Raising and Lowering Indices

Write down a vector on a curved manifold and you have specified a direction — an arrow tangent to some curve. Write down a one-form and you have specified something subtly different — a stack of level surfaces, a machine that eats a vector and returns a number. These two objects live in different homes: the tangent space $T_pM$ , from aq-gr-tangent-vectors-tangent-space, and its dual the cotangent space $T_p^*M$ , from aq-gr-cotangent-space-one-forms. The two spaces have the same dimension, so they are abstractly isomorphic, but on a bare manifold there is no canonical way to pair a particular arrow with a particular stack of surfaces — no preferred dictionary between the languages. The metric tensor supplies exactly the missing dictionary. Because $g_{\mu\nu}$ is a non-degenerate symmetric bilinear form (from aq-gr-metric-tensor-line-element), the map $V \mapsto g(V,\,\cdot\,)$ sends every vector to a definite one-form, and non-degeneracy guarantees the map is invertible. In components this is nothing but lowering an index, $V_\mu = g_{\mu\nu}V^\nu$ ; its inverse, raising, uses the inverse metric $g^{\mu\nu}$ . Differential geometers call the pair the musical isomorphisms, $\flat$ (flat, lowering) and $\sharp$ (sharp, raising), because $\flat$ pushes an index down and $\sharp$ pulls it up. Three consequences run through everything below. First, index placement is a representation of one geometric object, not a new object: $V^\mu$ and $V_\mu$ are the same arrow read two ways. Second, the operation respects the tensor transformation law, so raising or lowering any slot yields a genuine tensor. Third, because contracting an up-index against a down-index is coordinate-invariant, the metric is precisely what lets physics form scalars — norms, proper times, traces, and actions.

Key takeaways

The metric $g_{\mu\nu}$ and its inverse $g^{\mu\nu}$ give a canonical, invertible correspondence between vectors (tangent space) and one-forms (cotangent space) at every point.
Lowering converts a contravariant index to a covariant one, $V_\mu = g_{\mu\nu}V^\nu$ ; raising reverses it, $V^\mu = g^{\mu\nu}V_\nu$ ; the two undo each other because $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ .
Any slot of any tensor can be raised or lowered independently, so one object — the Riemann tensor, the stress-energy tensor — has many equivalent index arrangements chosen for convenience.
Raising and lowering preserve tensor character and let contractions build coordinate invariants like the norm $V^\mu V_\mu = g_{\mu\nu}V^\mu V^\nu$ ; in Lorentzian signature this norm can be negative, zero, or positive.
In Minkowski coordinates lowering flips only the time component: $V^\mu=(2,1,0,0)$ becomes $V_\mu=(-2,1,0,0)$ — same object, genuinely different number array.
Key idea: the metric establishes a canonical, invertible correspondence between vectors and one-forms, $V_\mu = g_{\mu\nu}V^\nu$ and $V^\mu = g^{\mu\nu}V_\nu$ , with $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ guaranteeing the two operations are exact inverses.

The core insight

The metric establishes a canonical, invertible correspondence between vectors and one-forms: $V_\mu = g_{\mu\nu}V^\nu$ and $V^\mu = g^{\mu\nu}V_\nu$ , with $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ guaranteeing the two operations undo each other.

Where this lives today

Index gymnastics stopped being a paper-and-pencil art the moment relativity went numerical. In numerical relativity — the simulations that produce the gravitational-wave templates matched against LIGO and Virgo detections — the Einstein equations are split into a 3+1 form (the ADM and BSSN formulations), and the codes spend a large fraction of their runtime raising and lowering indices on the spatial metric $\gamma_{ij}$ and the extrinsic curvature $K_{ij}$ at every grid point and timestep; getting the inverse metric $\gamma^{ij}$ right is a load-bearing correctness check in packages such as the Einstein Toolkit. The same operation underlies the ray-tracing behind the Event Horizon Telescope images of the M87 and Sgr A* black holes, where photon trajectories are integrated in the Kerr metric and the null condition $g_{\mu\nu}k^\mu k^\nu = 0$ is enforced by lowering the wavevector. The humble musical isomorphism is, quietly, in the inner loop of modern gravitational physics.

