Episode #122 - What's the Weather in Space? w/ Jose Arnal

Interactive Transcript

spk_0 It's the language of the universe.

spk_0 But I don't understand it.

spk_0 Welcome everybody to the Math and Physics Podcast.

spk_0 This is episode number 122.

spk_0 My name is Ray and today I am joined by my dear friend Jose Arnal, a senior PhD student here at the University of Toronto Institute for Aerospace Studies.

spk_0 And he's currently studying space physics and data assimilation where we're going to talk about essentially what he does, how he does it, and what is there to learn from it.

spk_0 So Jose, welcome to the show. How are you?

spk_0 Can you introduce yourself?

spk_0 Yeah, thanks Ray for having me. First of all, and doing great.

spk_0 It's really excited to be here to talk to you and explain to someone what I do to the audience a little bit.

spk_0 Alright, so you're a senior PhD student here. This is your fourth or fifth year?

spk_0 I don't usually like to talk about it too much. Just kidding. I started in 2018.

spk_0 I felt like it's taking me a little bit longer than I would have liked.

spk_0 I started at the master's level, but then transitioned to the direct entry PhD program.

spk_0 So I think this is technically my sixth year, I would say.

spk_0 So yeah, there's a lot of very different for the European students who a lot of times don't even really have direct entries. I don't think they do.

spk_0 So I think this is more the Western hemisphere where a lot of people have a chance to essentially skip your masters and go into strength of your PhD.

spk_0 So that's what you did.

spk_0 That's correct, yeah.

spk_0 Now, essentially my understanding and one thing I should also say is that he helps me in a lot of my research as well because one aspect of his is also mine.

spk_0 But let's talk about the one that isn't. Let's talk about space weather and space physics.

spk_0 So what got you into there? What kind of determined that you were going to do a PhD in that and just take us through your journey?

spk_0 Yeah, so if I can be fully honest, I saw some of the plots or some of the figures that you can produce when you're making predictions or simulations of space physics.

spk_0 Or space weather and it just looked really, really cool.

spk_0 You see all these colorful streams and what we call coronal mass ejections, which I'm sure we'll go into and it just looked really cool.

spk_0 But can I going broader than that?

spk_0 I don't know since when but I've always had a fascination with both space physics or maybe let's say space generally.

spk_0 But also with fluids, right?

spk_0 It's independently not not necessarily together, but these are not usually areas that you would think are tied, right?

spk_0 When we think of fluids, we might think of the ocean or we might think of airplanes.

spk_0 Such stuff like that and I thought that was really interesting.

spk_0 And when we think of space, we usually think of a vacuum, right?

spk_0 But actually space physics or space weather more precisely is actually a good example of how fluids can actually appear in space.

spk_0 And the equations of fluid dynamics and modeling them is actually really, really relevant in that sense.

spk_0 All right.

spk_0 I mean, it sounds a little weird to somebody that usually is probably thinking that space is almost a vacuum to now say that there are fluids in there.

spk_0 And not only there are fluids, but that essentially dictates how a lot of things work in space.

spk_0 So tell me some of the big things about, so when you say space weather, the first thing that I'm thinking about is what's the temperature up there?

spk_0 That's probably not what you're thinking about though, or maybe it is.

spk_0 So what exactly do you aim to model?

spk_0 What exactly do you aim to do?

spk_0 What are some of the challenges when you are trying to kind of dabble in space weather for space weather?

spk_0 Right. So I think temperature is actually one of the variables that we're interested in, but it's not, I would say the primary one.

spk_0 Let's say in the earth, maybe that is one of the primary variables that we're interested in.

spk_0 But we can think of the sun as a hot ball of plasma.

spk_0 So you can, okay, you can see that's a fluid probably.

spk_0 But this ball of plasma, the sun is essentially emitting or it's sort of plasma is escaping from the sun.

spk_0 And the stream of charged particles comes towards the earth, right?

spk_0 And these particles can actually have quite an effect on our technology and infrastructure.

spk_0 By sticking more on the science side, what we're doing is that by essentially using big computers to model how this plasma comes out towards the earth.

spk_0 And using the equations of fluid dynamics, we can quantify how, for example, the velocity field changes how we go from the sun to the earth.

spk_0 We can model the temperature and really important quantities to also model is the magnetic field.

spk_0 Because the magnetic field has quite a large impact on the geomagnetic events that happen at the earth.

spk_0 So when you say magnetic field, I'm assuming you're talking about magnetic field of the sun and how that impacts the earth.

spk_0 Yeah, so we call this the interplanetary magnetic field.

spk_0 But yeah, it's essentially the magnetic field that's coming out from the sun and how that's actually interacting with the earth's magnetic field.

spk_0 So this is, I guess, something I never actually even thought about.

spk_0 But all right, yeah, I guess the sun obviously also has a magnetic field.

spk_0 But in the earth, when you usually think about the magnetic field lines, you know, like the line thing of earth, like that doesn't span very wide, right?

spk_0 Like it's not very far.

spk_0 So are you saying like the earth is within the sun's influence of magnetic field?

spk_0 Or are you just saying that the magnetic energy, quote unquote, is transported to the earth?

spk_0 So I would say both in the sense that certainly the sun's magnetic field is much more powerful, right?

spk_0 But when you're near the earth, what's going to be important will be the earth's magnetic field, right?

spk_0 Right, right.

spk_0 But they actually influence each other quite heavily.

spk_0 More so, the sun's influencing the earth than vice versa.

spk_0 But it's actually because of the earth's magnetic field that the, that our atmosphere's not blown away, right?

spk_0 If we didn't have, if the earth didn't have a magnetic field, our whole atmosphere would be blown away by the solar wind, right?

spk_0 When you say the word solar wind, what is winding?

spk_0 What exactly is the wind?

spk_0 Are you talking about the magnetic?

spk_0 Are you talking about the plasma?

spk_0 You're talking about a combination?

spk_0 What exactly is the wind?

spk_0 Right, so what we mean by wind here is you can think of it as the ions and electrons, this matter that's in space and its movement through space.

spk_0 So I'm talking about here, its velocity, but I guess I'm talking about everything.

spk_0 Also, it's magnetic field.

spk_0 One of the unusual or maybe unintuitive aspects of, of plasmas, or at least these types of plasmas, is that we can say that the magnetic field is actually frozen in to the, to the plasma.

spk_0 So the magnetic field is transported with matter, if that makes sense.

spk_0 So you have thought of electrical energy being transported before?

spk_0 I don't think I've ever actually thought of a magnetic energy being transported before.

spk_0 Is there like any type of analogy or any kind of understanding that could help with understanding how, I mean I guess it's one and the same, right?

spk_0 I mean saying that is, it's obviously one and the same.

spk_0 But usually again, when you think of magnetic energy, you think it's more like in whatever's in the field that's mainly being affected.

spk_0 But now you're not only saying that it's in the field, but you're also saying it's transported.

spk_0 That's correct.

spk_0 So is there, is there any way that I could better understand the transportation of this magnetic field?

spk_0 Yeah, I think, I think maybe I'm making it more complicated than it needs to be.

spk_0 Think of just parallel magnetic field lines, okay?

spk_0 And they're, let's say there's this cloud of plasma in between the top and bottom of this image that we have.

spk_0 And let's say that we can somehow push, we're blowing this cloud away.

spk_0 You can think that the magnetic field lines will actually move with that cloud.

spk_0 So while they might start out, just straight from bottom to top, let's say, as the cloud moves from left to right, then the magnetic field lines will also be sort of pulled in that way.

spk_0 So at the top and bottom of this image, there'll be straight and parallel, but there'll be dragged along with this cloud of plasma.

spk_0 All right.

spk_0 So, and this is obviously only because of the fact that the, because the plasma is electrically charged and there's a relationship between that.

spk_0 That's actually transporting this whole thing.

spk_0 Yeah.

spk_0 So when we say that there's a high solar activity, first of all, I think this is a phenomenal time to also talk about this because, again,

spk_0 I don't know how many people are in the Western hemisphere here, but we had a, I mean, I guess the CMEs observed everywhere.

spk_0 It's just where you can see it best, but there are coronal mass ejections or CMEs as you pointed out in the beginning that occur quite often.

spk_0 Or maybe they don't.

spk_0 Maybe you can explain what those are, why they impact us, and why are we so interested in them almost?

spk_0 Right. Right.

spk_0 So I would say in this last year in 2024, we've seen a lot of genetic activity mainly due to coronal mass ejections.

spk_0 And I think for the first time, it's in a lot of people's mind where it wasn't before.

spk_0 So maybe some of the listeners would have seen the Aurora Borealis for the first time in their lives.

spk_0 I actually haven't been lucky enough to see it yet, but I'm looking forward to the day that I can see it.

spk_0 But these, I'm sure most people are familiar with the Northern Lights, right?

spk_0 Or Aurora Borealis. And this effect, this phenomenon usually happens much closer to the poles, but it's coming back a lot more south if you're in the north.

spk_0 And it's appearing, let's say in Toronto, where we're located, right?

spk_0 And this is due primarily because of coronal mass ejections.

spk_0 So what's happening is that at a high level, the sun has this 11-year cycle, where it will have more sun spots, let's say we call that solar maximum, and less sun spots, right?

spk_0 And maybe we'll call that when it's quiet, okay?

spk_0 So that's not called solar minimum?

spk_0 Perhaps it is. Yeah, maybe it should be, maybe it is.

spk_0 Okay.

spk_0 So what's happening right now is that we're right now at a solar maximum where there's many sun spots, and actually the phenomenon by which this occurs is not fully, fully, fully understood yet.

spk_0 But we're getting a whole host of coronal mass ejections, which you can think of these clouds or these eruptions of magnetic energy that are traveling in space.

spk_0 Now, when we're unlucky and the earth passes by the trajectory of one of these coronal mass ejections, then the magnetic field of the earth is compressed quite significantly.

spk_0 And this causes a through reconnection in the magnetic tail, which is again, a quite complicated process to go into, a host of charged particles will come streaming down towards the earth's poles.

spk_0 And we see this as, yes, as we're all right, but we're all this, for example.

spk_0 So the charged particles from, okay, so you said specifically it's a burst of magnetic energy.

spk_0 Yeah.

spk_0 So is there also, again, this is, might seem like a dumb question here, but like, there's also plasma in there.

spk_0 Yeah, it's just a heap of everything.

spk_0 It's a whole heap of plasma that's coming towards us, and it's actually coming towards with a big shockwave.

spk_0 So it's moving super, super sonically or super, alvanically towards us.

spk_0 Sorry, what was the word you just used?

spk_0 Yeah, so I use the word super alvanic, and that comes from the physicist who came up with what I call the MHD equations, which maybe we'll go into later.

spk_0 These equations describe the motion of plasmus.

spk_0 His name was, forget his first name, but his last name was Alvin, okay, kind of history lesson here.

spk_0 But to get to cut to the chase, we can think of, usually we think of normal fluid as being supersonic.

spk_0 Let's say we have a fighter jet traveling through the atmosphere, and it's going so supersonically, right?

spk_0 So that means that the fluid is moving faster than the speed of sound, right?

spk_0 But when it comes to plasmus, we can also have a supersonic flow, but we could also have a super alvanic flow, where the plasmus is actually moving faster than the alvanic speed.

spk_0 And that is essentially related to, let me see how we can discuss this.

spk_0 Okay, so the sound speed, let's think back to normal fluids.

spk_0 What is the sound speed in space though?

