36C3 - Grow your own planet

36c3 preroll music

Herald: Ok, I have to say, I'm always
deeply impressed about how much we already

learned about space, about the universe
and about our place in the universe,

our solar system. But the next speakers
will explain us how we can use

computational methods to simulate the
universe and actually grow planets. The

speakers will be Anna Penzlin (miosta).
She is PHC student in computational

astrophysics in Tübingen and Carolin
Kimmich (caro). She is a physics master's

student at Heidelberg University. And the
talk is entitled "Grow Your Own Planets

How Simulations Help us understand the
universe." Thank you!

applause

caro: So hi, everyone. It's a cool
animation right? And the really cool thing

is that there's actually physics going on
there. So this object could really be out

there in space but was created on a
computer. So this is how a star is

forming, how our solar system could have
looked like in the beginning. Thank you

for being here and that you're interested
in how we make such an animation. Anna and

I are researchers in astrophysics. And
we're concentrating on how planets form

and evolve. She's doing her PHD and in
Tübingen and I'm doing my masters in

Heidelberg. And in this talk, we want to
show you a little bit of physics and how

we can translate that in such a way that a
computer can calculate it. So, let's ask a

question first. What is the universe or
what's in the universe? The most part of

the universe is something we don't
understand, yet. It's dark matter and dark

energy and we don't know what it is, yet.
And that's everything we cannot see in

this picture here. What we can see are
stars and galaxies, and that's what we

want to concentrate on in this talk. But
if we can see it, why would we want to

watch a computer? Well, everything in
astronomy takes a long time. So each of

these tiny specs you see here are galaxies
just like ours. This is how the Milkyway

looks like. And we are living in this tiny
spot here. And as you all know, our earth

takes one year to orbit around the sun.
Now, think about how long it takes for the

sun to orbit around the center of the
galaxy. It's four hundred million years.

And even the star formation is 10 million
years. We cannot wait 10 million years to

watch how a star is forming, right? That's
why we need computational methods or

simulations on a computer to understand
these processes. So, when we watch to the

night sky, what do we see? Of course we
see stars and those beautiful nebulas.

They are a gas and dust. And all of these
images are taken with Hubble Space

Telescope. Oh, so there's one image that
does belong in there. But it looks very

similar, right? This gives us the idea
that we can describe the gases in the

universe as a fluid. It's really
complicated to describe the gas in every

single particle. So, we cannot track every
single molecule in the gas that moves

around. It's way easier to describe it as
a fluid. So remember that for later, we

will need that. But first, let's have a
look how stars form. A star forms from a

giant cloud of dust and gas. Everything
moves in that cloud. So, eventually more

dense regions occur and they get even
denser. And these clams can eventually

collapse to one star. So, this is how a
star forms. They collapse due to their own

gravity. And in this process, a disc
forms. And in this disc, planets can form.

So why a disc? As I said, everything moves
around in the cloud. So it's likely that

the cloud has a little bit of an initial
rotation. As it collapses, this rotation

gets larger and faster. And now you can
think of making a pizza. So when you make

a pizza and spin your dough on your
finger, you get a flat disc like a star,

like a disc around a star. That's the same
process, actually. In this disc, we have

dust and gas. From this dust in the disc
the planet can form. But how do we get

from tiny little dust particles to a big
planet? Well, it somehow has to grow and

grow even further and compact until we
have rocks. And even grow further until we

reach planets. How does it grow? Well,
that dust grows we know that. At least

that's what I observed when I took those
images in my flat. Well, so dust can grow

and grow even further and compact. But
when you take two rocks, we're now at this

in this stage. When you take two rocks and
throw them together, you don't expect them

to stick, right? You expect them to crash
and crack into a thousand pieces. So,

we're standing on the proof that planets
exist. How does this happen? And it's not

quite solved yet in research. So, this is
a process that is really hard to observe

because planets are very, very tiny
compared to stars. And even stars are only

small dots in the night sky. Also, as I
said, planets form in a disc. And it's

hard to look inside the disc. So this is
why we need computation to understand a

process that how planets form and other
astronomical processes. So let's have a

look at how this simulated on a computer.
miosta: OK. So, somehow we have seen

nature. It's beautiful and it's just like
a tank of water and a bubbly fluid we

already have. So, now we have this bubbly
fluid and here in the middle demonstrated.

