►
From YouTube: NUG Meeting 2014: Persson
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
Last
December
Scientific
American
devoted
issued
two
world-changing
ideas
and
they
are
right
on
the
front.
Cover
was
an
article
co-authored
by
our
next
speaker,
entitled
how
super
computers
will
yield
a
golden
age
of
material
science,
so
not
to
put
any
pressure
on
earth
but
I'm
pleased
to
introduce
Kristen
person.
You
will
tell
us
about
her
work
with
materials
project.
First,
thanks.
B
B
So
engineered
and
designed
materials
surround
us
and
they
enable
our
society.
There
is
a
reason
we
call
our
ages
by
materials,
the
Stone
Age,
the
Bronze
Age,
the
petroleum
age.
If
you
look
at
the
last
couple
of
decades,
we
now
make
airplanes
and
bridges
out
of
carbon
fiber
composites,
which
makes
them
lighter
easier
to
deal
with
and
work
arose
corrosion
resistant.
We
wouldn't
have
the
communication
age
that
we
have
today
the
Internet
age.
If
we
didn't
hadn't
come
up
with
materials
that
transmit
signals
across
long
distances,
without
changing
them
or
perturbing
them.
B
These
are
high
purity
glasses
in
fiber
optics
and
one
of
my
favorites
wenn
du
Pont
was
charged
with
coming
up
with
a
material
that
was
lightweight
and
strong
enough
for
better
parachutes.
They
developed
nylon,
which
of
course
was
quite
the
consumer
hits
as
well,
so
these
materials
changed
the
way
we
do
things.
But
how
do
we
and
we
come
up
with
them?
How
do
we
design
new
materials?
Well,
there's
the
edison
style,
and
it
was.
B
It
was
called
that
way
because,
when
Edison
was
looking
for
a
filament
for
his
new
invention,
the
light
bulb
he
painstakingly
went
through
three
thousand
materials
and
he
ordered
them
from
all
over
the
world.
They
took
him
years
and
he
used
his
intuition
right
and
whatever
was
available
around
him.
He
tested
three
thousand
of
them,
literally
one
by
one.
The
sad
thing
is,
he
didn't
actually
find
the
best
one
out
of
all
those
materials.
B
The
best
performing
one
was
the
cotton
based
bamboo
derived
material
and
a
cotton
of
a
decade
later,
some
Hungarian
scientists
patented
the
tungsten
filament.
Instead
that
we
use
even
today,
so
that's
the
edisonian
approach.
It
takes
a
quite
surprising
15
to
18
years
from
the
time
and
material
works
in
the
lab
to
being
commercially
successful,
and
that's
when
it
works.
That's
when
most
scientists
go
well,
I've
done
it.
I've
published
and
I
patent.
B
It
and
I'm
not
sending
it
off
to
industry
and
I'm
going
to
be
rich,
and
then
they
wait
and
they
wait
and
they
wait
and
it
takes
industry
a
very,
very,
very
long
time
to
actually-
and
in
some
cases
it
doesn't
even
work,
and
why
is
this?
Well,
it
tends.
It
turns
out
that
it's
because
there's
always
something
wrong
with
the
material
there's
always
something
they
have
to
engineer
around.
It
doesn't
quite
work
the
way
it
was
supposed
to
it.
B
It
corrodes
in
the
environments,
you
know
when
we
subjected
to
temperature
or
water
or
something
around
us
that
is
not
available
at
it.
That
is
not
in
the
controlled
environment
of
the
lab,
or
there
is
no
market
for
it.
Superconductors
were
invented
in
the
70s
and
we're
still
not.
We
still
don't
have
a
large
market
for
those
materials.
So
how
can
we
accelerate
this?
B
How
can
we
get
more
information
about
material
up
front,
so
we
either
don't
pick
the
ones
that
will
have
this
very
long
time
line
or
we
make
better
choices
depending
on
the
information
we
get
well
we're
a
little
bit
better
off
than
Edison
in
the
sense
that
some
things
that
happens
in
this
day,
we
today
have
the
physical,
the
laws
of
physics
that
tell
us
how
material
actually
behave.
We
can
compute.
B
It's
a
thought
provoking
that
if
I
had
all
of
hopper
for
one
day,
I
could
crunch
through
20,000
materials
and
one
property
like
the
basic
energy
of
the
material
I
never
get
all
of
hopper
wink
wink,
but
but
if
I
did
I
could
actually
crunch
through
a
significant
portion
of
the
known
materials
today,
so
that
it's
a
tractable
problem.
It's
not
like
biology,
even
though
they're
using
fewer
building
blocks
that
we
do,
we
have
the
whole
periodic
table.
They
have
more
combinatorial
problems,
because
these
molecules
can
go
from
yield
can
become
very
different.
B
The
known
world
of
inorganic
materials
is
somewhere
between
30
to
50,000,
it's
an
it's
a
confined
space,
and
then
you
can,
you
can
do
substitutions
and
you
can
try
to
find
the
madam
metastable
materials
that
live
outside
of
those
known
materials,
but
still
it's
not
the
order
of
millions.
