General Relativity
Home
  • The Equivalence Principle
  • Differential Geometry
  • Spacetime
  • Geodesics
  • Metric for a Gravitational Field
  • Black Holes: Schwarzschild Geometry
  • Black Holes: Falling In and Formation
  • The Einstein Field Equations
  • Gravitational Waves
  • Riemannian Spaces and the Metric Tensor
  • Curves and Motion
  • Manifolds
  • Cosmology
  • Lensing and Redshift
  • Tests of Relativity
Home
  • The Equivalence Principle
  • Differential Geometry
  • Spacetime
  • Geodesics
  • Metric for a Gravitational Field
  • Black Holes: Schwarzschild Geometry
  • Black Holes: Falling In and Formation
  • The Einstein Field Equations
  • Gravitational Waves
  • Riemannian Spaces and the Metric Tensor
  • Curves and Motion
  • Manifolds
  • Cosmology
  • Lensing and Redshift
  • Tests of Relativity
  • The Equivalence Principle
  • Differential Geometry
  • Spacetime
  • Geodesics
  • Metric for a Gravitational Field
  • Black Holes: Schwarzschild Geometry
  • Black Holes: Falling In and Formation
  • The Einstein Field Equations
  • Gravitational Waves
  • Supporting Chapters

    • Riemannian Spaces and the Metric Tensor
    • Curves and Motion
    • Manifolds
    • Cosmology
    • Lensing and Redshift
    • Tests of Relativity

Differential Geometry

Differential geometry studies smooth spaces using calculus. It gives us a language for describing directions, measurements, and curvature without making the answer depend on the coordinates we choose. We will build that language step by step, beginning with familiar geometry and ending with the tensors used in general relativity.

What Is Analytic Geometry?

Analytic geometry, also called analytical geometry, describes geometric objects using coordinates and equations. Choosing Cartesian axes turns a point in a plane into a pair of numbers (x,y). A circle of radius R centered at the origin then becomes the equation

x2+y2=R2.

The circle is the geometric object; the equation describes it in a particular coordinate system. In polar coordinates, the same circle has the simpler equation r=R. Changing coordinates changes the description, not the circle.

Coordinates also let us use algebra to answer geometric questions. Intersections become simultaneous equations, and the distance between two points in a Euclidean plane is

d2=(x2​−x1​)2+(y2​−y1​)2.

This formula assumes Cartesian coordinates and Euclidean geometry. One of our central questions will be how to express measurements when those assumptions no longer hold.

Extrinsic and Intrinsic Geometry

Extrinsic geometry studies a space through the way it sits inside a larger space. We can picture a sphere as a surface embedded in three-dimensional Euclidean space, and describe its bending using the surrounding space.

Intrinsic geometry studies measurements made within the space itself. Imagine observers confined to a surface who can measure lengths, angles, and the paths of freely moving objects. They can investigate its geometry without referring to a direction pointing away from the surface.

A flat sheet rolled into a cylinder illustrates the distinction. Imagine rolling a sheet of aluminium foil around a tube without stretching, crumpling, or creasing it. Unroll the foil and it lies flat again: lines drawn on it retain their lengths, intersect at the same angles, and enclose the same areas. Rolling changed how the sheet sits in three-dimensional space, but it did not change measurements made within the sheet.

This is the sense in which a cylinder is intrinsically flat. Cut it along a straight line parallel to its axis and unroll it, and the surface becomes a rectangle without distortion. If u measures distance around the cylinder and v measures distance along its axis, an embedding of a cylinder of radius R is

X(u,v)=(RcosRu​, RsinRu​, v).

The squared distance between nearby points on its surface is

dℓ2=du2+dv2,

exactly the Euclidean plane's distance rule in Cartesian coordinates. Its Gaussian curvature is therefore zero. The cylinder still has extrinsic curvature because its surface normal changes as we move around it. It also differs globally from the plane: moving far enough in the u direction returns to the starting point.

A sphere cannot be treated this way. Its intrinsic distance rule is

dℓ2=R2(dθ2+sin2θdϕ2),

and its Gaussian curvature is 1/R2. Flattening a spherical shell cannot preserve all distances and angles. If the material cannot stretch, it must buckle or tear apart; cutting it into narrow strips lets the strips lie flatter only by opening gaps between them. Every flat map of Earth therefore distorts lengths, areas, angles, or some combination of them.

