09:20:11 So, I would like to introduce our next tutorial.
09:20:15 Led by Marie Lau, from UC Riverside, giving a really useful I think tutorial on Quasar absorption line fitting, particularly for those of us who are theorists or computational people or even observers who operate in different realms of the electromagnetic
09:20:34 magnetic spectrum where there wouldn't necessarily be clean absorption line features to fit. So, and I encourage people to rather than asking questions in the zoom chat, doing so in the halo 21 tutorials chat, so that it's preserved, a bit longer than,
09:20:54 then, then the, the 40 minutes left in our zoom.
09:20:58 Okay, take it away Marie thanks.
09:21:02 Is there a tutorial channel on slack. Yes, hello 21 tutorials. Oh, I will look for it us thanks yeah hello I'm Marie, this tutorial was motivated by participants requests and a Slack channel Halo 21 analysis coax.
09:21:21 Thank you for joining. I'll begin with a brief review of The Voice profile. If you want a longer review a good reference would be Professor Mark chrome hoses graduate course material.
09:21:33 Then we'll talk about how to fit Quasar continuum with line Corps is a Python based software written by a professor expert Chaska and Professor Nicholas to host our elevator illustrated with that example.
09:21:48 And then I'll talk about fitting absorption in lines with Alice is a Python based software written by a DR yn Kook, Alice is his wife. Oh again illustrate with an example, or Asia Molly I thought about making a Jupiter not spoke about us, but I didn't
09:22:04 write the software as myself so it's better to to just illustrate what I did, which you may have to customize for your machine. At last hour hand over to Professor bird shots to demonstrate his Reaper, and Professor a bottle it demonstrates is our baby
09:22:22 fit.
09:22:24 Here's some review on The Voice profile.
09:22:27 Consider of serving a bright continuum stories, such as a quasar, the source produces a flux f new of zero, the light from this source passes through a cloud or foreground gas, the flux that's we observe is afternoon, have zero times exponential minus
09:22:57 This tower new is not negligible only over a narrow range in frequency outside this range we can directly measure avenue of zero by interpolating from one side of a line to the other, we can estimate avenue of zero in their frequency range where the line
09:23:06 is absorbed radiative transfer calculations gift the optical death, how new equals some constants times the oscillator strength of a given transition, times the column density of the species, through the cloud, times the line profile function.
09:23:22 The line profile function fight new is a voice profile. A voice profile is a complicated function that has a Gaussian core and Lauren see and dumping wings, if we focus on the court.
09:23:37 We can approximately five new by a pure Gaussian.
09:23:41 In a Gaussian form the be parameter is the Doppler broadening parameter, and is equal to square with two times the velocity dispersion.
09:23:51 So by modeling the absorption line profile we can derive this be value, the column density, and the line centers have served frequency.
09:24:02 As the optical depth increases, all the photons near the line center are absorbed the line center is sought to be saturated. And as the optical depth continues to increase the saturation region extends beyond the Doppler core of the line, and the damping
09:24:19 wings dominate.
09:24:21 For each one lyman alpha, the damping with Jim happens at 10 optical depth of 10 to the five, and that corresponds to a column density of a few times 10 to the 19 prescient squared.
09:24:37 What is called interpolating all for absorption lines and of course material is what observers called fitting a quasar continuum.
09:24:46 I will describe the line tours algorithm and the audience may correct me.
09:24:53 The spectrum is split into an arbitrary number of wavelength intervals about 10 rest Engstrom is wide, the flux media and is calculated in each wavelength interval.
09:25:05 The central wavelength and media flux of these intervals define a set of points, and a cubic spine is fitted through the points pixels that far more than 1.5 standard deviations below the spline or a mask, as outliers.
09:25:23 This process is iterative until no further pixels are removed as outliers, the fittest visually inspected and in regions where it affects up here is poor, the user can manually tweak it.
09:25:39 The spectrum can then be normalized by this unabsorbed continuum.
09:25:44 Here's a longer description of the line whose algorithm, I wrote this blurb for a co author art paper, I put it here for your reference in case you need to write a paper you're saying line tours, and I might going over this.
09:26:01 And I'm not going over this. But the slides will be preserved after the fact, so people can can read this at their leisure. Yes.
09:26:10 The example I use is a quasar expression of point nine, our users cost far you'll respect from from the Caspar survey. The data is stored in our fits file we're trading FQV underscore at CAST Bob outfits and the data includes the flux array, the era era,
09:26:28 and the wavelength array. The data is loaded into what's called an exit spectrum one the object in line toes. I named this expect from one the object SUVs back in this Jupiter notebook screenshots the first cell and a second cell are just loading the
09:26:46 data.
09:26:48 The third cell is where it continuum fitting method is called, and I give a gas to shift,
09:26:57 as I call it the fits continual method, a GUI window pops up. It shows that data and the automatic fits the gray curve is the flux array of the data versus that wavelength array, the curve is to continue them.
09:27:13 Direct circles are despite knots, the automatic fits is already quite good except for the go curl no emissions.
