## Cross product, curl, and higher-dimensional generalization
With [[Relative tensor and Levi-Civita symbol#The Levi-Civita symbols|the Levi-Civita symbols]], we may now write cross product and curl.
### The cross product
The **cross product** of two vectors $U^i$ and $V^i$ is defined using the Levi-Civita symbol:
$
W^i = \varepsilon^{ijk} U_j V_k
$
This expression gives a true contravariant vector $W^i$. It is coordinate-independent and valid in general curvilinear systems. To obtain the geometric vector $\mathbf{W}$, we write:
$
\mathbf{W} = W^i Z_i = \varepsilon^{ijk} U_j V_k Z_i
$
> [!Notice]
> This formulation confirms that the cross product is a **tensorial operation**, and does not rely on Cartesian coordinates.
> [!Proof]-
> **Divergence of cross product**
> Using the metrinilic property of the Levi-Civita symbol:
> $
> \nabla_i W^i = \nabla_i \left( \varepsilon^{ijk} U_j V_k \right)
> = \varepsilon^{ijk} \nabla_i (U_j V_k)
> = \varepsilon^{ijk} \left( \nabla_i U_j \cdot V_k + U_j \cdot \nabla_i V_k \right)
> $
> This is the full covariant divergence, consistent with tensor analysis.
> [!important]
> **Key geometric and algebraic properties**
>
> - **Antisymmetry**:
> $
> \varepsilon^{ijk} U_j V_k = -\varepsilon^{ijk} V_j U_k
> \quad \Rightarrow \quad
> U \times V = -V \times U
> $
>
> - **Orthogonality**:
> $
> W^i V_i = \varepsilon^{ijk} U_j V_k V_i = 0
> $
> The result is perpendicular to both inputs.
>
> - **Norm (length squared)**:
> $
> W_i W^i
> = \varepsilon_{ijk} U^j V^k \varepsilon^{irs} U_r V_s
> = \delta^{rs}_{jk} U^j V^k U_r V_s
> $
> Expanding the generalized Kronecker delta:
> $
> W_i W^i = U_j V_k U^j V^k - U_j V_k U^k V^j
> = \left( \mathbf{U} \cdot \mathbf{U} \right) \left( \mathbf{V} \cdot \mathbf{V} \right) - \left( \mathbf{U} \cdot \mathbf{V} \right)^2
> $
> That is:
> $
> |\mathbf{W}|^2 = |\mathbf{U}|^2 |\mathbf{V}|^2 - (\mathbf{U} \cdot \mathbf{V})^2
> $
Two examples for applying tensorial form are shown below.
> [!Example]-
> **In Cartesian and cylindrical coordinates**
>
> - **In Cartesian coordinates**, the basis vectors $Z_i$ are constant and orthonormal:
> $
> Z_i = \{ \mathbf{i}, \mathbf{j}, \mathbf{k} \}, \quad Z_{ij} = \delta_{ij}, \quad \sqrt{Z} = 1
> $
> So the geometric cross product becomes:
> $
> \mathbf{W} = \varepsilon^{ijk} U_j V_k Z_i
> = \sum_{i=1}^3 \left( \sum_{j,k=1}^3 \varepsilon^{ijk} U_j V_k \right) Z_i
> =
> \begin{vmatrix}
> \mathbf{i} & \mathbf{j} & \mathbf{k} \\
> U^1 & U^2 & U^3 \\
> V^1 & V^2 & V^3
> \end{vmatrix}
> $
>
> - **In cylindrical coordinates**, the basis vectors are position-dependent, and the metric tensor is:
> $
> Z_{ij} = \begin{pmatrix} 1 & 0 & 0 \\ 0 & r^2 & 0 \\ 0 & 0 & 1 \end{pmatrix}, \quad
> Z^{ij} = \begin{pmatrix} 1 & 0 & 0 \\ 0 & \frac{1}{r^2} & 0 \\ 0 & 0 & 1 \end{pmatrix}, \quad
> \sqrt{Z} = r
> $
> The basis vectors $Z_i$ are:
> $
> Z_1 = \hat{\mathbf{r}},\quad Z_2 = r \hat{\boldsymbol{\theta}},\quad Z_3 = \hat{\mathbf{z}}
> $
> Then the geometric cross product becomes:
> $
> \mathbf{W} = \varepsilon^{ijk} U_j V_k Z_i
> = \sum_i \left( \varepsilon^{ijk} U_j V_k \right) Z_i
> =
> \begin{vmatrix}
> r \hat{\mathbf{r}} & r \hat{\boldsymbol{\theta}} & r \hat{\mathbf{z}} \\
> U^1 & U^2 & U^3 \\
> V^1 & V^2 & V^3
> \end{vmatrix}
> $
> or more carefully:
> $
> \mathbf{W} =
> \begin{vmatrix}
> r Z^1 & r Z^2 & r Z^3 \\
> U^1 & U^2 & U^3 \\
> V^1 & V^2 & V^3
> \end{vmatrix}
> $
> where $Z^i$ is the geometric vector associated with basis direction $i$. The factor $r$ reflects the scaling from $\sqrt{Z}$ and ensures the correct geometric result.