But now we have to teach our computer to
deal with the bubbly fluid. And that's way

too much single molecules to simulate
them, as we already said. So there are two

ways to discretize it in a way that we
just look at smaller pieces. One is the

Lagrangian description, just like taking
small bubbles or balls of material that

have a fixed mass. They have a certain
velocity that varies between each particle

and they have, of course, a momentum
because they have a velocity and a mass.

And we've created a number of those
particles and then just see how they move

around and how they collide with each
other. That would be one way. And that was

described last year in a very good talk. I
can highly recommend to hear this talk if

you're interested in this method. However,
there's a second way to also describe

this. Not just going with the flow of the
particles, but we are a bit lazy, we just

box it. So we create a grid. And as you
see down here in this grid, you have the

certain filling level, a bit of a slope.
So, what's the trend there? And then we

just look for each box, what flows in what
flows out through the surfaces of this

box. And then we have a volume or a mass
filled within this box. And this is how we

discretize what is going on in the disc.
And actually, since we are usually in the

system of a disc, we do not do it in this
nice box way like this. But we use boxes

like those because they are already almost
like a disc and we just keep exactly the

same boxes all the time and you just
measure what goes through the surface in

these boxes. So, this is how these two
methods look like if you compute with both

of them. So, one was done by me. I'm
usually using this boxing method and the

other was done by my colleague. You see
this like when you look at them, at the

colors, at the structure here, you have
the slope inwards, you have the same slope

inwards here. You have even this silly
structure here. The same here. But what

you notice is you have this enlarge dots
that are really the mass particles we saw

before, these bubbles. And here you have
this inner cutout. This is because when

you create this grid, you have the very
region at the inner part of the disc where

the boxes become tiny and tinier. And
well, we can't compute that. So, we have

to cut out at some point in inner part So, here
when you go to low densities, these

bubbles blow up and distribute their mass
over a larger area. So, it's not very

accurate for these areas. And here we have
the problem we can't calculate the inner

area. So both methods have their pros and
cons. And are valid. But now, for most we

will focus on this one. Just so we have
this nice stream features. So, again,

going back to the boxes, we have to
measure the flow between the boxes. This

flow, in physics we call it flux, and we
have a density row one, density row too.

And the flux is the description of what
mass moves through the surface here from

one box to the next. So, if we write this
in math terms, it looks like this. This

says the time derivative of the density,
meaning the change over time. So how much

faster or slower, the velocity would be a
change in time. And then this weird

triangle symbol it's called nabla is a
positional derivative. So, it's like a

slope. So, how do we change our position,
actually. So, if we change, look at the

density over time, it should correlate to
what inflow we have over position. That is

what that says. So and then we have in
physics a few principles that we have

always to obey because that is just almost
common sense. One of them is, well, if we

have mass in a box. Well, like this, the
mass should not go anywhere unless someone

takes it out. So, if we have a closed box
and mass in that box, nothing should

disappear magically. It should all stay in
this box. So, even if these particles jump

around in our box with a certain velocity,
it's the same number of particles in the

end. That's again, the same equation just
told in math. So, a second very

rudimentary principle is if we have energy
in it, in a completely closed box. So, for

example, this nice chemicals here and we
have a certain temperature. So, in this

case, our temperature is low, maybe like
outside of around zero degree Celsius. And

then we have this nice chemicals down here
and at some point they react very heavily.

We suddenly end up with much less chemical
energy and a lot more thermal energy. But

overall, the complete energy summed up
here, like the thermal and the chemical

energy, also the energy of the movement
and the energy of potential added up to

this variable "U". That should not change
over time if you sum up everything.

Because our energy is conserved within our
clothed box. And then the third thing is I

think you all know this. If you have like
a small mass with a certain velocity, a

very high velocity in this case and it
bumps into someone very large, what

happens? Well, you get a very small
velocity in this large body and the

smaller mass stops. And the principle here
is that momentum is conserved, meaning

that the velocity times the mass of one
object is the same as then later for the

other one. But since it's larger, this
product has to be the same. That doesn't

change. And we have also in our
simulations to obey these rules and we

have to code that in so that we have
physics in them. So you say, ok, this is

really simple, these rules, right? But
actually, well, it's not quite as simple.

So, this is the Navier-Stokes equation, a
very complicated equation is not

completely solved. And we have here all
that is marked red are derivatives. Here

we have our conservation law that was the
nice and simple part. But now we have to

take other physical things into accounting
for pressure, accounting for viscosity,

for compression. So squeezing. And like
how sticky is our fluid? And also gravity.