09:27:22 In my Jupiter and don't smoke a menu of all the interactive options are printed, and I just list I feel important options here.
09:27:31 They are interactively zoom in and out. Set plot limits.
09:27:38 Add a new panda plot window left and right, add a new spine that deletes the nearest not and move to a nearest not, I can zoom into different spiritual regions.
09:27:50 And, in fact, if the Fed is poor or somewhere, I can move a not by putting my cursor at where I want the knots to be when I'm satisfied with the fits, I type q2 equates.
09:28:02 That was safety continuum in the same extraction one the object that host that data.
09:28:10 data. So the continuum is saved into DFQV spark object, I can call it the rights matter to your rights to contents back to the file from rich data is loaded.
09:28:21 I verify that this file has a new extension added, and the extension is named continuum.
09:28:30 At this point I'd like to ask, are there any urgent questions.
09:28:37 Um, I can't see the chat window so I assume no other questions, I think.
09:28:46 So the continuum.
09:28:48 So we just covered how to fit a quasar continuum, this continuum is used to divide the flux spectrum. We can don't fit at option lines, and the Slack channel Halo 21 analysis coax, a few Python packages have been mentioned, they are Alice Reaper and RBV
09:29:09 fit.
09:29:10 I'm going to illustrate, Alice. The example to us. So the cost bar up spectrum of the same Quasar. I'm going to fit a few office associated absorption lines.
09:29:23 The word associated means the absorption happens within several thousand kilometers per second of the equations redshift, and they're often and physically associated with a quasar I selected three lines.
09:29:44 Oh 460804787, and oh five six to nine. As this Quasar is part of Caspar Sufi. The Associated absorption have already been modeled by Professor taught trip, using Reaper.
09:29:54 I'm going to reproduce the results using Alice.
09:29:58 And I know it starts the Alice package depends on line pause for reading the cost lines breath functions, the inputs to be fat into the software is named something that mod is an ASCII file in the file you specify the data and an initial gas for the model.
09:30:18 The first line says run atomic atomic underscore Amara dot XML atomic underscore Merida XML is a file that contains the atomic data to be read by Alice.
09:30:32 The default atomic data table that comes with Alice is named atomic thought XML, and I added atomic data have several more transitions next few lines are some settings for running the fitting process and displaying the results.
09:30:50 And the next block of lines are for specifying my data insider data way to block. I gift a path to my data. I only have one data file and is named FQB underscore norm ASCII, the contents of this Fub underscore non ASCII comes from that fits file that
09:31:11 I saved from the work with line tours, Alice can only read the spectrum in ASCII formats and the spectrum has to be normalized.
09:31:21 I also need to specify wavelength ranges for fitting.
09:31:27 I visually inspect to the spectrum and I see absorption of all 468 in the range, 116 6.5 and strums 2116 9.5 x Rome's.
09:31:39 I see absorption of all four seven x seven. In another wavelength range, and I see absorption of oh five six to nine.
09:31:49 In another wavelength range.
09:31:50 I also specify the spectral resolutions, which for costs are a given by some predefined line spread functions OSF.
09:32:01 The next block of lines and does something that Marty file as far as specifying and initial gas off the model inside a moderate blog I specify some parameters of the fights profile that should be fixed or limited.
09:32:18 Then I specify an emission model, which for a normalized spectrum is just a constant value of 1.0.
09:32:27 Then I specify and absorption model online profile functions are void. For each Ayana GIFs option components I give a gas to his column density is redshift and the Doppler be value.
09:32:42 For example, let's look at the first line below the line that says absorption.
09:32:47 It says the ion is all for was a mass number of 16, the gas to the column density in large scale is 14.3, the gas to the rushes is 0.9192, and a gas two is be value is five kilometers per second.
09:33:05 There are many numerical values and the smaller rate block. If I want to specify that a certain numerical value should be fixed in the modeling. I put some uppercase letters, after it.
09:33:19 For example, the emission should be a consequence of 1.0, so I put uppercase letters CNS offer it. They can be any uppercase letters.
09:33:30 You may also notice that I put some lowercase letters, after the rush of gases and the Doppler be gases.
09:33:39 If I wanted to tie a certain all four components to a certain old five components, I would have quit the same lowercase letters, after their gases.
09:33:50 And this example, the components of all four and all five. Do not trace each other. So I'm not tying them.
09:33:58 After I finished writing this something that might file into terminal problems, I run this command. Run, underscore, Alice, something that mode.
09:34:10 So I run, Alice, the software searches for the local minimum of the total chi square it between the model and the data that we saw each iteration are printed to the terminal.
09:34:22 The final fitting results are printed to a file named something that mark out the Fitbit model is printed in the same format as the initial guests model, but with numerical values updated.
09:34:35 Here I showed the output model copied from something that might come out the errors of the model, come after the model itself, the software also outputs a figure showing the fits to the data.
09:34:54 Next, I can compare my fitting results today truth set for most components that two sets of results are very similar. The only exception is all five component for.
09:35:02 This is a week components and is heavily blended with neighboring components. So there is some degeneracy. Overall, our call the results quite good.