> [!Example]-
> **Tensor proof of cross product non-associativity**
>
> We prove that:
> $
> (\mathbf{U} \times \mathbf{V}) \times \mathbf{W}
> = (\mathbf{U} \cdot \mathbf{W}) \mathbf{V} - (\mathbf{V} \cdot \mathbf{W}) \mathbf{U}
> $
> using the Levi-Civita tensor in index notation.
>
> Let:
> $
> (\mathbf{U} \times \mathbf{V})^i = \varepsilon^{ijk} U_j V_k
> $
> Then:
> $
> \left[(\mathbf{U} \times \mathbf{V}) \times \mathbf{W} \right]^m
> = \varepsilon^{mnp} (\mathbf{U} \times \mathbf{V})_n W_p
> = \varepsilon^{mnp} \varepsilon_{nrs} U^r V^s W_p
> $
>
> Using the identity:
> $
> \varepsilon^{mnp} \varepsilon_{nrs}
> = \delta^m_r \delta^p_s - \delta^m_s \delta^p_r
> $
> we substitute:
> $
> \left[(\mathbf{U} \times \mathbf{V}) \times \mathbf{W} \right]^m
> = (\delta^m_r \delta^p_s - \delta^m_s \delta^p_r) U^r V^s W_p
> = U^m V^s W_s - V^m U^r W_r
> $
> So:
> $
> \left[(\mathbf{U} \times \mathbf{V}) \times \mathbf{W} \right]^m
> = (\mathbf{U} \cdot \mathbf{W}) V^m - (\mathbf{V} \cdot \mathbf{W}) U^m
> $
> which matches the vector identity:
> $
> (\mathbf{U} \times \mathbf{V}) \times \mathbf{W}
> = (\mathbf{U} \cdot \mathbf{W}) \mathbf{V} - (\mathbf{V} \cdot \mathbf{W}) \mathbf{U}
> $
> This confirms the non-associativity of the cross product using tensor notation.
### The curl
The **curl** of a vector field is an invariant differential operator defined using the Levi-Civita symbol. For a vector field $U_k$, its curl is given by:
$
V^i = \varepsilon^{ijk} \nabla_j U_k
$
In dyadic notation, this is simply written as:
$
\mathbf{V} = \nabla \times \mathbf{U}
$
> [!Note]+
> **Show expression with tensor identity: $\nabla \times (\nabla \times \mathbf{U}) = \nabla (\nabla \cdot \mathbf{U}) - \nabla^2 \mathbf{U}$**
>
> Consider the composition $\nabla \times (\nabla \times \mathbf{U})$ in tensor notation:
> $
> V_r = \varepsilon_{rsi} \nabla^s \left( \varepsilon^{ijk} \nabla_j U_k \right)
> $
> Using the **metrinilic property** $\nabla^s \varepsilon^{ijk} = 0$, we move $\varepsilon^{ijk}$ outside the derivative:
> $
> V_r = \varepsilon_{rsi} \varepsilon^{ijk} \nabla^s \nabla_j U_k
> $
> Now use the contraction identity for Levi-Civita symbols:
> $
> \varepsilon_{rsi} \varepsilon^{ijk} = \delta^{jk}_{rs}
> $
> Hence,
> $
> V_r = \delta^{jk}_{rs} \nabla^s \nabla_j U_k
> = \delta^j_r \delta^k_s \nabla^s \nabla_j U_k - \delta^k_r \delta^j_s \nabla^s \nabla_j U_k
> $
> Therefore:
> $
> V_r = \nabla^k \nabla_r U_k - \nabla^s \nabla_s U_r
> $
> Relabeling dummy indices yields the final result:
> $
> V_i = \nabla_i \nabla_j U^j - \nabla_j \nabla^j U_i
> $
> In vector notation:
> $
> \nabla \times (\nabla \times \mathbf{U}) = \nabla (\nabla \cdot \mathbf{U}) - \nabla^2 \mathbf{U}
> $
> Other expression can be treated in similarly manner.
