## I know this is a bit of a stretch, but is there anyway to encode the Ei(x) function in JMP.

Community Trekker

Joined:

Aug 8, 2012

The Ei(X) function is (I guess) a fairly rare function. At least I've never had to deal with it before. But it has popped up in an analysis I am doing, I am wondering if there is anyway to encode it        into a formula in JMP. Clearly, it isn't in there by default, but is there a way to manually enter this:

$\mbox{Ei}(x)=\int_{-\infty}^x\frac{e^t}t\,dt.\,$
Into a formula column? Thanks in advance.
--SurfaceThought
1 ACCEPTED SOLUTION

Accepted Solutions

Super User

Joined:

Jun 23, 2011

Solution

OK here's a JSL implementation of the Fortran code.  The code runs, but you'll need to test it out to make sure the numbers are good.  It's pretty much a straight conversion of the Fortran program.  Interesting exercise!

13 REPLIES

Community Trekker

Joined:

Jun 23, 2011

Not so rare.  The MATLAB page gives a lot of approximations, so there may be something there that can be ginned up from the functions in JMP.

Also look at the GNU library (http://www.gnu.org/software/gsl/manual/html_node/Exponential-Integral.html).  You could use this to generate an R object, and then pull that into JMP.

Steve Denham

Community Trekker

Joined:

Aug 8, 2012

Steve: Thanks for the quick reply:

I have looked at the approximations, and I am not sure that they will work for my purposes. The Taylor series approximation:

$\mathrm{Ei}(x) = \gamma+\ln |x| + \sum_{k=1}^{\infty} \frac{x^k}{k\; k!} \qquad x \neq 0$

is apparently slow to converge, and would need k = 30 in order to have three decimal point precision at the levels of x I would be dealing with. In addition, the wiki page states :

This formula can be used to compute $\mathrm{E_1}(x)$ with floating point operations for real $x$ between 0 and 2.5. For $x > 2.5$, the result is inaccurate due to cancellation.

Which is rather ominous, seeing as I will be dealing with x values larger than 2.5. I understand that "floating point operations" just means that the computing device rounds off to a certain number of decimal points after each step of the calculation. Does JMP use floating point operations? Is there even an alternative? I suppose I could transform x in some way so that the values would fit in between 0 and 2.5.

There is another approximation that is divergent, which isn't great because it is divergent, but also because the error seems to be very large for small values of x (less than one), which is also within the range of x values I may use. If I need to use this approximation, then I will have to, but I would like to save it for last option.

The problem with calculating the Ei(X) and importing it into JMP, is that the argument of the exponential integral includes a parameter that I want to solve for using non-linear least squares. Therefore it needs to actually be encoded into JMP so that JMP can change the parameter in order to iteratively solve for it.

Is it really true that there is no INT() function in JMP?

Thanks for the help,

--SurfaceThought

Community Trekker

Joined:

Jun 23, 2011

Not enough of a JMP user to know--much more of a SAS person.  But integration is not something SAS is very strong about at the function level, although there are some PROC FCMP approaches and macros that I know of.

What I thought was a possibility was using some of the gamma functions associated with Ei(x), like the incomplete gamma function and appropriately scaling the input and output.

Steve Denham

Community Trekker

Joined:

Aug 8, 2012

Steve,

Using the gamma function to calculate Ei(x) is so tantalizing, and so close to the solution, but after playing around with it I unfortunately think that it will not work.  The Gamma function as given in JMP allows you to compute either the complete gamma function or the lower incomplete gamma function. In other words, it allows you to integrate from zero to infinity, or zero to some positive x value, but you cannot change to lower limit of integration: it is always zero. The Exponential integral is related to the upper incomplete gamma function, as in order to calculate it you need to have a variable lower limit of integration and integrate up to infinity. Because the gamma function as it stands in JMP does not allow for a variable lower limit of integration, I think that this approach simply will not work.

Thank you though,

--SurfaceThought

Community Trekker

Joined:

Jun 23, 2011

Yeah, the more I looked at it (the incomplete gamma), the less happy I was as well, and I couldn't find an easy way to appeal to symmetry, especially for these functions.

Oh and as far as translating FORTRAN to JSL, the word "doomed" is quite appropriate. The MATLAB code would be easier to translate, I think.

Steve Denham

Super User

Joined:

Jun 23, 2011

In JMP use Floor(x) to get what is Int(x) in Excel.

Community Trekker

Joined:

Aug 8, 2012

Sorry, that was dumb of me. I meant like a integrate function, not like a round to integer function.

-- SurfaceThought

Super User

Joined:

Jun 23, 2011

Here's a Fortran implementation: http://people.sc.fsu.edu/~jburkardt/f77_src/toms385/toms385.html

Shouldn't be too hard to convert to JSL.

Community Trekker

Joined:

Aug 8, 2012

This is where I get way in over my head

I am much more of a statistician than a programmer ( although I am not much of either ). How close is FORTRAN to JSL? I have no knowledge of it. If I copy and pasted the entire thing into a JSL script window, and kept on trying to run it and fixing errors as they came up, would I be successful or would I be doomed without some knowledge of FORTAN?

Thanks,

--SurfaceThought