►
From YouTube: In the Shadow of the Black Hole - GitHub Satellite 2019
Description
In this panel talk, six members of the EHT team will walk through the entire EHT experiment from the individual telescopes that record the data through the calibration, imaging, and interpretation of the observations that lead to the first-ever direct image of a black hole released to the world on April 10th of this year.
Presented by:
Andrew Chael, Scientist (Event Horizon Telescope)
Kazu Akiyama, Scientist (Event Horizon Telescope)
Sara Issaoun, Scientist (Event Horizon Telescopes)
Lindy Blackburn, Scientist (Event Horizon Telescope)
CK Chan, Scientist (Event Horizon Telescope)
Roman Gold, Scientist (Event Horizon Telescope)
About GitHub Satellite 2019
A community connected by code
Explore our interconnected community—and how collaboration turns ideas into innovations.
Join us in November at San Francisco's Palace of Fine Arts for GitHub Universe - https://githubuniverse.com/
A
A
Hello,
everybody
and
welcome
to
the
EHT
panel
on
the
event
horizon
telescope
results.
So
we
have
a
panel
of
experts
here
to
to
help
you
dive
into
the
shadow
of
the
black
hole.
So
first
let
me
introduce
you
the
event
horizon
telescope
collaboration.
These
are
a
hundred
twenty
over
two
hundred
plus
member
collaboration.
The
EHT
is
truly
global
and
that
its
membership
is
comprised
of
people
from
Europe,
Asia,
Africa,
North
and
South
America.
The
goal
of
the
event
horizon
telescope
project
is
to
image
for
the
first
time,
the
shadow
of
a
black
hole.
A
So
if
you
look
up
at
the
night
sky
down
in
the
Virgo
galaxy
in
the
Virgo
constellation,
there
is
a
galaxy
called
m87
in
the
M&E
seven
galaxies,
there's
a
supermassive
black
hole
that
has
the
second
largest
apparent
size
as
seen
from
Earth
in
this
galaxy.
There
is
a
powerful
jet
of
plasma
that
it's
believed
to
come
from
the
supermassive
black
hole.
This
jet
pierces
through
the
entire
galaxy,
can
be
seen
in
visible
light
and
in
radio
radio
waves
are
unimpeded
by
dust
with
them.
We
can.
A
We
can
peer
through
the
entire
galaxy,
follow
the
jet
down
to
its
heart,
as
we
go
to
higher
resolution,
and
higher
observing
frequency
would
go
to
the
base
of
the
jet
where
there
is
the
shadow
of
the
black
hole,
surrounded
by
a
ring
of
light
image
for
the
first
time
by
the
event
horizon
telescope,
so
to
be
able
to
image
a
black
hole.
We
need
to
have
high
resolution
to
make
an
image
of
the
features
we're
trying
to
see
about
55
million
light
years
away.
A
So,
in
order
to
get
since
resolution
scales
with
the
size
of
the
telescope,
we
need
a
telescope,
the
size
of
the
earth
to
be
able
to
see
the
shadow
of
the
black
hole
in
m87.
Unfortunately,
nobody
would
let
us
fund
that,
so
we
have
to
get
around
it
by
multiple
telescopes,
so
the
EHD
of
the
computational
telescope.
It's
not
actually
a
telescope,
the
size
of
the
earth.
We
use
many
small
telescopes
around
the
globe
at
different
large
distances
and
connect
them.
A
We
use
a
principle
called
very
long,
baseline,
interferometry
or
VLBI,
which
basically
helps
us
connect
the
telescopes
in
our
network
and
synthesize
a
telescope
virtual
telescope,
the
size
of
the
earth.
So
say
we
have
a
distant
source,
far
away
because
it's
so
far
away
and
the
radio
emission
it
emits
arrives
as
plane
waves
at
the
telescopes
because
of
the
geometry
of
the
earth
and
the
distances
between
the
telescope's.
There
is
a
time
difference
in
their
arrival
time
of
the
signal.
This
time.
Difference
is
actually
timed
really
accurately
by
atomic
clocks
or
me
tzer's.
A
These
clocks
actually
lose
a
second
every
hundred
million
years,
so
they're
extremely
precise.
So
by
timing,
the
arrival
of
the
signal
we
actually
timestamp
the
signal
at
each
telescope
and
record
it
onto
our
hard
drives.
We
essentially
freeze
the
light
that
is
arriving
at
each
telescope
record
it
and
then
combine
it
later
on
in
post-processing.
