A comparison of FIT9 and TPOINT

Summary
A comparison of the old and new models is made in a TCS document "Treatment of Pointing in the new JCMT TCS", but an empirical approach is taken here.

Allsky pointing data was simulated and analysed using FIT9 and TPOINT . The introduction of specific changes to the 7 parameters of the model were in general perfectly detected by FIT9 , but were less perfectly determined by TPOINT . This indicates some difference in interpretation of the effects of these parameters within the two models - although these tests are not able to indicate which model is correct. The different algorithmic forms of the paramaters used by the two models are examined - there is a difference in the interpretation of 'azimuth residual' that explains apparent differences in the functional forms attributed to 4 of the 7 parameters, but confusion between the FIT9 parameters ##4 & 5 and the TPOINT parameters CA & IA remain unresolved.

Introduction
With the new JCMT TCS installed and being finally debugged, the default pointing analysis package must shift from the FIT9 package written by Sidney Kenderdine and in use since JCMT came on-line, to Patrick Wallace's TPOINT . It is important to compare the two packages, and the process, I thought, would be simple :

• simulate allsky pointing data :
  • for each of the 7-parameters separately, produce errors as specfied in the table for a grid of stars evenly scattered in (az,el) space. Spacing of 15 degrees in azimuth between azimuths 15 and 435, and 10 degrees in elevation between elevations 20 and 80 make for 203 simulated data.
• analyse with FIT9
• analyse with TPOINT
• compare suggested changes to the 7-parameter model - expecting minor differences only !

TPOINT provides standard errors on the parameter fits, and in each case they are very small in comparison to the derived values; eg. AN = 4.0 + 0.1 is the least significant of the determinations above.

• The above tests were repeated with the cos(latitude) terms to the FIT9 parameter #2 reinstated after a 2 year absence. (The reasons for their suppression in the October 1997 parameterization stem from a reconsideration of its geometrical roots ). However, the results were in essence unchanged, with the TPOINT derivation of AW worsening slightly to -3.8".

Of particular note are

1. a sign convention difference between parameter #2 of the FIT9 set & AW, and between #6 & IE.
2. the slightly less-than-perfect determinations by FIT9 in the cases of #3 and #7.
3. the confusion experienced even by FIT9 in separating changes in parameter 5 from those in 3 and 4 - although one's experience with reducing real data is that sets of large changes are sometimes suggested for parameters 3,4,5 that are clearly unrealistic (ie not related to real changes in the telescope hardware) and that some considerable cross-talk affects these three parameters.
4. the much larger confusion within TPOINT when analysing the same data.
5. the unrealistically large changes suggested by TPOINT for all of IA, NPAE and CA when parameter #3 was changed.
6. the confusion between the descriptive equivalence of FIT9 parameters ## 4 and 5 and their respective TPOINT couterparts, CA and IA, and the observed algorithmic equivalence between #4 and IA.

Conclusion

• The data tested here are generated using the FIT9 algorithms, so it is no surprise that FIT9 does pretty well in completing the circle and determining what it spawned. There may be interpretive differences between FIT9 and TPOINT , but to date (01 Nov 2000) the TPOINT model appears to be working quite satisfactorily, with one notable exception. The large value of NPAE (~100") for the JCMT has revealed a lack of rigour in some of the trigonometry near the poles. An updated, more rigourous, version of TPOINT , was created by Pat Wallace and is being installed.