09:35:12 Finally some tips on using Alice, the default atomic data table does not have some of the line site analyzed. So I augmented the atomic data table. Another tip is about the scenario where the F sober only partially covers the emitting stories.
09:35:29 There is no straightforward way to handle this partial coverage is not implemented in the master branch of the Alice repo, and get help to customize your way of handling partial coverage.
09:35:42 You can clone the cover factor branch and add it's the elephant on the skull voice stop high file.
09:35:48 And this is the end of my tutorial.
09:35:51 Thank you.
09:35:52 Great, thank you. Marie.
09:35:55 And so related to all of this.
09:35:58 That was a that was a great demonstration, and, and, and, honestly really good for for teaching this material to in, in courses graduate level courses and such, but to supplement this we are going to have.
09:36:13 Joe Birgit and Ron longboard Lloyd, give a couple slides, describing different programs. So, Joe Do you want to take over next.
09:36:29 Yes.
09:36:32 So grab the desktop here.
09:36:38 Okay. Um, so yeah I, along with the excellent tool that that really just demonstrated.
09:36:47 She asked me to and come up in the, in the appropriate Slack channel to discuss this piece of software that I started back in grad school, um, and, you know it's it's one of these classic.
09:37:04 I know I shouldn't reinvent the wheel, but, you know, now the wheels that had at my disposal. Did I feel really was meet all of the kind of needs that I needed in terms of of analyzing the data and in an expedited way so.
09:37:22 So I came up with what I call the beeper.
09:37:26 So this is a void profile fitting code, formerly called Joe bvp, the GitHub repository is here, and kind of what motivated me to, to write this, the software was that I needed to fit, lots of lines over, you know, multiple red shifts and do so.
09:37:50 Simultaneously, over a broad range of wavelengths so you know from the bluish part of the spectrum to to the red end.
09:37:57 And to do so in an efficient manner to get the line set parameters to prominently use the time that was one I think that was originally originally written by by Daniel lt who passed down through job hiring.
09:38:13 And so, I was I went ahead and plugged it into the PI GM package which includes a ton of stuff. And one of the really helpful things there is a line identification software and an accompanying workflow that was developed by Nicholas t hosts.
09:38:34 We've also added the ability to run fits in batch mode, so to speak. And so this is, this is really nice dovetailing with with IgM guesses because you have a workflow where you can identify all the lines in the spectrum extract aligned list and then run
09:38:52 the fits of all the lines that you've identified sort of in a non interactive way.
09:39:06 But I wanted to also have a means of of interactively fitting. And so I developed a gooey. And then the software plugs into line tools, which is a storehouse for line spread functions atomic data.
09:39:18 All kinds of stuff. A number of analysis script analysis tools, in and of its own. So, A lot of people have contributed that it's no longer job PvP because it's it's now.
09:39:31 The product of several people's efforts, some of whom I list here.
09:39:41 But I'm just moving to thinking about the, the IO for the software, so it's it's really driven by line lists that include.
09:40:02 Just some, some basic attributes, the name of the spectrum file that you're fitting rest frame wavelength of whatever line, it's redshift, and then some starting values for comments at be value, velocity centroid.
09:40:05 And then using some flags for all of those parameters, you can tie certain parameters to one another, such as for two transitions of the same species.
09:40:16 I'm an alpha I'm a beta here, or just allow them to very freely.
09:40:21 Okay, so
09:40:25 see this in action. So I have a spectrum that, to which I fitted a continuum using the line tools function that that Murray just just nicely demonstrated.
09:40:37 And so, I will load in my continuum for the spectrum. Along with this, my list.
09:40:49 And this is the graphical user interface. There are eight lines that are in the line list a couple of Milky Way silicon four lines, and a couple of components three components of each one detected in both live in alpha in live in beta.
09:41:05 And so the key here
09:41:10 is that all these lines have to be fitted at the same time why, because the silicon four lines are blended with the h1 lines, and we also have the Lyman beta minds here that are that are a little weaker.
09:41:25 So, I've looked at my parameter file. And with that I can go ahead and run the fit.
09:41:37 And so the resulting fit is seeing down here
09:41:43 can see, we nicely account for the blended component structure and the two components three components of each one and the blended Milky Way lines is common entity be value, the errors on all of these quantities appear in the.
09:42:00 In this window.
09:42:01 So, you can then add in a new bring in a new line list. So I have several that have created here.
09:42:11 Let's go ahead and get a different set of lines.
09:42:18 Can we get a nice quick sit for two components of each one, and both lyman alpha and beta.
09:42:26 It's also as a bad fit mode as I mentioned, so we're here I've created a have to line list group three and group for.
09:42:35 So I've just dumped those into a text file, and I can run.
09:42:40 The so called batch fit mode,
09:42:44 where it will just go through each of those mindless fit the lines.
09:42:51 And at the end, It will extract produce a nice
09:43:01 file for your inspection here this is the compiled fit inspection file for all of the lines that are identified in the spectrum, so I just.