So, we have a lot of additional factors,
additional physics we also have to get in

somehow. And all of these also depend
somehow on the change of position or the

change of time. And these derivatives
aren't really nice for our computers

because they well, they don't understand
this triangle. So, we need to find a way

to write an algorithm so that it can
somehow relate with these math formula in

a way that the computer likes. And one of
the way to do this is, well, the simplest

solution actually is just we say, OK, we
have now this nasty derivatives and we

want to get rid of them. So, if we look
just at one box now and we say that in

this box, the new value for the density in
this box would be the previous density,

plus the flux in and out times the time
stepover which we measure this flux,

right? So, and we have to somehow get to
this flux and we just say, OK, this flux

now is if we start here and the slope of
this curve, the trends so to say, where

this curve is going right now, it would
look like this. So, in our next step, time

step, we would have a density down here.
And well, then we do this again. We again

look at this point, where's the trend
going, where's the line going? And then we

end up here. Same here. So, again, we just
try to find this flax and this is the

trend at this position in time. So, this
goes up here. And then if we are here now,

look at this point, it should go up here.
So this is what our next trend would be.

And we do this over all the times. And
this is how our simulation then would

calculate the density for one box over a
different time steps. So, that kind of

works. So, the blue curve is the
analytical one, the red curve, well it

kind of similar, it works. But can we do
better? It's not perfect, yet, right? So,

what we can do is we refine this a bit,
taking a few more steps, making it a bit

more computationally heavy, but trying to
get a better resolution. So, first we

start with the same thing as before. We go
to this point, find the trend in this

point. That point like the line would go
in this direction from this point. And

then we go just half a step now. Sorry!
And now we look at this half a step to

this point now. And again, the same
saying, OK, where's the trend going now?

And then we take where this point would go
and added to this trend. So that would be

that. The average of this trend, of this
exact point and this trend, this dark

orange curve. And then we go back to the
beginning with this trend now and say this

is a better trend than the one we had
before. We now use that and go again and

search the point for half a time step. And
then again, we do the same thing. Now we

again try to find actually the trend and
average it with the arrow before. So it's

not exactly the trend. It's a bit below
the trend because we averaged it with the

arrow before. And now we take this
averaging trend from the beginning to the

top like this. Okay. This is already quite
good, but we can still do a little bit

better if we averaged with our ending
point. So, we go here, look, where is the

trend going that would go quite up like
this and we average this and this together

and then we end up with a line like this.
This is so much better than what we had

before. It's a bit more complicated, to be
fair. But actually it's almost on the

line. So, this is what we wanted. So, if
you compare both of them, we have here our

analytical curve. So, over time in one
box, this is how the densities should

increase. And now with it both of the
numerical method, the difference looks

like this. So, if we have smaller and
smaller time steps, even the Euler gets

closer and closer to the curve. But
actually the Runge Kutta this four step process

works much better and much faster.
However, it's a bit more computationally

and difficult.
caro: When we simulate objects in

astronomy, we always want to compare that
to objects that are really out there. So,

this is a giant telescope, well consisting
of a lot of small telescopes. But they can

be connected and used as a giant telescope
and it takes photos of dust in the sky.

And this is used to take images of discs
around stars. And these discs look like

this. So, these images were taken last
year and they are really cool. Before we

had those images, we only had images with
less resolution. So, they were just

blurred blobs. And we could say, yeah,
that might be a disc. But now we really

see the discs and we see rings here, thin
rings and we see thicker rings over here.

And even some spiraly structures here. And
also some features that are not really

radial symmetric like this arc here. And
it's not completely solved how these

structures formed. And to find that out a
colleague of mine took this little object

with the asymmetry here. And so, this is
image we just saw. And this is his

simulation. So, this is how the disc
looked like in the beginning, probably.

And we put in three planets and let the
simulation run. And so, what we see here

is that the star is cut out as Anna said.
So, the grid cells in the inner part are

very, very small. And it would take a long
time to compute them all. So, that's why

we're leaving out that spot in the middle.
And what we see here is three planets

interacting with the material in the disc.
And we can see that these planets can make

this thing here appear so that in the end
we have something looking very similar to

what we want to have or what we really
observe. So, we can say three planets

could explain how these structures formed
in this disc. It's a little bit

elliptical, you see that. That's because
it's tilted from our side of line. It

would be round if you watched at it face
on. But it's a little bit tilted. That's

why it looks elliptical.
miosta: So, we already saw we can put

planets in the gas and then we create
structures. One very exciting thing that

we found in the last year - or two years
ago it started but then we found more - is

this system PDS 70. In this system, for
the very first time, we found a planet

that was still embedded completely within
the disc. So, the gas and dust. Usually,

because the gas and dust is the main thing
that creates this signal of some radiation

because of heat. We only observe that and
then we can't observe the planet embedded.