spk_0 How does that even make sense?

spk_0 Yeah, so the sound speed in space is defined in the same way that the sound speed is defined in a normal nonconducting fluid.

spk_0 Yeah, but what was the compression?

spk_0 You were still compressing matter.

spk_0 So in the atmosphere, we're compressing a gas in space weather or in space, it's like we're compressing a plasma.

spk_0 And so that there's still a sound speed associated with that compression.

spk_0 Now, when it comes to the normal sound speed that we're used to, we're usually, it's defined in terms of, without getting too technical.

spk_0 And we can think of it as the magnitude of pressure.

spk_0 It's directly related to the pressure.

spk_0 Exactly.

spk_0 And the pressure and the density.

spk_0 So the alvanic sound speed, when you look at it in equation form, it looks very similar to the way that we defined the sound speed.

spk_0 But instead of being based on the thermodynamic pressure, it's actually based on the magnetic pressure.

spk_0 It's basically just, it's a fancy word, it's just related to the magnitude of the magnetic field.

spk_0 Okay, okay.

spk_0 So you're saying on earth, the relation is kind of, if you're going faster, then the sound, then the air can compress.

spk_0 Now here you're saying faster than the magnetic field can compress.

spk_0 That's a good analogy, yeah.

spk_0 All right, all right.

spk_0 So we're essentially, so, okay, so we've understood that these CMEs have an effect on earth.

spk_0 Now, before talking about various types of effects, would you care to discuss maybe how these magnetic events actually occur on the Sun?

spk_0 Yeah, I mean, that's a deep, deep question.

spk_0 Right, briefly, because I mean, so I think I've mentioned this on the podcast before, but in my,

spk_0 whoof, grade 11, I think this was, I did a co-op at my high school.

spk_0 I think I told you about this, too, where I went to York University and I modeled the Sun spots.

spk_0 Okay.

spk_0 And my whole quote unquote thesis in grade 11, obviously, so not thesis, but my research topic was essentially calculating the differential rotation of the Sun.

spk_0 Okay.

spk_0 Knowing that the list is out there, the Sun is, as he said, a ball of plasma.

spk_0 It's not a solid ball.

spk_0 So each latitude, almost you can think about it, rotates at a different velocity.

spk_0 Because essentially, it's not one solid ball.

spk_0 It's just all layers of plasma.

spk_0 So my whole research was essentially to model those Sun spots.

spk_0 Right.

spk_0 And while I understood that, hey, this is cool, I never really understood where those Sun spots kind of came from.

spk_0 What caused them?

spk_0 I know, if I correct me if I'm wrong, that it's essentially like regions of low magnetic activity, which is why they appear black.

spk_0 Or I could be wrong there, but maybe explain.

spk_0 Because when you look at the Sun, through obviously a solar filter, and you're trying to look for Sun spots, they're essentially just black spots on the Sun.

spk_0 Right.

spk_0 So maybe you could explain why they're black.

spk_0 Is it why they're occurring?

spk_0 Is there a...

spk_0 Essentially, what's the theory behind magnetic, uh, Sun spots?

spk_0 Yeah.

spk_0 And why these things occur?

spk_0 So you're definitely putting me in the spot here.

spk_0 Okay.

spk_0 I don't know about that.

spk_0 That's great.

spk_0 It's definitely good to be pushed.

spk_0 Now, let me caveat, saying I consider myself more of a fit of a...

spk_0 Sorry, of an engineer, rather than a physicist.

spk_0 I focused on modeling, like the...

spk_0 The practicals, you know, how do we do these calculations well and fast?

spk_0 So I think you're pushing me a little bit beyond.

spk_0 But I'll try my best, and if anyone can correct me, I'll be happy to make a correction.

spk_0 But I believe that actually the Sun spots are regions of high magnetic activity.

spk_0 And maybe...

spk_0 So I feel like we're at the blind, leading the blind here.

spk_0 One of us is wrong.

spk_0 We'll find out after which one of us it is.

spk_0 But essentially, you have these regions of very high magnetic activity,

spk_0 where you have magnetic field lines that are actually in opposite directions.

spk_0 So in the same Sun spot, you might have some magnetic field lines pointing out,

spk_0 and then some magnetic field lines pointing in.

spk_0 Okay.

spk_0 As opposed to...

spk_0 As opposed to just one direction, all in or all out.

spk_0 Okay.

spk_0 So you're saying most of the Sun is all in or all out.

spk_0 But magnetic field lines usually have some in some out.

spk_0 Yeah.

spk_0 And in the Sun spots, in particular, these magnetic field lines are opposed to each other.

spk_0 And this is creating a lot of, let's say, magnetic tension.

spk_0 Okay.

spk_0 And I don't fully understand how the process actually works.

spk_0 And I actually think this is still like a pretty active area of research.

spk_0 But one thing that happens is that you get something called a solar flare,

spk_0 which is essentially a burst of radio energy.

spk_0 So you can actually detect these flares before you can even...

spk_0 Before the CMEs actually come out.

spk_0 Okay.

spk_0 But at some point, the...

spk_0 At a certain point, and again, I don't fully understand how it occurs.

spk_0 Instead of just getting this energy that's being concentrated,

spk_0 matter, plasma in particular, will actually burst out as well.

spk_0 And so this is why Sun spots are very much related to corollum mass ejection activity.

spk_0 Okay.

spk_0 So solar flares are not CMEs.

spk_0 No.

spk_0 Right.

spk_0 So solar flares are simply just radio energy.

spk_0 Yeah.

spk_0 They're more related.

spk_0 More or less.

spk_0 But the CMEs or corollum mass ejections are more matter.

spk_0 We're talking about matter actually being transported out.

spk_0 Yeah.

spk_0 All right.

spk_0 Okay.

spk_0 So am I wrong in saying that the temperature of the Sun spot is significantly lower than the rest?

spk_0 Or was it so you might have noticed I didn't comment on that.

spk_0 Yeah.

spk_0 Okay.

spk_0 And again, I feel terrible because I'm supposed to be an expert in this.

spk_0 No, no, no.

spk_0 But again, I like to think of myself more of an engineer working on the data simulation side of things,

spk_0 which we haven't gotten into yet.

spk_0 But I believe, and man, I'm really putting myself out here.

spk_0 But I believe that it's actually higher temperature.

spk_0 What?

spk_0 Yeah.

spk_0 We're going to have to...

spk_0 We'll definitely talk about this later.

spk_0 Sorry to put you in the spot.

spk_0 Yeah, I know that's not really...

spk_0 But I think the best part about this conversation that we're having right now is...

spk_0 Of senior PhD student, still questioning things about what he himself is researching.

spk_0 So if anything, this is not a thing at you.

spk_0 This is more like everybody does this.

spk_0 Like no matter what year of research you're in, even if you're the most senior level of researcher you are,

spk_0 or you have, or you know, there's still more to know.

spk_0 Of course.

spk_0 There's always more to...

spk_0 And it's so refreshing that a senior level PhD student as yourself is still like...

spk_0 Oh, I'm not actually sure.

spk_0 And again, we've had professors, we've had all sorts of people do that as well.

spk_0 So it's always really nice to see that high level researchers also very curious in their whole research.

spk_0 Right.

spk_0 And asking...

spk_0 And asking still like, what is this?

spk_0 Why is this?

spk_0 Right.

spk_0 Okay, so we've spoken about CMEs and how they kind of come out of the sun.

spk_0 My very basic understanding of their effect on the Earth is that they affect satellites.

spk_0 Because that's I guess the only thing obvious in space.

spk_0 But what else does it affect?

spk_0 Does it affect the actual weather on Earth?

spk_0 Like the...

spk_0 I guess weather might not be the climate actually.

spk_0 The climate issue, broad, is weather?

spk_0 Yeah.

spk_0 I think that's a great question.

spk_0 So all of the...

spk_0 We can think of...

spk_0 Putting aside anthropogenic effects, which are obviously important.

spk_0 A lot of or most of the Earth's weather is driven by solar activity.

spk_0 If the sun were to pop out out of existence,

spk_0 then obviously the Earth would cool immediately.

spk_0

spk_0 Now in terms of how do you relate the activity of the sun,

spk_0 in the sense of coral mass ejections and the like to threshold weather,

spk_0 that I'm not sure about, and I think that's still a fairly active research area.

spk_0 Okay.

spk_0 Now I also just to...

spk_0 You comment, another comment, so you mentioned satellites.

spk_0 So certainly satellites are certainly affected by space weather.

spk_0 In recent news, you might have seen...

spk_0 I think it was either last year or the year before Elon Musk lost,

spk_0 I think, 40 of his satellites, starling satellites,

spk_0 due to essentially an increased drag, atmospheric drag,

spk_0 in the upper atmosphere that led to a bunch of,

spk_0 I believe, satellites, you know, coming down.

spk_0 Yeah.

spk_0 And then another big one, specifically for us Canadians,

spk_0 for us to think about,

spk_0 because we're a high latitude country,

spk_0 is actually that space weather can cause huge widespread blackouts.

spk_0 Okay. So what happens is that,

spk_0 due to...

spk_0 Basically, you can get electric currents on the surface of the Earth,

spk_0 and that induces high potential differences across electrical lines,

spk_0 like, total grid lines.

spk_0 And I think what happens is that you overheat capacitors.

spk_0 So you get a massive, massive, massive blackouts.

spk_0 For us, I mentioned a...

spk_0 a Canadian's one, not so recent, but significant or historic event,

spk_0 was in March of 1989,

spk_0 where a large region of Quebec lost...

spk_0 lost its...

spk_0 essentially had a big backout for nine hours.

spk_0 That was due to a GM magnetic event.

spk_0 So what can we do to prevent stuff like this?

spk_0 So we know...

spk_0 So, okay, first, actually, I had this conversation with my other lab partner recently,

spk_0 so maybe you can describe this to the audience as well,

spk_0 which is, how long does it take from seeing that,

spk_0 hey, first of all, how do we see that a CME exists?

spk_0 I guess you already said that we see a bunch of plasma,

spk_0 and I'm sure we have satellites that are looking at that constantly.

spk_0 So how long does it take for this to come to us?

spk_0 Because I'm assuming it's not traveling at the speed of light.

spk_0 No.

spk_0 So a fast CME would be, let's say, from 800 to 1,000 kilometers per second.

spk_0 All right.

spk_0 Okay, so that means we actually have days between the eruption from the Sun to the Earth,

spk_0 but that's not always necessarily easy to detect.

spk_0 Another sort of safety point that we have is we have satellites at Lagrange.1.

spk_0 So for those who are unfamiliar,

spk_0 there's a point between the Earth and the Sun,

spk_0 where their gravitational pull essentially cancels out,

spk_0 and means that we can bark as they were, satellite there to just observe the solar wind.

spk_0 So this gives us advanced warning of space weather activity,

spk_0 but it's pretty close to the Earth, so it's actually not great.

spk_0 So I think this is one of the reasons that we are actually researching,

spk_0 how do we predict space weather better,

spk_0 or we call it space weather prediction.

spk_0 Now, what can we do about it?

spk_0 So we can't do anything about it in the sense that we can't change space weather.

spk_0 In the same sense that we can't change the rest of the weather,

spk_0 there's no way that we can change space weather,

spk_0 but if we can model it or predict it well,

spk_0 then we can mitigate its risks.

spk_0 Can we minimize its impact on Earth?

spk_0 Yes.

