Musical isomorphism

In mathematics—more specifically, in differential geometry—the musical isomorphism (or canonical isomorphism) is an isomorphism between the tangent bundle $\mathrm{T}M$ and the cotangent bundle ${\mathrm{T}}^{*}M$ of a Riemannian or pseudo-Riemannian manifold induced by its metric tensor. There are similar isomorphisms on symplectic manifolds. These isomorphisms are global versions of the canonical isomorphism between an inner product space and its dual. The term musical refers to the use of the musical notation symbols $\mathrm{♭}$ (flat) and $\mathrm{♯}$ (sharp).Template:SfnTemplate:Sfn

In the notation of Ricci calculus and mathematical physics, the idea is expressed as the raising and lowering of indices. Raising and lowering indices are a form of index manipulation in tensor expressions.

In certain specialized applications, such as on Poisson manifolds, the relationship may fail to be an isomorphism at singular points, and so, for these cases, is technically only a homomorphism.

Motivation

In linear algebra, a finite-dimensional vector space is isomorphic to its dual space (the space of linear functionals mapping the vector space to its base field), but not canonically isomorphic to it. This is to say that given a fixed basis for the vector space, there is a natural way to go back and forth between vectors and linear functionals: vectors are represented in the basis by column vectors, and linear functionals are represented in the basis by row vectors, and one can go back and forth by transposing. However, without a fixed basis, there is no way to go back and forth between vectors and linear functionals. This is what is meant by that there is no canonical isomorphism.

On the other hand, a finite-dimensional vector space $V$ endowed with a non-degenerate bilinear form $⟨\cdot ,\cdot ⟩$ is canonically isomorphic to its dual. The canonical isomorphism $V\to V^{*}$ is given by

$v\mapsto ⟨v,\cdot ⟩$.

The non-degeneracy of $⟨\cdot ,\cdot ⟩$ means exactly that the above map is an isomorphism. An example is where $V={\mathbb{ℝ}}^n$ and $⟨\cdot ,\cdot ⟩$ is the dot product.

In a basis $e_i$, the canonical isomorphism above is represented as follows. Let $g_{ij}=⟨e_i,e_j⟩$ be the components of the non-degenerate bilinear form and let $g^{ij}$ be the components of the inverse matrix to $g_{ij}$. Let $e^i$ be the dual basis of $e_i$. A vector $v$ is written in the basis as $v=v^i{e}_i$ using Einstein summation notation, i.e., $v$ has components $v^i$ in the basis. The canonical isomorphism applied to $v$ gives an element of the dual, which is called a covector. The covector has components $v_i$ in the dual basis given by contracting with $g$:

$v_i=g_{ij}{v}^j.$

This is what is meant by lowering the index. Conversely, contracting a covector $\alpha ={\alpha }_i{e}^i$ with the inverse of $g$ gives a vector with components

${\alpha }^i=g^{ij}{\alpha }_j.$

in the basis $e_i$. This process is called raising the index.

Raising and then lowering the same index (or conversely) are inverse operations, which is reflected in $g_{ij}$ and $g^{ij}$ being inverses:

$g^{ij}{g}_{jk}=g_{kj}{g}^{ji}={{\delta }^i}_k={{\delta }_k}^i$

where ${\delta }_j^i$ is the Kronecker delta or identity matrix.

The musical isomorphisms are the global version of the canonical isomorphism $v\mapsto ⟨v,\cdot ⟩$ and its inverse for the tangent bundle and cotangent bundle of a (pseudo-)Riemannian manifold $(M,g)$. They are canonical isomorphisms of vector bundles which are at any point Template:Math the canonical isomorphism applied to the tangent space of Template:Math at Template:Math endowed with the inner product $g_p$.

Because every smooth manifold can be (non-canonically) endowed with a Riemannian metric, the musical isomorphisms show that a vector bundle on a smooth manifold is (non-canonically) isomorphic to its dual.

