Milano, March 28th 2003

Dear collaborators,

we have developed a new version of the h8 alignment, which takes 
into account the correlation between parameters.
 
The new version fits the x and y coordinates and the xy angle
in order to get a first rough alignment. Then all parameters 
(x,y,z coordinates and xy,xz and yz angles) are fitted together, 
building the full correlation matrix. 

The alignment strategy is defined in the main program h8main.F :

call align(19,2,4) // fit translations for planes 2-4 (strips)
call align(63,2,4) // fit all parameters for planes 2-4 (strips)
call align(19,5,5) // fit translations for plane 5 (pixel)
call align(63,5,5) // fit all parameters for plane 5

the first number is a bitted word which specifies which parameters are to be
fitted; the description of the bit pattern is found in align_planes.F

After some debugging, the problems we have had with the new version have 
finally been tracked to be due to a typing error. 
We have tested the new strategy with non 
irradiated single chips (Ge04,Ge09,Bonn7,Bonn9 at 0,10,30 tilt angles)
of May. The alignment converges rapidly, all residuals distribution are
fine and the best-fit values of the alignment parameters make sense. 

We have still to test the new version with irradiated chips, modules, 
and data taken in a magnetic field; we hope to provide a release 
of the program next week. 

As some of you need the values of the tilt angles for their analysis 
and simulation, and we are confident that the alignment works fine 
for May 2002 not-irradiated chip data, we report below the values 
of the tilt angle in the xz plane for the pixel detector (for the strips 
angles compatible with 0 are found).

The error in cos(alpha) is always 0.8 10^-3, but due to the non-linearity 
of the cosine this implies a sensitivity of 2 degrees for alpha = 0 
and less than 0.1 degrees at alpha = 30. In fact at normal incidence values 
compatible with 0 have been found for the angle, and we report 95% C.L. 
bounds. All other errors are 1-sigma values.

Run   Nominal tilt        dem.1        dem.2

1319      0              < 3.6          < 3.8         (2-sigma limits)
1320     10           9.7  +- 0.3     10.0  +- 0.3
1323     30          29.18 +- 0.08    29.99 +- 0.08
1327     40          39.38 +- 0.06    39.97 +- 0.06
1329      0              < 3.8          < 4.1         (2-sigma level)
1330     10           9.9  +- 0.3      7.6  +- 0.3
1333     30          29.73 +- 0.08    27.81 +- 0.08

Large deviations from nominal angle are observed only for Bonn7 data 
(run 1329-33 dem. 2) and can explain much of the cluster size 
difference between Bonn7 and Bonn9.

Best Regards, 

the Milan h8 team