The physical picture

Picture a single point $p$ of spacetime with two rooms attached to it. In one live the tangent vectors — arrows, velocities, directions you could move. In the other live the one-forms — not arrows but rulings, families of equally spaced level surfaces, the kind of thing $\mathrm{d}f$ produces from a function $f$ . A one-form acts on a vector by counting how many of its surfaces the arrow pierces. Both rooms are $n$ -dimensional, yet nothing about a bare arrow tells you which particular ruling it should be paired with. You need a rule that measures the arrow — and measuring is exactly what a metric does. Feed a vector into the metric in one slot and leave the other slot open: $g(V,\,\cdot\,)$ is now a machine waiting for a second vector, i.e. a one-form. That is $V_\mu = g_{\mu\nu}V^\nu$ . The surfaces of $V_\mu$ sit perpendicular to the arrow $V^\mu$ , and their spacing is set by the metric — a longer arrow (in the metric's own reckoning) makes a more tightly packed stack. Because the metric is non-degenerate you can always run the machine backwards with $g^{\mu\nu}$ , turning the ruling back into a unique arrow — that reversibility is the whole content of "the map is an isomorphism".

Figure

The metric is a two-way dictionary between the tangent space T_pM where a vector is an arrow and the cotangent space T_p*M where its partner is a stack of level surfaces; in Minkowski coordinates lowering V^μ = (2, 1, 0, 0) gives V_μ = (−2, 1, 0, 0), flipping only the time component

In ordinary Euclidean space in Cartesian coordinates the metric is just the identity $\delta_{\mu\nu}$ , so lowering does nothing to the numbers: $V_\mu = V^\mu$ component by component. That is precisely why introductory physics never bothers to distinguish vectors from one-forms — with $\delta$ as the dictionary, the two languages spell every word identically. The distinction turns on the moment the metric is non-trivial: in a Lorentzian metric the time component carries the opposite sign, so lowering flips it; in curved coordinates the diagonal entries rescale each component and off-diagonal entries mix them. The arrays $V^\mu$ and $V_\mu$ then look genuinely different even though they encode the same arrow.

Figure

Lowering weights each slot by the metric — in Schwarzschild coordinates at r = 2r_s the diagonal entries are g_tt = −1/2 and g_rr = 2, so V_μ = g_μμ V^μ turns V^t = 1 into V_t = −0.5 and V^r = 1 into V_r = 2

The deepest reason to care is that contracting an upper index against a lower one produces a scalar — a single number every observer agrees on. Without the metric, an expression like $V^\mu V^\mu$ (two upper indices) is not even defined: there is nothing to sum a vector against but a one-form, and you need the metric to make one. Lower one index first and $V^\mu V_\mu = g_{\mu\nu}V^\mu V^\nu$ is the squared length — how the metric grants spacetime its geometry of intervals, and why index gymnastics is the machinery that turns tensors into physics.

The core equations

The operation is defined by two mirror-image contractions. Lowering feeds a vector's index into the metric; the free index that survives is covariant, so the result is a one-form. This is the flat map $\flat$ written in components.

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

where: $V^\nu$ — contravariant (vector) components; $g_{\mu\nu}$ — metric tensor components; $V_\mu$ — the resulting covariant (one-form) components. Physical reading: the metric measures the arrow $V^\nu$ and returns the ruling $V_\mu$ that represents the same geometric object in the dual space. Raising is the inverse contraction, using the inverse metric to pull a covariant index back up into a contravariant one. It is the sharp map $\sharp$ , and it exists only because $g$ is non-degenerate.

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

where: $g^{\mu\nu}$ — the inverse metric; $V_\nu$ — covariant components; $V^\mu$ — the recovered contravariant components. Physical reading: raising turns a stack of surfaces back into the unique arrow that pierces them the metric's way. What makes (1) and (2) exact inverses is the single algebraic fact that defines $g^{\mu\nu}$ : it is the matrix inverse of $g_{\mu\nu}$ , nothing more and nothing less. This is a definition imposed independently, not something obtained by lowering indices on the metric.