This
is
a
tractable
problem.
It's
doable
so
going
from
the
edisonian
approach,
where
we
test
one
material,
one
material,
one
material.
We
have
to
synthesize
it
tested
since
the
start
of
justice
since
I
tested
until
we
hopefully
one
day,
come
up
with
the
best
one.
B
We
could
actually
try
all
these
materials
in
the
computer
throw
away
most
of
them
because
that's
they're
not
going
to
work
and
then
hopefully
give
more
a
better
selection
of
candidates
to
the
people
in
the
lab,
who
can
then
make
them
and
test
them
and-
and
the
idea
is
that
this
would
be
a
better
selection
than
going
through
all
them,
and
it
will
take
way
shorter
time,
so
I
wouldn't
be
standing
here
at
this
point,
everybody
goes
like.
Oh,
this
is
a
great
idea,
but
does
it
really
work?
Does
it
really?
Actually?
B
Is
it
just
one
of
those
blue
in
the
sky,
ideas
that
you
know?
Okay,
we
a
cell,
it
will
get
some
money
for
it
and
see
if
it
works,
so
we
tried
it
out
with
a
company.
Quite
quite
some
years
ago,
in
2004
we
built
the
first
database
in
the
first
way
of
doing
this.
High-Throughput
screening
of
materials
properties
at
the
order
of
duracell,
that's
owned
by
Procter
&
Gamble
that
pretty
much
owns
a
large
part
of
the
consumer
market
in
the
u.s..
B
They
came
to
us
that
MIT
I
was
at
MIT
at
the
time
and
they
asked
us.
Is
they
sell
these
guys,
these
alkaline
batteries
and
it's
a
very
old
chemistry?
They
make
a
lot
of
money
on
this
market
and
they
wanted
to
know.
Is
there
anything
better
out
there
is?
Are
any
of
our
competitors
like
energized,
are
going
to
come
up
with
something
as
steel
that
market
share
from
us?
B
So
they
want
to
know.
Can
you
screen?
Can
you
find
out
if
there's
a
better
caso
de
out
there
we'd
like
a
drop-in
replacement?
We
want
the
same
stuff,
the
same,
a
no
the
same
electrolyte,
but
we
like
a
better
anode
or
cathode.
It's
pretty
coming
so
we
said
sure
give
us
a
million
dollars
in
a
year
and
we'll
do
it.
B
So
they
did
that
and
we
assemble
a
team
of
seven
people
that
worked
on
it
and
during
that
year
we
screened
more
than
a
hundred
thousand
compounds,
not
all
by
first
principles,
not
all
quantum
mechanics,
but
we
generated
new
test
sets
and
we
learn
from
those.
We
did
data
mining
on
more
suggestions
and
then
we
submitted
more
and
more
calculations
in
the
end,
fifteen
hundred
of
those
100,000
were
better
in
terms
of
the
energy
they
could
store,
but
only
200
of
those
had
any
chance
of
surviving
in
the
environment
for
which
they
were
designed.
B
These
are
alkaline
batteries.
Its
pH
15
inside
of
those
the
electrolyte
is
pretty
corrosive,
is
basically
a
Drano
and
you
know
Drano
dissolves,
most
stuff
right,
that's
the
same
with
these
ceramics,
you
drop
the
material
in
there.
It
might
survive
for
a
bit,
but
if
you're
putting
your
battery
away
you'd
like
to
come
back
a
year
later
and
it
would
still
work
right,
the
shelf
life
is
important.
So,
in
the
end
200
of
those
hundred
thousand
we
suggested
to
Duracell.
These
would
be
good.
Candidates
go
and
try
to
make
them
and
then
test
them.
B
I
gotta
say
that
these
high
throughput
searches
they
typically
don't
spit
out
the
best
candidate
on
day,
one
or
even
on
day.
Two
hundred
and
fifty
what
you
learn
is
design
rules.
We
learn,
for
example,
how
to
stabilize
high
valent
compounds
bismuth
five
nickel
for
iron
six
manganese.
Seven.
These
things
are
typically
not
stable
and
estimate
a
lot
of
pH
15.
But
if
you
add
other
elements
to
them,
they
become
more
stable
than
otherwise
they
would
have
been.
These
are
the
green
spots
here.
It
tells
us
which
element
we
should
combine
them
with.
B
It
tended
to
be
low,
valence
large
cations,
like
lithium,
calcium.
This
data
mining
map
shows
us
that
everything
pretty
much
dissolves
at
ph
15,
all
the
blue
stuff,
the
more
blue
the
worse.
But
there
are
some
white
spots
and
those
two
those
white
spots
sold
as
well.
These
elements
are
less
prone
to
the
solution
and
the
other
ones.
So
if
we
add
a
little
bit
of
those,
that
may
actually
be
a
better
compound,
and
this
is
how
we
learn
and
we
generated
new
tests
that
new
candidate
tested
that
in
the
end,
gave
us
better
compounds.
