I considered the $k$th component of $\text{curl $f\mathbf{F}$}$. $f$ is a scalar field and $\mathbf{F}$ a vector field.

$\color{green}{[}\nabla \times (fF)\color{green}{]} _{\LARGE{\color{green}{k}}} = \epsilon_{ij\LARGE{\color{green}{k}}}\partial_i(f\mathbf{F})_j $
$= \epsilon_{ij\LARGE{\color{green}{k}}}\partial_i(fF_j) \qquad \qquad \qquad \qquad (\text{since $\mathbf{(F)}_j :=$ the $j$th component of $\mathbf{F} = F_j$})$
$= \epsilon_{ij\LARGE{\color{green}{k}}}(F_j\partial_if + f\partial_iF_j) \qquad \qquad (\text{since $f$ scalar})$
$= \underbrace{\epsilon_{ij\LARGE{\color{green}{k}}}F_j\partial_if}_{\Large{\bigstar}} + f\underbrace{{\epsilon_{ij\LARGE{\color{green}{k}}}\partial_iF_j}}_{\LARGE{\color{green}{[}\nabla \times \mathbf{F}\color{green}{]}_{\LARGE{\color{green}{k}}}}} $

Hereafter, I refer only to the term with the star underneath. Since [$\color{#007FFF}{F_j}$ corresponding to $\color{#007FFF}{\mathbf{F}}$] appears before [$\color{#FF00FF}{\partial_if}$ corresponding to $\color{#FF00FF}{\nabla f}$], thus ${\epsilon_{ij\LARGE{\color{green}{k}}}\color{#007FFF}{F_j}\color{#FF00FF}{\partial_if}} = {\color{green}{[}\color{#007FFF}{\mathbf{F}} \times \color{#FF00FF}{\nabla f}\color{green}{]} _{\LARGE{\color{green}{k}}}}$.
But the answer states $\color{green}{[}\nabla f \times \mathbf{F}\color{green}{]} _{\LARGE{\color{green}{k}}}$. What went wrong?

$\large{\text{Supplement to Andrew D's response :}}$

Here's my understanding of your answer : In ${\epsilon_{ij\LARGE{\color{green}{k}}}F_j\partial_if}, \; {(i, j, \LARGE{\color{green}{k}})}$ (in the subscript of the Levi-Civita symbol) denotes the order of the components. So the $i$th component must appear first, and the $j$th component second.

However, since $(i, j, k) = \color{brown}{(j, k, i)}$, therefore $\epsilon_{ijk} = \epsilon_{\color{brown}{\LARGE{jki}}}$. $\color{brown}{\text{Now, $j$ precedes $i$, so wouldn't this result in the wrong order of the components?}}$

$\large{\text{2nd Supplement to Andrew D's Comment beneath his Answer :}}$

$\color{#3EB489}{\text{The variable in the permutation succeeding the variable that's not summed}}$ corresponds to the first component to appear. Here, $k$ denotes the component being analysed so is not summed. Since I am looking at $\color{brown}{(j, k, i)}$, $\color{#3EB489}{i}$ succeeds $k$ so the $\color{#3EB489}{i}$th component is the first. Therefore, ${\epsilon_{ijk}F_j\color{#3EB489}{\partial_{\LARGE{i}}}f}$ = ${\epsilon_{\color{brown}{\LARGE{jki}}}\color{#3EB489}{\partial_{\LARGE{i}}}F_jf} = \color{green}{[}\color{#3EB489}{\nabla f} \times \mathbf{F}\color{green}{]} _{\LARGE{\color{green}{k}}} $

However, this appears to discord with Steven Stadnicki's 2nd comment, according to which: $ {\epsilon_{\color{brown}{\LARGE{jki}}}F_j\partial_if} = {\color{green}{[}\mathbf{F} \times \nabla f\color{green}{]} _{\LARGE{\color{green}{k}}}}$?

