Raising and Lowering Indices

A tensor lives intrinsically in a product of tangent and cotangent spaces, and the distinction between a vector (an element of $T_pM$ , drawn as an arrow) and a covector (an element of $T_p^*M$ , drawn as a stack of level surfaces) is real and basis-independent — one you push forward along a map, the other you pull back. Yet the moment a non-degenerate metric is chosen, these spaces become canonically isomorphic, and the everyday bookkeeping of physics — writing $V^\mu$ here and $V_\mu$ there — collapses into a question of which slot we happen to have filled. Raising and lowering indices is the operational name for applying that isomorphism, and it is arguably the single most repeated manipulation in all of theoretical physics. Concretely, the metric $g_{\mu\nu}$ is a symmetric, non-degenerate bilinear form, and non-degeneracy is exactly the condition that guarantees an inverse matrix $g^{\mu\nu}$ obeying $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ . Contracting a vector with $g_{\mu\nu}$ produces a covector $V_\mu$ ; contracting a covector with $g^{\mu\nu}$ regenerates the vector $V^\mu$ . Because $g$ and $g^{-1}$ undo one another, nothing is lost: $V$ is one geometric object wearing two component costumes. This article takes the metric duality established in aq-mp-metric-tensor and turns it into a working calculus — why the operations are exact inverses, what index position encodes physically, and how it survives the passage to a curved manifold because the Levi-Civita connection is metric compatible, $\nabla g=0$ . That last fact lets $R^\mu{}_{\nu\alpha\beta}$ and $R_{\mu\nu\alpha\beta}$ be used interchangeably throughout general relativity.

Key takeaways

A non-degenerate metric $g_{\mu\nu}$ and its matrix inverse $g^{\mu\nu}$ give a canonical isomorphism between a vector space and its dual, trading contravariant (upper) indices for covariant (lower) ones and back with no loss of information.
Lowering applies the metric, $V_\mu = g_{\mu\nu}V^\nu$ ; raising applies its inverse, $V^\mu = g^{\mu\nu}V_\nu$ . The two are mutual inverses precisely because $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ .
The rule extends to any rank: each index is raised or lowered independently at the cost of one factor of $g$ or $g^{-1}$ , e.g. $T^\mu{}_\nu = g^{\mu\alpha}T_{\alpha\nu}$ , and slot order must be tracked with explicit spacing.
The metric turns the inner product into a contraction, $\langle V,W\rangle = g_{\mu\nu}V^\mu W^\nu = V_\mu W^\mu = V^\mu W_\mu$ , unifying the geometric and algebraic pictures of length and angle.
Index position carries physics: in Minkowski signature lowering flips the time component, $V_0=-V^0$ , so the operation encodes the causal structure — for $V^\mu=(2,1,0,0)$ the invariant is $V_\mu V^\mu=-3$ .
On a curved manifold $g$ varies point to point, yet raising and lowering stay pointwise algebraic and commute with the covariant derivative because $\nabla_\alpha g_{\mu\nu}=0$ .
Key idea: a non-degenerate metric supplies a canonical isomorphism between a vector space and its dual; lowering applies $g_{\mu\nu}$ , raising applies its inverse $g^{\mu\nu}$ , and because $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ the operations are mutual inverses that lose no information.

The core insight

A non-degenerate metric supplies a canonical isomorphism between a vector space and its dual; lowering an index applies $g_{\mu\nu}$ , raising applies its inverse $g^{\mu\nu}$ , and because $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ the operations are mutual inverses that lose no information.

Where this lives today

Index raising and lowering is not a historical curiosity but the everyday reflex of every working relativist and field theorist. In numerical relativity — the codes that generated the gravitational-wave templates LIGO and Virgo matched against the 2015 binary-black-hole detection and its successors — the Einstein equations are recast in a 3+1 split where spatial indices are raised and lowered with a time-dependent three-metric at every grid point and timestep, and getting that bookkeeping right is essential to a stable simulation. In lattice QCD, discretized gauge fields live on a Euclidean metric where up-down contractions define the action supercomputers evaluate for hadron masses. Physics computer-algebra systems such as xAct and Cadabra implement canonicalization algorithms whose entire job is to track index positions and apply the metric automatically across expressions with hundreds of contracted indices. And every time a particle physicist writes the Lorentz-invariant $-\tfrac14 F_{\mu\nu}F^{\mu\nu}$ or a fermion bilinear $\bar\psi\gamma^\mu\partial_\mu\psi$ , the metric silently pairs each upper index with a lowered partner so the number computed is the same in every frame.

