►
From YouTube: NUG Meeting 2014: Almgren
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Our
next
speaker
and
Algren,
she
was
at
Livermore
and
moved
to
Berkeley
when
nursed
moved
to
Berkeley
in
1996.
So
she
was
at
Livermore
in
the
applied
math
group
and
she's.
Currently,
a
member
of
the
Center
for
computational
sciences
and
engineering
in
the
computational
research
division
here
at
the
lab
and
works
on
development,
implementation
of
algorithms,
for
solving
numerical
pdes
for
fluid
dynamics
with
applications
to
Astro
physics,
cosmology
and
low
Mach
number
flows,
she's
a
primary
developer
of
Castro
maestro
and
the
NYX
codes.
Who
all
of
us
who
work
at
nurse
know
very
well.
B
I
started
at
at
Livermore
in
1992,
I
was
a
graduate
student,
/
flea
and
then
I
went
to
livermore
as
a
postdoc
and
Mike
welcome
was
a
member
of
the
applied
math
group
that
I
joined
then,
and
this
is
this-
is
a
picture
apparently
from
1998.
So
this
is
the
way
I
did
I,
remember
Mike
and
we
had
a
lot
of
fun.
Actually
I
was
thinking
about
it.
I
was
remembering
the
way
we
worked
at
liverman.
B
We
had
these
offices
across
the
hall
from
each
other
and
of
course
nobody
had
two
screens
at
that
point,
which
I
think
many
of
us
do
now
and
the
debugger
would
take
it
pretty
much
one
screen,
and
so
we
had
debuggers.
Then
it
wasn't
that
ancient
history
and
Mike
and
I
would
each
bring
up
a
run
on
our
screens
and
we
yell
across
clogged.
B
Okay,
Mike,
you
read
the
left,
column
and
he'd,
read
his
numbers
and
then
I
say
stop
ok,
that
one's
different,
the
thing
I
was
trying
to
remember
and
I
want
to
pull
the
audience
here.
The
debugger
we
were
using
the
only
thing
I
remember
about
it,
was
that
you
could
set
the
palette
and
you
get
options
like
Monet
and
Cezanne.
Does
anybody
remember
this
one?
B
Okay,
if
you
think
of
it,
let
me
know
because
it's
driving
me
nuts,
because
I
just
remember
you
could
you
could
get
these
nice
pastel
colors
miu
debugger
and
they
had
such
a
selection
it
just
anyway,
so
we
will
all
miss
Mike.
Ok,
this
is
this
is
something
completely
different
from
what
David
talked
about.
This
is
the
one
of
the
application
areas
that
I
work
in
this
is
I
apologize.
It's
a
little
bit
of
a
revisit
of
a
talk.
B
I
gave
it
supercomputing
in
2012,
and
it's
an
example
of
one
types
of
things:
we
do
it
low
mach
number
models
in
computational,
astrophysicist
astrophysics
I
am
NOT
an
astrophysicist,
so
a
number
of
things
I
say
are
things
that
my
colleagues
have
taught
me
when
we
get
to
the
algorithms.
That's
where
I
really
know
what
I'm
talking
about.
So
the
question
is
so
what
do
we
do?
B
We
want
to
study
astrophysics,
we
want
to
study
certain
types
of
phenomena
and
when
you
think
about
astrophysics,
you
tend
to
think
of
things
that
blow
up,
and
so
the
first
thing
you
say
is
if
it's
blowing
up
sure
as
heck
not
low
Mach
number,
because
the
mach
number
is
the
fluid
speed
over
the
sound
speed
and
when
things
blow
up
they're
going
really
fast.
So
why
would
you
ever
use
a
low
mach
number
model
for
astrophysics?
And
the
answer
is
for
a
lot
of
simulations.
B
You
wouldn't
the
type
of
things
that
we
think
about
with
astrophysics.
We
think
about
novi.
We
think
about
super
novae.
We
think
about
gamma-ray
bursts,
or
maybe
we
did.
We
don't
think
about
those,
but
those
are
things
that
blow
up
x-ray
bursts.
Those
are
all
very
compressible,
very
explosive
phenomena
and
they're
the
ones
that
make
astrophysics
really
interesting.
The
one
I
want
to
focus
that
we
actually
have
modeled.
Several
of
those.
The
one
I
want
to
focus
on
is
type
1a
supernovae
and
there's
a
whole
hallway
of
people
actually
this
floor.
B
Even
you
know
a
whole
lot
about
these,
so
I'm
going
to
give
the
cartoon
version
of
the
science
and
why
you
would
care
so
what
are
they?
They
are
the
largest
thermonuclear
explosion
in
the
universe.
So
that's
kind
of
neat
the
brightness
of
a
single
star,
so
a
supernova
is
a
single
star
blowing
up
at
its
peak.
It
rivals
the
brightness
of
the
galaxy,
which
is
many
many
stars,
so
it
gives
you
an
idea
of
the
amount
of
energy
involved
and
the
first
question
I
asked,
of
course,
sometimes
a
type
1a.