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

where: $\delta^\mu{}_\nu$ — the Kronecker delta, the components of the identity map. Physical reading: raise a lowered index and the two metrics collapse to the identity, returning you to where you started — this is why $\sharp\circ\flat = \mathrm{id}$ . The rule extends slot by slot to a tensor of any rank: contract $g$ or $g^{-1}$ against exactly the index you wish to move, leaving the others untouched. A rank-2 tensor therefore has fully contravariant, mixed, and fully covariant faces, all describing one object.

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

where: $T^{\alpha\nu}$ — a rank-(2,0) tensor; $T_\mu{}^\nu,\,T^\mu{}_\nu$ — the same object with one index lowered. Physical reading: each free index is independent, so you can lower the first, the second, or both, in any order, and always land on a legitimate tensor. Contracting a raised index against its lowered partner delivers the fundamental invariant of the theory — the norm. In Lorentzian signature its sign classifies the vector as timelike, null, or spacelike.

V^\mu V_\mu = g_{\mu\nu}\,V^\mu V^\nu = \lVert V\rVert^2 \qquad (5)

where: $\lVert V\rVert^2$ — the squared length; negative for timelike, zero for null, positive for spacelike (in the mostly-plus convention). Physical reading: this single number is the same in every coordinate system — it is what proper time, energy-momentum normalization, and the line element are all built from. Because the metric is symmetric, the "raised-ness" in a contraction may be shifted freely between the two factors — both spellings are the same scalar $g_{\mu\nu}A^\mu B^\nu$ .

A_\mu B^\mu = A^\mu B_\mu = g_{\mu\nu}\,A^\mu B^\nu \qquad (6)

where: $A_\mu B^\mu$ and $A^\mu B_\mu$ — two ways of writing one invariant. Physical reading: you may move a metric from one factor to the other at will, a freedom used constantly to simplify tensor expressions. Finally, raising one index of the metric itself returns the identity — a useful sanity check and a reminder that the "mixed metric" is not a new object.

g^{\mu}{}_{\nu} = g^{\mu\alpha}\,g_{\alpha\nu} = \delta^{\mu}{}_{\nu} \qquad (7)

where: $g^\mu{}_\nu$ — the metric with one index raised, which carries no information beyond the Kronecker delta. Physical reading: the metric is, by construction, the object whose mixed form is the identity.

Derivation AThe metric induces a canonical isomorphism between vectors and one-forms

The goal is to prove that lowering with a non-degenerate metric is not merely a map from vectors to one-forms but a linear isomorphism — a bijection that canonically identifies the two spaces once a metric is chosen. Define $\flat:T_pM\to T_p^*M$ by $\flat(V) = g(V,\,\cdot\,)$ , i.e. $(\flat V)_\mu = g_{\mu\nu}V^\nu$ . Showing this map is linear, maps between spaces of equal dimension, and (by non-degeneracy) is one-to-one establishes that $\flat$ is invertible, with inverse the raising map $g^{\mu\nu}$ .

Step 1. Check linearity: for scalars $a,b$ and vectors $V,W$ , $g_{\mu\nu}(aV+bW)^\nu = a\,g_{\mu\nu}V^\nu + b\,g_{\mu\nu}W^\nu = a(\flat V)_\mu + b(\flat W)_\mu$ . Why: contraction with a fixed tensor is linear in its argument, so $\flat$ is a linear map and it suffices to study it with the tools of linear algebra.

Step 2. Note the dimensions match: $T_pM$ and $T_p^*M$ both have dimension $n = \dim M$ , because a space and its dual always have equal finite dimension. Why: for a linear map between two spaces of equal finite dimension, injective, surjective, and bijective are all equivalent, so we only need to rule out a non-trivial kernel.

Step 3. Show the kernel is trivial using non-degeneracy. Suppose $\flat V = 0$ , i.e. $g_{\mu\nu}V^\nu = 0$ for every $\mu$ . Non-degeneracy means the only vector orthogonal to every vector is the zero vector, equivalently $\det(g_{\mu\nu})\neq 0$ , so $V^\nu = 0$ . Why: a vanishing kernel makes the linear map injective, and by Step 2 injective implies bijective — $\flat$ is an isomorphism.