An unrolled sheet and a cylinder preserve the same intrinsic grid, while flattening a sphere requires gaps, tears, or stretching.

Flatness is thus an intrinsic claim about a metric, not a claim that an object looks visually flat in a surrounding space. On a flat surface, sufficiently small geometric experiments obey Euclidean relations, parallel transport around a contractible loop returns a vector unchanged, and the intrinsic curvature vanishes. On a sphere, a triangle bounded by arcs of great circles can have an angle sum greater than π, and parallel transport around a loop can rotate a vector. Observers confined to the sphere can detect these effects without looking outside it.

Parallel transport around two closed paths on a sphere: the loop enclosing the larger area returns the vector with a larger rotation.

For these geodesic loops on a sphere of radius R, the rotation angle is proportional to the enclosed area A: Δα=A/R2. The result depends on the path even though both journeys begin and end at the same point. This failure to return unchanged is called holonomy, and it provides an intrinsic test of curvature.

General relativity describes the intrinsic geometry of spacetime. It does not require spacetime to bend into an additional physical dimension.

What Is Differential Geometry?

Differential geometry combines smooth geometric spaces with differentiation and integration. Instead of asking only which points satisfy an equation, it asks how curves move through a space, what directions are available at a point, and how measurements change from point to point.

Calculus begins locally. A small segment of a smooth curve is approximated by its tangent line, and a small patch of a smooth surface is approximated by its tangent plane. Differential geometry makes these approximations precise and extends them to spaces of any dimension.

Different ingredients answer different questions. A smooth manifold lets us use coordinates and calculus locally. A metric supplies lengths and inner products. A connection lets us differentiate vector fields and compare directions along a path. Curvature then describes, among other things, how transporting a direction around a loop can change it.

We will first develop the objects defined at a single point. Comparing objects at different points comes later in Spacetime.

Why Study Differential Geometry in General Relativity?

In special relativity, spacetime has the flat Minkowski geometry. In general relativity, the spacetime metric is a field that responds to matter and energy. It determines the proper time measured by clocks, the light cones, and the geometry governing freely falling motion.

The equivalence principle motivates this description: in a sufficiently small freely falling laboratory, physics approaches its special-relativistic form. Across an extended region, however, neighboring freely falling objects can accelerate relative to one another. These tidal effects reveal spacetime curvature.

Coordinates are labels we choose for events. A prediction such as the elapsed time on a clock cannot depend on those labels. Differential geometry gives us quantities whose coordinate descriptions may change while their physical meaning remains fixed.

Throughout these notes, Greek indices label the four spacetime coordinates, x0=ct, and the spacetime signature is (+,−,−,−). When an index occurs once upstairs and once downstairs in a term, we sum over it. For example, ωμ​Vμ means ∑μ=03​ωμ​Vμ.

Manifold

An n-dimensional manifold is a space that locally looks like an open region of Rn. Around each point, we can label nearby points with n independent coordinates. The dimension counts the coordinates needed within the space, not those of any surrounding space used to draw it.

A circle is one-dimensional: near a point, one parameter locates another point on the circle. A sphere's surface is two-dimensional: two parameters locate a point on it. Spacetime is modeled as a four-dimensional manifold.

A coordinate map on a region is called a chart. A collection of charts covering the manifold is an atlas. Where charts overlap, their coordinates must be related by smooth, invertible transformations for the atlas to define a smooth structure.

No single chart need cover the entire space. Longitude and latitude, for example, fail to give a regular coordinate system at the poles. This is a failure of those coordinates, not a physical singularity of the sphere.

A smooth manifold by itself does not assign distances or angles. It gives us the setting for calculus; a metric will add a measurement rule.

Curves

A curve is a map from an interval of real numbers into a manifold:

γ:I⟶M,λ⟼γ(λ).

In a chart, this becomes a collection of coordinate functions xa(λ). The parameter λ labels points along the curve; it need not be a coordinate of the manifold or a physical time.

For example, a circle in the plane can be parametrized by

x(λ)=Rcosλ,y(λ)=Rsinλ.

Changing the parameter can describe the same path with a different rate of traversal. If λ=λ(σ), then

dσdxa​=dλdxa​dσdλ​.

Thus the traced path and the parametrized curve carry different information. A regular reparametrization preserves the path, while rescaling its tangent; a decreasing reparametrization also reverses its orientation.