B
B
B
There's
this
build
up
period,
there's
something
that
gets
the
start
to
the
point
of
exploding,
and
so
there
are
these
long
periods
of
very
non-explosive
fluid
dynamics
and
those
are
characterized
by
low
Mach
number
flows,
their
convective
flows,
which
means
what
you
care
about,
is
the
fluid
motion,
as
opposed
to
the
acoustics
the
sound
waves
traveling
around,
and
it
turns
out
that
we
actually
care
about
what
happens
before
the
explosions.
It
sets
the
the
initial
conditions,
and
so
we
need
to
be
able
to
understand
and
simulate
both
types.
So
we're
going
to
rest.
B
This
talks
can
be
talking
about
the
type
1a
supernovae,
but
a
lot
of
this
applies
to
other
things
that
we
do.
Ok
here
is
the
cartoon
versions,
the
physics:
why
do
you
care
about
type
ones?
Ok,
so
you
have
several
choices
for
the
universe.
It's
expanding
its,
not
expanding
its
step
and
in
any
one
of
those
cases
you
can
ask
if
it's
expanding,
is
it
getting
faster
or
slower?
If
it's
compressing,
is
it
going
to
turn
around?
B
So
we
want
to
know
what's
going
on
so
originally
back
about
a
hundred
years
ago,
I
said
no
at
step.
Okay,
that
that
kind
of
makes
sense,
that's
what
everyone
will
go
to.
First
1929
you
have
Hubble
observing
that
galaxies
are
in
fact
moving
away
from
us.
Okay,
so
that
says
the
universe
is
expanding.
So
that's
the
first
of
three
choices.
Now
we
want
to
know,
is
it
going
to
keep
is
expanding
a
constant
rate?
Is
it
going
to
get
faster?
Is
it
going
to
get
slower
so
now
you've
got
a
bunch
of
observations.
B
You
look
at
the
rotational
speed
of
galaxies,
orbital
velocity
of
galaxies
and
clusters,
gravitational
lensing.
All
of
these
things
point
to
the
fact
that
there's
something
else
out
there.
Besides,
what
we
see,
there's
mass
there
that
we
haven't
accounted
for
this
is
the
old
I
was
a
physics
majors.
The
undergrad
I
still
remember.
If
you
want
to
know
the
stone
spinning
around
the
string,
you've
got
this
MV
squared
over
R.
Well,
so
Em's
got
to
be
bigger
than
what
we
can
see.
B
B
This
is
that
they
hold
supernova
product.
The
supernova
cosmology
project,
high-z,
supernova
search
team.
Give
the
credit
where
credit's
due
this
is
the
very
short
cartoon
version.
If
you
say
that
all
supernovae
are
the
same,
if
you
believe
they're
the
same
there's
a
way
of
mapping
the
duration
to
the
duration
to
the
absolute
magnitude.
B
If
you
know
the
absolute
magnitude-
and
you
know
the
observed
magnet
to
the
apparent
brightness,
then
you
know
the
distance
you
how
far
away
it
is,
and
if
you
know
the
distance-
and
you
know
the
redshift,
you
put
those
on
a
plot-
you
get
a
Hubble
diagram
and
what
that
tells
you
is
that,
let
me
show
you
say
this
right:
the
older
objects
are
not
moving
away
from
you
as
fast.
Therefore,
the
universe,
the
rate
at
which
the
universe
is
expanding,
is
increasing.
B
So
this
is
huge,
because
now
there
has
to
be
something
to
balance
all
the
dark
matter,
and
so
because
we
don't
know
what
that
is,
we'll
call
it
dark
energy.
So
so
there's
all
these
fundamental
questions
of
the
universe
and
if
you
sort
of
look
back
there's
this
this
this
thing
in
red
their
standard
candles.
When
we
said
so
super
novae
behave
now
the
way
they've
always
behaved
and
they
will
behave.
It
would
be
really
nice
if
we
understood
how
they
blow
up
because
we're
putting
a
whole
lot
of
faith
in
them.
B
Concluding
that
there's
dark
energy,
but
we
really
don't
know
how
they
start
or
how
they
blow
up,
and
is
there
any
reason
to
assume
that
the
Stars
many
many
years
ago
blew
up
the
same
way
they
do
now.
So
if
we
understood
that
we
might
have
some
more
confidence
for
a
while,
there
was
a
this
was
sort
of
the
dominant
progenitor
theory.
I
would
say
in
the
last
couple
years
a
number
of
other
progenitors
have
been
explored.
This
is
the
one
that
we
have
looked
at.
We
are
actually
the
process
of
looking
like
several
others.
B
So
one
of
the
things,
if
you
want
to
do
a
simulation,
you
guys
know
what
your
time
scales
are:
we're
not
going
to
simulate
10
million
years
so
as
it
gets
bigger
and
heavier,
and
the
pressure
in
the
center
of
the
star
increases
over
a
period
of
centuries.