spk_0 So in the same sense that, let's say you're planning a big event,

spk_0 and you know that, okay, there's going to be crappy weather,

spk_0 there might be a storm.

spk_0 I'm talking about just normal, just normal, terrestrial weather.

spk_0 You might be able to better prepare for that.

spk_0 So in the same sense, we want to mitigate the risk of space weather

spk_0 by being able to predict it.

spk_0 So, and again, this is maybe another field where I'm not super familiar with,

spk_0 but electrical power companies,

spk_0 if they have advanced warning of space weather events,

spk_0 can prepare their capacitors, for instance,

spk_0 so that they don't overheat.

spk_0 And the techniques by which they do that,

spk_0 I'm not super well versed on.

spk_0 So CME, you mentioned, takes a couple days to essentially get here.

spk_0 Now, for, again, again, all the hype of CMEs and stuff has really been,

spk_0 especially this year, because we saw two big ones.

spk_0 So my question is this, both times that I've, again,

spk_0 this might not be again in one of your areas of expertise,

spk_0 because it is a physics-based question, but again, have at it.

spk_0 Number one, why, so why does it last for so long,

spk_0 if it just takes that long to come to us?

spk_0 When I, let me rephrase that question,

spk_0 if let's say it takes three days to come to us,

spk_0 usually I hear things like, oh,

spk_0 the Aurora Borealis will be looking better tomorrow than it will today.

spk_0 So how do we know how long it will last for?

spk_0 And I guess that's the first question.

spk_0 Let me just ask you that.

spk_0 How do we know how long this thing will last for?

spk_0 Right, so, okay, you can think of this ball of magnetic energy

spk_0 coming towards the earth with a shock, essentially.

spk_0 So, there's going to be a...

spk_0 When you say shock, maybe, maybe you're reaching that,

spk_0 because again, shock, so you mean, like, I almost am magnetic.

spk_0 Magnetic shock.

spk_0 So what I mean by that is, there's almost not instantaneous,

spk_0 but a very rapid increase in the, let's say,

spk_0 the magnitude of the magnetic field.

spk_0 Magnetic field, right?

spk_0 And that, because we're talking about the skills of space,

spk_0 these are...

spk_0 These don't come and go very quickly.

spk_0 So there's going to be an increased...

spk_0 An increase in the magnetic field magnitude,

spk_0 and that will remain high for some time.

spk_0 And now, when you're talking about all tomorrow, it might be better.

spk_0 That's actually caused because, for a lot of these events,

spk_0 we don't just get one CME that's coming towards us,

spk_0 but we actually get a whole host of CMEs, let's say,

spk_0 five that all merge and impact the earth.

spk_0 And that's why they can last for quite a while.

spk_0 All right.

spk_0 Okay, so it's basically a bunch of them that occur,

spk_0 usually, I'm assuming at the same time.

spk_0 Yeah, well, they happen to...

spk_0 Not necessarily right at the same time, but in close succession.

spk_0 Now, why is 2024 so special?

spk_0 Why are we getting so many this year?

spk_0 Is it a solar...

spk_0 Because I'm told it's not even a solar maximum,

spk_0 or not some completely...

spk_0 So we're not...

spk_0 We're not...

spk_0 We're not right at solar maximum, but we're close to it.

spk_0 We're close to it.

spk_0 But you're right in that there...

spk_0 I mean, there's been solar maximums before,

spk_0 but it seems at least to me that...

spk_0 Yeah, this is a little bit different than that.

spk_0 We're just having more and more of these events happening

spk_0 in our day-to-day life.

spk_0 I think it might just be a matter of chance.

spk_0 Really?

spk_0 All right.

spk_0 Like, I can't think of any reason as to why this...

spk_0 Solar...

spk_0 This year of solar activity...

spk_0 Is it any...

spk_0 ...different than, let's say, 11 years ago.

spk_0 All right.

spk_0 Let me ask you a question here.

spk_0 Once the CME gets released from the Sun...

spk_0 Is it...

spk_0 I'm assuming because it's...

spk_0 Okay, so I'm assuming it's getting released from the Sun Spot.

spk_0 I don't even know if this is a correct technical terminology here.

spk_0 Yeah, I'm making this...

spk_0 Like from that region.

spk_0 It's coming from the solar surface.

spk_0 Essentially my question being...

spk_0 That is it like a radially outward,

spk_0 or is it like going in a specific direction?

spk_0 Essentially saying that...

spk_0 Can there be a CME on the other side,

spk_0 like going on the other side of the solar system...

spk_0 ...where we just miss it?

spk_0 Or is it like a radially outward?

spk_0 No, it's more the first...

spk_0 Like you mentioned where you have...

spk_0 You can think of as many eruptions that are actually happening...

spk_0 ...at the same time or close to the same time.

spk_0 All right.

spk_0 And we might just be unlucky in that the Earth will be passing through the trajectory of that CME...

spk_0 ...at that point in time.

spk_0 So it's not like the whole...

spk_0 ...the Sun is all blowing up at once, as I were.

spk_0 But it's just an eruption at a particular place in time.

spk_0 All right.

spk_0 Okay.

spk_0 So...

spk_0 I'm just trying to think...

spk_0 ...because I've always been interested in understanding more about like solar activity...

spk_0 ...and how it really relates to the Earth.

spk_0 Can you maybe explain...

spk_0 ...and this probably is very, very related...

spk_0 ...my understanding is that Northern Lights occur 24-7.

spk_0 Because there's always activity...

spk_0 ...and there's always the charge particles going towards the poles.

spk_0 Right.

spk_0 I know you explain this briefly, but maybe you can take me on a journey one more time...

spk_0 ...until why they get...

spk_0 ...I mean, I guess there's the magnetic field that's concentrated towards the poles.

spk_0 So it goes there.

spk_0 But what is the process or what is kind of the transport of like the charge particles...

spk_0 ...from the Sun going to the poles?

spk_0 Is it just because they're like just attracted to high magnetic field?

spk_0 Or is there a little more than that?

spk_0 Yeah, so it's quite complicated, but you're writing that...

spk_0 ...they're happening, you know, the sort of...

spk_0 ...or our realists is happening constantly.

spk_0 But you were usually more talking about them when they come closer south.

spk_0

spk_0 Right.

spk_0 But the process by which had this happens is that...

spk_0 ...I explained that both the Sun and the Earth have...

spk_0 ...independently a magnetic field.

spk_0 And the solar wind is essentially compressing the Earth's magnetic field.

spk_0 Okay, so what this looks like if you can picture it is...

spk_0 ...a head of the Earth and what I mean by a head is in between the Sun and the Earth.

spk_0 So ahead of the Earth, you have a compression, you have a magnetic field that moves closer to the Earth.

spk_0 And then on the back side or the rear side of the Earth...

spk_0 ...you have something that's called a magnetotail, where you actually have...

spk_0 ...you can think of it as a wake if you're familiar with that terminology.

spk_0 So you have essentially a wake of magnetic field lines that are concentrated...

spk_0 ...back side of the Earth and these can extend quite a distance away.

spk_0 And then what happens is you get this thing called magnetic reconnection...

spk_0 ...where it appears that magnetic field lines are essentially combining...

spk_0 ...and it's through this process that you essentially have...

spk_0 ...a host of charged particles that are along these field lines...

spk_0 ...are shooting from this reconnection event all the way to the poles of the Earth.

spk_0 Oh, yeah.

spk_0 Oh, okay.

spk_0 And what dictates that connection? Like, why is that?

spk_0 So they're following, they're literally following the magnetic field lines.

spk_0 So because the magnetic field lines are joining at this reconnection event...

spk_0 ...and they're going all the way to the poles where the magnetic field lines emanate from...

spk_0 ...that's where they're concentrated at the poles.

spk_0 Oh, interesting.

spk_0

spk_0 Now, is it right to say...

spk_0 ...so obviously it is right to say that they're more concentrated on the magnetic poles.

spk_0 Sorry.

spk_0 The magnetic poles.

spk_0 Yes, that's a geographic point.

spk_0 That's a very good point.

spk_0 No, what's the whole ordeal with...

spk_0 Actually, this is not even really related to...

spk_0 ...I would just think of magnetic fields.

spk_0 Let me maybe not.

spk_0 Because I know the geomagnetic...

spk_0 Sorry, not the geomagnetic north is technically the geographic south.

spk_0 Yes, it's kind of amazing.

spk_0 But that is nothing to do with it.

spk_0 That's just like, once every, I believe, like, 12,000 years...

spk_0 ...or something the Earth just goes through a whole process.

spk_0

spk_0 Nothing really to do with solar wind or anything.

spk_0 I don't believe.

spk_0 You don't believe.

spk_0 Right, right, right.

spk_0 Okay.

spk_0 When you were describing CMEs, you explained the process of solar flares.

spk_0 Yes.

spk_0 Now, you were mentioning one of the biggest things about predicting these CMEs are...

spk_0 Well, I mean, sorry.

spk_0 The biggest thing with CMEs is predicting them.

spk_0 The hardest thing is predicting them.

spk_0 Now, you mentioned that usually solar flares occur quite a lot before.

spk_0 And obviously, solar flares are traveling at the speed of light.

spk_0 So, is there any kind of thought process with like, oh, maybe we can just measure solar flares that occur at certain times?

spk_0 That may cause a CME.

spk_0 Yes, so there's definitely a connection or correlation between flares and CMEs.

spk_0 But it's not necessarily one to one.

spk_0 Right, right.

spk_0 But it's certainly a part of the puzzle piece.

spk_0 Or a part of the puzzle, I should say, to track...

spk_0 Or predict that solar activity, right?

spk_0 So, is it right to say that every CME will have a solar flare, but every solar flare won't cause a CME?

spk_0 That's a good question.

spk_0 I don't know.

spk_0 That's a good question.

spk_0 I'm not sure.

spk_0 So, we've discussed why...

spk_0 I think I'm going to kind of stop the physics question soon, because I'm still very curious about...

spk_0 So, solar flares.

spk_0 Can you maybe talk a little bit more about them?

spk_0 Do you know why?

spk_0 Because we describe kind of why we think...

spk_0 So, why is there like a sudden burst of the radio energy?

spk_0 So, we said that there's a magnetic field lines possibly pointing in both directions near the Sun Spot.

spk_0 What kind of causes that first trigger that makes the solar flare...

spk_0 I guess I mean the plant...

spk_0 I mean, if there's a lot of energy, there's going to be some kind of burst, like a some kind of pressure bubble.

spk_0 But maybe you can...

spk_0 If you can talk about the trigger event...

spk_0 Man.

spk_0 You're really putting it on the side of the thing.

spk_0 I think I don't have a good answer, so I won't attempt it.

spk_0 That's all right.

spk_0 Because, I mean, in general, if anybody listening can put it in the comments, or that would be very lovely,

spk_0 because I've always wondered about solar events and just a little bit more about how this works.

spk_0 Okay, so we've spoken about CMEs and essentially space physics.

spk_0 Now let me ask you a little question about what you do with them.

spk_0 Sure.

spk_0 Now, my understanding is that...

spk_0 Well, I know that you model them.

spk_0 Yeah.

spk_0 And you very briefly described MHD equations, or the magneto-hydrodynamic equations.

spk_0 Right.

spk_0 Maybe you can explain a little bit more about what those exactly are, why you are able to model it using those equations.

spk_0 Right.

spk_0 And can you use a better set of modeling?

spk_0 Right.

spk_0 This is a great question.

spk_0 So let's dive into the modeling, which hopefully we'll have better answers for it.

