Forum der Vereinigung der Sternfreunde

Given the recent discussions on MIDAS and IRAF I thought I would introduce SPLAT-VO as I had not seen it discussed here.

While not for reducing spectra it has a number of excellent features for analysing spectra with a modern user interface.

While IRAF still has the advantage for debeleding lines I found SPLAT-VO much simpler for RV measurements through function fitting and the "mirror" method as well as for continuum removal and rectification where you can hand draw the continuum, followed by interpolation, or through polynomial fitting.

I have used it on Windows with an occasional problem with it not opening correctly but uninstalling and reinstalling fixed it for me.

The documentation is in English and not always clear (to me) but I have manged to do most tasks with a little trial and error.

The main page is here http://star-www.dur.ac.uk/~pdraper/spla ... at-vo.html and it is now being supported in Germany http://www.g-vo.org/pmwiki/About/SPLAT

Regards Andrew

Hi Andrew,
do you know the iraf commands rvidlines and rvreidlines?
Cheers
Christian

Yes thanks Christian, I have looked at them. However, I found doing continuum removal/ rectification much simpler in SPLAT-VO for my late type stars. Having done that it was simpler to continue with the tools in SPLAT-VO for Rv and EW.

Regards Andrew

Hello Andrew,
I normalized a spectrum of bet Cep with Splat-Vo and with Iraf. With splat I used a polynomial of degree 3, with Iraf a cubic spline of order 1 which means also a polynomial of degree 3. It is not surprising that the results are very similar. The advantage of Splat is clearly a better graphic interface. The advantage of Iraf is that I got the result automatically within 1 second. The only action was for me to confirm the result. If necessary you can follow with Iraf the same procedure as with Splat e. g. choosing relevant intervals and so on. But Splat let you choose only between polynomials of several degrees. With Iraf you can choose between legendre polynomials, chebychev polynomials, linear and cubic splines. For all this functions the order can be defined that means that you are able to divide the spectrum according the order in a given number of pieces where the spectrum is approximated separately. The function splot offers also a normalization procedure.
Cross correlation is possible with the FXCOR Task which I never used. splot offers five methods for determine the center of a line. With rvidlines and rvreidlines you can use several lines simultaneously to determine a rv. Splat allows to use only one line at a time. You get automatically a list with the error estimates. In my opinion the abilities for presentation are clearly better with Splat because of its more actual graphic interface. It is designed for analyzing large catalogs of astronomical objects. Anyway it is good to have an alternative and hopefully it will be get more features in the future. So it is worthwhile to watch the further development of Splat. Actually I see no advantage of Splat but the graphical abilities.
Cheers Christian

Good comparison Christian.

I will look at IRAF again as it has the advantage that you can automate the process. I have so far found it difficult with my spectra as there are only small regions of continuum.

I think that SPLAT does have some tools that I have yet to find in IRAF. They are both accessed from the Plot window:

1) Spectrum from interpolation - which allows you to hand draw and adjust a set of points to create a continuum or other spectrum.

2)Analysis -> Flip Compare - which allows you to mirror a spectrum and measure RV from the wings and/or core or a line without fitting.

My knowledge of IRAF is very limited so it may have these features as well built in or they could be replicated.

I think a strength of SPLAT and IRAF is that they are professional tools and not reliant on an individual for support.

Thanks for your continued support.

Regards Andrew

Hi Andrew set function = spline3 and order = 1. When you call continuum you will see a suggested curve and many data points which determines the curve. In many cases this is correct and you can confirm it by pressing q (=quit). When it is not satisfying you should increase the variable order by one and try again. With 's' you can indicate the regions which should determine the continuum. You can also delete data points which should not be relevant for the continuum. Look at the help file. With splot a continuum can also be determined. see also the help file. You should be aware that there is no exact definition of a continuum especially for late stars.
See above
I have no experience with cross correlation. I think you can determine by cross correlation only one line at a time. Please correct me when I am wrong in this point. This method is apparently very popular among the professionals but it is only one tool for measuring wavelenght. As long as the lines are symmetric you can use Gauss or Lorentz fits. I prefer the centroid method by using splot which determines an upright line through the center of gravity. It works also for asymmetrical lines. With splot you can deblend and determine mean, RMS,S/N, the line center, continuum at the region center, core intensity, flux above or below the continuum, and the equivalent width. splot is the working horse of iraf. All your activities are recorded in the splotlog file. The help file is very detailed. It takes time to read the stuff but it pays. It can do nearly all measurements but cross correlation.
I agree. Iraf has the advantage that it is unparalleled documented. SPLAT has the better graphic abilities because it programmed more recently.