B
So
in
2005
we
gave
the
set
of
200
compounds
to
Duracell,
they
patented
a
bunch
of
them
and
they
basically
blanketed
the
space.
These
patents
are
extremely
broad.
You
don't
really
learn
that
much
for
looking
at
them.
It's
everything
from
the
kitchen
sink
together
with
this
mat
5
and
is
everything
in
the
kitchen
sink
together
with
nickel
for
but
there
are
some
clues
in
there
and
they
are
actually
working
on
one
of
these
compounds
today.
I
know
that,
because
I'm
still
working
with
them,
it's
a
pretty
cool
compound.
B
It
was
among
those
200
it's
about
two
to
two
and
a
half
times
better
than
the
manganese
dioxide
is
in
your
alkaline,
and
the
last
thing
that
they're
working
on
now
is
the
shelf
life.
It
really
does
work
in
the
cell,
so
we'll
see
the
clock
is
ticking
right,
those
15
years
2005
I'm
holding
myself,
because
this
would
be
a
good
story
if
it
really
did
work.
In
the
end.
B
The
same
idea
of
screening
materials
using
high-throughput
computations
at
the
MIT
I
was
moving
to
Berkeley.
At
that
time,
they're
mighty
crowd
sold
the
same
idea
to
Boston
umicore
screening,
now
cathodes
rechargeable
cathodes
for
lithium
ion
batteries,
and
many
new
compounds
came
out
of
that.
This
is
the
carbonyl
phosphate
family,
which
is
a
completely
new
chemistry
and
I
challenge
any
battery
chemist.
Who
would
have
come
up
with
that
because
it
completely
new
anion-
and
this
is
the
benefit
also
another
one
of
using
computers,
they're
agnostic.
They
wouldn't
tell
us
that.
B
Oh
you
know
what
this
material
like.
This
one
has
never
worked
for.
Lucy
mine
batteries
before
don't
even
try
it.
The
computer
tries
everything
right
and
they
may
spit
out
things
that
are
outside
of
our
realm
of
intuition
and
that's
the
benefit
of
it.
It
gives
us
an
unbiased
view
of
the
world.
Another
one
that
came
out
was
a
vanadium
phosphate
that
was
designed
by
an
above
Jane.
B
So
this
is
great
right.
The
our
team
is
showing,
to
some
degree
that
we
can.
We
can
design
and
predict
new
materials
using
a
database
driven
or
a
data-driven
approach,
but
we
can
do
better
than
this.
We
we
live
in
the
information
age
and
if
we
give
material,
if
we
give
information
to
people,
they
will
do
stuff
with
it.
That
is
outside
of
the
intuition
of
these
two
groups
at
MIT
and
lbnl.
B
If
you
were
to
charge
me
if
I'm
a
material
scientist
that
works
in
an
industry
setting
or
in
academia
and
I
get
access
to
data.
Well,
you
know
how
would
I,
how
would
I
go
about
looking
for
something
that
is
outside
that
I?
Don't
have
accesses
today?
Well,
you
know
if
I
look
for
information,
I
know
where
to
go.
I,
google,
it
if
I,
don't
know
anything
if
I
want
to
know
about
a
restaurant.
B
If
I
know
a
little
bit
more,
if
I,
actually
am
a
battery
scientist
I
might
say
well,
I
need
a
bad
material
that
is
between
3
to
5
volts,
good
stability,
a
good
lithium
mobility,
but
Google
doesn't
do
this
very
well
for
you,
I've
actually
tried
it.
Oh
there
comes
up
a
bunch
of
page.
You
just
gives
me
some
some
mighty
of
something,
but
it
really
doesn't
give
me
a
list
of
materials
that
would
work
well
for
this.
So
this
is
what
we
said
about
to
do.
B
We
created
something
that
we
originally
called
the
materials
genome
project
and
now
it's
materials
project
where
you
can
actually
go
to
search
interface
and
enter
those
pretty
much
search
criteria.
I
want
the
specific
voltage
window.
I
want
a
stability
that
is
between
these
energy
values
and
I,
want
so
and
so,
for
example,
volume
difference
or
something
else,
and
we've
calculated
more
what's
available
in
this
database.
This
interface
is
more
than
38,000
inorganic
material.
B
So
it's
a
good
significant
chunk
of
the
known
world
and
some
extra
thousand
materials
that
we
came
up
with
during
our
own
designs
of
new
materials
and
a
lot
of
other
properties
that
just
fall
out
of
it.
You
can,
you
can
get
for
free
when
you
solve
the
Schrodinger
equation
and
many
tools
that
you
can
slice
and
dice
the
data
in
different
ways.
B
There
is
a
materials
Explorer
or,
if
you're,
looking
for
the
materials
that
have
a
particular
application
for
batteries
or
if
you
want
to
look
if
the
materials
have
a
specific
stability
at
alkaline
or
acid
conditions,
so
different
ways
of
analyzing
pretty
much.
The
same.
Data
in
different
ways
is
in
leveraging
the
data
for
different
applications.