spk_0 But, okay, where do we start?

spk_0 So, as I mentioned, we're modeling the plasma essentially as a continuum.

spk_0 Okay, so that's already quite a strong assumption.

spk_0 Okay.

spk_0 So we're assuming that this is a continuum of matter, and that we're assuming that this essentially behaves as a gas.

spk_0 Right?

spk_0 But not only is it a gas, we're assuming that it behaves as a perfectly electrically conducting gas.

spk_0 So there, we've baked in already quite a few assumptions.

spk_0 So one of them being that it's perfectly electrically conducting, that there's no resistance.

spk_0 And we're also assuming that the flow is inviscid, meaning that it has no viscosity.

spk_0 Okay.

spk_0 Or, sorry, I should say, yeah, no viscosity.

spk_0 So what I just described, those assumptions that I just gave you, if you start with essentially, you know,

spk_0 Newton's second law, F equals M A, through some fancy math, you're going to write a rive at something that we called Magneto Hydro Dynamics,

spk_0 or ideal Magneto Hydro Dynamics, which you mentioned already, which is MHD.

spk_0 So you can think of this as the oiler equations of gas dynamics, and these equations describe an inviscid flow.

spk_0 Okay.

spk_0 Sorry, just quickly.

spk_0 Yeah, oiler equations.

spk_0 The best way I like to describe it is literally the conservation equations.

spk_0 So when everything of conservation is basically conservation of mass, momentum, and energy.

spk_0 And if you just write them with a bunch of derivatives, make it look cool, you all of a sudden get these partial differential equations,

spk_0 which are often called the oiler equations.

spk_0 Now, these equations are differentiated from more famous Navier-Stokes equations, which we won't get into now, because of the fact that they're inviscid.

spk_0 So you're essentially dealing with an inviscid set of just conservation equations.

spk_0 That's right.

spk_0 And then what do you do?

spk_0 Do you add a couple equations to that?

spk_0 Do you just take those?

spk_0 What, what do you have to do?

spk_0 So you can think of, you've done a good description.

spk_0 You can think of the oiler equations of gas dynamics, coupled together with Maxwell's equations of electro dynamics.

spk_0 Okay.

spk_0 And now, how can we actually arrive at these?

spk_0 You can, essentially, without getting too into the nitty-gritty, you can, essentially,

spk_0 add an acceleration term that's due to the Lorentz force.

spk_0 And then, through that, because of this extra acceleration term, you're going to get some coupling in the momentum equation for the magnetic,

spk_0 you can think of as the magnetic momentum or the momentum associated with the magnetic field.

spk_0 You'll get an equation for the time evolution of the magnetic field.

spk_0 And you'll also get some components in the energy equation.

spk_0 So you essentially just add a bunch of Maxwell's stuff to where you're already.

spk_0 But I would say it's not a lot of hot.

spk_0 It's done very carefully.

spk_0 And this is actually why I've been one, I know, a price, I know, a price in physics for this work.

spk_0 Fantastic.

spk_0 Yeah.

spk_0 So when you say carefully, obviously don't go into the details, but any like standout laws or any standout things that must be followed,

spk_0 anything that, oh, you must satisfy this.

spk_0 Right.

spk_0 You must satisfy that.

spk_0 So the way I like to think about it is at a deeper level.

spk_0 So we can think of Euler or MHD as a result, a more specific result.

spk_0 But the broader sort of physics that's going on is we can think of Boltzmann's equation.

spk_0 So Boltzmann's equations describe essentially a transport of a particle distribution.

spk_0 Right.

spk_0 Now, if we take no acceleration terms and we take integrals of these equations, we can arrive at Euler's equations.

spk_0 And actually, we can arrive at Navier-Stokes equations as well.

spk_0 But we won't go into that.

spk_0 Now, if it's thinking back to Boltzmann's equations, if we add the fact that what I call the Lorentz force,

spk_0 we have this force that's essentially due to the cross product between the magnetic field and the velocity field.

spk_0 And this is called the Lorentz force.

spk_0 How big that is when it works.

spk_0 That's right.

spk_0 If you bake this in into the Boltzmann equation, you go through the same process of taking integrals and the like.

spk_0 You'll then arrive at what we call the ideal MHD equations.

spk_0 And again, you have to make a few assumptions and do a few sort of tricks.

spk_0 But it's done pretty carefully.

spk_0 Now, before we go into the nitty gritty of MHD because I have a couple of questions in how you model this.

spk_0 What are some different models that you can use?

spk_0 So first you said ideal.

spk_0 Right.

spk_0 So I'm assuming there is a non-ideal.

spk_0 And on top of that, just to ask you another question, are there any other models that might be used to model a solar wind?

spk_0 That's a great question.

spk_0 So yes, we can definitely relax.

spk_0 These assumptions quite a bit.

spk_0 So we can say, OK, resistive MHD.

spk_0 Right.

spk_0 And what is that important?

spk_0 And this is now where we instead of having a fluid that's perfectly conducting.

spk_0 Now there's actually resistance in the conduction of electricity.

spk_0 Right.

spk_0 That's one assumption we can relax.

spk_0 We could also relax the inviscid part.

spk_0 So we can say now we have viscosity.

spk_0 So now we have, let's say nav your stokes equations with Maxwell's equations.

spk_0 OK.

spk_0 What is this?

spk_0 Sorry, I don't like kind of tangent.

spk_0 What is viscosity in space, though?

spk_0 I guess it's the plasma itself that's viscous.

spk_0 Yeah, so this is like the inviscid assumption is actually quite a good assumption.

spk_0 Right.

spk_0 Because the viscosity of plasma in space will be minimal.

spk_0 Right.

spk_0 So it is actually quite a good assumption, but you can think of maybe laboratory plasmus.

spk_0 Yeah, definitely.

spk_0 Or fusion plasmus where perhaps these effects are more important.

spk_0 So the resistive part, how bad of an assumption is the perfectly electrical conductive?

spk_0 So I would say it is not the greatest assumption.

spk_0 All right.

spk_0 And this is really the surprising fact.

spk_0 OK, so I say mentioned where we're essentially model.

spk_0 I do image the assumes a continuum, right?

spk_0 So it's not very rare.

spk_0 But we know that space space plasmus are.

spk_0 And if you look at the, now I'm going to get really technical here.

spk_0 If you look at the mean free path between particles in space,

spk_0 they're in the order of one astronomical unit.

spk_0 That means they're in the order of the distance from the sun to the earth.

spk_0 Yeah.

spk_0 So if you know anything about fluid flow, this tells you that this approximation is terrible.

spk_0 And it shouldn't work at all.

spk_0 Quick, quick, quick description from mean free path for those that might not.

spk_0 Mean free path is essentially the average distance that any two particle will travel before hitting another particle.

spk_0 Correct.

spk_0 So you're essentially saying the average distance for one of these particles,

spk_0 they hit another one by this assumption is that it's one astronomical unit.

spk_0 No, so this is actually what's not the assumption, but what's actually.

spk_0 This is what actually, oh, sorry, I misunderstood.

spk_0 So this is what actually is taking place is that the mean free path for the average distance for a plasma.

spk_0 Is that correct?

spk_0 So a plasmaparticle is one astronomical unit before it's like another plasmaparticle.

spk_0 All right.

spk_0 So this should sort of point you to all this shouldn't work, right?

spk_0 But again, kind of diving deeper into the physics, there's this.

spk_0 What do you call it?

spk_0 There's a.

spk_0 Oh, man, I'm I'm I'm losing the words here, but there's.

spk_0 There's an effect that essentially read has a reduction in the effect of mean free path.

spk_0 Okay.

spk_0 And by that, I mean that there's I think it's actually a quantum effect that makes it so that it appears or the or the plasma behaves as if it had a much smaller mean free path.

spk_0 Okay.

spk_0 While I understand this, I don't really understand the context to the resistive part.

spk_0 Why is this advantageous or disadvantageous for assuming a perfectly electrically conducting fluid?

spk_0 Right.

spk_0 What does the mean free path have to do with that?

spk_0 Yeah.

spk_0 So that it's sorry, I guess it doesn't actually have much to do with resistive.

spk_0 I guess I just wanted to jump into.

spk_0 For the elephant in the room, the elephant in the room of ideal image D is that it's it shouldn't work.

spk_0 But when we.

spk_0 So there's some justification talking about this.

spk_0 This is the fact of being free path, but when you actually compare the predictions of ideal image D with observations, it actually matches up decently well.

spk_0 So it's not necessarily that the model is is built on go to groundwork as I were, but when you actually compare with observations, it does quite well.

spk_0 So when you say observations, obviously I'm assuming you're talking about solar observations, like only m8 like space weather based.

spk_0 So we're talking about works.

spk_0 Yeah, properties of space placements.

spk_0 So they're density magnetic field velocity.

spk_0 Now are there other equations, but non-MHD equations that you may be able to use the model.

spk_0 Yeah.

spk_0 So you can go simpler.

spk_0 So you can, for example, the very simplest you could go is Parker solar wind model, which is basically is not even a partial differential equation or ordinary differential equation.

spk_0 It's just an algebraic equation.

spk_0 That's that describes the solar wind is very, very, very, very simple.

spk_0 Okay.

spk_0 Now can you go more complicated non-MHD based?

spk_0 I don't know if it would be non-MHD based, but you can get, for example, multi multi species of HD.

spk_0 Okay, so one of the assumptions that we make with ideal MHD is that we're treating the electrons and ions as a single as a single fluid.

spk_0 But you can actually start treating ions and electrons.

spk_0 Like multi phase.

spk_0 Yeah, yeah, multi brain.

spk_0 Yeah, that's correct.

spk_0 So you can have that.

spk_0 You can have multiple temperatures, right?

spk_0 So in the same in the same way of thinking, right?

spk_0 Instead of having a single temperature, you know, have multiple temperatures.

spk_0 A pretty also popular technique is a simulation technique is called particle and cell.

spk_0

spk_0 Where you're now essentially keeping closer track of particles themselves rather than or things that behave like particles rather than this continuum approach.

spk_0 Yeah, yeah.

spk_0 But I guess beyond, I think you would always be tied in some way to MHD.

spk_0 All right.

spk_0 But it might not be ideal MHD.

spk_0 So MHD is essentially how you model this solar wind.

spk_0 Now, do you, okay, how do you start the modeling process?

spk_0 Because I know for, I mean, let me kind of give context that question.

spk_0 For most equations or most model or simulations, you would probably give a set of initial conditions somewhere.

spk_0 And it would kind of just take it from there.

spk_0 But my understanding of MHD or at least the way that you guys do your modeling, their solar wind, is that you use multiple models, depending on where you are modeling.

spk_0 That's right.

spk_0 So maybe you want to go over a little bit about how, and again, this is not just what you do.

spk_0 This is what most people do in the field of modeling solar wind.

spk_0 So how exactly just like start to finish, obviously brief, how do you model it?

spk_0 What models do you use and where?

spk_0 Yeah, it's a good question.

spk_0 So what we do, Utahias, is that we start at the surface of the sun.

spk_0 We actually have something called a magnetograms.

spk_0 You can think of these as images of the magnetic field, the radial magnetic field on the surface of the sun.

spk_0 And these are taken by satellites.

spk_0 Right. And these are actual, this is actually observational data.

spk_0 This is not a model. These are measurements.

spk_0 All right. So this is sort of our ground truth.

spk_0 So you need that to start start for our modeling capabilities.

