►
From YouTube: 2021-09-28 - Hawryluk, Carter, Wirth, Holland, Gerber - Powering the Future, Fusion & Plasmas
Description
NERSC Data Seminars Series: https://github.com/NERSC/data-seminars
Speaker:
Richard Hawryluk, PPPL
Troy Carter, UCLA
Brian Wirth, ORNL
Chris Holland, UCSD
Dave Humphreys, General Atomics
Title:
Fusion Long Range Plan and Fusion Energy Sciences Advisory Committee Report Briefing and Current and Future FES Needs at NERSC
Abstract:
Research in Fusion Energy Sciences comprises a big portion of the NERSC workload. Recently FES has developed a long-range plan to guide research priorities. The leaders of this effort will give a summary of these plans with an emphasis on how they involve NERSC and HPC. Outline: Troy Carter Brief overview of FESAC long-range plan with slide on HPC modeling (15 min), Rich Hawryluk main recommendations from NASEM report and strategic plan (8min), Brian Wirth role integrated design teams and integrated modeling (8min), Dave Humphreys report on FES/ASCR machine learning workshop (15 min Q&A).
Host of Seminar:
Richard Gerber
Senior Science Advisor | HPC Department Head
National Energy Research Scientific Computing Center (NERSC)
Lawrence Berkeley National Laboratory
A
We've
had,
of
course,
a
huge,
a
huge
community
of
fusion
and
plasma
physics,
researchers
using
nurse
and
they're,
one
of
our
very
very
important
communities
and
they've
recently
done
a
lot
of
work
to
plot
their
their
course
for
over
the
next
coming
years,
and
so
I
invited
this
group
to
give
us
some
summaries
of
some
of
these
reports,
and
you
know
we
really
want
to
understand
the
fusion
community
and
what
they
need
to
do
and
where
they're
going,
and
so
we
can.
We
can
better
support
him.
So
with
that,
why?
B
C
D
Okay,
hopefully
that's
visible
thanks
for
the
invitation,
so
we're
going
to
take
a
do.
A
short
presentation
here
and
chris
holland
from
ucsd
will
help
me
as
the
subject
matter
expert
for
high
performance
computing
as
part
of
the
subcommittee,
I
want
to
tell
you
about
the
the
report
of
the
fisac
called
powering
the
future
fusion
and
plasmas
I'll
point
out.
First
of
all,
this
is
not
just
this
report
not
just
focus
on
fusion.
D
Of
course,
qafps
has
a
broad
portfolio,
so
we'll
also
talk
about
plasma
science
and
technology
applications
and
chris
will
talk
a
little
bit
about
computing
needs
there.
So,
just
to
give
you
a
quick
overview
of
this
report.
First
of
all,
bottom
right,
right
corner,
you
can
download
a
pdf
of
the
reporter
see
a
web
version
at
us.
Fusionplasmas.Org.
D
There
we
go
so
this
process
started
in
2018,
with
the
charge
from
steve
bankley
from
the
office
of
science
to
psac,
and
the
charge
asked
for
a
strategic
planning
process
that
covers
the
entire
portfolio,
so
fusion
energy
development,
as
well
as
the
stewardship
of
broader
plasma.
Science.
Importantly,
this
is
a
two-part
process.
It's
modeled
after
the
the
high
energy
physics
p5
process
where
they
had
a
snowmass
part.
That
was
a
community
driven
piece
and
then
a
piece
led
by
p5
itself,
which
is
a
subcommittee
of
hepat,
so
the
equivalent
for
us.
D
We
had
a
budgetary
task
which
was
to
produce
over
10
years,
fiscal
22
to
31
optimized
program
for
fes,
and
so
with
three
budget
scenarios,
constant
level
of
effort,
which
is
just
accounting
for
inflation,
modest
growth
and
then
unconstrained,
but
prioritized
as
part
of
this.
The
eater
project
is
a
big
part
of
fes,
a
big
construction
project.
We
were
told
in
the
charge
to
assume
that
u.s
contributions
to
the
heater
construction
project
will
continue.
D
We
took
that
to
mean
that
our
job
was
not
to
project
that
and
we
focused
on
instead
the
non-eater
project
portion
of
the
dui
fps
budget,
just
to
give
the
context
so
again,
you're
going
to
hear
from
rich
in
a
second
and
brian
about
the
report
on
your
right,
but
this
process
happened
along
with
a
lot
of
other
activity
by
the
national
academies.