Surfaces

A smooth surface is a two-dimensional manifold.[1] When embedded in three-dimensional space, a patch can be described by a map X(u,v) whose two partial derivatives are linearly independent.

For a sphere of radius R, a familiar parametrization is

X(θ,ϕ)=(Rsinθcosϕ, Rsinθsinϕ, Rcosθ).

The pair (θ,ϕ) supplies coordinates on a suitable patch of the sphere. The three components of X describe the embedding. Keeping ϕ fixed traces a meridian; keeping θ fixed traces a circle of latitude.

At a regular point of the patch, ∂X/∂θ and ∂X/∂ϕ span its tangent plane. At a pole, the second vector vanishes, showing why this particular parametrization fails there. A different patch resolves the problem.

Surfaces give us accessible examples, but the intrinsic definitions below also work in higher dimensions without an embedding.

Tangents

A tangent vector records the direction and rate of passage of a curve through a point. In a Euclidean embedding we can picture it as an arrow. Intrinsically, we can define it by how it differentiates scalar functions.

For a curve through p at parameter value λ0​, its tangent V acts on a smooth function f as

V[f]=dλd​f(γ(λ))​λ0​​.

The chain rule gives

V[f]=Va∂a​f,Va=dλdxa​​λ0​​,∂a​=∂xa∂​.

We therefore write V=Va∂a​. The vectors ∂a​ form a coordinate basis: each differentiates in the direction of one coordinate while holding the others fixed.

All tangent vectors at p form the tangent space Tp​M. It is an n-dimensional vector space, even when the manifold itself is curved. We can add vectors at the same point and multiply them by numbers.

Vectors at different points belong to different tangent spaces. Comparing them requires additional structure; there is generally no preferred way to declare two arrows at separated points parallel.

A flat surface has tangent planes with the same direction everywhere, while tangent spaces at different points of a curved surface have different orientations.

For a flat affine plane, every tangent plane has the same direction as the surface itself. We can therefore identify each Tp​M naturally with the plane's underlying vector space—often stated informally as “the tangent space is the surface.” Strictly, they are different kinds of objects: the surface contains points, whereas Tp​M contains vectors based at the particular point p. On a curved surface, even this natural identification between different points is unavailable without choosing a rule for transporting vectors.

Coordinate Systems

Coordinates choose a basis as well as labels. On the Euclidean plane, Cartesian and polar coordinates are related by

x=rcosϕ,y=rsinϕ.

The coordinate basis vectors are written

∂a​:=∂xa∂​

for the vector in the xa direction. The same symbol also denotes the corresponding directional-derivative operator. Thus ∂a​f means “let the basis vector ∂a​ act on the scalar function f,” and the result is a number (the component of df), not another vector:

∂a​f=∂xa∂f​.

This is a useful notational compression: ∂a​ is a vector, while ∂a​f is that vector applied to a function.

The chain rule relates their coordinate bases:

∂r​=cosϕ∂x​+sinϕ∂y​,

∂ϕ​=−rsinϕ∂x​+rcosϕ∂y​.

The basis vector ∂ϕ​ has Euclidean length r, not one. An angular coordinate change of one radian corresponds locally to a distance proportional to the radius. A coordinate basis is therefore not necessarily orthonormal.

For overlapping charts xa and x′a, define the Jacobian by

Jab​=∂xb∂x′a​.

It must be invertible on the overlap. The coordinate bases transform with the inverse Jacobian:

∂a′​=∂x′a∂xb​∂b​.

Polar coordinates fail at the origin, where the angle is undefined. Such coordinate failures must be distinguished from singularities of the geometry itself.

Contravariance

We need only one fact about vectors here: their upper-index components transform with the Jacobian,

V′a=∂xb∂x′a​Vb.

This is the contravariant transformation law. The components compensate for the coordinate-basis change so that the vector itself stays the same.

For a simple example, replace a coordinate x by x′=2x. Then ∂x′​=21​∂x​ and Vx′=2Vx. Their product is unchanged. Contravariance is this compensation, not a statement that the underlying direction changes.

The same geometric vector measured with coordinates x and x-prime equals two x basis vectors or four shorter x-prime basis vectors.

Gradient

A scalar field f assigns a number to each point. Its value at a fixed point does not depend on the coordinates: f′(x′)=f(x). Its directional derivative along V is

V[f]=Va∂a​f.