B
Somebody
else
got
very
excited
about
this
idea
to
in
2011
OSTP
the
science
advisors
of
the
president
I
wanted
to
launch.
They
launched
an
initiative
called
the
G
materials
genome
initiative
and
they
asked
for
the
name.
So
we
gave
it
up
it's
when
the
president
asks
for
something
you
wrote,
don't
really
want
to
say.
No,
so
we
dropped
the
genome
part.
B
Some
of
the
things
I
just
like
to
give
an
example
of
some
of
the
properties
that
you
can
actually
get
from
the
materials
party
today
and
which
are
fairly
hard
to
find.
If
you
were
to
go
after
these
kinds
of
informations
by
just
looking
at
journal
articles,
one
after
the
other,
we
can
screen,
for
example,
for
things
like
safety
of
battery
materials,
we
can
calculate
how
easy
or
how
prone
and
material
is
to
releasing
oxygen.
When
you
heat
it
up,
that
comes
out
of
our
phase
diagram.
B
So
again
it's
one
of
those
things
that
once
you've
built
a
tool
it
gets,
gives
you
a
lot
of
data
for
free.
These
are
three
cathode
materials
that
are
suggested
or
our
use
the
lithium
ion
batteries.
This
one
is
used,
it's
known
to
be
fairly
safe,
and
indeed
it's
if
room
temperature
is
here,
it
releases
oxygen
at
much
higher
temperature.
This
is
one
of
the
ones
we
designed
the
manganese
carbonyl
phosphates.
It
releases
oxygen
a
little
bit
earlier
actually,
but
it
also
releases
co2.
So
it's
sort
of
its
own
built-in
fire
extinguisher.
B
This
one
is
one
of
the
materials
that
some
people
actually
like
a
lot,
because
it
has
a
great
voltage
profile.
It
is
she
it
has
the
right
energy
density
to
beat
current
ones,
but
it
releases
02
at
room
temperature.
It's
basically
a
ticking
bomb,
and
we
don't
want
these
situations.
They
cost
millions
and
millions
of
dollars
to
several
so
so.
This
is
one
of
the
things
that
we
can
compute
quite
readily
and
it's
easy
to
screen
on
and
gives
us
a
sort
of
a
yes.
B
In
my
opinion,
because
we've
been
doing
it
for
a
long
time
and
we've
built
newer,
newer
versions
on
top
of
the
other
ones,
and
one
of
the
benefits
of
moving
from
MIT
to
lbnl
for
me
was
to
get
in
contact
with
the
scientists
as
a
nurse
and
at
crd,
who
helped
us
to
make
this
environment
a
lot
better
than
it
was
and
was
just
in
the
academic
setting.
So
we
have
a
whole
workflow
on
how
to
handle
the
data
from
the
structure
that
we
start
with.
That
comes
from
any
database.
B
All
these
coats
that
drive
this
infrastructure
have
their
own
guardian
angels.
The
primatene
library
is
guarded
by
the
Hulk
easy
he's
an
assistant
professor
at
UC,
San
Diego.
The
custodian
doesn't
have
his
own
person,
but
he's
the
Terminator.
He
self
heels,
that's
a
guy
who
keeps
track
of
the
job.
So
if
they
fail,
he
he
puts
a
sort
of
a
recipe
band
aid
on
them
and
submits
them
back
in
and
see
if
it
works
and
the
fireworks
guardian
angel
is
Captain
America,
that's
the
alias
for
honor
of
Jane.
He
has
designed
fireworks.
B
It's
a
workflow
manager
that
keeps
track
of
all
these
calculations,
and
the
model
of
fireworks
is
no
material
left
behind
it's
it's
the
idea
that
no
we
never.
We
never
lose
the
material.
We
know
exactly
if
it's
from
start
to
end
where
it
ended
up.
If
it
failed,
if
it
needed
to
be
recomputed,
there's
a
wealth
of
information
stored
from
fireworks
on
how
these
computations
work,
or
they
don't
work
and
I'm
just
waiting
for
somebody
to
really
dig
into
that
and
and
use
it
and
feed
it
back
to
whatever
quantum
mechanical
code.
B
Of
course
they
do
it
much
better
than
the
person
before,
but
still
it's
an
enormous
waste
of
time.
So
every
all
of
our
tests
come
with
unit
tests
and
documentation
and
to
submit
anything
to
primacare.
You
have
to
go
through
the
hook
and
he's
pretty
strict
about
what
goes
in
and
what
doesn't
so,
I'm
and
and
we've
all
learned
I,
but
it's
good
it's
what
we
were
hoping
to
make
this
something
that
lasts
we're.
Also
seeing
this.
This
is
the
usage
of
primate
gin
and
some
of
our
coasts
of
the
world.
B
Interestingly,
have
no
idea
what
they're
doing
with
it
in
Australia
there's
a
lot
of
downloads
over
there,
but
it's
great
I'm
hope
they
hope
they
come
up
with
something
amazing
we're
doing
all
we
sort
of.
We
give
out
all
this
data
and
algorithms
for
free.