spk_0 Yes, you need these.

spk_0 All right. All right.

spk_0 Oh, yeah. Makes sense.

spk_0 So what you do is you start with these.

spk_0 And then in the lower corona, and now let me explain what that means, the corona is essentially the region that's in the atmosphere of the sun.

spk_0 So you can think of perhaps from the solar surface to let's say 20 sort of radii.

spk_0 20 solar radii.

spk_0 Yeah, that's approximately.

spk_0 Yeah, approximately.

spk_0 That's pretty big.

spk_0 Because usually when we think of atmosphere, we think like 100, like 100 kilometers away.

spk_0 Right. Right.

spk_0 You're saying, right, 25 solar radii away.

spk_0 So this is very far.

spk_0 Yes.

spk_0 Okay.

spk_0 Well, so what we do is from, let's say the surface of the sun to 2.5 solar radii in this lower region of the corona, we're assuming two things.

spk_0 We're assuming a, we're not two things.

spk_0 We're assuming one thing and we're assuming that the magnetic field is current free.

spk_0 Okay.

spk_0 That means that there's no, yeah, there's no electrical currents.

spk_0 And math and the way we do this release.

spk_0 It's motivated.

spk_0 I mean, it is motivated physically.

spk_0 If you look at, yeah, you can sort of look at observations and you can somehow justify this assumption.

spk_0 But mathematically, it actually makes the equations quite a bit simpler.

spk_0 All right.

spk_0 So what we do is that we take the divergence free property in the magnetic field.

spk_0 So this just means that the magnetic field doesn't have any sources or sinks.

spk_0 Right.

spk_0 It can never just be popping in or out of a single point.

spk_0 Okay.

spk_0 So we take that condition.

spk_0 We take the current free condition, which we can essentially summed up as the, the curl of the magnetic field is zero.

spk_0 So you have now divergence and curl are both zero.

spk_0 And we have now magnetic field, the magnetic field on the surface of the sun through the magnetograms.

spk_0 And we can actually now extrapolate the magnetic field from the surface of the sun to any other radial distance that we want.

spk_0 But we obviously don't do this to any radial distance that we want because we know that the assumption is won't hold.

spk_0 Exactly.

spk_0 Okay.

spk_0 So we do this essentially up until 2.5 solar radii.

spk_0 What we do beyond that point is that we essentially open up the magnetic field lines.

spk_0 So any magnetic field lines that are, yeah, that are close beyond this point by close.

spk_0 I just mean that they are originally from the sun and they terminate at the sun.

spk_0 So any magnetic field that that are still are closed beyond beyond this, this radius.

spk_0 We open them up, meaning that instead of pointing towards now the point I words or vice versa.

spk_0 And then we do this from 2.5 solar radii all the way to 25 solar radii.

spk_0 Okay.

spk_0 And is this like a specific model?

spk_0 Yeah, this model.

spk_0 These are all sort of built by a physicist through the 60s.

spk_0 Maybe you want to talk about, like, is there a name for this model?

spk_0 Yeah.

spk_0 So the one in the, in the solar in the lower corona is called the PFFFS.

spk_0 It stands for, I won't be able to do the PFFS model.

spk_0

spk_0 And then from 2.5 to 25 is the SCS or the shot in current sheet model.

spk_0 And it's a name after the physicist who came up with it.

spk_0 And it's still relying on similar assumptions.

spk_0 Okay.

spk_0 So now we've gone from observations of the magnetic field at the solar surface, now all the way to 25 solar radii.

spk_0 Sorry, what?

spk_0 I don't know if I missed this very, very briefly, but can you just summarize the difference between the, like, the 2.5 to 25 and the, like, what?

spk_0 The assumptions, just the assumptions.

spk_0 What are the assumption differences?

spk_0 The assumptions are the same.

spk_0 Okay.

spk_0 But in the SCS model, the, the shan, or the second model, we can actually have current sheets appearing.

spk_0 And this happens because we've opened, we've opened the magnetic field.

spk_0 And then this essentially results in something that we call a current sheet.

spk_0 All right.

spk_0 Okay.

spk_0 From that point onwards, now, let's say we're at 25 solar radii.

spk_0 Now, what we do is that we, we use essentially statistical correlations that where we correlate the magnetic field lines or the magnetic field to the rest of the plasma properties.

spk_0 So we have correlations for the, for the velocity field, we have correlations for the density, as well as for the, for the temperature.

spk_0 Okay.

spk_0 So once we have all these properties, we have magnetic field line from, from the two models that I just described, we have the rest of the plasma properties.

spk_0 Then this is where MHT can now kick in.

spk_0 So now we have the appropriate boundary conditions that we need for MHT.

spk_0 And then from 25 solar radii now to 1AU, let's say, or 225 solar radii, approximately, then we employ the MHT equations.

spk_0 And we model these with, with numerical methods.

spk_0 All right. So the MHT really kicks in at 25.

spk_0 And this is I'm assuming to, because you don't need the model MHT all the way through and through.

spk_0 Well, mainly, mainly, well, actually some people do model MHT all the way from the solar surface to, to, you know, 1AU.

spk_0 But it really simplifies the modeling to, to just focus on the magnetic field in this lower, in this lower region.

spk_0 Okay. Okay. So I'm just trying to gauge when would somebody use the MHT through and through versus when would they not do that?

spk_0 Is there an advantage? Is it just computational advantages?

spk_0 So a big advantage is computationally.

spk_0 But I would say it's probably more self-con, not probably, but it is more self-consistent to just have MHT the whole way through.

spk_0 Okay. Okay. But then you need, there's a few things that you have to take care of.

spk_0 For example, um, coronal heating in, in, in the corona that we don't have to take care of if, if we just use the magnetic field.

spk_0 Yeah.

spk_0 So just like, there's a few difficulties in the, in the magnetic field line is really the main driver of the dynamics in the solar wind.

spk_0 Yeah.

spk_0 Now, this doesn't, this seems like a pretty hefty calculation.

spk_0 It doesn't seem like something you can do on a laptop.

spk_0 How exactly do you do these calculations?

spk_0 Yeah. That's a good question. So certainly we could not do these on a laptop.

spk_0 So what we do is we make use of supercomputing facilities to put it simply.

spk_0 Let's say we have a computational domain of consisting of, let's say, 10 million computational cells.

spk_0 We're going to now break them up instead of just operating on these cells on a single CPU.

spk_0 We're going to break them up and we're going to make use of, let's say, 1000 CPUs to process them independently.

spk_0 And then there's some communication that has to happen at the boundaries of where you have two different CPUs operating on two separate cells.

spk_0 So what are some challenges when you, um, so what you essentially described is very important.

spk_0 I feel like for a lot of listeners, especially in today's day and age, where like artificial intelligence and machine learning are just taking super big advantage of parallelization essentially.

spk_0 Right. Right. So what you describe is essentially HPC or high performance computing, essentially using parallelization to your advantage.

spk_0 So what are some challenges that you noticed in actually modeling and actually writing the code for this model.

spk_0 And were and how did the parallelization kind of add into the challenge.

spk_0 Right. That's a good question. So whenever we're programming, let's say a Python script or whatever.

spk_0 We're just, we're not worrying about what each processor is doing. We're just have a single processor that's, that's doing everything. Right.

spk_0 Now when you have, let's say, 10,000 processors that are processing your data, one of the trickiest challenges is to, is the interface between two CPUs. Right.

spk_0 Because now we need to, let's say one CPU needs data that's sort of another another CPU's memory.

spk_0 So we need to start doing some communication. Okay. Right. We do this to do something called MPI or message passing interface.

spk_0 And it can get, it can get a little bit tricky and we just need to pay attention to how we're going to do this, this communication between CPUs in an efficient manner.

spk_0 Because if you don't do it efficiently, it can actually be quite what a big bottleneck just moving the data around so that each processor has what it needs.

spk_0 I'm assuming at one point, it almost becomes worse to use a certain number of CPUs simply because of how many messages are being passed or am I wrong?

spk_0 And you can just use, and like the more CPUs, the better. Or is there like a number?

spk_0 Well, you're right that there is a overhead, let's say, in using, you know, and splitting up your job or your calculation into multiple CPUs.

spk_0 In practice, we don't usually run into that bottleneck in that.

spk_0 Anytime we throw more CPUs at it, we will get faster speed. But, but, but in principle, let's say you just had one CPU per cell, then you're having to do massive amounts of communications.

spk_0 And in that case, I would imagine that, yeah, like the message passing itself would slow you down more than splitting up the job.

spk_0 Maybe something I can't explain because I guess I also dabble in something similar. Don't really research on MHD particularly, but also just have to model my own set of equations.

spk_0 And the way that that works in parallel and obviously the same as yours is, and this is, I believe, our most parallel code works, is it, is it not multi block for like most parallel code that they usually essentially, let's say you as, as you said, you have 10 million cells instead of putting one cell obviously on each processor.

spk_0 The best way to do this is essentially to have a certain number of quote unquote blocks. So let's say you have 10 million cells. Let's say you have 10 blocks again, not very idea, but let's say you do now you have a million cells per block.

spk_0 So now you just put a block on each processor.

spk_0 Right. You know, so that's that's kind of the way we do it. Obviously, assuming that's the way you do it as well.

spk_0 Correct. Yeah. And I think that's the way that most people using parallel code with modeling simulations and meshes and whatnot would do something like a multi block mesh.

spk_0 I think multi block approaches are pretty popular. Yeah. Right.

spk_0 Because I know that it can definitely get tricky when you, so like, for example, let's say I have 10 processors and I have 10 blocks.

spk_0 And I now say use all 10 processors. That would probably, I mean, that would be more ideal than putting, for example, two blocks on each processor. Right.

spk_0 But now, let's say I ask for 20 processors and I only have 10 blocks. Right. So you can actually make use of, you can't even make use of it. You can't make use of it.

spk_0 Oh, I just thought it would be slower. Okay. So what you could do in that situation to speed up your calculation is to take the 10 blocks, split them into 20. And now you have 20 blocks. How 20 processors?

spk_0 Oh, okay. So just split the blocks. So you would have each block would have less cells per block. So essentially you would want to maximize or you would want to ideally get one block per processor. Is that correct?

spk_0 Yeah. Essentially, that, I mean, that would be the fastest calculation. And then you also have to think about how many cells you want to pack in that block. Right.

spk_0 Now I'm assuming things get a little more complicated when they're a little like uneven numbers, like if let's say they're like 22 blocks in 23. So process.

spk_0 Absolutely. Yeah. Yeah. Now you've got one processor with two blocks. How does, how does the supercomputer or sorry, not a super, how do you the coder handle something like that? Yeah. So having multiple blocks in one processor is not a problem.

spk_0 Okay. But having odd number of blocks and the like can actually make things a little bit complicated in the sense that in the logic of, okay, how are we going to split these tasks?

spk_0 We're usually actually assuming even numbers. And I think I believe our code only handles even number of of cells, for instance.

spk_0 Okay. I actually did not know this. Yeah. All right. I mean, I guess it makes sense because odd number would be, I mean, I would just say it doesn't have to be that way. We could have coded it otherwise.

spk_0 But it makes the logic a little bit simpler. All right. So I'm assuming the biggest difficulty as well with the computing is not only like the passing of the messages, but also making sure that because again, we code in C++. So it's all very memory dependent.

spk_0 We also have to make sure that the memory is allocated correctly. Now, not that I've actually dived into the message passing interface myself too much. But is there any like again for those C++ developers or any developers that are dealing with allocation of memory.

