## Relative tensor and Levi-Civita symbol
### The volume element
Let $Z = |Z_{ij}|$ be the determinant of the covariant metric tensor $Z_{ij}$. The **volume element** is defined as:
$
\sqrt{Z}
$
Although $Z$ is **not a tensor**, it transforms according to a specific rule, making it a **[[#Relative tensors|relative tensor]] of weight 2**:
$
Z' = |Z_{i'j'}| = |Z_{ij} J^i_{i'} J^j_{j'}| = Z J^2
$
where $J = \left| J^\cdot_{\cdot'} \right|$ is the determinant of the Jacobian matrix, $Z=|Z_{\cdot \cdot}|$ and $Z'=|Z_{\cdot' \cdot'}|$. Since the inverse satisfies
$
J^{-1} = \left| J^{i'}_i \right|,
$
we also find that:
$
\sqrt{Z'} = |J| \sqrt{Z}
$
In Cartesian coordinates, $Z = 1$, so we recover the classical result $\sqrt{Z'} = |J| = \text{absolute value}(J)$.
> [!Notice]
> This identity holds because the inner product in Cartesian coordinates is Euclidean, i.e., $Z_{ij} = \delta_{ij}$.
> As a result, the determinant of the metric is identically $1$, and the volume element reduces to the Jacobian determinant.
> In this sense, Cartesian coordinates are geometrically special, serving as a baseline where the volume element becomes trivial.
> [!Note]
> **Geometric meaning**
> The volume element $\sqrt{Z}$ measures how much a coordinate system stretches or contracts space locally:
> - In 3D: $\sqrt{Z}$ gives volume scaling
> - In 2D: $\sqrt{Z}$ gives area
> - In 1D: $\sqrt{Z}$ gives arc length
>
> It also determines the correct integration measure $\mathrm{d}V = \sqrt{Z} \, \mathrm{d}x^1 \dots \mathrm{d}x^n$.
> [!Example]-
> Examples in common coordinates
>
> - Cartesian coordinates:
> $
> \sqrt{Z} = 1
> $
> - Polar / Cylindrical coordinates:
> $
> Z_{ij} =
> \begin{pmatrix}
> 1 & 0 \\
> 0 & r^2
> \end{pmatrix}
> \quad \Rightarrow \quad
> \sqrt{Z} = \sqrt{1 \cdot r^2} = r
> $
>
> - Spherical coordinates:
> $
>
> Z_{ij} =
> \begin{pmatrix}
> 1 & 0 & 0 \\
> 0 & r^2 & 0 \\
> 0 & 0 & r^2 \sin^2 \theta
> \end{pmatrix}
> \quad \Rightarrow \quad
> \sqrt{Z} = r^2 \sin \theta
> $
>
>
> >[!Proof]-
> > Verification using tensor contraction (δ-system)
> >
> > The determinant of the covariant metric $Z_{ij}$ in $n$ dimensions can be written using the generalized Kronecker delta:
> > $
> > \det(Z_{ij}) = \frac{1}{n!} \, \delta^{i_1 i_2 \dots i_n}_{j_1 j_2 \dots j_n} \, Z_{i_1 j_1} Z_{i_2 j_2} \dots Z_{i_n j_n}
> > $
> > Therefore, the volume element is:
> > $
> > \sqrt{Z} = \sqrt{ \frac{1}{n!} \, \delta^{i_1 \dots i_n}_{j_1 \dots j_n} \, Z_{i_1 j_1} \cdots Z_{i_n j_n} }
> > $
> >
> > In particular:
> >
> > - In 3D:
> > $
> > \sqrt{Z} = \sqrt{ \frac{1}{3!} \, \delta^{ijk}_{rst} Z_{ir} Z_{js} Z_{kt} }
> > $
> > - In 2D:
> > $
> > \sqrt{Z} = \sqrt{ \frac{1}{2!} \, \delta^{ij}_{rs} Z_{ir} Z_{js} }
> > $
> >
> >
> > **Examples**:
> >
> > - **Polar coordinates (2D)**:
> > $
> > Z_{ij} =
> > \begin{pmatrix}
> > 1 & 0 \\
> > 0 & r^2
> > \end{pmatrix}, \quad
> > \sqrt{Z} = \sqrt{ \delta^{ij}_{rs} Z_{ir} Z_{js} / 2 } = \sqrt{1 \cdot r^2 - 0} = r
> > $
> >
> > - **Spherical coordinates (3D)**:
> > $
> > Z_{ij} =
> > \begin{pmatrix}
> > 1 & 0 & 0 \\
> > 0 & r^2 & 0 \\
> > 0 & 0 & r^2 \sin^2 \theta
> > \end{pmatrix}, \quad
> > \sqrt{Z} = \sqrt{ \delta^{ijk}_{rst} Z_{ir} Z_{js} Z_{kt} / 6 } = r^2 \sin \theta
> > $
> >
> > These results match the known volume elements and confirm the tensor-based formulation.