There's
really
four
reports
that
have
bearing
and
provide
context
for
this
vsac
report,
two
of
them
well,
really.
D
One
of
them
is
exclusively
focused
on
on
the
plasma
science
area
and
that's
the
first
one
on
the
left.
Plasma
science
decadal
covered
that
too,
as
well
as
fusion,
but
then
the
burning
plasma
report
and
the
bringing
fusion
to
the
us
grid
were
focused
on
the
fusion
part,
and
these
reports
set
the
stage
by
conveying
exciting
opportunities
in
both
plaza
and
science
and
technology,
and
also
the
urgent
need
to
accelerate
development
of
fusion
energy.
D
That
was
really
started
with
the
2019
burning
plasma
report
and
amplified
tremendously
by
the
report
that
rich
led
and
that
those
reports
laid
out
the
goal
of
a
fusion
pilot
plan,
electricity
producing
pilot
plan
on
a
time
scale
sufficient
to
mitigate
climate
change,
to
impact
decarbonization
by
mid-century
all
right.
So
let
me
just
give
you
the
high-level
executive
summary
messages
from
the
psac
report
and
I'll
go
I'll
talk.
I
won't
take
you
through
all
the
recommendations.
I'll
just
give
you
a
few
highlights
and
I'll.
D
Let
chris
talk
specifically
about
how
high
performance
computing
is
what
what?
What
do
we
need
from
nurse
to
get
this
to
happen
all
right.
So
the
first
message
on
the
fusion
side.
Is
it
really
now's
the
time
to
move
aggressively
towards
development
and
deployment
of
fusion
energy?
And
why
is
now
the
time
there's
really
a
few
things
that
point
that
way,
first
and
foremost,
scientific
advancements
over
the
last
decade,
technological
innovations?
D
So
these
things
point
point
the
way
to
this
goal
of
fusion
pilot
plant
that
was
laid
out
by
the
2019
burning
plasma
report.
Another
driver
here
is
the
growth
of
the
fusion
energy
industry,
so
one
of
the
goals,
of
course,
is
to
commercialize
fusion
as
a
the
growing
that
industry
is
important
and
that
industry
has
been
seeded
by
significant
investment
in
startup
fusion
ventures.
So
around
2
billion
at
the
time
of
the
writing
report,
has
been
invested
in
a
large
number
of
companies.
D
You
can
see
the
logos
here
on
the
other
side
of
the
portfolio,
so
plasma
science
and
technology
says
here
has
far-reaching
impact
and
the
messages
plasmas
have
and
will
continue
to
transform
society
and
that
impact
spans
across,
of
course,
fusion
energy.
You
have
to
have
a
plasma
at
the
core
of
your
fusion
reactor,
so
really
understanding.
Plasma
physics
is
important.
D
D
Another
high-level
message
in
the
report
is
about
partnerships,
partnerships,
accelerate
progress
and
there's
a
few
categories
in
which
we've
talked
about
international
partnership.
The
eater
project
is,
is
a
critically
important
part
of
the
fusion
program
in
the
u.s
public.
Private
partnerships
is
an
area
given
the
opportunities
enabled
by
the
investment
in
the
private
sector.
So
there's
a
lot
of
companies
that
have
grown
up
over
the
last
decade
have
received
lots
of
investment.
Partnering
with
them
is
essential
for
a
variety
of
reasons,
one
to
enable
more
rapid
joint
progress,
but
also
one
of
the
goals.
D
Again.
If
we
want
to
commercialize
fusion
energy,
we
have
to
grow
the
industry,
so
some
focus
on
investing
there
to
accelerate
progress
is
important
and
finally,
especially
on
the
plasma
science
side,
but
also
this
is
important
on
the
fusion
side.
There's
a
multitude
of
agencies
that
fund,
in
addition
to
dewey
fes,
that
fund
research
into
plasma
science
and
fusion
energy
research,
we
need
better
partnerships
between
those
agencies.
There
are
barriers
that
kind
of
prevent
progress,
and
so
we
call
that
out
as
a
as
something
that's
important,
to
focus
on
all
right,
another
high-level
message.
D
D
So
one
thing
to
point
out
that
we
put
on
the
report
is
that
again
we're
poised
to
create
this
fusion
industry
that
could
be
world
leading.
We
have
to
take
the
opportunity
to
support
that.