  • 1
    $\begingroup$ Can you see that it makes no difference whether you write $\varepsilon_{ijk}F_j\partial_if$ or $\varepsilon_{ijk}(\partial_if)F_j$? $\endgroup$ Jul 2 '13 at 15:56
  • $\begingroup$ Also, to your last point: the critical element is that your final index $k$ needs to correspond to the last element of the Levi-civita symbol, not the middle one. Otherwise you would have $(a\times b)_k = \varepsilon_{ijk}a_ib_j = \varepsilon_{jki}b_ja_i = (b\times a)_k$. $\endgroup$ Jul 2 '13 at 15:59
  • $\begingroup$ @StevenStadnicki: Thank you for your comments. Wrt your first comment: Does it make no difference because we are just multiplying 0the components therein? Wrt your second comment: Could you please see the 2nd supplement that I've added to my original Question? $\endgroup$
    – NNOX Apps
    Jul 3 '13 at 14:02
  • $\begingroup$ I had hoped that for the rest of my life I would not have to witness such a cumbersome proof of an identity about ${\rm curl}(f{\bf F})$. In my opinion the "summation convention" was Einstein's biggest mistake. It prevented people from thinking what's going on here. $\endgroup$ Oct 3 '13 at 10:57
  • $\begingroup$ @ChristianBlatter Einstein is reputed to have quipped that it was his greatest contribution to mathematics; what would you consider a less cumbersome proof? $\endgroup$
    – Rax Adaam
    Mar 9 '18 at 0:54

It isn't the ordering of $F_j$ and $\partial_if$ in the product of terms which determines what order the terms are in the cross product, it is the ordering of the suffix's which does so - as we have $\{i,j,k\}$ as our right-handed set (taken from the ordering of the suffices from the Levi-Civita symbol given), it means that we take the $\partial_if$ as being the first component of the cross product and $F_j$ as our second, so we get $[ \nabla f \times \mathbf F]_k$ as required.

If this doesn't make much sense, say so and I'll try and clarify what I'm saying.

  • $\begingroup$ @LePressentiment Sure thing - I feel my original answer may have been slightly confusing: the important thing is that we are considering the order of the suffices after the k (or whatever the suffix we are not summing over may be), so we still take the $i$ first and then the $j$. Does that make sense? $\endgroup$
    – Andrew D
    Jul 2 '13 at 15:40
  • $\begingroup$ Thank you very much again. Yes, but another issue appears to have sprung. Could you please look at the 2nd supplement added to my original post? $\endgroup$
    – NNOX Apps
    Jul 3 '13 at 14:04

Recall that $\vec{a} \times \vec{b} = - \, \vec{b} \times \vec{a}$.

Now consider each, using your definitions (I have included the unit vector $\hat{e}_i$ explicitly, simply for additional clarity; see below):

$$ \vec{a} \times \vec{b} = \epsilon_{i{\color{red}j}{\color{blue}k}} \, \hat{e}_i \, a_{\color{red}j} \, b_{\color{blue}k} = - \, \vec{b} \times \vec{a} = - \epsilon_{i{\color{red}j}{\color{blue}k}} \, \hat{e}_i \, b_{\color{red}j} \, a_{\color{blue}k}.$$

Recall that $\epsilon_{ijk}$ is perfectly anti-symmetric, which means that interchanging any two indices changes its sign (there are only 6 nonzero terms, so it's not long to verify this by plugging in actual values), for example:

$$\epsilon_{1{\color{red}2}3} = - \epsilon_{13{\color{red}2}}.$$

Or, in general, $-\epsilon_{i{\color{red}j}k} = \epsilon_{ik{\color{red}j}}$. Put this together with our result for $-\vec{b} \times \vec{a}$, above, we have:

$$- \, \vec{b} \times \vec{a} = - \epsilon_{i{\color{red}j}{\color{blue}k}} \, \hat{e}_i \, b_{\color{red}j} \, a_{\color{blue}k} = \epsilon_{i{\color{blue}k}{\color{red}j}}\, \hat{e}_i \,a_{\color{blue}k} \, b_{\color{red}j} = \vec{a} \times \vec{b},$$

as expected.

N.B. the ordering of the $a_{j/k}$ and $b_{j/k}$ terms themselves doesn't matter, it's the relative position of their index in $\epsilon_{ijk}$ that matters. ie.

$$\begin{align*}\vec{a} \times \vec{b} & = \epsilon_{ijk} \, \hat{e}_i \, a_j \, b_k \\ & = \epsilon_{ijk} \hat{e}_i \, b_k \, a_j \\ & = \epsilon_{jki} \hat{e}_j \, a_k \, b_i \\ & = \epsilon_{kij} \hat{e}_k \, a_i \, b_j \end{align*} $$

In the beginning, it can be easy to conflate $i$, $j$ and $k$ with the unit vector in Cartesian space, but they aren't the same thing.