### The Voss–Weyl formula
The **Voss–Weyl formula** expresses the divergence of a contravariant vector field $T^i$ without invoking Christoffel symbols:
$
\nabla_i T^i = \frac{1}{\sqrt{Z}} \frac{\partial}{\partial Z^i} \left( \sqrt{Z} T^i \right)
$
This identity is valuable in general coordinates for the following reasons. First, it provides a divergence expression **without the need for Christoffel symbols**, making it algebraically simpler and often easier to compute. Second, the formula is valid in any curvilinear coordinate system, and it is constructed from invariant quantities. Finally, it plays a key role in simplifying expressions such as the Laplacian in differential geometry and physics.
> [!Proof]
> We begin by applying the product rule:
> $
> \frac{1}{\sqrt{Z}} \frac{\partial}{\partial Z^i} \left( \sqrt{Z} T^i \right)
> = \frac{\partial T^i}{\partial Z^i} + \frac{T^i}{\sqrt{Z}} \frac{\partial \sqrt{Z}}{\partial Z^i}
> $
>
> We compute the derivative of the volume element via the chain rule:
> $
> \frac{\partial \sqrt{Z}}{\partial Z^k}
> = \frac{1}{2 \sqrt{Z}} \frac{\partial Z}{\partial Z^k}
> $
>
> Then,
> $
> \frac{\partial Z}{\partial Z^k}
> = \frac{\partial Z}{\partial Z_{ij}} \cdot \frac{\partial Z_{ij}}{\partial Z^k}
> = Z Z^{ij} \left( \Gamma_{i,jk} + \Gamma_{j,ik} \right)
> = 2Z \Gamma^i_{ik}
> $
> Therefore,
> $
> \frac{\partial \sqrt{Z}}{\partial Z^k} = \sqrt{Z} \Gamma^i_{ik}
> $
>
> Substitute into the earlier expression:
> $
> \frac{1}{\sqrt{Z}} \frac{\partial}{\partial Z^i} \left( \sqrt{Z} T^i \right)
> = \frac{\partial T^i}{\partial Z^i} + T^i \Gamma^k_{ki}
> = \nabla_i T^i
> $
> [!note]
> **Laplacian of a scalar field**
> Let $F$ be an invariant scalar. The Laplacian is given by:
> $
> \nabla^i \nabla_i F = \nabla_i \left( Z^{ij} \frac{\partial F}{\partial Z^j} \right)
> $
> The Voss–Weyl formula avoids computing the full Christoffel expression:
> $
> \nabla_i \nabla^i F = \frac{1}{\sqrt{Z}} \frac{\partial}{\partial Z^i} \left( \sqrt{Z} Z^{ij} \frac{\partial F}{\partial Z^j} \right)
> $
> This form is easier to compute in non-Cartesian coordinates and ensures coordinate-invariant structure.
> [!Example]-
> **Spherical coordinates example: full derivation via Christoffel vs. Voss–Weyl**
>
> We compute the Laplacian of a scalar field $F$ in spherical coordinates using both the Christoffel-symbol-based method and the Voss–Weyl formula.