Another
high
level
message
is
one
of
the
reasons
we've
lost
leadership
is
really
because
of
the
absence
of
investment
in
major
new
facilities
to
address
critical
r
d
needs,
and
that
recommendation
is
really
focused
on
experimental
hardware,
but
it's
important
also
to
think
about
the
computational
resources
that
we
need
to
execute
the
strategy.
D
And
I
mentioned
this
two-phase
process,
so
this
was
very
important:
the
first
phase,
the
community
that
did
an
amazing
job
and
coming
together
and
really
generating
consensus
and
backing
the
the
ideas
that
were
put
forward
in
the
cpp
report
that
was
trans
transferred
into
this
fisac
report.
So
it's
important
to
note
that
really
the
community
is
backing
this
speaking
with
one
voice.
There
is
consensus
we
have
prioritized,
which
was
very
difficult,
and
so
this
this
is
a
very
important
message.
D
D
So
I'll
give
you
just
two
slides
on
recommendations:
I'm
not
going
to
go
through
all
of
them,
so
I
call
out
just
a
few
here
that
you
might
argue
were
highest
priority
within
the
fst
area.
The
report
again
lays
out
this
strategy
for
getting
to
the
fpp
by
2040s
important
parts
of
that,
first
and
foremost,
is
a
design
effort.
So
this
is
really
a
fusion
pilot
plant
design
effort
that
engages
all
stakeholders
in
the
community
to
help
establish
the
technical
basis.
D
What
do
we
need
to
do
to
close
critical
gaps
to
get
to
the
fpp,
and
this
again
this
is
where
you
guys
are
important.
You
guys
are
important
across
many
things
here,
but
theory
and
computation
is
essential
to
really
guiding
this,
as
well
as
engineering
design
tools.
We
view
this
as
kind
of
an
organizing
effort
within
the
program.
So
this
the
fpp
design
effort
really
helps
dictate
what's
important
in
other
aspects
of
the
program,
and
so
it's
very
is
essential.
D
Another
high
level
message,
so
I
mentioned
emphasis
on
the
technology
part
of
fusion
science
and
technology.
We've
had
less
investment
than
we
need,
especially
with
this
goal.
If
we
want
to
be
ready
for
the
fusion
pilot
plan,
we
need
to
invest
much
more
in
r
d
and
fusion
materials
and
technology,
and
so
going.
D
Of
high
level,
one
is,
is
to
enable
or
continue
steady
support,
provide
study
support
fundamental
plasma
science.
This
is
the
foundation
on
which
the
program
rests,
including
what
happens
in
the
fst
side
again
this.
This
also
sp
thinking
towards
this
angle
that
we're
highlighting,
which
is
the
opportunity
to
impact
sustainable
society
by
plasma
technologies.
D
D
F
F
Don't
know
if
it
came
through.
You
know
this
process
that
we
have
for
this.
We
actually
had
two
stages.
There
was
a
first
one.
We
called
the
the
cpp,
the
community
planning
process,
where
we
got
lots
of
feedback
and
workshops
from
everyone
across
the
entire
community.
Put
together
sort
of
you
know
a
version
of
sort
of
the
the
unconstrained
budget,
but
just
identifying
all
the
possibilities
of
what
we
could
do.
F
F
So
the
fusion
science
technology,
I
think,
is
troy
ludwig
I'll
talk
a
little
bit
more
as
well
we'll
come
up
later
on.
You
know
that's
going
beyond
plasmosity
to
all
the
the
non-plasma
but
nuclear
technology
engineering
that
you
need
to
build
a
power
plant,
whereas
the
plasma
science
technology
that's
going
into
all
the
kinds
of
plasmas,
you
might
study
for
things
beyond
fusion,
and
so,
but
you
know,
the
plasma
is
a
common
link
between
those
and
that's
sort
of.
Obviously
what
brings
us
together.
F
You
know
very
computational,
specific
recommendations
in
our
fiestac
report
that
you
know
the
first
was
just
as
a
general
cross-cutting
element
that
we
should
be
a
fps
needed
to
really
continue
supporting
both
foundational
theory
and
computation,
and
you
know
across
the
entire
program,
so
that
we
can
advance
our
understanding
testing
models,
but
you
know
also
start
leveraging
and
utilizing
all
these
new,
very
pretty
exciting
data
science,
things,
machine,
learning,
artificial
intelligence
started,
getting
us
ready
for
quantum
information
science.
