R help - Apr 2012 - Nlme not working on very similar initial values

Hello,

I am using CentOS Linux 6.0, R 2.14.1 and nlme 3.1-103

I am trying to fit some models using nlme, and what was happening was that it
would get to some datasets and just "stall", the memory usage would
grow and grow and eventually crash.

I looked carefully into one of the datasets and ran it separately. It worked. 
After narrowing down on all possibilities, I found that the main difference
between both was in the initial values. If one subtracts them:
> initialValuesOriginal - initialValuesThatWorks
          A.R           A.a           A.y           B.R           B.a
-3.194922e-08  0.000000e+00  0.000000e+00 -9.249630e-08  0.000000e+00
          B.y           C.R           C.a           C.y           D.R
 0.000000e+00 -1.713935e-06 -8.639821e-09  1.032083e-09  3.716880e-08
          D.a           D.y           E.R           E.a           E.y
 0.000000e+00  0.000000e+00  1.766460e-06  0.000000e+00  0.000000e+00

How come such a small difference can have such a big effect?  (working ok,
versus stalling the entire computer)

I also just played with rounding to 10 decimal digits and it worked fine.

I will look into what nlme is doing tomorrow, but was wondering if perhaps there
is something that I don't know about how R works, or using numbers with many
decimals in nlme for initial values, that would create such a phenomenon.

Thanks in advance,
Ramiro

	[[alternative HTML version deleted]]

Welcome to the wonderful world Of nonlinear model fitting in ill-conditioned
problems. See also  chaos theory.

Bert

Sent from my iPhone -- please excuse typos.

On Apr 3, 2012, at 3:03 PM, Ramiro Barrantes <ramiro at
precisionbioassay.com> wrote:
> Hello,
> 
> I am using CentOS Linux 6.0, R 2.14.1 and nlme 3.1-103
> 
> I am trying to fit some models using nlme, and what was happening was that
it would get to some datasets and just "stall", the memory usage would
grow and grow and eventually crash.
> 
> I looked carefully into one of the datasets and ran it separately. It
worked.  After narrowing down on all possibilities, I found that the main
difference between both was in the initial values. If one subtracts them:
> 
>> initialValuesOriginal - initialValuesThatWorks
>          A.R           A.a           A.y           B.R           B.a
> -3.194922e-08  0.000000e+00  0.000000e+00 -9.249630e-08  0.000000e+00
>          B.y           C.R           C.a           C.y           D.R
> 0.000000e+00 -1.713935e-06 -8.639821e-09  1.032083e-09  3.716880e-08
>          D.a           D.y           E.R           E.a           E.y
> 0.000000e+00  0.000000e+00  1.766460e-06  0.000000e+00  0.000000e+00
> 
> How come such a small difference can have such a big effect?  (working ok,
versus stalling the entire computer)
> 
> I also just played with rounding to 10 decimal digits and it worked fine.
> 
> I will look into what nlme is doing tomorrow, but was wondering if perhaps
there is something that I don't know about how R works, or using numbers
with many decimals in nlme for initial values, that would create such a
phenomenon.
> 
> Thanks in advance,
> Ramiro
> 
>    [[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.