Step 4. Identify the inverse explicitly. Since $\det g\neq 0$ the matrix $g_{\mu\nu}$ is invertible; call its inverse $g^{\mu\nu}$ , defined by $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ . Then $(\sharp\omega)^\mu = g^{\mu\nu}\omega_\nu$ satisfies $\sharp\circ\flat = \mathrm{id}$ , so $\sharp = \flat^{-1}$ . Why: this pins raising as the exact inverse of lowering, completing the proof that vectors and one-forms are in canonical one-to-one correspondence the moment a metric is present.

Go deeper · Derivation B · 1 minLowered components transform as a genuine covariant tensor

Derivation BLowered components transform as a genuine covariant tensor

The goal is to confirm that $V_\mu$ defined by lowering is not just a convenient array of numbers but a legitimate rank-(0,1) tensor — that it obeys the covariant transformation law whenever $V^\mu$ obeys the contravariant one; if it failed, index gymnastics would produce coordinate-dependent nonsense. We start from the transformation rules for a vector and for the metric under $x^\mu\to x^{\mu'}$ , substitute both into the lowering formula written in the primed frame, and watch a chain-rule product collapse to a Kronecker delta.

Step 1. Write how each ingredient transforms: the vector picks up one Jacobian, $V^{\nu'} = \dfrac{\partial x^{\nu'}}{\partial x^\nu}V^\nu$ , and the metric, being rank-(0,2), picks up two inverse Jacobians, $g_{\mu'\nu'} = \dfrac{\partial x^\alpha}{\partial x^{\mu'}}\dfrac{\partial x^\beta}{\partial x^{\nu'}}g_{\alpha\beta}$ . Why: these are the definitions of a contravariant vector and a covariant rank-2 tensor, established in aq-gr-tensor-transformation-law, and everything else follows mechanically from them.

Step 2. Lower in the primed frame by substituting both rules: $V_{\mu'} = g_{\mu'\nu'}V^{\nu'} = \dfrac{\partial x^\alpha}{\partial x^{\mu'}}\dfrac{\partial x^\beta}{\partial x^{\nu'}}g_{\alpha\beta}\,\dfrac{\partial x^{\nu'}}{\partial x^\nu}V^\nu$ . Why: this puts every transformation factor on the table so the redundant ones can be contracted away.

Step 3. Collapse the chain rule on the summed index $\nu'$ : $\dfrac{\partial x^\beta}{\partial x^{\nu'}}\dfrac{\partial x^{\nu'}}{\partial x^\nu} = \dfrac{\partial x^\beta}{\partial x^\nu} = \delta^\beta{}_\nu$ . Why: a Jacobian times its inverse is the identity, which is what removes the primed dummy index and leaves a clean expression.

Step 4. Read off the result: $V_{\mu'} = \dfrac{\partial x^\alpha}{\partial x^{\mu'}}\,g_{\alpha\beta}\,\delta^\beta{}_\nu V^\nu = \dfrac{\partial x^\alpha}{\partial x^{\mu'}}\,g_{\alpha\nu}V^\nu = \dfrac{\partial x^\alpha}{\partial x^{\mu'}}\,V_\alpha$ . Why: this is exactly the covariant transformation law for a one-form, so $V_\mu$ is a bona fide rank-(0,1) tensor and lowering is a tensorial operation. *The same argument applies when raising an index using $g^{\mu\nu}$ , which transforms with two forward Jacobians.*

Go deeper · Derivation C · 1 minLowering then raising a slot returns the original tensor exactly

Derivation CLowering then raising a slot returns the original tensor exactly

The goal is to show that $\flat$ and $\sharp$ acting on the same slot are perfect inverses, so no information is lost when you move an index down and back up — the round trip is the identity on each slot independently. We take a rank-(1,1) tensor $T^\mu{}_\nu$ , lower its upper index to a rank-(0,2) tensor, raise that index again, and use the inverse-metric relation to collapse the two contractions to a Kronecker delta. This is the algebraic guarantee behind treating $T^{\mu\nu}$ , $T^\mu{}_\nu$ , and $T_{\mu\nu}$ as one object.