But in this case, the planet was large
enough. And in the right position that we

actually were able to observe some
signature of accretion on this planet that

was brighter than the rest of the disc.
And then later, just this year, just a few

months ago, we actually found out well,
this is not the only object here. This is

very clearly a planet. But actually,
like this spot here is also something. So,

we can see it in different grains. Every
picture here is a different set of grains

observed. And we can see
this in five different kinds of

observations. So, there is a planet here.
And then there is also something we don't

know what it is yet, but its point like
and actually creates the feature that we

reproduce in different kinds of
observational bands or different kinds of

signals here. This is very interesting.
For the first time, we actually see a

planet forming right now within the disc.
And so a colleague of mine also is very

interested in the system and started to
simulate how do two planets in a disc

change the dynamics of a disc? So here we
have, of course, this disc is again tilted

because it's not phase on, it's like 45
degrees tilted, not like this, but like

this. And so he had it face on. This is
what a simulation looks like. So, there

are two planets: these blobs here, again,
as in this simulation. Here we have a

close up. You can actually see this little
boxes are actually our simulation boxes in

which we have our own densities. And then
he just looked at how the planets would

change the structure and the gas and also
how the gas would interact with the

planets, shifting them around. And it's
interesting. So, the planets tend to clear

out an area, open a gap, and within the
disk, that block has a lot of gas around

here. So, you have the brighter ring here
again and then clearing out more and more.

And at some point in the simulation you
saw they get a bit jumpy. So it's very nice.

You also see that planets induce in the
whole disc some kind of features like

spiral features. And so a single planet
will change the symmetry and the

appearance of a whole disc.
caro: So, the reason why the planet is

staying at this point is because we're
rotating with the planet. So it's actually

going around the disc, but the like camera
is rotating with the planet. So, it's

staying at that fixed place we put it in.
miosta: Exactly. But there's more because

as I already said, in the Navier-Stokes
equation, we have a lot of different kinds

of physics that we all have to include in
our simulations. One of the things, of

course, is we maybe don't have just a star
and a disc. We have planets in there and

maybe two stars in there. And all of these
larger bodies have also an interaction

between each other. So, if we have the
star, every planet will have an

interaction with the star, of course. But
then also the planets between each other,

they have also an interaction, right? So,
in the end, you have to take into account

all of these interactions. And then also
we have accretion just looking like this.

So, accretion means that the gas is bound
by some objects. It can be the disc, the

planet or the star that takes up the mass,
the dust or the gas and bounce it to this

object. And then it's lost to the disc or
the other structures because it's

completely bound to that. So, the
principle of this would be the simulation

I did last year and published, we have
here a binary star. So, these two dots are

stars. I kind of kept them in the same
spot. But every picture will be one orbit

of this binary, but since we have
interactions, you actually see them

rotating because of the interactions, with
each other. And then also we have here a

planet and here a planet. And the
interesting thing was that these two

planets interact in such a way that they
end up on exactly the same orbit. So, one

star's further out, the orange one, and then
very fast they go in. And they end up on

exactly the same orbit. If it now play nicely.
So, another thing is with the accretion here,

we actually see clouds from above dropping
down onto the new forming star here. So,

all of this, what you see here would be
gas, hydrogen. And it's a very early phase

so that disc is not completely flat. It
has a lot of material. And then we

actually have this infall from above
towards the star and then the star keeps

the mass. And we have to take this also
into account in our simulations. Another

thing we have to take into account up till
now, we just cared about masses and

densities. But of course what we actually
see is that stars are kind of warm,

hopefully. Otherwise, temperatures here
would also not be nice. And different

chemicals have different condensation
points. And this is also true in a system.

So, we start with the start temperature at
the surface of the star. We have a

temperature around 4.000 Kelvin. And then
we go a bit into the disc. And there is a

point where we for the first time reach a
point where we have any material at all.

Because it starts to condensate and we
actually have something solid like iron.

