yungman said:
I cannot verify this equation compare to Wolfram site. Using your equation:
(\vec A \cdot \nabla) \vec B = \nabla ( \vec A \cdot \vec B) -\vec A \times ( \nabla \times \vec B)
Your work is good - you can see right here that the starting equation is not right. The del operator must never operate on A, as it is doing on the first term to the right of the equal sign. When A, B, and C are vectors, there is a sum term which is rewritten:
\sum_j A_j B_i C_j = B_i\sum_j A_j C_j
and that is correct, but this does not work when B is the del operator:
\sum_j A_j \partial_i C_j \ne \partial_i\sum_j A_j C_j
In other words, you cannot just substitute the del operator into what is a vector identity. The correct equation is
(\vec A \cdot \nabla) \vec B = (\nabla \vec B)\cdot A -\vec A \times ( \nabla \times \vec B)
where \nabla \vec B is a tensor: (\nabla \vec B)_{ij}=\partial_i B_j
The incorrect step was in assuming (\nabla \vec B)\cdot A=\nabla(\vec B \cdot \vec A) which is fine when \nabla is replaced by a vector, but not when it is not.
I don't like messing around with these kinds of equations, it seems like there are a hundred different theorems, like recipes in a cookbook and, you try to solve problems by flipping through a bunch of recipes, hoping to find the right one. Using abstract index notation, you need only a few theorems, the notation is cleaner, and the core of the problem is isolated from the particular vectors or vector operators you are using. The problem is, its something new and different.
In abstract index notation (which is ultimately coordinate-system free!), the dot product is written
(\vec A \cdot \vec B)=A_i B_i
where double indices imply summation. The cross product is written:
(\vec A \times \vec B)_i=\epsilon_{ijk}A_j B_k
where \epsilon_{ijk} is the permutation pseudotensor (1 for even permutations of 123, -1 for odd permutations of 123 and zero otherwise). The above theorem is written:
A_i \partial_i B_j=A_j\partial_i B_j-\epsilon_{ijk}A_i(\epsilon_{kmn}\partial_m B_n)=A_j\partial_i B_j-\epsilon_{ijk}\epsilon_{kmn}A_i\partial_m B_n
The theorem solves itself when you realize one important theorem, that has only to do with the permutation pseudotensor, not the components of the vectors or operators involved:
\epsilon_{ijk}\epsilon_{kmn}=\delta_{im}\delta_{jn}-\delta_{in}\delta_{jm}
where \delta_{ij} is the Kronecker delta, 1 when i=j, zero otherwise. Of course, when the del operator (\partial_i) is involved you have to be careful to keep track of what it is operating on. \partial_i A_j B_k is nebulous, it should be written \partial_i (A_j) B_k or B_k \partial_i A_j when the del operator is operating only on A. When operating on both it should be written \partial_i(A_j B_k)