Notice this means you can cycle through the equivalent forms of epsilon until you find a form that "matches" the cross product. So, in your case:

$$\begin{align*} & \quad \epsilon_{ijk} F_j \partial_i f \\ & = \epsilon_{jki} F_j\partial_i f\\ & = \epsilon_{kij} F_j\partial_i f \end{align*}$$

Which we can now read as $\nabla f \times \vec{F}$ (or, equivalently: $-\, \vec{F} \times \nabla f$).

On including the unit vector in the definition of the cross product (because it seems there is some conflation between the indices $i$, $j$, $k$ and the basis vectors $\hat{i} = \hat{e}_1$, $\hat{j} = \hat{e}_2$ and $\hat{k}=\hat{e}_3$).

I chose to include the unit vector in my definition, because it removes this "components" expression. Instead, we can just give an expression for the (entire) cross product:

$$ \vec{a} \times \vec{b} = \epsilon_{ijk} \, \hat{e}_i \, a_j \, b_k.$$

This is the same as your definition; however, I've included the unit vector $\hat{e}_i$ explicitly. Notice, for example when $i = 1$, we get the first component of the cross-product:

$$ \epsilon_{123} \hat{e}_1 a_2b_3 + \epsilon_{132} \hat{e}_1 a_3b_2 = \hat{e}_1\, a_2b_3 - \hat{e}_1 \,a_3b_2 = (a_2b_3 - a_3b_2) \,\hat{e}_1,$$

and similarly for the other two components.

However, the naming of the indices is totally arbitrary, so we could equally write:

$$ \vec{a} \times \vec{b} = \epsilon_{lmn} \, \hat{e}_l \, a_m \, b_n, $$

or even,

$$ \vec{a} \times \vec{b} = \epsilon_{kji} \, \hat{e}_k \, a_j \, b_i, $$

because this represents every possible combination for all three indices equal to any one of 1, 2, or 3. (so $3^3 = 27$ different terms, where all but 6 are zero).

Now, once you've chosen your indices, order does matter, in particular all cyclic permutations [(1,2,3), (2,3,1), (3,1,2)] will have the same value, and all acyclic [(1,3,2), (3,2,1) and (2,1,3)] permutations will have the same value, i.e.

$$\vec{a} \times \vec{b} = \epsilon_{i{\color{blue}j}{\color{red}k}}\, \hat{e}_i \, a_{{\color{blue}j}} b_{\color{red}k} = \epsilon_{j{\color{blue}k}{\color{red}i}}\, \hat{e}_j \, a_{{\color{blue}k}}b_{\color{red}i} = \epsilon_{k{\color{blue}i}{\color{red}j}}\,\hat{e}_k \, a_{\color{blue}i} b_{\color{red}j}$$ whereas $$ \epsilon_{i{\color{red}k}{\color{blue}j}}\,\hat{e}_i \, a_{{\color{blue}j}} b_{\color{red}k} = \epsilon_{j{\color{red}i}{\color{blue}k}}\, \hat{e}_j \, a_{{\color{blue}k}}b_{\color{red}i} = \epsilon_{k{\color{red}j}{\color{blue}i}}\, \hat{e}_k \, a_{\color{blue}i} b_{\color{red}j} = \vec{b} \times \vec{a}$$

Hope this helps!


Terms as $\epsilon_{ijk}a_ib_i$ do not carry any information on operations between $a_i$ and $b_j$, hence no information on e.g. commutativity. They really are meant as summing symbols (maybe with no meaning at al)l $F_j$ over indices which appear twice, no matter in what order these symbols are arranged: $$\epsilon_{ijk}a_ib_j=a_i\epsilon_{ijk}b_j=b_ja_i\epsilon_{ijk}=\dots$$ (Maybe, writing those kind of sums explicitly for e.g. $i,j,k=1,2,3$ helps.)

Concerning your 2nd supplement, you only need to keep in mind that $$(\nabla f)_i:=\partial_if:=\frac{\partial f}{\partial x^i}$$ denotes one single symbol with index. Therefore, $\epsilon_{ijk}\partial_iF_jf$ does not make sense in this context.

Finally, Steven Stadnicki's first comment is most important.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.