The
carbon
burning
near
the
core
drives
convection.
So
slowly,
this
thing
starts
to
it's,
not
simmering,
yet
I'm
not
sure
what
the
appropriate
word,
but
slowly,
things
are
starting
to
move.
You've
got
heating
at
the
center
you're
starting
to
circulate
over
the
last
few
hours.
B
The
convection
becomes
more
vigorous.
You
have
the
heat
release.
The
heat
generation
of
the
center
is
intensifying,
the
convection
can
carry
the
way,
the
heat
so
fast,
and
initially
that's
enough
that
keeps
it
very.
Very
slowly
heating
up
heat
released
becomes
more
intense,
convection
becomes
more
vigorous.
Eventually,
the
convection
can't
take
away
the
heat
fast
enough
and
that's
what
we
call
ignition
so
ignition
is
where
you
eat
this
critical
temperature
and
then
within
seconds
it
blows
so
you've
got
really
disparate
timescales.
Here,
you've
got
this.
B
The
century
we've
got
this
millions
of
years,
centuries
hours
and
ignition.
So
we
are
not
so
ambitious
that
we're
going
back
two
centuries,
but
we
would
like
to
go
back
a
few
hours.
Okay.
So
if
you
follow
these
things,
you
know
that
people
have
announced
we've
solved,
supernova
we've
simulated
these
done.
We
know
how
they
blow
up
and
if
you
look
more
closely,
what
they
have
simulated
is
those
last
two
to
three
seconds,
and
so
how
do
you
do
it?
So
this
is
the
time
place.
This
is
not
our
work.
B
This
was
at
the
university
of
chicago,
and
the
red
here
is
probably
a
temperature
contour,
and
so
what
it
shows
is
you
have
an
initial
condition
of
some
sort.
You
let
it
evolve.
This
red
this
bubble
will
push
out
break
out.
The
whole
thing
will
explode
okay,
so
this
is
great.
This
is
an
exploding
star,
and
how
do
you
start
this?
Well,
you
have
to
know
what
a
star
looks
like
until
you
take
a
1d
profile
from
you:
have
a
1d
stellar
evolution
code.
B
You
map
that
into
a
3d
star
that
gives
you
the
radial
profile
of
the
star.
That's
probably
pretty
good.
Well,
then,
you
have
to
ignite
it
somewhere
and
the
thing
that
turns
out.
You
have
to
make
some
assumptions
here,
because
you
don't
know
where
it
ignites
and
the
problem
that
we
have
is
that
the
answer
depends
on
the
initial
conditions
which,
if
you've
ever
run
simulations,
you
find
sometimes
happens,
and
so
here
this
is
an
example.
B
Central
ignition,
you
say
stars
are
hottest
in
the
middle
because
that's
where
the
pressure
is
hottest,
and
that's
that's
where
you
typically
think
that
the
things
being
hot,
so
we
will
ignite
it
in
the
center
and
if
you
ignite
in
the
center,
you
get
a
very
specific
kind
of
explosion
and
it
turns
out,
if
you
ignite
an
off-center,
you
get
a
different
kind
of
explosion,
and
so
nobody
had
really
addressed
the
question
of
well.
How
do
we
know
where
it
excites
ignites
and
does
it
ignite?
B
You
know
10
kilometers
off
the
center
100
kilometers
off
the
center
that
matters
it
turns
out
in
terms
of
whether
whether
you
can
even
get
a
realistic
supernova
out
of
this
type
of
progenitor.
So
we
really
need
to
know
because
what
we
know
from
all
these
compressible
simulations
is
that
it
does
matter
how
you
start
it.
So
that's
why
we
care.
B
So
what
do
you
do?
Go
back
to
the
drawing
board.
You
say:
okay,
so
we
know
how
people
model
Superdome
and
basically
this
is
the
compressible
navier-stokes
equations.
We
can
write
them
down
very
easily.
We've
got
density,
velocity
pressure,
energy.
The
only
thing.
That's
actually
really
complicated
about
this
is
the
equation
of
state.
It's
not
a
regular
gamma
log.
Get
acid,
has
certain
degeneracies,
but
people
kind
of
know
what
that
is.
B
So
you
write
that
in
you
put
it
into
your
code
and
it's
a
really
nice
thing
to
program
and
run,
and
so
we,
in
fact
the
castro
code
solves
these
equations
and
it's
used
for
you
can
use
it
for
type
1a
explosions.
All
that's
been
done.
We
have
a
lot
of
collaborators
who
actually
using
it
for
type
2,
which
are
the
core
collapse.
Supernovae
and
I.
Think
wasn't
you.
Somebody
called
me
up
once
and
said
so
casters
running
on
nursing,
that's
taking
ten
percent
machine.
What
are
you
doing?
B
I
have
no
clue
class,
something
to
do
with
type
twos.