Discussion

Let Template:Math be a (pseudo-)Riemannian manifold. At each point Template:Mvar, the map Template:Math is a non-degenerate bilinear form on the tangent space Template:Math. If Template:Mvar is a vector in Template:Math, its flat is the covector

$v^{\mathrm{♭}}=g_p(v,\cdot )$

in Template:Math. Since this is a smooth map that preserves the point Template:Mvar, it defines a morphism of smooth vector bundles $\mathrm{♭}:\mathrm{T}M\to {\mathrm{T}}^{*}M$. By non-degeneracy of the metric, $\mathrm{♭}$ has an inverse $\mathrm{♯}$ at each point, characterized by

$g_p({\alpha }^{\mathrm{♯}},v)=\alpha (v)$

for Template:Mvar in Template:Math and Template:Mvar in Template:Math. The vector ${\alpha }^{\mathrm{♯}}$ is called the sharp of Template:Mvar. The sharp map is a smooth bundle map $\mathrm{♯}:{\mathrm{T}}^{*}M\to \mathrm{T}M$.

Flat and sharp are mutually inverse isomorphisms of smooth vector bundles, hence, for each Template:Mvar in Template:Mvar, there are mutually inverse vector space isomorphisms between Template:Math and Template:Math.

The flat and sharp maps can be applied to vector fields and covector fields by applying them to each point. Hence, if Template:Mvar is a vector field and Template:Mvar is a covector field,

$X^{\mathrm{♭}}=g(X,\cdot )$

and

$g({\omega }^{\mathrm{♯}},X)=\omega (X)$.

In a moving frame

Suppose Template:Math is a moving tangent frame (see also smooth frame) for the tangent bundle Template:Math with, as dual frame (see also dual basis), the moving coframe (a moving tangent frame for the cotangent bundle ${\mathrm{T}}^{*}M$; see also coframe) Template:Math. Then the pseudo-Riemannian metric, which is a 2-covariant tensor field, can be written locally in this coframe as Template:Math using Einstein summation notation.

Given a vector field Template:Math and denoting Template:Math, its flat is

$X^{\mathrm{♭}}=g_{ij}{X}^i{\mathbf{𝐞}}^j=X_j{\mathbf{𝐞}}^j$.

This is referred to as lowering an index, because the components of Template:Mvar are written with an upper index Template:Math, whereas the components of $X^{\mathrm{♭}}$ are written with a lower index Template:Math.

In the same way, given a covector field Template:Math and denoting Template:Math, its sharp is

${\omega }^{\mathrm{♯}}=g^{ij}{\omega }_i{\mathbf{𝐞}}_j={\omega }^j{\mathbf{𝐞}}_j$,

where Template:Math are the components of the inverse metric tensor (given by the entries of the inverse matrix to Template:Math). Taking the sharp of a covector field is referred to as raising an index.

Extension to tensor products

The musical isomorphisms may also be extended, for each Template:Math, to an isomorphism between the bundle

$\underset{i=1}{\overset{s}{⨂}}\mathrm{T}M\otimes \underset{j=1}{\overset{r}{⨂}}{\mathrm{T}}^{*}M$

of $(r,s)$ tensors and the bundle of $(r-k,s+k)$ tensors. Here Template:Mvar can be positive or negative, so long as Template:Math and Template:Math.

Lowering an index of an $(r,s)$ tensor gives a $(r-1,s+1)$ tensor, while raising an index gives a $(r+1,s-1)$. Which index is to be raised or lowered must be indicated.