The quantities ∂a​f transform differently from vector components. The chain rule gives

∂a′​f=∂x′a∂xb​∂b​f.

The natural object formed from these derivatives is the differential df. It takes a vector as input and returns the rate at which f changes along it:

df(V)=V[f].

Elementary vector calculus often calls the array of partial derivatives the gradient vector. This works directly in orthonormal Cartesian coordinates because the Euclidean metric identifies vectors with linear functions on vectors. More generally, we must distinguish the differential df from the gradient vector gradf, which requires a metric.[2]

For example, in the Euclidean plane let f=x2+y2=r2. Then

df=2xdx+2ydy=2rdr.

For f equals x squared plus y squared, the differential df vanishes along a level curve and the Euclidean metric identifies it with the outward gradient vector.

These expressions represent the same rule for computing changes in f. After introducing the metric, we will see precisely how to turn that rule into a vector.

Forms

A covector, or one-form, is a linear rule that takes a tangent vector and returns a real number. The differential df is the main example: it sends a vector V to the directional rate of change df(V)=V[f].

You can picture a one-form as a measuring tape laid across a surface. Its evenly spaced marks define one direction of measurement. Applying the one-form to a vector tells us the signed number of marks crossed by that vector: a vector parallel to the marks gives zero, while one pointing across them gives a nonzero result. A general one-form measures a component in this sense; it does not by itself return a vector's full length. A metric supplies the additional information needed to measure lengths and angles.

The one-form dx acts like equally spaced vertical measuring marks: a vector along the marks gives zero, while a vector crossing three intervals gives three.

In coordinates, the covector basis dxa picks out vector components: dxa(∂b​)=δab​, where δab​ is one when a=b and zero otherwise. We can therefore write a general covector and its action on a vector as

ω=ωa​dxa,ω(V)=ωa​Va.

Covariance

Covariance follows from a simple requirement: the number produced by a covector acting on a vector must not depend on the coordinates. In either chart,

ω(V)=ωa​Va=ωa′​V′a.

Insert the known transformation of the vector components:

ωa′​V′a=ωa′​∂xb∂x′a​Vb.

For this to equal ωb​Vb for every vector V, the covector components must transform with the inverse Jacobian:

ωa′​=∂x′a∂xb​ωb​.

This is the covariant transformation law, indicated by a lower index. Lower-index components transform in the opposite way to upper-index vector components so that their contraction remains a coordinate-independent scalar.

For x′=2x, the one-form ω=ωx​dx has ωx′​=21​ωx​. Together with Vx′=2Vx, this gives

ωx′​Vx′=ωx​Vx.

The covector measuring tape rescales with the coordinate: x-prime has twice as many marks, each worth half as much, so it gives the same measurement of the vector.

In several dimensions, the cancellation follows from the inverse-matrix identity

∂x′a∂xb​∂xc∂x′a​=δbc​.

The upper or lower position of an index therefore specifies a transformation rule, not just a typographical preference. The broader phrase coordinate covariance means that an equation retains its geometric meaning under coordinate changes; it does not mean that every index must be lower.

Metric

A metric g is a smooth assignment of symmetric, nondegenerate bilinear forms to tangent spaces. It takes two vectors at the same point and returns a number:

g(V,W)=gab​VaWb,gab​=g(∂a​,∂b​).

Symmetry means g(V,W)=g(W,V), and bilinearity means linearity in each argument separately. Nondegeneracy means that the matrix gab​ is invertible. On a Riemannian manifold it is positive definite, so it defines lengths and angles in the familiar way.

The Euclidean plane has line element

dℓ2=dx2+dy2=dr2+r2dϕ2.

Thus its polar metric components are grr​=1, gϕϕ​=r2, and grϕ​=0. A vector's squared length is (Vr)2+r2(Vϕ)2. Position-dependent metric components do not by themselves imply curvature: this plane is still flat.

In Cartesian Euclidean coordinates, the metric matrix is the identity, g=I. Having g=I on a coordinate region does mean the geometry there is flat. The converse is more subtle: a flat geometry need not have g=I in the coordinates currently being used, as polar coordinates show. Flatness means that there exists a coordinate system in which the metric takes the Euclidean identity form (at least locally).

Gradient from a Metric

The metric also defines the gradient vector by g(gradf,V)=df(V) for every V. Its components are

(gradf)a=gab∂b​f.