The physical picture

Picture a vector as an arrow and a covector as a stack of parallel, equally spaced planes — the level sets of a linear functional, densest where it grows fastest. These are genuinely different animals: an arrow has length and direction, while a covector "reads" a vector by counting how many of its planes the arrow pierces. Without extra structure there is no canonical way to turn one into the other. The metric is precisely that structure: given an arrow $V^\mu$ , it manufactures the unique stack of planes whose piercing-count of any test arrow $W$ equals the projection $g_{\mu\nu}V^\mu W^\nu$ . That stack is the lowered covector $V_\mu$ , and the inverse metric runs the machine backwards.

Figure

The metric as a dictionary between arrows and stacks of planes: a tangent vector V^μ (the arrow) is sent by the flat map ♭ using g_μν to the covector V_μ (the stack of level planes it pierces), and the sharp map ♯ using the inverse g^μν sends it back, so V_μ = g_μν V^ν and V^μ = g^μν V_ν are exact inverses

In Euclidean space in Cartesian coordinates the metric is the identity, $g_{\mu\nu}=\delta_{\mu\nu}$ , so $V_\mu$ and $V^\mu$ have identical numbers and the distinction becomes invisible — which is why elementary vector calculus never mentions it. The bookkeeping earns its keep only once the metric is non-trivial: curvilinear coordinates, a curved manifold, or Lorentzian signature. In the last case lowering can flip a sign. The second figure makes that concrete on the simplest four-vector: the mostly-plus Minkowski metric $\eta_{\mu\nu}=\mathrm{diag}(-1,+1,+1,+1)$ negates the time slot, leaves the space slots alone, and yields a frame-independent contraction.

Figure

Lowering in Minkowski space with η = diag(-1,+1,+1,+1): the contravariant vector V^μ = (2, 1, 0, 0) becomes V_μ = (-2, 1, 0, 0) — only the time component flips sign — and the invariant contraction is V_μ V^μ = (-2)(2) + (1)(1) + 0 + 0 = -3, a negative norm marking a timelike vector

Two habits make the rest effortless. First, read index gymnastics as type-checking: a correct tensor equation carries the same free indices, in the same up/down positions, on both sides, while every other index appears once up and once down as a summed dummy pair. If a final expression sprouts two free upper $\mu$ 's, an algebra error has crept in. Second, raising and lowering is local — it uses only $g$ at the point, never its derivatives — which is why it is algebraic and cleanly separate from differentiation, even though metric compatibility later lets the two be interchanged.

The core equations

The whole calculus rests on two operations and the identity linking them. Lowering an index contracts a contravariant slot against the metric, giving the covariant (dual) representation of the same object.

V_\mu = g_{\mu\nu}\,V^\nu \qquad (1)

where: $V^\nu$ — contravariant (upper-index) components of a vector; $V_\mu$ — its covariant (lower-index) components, the metric dual; $g_{\mu\nu}$ — metric tensor components, symmetric and non-degenerate. Reading: feed the vector into one slot of the metric, leave the other open, and out comes a covector. Raising reverses the move by contracting a covariant slot against the inverse metric. The inverse exists precisely because $g$ is non-degenerate, and it is the matrix inverse of $g_{\mu\nu}$ .

