spectro-aras.com

I have found the wavelength calibration procedure in ISIS using the LISA neon spectrum plus an A-type star spectrum to give rather erratic results. Sometimes it gives a good fit with rms <0.2, at other times the rms is >0.5 with no obvious reason. I have got results at both extremes on the same night under similar conditions using different A stars, both at reasonably high altitudes. I am careful with focusing to try to get the lines are sharp and clean as possible. The pixel size of the camera is set correctly.

To try to understand the sensitivity of the procedure, I reran the calibration for the same star varying the RV value assigned to the star to see what effect it had on the rms fit residual. The results are listed below. The best fit is for an RV value between -90 and -120 whereas the correct RV value for that star is actually -30. Does this point to a problem in the way I am doing the calibration, or in the calibration procedure itself, or something else?

Any suggestions on what might be happening here, or how to make the procedure more stable and reliable?

Thanks,
David

RV=0
point #1 x = 152.869 lambda = 3969.793 dlambda = 0.287
point #2 x = 189.929 lambda = 4101.998 dlambda = -0.248
point #3 x = 256.703 lambda = 4340.804 dlambda = -0.324
point #4 x = 401.590 lambda = 4860.816 dlambda = 0.524
point #5 x = 551.581 lambda = 5400.736 dlambda = -0.176
point #6 x = 702.433 lambda = 5945.058 dlambda = -0.228
point #7 x = 791.257 lambda = 6266.475 dlambda = 0.015
point #8 x = 857.378 lambda = 6506.435 dlambda = 0.095
point #9 x = 882.789 lambda = 6598.860 dlambda = 0.090
point #10 x = 915.196 lambda = 6716.927 dlambda = 0.113
point #11 x = 973.397 lambda = 6929.616 dlambda = -0.146
--------------------------------------------------------------------------
RMS : 0.331224 (in angstroms)
--------------------------------------------------------------------------

RV=-30
point #1 x = 152.869 lambda = 3969.441 dlambda = 0.242
point #2 x = 189.929 lambda = 4101.554 dlambda = -0.214
point #3 x = 256.703 lambda = 4340.301 dlambda = -0.255
point #4 x = 401.590 lambda = 4860.461 dlambda = 0.392
point #5 x = 551.581 lambda = 5400.630 dlambda = -0.070
point #6 x = 702.433 lambda = 5945.080 dlambda = -0.250
point #7 x = 791.257 lambda = 6266.497 dlambda = -0.007
point #8 x = 857.378 lambda = 6506.438 dlambda = 0.092
point #9 x = 882.789 lambda = 6598.856 dlambda = 0.094
point #10 x = 915.196 lambda = 6716.918 dlambda = 0.122
point #11 x = 973.397 lambda = 6929.616 dlambda = -0.146
--------------------------------------------------------------------------
RMS : 0.272064 (in angstroms)
--------------------------------------------------------------------------

RV=-60
point #1 x = 152.869 lambda = 3969.089 dlambda = 0.197
point #2 x = 189.929 lambda = 4101.109 dlambda = -0.180
point #3 x = 256.703 lambda = 4339.797 dlambda = -0.185
point #4 x = 401.590 lambda = 4860.107 dlambda = 0.260
point #5 x = 551.581 lambda = 5400.524 dlambda = 0.036
point #6 x = 702.433 lambda = 5945.101 dlambda = -0.271
point #7 x = 791.257 lambda = 6266.519 dlambda = -0.029
point #8 x = 857.378 lambda = 6506.441 dlambda = 0.089
point #9 x = 882.789 lambda = 6598.852 dlambda = 0.098
point #10 x = 915.196 lambda = 6716.908 dlambda = 0.132
point #11 x = 973.397 lambda = 6929.617 dlambda = -0.147
--------------------------------------------------------------------------
RMS : 0.225626 (in angstroms)
--------------------------------------------------------------------------


RV=-90
point #1 x = 152.869 lambda = 3968.736 dlambda = 0.152
point #2 x = 189.929 lambda = 4100.665 dlambda = -0.146
point #3 x = 256.703 lambda = 4339.293 dlambda = -0.116
point #4 x = 401.590 lambda = 4859.752 dlambda = 0.129
point #5 x = 551.581 lambda = 5400.417 dlambda = 0.143
point #6 x = 702.433 lambda = 5945.122 dlambda = -0.292
point #7 x = 791.257 lambda = 6266.541 dlambda = -0.051
point #8 x = 857.378 lambda = 6506.444 dlambda = 0.086
point #9 x = 882.789 lambda = 6598.848 dlambda = 0.102
point #10 x = 915.196 lambda = 6716.899 dlambda = 0.141
point #11 x = 973.397 lambda = 6929.617 dlambda = -0.147
--------------------------------------------------------------------------
RMS : 0.200933 (in angstroms)
--------------------------------------------------------------------------