For example, at a 1500 Kelvin. And then if
we go further in, we reach a point where

we have solid water and this is at 200
Kelvin. This is what we then would need

actually to have a planet that also has
water on it. Because if we don't get the

water in the solid state, it will not fall
onto a terrestrial planet and be bound

there, right? So, this is important for
our Earth, actually. And then if we go

even further out, we have also other gases
condensating to solids like CO2 or methane

or things like that. And since we only get
water on a planet when we have a

temperature that is low enough so that the
water actually forms is solid and it's

important for us to think about where that
is in our forming disc. Where do we start?

We have a planet like Earth that could
have some water, right? But it's not just

the simple picture, where we have all these
nice ring structures, where we have a clear

line. Actually, it gets more complicated
because we have pressure and shocks. And

thermodynamics is a lot like pogo dancing,
right? You crash into each other. And it's

all about collisions. So, the gas
temperature is determined by the speed of

your gas molecules. Like you bouncing and
crashing into each other, exchanging

momentum. So, there's two ways to heat up
such dance. First thing is you get a large

amount of velocity from the outside like a
huge kick, a shock into your system. A

second way would be if we have a higher
pressure, like more molecules, then also

you of course have more collisions and
then a higher temperature. So, if you

change - because you have a planet now in
the system - the pressure at some point,

you actually get a higher temperature. So,
that is not then we lose this nice line

because suddenly we have different
pressures at different locations. And a

colleague of mine also simulated this.
So, this is the initial condition we

just assumed: OK, if we have no
disturbance whatsoever, we have our nice

planet here at 1au. So, same distance as
earth to the sun. Here, too. But here we

assume that less heat gets transferred
from the surface of the disc. And here we

have the planet far out like Jupiter or
something. And now we actually let this

planet change the structure of the disc.
And what happens is - we found these spirals

and within these spirals, we change
pressure. And with this actually, if you

see this orange, everywhere where it's
orange it's hotter than the iceline. So,

we don't have water where it's orange. And
where it's blue we can have water. And the

interesting thing is, even if we put a
planet out here like Jupiter, we still

form these regions in the inner system
where we have less water.

caro: One problem in astrophysical
simulations is that we don't always know

how to shape our boxes or how small these
boxes have to be. So, we use a trick to

reshape the boxes as we need them. It's
called adaptive mesh. And this is a

simulation of the red fluid flowing in
this direction and the blue fluid in the

other one. So, at the boundary, the two
fluid shear and they mix up somehow and we

don't know how in advance. So, we start a
simulation and as the simulation starts,

we reshape those boxes here. So, in the
middle we don't need much. We reshape

because it's not that complicated here.
It's just the flow. But at the boundary we

see those mixing up of the two fluids. And
so, we reshape the cells as we need them.

This is done in a program, in an
astrophysical program called AREPO. We

will later show you some more programs to
use for simulations. But another

simulation I want to show you first is
also done with AREPO and it's a simulation

of the universe. So, from here to here,
it's very big. It's 30 million light

years. So each of these dots you see here
is the size of a galaxy or even more. And

here you can actually see that at some
regions it's very empty. So, we're

rotating around this universe, this
simulated universe here. And these regions

here are empty. And we don't need a lot of
boxes there. The big boxes are enough

here. But in this dense regions where we
have a lot of material, we need smaller

boxes. And this method I showed you where
we reshape the boxes as we need them is

used for this simulation.
miosta: So, actually, you see the

beginning of the universe there.
caro: Yes!

miosta: Basically, the initial mass
collapsing to the first galaxies and first

supernovae starting. Very beautiful
simulation.

caro: So, there are different programs, as
I already mentioned, in astrophysics.

Three of them, those three are all open
source, so you can download them and use

them on your own machine, if you like. But
there are more, a lot more. Some of them

are open source, some of them are not.
Sometimes it's hard to get them. In the

following, we will present the tool
FARGO3D and PLUTO in a detailed version or

a more detailed vision than AREPO
because we usually use those two for our

simulations. What I want to show you with
this slide is that depending on what you

want to simulate, you need to choose a
different program. And one thing is that

in astrophysics we sometimes call the
whole program code. So, if I use the word

code. Sorry about that. I mean, the whole
program. So, let's have a look at FARGO3D.