V^\mu = g^{\mu\nu}\,V_\nu \qquad (2)

where: $g^{\mu\nu}$ — inverse metric components. Reading: $g^{-1}$ is the machine that turns a stack of planes back into an arrow. The two operations are exact inverses because the metric and its inverse compose to the identity. This one relation is the engine of the entire subject.

g^{\mu\alpha}\,g_{\alpha\nu} = \delta^\mu{}_\nu \qquad (3)

where: $\delta^\mu{}_\nu$ — the Kronecker delta, components of the identity, equal to $1$ when $\mu=\nu$ and $0$ otherwise. Reading: lowering then raising runs $g$ into $g^{-1}$ , which cancel, so you land back on the original components. Nothing here is special to vectors. Each index of a higher-rank tensor sits in its own slot and is raised or lowered independently, each move costing one factor of $g$ or $g^{-1}$ , with slot order tracked by explicit horizontal spacing.

T^\mu{}_\nu = g^{\mu\alpha}\,T_{\alpha\nu},\qquad T_{\mu\nu} = g_{\mu\alpha}\,T^\alpha{}_\nu \qquad (4)

where: $T_{\alpha\nu}$ — fully covariant rank-2 components; $T^\mu{}_\nu$ — the mixed tensor with the first index raised, the staggered placement recording which original slot it came from. Reading: raise one index at a time, never stacking indices ambiguously. The metric's deepest job is to make lengths and angles computable, and lowering shows the inner product is a contraction in disguise. The three forms below name the same scalar.

\langle V,W\rangle = g_{\mu\nu}V^\mu W^\nu = V_\mu W^\mu = V^\mu W_\mu \qquad (5)

where: $\langle V,W\rangle$ — the metric inner product, a coordinate-independent scalar. Reading: lower one factor and the metric vanishes into a clean up-down contraction; symmetry of $g$ makes it irrelevant which factor you lower. Setting $W=V$ gives the invariant squared length, allowed in Lorentzian signature to be negative, zero, or positive — the algebraic fingerprint of timelike, null, and spacelike vectors.

V^2 = g_{\mu\nu}V^\mu V^\nu = V_\mu V^\mu \qquad (6)

where: $V^2$ — the invariant length-squared, whose sign is a Lorentz-invariant causal label. Finally, on a curved manifold the metric varies from point to point, yet the operations remain trivial to differentiate because the Levi-Civita connection annihilates the metric — metric compatibility — letting you slide $g$ freely through a covariant derivative.

\nabla_\alpha g_{\mu\nu} = 0 \quad\Rightarrow\quad \nabla_\alpha V_\mu = g_{\mu\nu}\,\nabla_\alpha V^\nu \qquad (7)

where: $\nabla_\alpha$ — the Levi-Civita (metric-compatible) covariant derivative. Reading: raising and lowering commute with differentiation, so index placement never obstructs the calculus of curvature.

Derivation ARaising and lowering as the canonical metric isomorphism

The goal here is to prove that raising and lowering are not merely a notational convenience but the component form of a genuine, basis-independent linear isomorphism between the tangent space $T_pM$ and its dual $T_p^*M$ , and to pin down exactly why non-degeneracy of the metric is the precise hypothesis that makes the map invertible. We define the flat map by contraction with the metric, establish its linearity, injectivity, and bijectivity, and show its inverse is forced to have components equal to the inverse metric.

Step 1. Define the flat map $\flat: T_pM \to T_p^*M$ by $\flat(V) = g(V,\cdot)$ , meaning the covector that acts on any test vector $W$ as $[\flat(V)](W) = g_{\mu\nu}V^\mu W^\nu$ . Because $g$ is bilinear, $\flat(aV+bU) = a\,\flat(V) + b\,\flat(U)$ . Why: linearity is what qualifies $\flat$ as a legitimate map of vector spaces, not just an arbitrary assignment, so all the tools of linear algebra become available.

Step 2. Read off components. Evaluating $\flat(V)$ against the basis vectors gives $[\flat(V)]_\mu = g_{\mu\nu}V^\nu \equiv V_\mu$ , which is exactly equation (1). Why: this identifies the abstract map with the concrete operation — "lowering an index" IS applying $\flat$ , so the geometric and computational pictures are literally the same statement.

