Collimations
Since the model update of 18 Feb there have been no obvious sudden shifts in collimation. The plots below are the 'g s n' and 'g e n' plots from TPOINT, plotting, respectively, dS (daz) and dE (del) residuals against 'number' - which is like 'time' but with equal interval between points.

In this concatenated set of data the raw rms scatters in dS and dE are 2.0" and 2.7" respectively, but some of this is due to nightly variations in the collimation offsets, as is reported below. However, the nightly differences are clearly small (of order 1"), and cannot convincingly be attributed to hardware problems at this level.

The nightly data also show a range in performance, from extremely good, as seen during the dedicated pointing run of 18 Feb, to poor(er), with rms residuals in (daz,del) approaching (2.0",2.5"). (Nothing quite as poor as was seen before the model update, thankfully). As I predicted I would (!), I feel I can blame any given night's poor performance on something other than the model ! - like the weather, for instance.

Temperatures
In the same period, many individual SCUBA datasets have shown systematic, temporal trends of the elevation residual, del. These may be causally related to the antenna temperature in one form or another - either T_mean, the mean temperature (of the legs, which is our only current available temperature measurement feasibly releveant to this problem) - or to T_diff, the temperature difference between the front and back legs. The trends from individual nights are listed below:

   Fitting   del = slope * X  +  const    to N points

   dataset    N   raw    ----- X=T_diff ----   ----- X=T_mean ----- 
                  rms    slope +-  resulting   slope +- resulting
                                      rms                   rms
                   "                   "                     "
  pt030218   21   1.6    -1.0  2.2    1.6      0.28  0.39   1.6
  pt030219   14   2.3    -3.0  1.9    2.2     -1.03  0.83   2.2
  pt030220   18   2.1    -5.6  1.1    1.3     -0.63  0.15   1.5
  pt030221   17   2.2    -2.4  2.4    2.2      1.21  0.58   2.0
  pt030222   17   1.9   -11.3  5.3    1.7      0.45  0.86   1.9
  pt030223   15   1.3    -1.3  4.3    1.3     -1.65  0.75   1.2
  pt030224   16   1.7    -5.8  2.3    1.4     -0.08  1.47   1.7
  pt030225   16   1.9     1.6  4.8    1.9     -2.20  0.48   1.3
  pt030226   24   2.3    -0.5  1.1    2.3     -0.08  0.20   2.3
  pt030227   30   2.3    -3.0  0.9    2.0     -0.11  0.16   2.3
  pt030228   26   2.7    -3.0  0.6    1.9     -0.46  0.08   1.8
  pt030301   29   2.3    -2.8  2.2    2.3      0.70  0.36   2.2
  pt030302   25   2.5    -4.6  1.5    2.2      0.19  0.38   2.5
  pt030303   21   2.1    -3.8  1.0    1.6     -0.78  0.13   1.3   
           ----        -----------           ------------
            289     mean -3.2  0.4            -0.43  0.11

Some nights strongly support a new value for the relevant coefficient, while others, of similar temperature range, do not ! Most frustrating, but indicative that the nightly behaviour of the elevation residuals, beyond that already accommodated by the temperature-dependent algorithms within the TEL task, may be only indirectly related to temperature. The weighted means of the slopes in the table above suggest that corrections could be made to the coefficients of either of the elevation-temperature algorithms : those relating del to T_diff or T_mean; although the former seems stronger.

The data from this period were collected in their entirety for similar analysis, and this was done in 2 ways :

1. by a simple concatenation (as done for the 'g s n' and 'g e n' plots above),
2. after adjusting each night's data for the mean collimations (pointing offsets)

The totality of data then look like this in each case :

The overall performance statistics are notably improved when the data are corrected for the nightly collimations : going from rms's in (daz,del) of (1.9,2.5) to (1.7,2.2).

These datasets were then analyzed as above :

   Fitting   del = slope * X  +  const    to N points

   dataset    N   raw    ----- X=T_diff ----   ----- X=T_mean ----- 
                  rms    slope +-  resulting   slope +- resulting
                                      rms                   rms
                   "                   "                     "
      1     289   2.5    -3.0  0.4    2.3     -0.10  0.05   2.5
      2     289   2.2    -2.7  0.3    2.0     -0.16  0.04   2.2

A change to the T_diff algorithm offers the greater improvement to the performance, and is statistically more significant (at the 8-sigma level cf. at the 4-sigma level).

The current value of TEMP_SLOPE in mt_teldir:TEL.ifl is 6.0"/deg, and the above analysis suggests changing this to 8.7"/deg. I'm reluctant to do this without good reason and revisited the comment that the 6"/deg value for this coefficient agreed well with "simple engineering calculations". Further calculations by Tomas Chylek (20030304) suggested that extreme values of 2"/deg and 40"/deg might pertain, depending upon the rigidity of the apex of the A-frame under front-leg-expansion.

Such a possible range makes the empirical determination (6"/deg) as valid as an 'engineering' value, with the corollary that any suggested (empirical) corrections may also be adopted (or not) without reference to the engineering/FEA modelling.

The case for adjusting the value of TEMP_SLOPE is fairly convincing, but, so far, my reluctance seems stronger. Watch this space.