Unknown error notation in Mathcad program.

[nobozoz-disable]

I am trying to get "NIST Regress(4).mcd" and or "FunctionFitting(1)_jr1.mcd" to work. I have mc2000 and mc14 - no version in between. mc14 will open either worksheet, but has a few "error flags" scattered throughout the spreadsheet like "This variable or function is not defined above." and "illegal l-value in assignment".

mc2000 will open only "FunctionFitting(1)_jr1.mcd", but the errors are the same as near as I can tell.

A snippet of the offending program code looks something like this: "f(>X<,NameVec(n,>V<))" (without the quotes).

I don't understand the ">X<" and ">V<" notation. The error message associated with this flagged notation is: "This variable or function is not defined above."

I'd post a gif of the offending areas, but I haven't figured out how to insert a gif in this message.

Thank you, Jim Regan

[TomGutman]

The NIST regression sheet definitely does not work in MC14 (MC12 introduced several incompatibilities, none of which have ever been resolved). I don't expect it to work in MC versions prior to MC11 either. Almost definitely not in MC2000, might work (perhaps with tweaking) in MC2001i.

The >X< is just a variable name. It is chosen to be very unusual, so as to minimize the chances that it might be used by the user. It is used in functions intended for symbolic processing, where it is important for some variables not to be defined. The numeric processor is unaware of this intent, and proceeds to complain about undefined variables. Complaints that are meaningless and should be ignored.
__________________
� � � � Tom Gutman

[nobozoz-disable]

Thanks for the prompt explanation and also for the warnings about trying to make this work in any version other than mc11. It's a shame all that work can't be supported across Mathcad versions.

[ptc-1368288]

On 11/21/2009 3:50:20 AM, nobozoz wrote:
>Thanks for the prompt
>explanation and also for the
>warnings about trying to make
>this work in any version other
>than mc11. It's a shame all
>that work can't be supported
>across Mathcad versions.
______________________________

I have two comments that may be of interest.

1. Based on the detailed explanation and the pertinence of the visit, this should go in the "recent programming feature". The case is in work sheet, then why the typical 11 structure does not work in 14 ? PTC = Explain. If that functionality can be restored in 14, what made the 14 structure incompatible ? From intentional design or a band-aid consequence left undetected. Again and in other words, what's the benefit of the 14 structure ?

2. At the time of the DAEP, some NIST data set were explored for comparison with Mathcad. Later on, my recollection is that NIST came in the collab for discussion. From visiting the NIST collection of their model and some justification, little or nothing is worth considering. They toke arbitrary models that best fit on the residuals, but is not necessarily the bet statistical fit vis the calculation of the parameters. The choice of the model is paramount, the library of the NIST models is very limited for general use in curve fitting. At this point, my comment is personal but it reads "NIST = useless". Now, I support and close my comment: after discovering from Paul W. the "PWMinerr", genfit is not needed anymore and lot more involved in construct than "PWMinerr" ... this is about fitting models. For the other general rational interpolating fit, the NIST method is unknown but it works. There are several methods for minimax. In the Mathcad DAEP [version 11], the rational fit that I have called "RobertGenfitMatrix" is robust. I have used it 1000's times (not necessarily for 1000's resulting fits, but a lot !). The work sheet is complete, stands alone, no guesses are needed, extract the coefficients, symbolic output the literal form, outputs the continued fraction form.

Conclusion:

Not to detract the Mathcad work sheet NIST, not to detract the NIST organisation/effort ... etc, The two have no use in Mathcad for the data set fitters. After "PWMinerr", no fit were done otherwise, and by same token declassifying NIST from a fitting reference, maybe except for the models that they may have augmented since my last visit, years ago.

jmG