Step 1. Lower the upper index by contracting with the metric: $T_{\alpha\nu} := g_{\alpha\mu}\,T^\mu{}_\nu$ , a rank-(0,2) tensor. Why: this is the flat operation applied to a single chosen slot, leaving the second index a spectator.

Step 2. Raise the same index back by contracting with the inverse metric: $g^{\rho\alpha}\,T_{\alpha\nu} = g^{\rho\alpha}g_{\alpha\mu}\,T^\mu{}_\nu$ . Why: sharp is applied to exactly the index that flat moved, which sets up the inverse-metric product that is about to simplify.

Step 3. Apply the defining relation $g^{\rho\alpha}g_{\alpha\mu} = \delta^\rho{}_\mu$ from equation (3). Why: the inverse metric was constructed precisely so this product is the identity, which is the pivot of the whole argument.

Step 4. Contract the delta: $\delta^\rho{}_\mu\,T^\mu{}_\nu = T^\rho{}_\nu$ , the tensor we started with. Why: the original component is recovered untouched, proving $\sharp\circ\flat = \mathrm{id}$ slot by slot — index placement is reversible bookkeeping, never a change of the underlying object. *The spectator index $\nu$ never participated, confirming the slots move independently.*

Limiting cases

Case	Result	Meaning
Euclidean Cartesian	$g_{\mu\nu}=\delta_{\mu\nu}$	Lowering changes nothing: $V_\mu = V^\mu$ . Vectors and one-forms coincide numerically — why elementary physics never distinguishes them.
Minkowski, mostly-plus	$g_{\mu\nu}=\mathrm{diag}(-1,+1,+1,+1)$	$V_0 = -V^0$ , $V_i = V^i$ . Only the time component flips sign; spatial components are untouched.
Diagonal metric	$V_\mu = g_{\mu\mu}V^\mu$ (no sum)	Each component is rescaled by its own diagonal entry, with no mixing between components.
Off-diagonal metric	$V_\mu = g_{\mu\nu}V^\nu$ (full sum)	Components genuinely mix; the lowered array is a linear combination of all raised components.
Metric self-check	$g^{\mu}{}_{\nu}=\delta^{\mu}{}_{\nu}$	Raising one index of the metric gives the identity — the mixed metric carries no information.

Common trapThe inverse metric is the matrix inverse of the metric, not a lowered or raised version of it

A common beginner error is to imagine $g^{\mu\nu}$ is obtained by "raising the indices" of $g_{\mu\nu}$ — but that would be circular, because raising already requires the inverse metric to be defined. The inverse metric is fixed independently as the matrix inverse: $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ . Concretely, lower both indices of $g^{\mu\nu}$ and you get $g^{\alpha\beta}g_{\alpha\mu}g_{\beta\nu} = g_{\mu\nu}$ right back — never anything new; and raising just one index gives $g^\mu{}_\nu = \delta^\mu{}_\nu$ , the Kronecker delta, not a fresh object either. So there are exactly three faces of the metric: $g_{\mu\nu}$ , its matrix inverse $g^{\mu\nu}$ , and the identity $\delta^\mu{}_\nu$ between them. Treat the "mixed metric" as a new tensor and you will double-count degrees of freedom.

Worked examples

1 · Lowering a four-momentum and reading off the invariant mass (numeric)

A particle in Minkowski spacetime (signature $-,+,+,+$ , units $c=1$ ) has rest mass $m=2$ and four-momentum $p^\mu = (E, p_x, 0, 0)$ with spatial momentum $p_x = 1.5$ . Lower the index, form the invariant $p^\mu p_\mu$ , and check the mass-shell condition.

Step 1. Fix the energy from the mass shell. The on-shell energy is $E = \sqrt{p_x^2 + m^2} = \sqrt{1.5^2 + 2^2} = \sqrt{2.25 + 4} = \sqrt{6.25} = \mathbf{2.5}$ , so $p^\mu = (2.5,\,1.5,\,0,\,0)$ .