We've
got
collaborators
running
these
things
pretty
often,
but
anyway.
This
is
a
great
thing
to
program
and
run
because
you
write
it
down
as
a
conservation
law.
You,
as
your
state
variable
you
update
it
with
fluxes.
You
can
use
a
time
explicit
method
to
solve
this.
It's
easy
to
program,
it's
easy
to
paralyze!
We've
done
this.
We
got
up
to
2000
cores
I
shouldn't,
get
that
not
on
Nurse
and
it's
straightforward.
B
So
we
do
a
lot
of
things
with
AMR
because
stars
it
turns
out
or
not
uniform,
and
you
really
want
to
zoom
in
on
the
center
of
the
star,
with
region
of
interest
and
not
all
of
space.
So
we
use
adaptive
mesh
refinement
for
these,
and
this
is
great.
So
we
know
how
to
do
this
and
it
works
great
for
modeling
those
two
seconds.
B
The
problem
is
those
two
second
calculations.
When
I
showed
you
the
cartoon
of
those
were
hero,
calculations
now
great
they're,
10
or
more
years
back
now,
so
they're,
not
so
hero
calculations.
Today,
but
two
seconds
was
the
limit,
so
now
I
say:
well.
We
really
need
to
go
back
two
hours.
Nobody
has
the
patience
for
that.
So
that's
that's
where,
for
me,
the
fun
begins.
Oh
one
more
side,
so
this
is
just
a
statement
that
well,
of
course
we
have
super
computer.
This
is
supercomputing
conference,
so
I
had
to
say
the
word
supercomputers.
B
So
you
know
we
have
a
mr,
isn't
that
the
answer
to
everything
you
just
zoom
in
and
you
don't
need
to
resolve
the
whole
thing.
Isn't
that
wonderful.
So
the
problem
is,
of
course,
that
when
you
write
an
explicit
code,
you
have
a
time
step
constraint
that
is
limited
by
so
Delta
X
is
the
mesh
facing
you.
Fluid
velocity,
C
Sound
speed
as
the
mesh
space
and
gets
smaller.
B
The
time
step
gets
smaller,
that
two
hours
takes
a
really
long
time,
and
so,
with
a
method
like
this,
you
can't
beat
the
probably
the
fact
that
you're
not
parallel
on
time
and
yes,
there
are
parallel
time
methods
and
that's
a
whole
separate
thing,
but
this
will
just
take
too
long,
even
with
lots
and
lots
of
processors,
even
with
perfect
scaling
so
I'm,
not
that
patient
of
person.
So
we
say:
ok,
let's
go
back
to
the
drawing
board.
I
showed
you
the
compressible
equations,
those
aren't
going
to
work
so
now
we
want
a
new
model.
B
This
to
me
is
where
it
gets
really
fun
because
we
say
ok,
what
do
we
care
about?
Well,
we
don't
care
about
the
acoustic
waves,
because
when
this
thing
is
churning
the
acoustic
waves
are
bouncing
back
and
forth
across
the
star.
They
serve
to
equilibrate
pressure
in
a
specific
way,
but
we
don't
need
to
resolve.
B
We
don't
need
to
resolve
in
the
same
way
that
if
you
want
to
do
a
model
of
the
heat
flow
in
this
room
and
you
care
about
well,
maybe
you,
as
individual
people
are
generating
a
little
bit
heat
and
that's
driving
convection
in
the
room.
You
might
want
to
have
a
great
model.
This
room,
you
don't
care
about
the
fact
that
I'm
talking
right,
there's
acoustic
waves
bouncing
around
they're
irrelevant
to
the
solution,
and
if
you
track
those,
you
would
never
find
out
whether
you
need
to
turn
on
the
air
conditioning.
B
Ok,
so
it's
the
same
thing
with
the
star.
We
just
don't
care
about
those,
but
they
kill
the
time
step
in
a
compressible
code.
So
what
we
want
to
do
is
analytically,
get
rid
of
those
ok,
we
want
to
get
rid
of
the
sound
waves,
but
we
don't
get
want
to
get
rid
of
analogously.
We
don't
want
to
get
rid
of
the
fact
that
the
individuals
in
the
room
are
generating
body
heat
that
is
making
the
airwaves.
So
the
same
thing
here
would
say:
okay,
what's
up
what's
our
wish
list?
B
Well,
obviously,
you
have
to
add
Boyan,
so
you
have
to
have
hot
things
rise.
You
have
to
allow
for
the
fact
that
this
star
is
regularly
stratified
it's
much
denser
at
the
center
of
the
star.
That
is
that
the
outside
you
have
an
equation
of
state,
that's
kind
of
funky,
so
you
can't
make
any
gamma
law
assumptions.
B
You've
got
heat
release
you
get
reactions.
You've
got
got
a
lot
of
heat
release.