Those
there
are
people
already
in
the
field.
F
You
know
we
also
definitely
need
to
take
advantage
of
the
investments
in
extra
scale
computing
and
so
there's
a
whole
I'll
talk
in
the
next
slide
about
some
possible
opportunities
for
that,
but
there's
across
the
entire
set
of
recommendations
and
research
there's
some
specific
there's
needs
for
all
of
that
and
of
course
we
want
to
continue
the
the
close
partnership
between
fps
and
oscar
on
things
like
sydex,
and
certainly
something
that
came
up.
You
know
in
our
discussions
right
now
the
sydex
are
pretty
much
focused
on
fusion.
F
You
know
magnetically
confined
fusion
plasmas,
but
there's
no
reason
that
it
was
obvious
to
us
that
you
know
they
had
to
stay
that
way.
So,
whether
you
you
could
expand
side
x
to
incorporate
other
fusion
systems
materials,
you
know,
non-fusion
studies,
you
know
the
low
temperature
plasmas
for
processing
solar
physics,
there's
a
whole
range
of
things
for
that
model
to
be
applied,
take
advantage
of
what's
already
going
on
and
strengthen
it.
So
you
know
at
the
high
level
those
are
the
two
in
the
fisa
report
that
call
out
computation.
F
You
know
as
troy,
lieutenant
one
of
the
top
level
ideas
was.
We
need
to
start
up
design
efforts,
design
teams
for
identifying
possible
pilot
plant
designs.
That's
what
fpp
is
fusion
pilot
plant.
So
if
we
want
to
in
you
know
that
we
need
to
we
having
validated
physics
models
that
are
suitable
for
doing.
That
is
going
to
be
absolutely
essential
to
be
this
on
the
time
scale.
F
We
want
that
if
we
want
our
power
plants
happening
on
that
20,
mid
2030s,
2040
time
scale,
you
know
we
don't
have
the
time
or
money
to
go
and
build
many
different
possibilities.
So
we
need
to
have
models
that
really
help
us
identify
what's
the
most
efficient
path
as
fast
as
we
can,
and
what
that
really
means
is
that
we
need
to
couple
predictive
models
that
are
simultaneously
quick,
efficient
and
accurate,
a
whole
lot
of
different
physics
phenomena
and
how
those
all
come
together
in
the
pile
plant.
F
If
so,
we're
going
to
scope
these
out
with
urgency,
and
so
certainly
that
includes
the
plasma
physics
models,
things
that
we're
doing
now,
but
that
goes
beyond
the
models
to
where
the
plasma
starts.
Interfacing
with
the
materials,
the
material
structures,
how
they
stand
up
to
the
neutron
environment.
Troy
mentioned
that
you
know
the
materials
experiments
we
need,
but
if
we
want
to
extrapolate
those
to
a
power
plant
scale
facility,
we'll
absolutely
need
computation
to
help
do
that
on
a
timely
fashion.
F
Looking
at
the
techno
economic
analysis,
it's
not
just
enough
that
your
power
plant
works.
Is
this
going
to
be
economically
attractive?
So
doing
all
those
kinds
of
computations
are
vital
and
so
to
develop
and
validate
those
models,
we're
going
to
need
both
pretty
extensive
capacity
and
capability.
I
don't
know
if
those
are
still
the
terms,
but
that's,
I
think,
that's
how
we
use
it
because
scope
out
that
range
of
conditions
to
build
the
hierarchy
of
validated
models
that
we've
talked
about
yeah.
F
We
have
to
run
many
many
simulations,
so
you
have
to-
and
you
know
those
just
by
necessity,
can't
always
be
the
heroic.
You
know
performance
pushing
ones,
so
you
have
to
have
you'll
be
able
to
do
many
simulations.
Many
different
conditions
call
many
miles
many
times.
At
the
same
time,
you
absolutely
still
need
the
highest
fidelity.
You
know
cutting
edge
simulations
to
help,
inform
those
models,
identify
what
might
be
changing
as
you
push
into
new
parameter
regimes
that
will
be
happening
in
these
reactors,
so
we
need
all
of
that
together.