>
> First write metric tensor
>
> The coordinate system is $(r, \theta, \phi)$. The metric tensor is:
>
> $
> Z_{ij} =
> \begin{pmatrix}
> 1 & 0 & 0 \\
> 0 & r^2 & 0 \\
> 0 & 0 & r^2 \sin^2 \theta
> \end{pmatrix}, \quad
> Z^{ij} =
> \begin{pmatrix}
> 1 & 0 & 0 \\
> 0 & \frac{1}{r^2} & 0 \\
> 0 & 0 & \frac{1}{r^2 \sin^2 \theta}
> \end{pmatrix}, \quad
> \sqrt{Z} = r^2 \sin \theta
> $
>
>
> **Calculate Laplacian via Christoffel symbols**:
>The Laplacian is given by:
> $
> \nabla^i \nabla_i F = Z^{ij} \left( \frac{\partial^2 F}{\partial x^i \partial x^j}
> - \Gamma^k_{ij} \frac{\partial F}{\partial x^k} \right)
> $
>
> We compute the required [[The Christoffel symbol#^657e8f|Christoffel symbols]] using:
> $
> \Gamma^k_{ij} = \frac{1}{2} Z^{kl} \left(
> \frac{\partial Z_{jl}}{\partial x^i}
> + \frac{\partial Z_{il}}{\partial x^j}
> - \frac{\partial Z_{ij}}{\partial x^l}
> \right)
> $
>
> Non-zero Christoffel symbols:
>
> - $\Gamma^r_{\theta\theta} = -r$
> - $\Gamma^r_{\phi\phi} = -r \sin^2 \theta$
> - $\Gamma^\theta_{r\theta} = \Gamma^\theta_{\theta r} = \frac{1}{r}$
> - $\Gamma^\theta_{\phi\phi} = -\sin \theta \cos \theta$
> - $\Gamma^\phi_{r\phi} = \Gamma^\phi_{\phi r} = \frac{1}{r}$
> - $\Gamma^\phi_{\theta\phi} = \Gamma^\phi_{\phi\theta} = \cot \theta$
>
> Compute each term of the Laplacian:
>
> $
> \nabla^i \nabla_i F =
> \frac{\partial^2 F}{\partial r^2}
> + \frac{2}{r} \frac{\partial F}{\partial r}
> + \frac{1}{r^2} \left(
> \frac{\partial^2 F}{\partial \theta^2}
> + \cot \theta \frac{\partial F}{\partial \theta}
> \right)
> + \frac{1}{r^2 \sin^2 \theta} \frac{\partial^2 F}{\partial \phi^2}
> $
>
> This is the full Laplacian in spherical coordinates derived with Christoffel symbols.
>
> **Alternative: Voss–Weyl formula**
>The Laplacian of a scalar using the Voss–Weyl formula:
> $
> \nabla^i \nabla_i F = \frac{1}{\sqrt{Z}} \frac{\partial}{\partial x^i}
> \left( \sqrt{Z} Z^{ij} \frac{\partial F}{\partial x^j} \right)
> $
>
> With $\sqrt{Z} = r^2 \sin \theta$, the expression becomes:
> $
> \nabla^i \nabla_i F =
> \frac{1}{r^2 \sin \theta} \left[
> \frac{\partial}{\partial r} \left( r^2 \sin \theta \cdot \frac{\partial F}{\partial r} \right)
> + \frac{\partial}{\partial \theta} \left( \sin \theta \cdot \frac{\partial F}{\partial \theta} \right)
> + \frac{\partial}{\partial \phi} \left( \frac{1}{\sin \theta} \cdot \frac{\partial F}{\partial \phi} \right)
> \right]
> $
>
> which simplifies to:
> $
> \nabla^i \nabla_i F =
> \frac{1}{r^2} \frac{\partial}{\partial r} \left( r^2 \frac{\partial F}{\partial r} \right)
> + \frac{1}{r^2 \sin \theta} \frac{\partial}{\partial \theta}
> \left( \sin \theta \frac{\partial F}{\partial \theta} \right)
> + \frac{1}{r^2 \sin^2 \theta} \frac{\partial^2 F}{\partial \phi^2}
> $
>
> This matches the classical spherical Laplacian and is obtained **without** computing any Christoffel symbols. This confirms that both approaches yield the same Laplacian, but the Voss–Weyl method avoids computing $\Gamma^k_{ij}$, making it much more efficient in curved coordinates.
### Relative tensors
A **relative tensor** of weight $M$ is an object that transforms under coordinate changes like a tensor, but with an additional factor involving the determinant of the Jacobian matrix. For a type $(1,1)$ relative tensor $T^i_j$, the transformation law is:
$
T^{i'}_{j'} = J^M T^i_j J^i_{i'} J^{j'}_j
$
where $J = \left| J^i_{i'} \right|$ is the determinant of the Jacobian matrix.
Ordinary tensors are simply the special case $M = 0$ and are referred as **absolute tensor**. A relative tensor of rank zero (i.e., a scalar function) is called a **relative invariant**. For instance, the determinant of the metric tensor, $Z = \det(Z_{ij})$ satisfies:
$
Z' = J^2 Z
$
and is a relative invariant of weight 2.