For instance, consider the (0, 2) tensor Template:Math. Raising the second index, we get the (1, 1) tensor

$X^{\mathrm{♯}}=g^{jk}{X}_{ij}{\mathrm{e}}^i\otimes {\mathrm{e}}_k.$

In other words, the components $X_i^k$ of $X^{\mathrm{♯}}$ are given by

$X_i^k=g^{jk}{X}_{ij}.$

Similar formulas are available for tensors of other orders. For example, for a $(0,n)$ tensor Template:Mvar, all indices are raised by:

$X^{j_1{j}_2\cdots j_n}=g^{j_1{i}_1}{g}^{j_2{i}_2}\cdots g^{j_n{i}_n}{X}_{i_1{i}_2\cdots i_n}.$

For a $(n,0)$ tensor Template:Mvar, all indices are lowered by:

$X_{j_1{j}_2\cdots j_n}=g_{j_1{i}_1}{g}_{j_2{i}_2}\cdots g_{j_n{i}_n}{X}^{i_1{i}_2\cdots i_n}.$

For a mixed tensor of order $(n,m)$, all lower indices are raised and all upper indices are lowered by

${X_{p_1{p}_2\cdots p_n}}^{q_1{q}_2\cdots q_m}=g_{p_1{i}_1}{g}_{p_2{i}_2}\cdots g_{p_n{i}_n}{g}^{q_1{j}_1}{g}^{q_2{j}_2}\cdots g^{q_m{j}_m}{X^{i_1{i}_2\cdots i_n}}_{j_1{j}_2\cdots j_m}.$

Well-formulated expressions are constrained by the rules of Einstein summation notation: any index may appear at most twice and furthermore a raised index must contract with a lowered index. With these rules we can immediately see that an expression such as $g_{ij}{v}^i{u}^j$ is well formulated while $g_{ij}{v}_i{u}_j$ is not.

Extension to k-vectors and k-forms

In the context of exterior algebra, an extension of the musical operators may be defined on Template:Math and its dual Template:Math, and are again mutual inverses:Template:Sfn

$\mathrm{♭}:\underset{i=1}{\overset{k}{\bigwedge }}V\to \underset{i=1}{\overset{k}{\bigwedge }}{V}^{*},$

$\mathrm{♯}:\underset{i=1}{\overset{k}{\bigwedge }}{V}^{*}\to \underset{i=1}{\overset{k}{\bigwedge }}V,$

defined by

$(X\wedge \dots \wedge Z{)}^{\mathrm{♭}}=X^{\mathrm{♭}}\wedge \dots \wedge Z^{\mathrm{♭}},$

$(\alpha \wedge \dots \wedge \gamma {)}^{\mathrm{♯}}={\alpha }^{\mathrm{♯}}\wedge \dots \wedge {\gamma }^{\mathrm{♯}}.$

In this extension, in which Template:Math maps k-vectors to k-covectors and Template:Math maps k-covectors to k-vectors, all the indices of a totally antisymmetric tensor are simultaneously raised or lowered, and so no index need be indicated: $$Y^{\mathrm{♯}}=(Y_{i_1\dots i_j}{\mathbf{𝐞}}^{i_1}\otimes \dots \otimes {\mathbf{𝐞}}^{i_j}{)}^{\mathrm{♯}}=g^{i_1{r}_1}\dots g^{i_j{r}_s}{Y}_{i_1\dots i_k}{\mathbf{𝐞}}_{r_1}\otimes \dots \otimes {\mathbf{𝐞}}_{r_s}.$$

This works not just for k-vectors in the context of linear algebra but also for k-forms in the context of a (pseudo-)Riemannian manifold:

$\mathrm{♭}:\underset{i=1}{\overset{k}{\bigwedge }}\mathrm{T}M\to \underset{i=1}{\overset{k}{\bigwedge }}{\mathrm{T}}^{*}M,$

$\mathrm{♯}:\underset{i=1}{\overset{k}{\bigwedge }}{\mathrm{T}}^{*}M\to \underset{i=1}{\overset{k}{\bigwedge }}\mathrm{T}M,$

Vector bundles with bundle metrics

More generally, musical isomorphisms always exist between a vector bundle endowed with a bundle metric and its dual.