The
only
thing
we
do
if
people
want
to
do,
we
ask
for
their
name
and
their
email
address
when
they
register
and-
and
we
have
a
few
Mickey
Mouse
and
Donald
Duck
there.
So
you
can
even
lie
so
people
wonder
why
do
we
do
this?
B
Why
do
we
give
it
for
free?
Because
so
we
do
it,
because
we
think
that
if
you
give
data
to
people,
they
do
amazing
stuff
with
it,
and
this
is
this
is
bigger
than
what
we
could
use
it
for
we're
already
doing
materials
design
within
our
own
groups,
and
we
could
keep
a
lock
on
it
and
use
it
for
own
tool,
but
it
really
doesn't
make
sense.
B
It's
amazing
to
see
what
people
do
with
data,
so
we
put
it
out
there
and
we
hope
that
this
will
be
the
seedling
of
materials
ecosystem
where
more
people
would
contribute.
Data
and
more
people
will
sort
of
build
into
this
idea
that
we
can
uncover
the
materials
genome
if
you
all
work
together,
but
we
do
occasionally
have
projects
that
are
funded
by
specific
entity
either.
B
If
it's
is
an
industry
partner
or
if
it's
an
government
grant
and
then
we
keep
the
data
separate,
we
call
it
sand
boxes
and
that
was
the
brainchild
of
Dan
gontran
crd.
Well,
we
basically
keep
you
know.
Sandbox
is
private,
but
they
interact
with
the
core
data.
That
is
public
and
they
leverage
all
the
tools
that
are
available
in
the
materials
project
and
one
one
example
of
those
sandboxes
is
what
we
recently
joined
or
work:
we're
partnering
the
energy
storage
hub
based
at
Argonne,
and
there
were
screening
for
electrolytes.
B
So
it's
a
bit
outside
of
what
our
beginning
goal
was
to
go
through
all
the
known
and
potentially
synthesizable
bulk
compounds
of
the
world.
But
this
is
any
solvent
and
salts
and
redox
active
molecules
that
could
enable
the
next
generation
of
energy
storage
solutions,
and
this
is
now
in
a
sandbox
that
is
only
available
to
jc's
or
partners
for
as
long
as
that
hub
goes
and
if
I'm
part
of
the
J
Caesar
and
I
log
in
to
the
materials
project.
B
Gotta
say
that
this
is
the
way
to
do
it
if
you're
going
to,
if
you're
going
to
have
a
hub
under
one
roof,
which
really
never
works
today,
right
they're,
all
under
one
roof
to
some
degree,
but
getting
data
through
the
internet
to
partners
and
enable
them
to
search
through
the
data
in
a
way
that
makes
sense
for
them.
That
is
the
way
to
do
a
real
collaboration
instead
of
sending
powerpoint
presentations
back
and
forth
or
excel
documents.
B
If
we
calculate
these
molecules-
and
we
enable
our
experimental
scientists
to
go
directly
to
that
and
search
through
the
data
themselves,
it's
a
huge
sufficiency
win
and
it
really
enables
them
also
to
start
thinking
about
the
computations
themselves
and
ask
questions.
How
accurate
is
this?
How
can
I,
how
can
I
use
this
data
in
the
best
way
for
my
research?
So
I
really
think
this
is
the
way
forward
to
do
these
large-scale
collaborations
that
are
being
funded
by
the
government,
but
always
have
a
hard
time
actually
realizing
this
under
one
roof.
B
To
give
you
an
example
of
how
we
can
leverage
the
data
in
many
different
ways,
we
just
in
October,
launched
the
probe
a
diagrams,
so
this
probe
a
diagrams,
is
basically
the
screening
tool
that
we
developed
in
the
alkaline
project
with
Durazo
to
find
out
where
the
materials
have
a
chance
of
surviving
at
very
alkaline
or
acid
conditions
in
Drano,
basically
or
NASA.
So
we
let
we
develop
this
tool
and
leverage
together
with
our
38,000
inorganic
materials
calculated
that
generates
about
more
than
70,000
for
a
diagram.
It's
unprecedented
there's
this
amount
of
information.
B
There
are
some
free
there's.
Some
commercial
software
for
generating
forebay
diagrams
out
there
that
you
have
to
buy,
but
then
you
have
to
feed
in
your
own
information
about
what
enthalpies
for
the
solids
and
go
in
there
and
they
just
provide
the
a
clue.
Science.
We
just
provide
all
this
for
free
and
it's
amazing
what
you
can
do
with
it.
This
is
just
the
tip
of
the
iceberg.
B
To
give
an
example
of
how
companies
are
using
our
sites,
they
don't
always
they
usually
don't
tell
us.
They
just
show
up
as
users
and
we
don't
snoop.
We
have
a
privacy
policy.
So
we
we
don't
look
at
what
they're,
what
searches
they're
doing,
but
occasionally
they
come
and
volunteer
information
about
what
they're
using
a
site.
For
so
intermolecular
did
this.