Step 3. Show injectivity. Suppose $\flat(V)=0$ , i.e. $g_{\mu\nu}V^\nu = 0$ for every value of $\mu$ . Non-degeneracy of $g$ is by definition the statement that the only vector annihilated this way is the zero vector, so $V=0$ and the kernel of $\flat$ is trivial. Why: this is the one place the hypothesis on $g$ is spent — degeneracy would leave a nonzero vector in the kernel, and the map would crush distinct vectors together, destroying invertibility.

Step 4. Promote to bijectivity. The tangent space and its dual have the same finite dimension $n$ , and an injective linear map between equal-dimensional spaces is automatically surjective, hence a bijection. Its inverse is the sharp map $\sharp: T_p^*M \to T_pM$ . Why: a bijection has a two-sided inverse, so every covector comes from exactly one vector — raising is well-defined and unambiguous.

Step 5. Force the inverse's components. Demanding $\sharp\circ\flat = \mathrm{id}$ means the matrix implementing $\sharp$ must satisfy $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ , precisely equation (3) — the raising matrix is the matrix inverse of the metric, $g^{\mu\nu}$ . Why: consistency alone dictates that raising must use $g^{-1}$ ; there is no freedom, which is why the construction is called canonical.

Go deeper · Derivation B · 1 minIndex gymnastics in Minkowski space and the mass shell

Derivation BIndex gymnastics in Minkowski space and the mass shell

The aim is to run the abstract machinery on the most important vector in relativistic physics — the energy-momentum four-vector — and watch a single correct index-lowering produce the relativistic energy-momentum relation, the invariant that defines a particle's rest mass. We lower the four-momentum's index with the Minkowski metric, contract to form the frame-independent scalar, and identify it with the mass; every sign traces to the signature.

Step 1. Write the contravariant four-momentum in mostly-plus signature $\eta_{\mu\nu}=\mathrm{diag}(-1,+1,+1,+1)$ as $p^\mu = (E/c,\ \vec p)$ , the time component carrying the energy and the space components the ordinary momentum. Why: fixing the signature convention up front is essential — the entire sign structure of the result, and even the sign of the invariant, depends on it.

Step 2. Lower the index by contracting with $\eta$ : $p_\mu = \eta_{\mu\nu}p^\nu = (-E/c,\ \vec p)$ . Only the time component changes sign because $\eta_{00}=-1$ , while $\eta_{ii}=+1$ leaves the spatial momentum untouched. Why: this is the physical content of lowering in Lorentzian signature — the covariant and contravariant time components genuinely differ, and forgetting the sign is the commonest error in relativistic kinematics.

Step 3. Form the invariant contraction $p_\mu p^\mu = (-E/c)(E/c) + \vec p\cdot\vec p = -E^2/c^2 + |\vec p|^2$ . This pairs a covector with a vector, so the result is a Lorentz scalar — the same number in every inertial frame. Why: contracting an up index with a down index is guaranteed to produce an invariant, which is why physically meaningful quantities are always built as fully contracted expressions.

Step 4. Identify the scalar with the rest mass. Evaluating $p_\mu p^\mu$ in the particle's rest frame, where $\vec p=0$ and $E=mc^2$ , gives $-m^2c^2$ , so $p_\mu p^\mu = -m^2c^2$ in every frame, which rearranges to $E^2 = (|\vec p|c)^2 + (mc^2)^2$ . Why: the mass shell is nothing but the statement that the length of the four-momentum is fixed, and that length is computed by exactly one index-lowering — the geometry hands you the dispersion relation for free.

Go deeper · Derivation C · 1 minRaising a higher-rank tensor and the invariant trace

Derivation CRaising a higher-rank tensor and the invariant trace

The purpose is to show that the trace of a rank-2 tensor — the scalar you get by contracting one upper against one lower index — is defined only once a metric lets you raise an index, and that the answer is independent of which index you raise when the tensor is symmetric. This is the operation that manufactures the Ricci scalar from the Ricci tensor, the scalar curvature at the heart of Einstein's equations.

