NLS results - 486
          ----NDigits Beta---- NDigits
 Model    Start1 Start2 Start3  SE(B)  Pass=1/Fail=0
 Misra1a     9.5   11.1   15.0    9.2   1.
 Chwirut2    8.6   10.9   15.0    9.0   1.
 Chwirut2   10.6    8.5   12.3   10.8   1.
 Lanczos3    8.3    9.3   10.5    8.3   1.
 Gauss1      8.8   10.6   10.6    8.6   1.
 Gauss2     10.3   10.3   10.3   10.4   1.
 DanielWo   10.2   10.8   11.2   10.2   1.
 Misra1b    10.9   10.9   11.0   10.9   1.
 Kirby2     10.6   10.5   10.5   10.8   1.
 Hahn1       8.6    9.8   10.8    9.4   1.
 Nelson      8.9    9.2   10.9    9.2   1.
 MGH17      -5.6L  10.3   11.4    9.9   1.
 Lanczos1   10.6   10.6   10.6    3.6*  0.
 Lanczos2   10.4   10.4   10.4   10.3   1.
 Gauss3      9.2   10.5   15.0    8.6   1.
 Misra1c    10.8   10.8   10.8   10.6   1.
 Misra1d    11.3   11.6   12.6   10.9   1.
 Roszman1    7.5   10.9   15.0    8.0   1.
 ENSO        7.4    7.4    9.7    8.5   1.
 MGH09      -3.4NC  8.1   12.1    8.2   1.
 Thurber     7.8    7.7   10.4    7.0   1.
 BoxBOD      8.4    9.3   11.7    8.4   1.
 Ratkows2    8.8    9.6   15.0    8.4   1.
 MGH10      -3.8L  10.9   10.9   10.6   1.
 Eckerle4   -3.2F   9.9   10.5    9.9   1.
 Ratkows3    7.4    9.0   11.0    9.1   1.
 Bennett5   11.0   11.0   11.0   10.4   1.
 NLS - 486 summary:  26./27 passed.

 NLS results - Sun 4
          ----NDigits Beta---- NDigits
 Model    Start1 Start2 Start3  SE(B)  Pass=1/Fail=0
 Misra1a    11.1   11.1   15.0   10.8   1.
 Chwirut2    8.7    9.9   15.0    9.1   1.
 Chwirut2    8.6    8.5   13.8    9.0   1.
 Lanczos3    8.3    6.5   10.5    8.3   1.
 Gauss1     10.6   10.6   10.6   10.6   1.
 Gauss2     10.0   10.0   10.3    9.9   1.
 DanielWo   10.3   11.2   11.2   10.4   1.
 Misra1b    10.9   10.9   11.0   10.9   1.
 Kirby2      8.9    8.3   10.5    8.7   1.
 Hahn1      10.0   10.7   14.4   10.3   1.
 Nelson     10.7    7.6   10.9   10.6   1.
 MGH17      -7.2L   7.2   11.2    6.8   1.
 Lanczos1   10.6   10.6   10.6    3.5*  0.
 Lanczos2   10.3   10.4   10.4   10.2   1.
 Gauss3     10.5   10.5   10.5   10.5   1.
 Misra1c    10.8   10.8   10.8   10.6   1.
 Misra1d    11.3   12.6   15.0   11.1   1.
 Roszman1    7.0    8.9   15.0    7.5   1.
 ENSO        6.8    6.5   15.0    8.0   1.
 MGH09      -3.4NC  7.3   10.4    7.4   1.
 Thurber     7.3    7.6   10.8    6.6   1.
 BoxBOD      8.4    9.3   11.7    8.4   1.
 Ratkows2    8.9   10.7   15.0    8.5   1.
 MGH10      -3.8L  10.9   11.0   10.8   1.
 Eckerle4   -3.2F  10.5   10.5   10.5   1.
 Ratkows3    8.3    8.4   11.1    8.4   1.
 Bennett5    9.2   11.0   15.0    9.1   1.
 NLS - Sun summary:  26./27 passed.

Convergence notes:
F  = Failure to find increasing step, with norm of gradient > .1
NC = No Convergence, after 100 iterations (may be heading for
     the local optimum at the boundary for MGH09)
L  = convergence to Local optimum (inferior to the global optimum)
     In both cases, these involved singularity of the parameters:
       MGH17:  all terms on the RHS became zero except the intercept.
       MGH10:  the RHS became zero.
If no note was given, then it either converged normally, or failed
to find an increasing step, but with norm of gradient < 1E-10 .

* For Lanczos1, TSP had to be enhanced (9/2/97) to keep iterating when the
SSR became less than 1D-21 -- previously it would declare a
perfect fit at around that point.  On both 486 and Sun, it does not
appear possible to reproduce the NIST certified SEs to a full 4 digits,
with double precision data, due to the large number of significant
digits (14) in the data.  The SSR at the certified coefs is 3.98E-21 (vs. 
1.4308E-25 given by NIST with extended precision data), so the SEs
are different by a factor of 170 at the certified coefs.  If we iterate as
far as possible, we can get a SSR of 1.4303E-25 on the Sun4,
and 1.4277E-25 on a 486.  These SSRs are also too different from the NIST
value to get a full 4 digits of agreement in the SEs.

Double precision data, and a convergence tolerance of 1E-11 were used
(since the goal was to obtain 11 correct digits).

SE(Beta) is computed from the Standard Errors for the Betas that
achieve the lowest SSR of the first 2 sets of starting values.
A "Pass" is defined as at least 4 correct digits in Beta and SE(Beta),
for the best Betas just described.  TSP would also pass the same tests
with at least 6 digits of both Beta and SE(Beta).