You've
got
to
have
that
make
the
fluid
less
compressible
and
therefore
buoyant
and
you've
got
let's
start
puff,
because
these
things
do
actually
puff
out
before
they
explode,
there's
just
an
overall
expansion
of
the
star,
and
so
you
want
a
model
that
does
this,
and
the
whole
point,
of
course,
is
that
you
can
run
it
with
a
bigger
DT
and
therefore
you
can
solve
it
in
your
lifetime
so
lower
time.
It's
all
that
time
to
solution.
B
I
want
to
get
the
answer.
So
that
I
can
use
all
those
neat
techniques
to
understand
what
the
science
is
telling
me.
Okay
and
the
question
is
you
know
gosh
I've
heard
about
these
things
and
don't
these
models
exist.
This
is
just
a
quick
chart
to
say.
Yes,
there
are
a
whole
lot
of
models
along
this
spectrum.
If
you
look
so
the
general
categories,
low
mach
number
models,
incompressible
is
one
limit
of
that
and
then
an
elastic
is
used
a
lot
in
atmospheric
modeling.
B
You
might
hear
that
one
a
boost
nest
is
a
simpler
version
of
an
elastic
pseudo.
Incompressible
is
another
something
that's
very
common
in
MLS
in
them
atmospheric
modeling,
and
each
of
those
fails
to
to
allow
for
one
of
the
features
that
we
need
and
in
fact,
in
atmospheric
modeling.
It
turns
out
that
the
temperature
differences
are
not
that
high.
So
you
know
in
modeling
climate,
flower
or
just
weather
prediction,
you
don't
have
the
kinds
of
extremes
in
temperature
that
you
do
when
a
supernova
is
about
to
at
night.
B
Okay,
so
that
works
great
for
the
atmosphere.
It
doesn't
work
when
you're
trying
to
explode
something
and
again
the
background
states,
something
that
was
new.
Okay.
This
example
is
just
to
show
that
if
you
aren't
careful,
you
can
be
really
wrong.
Okay,
so
this
is
just
somebody's
point.
Well,
people
point
out
to
me
more
than
once
it's
a
bubble
rising.
B
The
pictures
at
the
top
should
be
at
the
bottom,
so
you
get
the
sense
of
rising
up,
so
this
is
a
times
where
time
goes
down
and
there
are
four
different
models
here
on
the
left
is
what
a
compressible
code
would
give
you,
so
that
is
just
a
classic
bubble.
If
you
go
from
the
upper
left
to
middle
left
to
bottom
left,
that's
a
classic
bubble
rising
in
it
and
it
kind
of
mushrooms
like
that,
this
one's
just
a
2d.
B
In
the
second
column,
you
see
the
picture
of
the
low
mach
number
method,
and
what
you'll
notice
is
that
the
shape
is
roughly
the
same.
The
height
of
the
bubble
is
roughly
the
same,
and
in
fact,
if
you
look
carefully
it's
actually
interesting,
you
can
see
that
the
compressible
bubble
is
wider.
This
drove
us
nuts
for
years
and
in
fact
this
year
it
finally
figured
out
why
she's
somebody
else
figured
out
why?
But
if
you
say
well,
you
know
this
other
models,
weren't
worth
a
good
enough.
B
They
were
already
out
there
so
yeah
an
elastic
model.
An
elastic
model
puts
in
a
linearization
of
the
effect
of
temperature
on
density.
It
turns
out
the
bubble.
Just
doesn't
rise
enough,
so
you
put
in
that
model.
You'll
get
the
answer
wrong.
Incompressible
gets
the
buoyancy
right,
but
doesn't
let
the
expansion
happen
correctly
and
it
basically
well
it
doesn't
it
completely
on
physicals.
B
So
the
point
of
this
is
that
simple
models
weren't
enough.
We
need
to
come
up
with
a
new
one
and
what
happens
the
way
all
the
mach
number
models
are
drive.
Basically,
is
the
observation
that
if
the
flow
is
slow,
if
the
mach
number
is
small,
then
we
know
by
asymptotics
that
the
pressure
has
to
be
relatively
uniform,
I
say
relatively
uniform,
because,
if
you're
doing
a
combustion
calculation,
the
pressure
is
a
constant
in
space.
B
You,
don't
you,
don't
have
large
gravitational
effects,
so
there's
a
pressure
that
that
may
grow
in
time,
if
you're
in
a
closed
chamber,
but
it
has
no
spatial
variation
if
you're
doing
a
star,
the
pressure
is
very
stratified,
has
dependence
on
the
radius
or
the
elevation
in
the
atmosphere
and
time,
and
then
you
had
a
pressure
perturbation.
So
this
is
true
for
all
know:
Mach
number
flows-
if
it
wasn't
true,
then
that
pressure
probation
would
accelerate
the
flow.
B
It
wouldn't
be
low,
knock
number,
so
you're,
safe
and
the
the
assumption
you
make
is
that
that
background
pressure
is
what
enters
the
thermodynamic
relationships
and
the
the
probational
pressure
pie
is
what
affects
the
dynamics
physically?