Step 2. Lower with $g_{\mu\nu}=\mathrm{diag}(-1,1,1,1)$ : only the time slot flips, giving $p_0 = g_{00}p^0 = -2.5$ and $p_i = p^i$ . Hence $p_\mu = \mathbf{(-2.5,\,1.5,\,0,\,0)}$ — the same four-momentum, read as a one-form.

Step 3. Contract to form the scalar: $p^\mu p_\mu = (2.5)(-2.5) + (1.5)(1.5) = -6.25 + 2.25 = \mathbf{-4}$ . Compare with $-m^2 = -4$ . The invariant equals $-m^2$ exactly, confirming the mass-shell relation $p^\mu p_\mu = -m^2$ . The lowering plus contraction reproduces the invariant mass with no reference to the observer's frame — that frame-independence is the entire point.

2 · Lowering in a curved metric: Schwarzschild at r = 2r_s (numeric)

In Schwarzschild coordinates the metric is diagonal with $g_{tt} = -(1 - r_s/r)$ and $g_{rr} = (1 - r_s/r)^{-1}$ , where $r_s$ is the Schwarzschild radius. Evaluate the lowering of a vector $V^\mu = (1, 1, 0, 0)$ (components in $t$ and $r$ ) at $r = 2r_s$ , and compute its squared norm.

Step 1. Evaluate the metric factor at $r = 2r_s$ : $1 - r_s/r = 1 - \tfrac{1}{2} = \tfrac12$ . So $g_{tt} = -\tfrac12$ and $g_{rr} = (\tfrac12)^{-1} = 2$ . The two diagonal entries are reciprocals in magnitude, a hallmark of the Schwarzschild form.

Step 2. Lower each component with $V_\mu = g_{\mu\mu}V^\mu$ (no sum, diagonal metric): $V_t = g_{tt}V^t = (-\tfrac12)(1) = \mathbf{-0.5}$ and $V_r = g_{rr}V^r = (2)(1) = \mathbf{2}$ . The raised array $(1,1)$ becomes the lowered array $(-0.5,\,2)$ — a vivid reminder that the two look nothing alike once the metric is non-trivial.

Step 3. Form the squared norm $\lVert V\rVert^2 = V^\mu V_\mu = V^t V_t + V^r V_r = (1)(-0.5) + (1)(2) = -0.5 + 2 = \mathbf{1.5}$ . It is positive, so this particular $V$ is spacelike at $r=2r_s$ : the radial contribution outweighs the temporal one here. The same components would give a different norm at a different radius, because the metric that measures them is itself a function of position.

3 · Raising an index of a rank-2 tensor and taking its trace (numeric)

Work in 2D Minkowski, $g = \mathrm{diag}(-1, 1)$ , so $g^{-1} = \mathrm{diag}(-1, 1)$ . Given the symmetric contravariant tensor $T^{\mu\nu} = \left(\begin{smallmatrix} 2 & 3 \\ 3 & 5 \end{smallmatrix}\right)$ , lower its second index to get $T^\mu{}_\nu$ , read off the trace, and confirm the round trip.

Step 1. Lower the second index: $T^\mu{}_\nu = T^{\mu\alpha}g_{\alpha\nu}$ . Since $g$ is diagonal this multiplies column $0$ by $g_{00}=-1$ and leaves column $1$ unchanged, giving $T^\mu{}_\nu = \left(\begin{smallmatrix} -2 & 3 \\ -3 & 5 \end{smallmatrix}\right)$ . Note it is no longer symmetric — a mixed tensor need not be, even when its parent was.

Step 2. Take the trace on the mixed indices: $T^\mu{}_\mu = (-2) + 5 = \mathbf{3}$ . Cross-check against the direct contraction $g_{\mu\nu}T^{\mu\nu} = -T^{00} + T^{11} = -2 + 5 = 3$ . The two agree, as they must — tracing a mixed tensor is the same as contracting the fully upstairs tensor with the metric.

Step 3. Confirm the round trip by raising the index back with $g^{\nu\beta}$ : multiplying column $0$ of $T^\mu{}_\nu$ by $g^{00}=-1$ restores the original $T^{\mu\nu}$ . Nothing was lost: the trace scalar $3$ is invariant, and $\sharp\circ\flat$ returns the original tensor exactly, as the third derivation guaranteed.