A
So
the
distances
between
the
telescope's
tell
us
about
the
source
structure.
We're
seeing
the
closer
telescopes
are
together
the
more
signal
they
can
see
in
common.
So
they
actually
tell
us
about
the
large-scale
structure
of
the
source.
As
you
take
telescopes
further
and
further
apart,
they
see
less
signal
in
common
and
so
tell
us
about
the
finer
scale
structure
of
the
of
the
source,
and
so
in
order
to
make
an
image,
you
need
a
large
range
of
scales
and
a
large
range
of
orientations
of
the
telescope's.
A
So
to
do
that,
you
need
to
have
many
telescopes
at
different
distances
and
different
orientations
on
the
earth.
So
the
EHD
observes
at
a
frequency
of
230
gigahertz
radio
waves
that
230
gigahertz,
unfortunately
can
get
scrambled
or
attenuated
by
water,
vapor
and
disturbance
in
the
atmosphere.
The
EHD
gets
around
that
by
borrowing
telescopes
that
are
very
high
in
dry
sites.
At
some
of
the
most
extreme
remote
locations
on
earth
here
are
some
of
the
telescope's
of
our
2017
campaign.
A
There's
the
SMT
in
the
mountains
of
Arizona,
the
JC
MT
on
Mauna
Kea
in
Hawaii,
the
LM
T
in
this
on
the
volcano,
sierra
negra
in
mexico,
the
elan
30
meters,
telescope
on
pico
Valletta
in
the
sierra
nevada
and
spain,
the
pecs
telescope
and
the
Alma
telescope
in
the
Atacama
Desert
in
Chile,
one
of
the
driest
deserts
on
earth,
the
SBT
at
the
geographic
South
Pole
in
Antarctica,
and
the
SMA
on
the
Amano
Kea
in
Hawaii.
As
you
can
see,
these
telescopes
were
extremely
different.
They
look
nothing
like
each
other.
A
So,
although
our
telescopes
have
different
designs
and
operations,
they
have
one
thing
in
common
or
actually
two
things.
One
thing
is
the
atomic
clocks
that
record
and
timestamp
the
signal,
and
then
we
have
our
amazing
EHT
custom
recording
equipment
with
this
recording
equipment
which
you
can
see
on
the
left
here.
This
signal
is
sampled
recorded
in
these
little
boxes
that
are
called
modules.
Each
module
is
stack,
is
a
stack
of
standard
hard
drives
in
our
2017
campaign.
We
use
92
of
these
modules
across
the
entire
array
or
736
standard
hard
drives.
A
During
our
observations,
we
have
to
keep
a
very
close
eye
on
the
weather,
so
we
use
our
VLBI
monitor,
which
actually
keeps
track
of
the
weather
and
forecast
at
each
site.
In
order
to
make
observations,
we
basically
need
to
keep
a
close
eye
that
all
the
weather
is
excellent
at
all
sites
to
guarantee
a
strong
signal
recorded
at
each
telescope
because
of
the
atmosphere
in
our
2017
campaign.
We
had
a
10
day
window
and
we
had
five
days
where
we
triggered
observations
with
the
best
weather
possible.
A
Of
course,
none
of
this
would
be
possible
without
our
instrumentation
hardware
and
software
teams
and
our
operators
and
observers
that
go
to
these
remote
sites
and
carry
out
our
observations,
test
our
equipment
and
gather
these
data
for
us.
So
now,
I'm
gonna
pass
on
to
Lindy.
Who
will
talk
to
you
about
how
we
get
this
recorded
data
and
transform
it
to
perfect
data
for
data
analysis?.
B
So
I'll
be
talking
about
the
data
pathway,
so
this
is
how
we,
from
these
signals,
recorded
telescopes
to
the
set
of
data
on
the
right.
That's
used
for
imaging,
and
this
is
a
bird's-eye
view
of
the
data.
So
we
begin
with
all
the
telescope's
pointed
at
the
same
source
recording
signals
at
the
same
time
straight
to
hard
disks.
B
You
know
about
a
thousand
hard
disks
as
Sarah
mentioned,
then
it
turns
out
the
fastest
way
to
get
them
all
the
same
place
is
physically
picking
them
up
and
moving
them,
so
he
shipped
them
all
to
our
correlation
Center
and
the
correlations
that
are
plays
back
this
frozen
light.
It
really
forms
the
virtual
mirror
here
of
our
telescope
in
a
very
fundamental
way
during
another
stage
of
calibration.