What
happens
is
the
sound
waves
instead
of
tracking
them
as
they
move
across
the
room
or
across
the
star?
You
just
say
they
equilibrate
instantly.
B
So
every
time
step
you
equilibrate
the
sound
waves,
they
still
affect
the
solution,
but
they
affect
them
instantaneously,
you're,
not
tracking
them,
as
they
march
across,
and
so
we
say,
they're
filtered
out,
because
sound
waves
are
no
longer
a
solution
to
the
equations.
Okay,
mathematically,
what
happens?
B
You've
changed
your
equation,
set
you've
introduced
to
constraint,
and
that
changes
the
character
of
that
and
that's
going
to
require
some
additional
solution
techniques,
but
computationally
you've
got
what
you
want
if
you
advance
the
method
design
on
these
equations,
you
march
it
at
the
fluid
velocity,
not
the
acoustic
velocity
and
that's
what
we
wanted.
Ok,
what
does
it
look
like?
It
looks
very
similar.
In
fact,
the
first
three
equations
are
pretty
much
identical
pretty
much.
The
fourth
one
is
a
constraint
on
the
velocity,
and
that
says
the
velocity
can
expand
due
to
a
bubble
rising.
B
It
can
expand
due
to
heat
release.
It
cannot
expand
because
of
the
coup.
Stick
wave
came
through
and
tried
to
to
push
on
it
and
these
other
variables
just
have
to
do
with
the
heat
release
and
the
compositional
change
and
everything,
and
then
at
the
bottom.
We
want
to
expand
the
star
as
well:
okay,
oh
okay.
So
there
were
three
codes
mentioned
in
the
introduction.
Castrum
is
a
compressible
code
that
we
wrote
that
people
are
using
forum
supernova,
mostly
maestro's,
the
code
that
we
wrote
for
this,
and
then
Nix
is
a
code.
B
That's
now
I
think
burning
the
most
hours
of
nuh-uh
NESN.
Look
at
Dave
nice,
not
looking
I
think
it
was
Nick's
was
a
code
that
was
just
featured
in
the
Edison
early
science
results
and
it
turns
out
that
that
they
share
a
lot
of
infrastructure,
which
is
a
convenient
thing
for
us
anyway.
So
the
numerical
approach
is
based
on
generalized
projection
method,
which
may
or
may
not
mean
anything
to
you.
It's
second
order
accurate,
which
is
what
you
need.
It
has
the
same
way
of
handling
the
reactions
as
the
other
code.
B
So
we
kind
of
know
that
that
works,
and
then
this
last
part
is,
is
just
a
plug
that
in
si
si
se,
we
have
a
software
framework
called
box
slide
and
again
going
back
to
Mike
welcome
box
I'd
was
something
that
we
developed
back
at
Livermore
and
the
funny
things
I
remember
the
discussion
said
well,
we
really
need
to
make
box
like
parallel.
Remember,
saying:
yeah,
parallel
I,
don't
know
that's
it.
We
can
put
that
off
for
a
while,
but
so
for
a
while.
B
We
had
P
box
like
the
parallel
version
of
that
and
it's
it's
kind
of
hard
to
imagine
now
that
we
ever
had
a
non-parallel
version.
I
I
don't
remember
that
actually,
but
it's
a
code
that
supports
block
searched
a
amar,
so
it
has
the
structure
to
you
can
do
anything
with
with
unions
of
grids.
You
can
do
anything
with
unions
of
grids
across
multiple
levels.
It
means
you
don't
have
to
rewrite
all
those
parts
of
the
code.
Every
time
you
write
a
new
one.
This
is
the
castro
code
scaled
over
200,000
processors.
B
It
has
these
linear
solvers
for
solving
elliptic
and
parabolic
equations.
It
turns
out
those
don't
scale
as
well.
So
that's
the
price
you
pay,
there's
always
there's
always
a
price
we
march
at
a
much
higher
time
step,
but
now
we
have
to
do
linear,
solvers,
and
anybody
who
does
this
knows
that
conservation
laws,
work
really
well.
B
Linear
solvers
have
issues,
but
we
I'd
say
we
scale
up
to
maybe
let's
say
50,000,
so
we
do
a
pretty
good
job,
hybrid,
MPI,
openmp
and
then
because
you
never
basically
a
plug-and-play
equation
of
state
reaction
networks
because
you
never
quite
sure.
Well,
you
know
this
was
a
carbon-oxygen.
You
white
dwarf.
What
happens
if
we
put
in
a
little
bit
this
you
know.
What's
the
magnesium
reaction,
you
want
to
be
able
to
play
with
that?
Okay,
so,
physically
again,
we
said
acoustic
waves
are
no
longer
there.
You've
got
the
other
physics
that
you
want.
B
Mathematically
you've
changed
the
character.