spk_0 How does that impact or how does the parallelization impact that? Yeah. So I think what you're, what you're asking me is how does realization increase or change your memory footprint?

spk_0 Yeah. Yeah. Yeah. Yeah. Essentially. Yeah. And how do you deal with that? How do you deal with right? Right. Right. Memory everywhere and putting it all in. Right. So if we had a serial program that didn't paralyze, then our memory footprint would actually be quite a bit smaller.

spk_0 And the reason for that is because we need to essentially allocate or to use a different word. We need to give space so that we can store information about our neighbors or by neighbors. I mean, the information on different CPUs.

spk_0 So if we had a serial code, we wouldn't need that memory location. But because we need to transfer messages, we actually need to dedicate a little space of memory to put the information that belongs to our neighbors so that we can use it.

spk_0 So certainly it will increase the memory footprint. So essentially each processor must have information about all the other processors. Or is it just like one processor that needs to have this information?

spk_0 So wouldn't that be more efficient? So you need information about your neighbor. So let's say let's go back to what we're talking about blocks. Right. Right. So let's say we have two blocks belonging to, let's say we have four blocks. Okay. And there's there's two processors. So I know for the audience, this can get a little bit complicated when we talk about a little bit. A lot of numbers. Let's say we have four blocks, two processors. That means that each processor has two blocks on it. Right.

spk_0 So that means that the blocks that are at an interface and by interface, I mean that one block belongs to one CPU and the other block belongs to this other CPU needs to have information about that other block so that it can essentially do calculations with its own information.

spk_0 I think I kind of that was a big confusing. I don't know if I really answered your question.

spk_0 No, I mean, I kind of I kind of get like you essentially have to save memory for just like having the fact that there are more than there's more than one block. So that memory itself will be adding to the fact that if you did it in cereal, you would need that. So you would be saving that memory. Now is adding this memory.

spk_0 Is there a point where it gets too much? Is that because I'm assuming like for every block, you know, how much memory would you really need to allocate? What if your computer is low on memory, high on power or maybe high on power, low on, or sorry, low on power, high on memory? How would it, how would it deal with that?

spk_0 Yeah, it's a good question and this is very much related to your earlier question regarding do you always benefit from the polarization and like I said, the answer is no. Yeah, because of this overhead related to storing information about your neighbors.

spk_0 True.

spk_0 Yeah. So there's a lot of there's a lot of just communication that you have to do and you have to make sure it's done correctly. That's correct. Yeah.

spk_0 Okay. Now before we get into the nitty gritty of your thesis, I do want to just close up with if you have any or sorry, not if, but I'm sure you do any challenges with the modeling of your MHD code because I would assume that or I don't know, maybe you tell me I'm wrong.

spk_0 I don't know how many how is space what I'm assuming space whether is getting popular only very recently because of all our satellites and our technology or maybe it's been popular for many years. I don't know.

spk_0 I mean, depends what you mean by many years, but our supervisors, so Ray and I both are supervised by the same by the same guy and he started his MHD work in the late, I would say in the late 90s.

spk_0 So is that recent, I'm not sure, but certainly people have been working on this for a few decades now.

spk_0 Okay, so space whether it's not new for sure, it's definitely not new and modeling it modeling is not you either. Okay, okay.

spk_0 So any new challenges that you've experienced while modeling MHD as opposed to, okay, first of all, maybe you can discuss some general challenges which is modeling.

spk_0 Let's not right. Maybe discuss HPC anymore, just modeling challenges and then any specific to MHD that you went through.

spk_0 Yeah, so I think in the context of space whether the biggest challenges are the appropriate specification of the boundary conditions, which we talked about the whole all the different models that go into it.

spk_0 And how we specify the plasma properties at the inner boundary. So that's quite a big challenge. And that's and there's still.

spk_0 There's still lots of effort that's dedicated to that. The second challenge, I would say that's more on the solved side. I would say is the.

spk_0 How do you deal with the solenoidal or the divergence free property of the magnetic field? And again, I did, I've mentioned that a little bit.

spk_0 But the the MHD the magnetic field from Gauss's law, we know cannot have any sources or sinks. Okay, that's expressed mathematically as divergence free.

spk_0 So how do we bake that in so that our code also be a respect stock property. And in the early 2000s, there was quite a bit of effort that was dedicated to dealing with this difficulty.

spk_0 That's particular to MHD and maybe you don't see in other fluid descriptions. But I would say that's pretty much handled at this point.

spk_0 So I would say now we're in the sort of let's call it the third generation of MHD modeling where we've solved the by the magnetic field divergence free property.

spk_0 We have a good grasp of the boundary conditions. Okay, or that the the data driven boundary conditions and the next frontier is really uncertainty quantification and data simulation.

spk_0 And this is where we want to not only just reduce our model, but we want to know, okay, well, how accurate is our model and we want to make use of all the observational data that we have to better inform our model.

spk_0 Okay, I think this is the perfect way to ask you because you mentioned observational data to talk about the actual nitty gritty of your thesis, which is data simulation.

spk_0 And that's actually fun fact. That's not recording anymore. That's fun fact how I got to know him because I'm currently also studying data simulation for my thesis, which I always have promised I'll make a video on that I've not done yet.

spk_0 And that's actually how I met Jose because he has dabbled in data simulation for quite a while. So maybe you can start with explaining why what is data simulation, why does it connect to MHD and I'm assuming its advantages, which is why you're using it. Why is it advantage?

spk_0 Yeah, it's a great question. So you mentioned, okay, what are some of the challenges that we deal with when we do MHD modeling and I went through that, but let's say now we have the perfect model or model behaves exactly how we would like it to even if that were the case.

spk_0 It's very likely and in fact, it is not only likely, but it occurs that that our models don't actually match the observations that we record after the fact.

spk_0 Yeah, okay. And there's different reasons for that. But one of them once we're trying to address with data simulations that there's actually a lot of uncertainty in the input that we give our models.

spk_0 So again, let's say you have a perfect model, but if you give it bad data or bad information, it'll, the questions might be correct, but you're not actually predicting what you want to predict.

spk_0 Right, or you can think of more simply garbage in garbage out. Right. Okay. And this is a problem that's not just relevant to MHD. This is a problem that's relevant to any type of predictive science. Okay.

spk_0 Okay. So any kind of, when is your predictive science? I think when a lot, I think what we've been saying in the word modeling a lot. Right. And I think maybe, I don't know if this is the perfect time to transition to the word forecasting.

spk_0 That's a good question. Yeah. Because forecast, I think, makes a lot of sense to a lot of people that are thinking about what is the point of modeling when you say, Ma, sorry, when you say modeling, what are you doing? You're most of the goal, I shouldn't say all modeling, but most of the goal is the forecast or predict what something will be best example of that.

spk_0 Let's say I want to know the temperature tomorrow. Right. And this is actually kind of important too because data simulation is used everywhere in weather prediction as well. How do we know are the weather app is accurate because of data simulation.

spk_0 So knowing the temperature tomorrow, if let's say I wish to know it exactly, no matter essentially how good my model would be, there's some information that I have to put about it today or to run the model.

spk_0 So my whole goal is essentially to predict the temperature tomorrow. Now what Jose is explaining is that data simulation. I mean, oral actually not only data simulation, but just the fact about all models is that no matter what I predict, because I did not have exactly perfect information today, because you never will.

spk_0 It won't be perfect. So let's say I predict is 17 degrees tomorrow. Let's say at 16.3 is good enough, but it's not perfect. Right. Right. So that is kind of where data simulation comes in.

spk_0 That's correct. And so we I think there's sort of two main areas where this is really, really relevant. One is and you already alluded to it is a atmospheric weather prediction. Okay. So there's actually this huge, very expansive network of observations involving millions and millions of observations that are assimilated or integrated into weather prediction models every single day.

spk_0 So this is essentially for those who are interested, we're solving these huge optimization problems every single day when we do weather prediction.

spk_0 And we say optimization problems that's basically saying like you're trying again, we haven't really gone into the nitty gritty yet, but yeah, essentially you're saying you're trying to optimize for the temperature or whatever you're trying to find, given the data.

spk_0 Exactly. Right. Exactly. Yeah. Briefly, what is what is some of this data that's generally is it just like temperature data? Is it it's wind data? I have no idea. It's a whole host. It's a whole host. It's like you said, like velocity wind is a good one temperature pressure. There's weather balloons. There's physical stations that are stationed that are, you know, fixed to the ground. There's a whole host. There's satellites that are taking radar measurements of the earth.

spk_0 And all those measurements in combination with the modeling capabilities that weather prediction centers have is what makes weather forecast work today. If you didn't have the data and you only have the model, there's no way that we could make any meaningful predictions.

spk_0 Okay. And I think that's one of the differences between weather forecast, sort of threshold or atmospheric weather forecast against space weather forecasting is that atmospheric weather forecast is highly chaotic, meaning that a very small change to the initial conditions will lead to a huge change in the output.

spk_0 So you can think some people call this the butterfly effect. Right. Now this is also the case in space weather, but in my experience, it's it's effect is much less, it's much more reduced. So it's not as chaotic. This is not a chaotic as far as I've experienced. Yeah. So data simulation. Right. You have a bunch of parameters and you have to optimize them. Explain what that means in especially in relation to MHD. How are you relating?

spk_0 Or I guess maybe I can give a little. So essentially the way that best I've understood data simulation works is that you essentially have a parameter that you wish to optimize for. And I'm going to continue with the temperature example with the weather tomorrow.

spk_0 Let's say I want to know the weather tomorrow perfectly. Now to know that weather tomorrow, I probably have to know let's say I need to know the pressure profile today. Right. Right. Let's say I need to know that like essentially my model, my model to predict the weather tomorrow to give me that result. That simulation, it needs some input data.

spk_0 And let's say one of those input data is pressure like a pressure profile. Now I put in a bogus pressure profile. It's obviously going to give me a bogus temperature result. You said garbage in garbage out.

spk_0 So my whole goal with data simulation is essentially to take observations of the maybe the pressure profile, but maybe not the pressure profile. Right. That's the biggest advantage of data simulation that you don't actually need to know exactly what you are trying to optimize for.

spk_0 So for example, let's say I don't know nothing about weather. So please do not like site any of this. But let's say the density profile was directly related to the pressure profile, which I'm sure it is.

spk_0 And let's say there was no way for us to measure this pressure. So data simulation, the advantage that it gives us is that we can take data. And as long as it's related to whatever parameter we want to optimize for, which in this case is the pressure profile for today.

spk_0 We can take that data. The goal of this data assimilation algorithm then is essentially to take that data and optimize for this pressure of today for this pressure profile today.

spk_0 Now most most people think, oh, aren't you trying to optimize for the temperature. Yes, but the whole goal is the only way that I get that temperature is by perfecting the input.

spk_0 Right. So one big thing about data assimilation that we will talk about or at least that we are using in our codes that makes the programming and the math a lot easier is we are essentially assuming a perfect model.

spk_0 We're essentially assuming that the model that we're using. That means if we give it a pressure profile today, it will give like the exact temperature that we expect.

spk_0 That's what we're assuming not that is that that is not a right assumption whatsoever. But it's significant. I mean, not whatsoever. It's definitely it's not a correct assumption, but it can be taken. It significantly reduces the mathematics required to do a lot of this.