RV=-120
point #1 x = 152.869 lambda = 3968.384 dlambda = 0.107
point #2 x = 189.929 lambda = 4100.220 dlambda = -0.112
point #3 x = 256.703 lambda = 4338.789 dlambda = -0.046
point #4 x = 401.590 lambda = 4859.397 dlambda = -0.003
point #5 x = 551.581 lambda = 5400.311 dlambda = 0.249
point #6 x = 702.433 lambda = 5945.144 dlambda = -0.314
point #7 x = 791.257 lambda = 6266.563 dlambda = -0.073
point #8 x = 857.378 lambda = 6506.447 dlambda = 0.083
point #9 x = 882.789 lambda = 6598.844 dlambda = 0.106
point #10 x = 915.196 lambda = 6716.889 dlambda = 0.151
point #11 x = 973.397 lambda = 6929.618 dlambda = -0.148
--------------------------------------------------------------------------
RMS : 0.205960 (in angstroms)
--------------------------------------------------------------------------

RV=-150
point #1 x = 152.869 lambda = 3968.031 dlambda = 0.062
point #2 x = 189.929 lambda = 4099.775 dlambda = -0.078
point #3 x = 256.703 lambda = 4338.285 dlambda = 0.023
point #4 x = 401.590 lambda = 4859.043 dlambda = -0.135
point #5 x = 551.581 lambda = 5400.205 dlambda = 0.355
point #6 x = 702.433 lambda = 5945.165 dlambda = -0.335
point #7 x = 791.257 lambda = 6266.585 dlambda = -0.095
point #8 x = 857.378 lambda = 6506.450 dlambda = 0.080
point #9 x = 882.789 lambda = 6598.840 dlambda = 0.110
point #10 x = 915.196 lambda = 6716.879 dlambda = 0.161
point #11 x = 973.397 lambda = 6929.618 dlambda = -0.148
--------------------------------------------------------------------------
RMS : 0.238838 (in angstroms)
--------------------------------------------------------------------------

Hi David,

I do not have any direct experience with the LISA calibration lamp but I know from experience with the LHIRES that offsets between the lamp lines and stellar lines are not uncommon due to the different optical path (if the neon is slightly offset from vertically above the slit, this can move the line for example or the line shape of one or other of lamp or star lines may be distorted asymetrically by the spectrograph optics, particularly at the edges of the field and this effect might be different for lamp and star lines. This sort of effect might explain why the RV offset of the A star for best fit is not as published. You could check this for example by calibrating just using the neon lines and checking the wavelength of the star H alpha line (which is bracketed by strong neon lines.)

Cheers
Robin

I'm likely missing something here - but the residuals you're showing, something like .2 Angstroms, given your dispersion, which is around 3.6 Angstroms per pixel, is around 0.05 pixels. You don't mention what the S/N is per pixel for the measured lines, but overall I'd think consistently measuring line positions to around 1/20 pixels is pretty good. Am I missing something?

Mike

Hi
Forgive the novice question but what`s the RV value that David refers to please?
thanks
Steve

Radial Velocity
Cheers,
François

Robin,
Thanks for the suggestion. I tried calibrating with just the neon lamp in ISIS by specifying the Predefined mode with the LISA (internal neon lamp) option and then asking it to process an A star spectral image. From the running report as ISIS processes it appears to start with a primary wavelength fit using Balmer lines, presumably from the A star spectrum, and then does a fit to what appear to be neon lines - so I'm not sure how good a test this is. The wavelength of the H-alpha line in the calibrated spectrum was 6561.5A, not very good.

Mike,
I realise this example looks like a good fit, as indeed it is. This was for a bright A star with good S/N. My problem is that I get a spectrum like this which calibrates well but then another A star spectrum with similarly good S/N taken an hour later has an rms 3x larger for no obvious reason. It feels as thought there is an instability about the calibration process that I don't understand yet. The reason I included the details of this example was to show that, although the fit rms looks good, the best fit (smallest rms) occurs at the wrong RV value for the A star. Again I get the feeling something is not right. When I showed these results to Olivier Thizy at the show in Warwick a few weeks ago he suggested I post this to the list. I would like to be able to make wavelength calibration a more predictable and reliable process so all suggestions are welcome.