09:43:11 We went through IgM guesses identified all the lines extracted the line lists and then ran the fit and batch mode, non interactively, so you can see it does pretty good job.
09:43:21 And then you can go back and tweak any parameters as necessary to get to get a better fit if it.
09:43:28 If it mangoes some kind of be value or something, you don't get such a great fit. So that's the tool. It's available on GitHub. And feel free to contact me if you want any help setting up.
09:43:45 Terrific, thank you Joe. Yeah, we'll have to write down links for all the various tools for doing this from both Murray, as well as Joe and and now, wrong bonds demonstration online.
09:43:58 Would you like to share screen and demonstrate yet another method for doing all this, this business. Sure. All right, thank you so much.
09:44:09 So again, I guess it's for every question is kind of a rite of passage to write their own point profile fitter, that's how I got started I wrote my own thing.
09:44:18 And I know only astronomer, in, in this field who I know used to use MATLAB, so I had to write everything from scratch because nobody else was easy. So anyhow it evolved to a Python based software right now.
09:44:32 And my goal here was to really use a simplified user friendly mC mC for profile fitter and the heart of the code is really a class which arbitrarily allows you to generate a bunch of for profile for any redshift.
09:44:51 The code is available in GitHub.
09:44:54 It is pretty much using other code, obviously, for example, to fit.
09:45:02 What have to do nm CMC fit I use the MC code package, as well as to the corner ports I use a cornerback etc with your standard. I use line tools, just to grab, let's say, a fits file, or to grab the call center serve because I was too lazy to direct mail
09:45:19 to court for that.
09:45:21 Otherwise, one thing I wanted to point out the beautiful example that Maria and Joel show are with amazing data. Those are the kinds of spectrum cream of heavy, but reality is not that good.
09:45:36 Humble doesn't always get HTC 10 doesn't always give you so much time. So you might end up with a spectrum, which looks like this.
09:45:44 You know, so it's a little bit much noisier than the beautiful spectrum that they were showing we can always come in.
09:45:55 And you can see that signal noise is not that great. However, there's a lot of absorption there. And when you fit, such noisy data with a standard square feeding routine, you might be stuck in a local minimum.
09:46:03 That was and plus there'll be other like contaminations there that's why I wanted to try to fit it using a patient method.
09:46:12 And so here what in this demo I'm showing you an actual example of a prospector which is pretty noisy. And what I'll try to show you is fit a series of linemen alpha 11 billion I haven't gotten the lines here simultaneously.
09:46:26 This basically the there too. I bet the notebook files in the repository, which has these examples so you should be able to in principle run it, without any issues, as long as you have the dependence installed.
09:46:42 So I wanted to show you a little bit of pre processing.
09:46:46 Simply I am taking the spectra and using a non by way of statement to slicing the relevant pieces of the spectrum wherever to perform the effect. That's it.
09:46:56 So, I am choosing to fit at a redshift of redshift point four six where there happens to be an absorber. And I want to fit three lines so I'm slicing, they're relevant way linkery you can slice and dice Hi everyone, this is really, you can substitute
09:47:12 this part with your own personal input file where you have a very well process file. This is kind of the important part here. I have a simple gooey where, which allows you to kind of guess what are your input parameters.
09:47:31 For example I want to fit a each one calm density profile to this absorbs consumption trough, so I can simply say that what is the redshift, what are the relevant Richards that I'm interested in, and call this GUI.
09:47:46 And then I can let's say I guess there is an absorber here. so if a cloud here, I guess there's another club here, I guess there's another field here.
09:47:54 So it allows you to interactively identify where the terms are, and then I can also guess out for this clump here what is the Doppler be parameter guess let's see I'm seeing any quantum specific, why not.
09:48:08 For the second clump, I want to say it's 30. The third clump, I want to say it's again 30, because why not, then some bookkeeping is done so that it can be fed in, and this is really the heart of the program.
09:48:21 This will effectively generate any arbitrary board profile you one for any set of redshift. If you have, let's say, interlocking absorbers You can also add here, if you have some nuisance parameters you can add it here.
09:48:37 I, in this case I have chosen to have a garden forward half max Gaussian, as a convolution parameter which is 6.5 pixels. You can also change it.
09:48:48 It just with HST cost setting to grab the corresponding line split function.
09:48:53 So, if I just plot here it will.
09:48:57 Oops, I did not execute this.
09:49:03 If I just want here you'll see in a big pot where I'm showing the tree little slices of spectrum that I'm trying to fit here.
09:49:11 And these are the initial guesses that I'm using. And the second part, again you can substitute this part with your favorite filter, but all I'm doing is I'm taking a model and running it to fit in an mC mC manner.
09:49:25 I'm not going to run it because it takes like 30 seconds but ignore the errors, there are always those coming up.
09:49:32 It does the fit and you're pretty much done, then I can show all affiliate parameter in a nice corner plot, and it's kind of important to show that I have fitted a rotter noisy spectrum, and I have three columns density estimates, and you can sort of
09:49:54 you have multimodal austere approach to distribution for your best guest, velocity etc etc. So that's it you're fitting is done. And if I go ahead and plot.