spk_0 But essentially the idea is we're assuming a perfect model. We're obviously assuming an imperfect data set. We're assuming imperfect predictions. So we have uncertainties everywhere else except in the model. So the model essentially is assumed to be perfect.

spk_0 So we're taking this density profile. Let's say that I've recorded today and the whole goal, as I mentioned again, of the data assimilation program is essentially to optimize for the pressure for today.

spk_0 So once I feed all this density data into my program, it's going to give me some pressure profile. Now I feed this pressure profile back into my model to then predict the temperature for tomorrow.

spk_0 And now that is kind of how modern weather prediction is done. We take all this data to better predict factors that we can then put in our model to then predict what we actually want to predict.

spk_0 Yeah, right. It's like the description rate. And I think something you touched on is also very key. And that's this art of weighing the data and the model. So we have we don't know we don't we don't we're not going to take all the data points as if they're equally like like the data.

spk_0 Likely or equally good. We're going to say, okay, we know some of our measurement tools. Let's say our radar, our satellites are more accurate than some of the other ones. And we want to account for that accuracy in our measurements.

spk_0 And the same is true of our what we call the background inputs. We might think, okay, let's go in with with Rayhands example.

spk_0 We might think of well, this part of the pressure profile, we we're not sure, but we're actually pretty pretty we have some idea of what it looks like.

spk_0 But other parts of the pressure profile are maybe much more noisy. We don't really know what it looks like. So we want to bake in that information that parts of parts of our inputs are more accurate.

spk_0 We're more confident and parts of our measurements are more as well more accurate, more confident. And by, you know, making or leveraging that information, we can actually get much better results as well.

spk_0 So what is so now that we've I think understood the goal of data assimilation.

spk_0 What is your use with it? How do you use data simulation with MHD? What are you trying to optimize for? Right. And yeah, what are some challenges?

spk_0 Yeah, so with let's before I answer that directly, let's go back to to weather prediction. All right. So the goal of data simulation for weather prediction is really to come up with improved estimates of the in of the initial conditions.

spk_0 Okay, so the initial conditions is everything. Okay, now when it comes to MHD or space weather modeling.

spk_0 The initial conditions, finally enough are actually almost irrelevant. Okay, what's really important for us is actually the boundary conditions.

spk_0 And there's a mix. This makes the problem a little bit different. So what we want to do then is we want to make have these take these observations. Like, for example, like I said, from Lagrange 0.1 or perhaps other satellites and update or correct the boundary conditions that we talked about at this at 25.

spk_0 Solar radii. Right. Okay. So I guess in my example, the pressure profile example that I was giving is more related to the initial condition that, oh, this particular model, or as I said, forecast needs the pressure to start the modeling.

spk_0 Right. Starts with some pressure. And then again, this is all completely bogus. I'm just making stuff up here. So don't actually this is not how weather prediction is done. This is just an explanation for the assimilation part.

spk_0 So let's say you take some pressure profile and then you simulate it. So that is an initial condition. Now what you're explaining is essentially instead of predicting or essentially.

spk_0 So are you optimizing for the state essentially at the boundary or are you optimizing for different variables at the boundary.

spk_0 So I don't know what you mean exactly by that. Do you mean some parameters? Yeah. So are you optimizing like a certain parameter at the boundary or the entire state variable at the boundary?

spk_0 So what I'm doing in my thesis is just looking at the whole state variable at the boundary. Right. That's related. That's associated with the sun. Right.

spk_0 But there is also some work being done on perhaps some of the underlying parameters. How can we optimize or update those underlying parameters? Yeah.

spk_0 Yeah, because currently in my research, again, not that I've got into this in the podcast yet, but I will.

spk_0 I'm also doing boundary data assimilation. So even in my problem, I have the whole deal of the boundary needs to be.

spk_0 I don't know if I simulated is the right word, but optimized for. So in your case, it's if I'm not mistaken, the 25 are not boundary. That's correct. Right. So what are you up to?

spk_0 Like, okay, I didn't truly understand. Still you said you're optimizing the state variable, but maybe explain a little bit more just to somebody who doesn't really understand what that means.

spk_0 Yeah. And what observations are using? Yeah, it's a good question. So let's say let's make this very simple. Let's say we have two circles.

spk_0 One at a radius of 25 and another one at a radius of 225. Right. Okay. So the inner circle is analogous to the surface of the sun.

spk_0 And the bigger circle is analogous to the location of the earth in space relative to the side. Okay. So what we want to know is what is the density, the plasma density, what is the plasma pressure, what is the magnetic field, and what is the velocity field at this inner circle.

spk_0 Once we know that, then again, assuming that our model is perfect, which we know is not true. But once we know that information, then we can predict what the solar wind will look like at the larger circle or at the surface of the earth.

spk_0 And in terms of observations, we're mainly using observational data from a satellite called ACE, which measures all plasma properties at where we're at exact at the Lagrange point one.

spk_0 No, but where does it measure these properties at Lagrange point one at Lagrange point one. All right. All right. So essentially you're using data close to the earth. Yes.

spk_0 To optimize for parameters close to the sun. Yes, that's exactly right. That's the advantage of data simulation that you don't, especially with again, we might get into this with the method that you're using for data simulation, which is your your variational data simulation.

spk_0 Correct. And if I'm not mistaken in other methods, you're not you're not able to do that, right. That you can't optimize for a parameter that you're not measuring or am I wrong.

spk_0 So let's maybe we should just dive into this. Okay, sure. So let's let's let's break up data simulation into two main families. One, we call variational. We also sometimes call it I joined paste, but let's just call it variational for now.

spk_0 And the second one, we call it sequential data simulation. Right. Okay.

spk_0 In the in the variational approach, we it's named after the calculus of variations. So we're essentially and this is the language that Ray and I have been using. We're treating this data simulation problem as one very big optimization problem where we want to optimize for the inner boundary conditions or whatever we've gone into that quite a bit.

spk_0 Okay, with the second approach with the sequential approach, we take more of a statistical kind of framework where we say, okay, given these observations and given this model and given these uncertainties, what is the most likely state.

spk_0 Given given the data, okay, and actually we're we're also asking a question in the variation approach, but we're just doing so in a slightly different way. Now the difference is that because we can in the variational approach, we can track how does this point that's not of the surface relate to the surface of the of the, let's say the of the sun or it's not quite at the sun. Right.

spk_0 With this sequential approach, you could also ask this question, okay, but it would be a lot more complicated to essentially have this information travel back in space and time.

spk_0 So I'm not giving a good explanation because it's a little bit challenging to maybe maybe I'll give an analogy. Okay.

spk_0 So let's say our data simulation problem is very simple. Overdoing is we're throwing a ball in the air. Okay. And we have a single snapshot of that ball. Okay.

spk_0 We want to predict where it lands. So with the variational approach, what we're going to do is we're going to pose an optimization problem for let's say the initial velocity of the ball. Okay.

spk_0 And we're going to look at essentially all the trajectories of this ball could have taken and and figure out what's the most likely one by by doing this optimization problem in this sequential approach.

spk_0 What we're doing is, okay, given this snapshot right now and given my that my model tells me the position of the ball at this at the same time when the observation was taken, we're going to now update our model so that it better matches the observations.

spk_0 And then continue from there from there. Right. So just again to nail down on the difference the variation approach is really just optimizing for the initial conditions or let's say boundary conditions.

spk_0 Well, the sequential approach is is changing your state at the point of where when the observation was taken. Is that the way we explain it.

spk_0 It seems like sequential is far more is far more accurate, not accurate is wrong word, but far more useful because you can predict it at every state. Am I wrong in making that assumption or what are the advantages and disadvantages of each method.

spk_0 Yeah. So I would say that it's not always clear which method will win out and you kind of have to do the difficult work of really quantifying the merits of each four year problem.

spk_0 Turns out that what we've done this a little bit for space weather and it seems to show that the variational approach will be much better.

spk_0 Okay. Now both methods are using the same amount of information. Okay. So let's say we have 10 snapshots of this ball.

spk_0 We're going to use all 10 snapshots to now reconstruct what the initial velocity of the ball was when I was thrown with this sequential approach.

spk_0 At each time that the observation is taken we're going to update the state. So but at the end of that simulation or that forecast.

spk_0 Both the sequential approach and the variation approach used all 10 observations to update its information.

spk_0 Yeah. Okay. So that's all right. But is it a computational advantage? I still don't truly understand.

spk_0 So I get it that it's problem dependent and you see advantages in both. But is there just like a specific type of problem where you just can't use one method over the or like you just cannot use one.

spk_0 I would say there's not a problem where you cannot use one. I think there's there's problems where it's a lot easier to use one or the other.

spk_0 And I think for the types of problems that we're doing where we're really just interested in boundary data. The variational approach is easier in that sense because we with the sequential approach you would essentially need a way to track information from the observation location to the boundaries.

spk_0 And with the variational approach we actually get this type of information transport through something called the adjunct equations, which I don't really want to get into because it's quite complicated.

spk_0 But it's related to the characteristics of our PDs. All right. Okay. That was actually a nice. I think I think I really like that last line where you explained.

spk_0 Where like how the observation is transporting to our boundary or whatever our thing of choice. Now let's say I wanted to know the state in the middle. I'm assuming variational is useless.

spk_0 So what do you mean by the state in the middle? That was not a very good question.

spk_0 Let's say I wish to optimize. Let's say I have an unsteady problem. In this case, this is referring to a time evolving system. Let's say I wanted to know the state at every point in time. And I wanted to optimize for that state in every point in time.

spk_0 Right. I'm assuming the variational would not be very good there. No, so the variational I would disagree. The variational approach is is doing that actually is doing that as well.

spk_0 So maybe like let's go a little bit deeper than so the way that we construct this. We've talked a lot about constructing an optimization problem. Right. So what we're doing is that let's say we have let's stick to our ball analogy.

spk_0 So let's say we have 10 snapshots of the ball. Now we're going to construct this function. That's going to tell us how good or bad our model is. And all we can think of is very simple.

spk_0 So what we're doing is we're going to just take the difference between our model and the observation and square it. Okay. And this this function is updated with time. So if we have 10 observations, we're going to update this function 10 times.

spk_0 Right. Okay. So now what we're doing is we want to minimize this essentially this misfit function by choosing the best parameters so that our model best matches the data.

spk_0 Now let me explain briefly because the misfit function was a good use of word there. I really like that word never never heard a misfit function. Right. I really like that.

spk_0 Because essentially what the square is doing for those of you that I'm sure have seen like an L1 norm or an L2 or I guess an L2 norm error is essentially I don't know if it's the right term, but it kind of grabs the error of your state versus what you already have.

spk_0 So let's say you're seeing the difference between the system and the observation and you square it. That's essentially an analogy for trying to get the error estimate for how off either one of them are.

spk_0 Yes. And what you're now saying is essentially we wish to minimize that exactly right. So in the best case scenario, let's say we picked the most optimal.

spk_0 The most optimal parameters then our model would perfectly match the observations. Now we won't actually see this in practice because observations have noise.

spk_0 So there's always going to be a misfit, but we like to minimize that misfit is best possible.

spk_0 So now kind of getting it back into image D and observations. What are so you said you mainly use ace for solar.

spk_0 Sorry, was it not surface observations? It was more like so ace is measuring the properties. Yeah, I know location at L1.

spk_0 Any other observations that you can also add in there and maybe take advantage of because I mean, I think somewhere in this podcast we mentioned that there are many, many, many, many satellites.

spk_0 So unfortunately, there's not many, many, many. Unfortunately, there's only a few.