F
F
So
there's
an
existing
network
called
lasernet
us
that
the
various
laser
science
groups
on
the
usu
they
have
small
facilities
around
the
country
that
you
can
go
and
do
research
on
you
put
in
proposals
and
within
our
report
we
one
called
for
expanding
those
networks
to
become
additional
new
networks
for
things
like
low
temperature,
plasmas
and
magnetized
plasmas
or
pulse
power
systems,
so
different
kinds
of
plasma
systems
that
are
being
studied
on
campuses
and
labs.
These
are
smaller
scale,
often
than
the
fusion
systems,
but
also
couple
in
modeling
support
to
them.
F
So
this
is
kind
of
a
common
thing
we've
seen
in
fusion.
I
imagine
many
other
areas
where
you
may
do
an
experiment
or
something
you
get.
You
know
someone
gets
a
grant
to
go,
do
their
experimental
research
they
do
that,
but
then
they
want
to
do
some
modeling
and
then
you
have
to
go
through
a
whole.
Another
process
of
okay
well
who's
going
to
do
this
simulation.
Where
do
I
get
the
computing
time
for
that
and
so
building
in
that
computing
support
and
the
resources?
So
it's
not.
F
You
know
the
time,
but
also
the
expertise
into
those
networks,
so
you
can
tie
in
they
take
it.
Different
groups
can
take
advantage
of
the
modeling,
get
the
support
they
need
to
make
those
more
effective-
and
you
know
just
calling
out
here
like
things
like
the
low
temperature
plasmas,
there's
a
plasma
technology
program,
that's
no
longer
recommendations
and
there's
a
whole
host
there.
You
know
non-fusion
plasma
applications
in
industries
now
so
there's
you
know
a
lot
of
opportunities
for
you
know
things
like
this.
That
impact
things
now
pretty
immediately.
F
So
and
you
I
think,
because
those
areas
people
are
having
necessarily
always
had
a
whole
lot
of
computing
support.
That,
to
my
mind,
means
that
there's
a
whole
lot
of
potential
there.
For
you
know,
even
a
small
investment
you
know
could
give
them
a
big
leverage
in
terms
of
taking
advantage
of
both
the
data
science
and
the
kind
of
more
traditional.
You
know.
High
fidelity,
you
know
high
performance
computing.
F
F
D
B
B
B
The
second
recommendation
there
was
the
u.s
should
start
a
national
program
with
accompany
research
and
technology,
leading
to
construction
of
a
compact
pilot,
reduce
electricity,
confusion
at
the
lowest
possible
capital
cost.
Now
this
second
recommendation
actually
was
a
motivator
for
this
study,
and
the
other
motivator
for
this
study
was
what
troy
mentioned.
B
B
I'm
going
to
give
you
the
key
takeaways
and
then
show
you
a
couple
of
slides
and
a
few
slides
as
to
how
we
got
there.
Our
first
recommendation
was
from
youtube's
united
states
to
be
a
leader
in
fusion
to
make
the
impact
on
the
transition
to
a
low
carbon
emission
electrical
system
by
2050
department,
energy.
The
private
sector
should
produce
lead
electricity
from
in
the
fusion
power
plant
in
the
united
states
in
2035
to
24
time
scale.
B
The
second
recommendation:
you
should
move
forward
now
to
foster
the
creation
of
national
teams,
including
public
private
partnerships,
that
will
develop
conceptual
pilot
plant
designs
and
technology
roadmaps
that
will
lead
to
an
engineering
design
that
will
bring
fusion
to
commercial
viability,
and
our
conclusion
is
to
accomplish
this
successful
operation
of
pipeline.
In
this
time
frame
requires
urgent
investment
by
the
eoe
and
private
industry,
both
to
resolve
the
technical
and
scientific
issues
to
design
construct
and
commission
a
pilot
plan.
B
This
is
an
outline
of
our
report.
I'm
going
to
talk
about
the
four
bullets
on
the
bottom.
The
first
one
is
the
role
of
pilot
plant
in
in
the
u.s
marketplace,
and
one
of
the
things
we
obtain
in
the
course
of
this
is
information.
What
the
utilities
are
doing
and
utilities.
We
see
a
transition
to
low
carbon
electrical
generation
by
2050.
B
B
The
implications
of
this
actually
quite
profound.
If
we
go
to
the
generation
of
low
co2
emission,
the
average
cost
of
electricity
could
go
up
and
you
can
see
that
on
the
plot
of
the
right,
if
you
don't
have
what's
called
firm
low
carbon
electrical
sources,
in
particular
things
which
are
dispatchable,
you
can
avoid
the
cost
increase
by
having
firm,
low
carbon
electrical
source,
and
this
has
actually
resulted
in
our
first
major
record.