B
We
extract
those
signals
from
the
data
and
measure
their
properties
and,
finally,
after
this
we're
ready
for
forming
the
image,
and
this
is
where
the
image
is
actually
made
during
image
reconstruction.
One
of
the
stories
of
the
HT
is
just
the
massive
amount
of
data
we
processed.
So
you
know
that
the
image
that
you've
seen
now,
it's
pretty
pretty
famous.
It's
really
a
little
blurry,
you
know
maybe
characterised
by
about
a
hundred
pixels,
or
so
it
fits
on
a
kilobyte.
B
You
could
send
it
by
email
and
we're
beginning
with
petabytes
of
data
that
were
your
chords.
So
this
is
about
twelve
orders
of
magnitude.
Indeed,
a
reduction,
a
huge
amount,
and
it
really
reflects
the
amount
of
noise
we
have
to
mitigate
and
the
amount
of
complexity
we
have
to
remove
for
my
data
before
we
can
actually
make
the
image
so
stepping
back
a
little
bit
from
from
the
beginning
and
the
signals
I
want
to
explain
the
correlation
process,
because
it's
so
fundamental
to
the
way
interferometer
like
the
EHT
works.
B
Now
there
are
a
couple
things
that
make
this
process
hard,
and
this
is
why
this
is
not
so
straightforward.
One
is
that
the
correlation
coefficients
that
we
expect
are
actually
quite
small,
and
the
reason
is
because
the
signals
are
noise
dominated
there's,
a
lot
of
atmospheric
background,
thermal
noise
that
dominates
and
the
correlation
coefficient.
We
expect
this
could
be
one
part
in
a
million,
and
we
need
to
Everage
many
millions
of
samples
just
to
be
able
to
get
a
technical
signal.
B
To
top
of
one
of
these
panels
is
one
wrap
of
phase
and
a
230
gigahertz,
that's
about
4
Pico
seconds,
so
we're
really
talking
about
very
small
amounts
of
timing
jitter,
but
causing
a
very
significant
effect
in
our
data.
We
use
our
methods
to
model
this
out
and
get
something
nice
and
stable,
after
which
we
can
collapse
all
this
data
into
one
single
point.
So
this
is
one
example
of
how
the
data
is
simplified.
During
the
calibration
process,
we
developed
three
independent
pipelines
to
do
this.
This
is
a
theme
during
our
data
reduction.
B
Another
aspect
of
calibration-
that's
very
important,
is,
is
folding
in
the
sensitivity
of
the
stations,
so,
as
I
mentioned
before,
the
Kolar
correlation
coefficient
is
a
number
between
0
&
1,
but
to
translate
and
there's
something
physically
meaningful.
We
fold
into
sensitivity,
estimates
and
because
we're
using
such
different
stations
around
the
earth.
This
is
important
for
physically
interpreting
our
data.
This
is
just
one
representation.
B
The
data
before
before
processing
before
folding
in
sensitivity
and
what
we
have
here
is
a
degree
of
similarity
between
signals
recorded
to
sites
versus
the
length
of
the
baseline
between
the
stations
and
after
folding
in
station
sensitivities.
We
get
something
that
reflects
a
very
simple
and
elegant
structure
to
the
data
and
if
we
overlay
what
we
expect
on
this
spectrum
of
the
data
from
just
a
simple
ring
model
of
a
source
on
the
sky,
we
see
that
it's
a
pretty
good
match.
So
we
have
this
amplitude
spectrum.
C
So
so
Sarah
and
Mindy
provided
an
excellent
introduction
of
how
the
EHC
works.
So
our
HD
is
assembling
eight
Tests
of
in
this
planet
that
create
a
planet
sized
telescope
with
incredibly
high
on
your
resolution,
capable
of
imaging
a
black
hole
for
the
first
time.
So
now
you
can
see
the
the
RS
seen
from
m87
and
you
can
see
that
with
art
notation.
A
new
telescope
is
joining
the
array
and
you
can
see
that
a
battery
our
size
telescope
have
a
universal
mirrors
with
the
off
notations
and
we
got
a
corresponding
data
set.
C
But
you
can
also
realize
that,
because
we
have
a
limited
number
of
station
in
this
planet,
so
our
virtual
mirror
is
really
sparsely
sampled.
So
that
create
a
situation
that
so
there
are
infinite
number
of
images
that
can
actually
perfectly
explain
our
data
set.
