Don't we have an ocean optics spectrum to source from too? I don't remember if we took a cfl spectrum with it, but we could maybe ask Mary to take one... Or one of the folks in the Oil test kit beta program, as a couple have access to lab specs.

Chris,

Have you ever looked at "The Elements Spectra"; it gives you numerical values of the peaks, you can get it from: www.infini-t.com or email : custserv@infini-t.com; or just enter The Elements Spectra into Google. I have glanced at Terbium and there seems to be a lot of peaks noted.

Gwill

Thanks, Gwill -- good resource.

Chris, I wanted to get some more clarity from @gretchengehrke on the mercury oxidation states in the comments of my post. Hopefully that'll answer some questions, but it seems like your suspected 8,000,000-intensity line is listed as Hg II. I also wonder if there could be a blend of different oxidation states... I really don't know the chemistry.

Hi Gwill, there do seem to be a lot of terbium peaks. I have not yet figured out how to know which of those peaks are expected in the light from a CFL. Same with Hg peaks which are also numerous. Here is a figure which labels a terbium peak at 542 nm:

Jeff, A fluorescent bulb works because the electric current turns the gas in the bulb into a plasma which is by definition a bunch of highly ionized atoms. So I guess we should expect lots of different ions of all the components of the gas (typically argon and mercury). So lots of Hg emission lines are possible, but only a few are prominent in CFL spectra. The fact that it is difficult for us to figure out which of those dozens of peaks to expect suggests that something sort of complex is happening. It does not help that both Hg and Tb have emission peaks at very close to 436 nm and 542 nm. In fact that is sort of unfair (but maybe the basis for a good exam question in freshman chemistry).

I don't know of a Hg spectra from Mary and John's spectrometer. I missed the whole spectro session at the last LEAFFEST. Too bad because there are lots of things it would have been good to get spectra of. You would think that good spectra of CFLs would be common.

If it's helpful, we can probably ask some of the OTK beta participants for one, since a couple have access to lab spectrometers.

In your image/image comparison, I'm curious -- since we calibrate based on the 436nm blue line, what does that comparison look like if 546 and 436 are aligned, instead of the 405nm blue line? I'm interested in if the distortion in our spectra is accentuated towards the wavelength extremes, and if our match is better for the central lines.

This is tough. I think perhaps the best we can do is to try to get a dataset from a calibrated lab spectrometer and treat that as our reference, even if we don't know the identities of all the lines. Does that make sense?

That's a very important observation. Had I originally calibrated this spectrum on the 436 line and the 611 line (instead of 405 and 650), all the other lines would have matched much better. I think it might be a mistake to use the outer lines for calibration because they might have more distortion for some reason (maybe the effect stoft was describing). Nevertheless, the polynomial was able to correct the lines well, so the final spectrum is probably better than any uncorrected one.

Would you be willing to try a recalibration using the 436/546 lines as a starting point? Would that be useful as a point of comparison?

Also - I believe I've gotten a root mean square error system running but its results seem inconsistent to me. I can show you at LEAFFEST, but maybe some tweaking will help -- it's possible I've made a mistake. It's not notably better than the old "fitness" system.

If the original Snowy Sky CFL is calibrated by pinning down the 436 nm and 546 nm peaks (using a macro) it looks like this:

Most of the peaks do not match with their known wavelengths. If the 436 nmm and 611 nm peaks are used it looks like this:

Those two pinned peaks are farther apart and should allow for better calibration. I think on average the match with known peaks is closer, but only a couple of peaks other than the pinned ones match within a nm. So a non-linear correction like the third order polynomial seems to be required to make the major peaks of this spectrum line up with reality:

Hmm. I guess there are pros and cons to each approach.

I agree that the 546 line doesn't result in a great linear calibration -- some end up about 5 nm off. For example, the OTK test calibration I used in July has the red line appearing at 613.14nm instead of 611.60nm. That's ~1.5nm off.

But it's hard to say -- i cloned and recalibrated it with the new interface, and got 436.32nm, 542.7nm, and 609.65nm -- manually placing the lines using the image reference as a guide. That's approximately 0.5nm, 3.3nm, and 2nm off, respectively. That's not too bad, although it may get worse towards the extremes.

