Proper usage of Einstein sum notation

Tags:
1. Apr 21, 2017

Gan_HOPE326

1. The problem statement, all variables and given/known data

I'm dealing with some pretty complex derivatives of a kernel function; long story short, there's a lot of summations going on, so I'm trying to write it down using the Einstein notation, for shortness and hopefully reduction of errors (also for the sake of a paper in which I have to write all this stuff down and possibly do it without blowing past the page's margins). Right now I was testing something that's relatively simple, but I'm not sure I'm using this correctly.

2. Relevant equations

My test example was a relatively simple derivative. For reference, these are the symbols I am using:

$$P_{ij} = exp[-(x_i-x_j)^2] \qquad P_{ij}' = \frac{dP_{ij}}{dx_i} = -\frac{dP_{ij}}{dx_j} \qquad P_i = P_{ij}\delta_{jj} \qquad P_i' = P_{ij}'\delta_{jj}$$

I'm already unsure about the use of $\delta_{jj}$ there, but then comes the problem. As a first exercise I'm trying an example of a derivative, with an additional index $n$:

$$\frac{d(P_iP_i)}{dx_n}$$

3. The attempt at a solution

Here's my solution:

$$\frac{d(P_iP_i)}{dx_n} = 2P_i\frac{dP_i}{dx_n} = 2P_i\left[\frac{dP_i}{dx_i}\delta_{in}-\frac{dP_i}{dx_j}\delta_{jn}\right] = 2P_nP_n' - 2P_iP_{in}'$$

Which actually works (tested numerically), but seems ugly and wrong to me due to those repeated $n$ indices which seem to imply a summation that isn't really there. Did I do something wrong? Is there some other symbol I'm disregarding or some rule I don't know? Thanks!

2. Apr 21, 2017

vela

Staff Emeritus
One error is that the same index shouldn't show up more than twice in a term, so $P_i = P_{ij}\delta_{jj}$ doesn't make sense because $j$ appears three times. It's not clear to me what you're trying to do there. What is $P_i$ supposed to be equal to in normal summation notation?

3. Apr 21, 2017

Gan_HOPE326

In regular notation,

$$P_i = \sum_j P_{ij}$$

I suppose I could get the same result by multiplying by an array of ones with a single index, I just don't know if there's a conventional symbol for that.

4. Apr 21, 2017

vela

Staff Emeritus
That's probably the most straightforward way. You can define a vector of ones, say $e = (1, 1, \dots, 1)$, then $P_i = P_{ij}e_j$.

You also need to clean up the notation for the derivative. The chain rule gives you (with no implied summation here)
$$\frac{d}{dx_n} P_{ij} = \frac{\partial P_{ij}}{\partial x_i}\frac{dx_i}{dx_n} + \frac{\partial P_{ij}}{\partial x_j}\frac{dx_j}{dx_n}.$$

5. Apr 21, 2017

Gan_HOPE326

Yes, right, I'll fix that. The minus sign came from me knowing it appears in the end but it's not correct there.

EDIT: apparently I can't edit the first post in the thread? Sorry for that.