So
at
the
imaging
process,
each
imaging
technique
or
each
managing
algorithm
that
create
image
from
observing
data
set
delay,
try
to
derive
the
most
likely
image
from
the
sparsely
sampled
data
set
based
on
its
prior
knowledge
or
its
prior
assumptions.
C
So
in
the
history
of
the
radio
astronomy,
their
new
two
classes
of
imaging
techniques,
the
first
one
is
the
most
popular
and
the
most
traditional
technique,
so
called
the
Queen.
The
queen
is
an
inverse
modeling
approach
that
that
try
to
drive
a
reasonable
solution
from
from
the
image
domain
in
an
interactive
way
by
the
input
of
users,
but
Creon
has
a
long
history
of
more
than
40
years
and
it
has
been
tested
by
data
and
used
by
community
for
a
long
time.
C
That's
of
really
good
advantage,
on
the
other
hand,
that
interactive
nature
of
the
Korean
actually
needs
a
lot
of
user
input
and
also
expertise
of
some
professionals,
knowledges
and,
and
also
it's
also
a
little
bit
difficult
to
incorporate
a
many
of
serving
effect
directly
in
the
imaging
process.
So
that's
why,
if
we
develop
another
class
of
new
imaging
techniques
that
take
afford
millennia
loads,
so
that
is
a
Bayesian
inspired
method
that
can
include
a
lot
of
prior
assumptions
of
or
prior
notices
of,
the
images
with
sometimes
combination
of
them.
C
We
can
also
incorporate
a
lot
of
observing
effect
of
suffering
instance.
What
are
the
systematic
errors
remain?
Even
after
the
calculations,
so
the
new
imaging
technique
is
more
flexible,
imaging
process
that
can
match
it
more
with
the
actual
observing
process.
So
such
a
development
of
new
imaging
techniques
is
actually
one
of
the
prime
innovation
in
even
HD
corporations.
So
there
are
a
number
of
imaging
techniques
that
develop
in
the
last
several
years
to
make
the
far
shaver,
buccal
images
and
most
of
images.
Most
of
imaging
techniques
are
implemented
in
two
independent
imaging
libraries.
C
The
first
one
is
a
huge
team
urging
library
and
another
one
is
a
smiley,
so
both
two
new
libraries
are
a
completely
open
source
and
Python
interface
and
indeed
have
and
and
as
not
mentioned
in
the
beginning
of
this
conference.
Actually
this
two
new
imaging
libraries
are
using
a
lot
of
libraries
actually
developed
by
many
communities.
So
that
means
you
know.
The
number
of
people
in
many
communities
actually
contributed
to
make
these
two
new
imaging
libraries
that
actually
created
the
first
ever
images
of
buccal.
This
is
a
really
really
cool
thing.
C
On
the
other
hand,
the
latter
two
teams
have
a
modest
focus
on
the
traditional
techniques
anyway,
so
each
four
team
actually
worked
on
m87
data
independently
without
any
exchange
of
information
with
outside
of
the
team
and
seven
weeks
later,
we
united
together
at
the
Harvard
University
at
the
end
of
July,
2018
and-
and
this
photo
actually
is
showing
the
exact
moment
when
we
compared
images
across
four
teams.
For
the
first
time-
and
you
cannot
see-
is
a
hook,
unfortunately,
but
you
can
see.
C
Actually
what
are
the
results
from
the
people's
face,
particularly
you
know,
Tate
is
really
really
small
a
face.
You
know
in
front
of
participants,
so
here
is
the
first
images
compared
across
the
four
teams,
so
these
images
are
made
of
made
from
the
engineering
data
Lele
so
that
are
not
coming
from
the
final
data
data
have
still
some
issues
but
but
being
actually
got
a
very
consistent
basic
structure.
So,
for
instance,
all
images
have
a
link
with
a
very
consistent
diameter
and
then
also
this
Ning
is
not
symmetric.
C
You
know
you
can
you
can
see
that
the
clear
asymmetry
where
the
lower
side
of
a
southern
side
is
wider
and
this
basic
structure
is
not
is
even
seen
in
the
you
know,
our
final
images
that
person
is
on
April
10th,
so
we
got
it
well,
basically,
at
the
basic
structure.
At
this
time
with
engineering
in
data,
we
have
independent
four
teams
that
actually
provide
a
certain
confidence
about
our
results.
So
this
is
another
memorial
photo
taken
after
the
first
comparison
of
images.