The cons of tackling polynomial calibration right now are that:

1. I desperately need to finish everything else on this list AND this list (58 remaining issues) and am already past the planned deadline,
2. the images won't line up anymore, since we won't warp the image, and
3. to do it right, we may need to warp the images as the user drags, or re-design the calibration system to use only the graph

I think that if we quantify the error using root-mean-square-error assessment, we should stick to 436/546 linear calibration for the time being. Now that it's possible to do calibrations completely in JavaScript, I should finish the API and documentation so that it's easier for anyone (even if it is just me, later) to add a new calibration module on top of the existing infrastructure.

What's more, the RMSE assessment plus the visual image-comparison reference in the new interface allow people to make different decisions -- not just using 436/546, but balancing the errors out along the span of the whole dataset visually. How's that sound for now?

Oops, meant to share links to my tests -- here's the one I used the new calibration system on, and listed 3 approximate errors for: https://spectralworkbench.org/spectrums/show2/58942

Hmm. So I'm wondering if part of the fit problem is because we're simply comparing intensity at the 14 points and 13 troughs. Take this example:

Though I equalize the max heights of the two before running RMSE, I bet the baseline variation makes the match not great. I could try not measuring troughs? But that doesn't seem right.

Or, we could "search around" each of the 14 expected peaks for where peaks actually occur, find their exact real locations, and add up the error for each peak. This could fail if not every peak is found, but we could narrow it to fewer, more prominent peaks.

I may simply have made some dumb mistakes in implementing RMSE, but if the approach has more fundamental flaws, maybe this is an alternative.

@warren , @cfastie I see i'm kinda late to the party, i can offer source verification (you can ship CFL source to me) with 3rd party device, Ocean optics and LR1 from aseq http://www.aseq-instruments.com/LR1.html i have access to few devices... Sorry for not joining barnraising this year 8-/

Oh thanks! I believe, but Chris may correct me, that it'd be sufficient to measure a cold long tube fluorescent lamp and compare it to pretty much any warm CFL. That'd be interesting to see!

Hi Robert,
In this note and comments we were lamenting the lack of replicate spectra of CFL lamps to confirm the information at the Wikipedia page. It's still startling that there aren't plenty of reliable spectra of CFLs online, but I have not found any to compare to the Wikipedia spectrum. So it would be really helpful to have a spectrum and data file for a CFL from a well calibrated spectrometer (or two). If the near infrared can be included that would be helpful, so maybe 300 to 1000 nm. It would also be good to have spectra for different types of fluorescent lamps to see if there is a consistent difference among lamp types.

Thanks,
Chris

I will start with long CFL tubes that i just did... Unfortunately i don't have bare tube (diffuser in front/office lights) that might play same role in peak smear. this is full spectra

Zoom with readout on peak@544.6nm

Zoom with readout on peak@545.9

Oh, great -- do you have raw data for these as well, like CSVs?

Yes i do, here:

FC__long.txt

Oh and i suggest using QTI plot (you can zoom into graph data way better then excel, plot traces with just right click/plot) very nice open source tool. https://intranet.cells.es/Members/cpascual/docs/unofficial-qtiplot-packages-for-windows

Thanks @zega!

By comparing my spectra of the Sun (more lines for precise polinomial callibration) with that of the CFL lamp, I can confirm that the green line is almost certainly sodium 546.074. The other peak is at 541.68 and is probably Tb. Here is my spectrum:

And here is the data, if anyone is interested. cfl.txt (space separated) raw intensity: 1,3 calibrated intensity: 1,3/4*5

I didn't know sodium was used as a phosphor in CFL lamps. That green line is one of the brightest in CFL spectra and in all mercury spectra, and mercury is always present in CFLs.

Ups, I meant mercury, not sodium, of course. The line I can't identify is the one at 445.1 and 445.9nm. Any ideas? Also, are the two faint lines on each side of the mercuty/terbium green line it's 'satelites' or something else?