It's a hydro dynamics code and what you
see here is an input parameter file. There

you define how the disc looks like. How
much mass does it have? How big is it? And

what planet? So, here at Jupiter, do you
see that? Jupiter is put in. And we also

define how big our boxes are. This
program is written in C, which is quite

nice because a lot of astrophysical
programs are still written in Fortran. So,

this is good for me because I don't know
any Fortran. We can run this and what's

typical for FARGO3D. So that's a compilation
actually on my computer. So, I don't need

a fancy computer. I just did it on my
small laptop and now we run it. Now,

typical for FARGO3D, as you will see are lot
of dots. So, here it will print out a lot

of dots and it will create at certain
times some outputs. And these outputs are

huge files containing numbers. So, if you
look at them they are not really

interesting. They just are a numbers in
something like a text file. So, a big part

of astrophysics is also to visualize the
data. Not only to create it but also to

make images so that we can make movies out
of them. For that, I prefer to use Python

but there are a lot of tools Python
Matplotlib, but there are a lot of

different tools to visualize the data. So,
this is actually that output. That first

one we just saw. The Jupiter planet in the
disc that I defined in this parameter file

and it's already started to do some
spirals. And if I would have let it

run further than the spirals were more
prominent. And yeah, now we have a planet

here on our computer.
miosta: OK, so we also have PLUTO. PLUTO

somehow has a bit more setup files. So,
what I need is three files here. Looks a

bit complicated to break it down. This
file defines my grid and initial values.

And this simulation time here we input
actually what physics do we want to need?

What is our coordinate system? So, do we
want to have a disc or just like spherical

boxes or like squared boxes? And how is
the time defined? And here we then

actually write a bit of code to say, OK,
now how do I want a gravitational

potential? So, what's the source of
gravity or what will happen at the inner

region where we have this dark spot? We
have somehow to define what happens if gas

reaches this boundary. Is it just falling
in? Is it bouncing back or something? Or

is it looping through the one end to the
next? This is also something we then just

have to code in. And if we then make it
and let run, it looks like this. So,

again, our nice thing we hopefully put in
or wanted to put in: the time steps, what

our boundaries were, parameters of
physics. Hopefully, the right ones and

then nicely we start with our time steps
and then we see this. It's hooray! It

worked actually. Because it's actually not
quite simple usually to set up a running

program. A running problem, because you
have to really think about what should be

the physics. What's the scale of your
problem? What's the timescale of your

problem? And specify this in a good way.
But in principle, this is how it works.

There are few test problems if you
actually want to play around with this to

make it easy for the beginning. And this
is how we do simulations. So, as I already

set, we can just start them on our
laptops. So, here this is my laptop. I

just type a dot slash FARGO3D and that
should run, right? And then I just wait

for ten years to finish the simulations of
500 timesteps or outputs. Well, that's not the best

idea. So, we need more power. And both of
us, for example, are using a cluster for

Baden-Württemberg and that takes down our
computation time by a lot. Usually, like a

factor of maybe 20, which is a lot. So, I
would need on my computer maybe a year and

then I just need maybe 5 hours, a few days
or a week on this cluster, which is

usually the simulation time about a week
for me.

caro: So, what you see here is that we use
GPUs, yes. But we do not or mostly not use

them for gaming. We use them for actually
actual science. Yeah, would be nice to

play on that, right? That just said!
miosta: So, back to our Earth, actually.

So, can we now? We wanted to grow our own
planet. We can do that, yes of course. Can

we grow Earth? Well, Earth is a very
special planet. We have a very nice

temperature here, right? And we have not a
crushing atmosphere like Jupiter, like a

huge planet that we could not live under.
We have a magnetic field that shields us

from the radiation from space and we have
water. But just enough water so that we

still have land on this planet where we
can live on. So, even if we fine tune

simulations, the probability that we
actually hit Earth and have all the

parameters right is actually tiny. It's
not that easy to simulate an Earth. And

there are a lot of open questions, too.
How did we actually manage to get just

this sip of water on our surface? How did
we manage to collide enough mass or

aggregate enough mass to form this
terrestrial planet without Jupiter is

sweeping up all the mass in our system?
How could we be stable in this orbit when

there are seven other planets swirling
around and interacting with us? All of

this is open in our field of research
actually, and not completely understood.