In the Euclidean polar plane this becomes

gradf=(∂r​f)∂r​+r21​(∂ϕ​f)∂ϕ​.

The familiar Cartesian gradient is the flat-space special case of this rule. On a curved surface, the same differential df becomes a gradient vector only after the intrinsic metric supplies gab. For example, on a sphere of radius R with f=cosθ,

df=−sinθdθ,gradf=−R2sinθ​∂θ​.

The gradient remains perpendicular to the level curves, but its direction belongs to the tangent plane at each point and its magnitude varies with latitude. It vanishes at the poles, where the height function has a maximum or minimum.

On a Euclidean plane the gradient of f equals y is constant, while on a sphere the gradient of f equals cosine theta is tangent to the sphere and varies with latitude.

A spacetime metric is Lorentzian rather than positive definite. With our signature, g(V,V) is positive for timelike vectors, negative for spacelike vectors, and zero for null vectors. A nonzero vector can therefore have zero norm without the metric being degenerate. Along a timelike worldline, ds2=c2dτ2 relates the metric to a clock's proper time.

Tensors

We can meet tensors first as operators on vectors. In familiar linear-algebra notation, a linear operator T takes an input column vector v and returns an output column vector w:

w=Tv.

On a manifold, v and w are the coordinate columns of tangent vectors V,W∈Tp​M, and T is represented by components Tab​:

W=T(V),Wa=Tab​Vb.

The lower index b identifies the input component of V being used, while the upper index a identifies the component of the output vector. The repeated b is summed—exactly the row-by-column multiplication in w=Tv. Thus Tab​ is the coordinate description of a linear operator, not merely a rectangular array of numbers.

For example, in two Cartesian dimensions, the operator

T=(01​−10​)

rotates every vector counterclockwise by a right angle:

w=Tv=(01​−10​)(vx​vy​​)=(−vy​vx​​).

The matrix product above is the familiar notation for a contraction. In index notation, applying the operator to the vector is

Wa=Tab​Vb,

where the repeated index b is summed. The operator's lower index and the vector's upper index are paired, leaving the output index a.

A tensor such as Tab​ is called a mixed tensor, or a tensor of type (1,1). Its upper index is contravariant and its lower index is covariant. For now, we will simply use this mixed notation; we have not yet explained why an index can be moved from one position to the other.

The operator must not depend on the coordinates used to describe it. Under a coordinate change, its components transform once like a vector and once like a covector:

T′ab​=∂xc∂x′a​∂x′b∂xd​Tcd​.

This ensures that transforming V, applying T, and then transforming the result gives the same geometric vector as applying T first. An array of numbers is a tensor only if it obeys the appropriate transformation law. The position of each index is important, even though we are asking you to trust that fact for now. In Cartesian Euclidean coordinates, the metric is the identity, so the numerical distinction between Tab​, Tab​, and Tab is hidden; general relativity makes that distinction visible.

Vectors, covectors, metrics, and operators are all examples of tensors. More generally, a tensor of type (r,s) is a multilinear map that takes r covectors and s vectors as inputs and returns a scalar. Its components carry r upper and s lower indices. A vector has type (1,0), a covector has type (0,1), the metric has type (0,2), and Tab​ is type (1,1). Scalars are tensors of type (0,0).

Contraction

Contraction pairs an upper and a lower index and sums over them. It reduces the type by (1,1); for example, Taa​ is a scalar. The matrix product for composing two operators is another contraction:

(ST)ac​=Sab​Tbc​.

The repeated index b is the contracted row-column index, while a and c remain as the output and input indices of the composite operator. This is why matrix multiplication is naturally expressed as contraction of mixed tensors.

Why does the result remain a tensor? Under a coordinate change, the contracted upper and lower indices contribute inverse Jacobian factors:

S′ab​T′bc​=∂xd∂x′a​Sde​δef​∂x′b∂xe​∂xf∂x′b​​​Tfg​∂x′c∂xg​.

The middle factors cancel by the chain rule, leaving

(ST)′ac​=∂xd∂x′a​(ST)dg​∂x′c∂xg​,

which is exactly the transformation law for another mixed tensor. Contraction is therefore a coordinate-independent operation, not an accidental property of one matrix representation.