C
You
can
see
a
single
m87
image
here,
so
this
is
the
average
of
all
four
teams
images.
This
is
actually
really
similar
to
what
we
presented
in
a
press
conference
and
then
also
you
can
also
see
you
know
fantastic
international
team
that
actually
spent
several
years
to
develop
new
imaging
techniques
and
also
work
hard
to
create
the
first-ever
picture
of
all
black
holes.
So
this
is
a
tree
acrobatic,
relational
effort
and
and
obviously
we
are
proud
of
our
fantastic
imaging
team.
D
Thanks
Kazu
so
yeah,
once
we
had
done
this
imaging
process
with
four
different
blind
imaging
teams,
we
have
a
pretty
good
confidence
that
we
weren't
sort
of
deluding
ourselves
into
thinking
that
this
image
was
there
in
the
data
and
that
different
imaging
methods,
designs
with
different
techniques,
could
reproduce
the
same
fundamental
image.
But
then
we
still
have
the
question
of
how
is
we
best
tune?
The
knobs
on
our
virtual
camera
and
to
get
us
the
best
possible
image
or
the
most
reliable
image
from
all
these
different
algorithms
I
mean.
D
So
we
have
three
different
software
packages
that
we
used
in
this
next
step
of
imaging
one
based
on
the
traditional
clean
algorithm
cause
we
described,
and
then
the
two
that
were
developed
specifically
for
the
HT
and
really
built
on
all
of
this
open-source
Python
software
is
cause
we
describe.
She
wanted
to
objectively
strive
determine
what
the
best
way
to
tune
all
these
different
knobs
in
the
the
image
would
be
so.
The
way
that
we
did
this
is
we
tested
tens
of
thousands
of
different
imaging
parameter
combinations
systematically.
D
So
we
did
this
on
a
suite
of
synthetic
source
images
and
we
demanded
that
our
different
imaging
parameters
could
really
distinguish
between
these
sources
and
all
the
different
ways
that
these
sources
might
be
fooling
us
into
giving
us
similar
images.
We
only
chose
the
parameters
to
apply
to
m87
that
were
the
most
trustworthy
at
the
end
of
the
day,
while
these
dip
imaging
methods
have
different
resolutions
and
socially
slightly
different
images,
when
we
blur
them
to
a
fundamental
equivalent
resolution,
we
get
very
similar
images,
I'm,
really
giving
us
confidence
that
these
images
are
robust.
D
So
all
the
different
imaging
choices
that
we've
made
and
then
these
images
combine
together.
We
actually
simply
average
them
to
give
us
the
final
image
of
m87
that
we
share
with
the
world
on
April
10th.
So
the
next
part
of
my
discussion,
I
talked
a
little
bit
on
measuring
the
mass
and
the
size
of
the
black
hole,
so
it
turns
out
that
measuring
black
hole
masks
actually
very
difficult
on,
because
you're
really
indirectly
looking
at
it.
D
So
the
best
way
to
can
do
it
is
with
these
direct
images
in
previous
methods,
have
really
disagreed
and
are
very,
very
difficult
to
do
so.
The
massive
black
hole
is
proportional
to
the
size
of
its
shadow.
So
if
we
can
measure
the
size
of
the
shadow,
we
could
really
accurately
measure
the
de
mass
and
the
exact
value
depends
on
the
details
on
the
surrounding
material
and
on
the
calibration.
So
we
really
needed
to
approach
this
problem
systematically
and
the
way
that
we
did
this.
D
It's
a
common
theme
in
the
EHT,
whose
three
different
imaging
techniques
to
measure
three
different
techniques
to
measure
the
size
of
the
shadow.
So
we
could
really
verify
that
we're
doing
this
correctly
and
that
our
answer
made
sense,
the
first
ones
we
can
just
measure
directly
from
the
image.
We
can
just
take
a
circle
around
the
ring
and
measure
the
size
this
way
and
an
uncertainty
on
that
size.
The
next
is,
we
can
model
the
data
using
simple,
geometric
shapes.
D
So
in
this
framework,
we
basically
decompose
the
image
into
a
simple
collection
of
geometric
shapes
and
fit
these
shapes
on
to
the
to
the
underlying
data,
and
we
do
this
using
a
sort
of
Bayesian
sampling
and
parameter
estimation
framework.
So
just
this
one,
which
is
really
samples.
The
stayce
of
uncertainties
of
the
different
parameters
of
our
image
that
we're
sampling
in
the
third
ways
that
we
can
fit
individual
frames
from
supercomputer
simulations.