B
They
told
us
that
they
were
using
it,
for
example,
for
designing
new
photocatalyst
using
some
of
our
predicted
materials
that
have
never
been
made
and
they
were
oxynitride
in
the
oxynitride
space
and
they
took
some
of
those
and
they
synthesize
them
as
thin
films
and
actually
some
of
our
predictions
in
terms
of
band
gaps,
pretty
much
lined
up
with
what
they
saw
as
well.
So
it
was
a
nice,
a
nice
story
to
see
what
they're
doing
with
the
data.
B
In
a
nutshell,
we
launched
in
october
two
thousand
eleven.
We
have
a
little
bit
more
than
5,600
users
today,
we're
de
sentir
were
part
of
the
JC's
or
hub,
where
partner
and
then
at
that
centered
madison,
and
we
just
recently
joined
jacob,
see
if
we
can
make
a
difference,
we're
using
course
we're
all
over
this
country
and
I,
don't
know
it's
outside
in
the
world
as
well,
which
is
great,
because
I
think
this
is
a
very
intuitive
tool
for
students
to
be
able
to
browse
through
phase
diagrams
and
see
in
real
time.
B
You
know
if
I
change
this
element
that
element.
How
does
that
change
the
phase
diagram?
What
is
the
stability
metrics
between
chemistry's
and
structures
in
different
spaces?
We
have
a
number
of
different
partner
institutions
that
contribute
either
data
algorithms
are
both
and
a
lot
of
companies
are
using
our
website,
but
I
gotta
say
in
most
cases
we
have
no
idea
what
they're
doing
with
it
so
I'll
just
end
with
a
little
bit
of
a
sort
of
a
vision.
I
think
this
is
how
we
usually
work
a
scientist.
B
But
if
we
leverage
the
data
in
a
better
way,
let's
say,
for
example,
instead
of
instead
of
just
publishing
our
data
to
a
journal,
and
then
I
have
to
go
to
a
journal
and
get
that
information,
and
maybe
I'm
eating
sitting
with
a
ruler
to
find
out
what
the
points
in
the
diagram
is.
Instead,
we
share
the
data
through
some
internet-based
way
and
we
can
download
that
data
and
have
access
to
that
data.
B
I
can
completely
revolutionize
how
the
single
investigator
goes
about
its
work,
because
I
could
I've
seen
it
I've
seen
how
it
changes
people's
way
of
thinking.
If
they
have
access
to
data,
they
can
ask
the
question
and
directly
get
an
answer.
How
does
the
voltage
of
a
material
correlate
with
the
cation
can
I
see
the
trend?
How
does
it
if
I
combine
two
cations
in
a
material
with
each
other?
How
does
that
change
the
stability
of
the
material?
If
I
can
get
an
answer?
B
Those
questions
fairly
rapidly
from
a
data
set
that
changes,
how
I
do
material
science
and
that's
that's
bigger
than
anything.
It's
the
beginning
of
I
know
this
world
is
tired
and
old,
but
it
is
a
paradigm
shift
in
material
science
and
maybe,
if
we
all
do
it
and
we
all
enable
the
sort
of
sharing
of
data
in
a
comprehensive
way,
we
can
actually
uncover
the
genome
of
materials
to
know
exactly
where
the
building
blocks
go
in
to
design
exactly
the
properties.
We
want
to
optimize
the
performance
so
I'll
just
end
with
some
happy
quotes.
B
It's
pretty
awesome
to
occasionally
get
some
feedback
from
users.
I'm
sure
you
guys
know
that
as
well.
Thank
you
for
enabling
my
research
there's
a
guy
in
Nigeria
who
can't
afford
to
buy
any
journals
who
has
no
access
to
the
ICSD.
So
when
he
publishes
a
paper,
he
cites
the
materials
project
ID
for
the
structure
where
it
got
it
so
heartwarming
and
you
know,
people
who
use
it
in
different
ways
and
I
could
have
done
it
with
all
the
support
of
lbnl
as
well
as
in
outside
and
entities.
B
We
we
started
out
as
an
ldr
d
and
a
very
small,
enthusiastic
group.
The
david
skinner
was
part
of
it
and
I
would
not
be
able
to
be
here
if
nurse
hadn't
helped
us
and
crd
dan
Gunther
said
in
for
free
and
many
of
our
meetings,
we're
now
doing
fairly
well,
we
have
funding,
but
you
know,
without
that
enthusiastic
shoestring
budgets,
core
team,
we
could
not
have
made
it.
So
thank
you
to
all
and
thanks
to
the
team
and
very
happy
to
take
questions.
C
B
Yeah,
it's
it's
it's
a
difficult
one,
so
we've
tried
intermolecular
has
tried,
for
example,
you
saying
they
they
leverage
all
of
our
data.
They
basically
go
straight
to
the
database
through
the
API
and
just
download
it
all
and
they've
tried
to
see
if
they
can
make
some
of
their
data
public,
but
it's
it
depends
very
much
on
their
privates
on
their
own
customer
like
how
they
how
they
basically
make
money.