Thanks, David

This is an interesting subject, Mike.
But, the solution should be that Shelyak produce a *real* calibration lamp ... The question has been raised, but no result todate.
Cheers,
Francois

Hi David,

David Boyd wrote:Robin,
Thanks for the suggestion. I tried calibrating with just the neon lamp in ISIS by specifying the Predefined mode with the LISA (internal neon lamp) option and then asking it to process an A star spectral image. From the running report as ISIS processes it appears to start with a primary wavelength fit using Balmer lines, presumably from the A star spectrum, and then does a fit to what appear to be neon lines - so I'm not sure how good a test this is. The wavelength of the H-alpha line in the calibrated spectrum was 6561.5A, not very good.

Actually I think this explains RV discrepancy quite well. The ~ -1.5 A error in the H alpha line equates to -70 km/s RV which is similar to the correction in RV you are having to apply to the Balmer lines get a good fit when mixing neon and star lines. I dont know how close the LISA neon lines normally are to the correct absolute value but misalignment of the neon in the LHIRES can certainly produce shifts of equivalent magnitude (a pixel or so) Perhaps this sort of error is close to the limit of absolute calibration accuracy of the LISA ? High absolute wavelength accuracy is tough to achieve but as long as it is repeatable though it does not affect the precision.

Mike makes a good point about the RMS value of your good fit being only 1/20 of a pixel. Even your "bad" fit is then only ~1/7 pixel. With the LHIRES I dont tend to worry too much as long as the residuals are <1/5 pixel so this might be as good as it gets. (It does not take much noise/distortion of the line shape etc to introduce a 1/5 pixel shift) How much is it affecting the fit? eg if you overlay your "good" and "bad" fit star spectra can you detect any calibration errors ?

Cheers
Robin

Hi Robin,

Comparing the "good" fits (low rms in wavelength calibration) with the "bad" ones, it seems to be the H-alpha line which moves the most with the line blue-shifted more for the worst fits. The other H Balmer lines tend to match better in all cases. Possibly this does indicate a mismatch between the neon and Balmer line components of the calibration within the usual LISA procedure?

Maybe I am being unnecessarily concerned about this. I accept these are small errors in an absolute sense. It was the inconsistency which bothered me more, but that could just be statistical variation within what is an acceptable distribution of calibration errors. As you say, maybe this is as good as it gets.

Thanks again for the feedback,
David

David Boyd wrote:Robin,
Thanks for the suggestion. I tried calibrating with just the neon lamp in ISIS by specifying the Predefined mode with the LISA (internal neon lamp) option and then asking it to process an A star spectral image. From the running report as ISIS processes it appears to start with a primary wavelength fit using Balmer lines, presumably from the A star spectrum, and then does a fit to what appear to be neon lines - so I'm not sure how good a test this is. The wavelength of the H-alpha line in the calibrated spectrum was 6561.5A, not very good.

Mike,
I realise this example looks like a good fit, as indeed it is. This was for a bright A star with good S/N. My problem is that I get a spectrum like this which calibrates well but then another A star spectrum with similarly good S/N taken an hour later has an rms 3x larger for no obvious reason. It feels as thought there is an instability about the calibration process that I don't understand yet. The reason I included the details of this example was to show that, although the fit rms looks good, the best fit (smallest rms) occurs at the wrong RV value for the A star. Again I get the feeling something is not right. When I showed these results to Olivier Thizy at the show in Warwick a few weeks ago he suggested I post this to the list. I would like to be able to make wavelength calibration a more predictable and reliable process so all suggestions are welcome.

Thanks, David

David
I found some variation in the calibration lies using the internal neon light with my LISA also.
The screws holding the calibration unit in the spectrograph had worked themselves a bit loose. I think this may have led to a small change in the position of the neon light above the slit and this changed the position of the lines.
I currently don't rely on the internal neon light but use an external filly lamp in front of the scope. I take a 30 sec exposure either constantly moving the lamp around the front of the scope or holding it still aligned with the slit as described by Christian here http://www.astrosurf.com/buil/isis/He_c ... method.htm
This method seems to give me a more consistent result.

Cheers

Terry