Physics of the portrait

Raising and lowering indices is the visible surface of a much larger machine — the absolute differential calculus, better known today as tensor calculus. Its inventor turned a scattered set of nineteenth-century techniques into a systematic language in which the metric, and the operation of moving indices with it, is the central grammatical rule.

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

Gregorio Ricci-Curbastro

1853–1925 · Italian mathematician, founder of tensor calculus

Gregorio Ricci-Curbastro spent the better part of two decades at the University of Padua building what he called the absolute differential calculus — a way of writing geometric and physical laws so that they hold in every coordinate system at once. Starting in the 1880s he generalized Riemann's metric and Christoffel's transformation of quadratic differential forms into a complete formalism of covariant and contravariant components, in which the metric $g_{\mu\nu}$ and its inverse $g^{\mu\nu}$ are the tools that convert between the two. Raising and lowering indices, the musical isomorphisms, are exactly his device for keeping equations coordinate-free while still computing with concrete components. With his former student Tullio Levi-Civita he published the definitive account in 1900, Méthodes de calcul différentiel absolu et leurs applications, in the Mathematische Annalen — a paper almost no physicist read at the time. Its moment came fifteen years later, when Einstein, guided by his friend Marcel Grossmann, found in Ricci's calculus the only mathematics capable of expressing gravity as spacetime curvature. The Ricci tensor $R_{\mu\nu}$ on the left of the Einstein field equations carries his name, and every time a relativist raises an index they speak the language he built — though Ricci-Curbastro himself, a quiet and devout man, never worked on relativity.

History

1854

— Riemann founds the intrinsic geometry of manifolds and the metric in his habilitation lecture on the hypotheses underlying geometry

1869

— Christoffel introduces his symbols and the covariant transformation of quadratic differential forms

1900

— Ricci-Curbastro and Levi-Civita codify the absolute differential calculus in the Mathematische Annalen

1908

— Minkowski recasts special relativity as the geometry of a spacetime with an indefinite metric

1916

— Einstein deploys the Ricci calculus to write general relativity and adopts the summation convention on repeated indices

Going further

Texts. Misner, Thorne & Wheeler, Gravitation, §3.2–3.5 (index gymnastics and the metric) · Wald, General Relativity, §2.3 (the metric and its inverse, abstract-index form) · Carroll, Spacetime and Geometry, §2.5 (raising and lowering, musical isomorphisms) · Schutz, A First Course in General Relativity, ch. 3 (components and the metric) · Hartle, Gravity, ch. 8 (the metric in practice). The key identities and anchor numbers are: (1) the inverse-metric relation $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ and the round trip $\sharp\circ\flat=\mathrm{id}$ ; (2) the covariant transformation law for $V_\mu$ via the chain rule; (3) the Minkowski example $V^\mu=(2,1,0,0)\to V_\mu=(-2,1,0,0)$ and the four-momentum invariant $p^\mu p_\mu=-m^2$ with $m=2,\,p_x=1.5,\,E=2.5$ ; (4) the Schwarzschild lowering at $r=2r_s$ giving $g_{tt}=-\tfrac12,\,g_{rr}=2$ , hence $V_\mu=(-0.5,2,0,0)$ and norm $1.5$ ; (5) the 2D raise/trace example with $T^{\mu\nu}=\left(\begin{smallmatrix}2&3\\3&5\end{smallmatrix}\right)$ , trace $3$ .

Builds on. The Metric Tensor and the Line Element · Tangent Vectors and the Tangent Space · Cotangent Space and One-Forms · General Tensors and the Tensor Product · The Tensor Transformation Law. Leads to. Tensor Algebra: Contraction, Symmetrization and Antisymmetrization · Metric Signature and Lorentzian Geometry · The Invariant Volume Element.

Ask the tutor: Want to see how these same musical isomorphisms turn the Riemann tensor's many index arrangements into one object, or how the norm's sign builds the light cone? Ask, and we will work it through.

Are V^μ and V_μ two different vectors, or the same one?