In
their
case,
they
make
money
off
of
materials
like
IP
and
and
carving
out
data.
That
is
not
protected.
It's
really
tough
for
them.
B
Also,
there
is
an
issue
of
how
do
you
compare
computers
and
experimental
data?
That's
an
open
question.
That's
one
of
those
things
that
we
really
have
to
I
think
address,
because
materials
to
experimental
materials
data
is
much
messier
than
computed.
I
know
exactly
where
the
atoms
are.
I
can
correlate
exactly
the
properties
to
the
specific
arrangement
of
atoms
in
the
material.
When
you
make
a
material,
it's
not
that
perfect.
Some
atoms
are
out
of
place.
B
There
are
grain
boundaries,
the
crystallites
are
kind
of
like
disorder
in
certain
places,
and
you
don't
know
if
the
properties
that
you're
measuring
are
related
to
the
inherent
crystal
structure
or
to
that
messiness.
So
having
and
we've
written
proposals
on
this,
but
we
haven't
gotten
traction
yet
having
in
a
translation
layer
between
what
is
the
experimental
data.
What
is
the
computer
data
that
I'm
making
them
talk
to
each
other
in
a
more
comprehensive
way
that
is
needed?
We
don't
have
that
yet.
B
We
just
take
but
they're
very
easy
to
measure
experimentally
I,
don't
care
about
this
stuff
so,
but
we
can
compute
solace
very
easily
and
that's
hard
and
mundane
and
boring
work
and
nobody
wants
to
do
it
from
the
experimental
side
so
taking
those
two
data
set
and
combining
that
was
actually
the
really
the
best
of
both
worlds
and
I
can
see
that
and
should
be
able
to
be
to
do
that
in
other
fields
as
well
in
material
science.
We
just
it
just
requires
people
to
look
at
both
datasets
and
developing
that
translation
layer
but
yeah.
D
To
ask
how
much
sort
of
the
semantic
areas
of
focus
like
you
mentioned
a
lot
about
batteries
and
I
mean
I,
have
a
colleague
that
works
on
tires.
You
know
for
cars
when
I
just
mean
there's
all
sorts
of
things,
everything's
made
of
materials,
and
so
how
much
do
those
areas
of
attention
influence?
What
you
know,
what
analysis
you
do
and
what
goes
into
your
data
yep.
B
That's
a
good
question
too
so
yeah
materials.
Is
it
it's
a
very
broad
term
right?
It
goes
everything
from
rubber
and
and-
and
you
know,
glasses
to
real
crystalline
systems
and
the
crystalline
systems
are
the
easiest
for
us
to
cathode.
That's
what
we
started
there,
we're
now
going
after
liquids,
which
is
a
little
harder
because
you're
they
move
around
and
stuff
which
is
messy
and
polymers,
and
rubber
and
stuff
like
that,
has
another
way
of
dealing
with
it
and
there
are
people
doing
computations
of
those
too.
B
So
every
time
you
go
after
new
class
of
materials,
you
kind
of
have
to
tune
your
approach
of
it.
It's
not
worldly.
It's
not
widely
applicable
to
you
can't
just
address
everything
exactly
the
same
machinery
has
to
be
tuned
and
then,
if
you're,
combining
materials
infrequently
frequently
the
materials
you're
talking
about
are
not
just
one
material.
There
are
several
materials
together
and
then
you
have
to
develop
something
for
that.
So
it's
a
length
scale
and
it's
an
it's
a
chemistry.
B
You
have
to
tune
both
the
length
scale,
as
well
as
the
chemistry,
your
approach
to
those.
So
no
it's
not
it's
not
a
cure
for
all
the
approach.
We're
doing,
but
I
think
that
if
you,
if
you
enable
the
sort
of
atomistic
the
bottom
approach
for
all
those
44
polymers
for
for
plastics,
for
for
liquids
for
for
solids,
you
can
build
the
other
length
scales.
B
On
top
of
that,
that's
the
idea
and
I
can
see
to
some
degree
that
that's
happening
we're
starting
to,
for
example,
to
calculate
the
elastic
constants
for
a
lot
of
the
solids
that
goes
into
higher
length
scale
models.
And
if
you
combine
two
materials
you
can
use
those
to
predict
the
strengths
of
materials.
For
example,
there
are
higher
order
models
that
use
that
information.
So
that
answer
a
question
to
some
degree.
E
B
We
when
we
got
the
center
gravity,
we
we
specifically
incorporated
a
statistician
in
the
team
that
had
no
knowledge
of
material
science
whatsoever,
so
he's
not
working
with
us
he's
also
at
UC
San
Diego
to
extract
those
trends
and
work
with
us
in
its
it
takes
time,
because
even
when
we
want
him
to
be
agnostic
and
apply
the
best
tools
from
his
field,
but
it
also
takes
time
for
us
to
sort
of
say
well,
you
know
this
doesn't
make
any
sense
right,
but
he
has
a
very
open-minded
approach.
He
just
throws.
B
You
know
all
the
stuff
you
can
think
about.