Finding
a
recommendation.
B
It
must
be
attested
to
a
public
confidence
in
technology.
The
success
of
the
commercial
plans
that
will
follow
also,
as
we've
seen
so
much
in
the
last
20
years
and
I'm
sure
it'll
be
true
in
x20
due
to
the
evolving
energy
marketplace,
the
characteristics
of
a
future
power
plant
must
be
periodically
reviewed
by
energy
experts
and
updated
decreasing
the
likelihood.
The
diffusion
concept
will
successfully
contribute
to
the
needs
of
society.
B
So
a
pilot
plant
must
produce
an
amount
of
fusion
power
and
energy
that
is
sufficiently
represented.
The
market
needs
in
order
to
meet
the
pilot
plant's
goals
of
demonstrating
integrated
performance
and
costs,
while
demonstrating
net
electrical
electricity
gain
greater
than
one
and
produce
at
least
50
megawatts.
In
that
electrical
power.
B
Another
important
consideration
for
a
pilot
plant
is
economics
on
the
basis
of
today's
energy
market,
united
states.
The
fusion
first
of
all
power
plant
will
need
to
have
a
total
overnight
construction
costs
less
than
five
billion
to
six
billion
dollars
in
order
to
be
viable,
present,
u.s
electrical
marketplace
for
a
projected
operating
lifetime
for
at
least
four
years.
B
B
For
power
plant
and
the
construction
decisions
to
be
accelerated
by
combination
of
government
private
funding,
this
resulted
in
our
strategy.
We
identified
what
the
key
gaps
were
as
far
as
engineering
and
research,
and
these
will
be
addressed
in
the
near
term
during
the
so-called
conceptual
and
preliminary
design
phase.
During
that
phase
we
performed
as
national
teams
and
we're
talking
two
or
four
teams
about
conceptual
design.
Technology
roadmap.
B
B
We
have
to
define
the
regulatory
framework,
possible
side
options
and,
very
importantly,
perform
the
preliminary
design.
You
need
to
preliminary
design
to
obtain
a
good
cost
to
schedule
basis,
so
you
know
what
you're
going
into
as
you
move
towards
final
design
and
construction
and
startup
operations.
B
So
we
hope
that
we
could
do
this
by
2028
and
by
2035
the
device
would
be
built,
commissioned
and
operating
non-nuclear
phase.
Then
you
go
into
this
first
operating
phase
to
demonstrate
sufficient
fusion
type
of
energy
gain
that
electricity
and
the
second
operating
phase,
and
this
is
absolutely
critical.
B
So
we
recognize
something
like
this.
You
just
as
troy
mentioned
more
aggressive
than
the
keysight
report.
There
are
risks
and
opportunities
the
risk
guard,
the
level
of
scientific
and
technical
technology,
readiness
resulting
in
schedule.
Risk
has
to
be
acknowledged
and
there's
another
risk
that
we
won't
support.
Electric
electricity
transition,
people
will
say
well,
that's
this
time
scale.
Fusion
is
not
needed.
B
B
One
is
the
engagement
of
the
private
sectors
we've
seen,
which
I
mentioned,
which
is
very,
very
important.
Another
one
is
the
impact.
The
transition
to
a
low
carbon
emission
electricity
society
is
moving
ahead
to
make
this
transition
and
the
question
is:
can
we
engage
fusion
in
that
transition
and
for
us
to
be
a
leader
in
development
of
fusion
energy?
As
a
country,
we
feel
that
the
risks
we've
identified
and
mitigated
by
performing
research
and
development
and
parallel
design,
and
we
have
clear
decision,
points
to
evaluate
progress
and
moving
ahead.
B
B
E
Brian
thanks
rich,
so
I'm
gonna
build
off
of
what
rich
just
said,
and
I
wanna
highlight
just
a
little
bit
more
the
aspects
of
high
performance
computing
and
then
I'm
going
to
hand
off
I'll
be
relatively
brief
and
hand
off
to
dave
humphries
who's,
going
to
tell
you
about
the
2019
report
that
fes
did
in
collaboration
with
oscar
on
machine
learning
and
artificial
intelligence.
So
for
us,
integrated
simulation
is
a
key
part.
Much
like
troy
and
chris
talked
about
in
the
community
plan
and
the
fee
sec
long
range
plan.