Step 1. Start from a fully covariant rank-2 tensor $T_{\alpha\nu}$ and raise its first index with the inverse metric: $T^\mu{}_\nu = g^{\mu\alpha}T_{\alpha\nu}$ . Why: a "trace" needs one up and one down index to contract, but a fully covariant tensor has none upstairs — the metric is what supplies the missing upper index in the first place.

Step 2. Contract the mixed tensor by setting the upper and lower indices equal and summing: $T^\mu{}_\mu = g^{\mu\alpha}T_{\alpha\mu}$ . This up-down contraction is a scalar. Why: only an up-down pair contracts to something coordinate-independent; contracting two lower indices without a metric would not transform as a scalar.

Step 3. Check that the choice of index does not matter. Because $g^{\mu\alpha}=g^{\alpha\mu}$ is symmetric, $g^{\mu\alpha}T_{\alpha\mu}=g^{\alpha\mu}T_{\alpha\mu}$ , so raising the second index instead gives the same number for symmetric $T$ . Why: a well-defined trace must not depend on an arbitrary bookkeeping choice, and the symmetry of the metric is exactly what guarantees that.

Step 4. Write the trace as a full double contraction, $\mathrm{tr}\,T = g^{\mu\nu}T_{\mu\nu}$ , a single coordinate-independent scalar. Applied to the Ricci tensor this reads $R = g^{\mu\nu}R_{\mu\nu}$ , the Ricci scalar. Why: this is the concrete route by which curvature becomes a number — the scalar that Einstein's field equations balance against the trace of the stress-energy tensor.

Limiting cases

How raising and lowering behave in the special cases that matter most:

Case	Result	Meaning
Euclidean Cartesian	$g_{\mu\nu}=\delta_{\mu\nu}=g^{\mu\nu}$	Components are numerically unchanged, $V_\mu=V^\mu$ . The up/down distinction is invisible, which is why introductory vector calculus can ignore it entirely.
Minkowski (mostly plus)	$\eta_{\mu\nu}=\mathrm{diag}(-1,+1,+1,+1)$	Lowering flips the time component, $V_0=-V^0$ , and leaves space components, $V_i=V^i$ . Here $\eta^{\mu\nu}=\eta_{\mu\nu}$ as matrices because $\eta$ is its own inverse.
Diagonal curvilinear	$V_i = g_{ii}V^i$ (no sum)	In spherical coordinates $g_{rr}=1$ , $g_{\theta\theta}=r^2$ , $g_{\phi\phi}=r^2\sin^2\theta$ ; lowering rescales each component by its diagonal metric factor.
Conformal rescaling	$g_{\mu\nu}\to\Omega^2 g_{\mu\nu}$	Lowering picks up $\Omega^2$ , raising picks up $\Omega^{-2}$ ; a mixed tensor's fully traced scalar changes, which is why conformal weight bookkeeping matters.
Degenerate / no metric	$\det g = 0$	The inverse $g^{\mu\nu}$ does not exist; raising is impossible and vectors cannot be canonically identified with covectors. A non-degenerate metric is the prerequisite.

Common trapThe up/down distinction is "just notation" and index order can be ignored

Neither is safe. Raising and lowering only work because the metric is non-degenerate; strip it away, or let it degenerate, and there is no canonical way to identify a vector with a covector at all — the operation is undefined, not merely inconvenient. And index position is physical: in Minkowski space $V_0 = g_{00}V^0 = -V^0$ , so writing $V_0$ where you meant $V^0$ flips a sign and can turn a correct energy into its negative. Slot order is equally load-bearing for non-symmetric tensors: $T^\mu{}_\nu = g^{\mu\alpha}T_{\alpha\nu}$ raises the first slot, $T_\nu{}^\mu = g^{\mu\alpha}T_{\nu\alpha}$ the second, and these are different tensors. Always stagger indices explicitly — write $T^\mu{}_\nu$ , never an ambiguous stacked $T^\mu_\nu$ .