Computationally
you've
got
the
dt
you
want,
but
you
now
have
this
this
linear,
the
elliptic
equation
you
need
to
solve,
and
so
we
do
use.
Multigrid
multigrid
plays
very
well
with
AMR,
because
AMR
levels
can
look
a
lot
like
multigrid
levels.
So
it's
a
very
nice
iterative
technique
and
it
turns
out
that
that
is
the
linear.
Solver
is
what
takes
most
of
the
time,
not
surprisingly,
but
I
made
the
statement
in
the
end
we
care
about
time.
B
Dissolution
and
I
will
tell
you
that
we
can
do
a
calculation
of
Maestro
that
we
cannot
do
with
Castro.
So
that
is
the
question.
Is
it
worth
it?
Okay,
so
now
for
the
fun
part?
How
many
doing
a
time
we
we're
good
so
there's
a
question
that
we
want
to
answer,
which
is
what
are
the
initial
conditions
for
these
explosions?
So
we
go
out.
We
say:
okay,
we're
going
to
make
a
new
method
and
build
a
new
code
to
do
this.
B
We
do
that
we
verify
and
it
up
validate
and
verify
along
the
way,
and
then
we
run
a
simulation
and
some
of
the
things
that
data
data
was
talking
about.
That
David
was
talked
about
with
data.
There's
always
this
trade-off
of.
You
really
want
to
be
able
to
just
look
at
the
data.
I
love.
Looking
at
the
dancing,
you
know
speak
to
me.
What
do
I
see?
Well,
okay,
so
now
I've
got
too
many
numbers
to
do.
That.
B
That
are
very
important,
but
you
want
to
get
a
big
picture
and
you
it's
3d,
so
you
don't
want
to
block
out
something
that's
happening
behind,
and
this
was
actually
I'd
say
we
in
some,
since
we
hand
did
it
so
we
played
with
visit
and
we
just
kept
doing
it
over
and
over
until
when
we
made
movie
after
movie
to
say,
can
you
tell
what's
happening
and
if
you
can
come
up
with
a
service
that
will
just
do
that?
That
would
be
awesome.
B
Unfortunately
I
I
would
say
we
didn't
know
what
we
were
looking.
We
were
looking
for,
we
post
some
hypotheses
and
then
said:
okay,
does
this
look
like
it?
No
doesn't.
Look
like
so
anyway.
These
are
both
from
one
simulation.
We
wanted
to
know
in
the
inner
thousand
kilometers
of
the
star.
What
what
does
the
flow
look
like?
So
it
turns
out
that
people
had
done
some
some
very
modified,
some
small-scale
approximations
to
this.
B
They
cut
out
the
center
of
the
star
and
then
had
flows
that
went
through
the
center
to
start,
which
I
never
quite
understood.
So
there's
a
question
is:
does
ignition
happen
right
at
the
center
because
that's
where
its
hottest
and
then
it
just
everything
just
kind
of
goes
radially
outward
at
that's
one
picture
of
it
and
the
answer
see
if
I
can
make
this
go
there
we
go
so
believer.
This
is
the
best
movie
that
we
came
up
with
red
is
outward
flowing,
blue
is
inward
flowing,
so
you
can
get
a
general
idea.
B
There
is
a
sense
of
a
dipole.
Okay,
that's
something
that
had
been
predicted
from
some
earlier
simulations,
not
by
us.
They
said
we
think
we
see
this
dipole
flow
and
in
fact,
if
you
look
there
is
a
coherent
red
structure
coming
out.
There
is
not
a
coherent
blue
structure
going
in.
So
what
you'd
imagine
is
the
the
flow,
the
fluid
just
don't
get
sucked
in
from
everywhere
else.
It
gets
pulled
in
and
jets
out
through
the
center.
B
B
You
might
imagine
that
again
we
just
play
that
one
more
time
that
one
doesn't
tell
you
a
whole
lot,
except
that
things
are
moving
out
from
the
center
so
anyway,
what
we
finally
ended
up
doing
so
then,
how
do
you
define
ignition?
Because
you
can't
let
the
maestro
code
ignite
because
it
doesn't
solve
the
equations
once
it
ignited,
so
we
had
to
define
ignition.
So
we
said,
okay
ignition
is
where
it
gets
above
six
times,
ten,
nine
degrees
Kelvin,
and
we
want
to
know
we
say:
okay,
so
the
minute
that
happens.
B
We
have
to
stop
the
simulation
because
then
we're
outside
our
range
of
validity.
They
say
well
gosh.
If
you
get
pretty
close,
then
maybe
you're
a
potential
ignition
site,
and
so
what
we
did
is
through
the
simulation
we
monitored
the
peak
temperature
and
where
it
was,
and
we
made
this
histogram.
So
we
filtered
this
we've
been
the
data
and
what
you're
seeing
here
is
the
color
coding
is
by
the
temperature.