This is the reason why we still need to
do astrophysics and even in all our

simulations there is no planet B. And the
earth is quite unique and perfect for

human life. So, please take care of the
Earth and take care of yourself and of all

the others people on the Congress. And
thank you for listening and thank you to

everyone who helped us make this possible.
And to the people who actually coded our

programs with which we simulate.
Thank you!

applause

Herald: Thank you for the beautiful talk
and for the message at the end, the paper

is open for discussion, so if you guys
have any questions, please come to the

microphones. I'm asking my Signal Angel?
No questions right now. But microphone two

please!
Mic2: Oh, yeah. Thank you very much.

Really beautiful talk. I can agree. I have
two questions. The first is short. You are

using Navier-Stokes equation, but you have
on the one hand, you have the dust disc

and on the other hand, you have solid
planets in it. And so are you using the

same description for both
or is it a hybrid?

miosta: It very much depends. This is one
of the things I showed you that for PLUTO,

we write this C file that specifies some
things and about every physicist has

somewhat his or her own version of things.
So, some usually the planets, if they are

large, they will be put in as a gravity
source. And possibly one that can accrete

and pebbles are usually then put in a
different way. However, also pebbles are

at the moment a bit complicated. There are
special groups specializing in

understanding pebbles because as we said
in the beginning, when they collide,

usually they should be destroyed. If you
hit two rocks very together, they don't

stick. If you hit them hard together, they
splatter around and we don't end up with an bigger object

caro: Just to explain pebbles are small
rocks or like big sand stones or something

like that. Yeah. So bigger rocks,
but not very big, yet.

miosta: Yes!
caro: It depends on which code you use.

Mic2: Thank you. Very short, maybe one.
Do you also need to include relativistic

effects. Or is that completely out?
miosta: It's a good question. Mostly if

you have a solar type system, you're in
the arrange where this is not necessary.

For example, with the binaries, if they
got very close together, then at the inner

part of the disc, that is something we
could consider. And actually, I know for

PLUTO, it has modules to include
relativistic physics, too, yes!

Mic2: Thank you!
Herald: OK, we have quite some questions,

so keep them short. Number one, please!
Mic1: Thank you. Yeah. Thank you very

much for your interesting talk. And I
think you had it on your very first slides

that about 70 percent of the universe
consists of dark matter and energy. Is that

somehow considered in your
simulations or how do you handle this?

caro: Well in the simulations we make, we
are doing planets and discs around stars.

It's not considered there. In the
simulation we showed you about the

universe at the beginning, the blueish
things were all dark matter. So, that was

included in there.
Mic1: OK, thank you.

Herald: OK. Microphone 3.
Mic3: Hi, thanks. Sorry, I think you

talked about three different programs. I
think PLUTO, FARGO3D and a third one. So,

for a complete beginner: which program
would you suggest is like you more use

like if you want to learn more?
Which one is user friendly or good?

miosta: I would suggest FARGO3D first. It's
kind of user friendly, has a somewhat good

support and they are always also very
thankful for actual comments and additions

if people actually are engaged in trying
to improve on that. Because we are

physicists. We're not perfect programmers
and we're also happy to learn more. So

yeah, FARGO3D I would suggest, it has some
easy ways of testing some systems and

getting something done.
caro: And it also has a very good

documentation and also a manual "How to
make the first steps on the Internet". So,

you can look that up.
Mic3: Awesome. Thank you.

Herald: Let's get one question from
outside, from my Signal Angel.

Signal Angel: Thank you for your talk.
There's one question from IRC: How do you

know your model is good when you can only
observe snapshots?

caro: Oh, that's a good question. As we
said, we're in theoretical astrophysics.

So, there are theoretical models and these
models cannot include everything. So,

every single process, it's not possible
because then we would calculate for years.

Yeah, to know if a model is
good you have to…

miosta: Usually, you have a hypothesis or
an observation that you somehow want to

understand. With most of the necessary
physics at this stage to reproduce this

image. So, also from the observation we
have to take into the account what our

parameters kind of should be, how dense
this end of the simulation should be and

things like this. So, by comparing two
observations, that's the best measure we

can get. If we kind of agree. Of course,
if we do something completely wrong, then

it will just blow up or we will get a
horribly high density. So, this is how we

know. Physics will just go crazy if we do
too large mistakes. Otherwise, we would

try to compare two observations that it
actually is sensible what we did.

caro: Yeah, that's one of the most
complicated tasks to include just enough

physics that the system is represented in
a good enough way. But not too much that

our simulation would blow up in time.
Herald: Number two, please.

