Re: TASS Database, Store ra,dec Sigma or not?

[Chris Albertson <chris@topdog.pas1.logicon.com>]

aah@nofs.navy.mil wrote:
> 
>   I disagree with Michael (that's ok, isn't it?).
>   We need coordinate sigmas to be sure that an object is properly matched
> with a mag16 object, for moving objects, etc.  The easiest way to calculate
> the coordinate errors for the master tass_cat catalog is to have three fields:
> number_observations, sum_coordinate and sum_coordinate^2.  Then, at any time,
> one can calculate the errors from those three quantities.  Unfortunately,
> the sums have to be 64-bit floats to get the proper scaling and precision;
> the number_observations field can certainly be a 16-bit integer.

I don't see any disagreement here.  I thought that is what Mcheal was
saying.  To quote Michael:
>>  As long as we store the sum of the squares of the positions,
>>  then we can re-construct the standard deviation as needed.

Here is my current plan as based on comments I've read to date,
Store the following for RA and DEC:
  (1) Mean, 32-bit int, units mas
  (2) Count, 16-bit int, 
  (3) Sum of squares. 64-bit float, units (Degress)^2 not (mas)^2 
 
I am not currently storing the Sum.  I figure  Sum = Mean * Count.  
There is some loss of precision but to how many decimal points do 
we need to know the standard deviation?

I think we need the mean, even if it can be computed from Sum/Count.
We _will_ search on mean RA,DEC.  So if we want to store Sums it is
only to make the Std. Dev. better. So rather then Sum which requires
a 64 bit float, why not just store Std Dev. in a 32-bit float?  




>   Sorry -- my philosophy with scientific data is to always quote errors
> with any published value.  If it requires extra storage in the master
> catalog, then I feel it is a necessary evil.
> Arne

-- 
--Chris Albertson

  chris@topdog.logicon.com                Voice:  626-351-0089  X127
  Logicon RDA, Pasadena California          Fax:  626-351-0699