This paper covers the VO tools for spectral analysis including SPLAT it makes some reference to IRAF and MIDAS - https://arxiv.org/pdf/1112.2787.pdf

Happy new year

Regards Andrew

Hi,
this paper gives a good overview over some popular software packages. However I doubt the statement that Hermitean polynomials are appropriate for approximation of functions defined on finite intervals. I can't see any reason why this should be.
Happy new year
Christian

Christian, It stems from a paper referenced here http://adsabs.harvard.edu/full/1982PDAO...16...67H but I could not find the original paper by Tsipouras and Cormier only an abstarct. However, it seems it was judged to be good subjectively based on what a skilled draftsman would draw freehand.

When I have tried it it did seem to do this and the examples in the attached paper seem "good".

Regards Andrew

Andrew,
the mathematical literature, to knowledge, shows nothing which support the use of Hermitean polynomials. Bronstein, Semendjajew, Taschenbuch der Mathematik, which also translated into the Englisch language (Handbook of Mathematics) recommends the use of Legendre or Tschebyscheff (German transscription, english Chebycheff) polynomials for finite Intervalls. Both systems are orthogonal systems for the Hilbert space of functions on <-1,1>. Both systems use different metrics. I am not completly sure because I am not a specialist for approximation theory.
When you find some literature about this subject please let me know. I appreciate when you would send me one example of a spectrum of a late type star which you regard as difficult to normalize.
Christian

Hi Christian,

Your knowledge of this area is well beyond mine so I appreciate your interest. I attach an example from Rho Cas it is from my MRes echelle and has been processed in IRAF with Blaze removal using a tungsten flat.


I look forward to your results.

Regards Andrew

Hi Christian, I found this paper http://www.sciencedirect.com/science/ar ... 2700003678 and these lecture notes http://www.math.univ-toulouse.fr/~calvi ... rs/MPI.pdf on Hermite interpolation.

Google gives many hits on this area with Lagrange and Hermite interpolation by algebraic polynomials being discussed together. It is clearly a big topic in numerical methods. I don't have the mathematical background to follow the papers and notes but I hope they mean more to you.

Regards Andrew

Hi Andrew,
many thanks for the paper. I will have a closer look on them. Here are the results of normalization. Both were generated with the iraf function continuum and cubic splines. n1Rho_Cas_... was normalized with order=1, n2Rho_Cas_... with order=2. order=1 means that one cubic polynomial covers the full range. order = 2 means that the range was split into two pieces. Each piece was covered by one cubic polynomial. At the border between the pieces both polynomials have the same y and y' (derivative). It is a matter of opinion which one is preferable. I would choose n1.
Please let me know your opinion about the results.
Regards Christian

Hi Christian,

I attach a comparison of the two rectified spectra you produced and one I had done in SPLAT-VO. They differ to a degree but it is impossible to say which is best.

I would be pleased if you can tell me all the parameters you had set in "IRAF continuum" so I can try to replicate your results.

In the end I think consistency is important to ensure good EW measurements.

Once I have the continuum parameters from you I will process all my existing spectra to test how robust it is compared to my splat proceedure.

Regards Andrew

Hi Andrew,
for n1 I took the default values ( function = spline3 !), for n2 I set order = 2 and pressed the f-key to trigger a new approximation.
Which is yours?
I had a closer look into several papers and I see no need to change my opinion. By the way interpolation and approximation are not the same. An approximated curve has not to meet the sample points as opposed to an interpolated curve. This means that the set of approximations includes more curves than the set of interpolations. In the paper you sent me the name of Hermite is mentioned but not in the context of hermitian polynomials.
Regards Christian

Sorry in order from the top n1 ,n2 and third in the list mine (Ratio: ...)

Regards Andrew

Hi Christian,

I looked at the ELODIE archive and plotted the region we have been looking at rectifying.
Then using this as a guide I used the I used the SPLAT interpolation tool to generate a new rectification (black) and compared it with your n1 (red).

I used a set of points which I estimated from the ELODIE to be the continuum and then the hermite curve type to interpolate a continuum.
As expected this gives a similar result to the ELODIE spectrum and has a flatter ceiing than our previous attempts.
Regards Andrew