Trace of a tensor

Given a (0, 2) tensor Template:Math, we define the trace of Template:Mvar through the metric tensor Template:Mvar by $${\mathrm{tr}}_g(X):=\mathrm{tr}(X^{\mathrm{♯}})=\mathrm{tr}(g^{jk}{X}_{ij}{\mathbf{𝐞}}^i\otimes {\mathbf{𝐞}}_k)=g^{ij}{X}_{ij}.$$

Observe that the definition of trace is independent of the choice of index to raise, since the metric tensor is symmetric.

The trace of an $(r,s)$ tensor can be taken in a similar way, so long as one specifies which two distinct indices are to be traced. This process is also called contracting the two indices. For example, if Template:Mvar is an $(r,s)$ tensor with Template:Math, then the indices $i_1$ and $i_2$ can be contracted to give an $(r-2,s)$ tensor with components

$X_{j_1{j}_2\cdots j_s}^{i_3{i}_4\cdots i_r}=g_{i_1{i}_2}{X}_{j_1{j}_2\cdots j_s}^{i_1{i}_2\cdots i_r}.$

Example computations

In Minkowski spacetime

The covariant 4-position is given by

$X_{\mu }=(-ct,x,y,z)$

with components:

$X_0=-ct,X_1=x,X_2=y,X_3=z$

(where Template:Mvar,Template:Mvar,Template:Mvar are the usual Cartesian coordinates) and the Minkowski metric tensor with metric signature (− + + +) is defined as

${\eta }_{\mu \nu }={\eta }^{\mu \nu }=\left(\begin{array}{cccc}-1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right)$

in components:

${\eta }_{00}=-1,{\eta }_{i0}={\eta }_{0i}=0,{\eta }_{ij}={\delta }_{ij}(i,j\ne 0).$

To raise the index, multiply by the tensor and contract:

$X^{\lambda }={\eta }^{\lambda \mu }{X}_{\mu }={\eta }^{\lambda 0}{X}_0+{\eta }^{\lambda i}{X}_i$

then for Template:Math:

$X^0={\eta }^{00}{X}_0+{\eta }^{0i}{X}_i=-X_0$

and for Template:Math:

$X^j={\eta }^{j0}{X}_0+{\eta }^{ji}{X}_i={\delta }^{ji}{X}_i=X_j.$

So the index-raised contravariant 4-position is:

$X^{\mu }=(ct,x,y,z).$

This operation is equivalent to the matrix multiplication

$\left(\begin{array}{cccc}-1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right)\left(\begin{array}{c}-ct\\ x\\ y\\ z\end{array}\right)=\left(\begin{array}{c}ct\\ x\\ y\\ z\end{array}\right).$

Given two vectors, $X^{\mu }$ and $Y^{\mu }$, we can write down their (pseudo-)inner product in two ways:

${\eta }_{\mu \nu }{X}^{\mu }{Y}^{\nu }.$

By lowering indices, we can write this expression as

$X_{\mu }{Y}^{\mu }.$

In matrix notation, the first expression can be written as

$\left(\begin{array}{cccc}{X}^0& X^1& X^2& X^3\end{array}\right)\left(\begin{array}{cccc}-1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right)\left(\begin{array}{c}{Y}^0\\ Y^1\\ Y^2\\ Y^3\end{array}\right)$

while the second is, after lowering the indices of $X^{\mu }$,

$\left(\begin{array}{cccc}-X^0& X^1& X^2& X^3\end{array}\right)\left(\begin{array}{c}{Y}^0\\ Y^1\\ Y^2\\ Y^3\end{array}\right).$

In electromagnetism

For a (0,2) tensor, twice contracting with the inverse metric tensor and contracting in different indices raises each index:

$A^{\mu \nu }=g^{\mu \rho }{g}^{\nu \sigma }{A}_{\rho \sigma }.$

Similarly, twice contracting with the metric tensor and contracting in different indices lowers each index:

$A_{\mu \nu }=g_{\mu \rho }{g}_{\nu \sigma }{A}^{\rho \sigma }$

Let's apply this to the theory of electromagnetism.