09:49:59 Again, the code does it for you. All I have to do is take your posterior distributions block the output and you can immediately see you know, three parameters are fitted with a best fit login, which I asymmetric error bars.
09:50:12 be parameter asymmetric error bar, and also the velocity central it's multiple iterations are being shown here which shows you know what is your uncertainty in filling, let's say a noisy parameter like this.
09:50:23 So I believe a mC mC method captures your uncertainties a bit more accurately, specially if you have blending and you have to fit. Let's say that if a another object as a different restaurant, etc.
09:50:37 And then you can go. Have your output Finally, execute this cell, and it will save it in a pipe in a Python dictionary format, but you can change it to whatever you want.
09:50:47 So this is the core of the program.
09:50:51 There is also another example, given where instead of doing an interactive fit and just by hand defining what are the initial guesses that I'm using to fit the profile, everything else remains to see.
09:51:04 So, essentially this program was inspired, and I had to write it in Jupiter notebook because most of my students prefer to work in Jupiter notebook and they're not very well versed in terminals.
09:51:16 But yeah, it's pretty user friendly and if you decide to use it, please let me know.
09:51:23 Awesome. Very cool.
09:51:26 So now, everyone has tools for being able to absorption line fitting in a variety of different ways. Yeah, I will. Kind of get all these relevant links and put them in the tutorials channel.
09:51:40 So, for, for future use. But of course this is that the whole tutorials being recorded so we'll have that indefinitely.
09:51:57 Well, excellent work. Thank you very much Murray, Joe and Ramon, that those were really eye opening in terms of this new this this methodology.
09:52:00 I suppose we don't have to move to a breakout room.
09:52:04 Since this is the final one. So if there are any questions for any three of our, our speakers on that last tutorial, I encourage people to ask them. Now if they wish.
09:52:17 So I'm a little bit interested in it perhaps it's a too much of a theoretical question but the.
09:52:27 The Laurentian profile is derived from a completely classical model of a damned romantic.
09:52:34 Damn, come on Acosta later with a charge on it driven by any field.
09:52:39 And so in dreams book he talks about like the obvious point that it that the profile should be in principle be the drive from the commerce Heisenberg formula.
09:52:51 Do any of you guys that have experienced with this kind of fitting have a sense of what kind of errors, there are between the currency and file in the commerce Heisenberg profile, I mean I presume it's small, but it's,
09:53:07 I think in almost all cases we're talking about that it's super super small like that's not something that's actually dominating the profile in fact by and large the the Gaussian core is dominating in almost all cases unless you're dealing with like a
09:53:23 DLA or you're measuring, you know, trying to measure like deuterium or something like that, then you might have to worry about that but, um, I would say for like most applications.
09:53:32 This isn't something that is that's certainly not what's dominating the uncertainty I think that the point of like you know they're being uncertainty in the line identifications themselves, or how much like blends are occurring that's not captured by
09:53:48 the actual errors and so you really do have to worry about that, I would say, much, much more.
09:53:57 And I guess my final question was, like, you want to leave pointed out when when ping showed some of his absorption spectra. Like many many days ago. Some of these cemeteries that were in that he showed we're not at all like the mean the standard a cemetery
09:54:16 these that I would think that would be legitimately fitted by a series by a series of absorption lines. And so, I guess the question is when you get a line like that mean what is the physics that's causing that asymmetry and how do you deal with that
09:54:31 or how do you remove that so that you can fit the spectra sensibly.
09:54:38 I might not have seen that spectrum you're referring to.
09:54:45 Yeah, it would be hard for me to pull that back up again but instead of it being instead instead of the line being symmetric, there was this long fall off like this, and then it came up to like what you would, so maybe it's a convolution of two different
09:54:58 lines right. Like, I mean, the profile itself is always symmetric, but you know, a total thing that you see in your spectrum can look super asymmetric and you just have to figure out the combination of all these symmetric profiles to make it work out.
09:55:14 I guess that's what I'm questioning I mean in general I agree with you just but but it at least in some of the spectrum, you've got a tail that's going like this I mean you would have to have like such a huge number of absorption to model something like
09:55:31 what penguin was showing that I don't know, I had always, it's usually pretty straightforward. Yeah. The example I showed had quite an asymmetric wing on the, on the red end due to just a superposition of the Milky Way silicon for and the lyman alpha
09:55:50 of one of the components was fitting so you know the superposition does get you that.
09:55:56 Yes As Chris was just saying the way profiles are just basis functions and so.
09:56:06 So adding those two together and then involving with the widespread function of the instrument. You know, it does generally give good fits with with few components. Now, we can argue all day about if the number of components we choose is the right number and
09:56:15 and this goes back to cloud size and everything else, but it can never be unique right but, yeah.
09:56:22 So the line broadening from the Doppler me parameter is a combination of both the thermal and peculiar velocity motion that could range from let's say 20 kilometer per second 100 kilometers per second so you don't need a lot of components to fit the whole
09:56:35 profile.