> [!important]
> **Key properties of relative tensors**
>
> - Multiplying relative tensors of weights $M$ and $N$ gives a tensor of weight $M + N$.
> - Contracting indices preserves the weight.
> - The determinant of an $n \times n$ relative covariant tensor of weight $M$ is a relative scalar of weight $2 + nM$.
> - The inverse of a relative contravariant tensor of weight $M$ is a relative covariant tensor of weight $-M$.
> [!Proof]-
> **Product rule for weights**
> Let $S^i_j$ be a relative tensor of weight $M$ and $T^k_l$ of weight $N$:
> $
> S^{i'}_{j'} = J^M S^i_j J^i_{i'} J^{j'}_j, \quad
> T^{k'}_{l'} = J^N T^k_l J^k_{k'} J^{l'}_l
> $
> Then the product $S^i_j T^k_l$ transforms as:
> $
> S^{i'}_{j'} T^{k'}_{l'} = J^{M+N} S^i_j T^k_l J^i_{i'} J^{j'}_j J^k_{k'} J^{l'}_l
> $
> so the product has weight $M + N$.
>
> **Contraction preserves weight**
> Suppose $T^{ij}_{kl}$ is a relative tensor of weight $M$, and we contract over $i$ and $k$:
> $
> T^{i'j'}_{i'l'} = J^M T^{ij}_{kl} J^{i'}_i J^{j'}_j J^k_{i'} J^l_{l'}
> $
> The contraction involves $J^{i'}_i J^k_{i'} = \delta^k_i$, cancelling the extra Jacobian factors. Thus, the result still has weight $M$.
>
> **Weight of the determinant**
> Suppose $a_{ij}$ is a relative covariant tensor of weight $M$:
> $
> a_{i'j'} = J^M a_{ij} J^i_{i'} J^j_{j'}
> $
> Then the determinant transforms as:
> $
> \det(a') = \det(J^M a_{ij} J^i_{i'} J^j_{j'}) = J^{nM} \cdot J^2 \cdot \det(a) = J^{2 + nM} \det(a)
> $
> Here, $J^{nM}$ accounts for the $J^M$ factor on each row, and $J^2$ comes from the determinant of the two Jacobians. As a direct consequence, the [[#The volume element|volume element]] $\sqrt{Z}$, introduced in the previous section, is a relative scalar of weight 1.
>
> **Inverse weight flips sign**
> Suppose $A^i_j$ is a relative contravariant tensor of weight $M$:
> $
> A^{i'}_{j'} = J^M A^i_j J^{i'}_i J^j_{j'}
> $
> Let $B^j_k$ be its inverse, satisfying $A^i_j B^j_k = \delta^i_k$. Then:
> $
> A^{i'}_{j'} B^{j'}_{k'} = J^M A^i_j J^{i'}_i J^j_{j'} \cdot B^j_k J^{j'}_j J^k_{k'} = J^M J^{-M} A^i_j B^j_k J^{i'}_i J^k_{k'} = \delta^i_k J^{i'}_i J^k_{k'}
> $
> so $B^{j'}_{k'}$ must transform with weight $-M$. Hence, the inverse of a tensor of weight $M$ transforms as a tensor of weight $-M$.
### The Levi-Civita symbols
The **Levi-Civita symbols** are defined by rescaling the relative tensors $e^{ijk}$ and $e_{ijk}$ using the volume element $\sqrt{Z}$:
$
\varepsilon^{ijk} = \frac{e^{ijk}}{\sqrt{Z}}, \qquad
\varepsilon_{ijk} = \sqrt{Z} \, e_{ijk}
$
They are **absolute tensors** under orientation-preserving coordinate changes, and play a central role in describing orientation, volume, and antisymmetry.
The covariant symbol $\varepsilon_{ijk}$ can also be obtained by lowering the indices of $\varepsilon^{ijk}$ using the metric tensor:
$
\varepsilon_{ijk} = \varepsilon^{rst} Z_{ir} Z_{js} Z_{kt}
$
This agrees with the definition $\varepsilon_{ijk} = \sqrt{Z} e_{ijk}$, confirming consistency.