Worked examples

1 · Lowering a Minkowski four-vector and reading its causal character (confidence builder)

In Minkowski space with mostly-plus metric $\eta=\mathrm{diag}(-1,1,1,1)$ , a vector has contravariant components $V^\mu=(2,1,0,0)$ . Find its covariant components $V_\mu$ , its invariant norm $V^2$ , and classify the vector.

Step 1. Lower each index with $\eta$ , using $V_\mu=\eta_{\mu\nu}V^\nu$ . The metric is diagonal, so each component is simply scaled by the matching diagonal entry: $V_0=\eta_{00}V^0=(-1)(2)=-2$ , and $V_1=\eta_{11}V^1=(+1)(1)=1$ , with $V_2=V_3=0$ . So $V_\mu=(-2,1,0,0)$ — only the time component changed sign.

Step 2. Form the invariant norm as the contraction $V^2=V_\mu V^\mu$ , pairing the covariant components just found with the original contravariant ones: $V^2=(-2)(2)+(1)(1)+(0)(0)+(0)(0)=-4+1=\mathbf{-3}$ . Equivalently $V^2=\eta_{\mu\nu}V^\mu V^\nu=-(2)^2+(1)^2=-3$ , the same number by construction.

Step 3. Classify. In mostly-plus signature a negative norm means the vector is timelike — its time part dominates its space part, the character of a particle's four-velocity or four-momentum. Had $V^2>0$ it would be spacelike, and $V^2=0$ null. The single sign is a Lorentz-invariant physical verdict: $V_\mu=(-2,1,0,0)$ , $V^2=-3<0$ , timelike.

2 · Lowering a vector in spherical coordinates

In flat 3-space in spherical coordinates the metric is diagonal with $g_{rr}=1$ , $g_{\theta\theta}=r^2$ , $g_{\phi\phi}=r^2\sin^2\theta$ . At $r=2$ , $\theta=\pi/2$ a vector has contravariant components $V^\mu=(V^r,V^\theta,V^\phi)=(3,\ 0.5,\ 0.25)$ . Find $V_\mu$ , and compare the physical (orthonormal) $\theta$ -component.

Step 1. Evaluate the metric factors at the point. With $r=2$ and $\theta=\pi/2$ (so $\sin\theta=1$ ): $g_{rr}=1$ , $g_{\theta\theta}=r^2=4$ , and $g_{\phi\phi}=r^2\sin^2\theta=4$ . The metric is diagonal, so lowering acts componentwise with no sums.

Step 2. Lower each index via $V_i=g_{ii}V^i$ (no sum): $V_r=(1)(3)=3$ , $V_\theta=(4)(0.5)=2$ , and $V_\phi=(4)(0.25)=1$ . So $V_\mu=(3,\ 2,\ 1)$ — the covariant components differ dramatically from the contravariant ones because the angular metric factors are large at $r=2$ .

Step 3. Contrast with the physical component. The orthonormal $\theta$ -component a lab reports is $V^{\hat\theta}=\sqrt{g_{\theta\theta}}\,V^\theta=\sqrt{4}\,(0.5)=1$ , which sits between the contravariant $V^\theta=0.5$ and the covariant $V_\theta=2$ . The three numbers $\{V^\theta, V^{\hat\theta}, V_\theta\}=\{0.5, 1, 2\}$ differ by successive factors of $\sqrt{g_{\theta\theta}}=2$ : contravariant, physical, and covariant are one vector scaled by powers of the metric factor.

3 · The trace of the metric with its own inverse

Show that contracting the metric fully against its inverse returns the dimension of the space.

Step 1. Raise one index of the metric with its inverse: $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ by equation (3) — the mixed form of the metric is just the identity.

Step 2. Contract the remaining pair by setting $\mu=\nu$ and summing: $g^{\mu\nu}g_{\mu\nu}=\delta^\mu{}_\mu=n$ , since summing the Kronecker delta over its diagonal counts the coordinate directions. In spacetime this is $g^{\mu\nu}g_{\mu\nu}=4$ , the identity used whenever one traces the Einstein tensor.

Physics of the portrait

Raising and lowering indices is the working grammar of a language invented to write the laws of physics in a form independent of coordinates — and its inventor spent years being ignored before Einstein made his calculus indispensable.