09:56:38 You want has been patiently raising her hand.
09:56:41 I'll bet.
09:56:42 Okay, so my thought was actually about whether or not there could be just intrinsic asymmetry in the Broughton and ally. And I was really just thinking about one example I saw in, I believe, Josh winners paper about cosmic ray accelerated cold blobs.
09:57:02 I'm sure there are other cosmic ray experts here that can cracked me.
09:57:08 But I believe that in the case where a cult blob is accelerated by cosmic rays. There's a symmetry within a call Bob cosmic refresher on what that result in a intrinsic asymmetry of the broadening of the lines, because I understand that, with thermo brightening
09:57:30 you just automatically get just Gaussian plus whatever.
09:57:35 But what if you physically have a symmetric brawny, and I can imagine was non classical turbulence. You can also have it, a symmetric brought me, that even a consideration or observers just don't even care.
09:58:02 Because we talked about how we distinguish between different models, right. So, I guess I just want to think about. Hmm. how big of a difference would it be if you have a symmetric colossal felt in one cloud let's say is it five kilometer per second,
09:58:10 10, that I don't know I didn't do that work I don't, I don't know if they did mock ups operations with maybe Trident that might help right
09:58:20 so that I don't know.
09:58:22 So maybe it's all within the noise level, and when even pop up. Even if you try to fit with some a symmetric profile, but I'm not sure it's just something I've been thinking about and I don't think I've ever actually thought that question through some
09:58:41 of the people, because they will just say, oh yeah well you just add another Gaussian component but that's not what I was asking.
09:58:52 So I guess one question I had that may kind of speak a little bit to it you only is talking about is, so I've.
09:59:00 I have some laboratory experience with getting instrumental profiles but if I'm do, does anybody in this breakout room or what's left and today know how, what the standard practices for getting the light of the instrumental profile on a telescope.
09:59:21 Is that just done at the beginning of every observation or
09:59:27 we can use an article I am, I guess.
09:59:32 So could you just repeat that.
09:59:35 I'm thinking if you want to measure the intrinsic line profile what people will do is to use an arc lamp.
09:59:43 Is that what people do.
09:59:46 Todd. Todd has his hand up and I think he'll comment on the line spread function and how people try to constrain that go ahead talk.
09:59:56 Right. So what happens at the beginning of the deployment of a space instrument is that, for example with costs observations of some star or obtained, and then compared to much higher resolution observations that were obtained say with space telescope
10:00:14 Imaging Spectrograph, and that comparison provides constraints on the line spread function. And then that line spread function is used, pretty much. Subsequently, exclusively.
10:00:27 It's not revisited and maybe. Well, sometimes it is but not very often because it's hard to get Hubble Space Telescope time even for calibrations unfortunately.
10:00:38 Now, if you, you know, you have to sort of drill down and look at the history of the things sometimes there are derivations that are done on the ground before the thing is deployed with the ground based instruments.
10:00:53 Uh huh. It's usually possible to, to build an instrument that has a really nice line spread function so it's usually approximated as a Gaussian.
10:01:02 But what I really like to return to is this business of an asymmetric profile. It's really not fair to say that, that observers don't care. We're quite interested in that.
10:01:13 And
10:01:16 you can think of what we do as just some way of parameter rising the data, and so you can take the extra Gaussian or whatever that we put in, and then you can look at whether or not that's consistent with your model that requires some intrinsically a
10:01:32 symmetric profile.
10:01:34 So you might ask, Well, why don't you just put in the intrinsically asymmetric profile in the first place and the answer is because it's totally unconstrained you know you can you can do that, and you don't really have any way of determining typically,
10:01:49 whether the line is really intrinsically asymmetric or if it's just a juxtaposition of multiple components, and if the juxtaposition of multiple components is entirely reasonable.
10:02:03 then there's no.
10:02:06 I mean, unless you have some
10:02:11 theoretical desire to require those asymmetric profiles there's, there's no real reason to say okay, it must be intrinsically asymmetric I mean if you, if you can explain some reason that this profile has to be intrinsically symmetric then that that's
10:02:28 exciting. But typically, you know, there's no real indication that the addition of multiple components is unreasonable that be values and such as other folks have said, Are you know quite plausible.
10:02:43 So that's the that's the problem and, and let me just add, when we find something that is isolated and observed a very high spectral resolution. And let me just point out in space we can get spectral resolution of 200,000, which is often not available
10:03:01 on the ground.
10:03:02 The profiles look quite symmetric net the looked at many many of these things, the housings actually, and I haven't seen any compelling examples where, when you have that nice isolated.
10:03:18 Well, observe high Signal to Noise high resolution profile that there's any compelling indication of a symmetry so that's really have a quick defense of what observers have done.
10:03:32 What I'd love to see it, it would be fascinating Of course anything new, is very interesting, but so far. I personally haven't seen any evidence that a symmetric profiles are required.
10:03:44 When you really have a clean and well isolated case to study.
10:03:49 I hope that's helpful. Yeah, I think that's.