spk_0 Oh, few. Alright. So I think you were talking about many, many when you were talking about the weather.

spk_0 That's right. Sorry. Sorry. My apologies. No, no, that's. So you've got a few satellites. Yeah. So I think ace is one of the primary ones that we want to certainly look at.

spk_0 And two other satellites are the stereo satellites and actually one of them is not an operation anymore, but these are stereo A and B.

spk_0 And these are located. You can not there. They are at a sort of an L1 distance, but they have a different sort of angle, angular distance.

spk_0 So one would be like a head of the earth in terms of the angle. And the other one would be further than the earth in terms of the angle.

spk_0 Alright. So how do you account for that difference when you assimilate stuff? Yeah. So thinking back to our misfit function or misfit function just cares about the difference between the model and the data.

spk_0 And it's going to grab the model data at the right location so that it matches with the location of the satellites.

spk_0 Yeah. Has there ever been a situation because you said there were few satellites where satellites don't agree with each other. And there's like a little bit.

spk_0 Yeah. Right. So how do you deal with that in data assimilation? So I would say there's sort of two sort of parts to this answer. One would be, okay, well, there might be a bias in the measurements. Right.

spk_0 And certainly we don't want that because a bias is really hard to account for. Then there could also then there's also random noise. Right. So when we compare two satellites, two measurements of anything, it doesn't have to be space related.

spk_0 Right. When we compare two measurements of anything, almost certainly there's going to be noise that's going to cost these observations to be different.

spk_0 So accounting for that noise and weighing which one you prefer more is kind of what we talked about where we want to give more weight to the ups to the measurements that are more accurate.

spk_0 Right. Right.

spk_0 And then the second part of that would be that well, these measurements might be taking place at different points in space and time. Right.

spk_0 So they don't necessarily actually need to agree perfectly with each other. Right. And this and because they're located differently in space and time, we can actually account for this difference with our model.

spk_0 Okay. Right. Right. Because, okay, that makes sense.

spk_0 Talking about data simulation, the first thing that got me kind of excited about it was that I can convince an employer that it's basically machine learning.

spk_0 No, obviously I'm kidding. But I mean, it basically is in a very low level.

spk_0 I guess question number one can be, is there a difference? I mean, I'm not that you know you're a machine learning expert, obviously not. But like what kind of description can does this or similarities does this have with machine learning?

spk_0 And if not or just to add to that question entirely, can machine learning be used for data simulation for optimizing these parameters?

spk_0 Yeah, these are great questions. So I would say there's actually quite a sizable similarity between machine learning approaches and data similar like more traditional data simulation. Right.

spk_0 So because machine learning is such a wide term, I'm going to just focus on neural networks for this answer. Okay. And I think that's what I have the most familiarity with. So if you were to take a neural network or let's say, let's say you were to take my my MHT model, throw away the MHT throw away the physics and replace it with a neural network.

spk_0 And you use the same misfit function and you try to train or usually the language that uses train, but I say you were to try to optimize the parameters of your neural network so that it matches the data well.

spk_0 Yeah. Now we call this machine learning. Right. Right. Now let's remove the neural network throwing physics. We're still training in the sense that we're finding these optimal parameters to best fit the data.

spk_0 So we can think of data simulation as a much more physics, physics informed or physics heavy form of machine learning. If that makes sense. No, it does. All right.

spk_0 So simply because of the addition of physics is what makes the data simulation kind of stand out for machine learning.

spk_0 Yeah. Like for example, what they use in neural networks. And again, I'm not a machine learning expert, but I've read a little bit about it. And one of the main algorithms they use for training their neural networks is something called back propagation where they're essentially taking differences between their output and their prediction and the data.

spk_0 And then they're tracking how that information is related to the inputs. That's basically an easier version of what we call the adjunct method.

spk_0 Okay. So the adjunct method that we use to calculate sensitivities or great or model gradients is essentially just a chain rule, but applied to our to our physics.

spk_0 Whereas in the neural network approach is the chain rule applied to the neural network parameters themselves. All right.

spk_0 So there's a second question. How does machine learning or all of artificial intelligence can it help doing? Now, I know you mentioned that replace this with the neural network. It's machine learning.

spk_0 Has anybody tried that? Is there is there because I mean, is there I mean that's a separate question, but is there any.

spk_0 Forgetting the word for it, but like any push towards trying machine learning algorithms or artificial intelligence algorithms with MHT.

spk_0 Yeah, I think I think right now there's a big push in the scientific community. I'd large to try to use machine learning approaches for all types of science questions.

spk_0 And I and I welcome that I think I think there's room for it and it could certainly help us quite a bit.

spk_0 I'm not sure if I'm answering your question fully, but in terms of sort of plausible areas that I could see this helping a lot with.

spk_0 I would say one that I'm thinking of is using Bayesian optimization, which some people would sort of put in the camp of machine learning to replace.

spk_0 The the Washington optimization that we use. So what we do in sort of what I've done in my thesis is just use sort of pre-canned optimization packages that are essentially just rely on Newton's method.

spk_0 But you can actually replace these optimization techniques with more machine learning, focus techniques that essentially build a surrogate.

spk_0 But so you're essentially talking about replacing the technique to do the data simulation with machine learning.

spk_0 I'm saying scrap that entirely. What have you just like has there been any push towards just using data. I mean just using machine learning towards like MHT or just using artificial intelligence or are the model back approaches.

spk_0 Yeah, so I haven't seen that as much because it's a challenging problem. I think you could maybe someone could do some research on starting just from like let's say solar flare data or just observations of the sun and going all directly to okay, what are the conditions at earth.

spk_0 And I think that's actually been tried, but I would be a little bit skeptical of those techniques.

spk_0 All right, all right. Maybe as as like a.

spk_0 Closer I can might maybe ask you what is your opinion in general about artificial intelligence in what we like to call CFD codes or computational fluid dynamics which is kind of the field in which we're both dabbling in.

spk_0 What do you think about artificial intelligence and machine learning kind of coming in there is it do you not like the lack of physics based modeling or are there advantages maybe computationally.

spk_0 Yeah, so from the work that I've seen so I would just base my answer. Okay, what does what does the science tell us tell us? Okay, what what are the experiments show and people have shown that if you throw away the cost for training.

spk_0 Then using machine learning approaches can significantly accelerate CFD but the training is the problem now the training is a problem that's not always necessarily doesn't mean that we should throw away this technique right so there are certain circumstances where your problem looks very similar like let's say you need to solve the same problem many times with slightly different variations.

spk_0 Maybe this is an avenue where machine learning can really help you out right because you might it might be expensive to train the model at first but then when you just need to reuse it for slightly different variations then that then it'll be be very performance right so I would just like to highlight that.

spk_0 From my experience from the work that I've seen machine learning is very I would think of it as very very very good at interpolating yes completely trash at extrapolating so if you need to extrapolate you need to use real physics if you need to interpolate I think machine learning could be a good avenue.

spk_0 So my next question that I think now you just answered with that was going to be oh we have so much data for terrestrial weather why don't we use artificial intelligence models for that but now you just said that it's not very good for extrapolating.

spk_0 In fact it actually has been done because I would just think with terrestrial weather there's just so much information that it's just got to perform at least somewhat good.

spk_0 So there's I saw some recent papers this summer about exactly that so essentially replacing numerical weather prediction with machine learning okay and they've got some very encouraging results and I encourage yeah to look further and is that mainly just because of the amount of data available.

spk_0 I think that's part of it I think part of it is that I think the argument that they make is that we've seen the earth's weather enough times that we can now start interpolating rather than extrapolating.

spk_0 Yeah that's okay yeah and so for us it's a bit scary because there are maybe these technologies will take our will render our techniques useless I don't think so but it's it's exciting and it is different.

spk_0 Artificial intelligence the new leaf you know yeah I've been definitely thinking about it in CFD for a while in like how it can impact any of us really in our jobs yeah but as you very well said

spk_0 at least any time in the near future I don't really seeing it taking up physics based models especially for physics based modeling yeah yeah yeah so I definitely see a little bit of there but I can also see in a lot of just numerical methods in general where physics the physics of it may not be that important where artificial intelligence could definitely shine not that I know specific examples of this but just from a little friends of mine that are not in the world.

spk_0 So I think that's not necessarily doing highly physics based modeling maybe artificial intelligence could help them maybe here or there you know especially one of the biggest challenges that we hear with CFD engineers is we're not challenges but like I guess annoyance is like mesh at attention and stuff like that right so maybe like a lower level task could be a given off to a machine learning algorithm and then the actual physics based calculations you can just do yourself yeah yeah so there's a lot of avenues for the future of where this whole

spk_0 new era will take us hopefully we'll keep our jobs but I'm very excited maybe just a close up do you want to maybe talk about any future aspirations for what you maybe want to do with this current PhD any maybe you want to continue in MHD maybe you don't anything that you see that the MHD feel particularly space weather is going to do in the future anything coming up right so I think in general the way that I would like to steer my career is to stick to the future.

spk_0 With this dual approach of let's take real observations and let's take advanced model techniques and let's marry them right so I want to stick with that and if I can that be great I think I would like my sort of goal or or the direction I like to take is to become a faculty at a Canadian university as we all know that's quite a quite a big goal so we'll see how it goes right.

spk_0 But yeah if I could if I were to sell you know be selling my research program or advertising my research program to do a faculty hiring committee I would frame it in the sense of both observational data and numerical modeling used for predictive science.

spk_0 All right so where do you see the future of space weather right so sorry I forgot to answer that.

spk_0 I think right now because we're in near solar maximum there's going to be more funding and more research conducted I fear that in the somewhat near future let's say five years from now we're going to be approaching more of a solar minimum so there might be less funding in in modeling capabilities saying that I think every single day or every single year we rely more and more and more on the infrastructures that can be affected by space weather phenomena.

spk_0 Right like we use GPS systems more and more like we rely on electricity more and more not just in the first world countries but in other countries as well.

spk_0 So I think if we sort of think at a wider scale I think predicting space weather will become more and more relevant my time goes on.

spk_0 Right well thank you so much for coming on any last remarks you might want to make to the lovely audience that we have any final quotes.

spk_0 Perhaps in the in the spirit of one of the things that came up we talked about senior PhD students asking questions and still not having the answer I would just encourage the listeners to to maintain a spirit of curiosity and to study hard and learn a lot of math.

spk_0 Oh that's a lovely way to finish it all right definitely agree with the curiosity part and the math part I guess not enjoyable for everyone I did I hope everyone does as well I think through this podcast we've communicated that we've been doing.

spk_0 Love math but it's kind of important for every scientist at least in our field but yeah it was lovely having you on I think we've all learned a lot about I mean I've learned a lot about my own topic today I think so that's been lovely and space weather I think is one of those niches that a lot of people reading it would kind of be like oh my that sounds cool and then would kind of get into it so I'm glad we had you talk about that as well.

spk_0 Thank you for coming on and take care everyone have a lovely day this has been the math and physics podcast episode number 122 my name is Ray and we'll see you soon peace.

Episode #122 - What's the Weather in Space? w/ Jose Arnal

Interactive Transcript

Topics Covered

Related Episodes

Reimagining Midlife: My 2x20 Project™ Update & the Power of Experiments

Stranger Things Was REAL: Inside the Montauk Project

the magician who invented the modern world: john dee and the lost birth of science

Essentials: Erasing Fears & Traumas Using Modern Neuroscience

Share Episode