Origins · 1 minGregorio Ricci-Curbastro and how this was discovered

Gregorio Ricci-Curbastro

1853–1925, Italian mathematician · the absolute differential calculus (tensor analysis)

Gregorio Ricci-Curbastro created, almost single-handedly, the calculus in which raising and lowering indices is the fundamental move. Influenced by Riemann's 1854 vision of a manifold carrying a metric and by Christoffel's work on quadratic differential forms, Ricci set out in the 1880s to build a systematic machinery for quantities that keep their meaning under arbitrary change of coordinates. By 1892 he had it: the "absolute differential calculus," which introduced the systematic distinction between covariant and contravariant components and the operations — using the metric $g_{\mu\nu}$ and its inverse $g^{\mu\nu}$ — that convert one into the other. For years the work was dismissed as an obscure formalism. The turning point was the great 1900 memoir he wrote with his student Tullio Levi-Civita, "Méthodes de calcul différentiel absolu et leurs applications," which laid out the index calculus and transformation rules in the form still taught today. Its vindication came in 1915, when Einstein — tutored in the formalism by Marcel Grossmann — found that Ricci's tensor calculus was precisely the mathematics general relativity required, with the contraction now called the Ricci tensor and its metric trace the Ricci scalar at the centre of the field equations. What had looked like abstract bookkeeping turned out to be the natural language of gravitation, and the humble act of trading an upper index for a lower one became one of the most-performed operations in all of physics.

History

1854

— Riemann's Habilitation lecture introduces the metric of a curved manifold

1869

— Christoffel studies quadratic differential forms and covariant differentiation

1892

— Ricci-Curbastro founds the absolute differential calculus with covariant and contravariant components

1900

— Ricci and Levi-Civita systematize the index calculus in their Méthodes memoir

1915

— Einstein's general relativity makes metric index gymnastics the language of gravitation

Going further

Texts. Stone & Goldbart, Mathematics for Physics, chapter on tensors and the metric · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., tensor-analysis chapter · Nakahara, Geometry, Topology and Physics, 2nd ed., §5.4 (Riemannian geometry) · Hassani, Mathematical Physics, tensor chapters · Byron & Fuller, Mathematics of Classical and Quantum Physics, tensor sections · Misner, Thorne & Wheeler, Gravitation, §3 · Wald, General Relativity, §2.3. The key results to keep in mind: $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ defines the inverse metric and makes raising and lowering mutual inverses; the mostly-plus Minkowski sign flip gives $V_0=-V^0$ , so $V^\mu=(2,1,0,0)$ lowers to $V_\mu=(-2,1,0,0)$ with invariant $V_\mu V^\mu=-3$ ; the energy-momentum relation $E^2=(|\vec p|c)^2+(mc^2)^2$ follows from $p_\mu p^\mu=-m^2c^2$ ; the spherical-coordinate factors $g_{\theta\theta}=r^2$ , $g_{\phi\phi}=r^2\sin^2\theta$ give $V_\mu=(3,2,1)$ for $V^\mu=(3,0.5,0.25)$ at $r=2$ , $\theta=\pi/2$ ; the full metric contraction is $g^{\mu\nu}g_{\mu\nu}=n=4$ in spacetime; and metric compatibility $\nabla_\alpha g_{\mu\nu}=0$ holds for the Levi-Civita connection.

Prerequisites. Covariant and Contravariant Tensors · The Metric Tensor · Tensor Algebra and the Tensor Product · Tensor Contraction and the Trace. Leads to. The Covariant Derivative · Hodge Star and the Codifferential · The Riemann Curvature Tensor · Differential Forms.

Ask the tutor: Ask me to lower the indices of any four-vector you name and read off whether it is timelike, spacelike, or null, or to show step by step how the Ricci scalar is built by raising and contracting the Riemann tensor.

Why bother distinguishing upper and lower indices at all?