10:03:52 I hope that's helpful. Yeah, I think that's a thank you for chiming in on that because yeah it's interesting to see what observers see and see if Yeah, if you're if you're fitting these things as multiple components when there's some intrinsic a cemetery
10:04:07 but it sounds like there isn't. So, thank you for for chiming in on that every now blocky has a comment or question and I think it's related to all of this.
10:04:18 I just wanted to circle back to John's point about cosmic ray accelerated clouds.
10:04:24 So, as far as I know I don't think there have been any synthetic spectra like pointed directly at simulations of this.
10:04:32 But, so the physical kind of model you'd have is like a cloud, and cosmic ray pressure buildup on one side of it and that's accelerating to that other side.
10:04:42 And so, I don't think that cosmic ray pressure would explicitly make it like add to the absorption line signature, but it would change the temperature and the density of the gas, and it would heat that the cloud boundary layer.
10:04:58 And so, you would expect, maybe to see multiple phases of gas at a similar velocity or like, you know, if you can actually resolve the scales of you know this is probably a small cloud let you know and trained in a hotter medium so if you can resolve
10:05:14 the you know the medium the body of the cloud and the that line maybe you might see slight velocity differences but I'm not sure that this is something we can resolve in current observations, but actually I don't think 3d simulations have been run over
10:05:30 this so this is certainly something we should consider looking into making predictions for.
10:05:36 Yeah, I don't know, have there been a synthetic spectrum made for for kind of non standard physical systems like including cosmic ray physics in these clouds, for instance I I'm not aware of any.
10:05:52 But
10:05:55 perhaps other people know.
10:05:57 Yeah, that'd be cool. Someone could produce that it'd be very interesting, is actually something we're looking into with Justin.
10:06:08 Some collaborators, but these are. This is for a simulation of like a cosmic ray dominated Halo not individual cloud blitz that are accelerated by cosmic rays and, like, cosmic rays streaming, which would be slightly different applications.
10:06:29 I'd like to add a comment that's, um, that's related to a symmetric line profile so of course when we talk about absorption line spectroscopy. The assumption is that your gases diffuse and them, and that does relative to brights background continuum source.
10:06:48 So, I'm in response to Dr. Young Lee's question, what about if you data. So, what happens is that when your gases diffuse and then you don't actually get the kind of complex and detailed absorption nine profiles that we've been demonstrating, and you
10:07:11 may be spatially integrating the line that dimension line to just get a sense of that kinematics. And the other thing is that when you have if you data, that means some of your emitting gas is not directly in front of that bright background source and.
10:07:33 And so you need to care about radiative transfer modeling, and especially if your line is say Limonov is a strong and resonance mine, and so you have complicated profiles that look a bit like the piece that may profile and.
10:07:51 And there are no unique answers to the parameters to put into this model and yeah so there's a lot of degeneracy and I put it paper in the chat us. Lee at all, 2020 and that's a good example of showing the amounts of the Geneses.
10:08:15 Yeah.
10:08:17 Great. Thank you, Marie.
10:08:22 Are there any other topics that people want to discuss related to the absorption line fitting.
10:08:30 Now this seemed to cover I think a lot of the questions that you want Leah and I had at a couple of weeks ago so thank you very much for organizing this it's excellent, Excellent, thanks Johnny, and anyone else.
10:08:51 Oh, I make a comment. So I have some undergrad students who are beginning to learn the ins and outs of this stuff and one thing that's challenging for undergrads is we're hearing that everyone's writing their own like fitters.
10:09:05 And so, like, it's very overwhelming when you try to provide resources for a student because there's literally 1000 different places you can point them to for this boy profile stuff so having, like this video and all the resources that are being made
10:09:22 available is going to be super helpful. So it's awesome.
10:09:25 Excellent. Yeah, I think that's one of the main points of these tutorials is is breaking down these boundaries like this So, making it good introduction for people who are new to the field or from outside subdiscipline, yeah.
10:09:43 Nicola.
10:09:42 Yeah, has a very small maybe technical question regarding wrong to using EMC which has some limitations for one the number of parameters become large is probably not example use cases but the primary reason is because the proposal distribution in EMC
10:10:02 is a simple garden which does not allow the code to explore the parameter space very quickly so it's very slow and conversion.
10:10:12 And I presume it would be interesting to to extend this to other other code that are more robust and faster.
10:10:39 And my goal is really designed to do small chunks for me.
10:10:32 Just to jump in that's factually untrue EMC does not use a Gaussian proposal distribution.
10:10:41 By default it does
10:10:44 that, what does. But, by default, it does I looked into it, and you can change to two other proposal distribution.
10:10:54 It EMC the offline and variant sampler Yes, it doesn't use the Gaussian proposal distribution and uses and I find it very and stretch move, which is not the same thing as a galaxy and proposal distribution.
10:11:08 But I think the point still hold that it's computationally expensive. Once the profiles become complicated or. There are a lot of lines are a large number of properties.
10:11:19 Yeah, is that that's that's a challenge with an mC mC Basie and approach to this whole business, which is one reason we haven't gone to that wholesale.