In a tensor equation, each unsummed, or free, index must occur in the same position on both sides. For example,

Aa=Tab​Vb

has free index a and summed index b. Tensor notation makes coordinate independence explicit, but that mathematical consistency alone does not establish that an equation describes nature.

Raising and Lowering Indices

Recall that a covector, or one-form, is a linear map that takes one vector as input and returns a number. The metric is a bilinear pairing: it takes two vectors and returns a number. If we fix one argument of any bilinear map, the remaining slot is linear, so it becomes a covector. Given a vector V, we use the metric to define one by keeping the second slot open:

V♭(W):=g(V,W).

This is the metric-induced musical map V↦V♭. It is canonical once the metric has been chosen, but there is no canonical vector-to-covector map on a bare vector space without such extra structure. The symbol ♭ is a mnemonic for the name lowering.[3]

To read the component formula, first expand the vector in the coordinate basis. The numbers Vb are its vector components, and ∂b​ are the basis vectors:

V=Vb∂b​.

Let ω:=V♭ denote the resulting covector, expanded in the dual basis dxa:

ω=ωa​dxa,ωa​=gab​Vb.

Derivation of the component formula

Here the metric components are defined by its action on the coordinate basis:

gab​:=g(∂a​,∂b​).

Using V=Vb∂b​, the covector ω=V♭ acts on the basis vector ∂a​ as

ωa​=ω(∂a​)=V♭(∂a​)=g(V,∂a​)=Vbg(∂b​,∂a​)=gab​Vb,

where symmetry of the metric lets us write g(∂b​,∂a​)=gab​. Thus gab​ is the rule that takes the vector component column Vb and calculates the covector component column ωa​.

In polar coordinates on the Euclidean plane,

dℓ2=dr2+r2dϕ2,

so a vector

V=Vr∂r​+Vϕ∂ϕ​

becomes the covector

ω=V♭=Vrdr+r2Vϕdϕ.

The factor r2 is the nontrivial metric component gϕϕ​; it is exactly why the lower and upper ϕ components differ.

The matrix G=(gab​) represents this metric pairing, so the operation is ω=GV. Any invertible matrix could define a reversible map between columns, but the metric is the particular symmetric, nondegenerate pairing chosen to define lengths, angles, and spacetime intervals.

The inverse metric gab performs the reverse operation:

Va=gabωb​,gabgbc​=δac​.

To verify this, start with the covector components ωb​=gbc​Vc and multiply by gab:

gabωb​=gabgbc​Vc=δac​Vc=Va.

Here b is a repeated, or dummy, index, so we sum over it. The index a appears once on each side and is therefore the component we are solving for. The identity

gabgbc​=δac​

plays the same role as G−1G=I for an ordinary matrix G. The Kronecker delta δac​ is the identity matrix in index notation: when it acts on Vc, it leaves the component labelled by a unchanged,

δac​Vc=Va.

In matrix language, lowering is ω=GV, while raising is V=G−1ω. The index calculation above is precisely this matrix-inverse calculation written component by component.

This is called raising the index. These are not changes to the underlying geometric object; they are two metric-dependent ways of representing related objects. In Cartesian Euclidean coordinates, where gab​=δab​, the numerical components happen to be unchanged. In polar coordinates, however,

Vr​=Vr,Vϕ​=r2Vϕ,

so the distinction is visible.

Tensor Fields

A tensor field assigns a tensor smoothly at each point. In general relativity, the metric gμν​, stress–energy tensor Tμν​, and curvature tensor Rρσμν​ are central examples. Their components vary with position and coordinate choice while the geometric fields remain well defined.

Looking ahead. The tensors introduced here are the language needed for curvature. In the subsequent pages, we will develop the metric, connection, and curvature tensors in more detail, and see how they describe gravity in general relativity.


  1. A hypersurface is a submanifold with one dimension fewer than its ambient space. Thus a surface in three-dimensional space is a hypersurface, while a hypersurface in four-dimensional spacetime is three-dimensional. ↩︎

  2. The common notation ∇f is also used for the metric gradient. We write gradf here to avoid confusing it with the covariant derivative ∇a​, introduced later. ↩︎

  3. In music, a flat lowers a note by a semitone; the same visual symbol is used here as a reminder that the metric lowers an index. See Flat (music). ↩︎

Last Updated: 9/25/26, 7:13 AM
Contributors: Hanh Huynh Huu
Prev
The Equivalence Principle
Next
Spacetime