B
What
this
shows
you,
if
you
look
at
this,
is
that
the
hot
spots
occur
in
a
relatively
narrow
band
radio,
and
so
you
can
say
with
some
confidence
that
the
supernova
is
most
likely
to
ignite
somewhere
around
point
five,
that's
point
five
times
ten
to
the
seven
centimeters
from
the
center
of
the
star.
This
is
something
that
had
not
been
known
at
all
and
again
I.
The
thing
I
showed
you
before
was
a
central
United
ignition.
Where
you
say
it's
going
to
night
right
in
the
center
will
watch
it
move
out.
B
It
turns
out
it
doesn't
okay.
The
other
thing
that
you
want
to
know
is
these
locations
where
it's
igniting.
What's
the
flow
like?
Is
it
in
fact,
caught
up
in
the
middle
of
the
dipole?
Is
it
off
to
the
side?
Is
it
dropping
in
words,
and
just
because
this
was
not
actually
an
astrophysics
talk,
we
again
tried
to
get
a
feel
for
what
the
velocity
field
looks
like,
and
what
you
can
see
here
is.
This
is
a
cartoon
that
we
made
there.
B
Is
this
dipole
coming
out
and
it
turns
out
that,
yes,
the
ignition
does
happen
in
that
range
and
there's
a
bunch
of
circumferential
flow
at
the
outside
you've
got
some
buoyancy.
You've
got
a
little
bit
of
turbulence.
It
turns
out,
through
some
other
studies,
that
the
turbulence
is
not
as
big
of
a
factors
that
we
thought
it
might
be.
But
the
point
is
we
got
a
cartoon
of
how
this
works
and
I
hesitate.
B
The
shows,
because
it
was
not
a
nurse
simulation,
but
what
we
were
then
able
to
do
was
to
say:
okay,
let's
take
that
information,
the
ignition
radius
and
the
velocity
field.
Let's
use
that
to
initialize
the
castro
simulation,
which
is
the
compressible
code,
and
so
we
take
it
from
convection
to
break
out.
This
is
something
that
was
done
under
blue
waters,
grant
and
actually
initialized.
It
quote
correctly
to
say
how
do
these
things
blow
up
and
that's
when
it
was
done
by
chris
malone,
who
is
starting
at
los
alamos?
B
I
think,
as
we
speak
and
then
just
as
a
summary.
So
again
we
had
the
science
problem
said:
okay,
can't
do
it
with
the
existing
codes,
new
model,
new
code-
and
you
know
all
the
bells
of
whistles-
am
are
hybrid
parallel
and
then,
of
course,
once
you've
built
this
code,
you
say:
what
else
can
we
do,
and
so
that's
one
of
things
maestro
is
still
running.
B
We've
we've
put
the
particular
type
one
to
rest,
I
think
and
we're
looking
at
some
of
the
other
progenitor
simulation
of
progenitor
candidates
to
say:
can
those
create
the
type
ones
that
we
see
Chris
Malone's
again,
a
bunch
of
type
1
X
ray
bursts?
We
had
a
graduate
student
working
with
us
that
did
convection
and
main
sequence
stars
because
stars
are
always
conducting,
even
if
they
don't
blow
up,
and
you
would
like
to
understand
what
that
convection
looks
like.
B
A
B
C
B
It
is
a
it
is
a
process
right,
so
so
what
we?
Yes,
so
we
generate
the
initial
results
or
the
compressible
simulation.
The
compressible
simulation
generates
an
explosion,
the
thing
that
we
observe
the
light
curves
and
so
there's
another
step
which
to
say
take
the
simulation
results,
generate
the
light.
Curves
compare
those
two
specific
observations
that
part
of
the
puzzle.
We
have
have
not
been
involved
in
the
light.
Curves
and
kayson
does
a
lot
of
that.
So
we
have
not
followed
it
through
to
say
wow.
This
simulation
matches
this
supernova
observation.
B
B
So
we
have
not
what
we
would
ideally
do
if
they
kept
finding
us
to
do
astrophysics
is
to
create
that
end
to
end
where
we
could
do
a
simulation,
we
could
say
doesn't
match
any
of
these
actual
observations,
because
that's
the
whole
point,
that's
what
we're
doing.
Is
you
get
these
observations
and
you
want
to
use
those
to
tell
you
how
the
supernova
exploded
and
what
you
have
to
do
is
you
have
to
run
simulation,
say:
okay,
here's
one
version
doesn't
match:
okay,
we'll
tweak
it.
B
A
D
B
A
non-directional
right,
so
the
quite
the
thing
that
you
really
so
what
you
see
in
the
light
curves
is
how
much
of
the
star
has
burned
before
it's
loaded.
So
you
can
imagine
if
you
had
a
flame
propagate
out
from
the
center,
a
flame
that
went
out
like
this
burned
through
the
star
and
then
the
star
exploded.
Then
what
then,
the
star?
The
state
of
explosion
is
actually
very
different
than
if
you
maybe
shoot
something
shoot
a
burning
bubble
out.