10:11:31 He's just computationally demanding and, and so it's it, I think it's clearly a nice, it's the best way to do this but it's hard with computational facilities that many of us have access to.
10:11:48 If you have something messy I mean with the cosmic data, it's really hard because there are so many different lines, different red shifts and components and it's just, it's hard to do it that way.
10:12:01 Yeah, I wrote an algorithm to group lines together, being a big line list that has to be fitted together, um, you know, so it's it's this kind of rabbit hole thing where, you know, you, you have one line that's blended with another, and that other line
10:12:18 has, you know, three other transitions that all can be blended with other things. And, you know, like, with data set like cosmetic like Todd's been beating his head against for for a while.
10:12:32 It's it's just, I mean it would be great if we could somehow come to a means to to leverage mC mC but it's how many lines you have.
10:12:43 There's a whole thing.
10:12:45 It just depends.
10:12:47 What's a big number, hundreds, hundreds, hundreds and computers. That is certainly need a supercomputer, but you do need, right. What is true is that EMC is not particularly efficient for large problems.
10:13:05 And so, gradient and Hessian aware samplers are gradient and our proposal distributions are
10:13:16 going to be necessary to do that in a reasonable amount of time not because of those like purely purely numerical computational considerations but Geeta sort of algorithmic stuff but.
10:13:29 So, another interesting thing is, there are a lot of purpose built there basically there's a lot of tools that exist.
10:13:38 That could make this a lot faster. So for example, something like numb Pyro one of these new ways of doing of combining influence, auto grad, you know GPUs GPUs etc.
10:13:55 might be the way forward for something like you know hundreds of lions, where it has all sorts of interesting proposal distributions and like, it's very.
10:14:06 There's some really advanced stuff there, and it would be a little bit, you know, it would take a bit of courting to get some of these codes to interface with that.
10:14:15 I'm Carol, Nicola if you guys know a better sampler, can you put it on Slack, I'd love to know
10:14:26 talk.
10:14:27 Oh, you're just applauding you're not raising your hand. Okay, yeah, I mean it's great this is what we need is to find ways to do this faster that I'm very interested.
10:14:40 One thing that this is all related to in the halo 21 medalists at group over the last few weeks, there's been an effort to do generation of synthetic spectra for very idealize systems where we know the middle of the cities and then getting observers to
10:14:58 do absorption line fitting to the output spectra to see if we can pull out and extract the exact same medalist city that's put into the simulation and make sure that there's crosstalk between the observers and the simulators.
10:15:12 Right now, there have only been a couple of simulate or sorry, a couple of observers who've taken taken up the challenge, but this seems like the right group of people to pitch this to, given the people who are sticking around 15 minutes after the end
10:15:26 of our session talking about weight profile fitting of absorption issues. So, I'm Zach Haven and Jane Charlton are going to present on this on Friday for about 15 minutes in the morning, before we go into the structured discussion so I encourage people
10:15:40 to check it out. Because it'll provide a little bit of context for what's going on and and maybe you guys could could get involved with this because it seems like the tools that you presented both Murray and Joe and Rahman, are, are super awesome, and,
10:16:03 and, and maybe could make quick work of the simplistic spectra.
10:16:03 Anyway, I didn't mean to derail the discussion, just wanted to throw that in there because I think it's, it's cool for making sure we're all on the same page between the simulations and the observations.
10:16:15 Any other comments or questions regarding this tutorial.
10:16:22 Yeah, this is going on a little long but since Todd is on the line that was going to maybe ask him verbally what I had put into one of the slack channels.
10:16:32 So, Todd it when doing astrophysics doing absorption spectra in the laboratory.
10:16:39 Sometimes people can take an original spectrum and do a fully ad convolution of it. If it looks like a line might, might be more highly resolve that was wondering if, if that kind of D convolution is ever used in.
10:16:58 In the kind of spectra that, that you're fitting.
10:17:02 So Ed Jenkins has experimented with that quite a bit, and I'd have to go back and look at some of the stuff that he has written. I think it turned out to be hard to really with the quality of the data that we typically have applied those kinds of methods,
10:17:25 but he's, he's looked into that a fair bit I'm afraid I don't have spent some years since I looked at what, and found so I don't have it.
10:17:36 You know at my right well I can go look you said, Ted Jenkins, Edward Jenkins. Oh, yeah. I don't know how much of it, he has published, it's sort of you know he he's a guy who does a lot of stuff and shares his results with his team's, some of it gets
10:18:01 Yeah, I know that in, in the laboratory had a lot of experience with, it's a really powerful tool, but if you don't use good judgment, you can all of a sudden get like totally false lines and so, got to be very careful.
10:18:15 Okay. All right. Thanks.
10:18:19 Okay.
10:18:20 I were about 20 minutes past the hour, thank you very much to the tutorials, Joe, Evan Murray, Joe again and wrong been wonderful presentations I learned a lot and we'll going back, we'll go back over these tutorials to really catch up on the details
10:18:41 but very much appreciate this.