Subject: Re: jemx mosaics Hi - sorry to disturb you while you were away, I hope you had a nice trip ! I had a bit more of a think about this from my end: I understand about not wanting to renormalise the images, but for a different reason: to me it seems that there is something not right - the significance should simply be the flux measured, divided by the error on the flux (sqrt of the variance), it shouldn't need normalising if both of those are correct, the distribution should have a width of 1 sigma by definition. The distribtion of the background significances should remain the same gaussian (mean of 0, std of 1 sigma, area of number of pixels in the image) although it should look smoother as exposure increases. In plot3.ps, I show distributions from 2 observations - the 20scw (70 ksec; solid histogram) and 10scw (36ksec; dashed histogram) ones. I chose these as both have at least 1 full hex dither pattern in and so the output mosaics are both about the same size (about 209,000 pixels in both). You can see how the two distributions have different widths - (4.8 sigma from 20 scw, 3.5 sigma for 10 scw). In fact looking at the high significance tails in this plot, which is where we expect anything 'non-background' to be (i.e. the sources), then for the 20scw one we might think we could believe anything above 20 sigma as a source, whereas if we look for half as long we might believe anything above about 15 sigma... which implies we are more sensitive to faint sources the shorter we look at them ! The dot-dash line is a gaussian with a std of 1 sigma and normalised to have the same area (= no of image pixels) as the 10 scw image - I think this is what these distributions should look like; regardless of the length of the observation, the distributions should look like the ideal curve although the longer the observation the smoother match to this ideal curve they should be. The distributions from the mosaics have a huge std compared to the ideal curve. In plot.ps there are two plots: the first shows the value of the significance of the detection of GRS 1915 (from rev 0295 when it was in a steady state) as extracted from the mosaics with jmx_src_locater, and the values of the width of the background distribution with exposure; the problems then become clear in the second plot, where I show the ratio of those two earlier curves (signal to noise if you like) - this seems to show that there isn't much point in looking at a particular source for more than ~40 ksec as it doesn't actually gain anything in sensitivity or detection significance ! So, I think there is something wrong in the calculation of the significance images. I have checked a couple of images by dividing the reconstructed flux image by the square root of the varianece image, and this seems to give exactly (near enough) the same as the significance image, so perhaps there is something wrong with the calculation of the variance. I suppose there could be an effect if there is a large random systematic on every individual Scw image, but if thats true then we should really look at ways of getting rid of it as it is having a very large effect. This also causes a problem for using jmx_src_locator in the barn, since to get meaningful results out we'd have to use a different value of min sig depending on the observations, so it would be good to get to the bottom of it. What do you think ? Cheers Simon Jérôme Chenevez wrote: > Hello Simon, > > Sorry not having answered your mail sooner, but I am in holidays this week > (just having a look from home to the Bulge data). > I have to think about your question, but I remember we decided long ago > during a JEM-X SW meeting not to normalise the significance artificially > to 1, letting the user to take account of the FWHM of the significance > distribution, as you did I guess. > > Cheers, > Jérôme > > >> Hello again, >> >> Regarding my latest Infrequently Asked Question I ran a bit of a test, >> to see what happens if I simply take the set of INTEGRAL data (GRS 1915 >> from rev 0295 - total 28 ScW) and then make mosaics for subsets of that >> (I did only the first 2 scws (~7ks), first 5 (18ks), first 10 (36ks) and >> all 28 of them (100ks)). The test for 20 ScWs is still running, but >> from the results so far I'm afraid it doesn't seem like much point in >> waiting for it to finish. >> >> So I have used fimgstat to report the statisics of the significance >> images in the 3-10 keV band and plotted the value returned for the >> standard deviation of the images against the length of the observation >> (I have only read TELAPSE from the header, not the GTI information, but >> the effect is clear). >> >> There seems to be a seriously strong linear correlation between the >> standard deviation of the distribution of all the pixel values in the >> significance image and the length of the observation - if the >> significance images were correct then the points should be scattered >> around a std of 1 regardless of observation length. >> >> My GB observations are all about 12ksec in length so there isn't much >> range on time and they seem to have std around 1.5, which is a bit lower >> than for the GRS observations, so I'm sure there is also an effect >> based on what is in the field of view. However from the GRS data there >> clearly seems to be a problem involving the time in the calculation, >> and thats much more serious. >> >> Sorry to drop this on you on a Friday afternoon - I'll be happy if you >> can prove me wrong !! >> >> Simon >> >> >> Carol Anne Oxborrow wrote: >> >>> Dear Jerome, >>> >>> could you please send me a copy of your reply to Simon? I'll try and >>> make another question and answer for the JEM-X FAQ from his and your >>> emails - that way we'll get the FAQ populated by actual (Frequently - >>> is once `frequently'?) Asked Questions. >>> >>> Thanks, >>> >>> CAO >>> >>> DANISH NATIONAL SPACE CENTER >>> Dr. Carol Anne Oxborrow >>> Email: oxborrow@space.dtu.dk Homepage: >>> http://www.spacecenter.dk/~oxborrow Telephone (direct): +45 35 32 57 >>> 33 >>> >>> Main teleph.: (+45) 35 32 57 00 >>> Fax: (+45) 35 36 24 75 >>> Main E-mail: office@spacecenter.dk >>> >>> A strengthened Danish National Space Institute, DTU-Space >>> >>> Read more about the Danish National Space Institute at: >>> www.spacecenter.dk >>> >>> -------------------------------------------------------------------------------- >>> ----------------- >>> >>> >>> >>> >>> >>> >>> >>> > > >