Because they label genuinely different objects that transform in opposite ways under a change of coordinates: an upper index transforms with the Jacobian, a lower index with the inverse Jacobian. Keeping them apart makes the Einstein summation convention a reliable type-check — a contracted pair, one up and one down, transforms into a coordinate-independent scalar, whereas summing two indices in the same position gives something coordinate-dependent and physically meaningless. The distinction only becomes invisible in Cartesian coordinates on flat space, where the two transformation laws happen to coincide.

What does "non-degenerate" mean, and why is it required?

A metric is non-degenerate at a point if the only vector $V$ satisfying $g_{\mu\nu}V^\nu=0$ for all $\mu$ is the zero vector — equivalently, if $g_{\mu\nu}$ has nonzero determinant and hence an inverse. Lowering works with any bilinear form, but raising needs the inverse metric $g^{\mu\nu}$ , which exists only when $\det g\neq0$ . If the metric were degenerate, lowering would collapse distinct vectors onto the same covector with no way to invert it, and vectors and covectors could no longer be identified. Non-degeneracy is thus not a technical nicety but the precise condition for the whole isomorphism to exist.

Why does lowering flip the sign of the time component but not the space ones?

Purely because of the metric's signature. In the mostly-plus convention $\eta=\mathrm{diag}(-1,+1,+1,+1)$ , so lowering with $V_\mu=\eta_{\mu\nu}V^\nu$ multiplies the time component by $\eta_{00}=-1$ and each space component by $\eta_{ii}=+1$ . The single minus sign encodes that time is different from space — it is what makes the interval $-c^2t^2+x^2+y^2+z^2$ Lorentz-invariant and gives light cones their structure. So the sign flip is not an arbitrary rule; it is the causal geometry of spacetime showing up in the bookkeeping. In the mostly-minus convention the roles reverse, which is why stating your signature is mandatory.

Is T^μ_ν the same tensor as T_ν^μ?

Only if the tensor is symmetric; in general they are different objects and the notation must distinguish them. The expression $T^\mu{}_\nu = g^{\mu\alpha}T_{\alpha\nu}$ raises the first slot of $T_{\alpha\nu}$ , while $T_\nu{}^\mu = g^{\mu\alpha}T_{\nu\alpha}$ raises the second. These agree when $T_{\alpha\nu}=T_{\nu\alpha}$ but differ otherwise — for an antisymmetric tensor they are exact negatives. This is why you must never write a lazy stacked $T^\mu_\nu$ with the indices vertically aligned: the horizontal staggering records which original slot the raised index came from, and losing it can silently change the value of your result.

Do raising and lowering commute with differentiation?

With the covariant derivative, yes — one of the most useful facts in the subject. The Levi-Civita connection is metric compatible, $\nabla_\alpha g_{\mu\nu}=0$ : the metric is constant as far as covariant differentiation is concerned. So differentiating a lowered vector, the Leibniz rule gives $\nabla_\alpha(g_{\mu\nu}V^\nu)=(\nabla_\alpha g_{\mu\nu})V^\nu + g_{\mu\nu}\nabla_\alpha V^\nu$ , the first term vanishes, and $\nabla_\alpha V_\mu = g_{\mu\nu}\nabla_\alpha V^\nu$ — you may raise or lower before or after differentiating and get the same answer. Note the qualifier "covariant": for the ordinary partial $\partial_\alpha$ this fails on a curved manifold, since $\partial_\alpha g_{\mu\nu}\neq0$ in general.

What are the "musical" isomorphisms, sharp and flat?

They are the coordinate-free names for exactly the operations this article describes. The flat map $\flat$ (as in lowering a musical note) sends a vector to a covector by $V^\flat{}_\mu = g_{\mu\nu}V^\nu$ ; the sharp map $\sharp$ sends a covector back by $\omega^\sharp{}^\mu = g^{\mu\nu}\omega_\nu$ . They are inverse isomorphisms between the tangent and cotangent bundles — the same maps physicists invoke every time they raise or lower an index. The gradient of a function, for instance, is naturally a covector $\partial_\mu f$ ; applying sharp gives the gradient vector $g^{\mu\nu}\partial_\nu f$ you actually point along.