelike condition v2 < 0 tells us that

be orthogonal to a timelike vector. A timelike or null vector cannot be orthogonal

Velocity and momentum in spacetime

in which case (3.74) tells us that

Expanding γ for small β gives

which is what we get in the limit β → 0 if we make the assignment p 0 = E/c2 =

Lorentz boost of velocity

The components in some frame S̃ moving at velocity β (as measured in S) relative

The space components transform in a more complicated manner, with

This is true if and only if l = α k,  then since

k are orthogonal if and only if l is a constant multiple of k.

3.4 Tensors and forms

explanation of why momentum is properly treated as a one form, rather than a

By this definition we can see that

ᾱ ∧ β̄ = −β̄ ∧ ᾱ. (3.96)

Therefore if β̄ = cᾱ, where c is a constant, then β̄ ∧ ᾱ = cᾱ ∧ ᾱ = 0.

x(t) = x(t)êx + y(t)ê y + z(t)êz . (3.14)

(v 0 )2 ≤ (β · v)2 ≤ |β|

The components in some frame S̃ moving at velocity β (as measured in S) relative

This is true if and only if l = α k, then since