D
That
Roman
will
talk
about
next
to
the
underlying
image,
and
so
this
week,
yeah,
and
so
all
three
of
these
different
methods
actually
use
three
new
different
software
packages
that
were
designed
the
first
to
design
for
the
EHT.
The
first
modular
MCMC
framework
called
temas.
That
roman
will
talk
about
a
second
gen,
a
which
relies
on
a
machine
learning,
algorithms,
genetic
algorithms
and
then
dynasty.
Actually
is
a
software
package
developed
by
a
grass
root
at
Harvard
that
we
and
he's
designed
for
a
completely
different
problem.
D
D
E
All
right,
thanks
Andrew,
so
so
I
think
now
you've
gotten
a
really
good
idea
and
a
good
appreciation
for
how
difficult
and
challenging
it
is
to
to
make
these
measurements
to
then
understand,
calibrate
and
therefore
reduce
the
data
and
then
to
reconstruct
an
image
right
and
now.
I
would
like
to
pose
to
you
a
couple
of
additional
questions
that
you
might
be
interested
in.
So
let's
take
a
look
at
this
image
again
and
I
would
like
to
start
at
square
one
and
ask
what
do
you
see
in
this
image?
E
E
Otherwise,
it
wouldn't
be
a
ring,
but
now
you've
discussed
features
of
the
image,
but
now
the
questions
that
that
that
theorists
are
posing
themselves
is:
how
do
we
know
that
it's
a
black
hole
or
why
does
the
image
look
the
way
it
does,
or
in
other
words,
what's
actually
going
on
that
gives
rise
to
this
to
this
emission
and,
of
course,
the
theoretical
implicit
implications?
What
does
this
all
mean
and
for
all
those
three
questions
and
related
questions?
E
You
need
a
theory,
so
Andrew
in
the
previous
talk
told
you
already
one
important
ingredient
in
answering
question
number
one:
how
do
we
know
it's
a
black
hole?
A
very
important
ingredient
is
well.
You
first
have
to
measure
the
data,
but
then
second,
you
need
a
theoretical
prediction.
Andrew
told
you
that
the
mass
of
the
black
hole
is
related
to
the
angular
size,
but
that's
only
true,
because
you've
assumed
a
theory
that
makes
this
prediction
without
this
theoretical
prediction.
E
You
you
look
at
this
image
in
a
completely
different
way
and
it's
much
harder
to
gain
any
confidence
in
the
fact
that
you're
actually
staring
at
a
black
hole.
It
could
be
an
exploding
envelope
of
a
star,
for
example,
or
something
like
that.
But
there
is
a
theory
that
tells
you,
if
you
have
a
mass,
if
you,
if
you
have
a
black
hole
of
a
certain
mass,
it
should
have
this
size.
Now.
Why
does
the
image
look?
The
way
it
does?
E
E
So
one
of
the
strange
and
perhaps
defining
properties
of
black
holes
is
what
we
call
the
event
horizon
a
surface
of
no
return'
a
surface
where,
if
you
dip
things
into
it,
they
are
lost
to
the
exterior
world
forever,
even
light.
So,
if
light
gets
too
close
to
a
black
hole,
it
gets
absorbed,
falls
into
the
event
horizon
and
never
gets
out.
So
that's
very
useful.
If
you're
trying
to
explain
a
dark
region
in
an
image-
okay,
it's
not
so
great
if
you
have
a
bright
ring
around.
So
what
about
that
bright
ring?
E
E
So
you
see
in
this
lower
image
that
these
light
rays
that
are
getting
closer
and
closer
to
the
black
hole
and
the
closer
they
get
the
more
they're
deflected
by
the
black
hole
until
they
get
too
close,
and
then
they
get
absorbed
by
the
black
hole
and
that's
what's
giving
rise
to
what
we
call
a
shadow,
the
lens
structure
of
the
event
horizon
that
we
see
it
infinity
in
our
telescopes
all
right.
So
so
what
what?
What
happens
around
this
black
hole?
What's
actually
physically
going
on
that
gives
rise
to
this
emission?
E
Well,
let
me
take
you
a
second
time
on
a
journey
to
m87
we're
going
to
zoom
in
and
we're
going
to
go
through
the
same
scales
that
you
saw
before
now,
we're
approaching
scales.
We
can
see
with
a
Hubble
telescope,
this
fantastically
bright
jet
region.