Mic2: I've got a question about the
adaptive grids. How does the computer

decide how to adapt the grid? Because the
data where's the high density comes after

making the grid...
miosta: Yes, this is actually quite an

interesting and also not quite easy to
answer question. Let me try to give a

breakdown nutshell answer here.
The thing is, you measure and evaluate the

velocities. Or in the flux, you also
evaluate the velocity. And if the velocity

goes high, you know there's a lot
happening. So, we need a smaller grid then

there. So, we try to create more grid
cells where we have a higher velocity. In

a nutshell, this is of course in an
algorithm a bit harder to actually

achieve. But this is the idea. We measured
the velocities at each point. And then if

we measure a high velocity,
we change to a smaller grid.

Mic2: So, you can predict where the mass
will go and whether densities are getting high.

miosta: Exactly. Step by step so to say.

Mic2: Thanks
Herald: We stay with Microphone 2.

Mic2: Okay. I've got a bit of a classical
question. So, I guess a lot relies on your

initial conditions and I have two
questions related to that. So first, I

guess they are inspired by observations.
What are the uncertainties that you have?

And B, then what is the impact if you
change your initial conditions like the

density in the disc?
miosta: Yeah, right now my main research

is actually figuring out a sensible
initial conditions or parameters for a

disc. If you just let it have an initial
set of conditions and a sensible set of

parameters and let it run very long, you
expect a system hopefully to convert to

the state that it should be in. But your
parameters are of course very important.

And here we go back to what we can
actually understand from observations. And

what we need for example is the density,
for example. And that is something we try

to estimate from the light we see in these
discs that you saw in this nice grid with

all these discs we estimate OK, what's the
average light there? What should then be

the average densities of dust
and gas in comparable disks.

Mic2: Okay, thanks.
Herald: Okay, one more at number two.

Mic2: Yes. Thank you for the talk. When
you increase the detail on the grid and

you learn more. When you want to compute
the gravitational force in one cell, you

have to somehow hold masses from the all
the other cells. So, the complexity of the

calculus grows.
miosta: Yes

Mic2: Quadratically, at the square of the...
how do you solve that? With more CPUs?

caro: Well, that would be one way to do
that. But there are ways to simplify if

you have a lot of particles in one
direction and they are far away from the

object you're looking at. So, yeah. So, if
you have several balls here and one ball

here, then you can include all these balls
or you can think of them as one ball. So,

it depends on how you look at it. So, how
you define how many particles you can take

together is when you look at the angle of
this... many particles we'll have from the

seen from the object you're looking at.
And you can define a critical angle. And

if an object gets smaller or if lot of
objects get smaller than this angle, you

can just say, OK, that's one object. So,
that's a way to simplify this method. And

there are some, yeah,
I think that's the main idea.

Herald: Okay, we have another one.
Mic2: Do you have a strategy to check if

the simulation will give a valuable
solution or does it happen a lot that you

wait one week for the calculation and find
out OK it's total trash or it crashed in

the time.
caro: So, that also depends on the program

you're using. So, in FARGO3D, it gives
these outputs after a certain amount of

calculation steps and you can already look
at those outputs before the simulation is

finished. So, that would be a way to
control if it's really working. Yeah, but

I think...
miosta: It's the same for PLUTO. So, there

is a difference between timesteps and
actually output steps. So and you could

define your output steps not and as the
whole simulation, but you can look at each

output step as soon as it's produced. So,
I usually get like 500 outputs, but I

already can look at the first and second after
maybe half an hour or something like that.

caro: Yeah, but it also happens that you

start a simulation and wait, and wait, and
wait and then see you put something wrong

in there and well then you have to do it
again. So, this happens as well.

Mic2: Thanks.
Herald: Okay. One final question.

Mic2: Yeah, OK. Is there a program in
which you can calculate it backwards? So

that you don't have the starting
conditions but the ending conditions

and you can calculate how it had started?
miosta: Not for hydrodynamic. If you go to

n-body, there is a way to go backwards in
time. But for hydrodynamics, the thing is

that you have turbulent and almost like
chaotic conditions. So, you cannot really

turn them back in time. With n-body you
can it because actually it's kind of... Well,

it's not analytically solved, but it's
much closer than like turbulences,

streams, spirals and all the
things you saw in the simulations.

Herald: OK, I guess that brings us to the
end of the talk and of the session. Thank

you for the discussion and of course,
thank you guys for the presentation.

36c3 postroll music

Subtitles created by c3subtitles.de
in the year 2021. Join, and help us!