They are the same geometric object written in two different spaces. $V^\mu$ lives in the tangent space as an arrow; $V_\mu$ lives in the cotangent space as a stack of level surfaces. The metric is the dictionary that pairs them uniquely, and because it is invertible you can always go back and forth without losing anything. What differs is only the component array: in a non-trivial metric the numbers $V^\mu$ and $V_\mu$ look unlike each other — in Minkowski the time component flips sign, in Schwarzschild each component is rescaled. But they encode identical physical information, so it is fair to call them the vector and its metric-dual one-form, not two independent vectors.

Why can't I just sum V^μ V^μ over two upper indices?

Because summation that produces an invariant must pair a contravariant index with a covariant one, and $V^\mu V^\mu$ pairs two contravariant indices. Geometrically, a vector can only be fed to a one-form — it needs a machine that eats arrows and returns numbers, and a second arrow is not such a machine. The Einstein summation convention deliberately signals a valid contraction only when one index is up and one is down for exactly this reason. To get a number you must first lower one index with the metric: $V^\mu V_\mu = g_{\mu\nu}V^\mu V^\nu$ . Written with two upper indices the expression is not coordinate-invariant and, strictly, is not defined at all without an implied metric.

Does raising or lowering an index ever lose information?

No — the operations are exact inverses, so a round trip returns the original tensor untouched. This is guaranteed by $g^{\mu\alpha}g_{\alpha\nu}=\delta^\mu{}_\nu$ : lower a slot with $g$ , raise it back with $g^{-1}$ , and the two metrics collapse to the identity. The correspondence could fail only if the metric were degenerate ( $\det g = 0$ ), where the inverse metric does not exist. On any genuine non-degenerate metric — Riemannian or Lorentzian — every index moves down and up freely, on any slot, in any order, with the underlying object preserved.

What is the musical isomorphism, and why the flat and sharp symbols?

It is the coordinate-free name for raising and lowering. Lowering a vector to a one-form is written $V^\flat$ , using the flat symbol $\flat$ because it "lowers the pitch" of the index — pushes it down. Raising a one-form to a vector is $\omega^\sharp$ , the sharp symbol $\sharp$ "raising the pitch" — pulling the index up. The two together, $\flat:T_pM\to T_p^*M$ and $\sharp:T_p^*M\to T_pM$ , are inverse isomorphisms, and the musical metaphor is a memory aid for which one does which. In components $\flat$ is contraction with $g_{\mu\nu}$ and $\sharp$ is contraction with $g^{\mu\nu}$ ; the abstract notation just emphasizes that the maps exist independently of any coordinate choice.

Why does the mixed metric carry no information?

Because raising one index of the metric is, by the very definition of the inverse metric, the identity: $g^\mu{}_\nu = g^{\mu\alpha}g_{\alpha\nu} = \delta^\mu{}_\nu$ . The inverse metric was constructed precisely so that this product equals the Kronecker delta, so the "mixed metric" is not a new tensor — it is the identity map wearing metric clothing. This is a useful consistency check: if a calculation ever produces a $g^\mu{}_\nu$ , you may replace it by $\delta^\mu{}_\nu$ on sight. It also explains why there are only three genuinely distinct faces of the metric — $g_{\mu\nu}$ , $g^{\mu\nu}$ , and the delta between them — rather than four.

In a Lorentzian metric, why does only the time component flip sign?

Because in Minkowski coordinates the metric is diagonal, $g_{\mu\nu}=\mathrm{diag}(-1,+1,+1,+1)$ , so lowering multiplies each component by its own diagonal entry with no mixing: $V_\mu = g_{\mu\mu}V^\mu$ (no sum). The spatial entries are $+1$ , so $V_1,V_2,V_3$ are unchanged; the time entry is $-1$ , so $V_0 = -V^0$ . This single sign flip is the algebraic root of the causal structure of relativity — it is why the norm $V^\mu V_\mu = -(V^0)^2 + \lvert\vec V\rvert^2$ can be negative (timelike), zero (null), or positive (spacelike), instead of a sum of squares as in Euclidean space. In curved coordinates the metric is no longer this simple diagonal, and lowering generally rescales and mixes all components.