Skymaker oversampling

[lino]

Hi Emmanuel,

I have two questions about skymaker.
1) playing with the skymaker input params I found that if you increase the PSF_OVERSAMP value you get a FWHM smaller then the input SEEING_FWHM. That sounds strange to me.
Any hint about a reasonable value for that parameter?

2) setting SEED_MOTION and SEED_STARPOS to non zero values I found that the images produced without changing any params are not exactly the same (few stars in the difference image).

Thanks in advance for the answer

Lino

[Emmanuel Bertin]

Hello Lino,

thanks for posting on the forum.

Quote:1) playing with the skymaker input params I found that if you increase the PSF_OVERSAMP value you get a FWHM smaller then the input SEEING_FWHM. That sounds strange to me.
Any hint about a reasonable value for that parameter?

One side effect of PSF_OVERSAMP is that it decreases the extent of the PSF model on the simulated image (if PSF_MAPSIZE is kept constant). So if the PSF FWHM is measured with SExtractor's FLUX_RADIUS, with a PSF_MAPSIZE too small for such a high sampling, there is definitely the possibility to have an impact on the half-light radius (mostly because of missing wings).
To simulate ground-based images, for a given PSF_MAPSIZE (which basically sets the speed at which each object is computed), it is sufficient to adjust PSF_OVERSAMP to an odd value, in such a way that the upper boundary of the pupil range is higher than Telescope_diameter + 2 times "r0" (Fried radius). In other words, increase PSF_OVERSAMP until the WARNING message disappear!

Quote:2) setting SEED_MOTION and SEED_STARPOS to non zero values I found that the images produced without changing any params are not exactly the same (few stars in the difference image).

This should not happen. You simulate only stars?

Cheers
- Emmanuel.

[lino]

Hi Emmanuel,

Emmanuel Bertin Wrote:Hello Lino,

thanks for posting on the forum.

Quote:1) playing with the skymaker input params I found that if you increase the PSF_OVERSAMP value you get a FWHM smaller then the input SEEING_FWHM. That sounds strange to me.
Any hint about a reasonable value for that parameter?

One side effect of PSF_OVERSAMP is that it decreases the extent of the PSF model on the simulated image (if PSF_MAPSIZE is kept constant). So if the PSF FWHM is measured with SExtractor's FLUX_RADIUS, with a PSF_MAPSIZE too small for such a high sampling, there is definitely the possibility to have an impact on the half-light radius (mostly because of missing wings).
To simulate ground-based images, for a given PSF_MAPSIZE (which basically sets the speed at which each object is computed), it is sufficient to adjust PSF_OVERSAMP to an odd value, in such a way that the upper boundary of the pupil range is higher than Telescope_diameter + 2 times "r0" (Fried radius). In other words, increase PSF_OVERSAMP until the WARNING message disappear!

Quote:2) setting SEED_MOTION and SEED_STARPOS to non zero values I found that the images produced without changing any params are not exactly the same (few stars in the difference image).

This should not happen. You simulate only stars?


Cheers
- Emmanuel.

Yes, only stars.

ciao

Lino

[Emmanuel Bertin]

Gloops. I will investigate and report back to you!
Cheers
- Emmanuel.

[natsushi]

Hi Emmanuel and Lino,

(08-01-2007 17:11)Emmanuel Bertin Wrote:  Gloops. I will investigate and report back to you!
Cheers
- Emmanuel.

I also have same problem on random seed of stars; with exactly same random seed, the star positions differ run by run.

This happens if one compiles with multi-threads mode(maybe default?).
The quick solution is explicitly disabling the multi-threads option, i.e. ./configure --disable-threads.
Also be sure to have single thread mode FFTW library.

-Atsushi,

[Emmanuel Bertin]

Hi Atsushi,

what version of SkyMaker are you referring to? This should have been fixed since V3.2.1.
- Emmanuel.

[bmv]

(08-01-2007 15:54)Emmanuel Bertin Wrote:  One side effect of PSF_OVERSAMP is that it decreases the extent of the PSF model on the simulated image (if PSF_MAPSIZE is kept constant). So if the PSF FWHM is measured with SExtractor's FLUX_RADIUS, with a PSF_MAPSIZE too small for such a high sampling, there is definitely the possibility to have an impact on the half-light radius (mostly because of missing wings).
To simulate ground-based images, for a given PSF_MAPSIZE (which basically sets the speed at which each object is computed), it is sufficient to adjust PSF_OVERSAMP to an odd value, in such a way that the upper boundary of the pupil range is higher than Telescope_diameter + 2 times "r0" (Fried radius). In other words, increase PSF_OVERSAMP until the WARNING message disappear!

Cheers
- Emmanuel.

Very important note! Thank you.
Probably should introduce a new WARNING message.
That oversampling is too high.
Too high oversampling leads to a shift in position:

[bmv]

There are some problems with PSF_OVERSAMP <5 and large PSF_MAPSIZE.

Quote:sky -M1_DIAMETER 0.508 -M2_DIAMETER 0.1778 -ARM_COUNT 0 -MAG_ZEROPOINT 20.8208 -EXPOSURE_TIME 300 -PIXEL_SIZE 0.1 -SEEING_FWHM 0.5 -IMAGE_SIZE 4096 -STARCOUNT_ZP 5e3 -BACK_MAG 21 -WAVELENGTH 0.5 -PSF_MAPSIZE 8192 -MAG_LIMITS 12.0,20.0 -PSF_OVERSAMP 5

> WARNING: sky.conf not found, using internal defaults

----- SkyMaker 3.10.4 started on 2013-07-20 at 17:54:48 with 8 threads

Pupil range: 0.6-5156.6mm / Subsampling: 1X / ro=20.1cm / motion=0.00'' rms
57 sources added
> All done (in 50.8 s: 1.1 sources/s)

Quote:sky -M1_DIAMETER 0.508 -M2_DIAMETER 0.1778 -ARM_COUNT 0 -MAG_ZEROPOINT 20.8208 -EXPOSURE_TIME 300 -PIXEL_SIZE 0.1 -SEEING_FWHM 0.5 -IMAGE_SIZE 4096 -STARCOUNT_ZP 5e3 -BACK_MAG 21 -WAVELENGTH 0.5 -PSF_MAPSIZE 8192 -MAG_LIMITS 12.0,20.0 -PSF_OVERSAMP 3

> WARNING: sky.conf not found, using internal defaults

----- SkyMaker 3.10.4 started on 2013-07-20 at 18:00:32 with 8 threads

Pupil range: 0.4-3094.0mm / Subsampling: 1X / ro=20.1cm / motion=0.00'' rms
> Adding 79 stars...
Floating point exception (core dumped)

[Emmanuel Bertin]

Hi bmv,

thanks for the bug report. It is now fixed in the latest development version (v3.10.5).
- Emmanuel.

[Emmanuel Bertin]

(07-20-2013 09:06)bmv Wrote:  Very important note! Thank you.
Probably should introduce a new WARNING message.
That oversampling is too high.
Too high oversampling leads to a shift in position:

Hi bmv,

I am confused, what does the plot show exactly (especially on the y axis)?
- Emmanuel.