That
could
be
descriptors,
and
then
we
occasionally
go
back
and
tell
like
know
that
really
shouldn't
correlate.
Why
does
it
and
then
we
usually
find
there's
a
second
order,
correlation
somewhere,
but
yeah?
We
are
trying
to
do
that.
It
takes
a
bit
of
time,
though,
and
he
typically
thinks
we're
pathetic
in
terms
of
data
when
we
say
well,
we
have
300
data
points,
it's
better
than
any
of
the
experimental.
B
B
Well
many,
I
would
say
at
this
point
where
some
degree
is
still
limited
by
that
whenever
we,
so
we
learned
about
computations
today
for
something
we
were
you
know.
Nurse
has
been
fantastic
he's
giving
us
access.
But
if
we
again
God
hopper
for
a
week,
we
would
write
on
the
web.
We
could.
We
could
totally
go
to
town
with
that
and
get
another
20,000
materials
on
there.
So
that's
one
thing,
which
is
some
degree
still
limited
by
infrastructure,
did
still
some
you
know
plugs
that
people.
B
B
So
there's
still
some
tinkering
outs
people's
time,
basically
and
and
when
it
comes
to
new
data,
so,
for
example,
right
we're
about
to
go
high
throughput
with
elastic
constants
that
is
taken
as
about
a
year
and
a
half
to
take
algorithms
that
are
known
in
the
literature
but
making
them
robust
enough
that
you
can
run
them
in
high
throughput,
which
isn't
what
this
typical
investigator
does.
They
calculate
30
compounds
and
they
go
like.
B
Oh,
this
is
a
great
method
go
do,
but
it's
not
the
same
as
doing
10,000,
because
then
a
bunch
of
them
fail,
and
you
don't
know
why
trying
to
figure
that
out
and
you
have
all
kinds
of
fixes
for
it.
So
it's
taken
as
a
year
and
I
have
to
take
sort
of
an
existing
methodology
and
bringing
that
to
high-throughput.
So
we
could
really
blow
the
town
when
elastic
constants
and-
and
you
know
there
are
300
known
in
the
world
today
from
experiments.
G
B
Well,
we
have
some
experimental
data
whenever
we
can
lay
our
hands
on
it.
For
example,
in
the
reaction
when
the
calculate
reaction
energies
we
have
downloaded
whatever
is
known
from
from
works
like
kuba,
jeff's
p
and
compare
it
to
experimental
data,
but
you
know
they're,
there
are
very
few
free
sources
of
experimental
data
available
can
get
our
hands
on
so.
G
B
Important
yeah,
so,
first
of
all,
yes
going
back
to
the
edisonian
way,
yes
and
no
I
suppose
that
Edison
must
have
learned
something
from
his
3000
materials
right,
but
he
still
tested
every
single
one
of
them.
You
could
have
argued
that
he
could
have
tested
one
carbon
and
go
like
well
I'm,
not
gonna,
try
the
carpus
again,
but.
G
B
Went
through
every
single
one
of
them,
what
we're
trying
to
do
is
by
using
data
that
can
be
easily
assembled.
We
can
learn
from
it
and
become
smarter
about
what
to
actually
test
in
the
real
life.
So
it's
not
like
weird,
you
know
again:
it's
not
the
funnel
that
spits
out
the
perfect
material
you
go
and
take
a
vacation
for
three
weeks.
It
really
is
about
learning
the
data,
mining
and
understanding
the
underlying
principles
and
I
I
stand
by
that.
B
So
I
do
think
it's
a
more
clever
way
of
using
data
to
become
better
a
better
material
scientist.
But
what
don't
we
know?
Well,
the
higher
at
length
scales
is
still
a
missing
piece
and
there
are
people
using.
You
know,
trying
to
go
to
higher
length
scales
using
basically
multiscale
modeling.
We've
been
talking
about
it
for
a
long
time,
but
it
really
hasn't
caught
on
yet
because
it's
so
hard,
if
you're,
using
metrics
from
many
different
length
scales
and
trying
to
actually
model
a
material
on
a
real
macroscopic
level.
B
Well,
you
have
all
kinds
of
messiness
is
in
the
in
the
material
that
is
hard
and
we're
not
there
yet,
and
you
can't,
you
can't
do
that
high
throughput.
We
can't
calculate
that
to
any
degree
where
you
can
learn
from
the
data,
so
but
hopefully
by
by
giving
access
to
the
atomistic
level
of
lots
of
data
that
can
enable
better
systematic
learning
curves
of
the
higher
order
models
as
what
we're
hoping
that
by
building
bottom-up.
But
that
will
be
better,
but
we
can't
do
that
today.
B
If
you
ask,
for
example,
like
you
know
in
that
airplane,
whereas
it
going
to
crack
the
carbon
composite
fiber,
which
is
a
fairly
complicated
material
that
can't
be
predicted
from
whom
computations
we
can
get
an
idea,
but
most
of
the
time
people
are
still
trying.
How
long
is
it
going
to
last
10
20
years,
we're
pretty
bad
at
durability,
tests.