> [!Proof]-
> Starting from the identity $\varepsilon^{ijk} = e^{ijk} / \sqrt{Z}$, substitute into the right-hand side of the index-lowered form:
> $
> \varepsilon^{rst} Z_{ir} Z_{js} Z_{kt}
> = \frac{e^{rst}}{\sqrt{Z}} Z_{ir} Z_{js} Z_{kt}
> = \frac{Z e_{ijk}}{\sqrt{Z}} = \sqrt{Z} e_{ijk}
> = \varepsilon_{ijk}
> $
>
> Thus, both definitions agree. As a corollary, their product gives the generalized Kronecker delta:
> $
> \delta^{ijk}_{rst} = \varepsilon^{ijk} \varepsilon_{rst}
> $
> The generalized Kronecker delta $\delta^{ijk}_{rst}$ acts as an identity object for antisymmetric tensors, and since it is defined entirely via absolute tensors, it is itself an absolute tensor.
> [!Notice]
> While the components of $e^{ijk}$ and $e_{ijk}$ are numerically equal in Cartesian coordinates, they are **not** related by [[Index juggling|index lowering]] due to their differing transformation properties: $e^{ijk}$ has weight 1, $e_{ijk}$ has weight $-1$.
### The metrinilic property of Levi-Civita symbol
*This section is mainly the proof. One may skip it if not interested.*
We now prove that the Levi-Civita symbol with all lowered indices satisfies the **metrinilic property**, that is:
$
\nabla_i \varepsilon_{rst} =\nabla_i \varepsilon^{ijk}= 0
$
This means the Levi-Civita symbol behaves like a true metric: it is compatible with the covariant derivative and invariant under parallel transport.
We begin with the definition:
$
T_{irst} := \nabla_i e_{rst}
$
Since $e_{rst}$ is a constant symbol (its partial derivatives vanish), we compute the covariant derivative purely from the Christoffel correction terms:
$
T_{irst} = -\Gamma^m_{ir} e_{mst} - \Gamma^m_{is} e_{rmt} - \Gamma^m_{it} e_{rsm}
$
This expression is manifestly skew-symmetric in the indices $r$, $s$, $t$. To simplify, consider a specific case:
$
T_{i123} = -\Gamma^m_{i1} e_{m23} - \Gamma^m_{i2} e_{1m3} - \Gamma^m_{i3} e_{12m}
$
Only one term contributes in each sum:
$
T_{i123} = -\Gamma^1_{i1} e_{123} - \Gamma^2_{i2} e_{123} - \Gamma^3_{i3} e_{123}
= -\Gamma^m_{im} e_{123}
$
Generalizing:
$
T_{irst} = -\Gamma^m_{im} e_{rst}
$
Now consider the Levi-Civita symbol with volume element:
$
\varepsilon_{rst} = \sqrt{Z} \, e_{rst}
$
Apply the product rule:
$
\nabla_i \varepsilon_{rst}
= \nabla_i \left( \sqrt{Z} \, e_{rst} \right)
= \frac{\partial \sqrt{Z}}{\partial Z^i} e_{rst} + \sqrt{Z} \, \nabla_i e_{rst}
$
Using the identity:
$
\frac{\partial \sqrt{Z}}{\partial Z^i} = \sqrt{Z} \Gamma^m_{im}
$
and our previous result for $\nabla_i e_{rst}$:
$
\nabla_i \varepsilon_{rst}
= \sqrt{Z} \Gamma^m_{im} e_{rst} - \sqrt{Z} \Gamma^m_{im} e_{rst} = 0
$
> [!note]
> The Levi-Civita symbol $\varepsilon_{ijk}$ satisfies the same metrinilic property as the metric tensor:
> $
> \nabla_i \varepsilon_{jkl} = 0
> $
> This ensures its consistency in curved coordinates and is essential in geometric identities involving determinants, volume forms, and antisymmetric operations.
> [!important]
> **Consequences of the metrinilic property** (confirms [[Properties of covariant differentiation#The metrinilic property]])
>
> - The covariant Levi-Civita symbol satisfies:
> $
> \nabla_m \varepsilon_{rst} = 0
> $
>
> - The contravariant version satisfies:
> $
> \nabla_m \varepsilon^{ijk} = 0
> $
>
> - As a consequence, all Kronecker delta systems built from their products are covariantly constant:
> $
> \nabla_m \delta^{ijk}_{rst} = \nabla_m \delta^{ij}_{rs} = \nabla_m \delta^i_r = 0
> $