That
points
you
to
the
center
of
the
galaxy,
and
now
you
don't
see
an
image,
but
you
see
a
simulation
that
shows
you
what
happens
to
the
gas
as
it
orbits
the
black
hole
and
as
it
falls
onto
the
black
hole,
it
turns
out.
It's
a
very
turbulent,
very
wild
environment.
E
E
You
see
in
the
middle,
an
image,
a
theoretical
image
that
was
generated
from
one
of
the
simulations
you
just
saw,
and
you
see
on
the
right,
a
blurred
version
of
that
theoretical
image,
and
the
amazing
thing
is
that
the
observed
image
and
the
blurred
version
of
the
theory
image
look
very
similar,
and
that
gives
us
a
great
degree
of
confidence
that
we
have
some
idea
of.
What's
going
on.
E
So
you
have
to
turn
the
crank
in
comparing
theory
and
data,
and
you
have
to
do
this
many
many
times,
and
here
you
see
a
whole
library
of
these
simulations
that
I
showed
you
and
each
one
of
those
takes
a
week
or
two
to
run
on
thousands
of
computational
cores,
and
you
have
to
turn
these
into
images.
Many
images
that
depend
on
viewing
angle
and
spin
on
magnetic
flux
and
all
kinds
of
things,
and
you
need
to
very
and
constantly
compare
data
and
theory
to
figure
out.
E
E
Okay,
so
here's
here's
a
view
of
demos
on
github.
Obviously,
it's
very
useful
for
us
to
track
changes
in
the
code,
monitor
progress
and
even
helps
with
validation
of
the
code
and
so
on,
so
about
the
theoretical
implications.
Just
to
summarize
real
quick,
so
the
first
ever
black
hole
image
and
in
particular
the
agreement
of
theory
and
data,
is
actually
a
game-changer
for
understanding
of
gravity
and
Jets,
and
so
far
Einstein's
theory
seems
to
be
supported
by
the
data
and
also
the
black
hole.
Nature
of
the
source
is
supported.
E
In
fact,
we
tried
some
black
hole,
mimickers
that
are
not
quite
black
holes,
but
behave
in
a
similar
way
and
they're
disfavored
by
the
data,
and
the
question
is:
can
we
look
for
deviations
from
Einstein's
theory,
and
this
is
something
that
the
next
speaker
CK
is
going
to
tell
you
more
about
and
also
about
some
of
the
other
codes
and
software
that
were
used
in
the
process?
Thank
you.
F
Okay,
so
in
this
last
few
minutes,
I
want
to
go
back
to
a
few
very
simple
points.
The
first
one
years,
the
event
horizon
telescope
is
a
scientific
experiment.
So
you're
taking
the
image
of
the
black
hole
is
just
the
first
step
in
an
era
of
new
gravity
experiments.
So
the
big
question
here
is
how
gravity
works
and
if
Einstein
is
correct,
a
lot
and
it
turns
out
you're,
given
all
the
confirmation
that
Roman
just
told
us
there
is
a
signature
in
this
black
hole
image
that
can
answer
these
big
questions.
F
So
you
know
I
want
to
show
you
a
software
that
we
developed
a
few
years
ago
that
we
follow
the
geodesics
of
photon
or
on
curved
space-time.
So
the
idea
there,
if
you
know,
if
we
send
a
quarter
million
photons
to
a
black
hole,
we
will
see
your
the
algorithm
will
integrate
here
according
to
Einstein's
equations,
and
some
of
them
will
hit
the
UN
horizon
and
got
stuck
there.
F
Some
of
them
you
passed
around
the
black
hole,
you
walk
around
a
black
hole
a
few
times,
and
some
of
them
will
just
fight
through,
and
you
know
just
to
give
you
a
perspective
in
order
to
create
an
image
in
the
interstellar
movie.
It
takes
a
hunger
CPU
hours
to
render
a
single
frame.
But
with
this
software
that
we
develop
fans
for
the
GPU
technology,
we
are
able
to
compute
a
single
image
in
three
seconds
and
using
this
kind
of
calculation
and
follow
how
like
orbit
around
black
hole.
F
We
realized
that,
if
I
stein
is
correct,
it
turns
out
that
the
shape
of
the
black
hole
shadow
is
independent
of
the
black
hole's
being
and
the
inclination
angle
that
we
look
at
it
and
it's
always
a
circle.
So
this
is
a
very
unique
signature
that
we
can
use
and
there
are
other
kind
of
your
gravity
theories
out
there
and
they
actually
predict
where
different
shape
of
the
black
hole
shadow.