### Other dimensions generalization
The [[Relative tensor and Levi-Civita symbol#The Levi-Civita symbols|Levi-Civita symbol]] and the delta system naturally generalize to arbitrary dimensions. In two dimensions, indices take values 1 or 2. The permutation symbols $e_{ij}$ and $e^{ij}$ are defined analogously to the 3D case:
$
e_{ij}, \, e^{ij} =
\begin{bmatrix}
0 & 1 \\
-1 & 0
\end{bmatrix}
$
From these, we define the full delta system:
$
\delta^{ij}_{rs} = e^{ij} e_{rs}
$
which satisfies the identity:
$
\delta^{ij}_{rs} = \delta^i_r \delta^j_s - \delta^j_r \delta^i_s,
\qquad
\delta^i_j = \delta^i_{jr}
$
The determinant of a $2 \times 2$ matrix $a^i_j$ is given by:
$
A = \frac{1}{2} \delta^{ij}_{rs} a^r_i a^s_j
$
The Levi-Civita symbols are made absolute using the volume element $\sqrt{Z}$:
$
\varepsilon_{ij} = \sqrt{Z} e_{ij}, \qquad
\varepsilon^{ij} = \frac{e^{ij}}{\sqrt{Z}}
$
With this, we define a 2D cross product involving a single vector:
$
V_i = \varepsilon_{ij} U^j
$
It satisfies the orthogonality condition:
$
V_i U^i = \varepsilon_{ij} U^i U^j = 0
$
> [!note]
> The 2D cross product maps a vector to an orthogonal vector.
> The Levi-Civita symbol ensures antisymmetry, and this construction is crucial in 2D vector calculus.
We may also define a scalar-valued curl:
$
V = \varepsilon^{ij} \nabla_i U_j
$
To generalize to $n$ dimensions, we define the permutation symbols $e_{i_1 \cdots i_n}$ and $e^{i_1 \cdots i_n}$, which take values $1$ or $-1$ for even and odd permutations, and $0$ otherwise.
The full delta system becomes:
$
\delta^{i_1 \cdots i_n}_{j_1 \cdots j_n} = e^{i_1 \cdots i_n} e_{j_1 \cdots j_n}
$
and can be expressed by the determinant:
$
\delta^{i_1 \cdots i_n}_{j_1 \cdots j_n}
=
\begin{vmatrix}
\delta^{i_1}_{j_1} & \cdots & \delta^{i_1}_{j_n} \\
\vdots & \ddots & \vdots \\
\delta^{i_n}_{j_1} & \cdots & \delta^{i_n}_{j_n}
\end{vmatrix}
$
Delta systems satisfy contraction identities:
$
\delta^{i_1 \cdots i_{k-1}}_{j_1 \cdots j_{k-1}} = (n-k) \, \delta^{i_1 \cdots i_{k-1} i_k}_{j_1 \cdots j_{k-1} i_k}
$
and in particular:
$
\delta^{i_1 \cdots i_k}_{i_1 \cdots i_k} = k!
$
The determinant of an $n \times n$ matrix is:
$
A = \frac{1}{n!} \delta^{i_1 \cdots i_n}_{j_1 \cdots j_n} a^{j_1}_{i_1} \cdots a^{j_n}_{i_n}
$
In higher dimensions, the cross product generalizes to a construction involving $(n-1)$ contravariant vectors. The result is a covariant vector:
$
V_i = \varepsilon_{i j_1 \cdots j_{n-1}} U^{j_1}_{(1)} \cdots U^{j_{n-1}}_{(n-1)}
$
where each $U^{(k)\,j_k}$ is a contravariant vector. This generalization is well-defined in all dimensions but does **not** produce a unique direction unless $n = 3$.
> [!warning]
> The cross product as a binary operation is **not well defined** in dimensions other than 3. (Although it's possible, but have completely different meanings.)
> In $n > 3$, the Levi-Civita symbol still defines orientation and antisymmetry,
> but the curl becomes a differential form or tensor rather than a vector.