[daip] CONVL question

[Samuel Conner]

To Whom it may concern,

    The HELP/EXPLAIN info in 31DEC04 CONVL says that only the inner 1/4 of the map is properly convolved in both OPCODE = 'IMAG' and OPCODE = 'GAUS' mode.

     I infer from this that if I want to do a 'GAUS' convolution and convolve the entire FOV of my image, I should first PADIM with zeros to twice the image size, CONVL, and then SUBIM to recover my original geometry.

     My question is this:

     Out of curiosity, I wanted to see what effect there is if I DON'T PADIM to twice the size. So I CONVL'd both ways and computed the ratio of the two results. My input image is 1024 square.

     To my surprise, the ratio is 1.000 over the entire FOV, not only in the inner 1/4, but also the outer 3/4, with the exception of a sprinkling of individual pixels which differ from 1.000 by small amounts. 

     The number of these pixels is

     outside .99 - 1.01             2 pixels (.96 in the outer 3/4; and 1.15 in column 2)
     outside .995 - 1.005           several
     outside .999 - 1.001           dozens
     outside .9999- 1.0001          hundreds

     It seems plausible to me that the discrepancies I am seeing, which are mostly below the 10th bit, could
     simply be the generic limitations of single precision arithmetic within AIPS. 

     Am I right to interpret this to imply that CONVL is doing the right thing in 'GAUS' mode with images that
     have not been double-sized?

     Thanks for your time and help.

Sincerely yours, 

Sam Conner




   The number of these pixels is THe number of these pixels
is of order. Most of the deviations are at less than 1