So
your
email
collaboration,
many
of
us,
actually
want
to
disprove
Einstein.
F
But
the
fact
that
you
know
the
black
hole
shadow
is
consistent
with
the
ship
with
a
circle
is
indicating
that
Einstein's
right
again,
but
it's
always
right
but
you're
in
the
future.
In
the
future,
when
we
add
more
telescopes
to
our
way,
you
maybe
even
go
to
space,
we
will
blue.
We
will
be
able
to
place
your
tighter
constraint
on
general
relativity
and
understand
gravity
better
and
maybe
at
some
point
we
will
you'll
be
able
to
tell
which
part
is
wrong.
F
F
So,
thanks
again
and
at
the
end
of
the
day,
yoky
helped
bring
everything
together
to
form
a
community
and
in
that
sense
the
yeast
here
in
scientist
in
you,
we
are
just
part
of
the
community
and
we
are
trying
our
best
to
contribute
back
and
pick
our
software
open-source.
So
this
is
smiley
that
castle
showed
you
earlier.
F
Carthoris
could
be
in
this
software,
and
this
is
HD
imaging
and
who
is
meeting
there
and
Lindy,
and
myself
and
other
people
in
the
East
here
contributing
the
one
beautiful
thing
of
putting
this
on
github
is
we
receive
contribution
from
people
that
will
ever
met?
They
may
not
be
astronomer,
but
they
are
expert
in
Python
the
expert
in
dr.
and
their
contribution
helped
us
improve
the
software
and,
at
the
same
time,
our
team
also
used
keep
to
version
control
our
data.
The
result
is
our
data
analysis
of
reproducible.
F
F
And
create
your
own
m87
image
and
actually
they're
already
teachers
out
there,
use
it
software
and
data
and
put
them
into
astronomy
classes,
and
we
are
here
just
where
we
excite
you
to
see
how
this
can
inspire
a
new
generation
of
scientists.
So
the
last
thing
I
want
to
say
is
your
science
and
technology
are
converging.
The
next
big
discovery
is
going
to
happen
at
the
intersection
of
science
and
technology
and
as
open-source
software
developer
of
you,
a
contribution
of
you're
contributing
to
this,
and
all
of
you
a
part
of
this.
A
A
A
C
C
So
so,
for
instance,
if
we
launch
a
satellite
on
the
mid
Earth's
orbit
where
the
GPS
satellites
are
then
actually
we
can
increase
angular
resolution
by
factor
3
or
4
that
actually
provide
maybe
another
four
eyes
and
scale
images
of
other
supermassive
local
in
other
galaxies,
or
we
can
also
get
much
finer
images
for
the
m87
yeah.
So
that's
actually
one
of
our
scope
in
the
next
decade.
G
A
So
actually
so
we
take
into
account
the
rotation
of
the
earth
when
we
do
our
calibrations
or
when
we
combine
the
data
together,
we
actually
take
into
account
how
the
earth
moves
and
the
telescope's
move
apparent
to
each
other,
but
the
Earth's
rotation
synthesis
which
we
use
in
imaging,
actually
help
us
fill
our
coverage.
So
we
basically
use
the
earth
as
our
telescope
to
basically
rotate
beneath
our
feet
and
kind
of
with
it.
A
E
So
you
can
really
pinch
through
it
and
you
can
get
this
nice
ring
and
we've
looked
at
other
sources,
as
you
may
have
heard,
but
this
one
was
the
one
that
that
looked
promising
right
from
the
start,
as
you
could
see
the
moments
and
the
slides
it
just
looked
very
promising,
and
so
we
focused
all
of
our
attention
and
it's
been
a
lot
of
work,
but
there's
more
to
come
for
sure.
Yes,.
A
We
observed
another
black
hole,
Sagittarius
a
star
which
is
at
the
center
of
our
own
galaxy.
It's
the
supermassive
black
hole
in
our
Milky
Way,
but
that
one
is
a
lot
smaller.
So
it's
a
lot
more
variables.
M87
is
like
taking
a
picture
of
a
person.
It's
a
long.
Exposure
of
a
person
ready
to
take
a
portrait
sign
a
star
is
basically
taking
a
picture
of
a
running
toddler
in
long
exposure.
So
that's
the
kind
of
challenge
we're
dealing
with
next
is
learning
how
to
make
movies
and
we're
going
to
need
a
lot
more.