The contravariant electromagnetic tensor in the Template:Math signature is given by

$F^{\alpha \beta }=\left(\begin{array}{cccc}0& -\frac{E_x}{c}& -\frac{E_y}{c}& -\frac{E_z}{c}\\ \frac{E_x}{c}& 0& -B_z& B_y\\ \frac{E_y}{c}& B_z& 0& -B_x\\ \frac{E_z}{c}& -B_y& B_x& 0\end{array}\right).$

In components,

$F^{0i}=-F^{i0}=-\frac{E^i}{c},F^{ij}=-{\epsilon }^{ijk}{B}_k$

To obtain the covariant tensor Template:Mvar, contract with the inverse metric tensor:

$\begin{array}{rl}{F}_{\alpha \beta }& ={\eta }_{\alpha \gamma }{\eta }_{\beta \delta }{F}^{\gamma \delta }\\ & ={\eta }_{\alpha 0}{\eta }_{\beta 0}{F}^{00}+{\eta }_{\alpha i}{\eta }_{\beta 0}{F}^{i0}+{\eta }_{\alpha 0}{\eta }_{\beta i}{F}^{0i}+{\eta }_{\alpha i}{\eta }_{\beta j}{F}^{ij}\end{array}$

and since Template:Math and Template:Math, this reduces to

$F_{\alpha \beta }=({\eta }_{\alpha i}{\eta }_{\beta 0}-{\eta }_{\alpha 0}{\eta }_{\beta i})F^{i0}+{\eta }_{\alpha i}{\eta }_{\beta j}{F}^{ij}$

Now for Template:Math, Template:Math:

$\begin{array}{rl}{F}_{0k}& =({\eta }_{0i}{\eta }_{k0}-{\eta }_{00}{\eta }_{ki})F^{i0}+{\eta }_{0i}{\eta }_{kj}{F}^{ij}\\ & =(0-(-{\delta }_{ki}))F^{i0}+0\\ & =F^{k0}=-F^{0k}\\ \end{array}$

and by antisymmetry, for Template:Math, Template:Math:

$F_{k0}=-F^{k0}$

then finally for Template:Math, Template:Math;

$\begin{array}{rl}{F}_{kl}& =({\eta }_{ki}{\eta }_{l0}-{\eta }_{k0}{\eta }_{li})F^{i0}+{\eta }_{ki}{\eta }_{lj}{F}^{ij}\\ & =0+{\delta }_{ki}{\delta }_{lj}{F}^{ij}\\ & =F^{kl}\\ \end{array}$

The (covariant) lower indexed tensor is then:

$F_{\alpha \beta }=\left(\begin{array}{cccc}0& \frac{E_x}{c}& \frac{E_y}{c}& \frac{E_z}{c}\\ -\frac{E_x}{c}& 0& -B_z& B_y\\ -\frac{E_y}{c}& B_z& 0& -B_x\\ -\frac{E_z}{c}& -B_y& B_x& 0\end{array}\right)$

This operation is equivalent to the matrix multiplication

$\left(\begin{array}{cccc}-1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right)\left(\begin{array}{cccc}0& -\frac{E_x}{c}& -\frac{E_y}{c}& -\frac{E_z}{c}\\ \frac{E_x}{c}& 0& -B_z& B_y\\ \frac{E_y}{c}& B_z& 0& -B_x\\ \frac{E_z}{c}& -B_y& B_x& 0\end{array}\right)\left(\begin{array}{cccc}-1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right)=\left(\begin{array}{cccc}0& \frac{E_x}{c}& \frac{E_y}{c}& \frac{E_z}{c}\\ -\frac{E_x}{c}& 0& -B_z& B_y\\ -\frac{E_y}{c}& B_z& 0& -B_x\\ -\frac{E_z}{c}& -B_y& B_x& 0\end{array}\right).$

See also

• Duality (mathematics)
• Template:Section link
• Einstein notation
• Flat (music) and Sharp (music) about the signs Template:Music and Template:Music
• Hodge star operator
• Metric tensor
• Vector bundle