E
So
this
first
bullet
here
is
is
the
recommendation
that
our
that
rich
already
noted
that
we
think
we
have
a
it's
an
ambitious
but
achievable
plan,
but
we're
going
to
need
to
move
forward
rapidly,
both
in
terms
of
defining
what
the
fusion
power
plant
confinement
concept
is
and
move
forward
on.
The
technology
to
extract
the
fusion
power
and
close
the
fusion
pop
fill
cycle
and
again
modeling
and
simulation
is
going
to
play
an
important
role
there,
because
a
lot
of
these
are
excuse
me.
E
I
just
jumped
forward
instead
of
back
because
a
lot
of
these
are
multi-physics
aspects
of
phenomena
and
so
we're
going
to
need
both
the
leadership
scale
computing
through
scidac
projects
to
support
high
fidelity
simulations
that
can
serve
the
basis.
Well,
we
have
limited
experimental
capability
available
to
benchmark
and
determine
the
reduced
parameter
models,
possibly
including
artificial
intelligence
and
the
more
detailed
physics
models
will
also
play
a
role
with
system
and
process
models
that
can
be
combined
into
a
full
device
model,
a
fairly
comprehensive
engineering
based
tool
that
can
contribute
to
evaluating
design
operations
and
maintenance.
E
Now
again,
the
significant
recommendations
there
are
that
there's
a
number
of
technology
challenges
that
we
have
to
address,
simultaneous
with
the
fusion
power
plant
design
or
the
confinement
concept
and
those
challenges
are
both
in
terms
of
the
plasma
facing
components
that
are
mediating.
The
plasma
material
interactions
and
extracting
heat
potentially
also
looking
at
the
possibility
of
using
liquid
metals
rather
than
solid
components,
and
that
correspondingly,
we
need
to
have
a
significant
amount
of
study
in
research.
E
That's
there
and
a
number
of
new
facilities,
but
the
new
facilities
will
be
complicated,
complemented
not
complicated,
complemented
with
the
sort
of
high
performance
computing,
as
well
as
more
capability
computing.
So
among
the
new
facilities
that
we're
going
to
need
are
heat,
flux,
testing
platforms.
E
Troy
already
mentioned
the
prototypic
neutron
source,
linear
devices,
for
testing
plasma,
facing
components
and
add
to
that
advances
in
high
temperature
superconducting
magnets
and
how
the
magnets
and
the
shielding
materials
for
the
magnets
and
structural
support
degrades
under
neutrons,
structural
and
functional
materials
and
then
blankets.
So
that
is
all
in
our
report.
I
want
to
take
my
national
academy's
hat
off
for
a
moment
and
give
you
a
little
bit
of
personal
perspective
since
I
leave.
E
Stop
shares
thanks
for
letting
me
know
debbie,
I
you
know
we're
all
having
zoom
fatigue
and
yeah
some
days.
It
works,
and
some
days
it
doesn't
so
hopefully
you're
seeing
worse
personal
perspective
on
plasma
material
interaction
slide
super.
So
the
the
plot
on
the
on
the
right
here
really
highlights
the
multi-scale
multiphysics
aspect
of
this,
because
there
are
processes
that
are
occurring
in
the
plasma
scrape
off
layer
or
the
boundary
plasma
where
it's
going
to
interact
with
the
material
surface.
E
There
are
processes
going
on
in
the
material
surface,
driven
by
ion
implantation
and
neutrons
that
are
passing
through
that
are
modifying
properties
and
microstructure,
and
then
there's
processes
at
the
interaction
sputtering
gas
implantation
that
are
influencing
essentially,
the
hydrogen
tritium
fuel
saturation
and
fuel
permeation,
and
all
of
these
complex
processes,
whether
you
think
about
it
as
a
multi-scale,
multi-time,
multi-scale
phenomena
or
you
think
about
it.
More
sort
of
cartoon
pictorial
process
are
highly
coupled
to
each
other,
and
so
what
we
have
done
in
our
scidac
project
is
develop
a
framework
for
integrating
multiple
codes
together.
E
These
arrows
here
are
shown
as
if
they
need
to
go
back
up
and
what
we've
demonstrated
in
our
scidac
is
that
we
can
do
sort
of
the
one-way
flow
with
a
little
bit
of
two-way
coupling
here.
So
we
model
the
interior
surface
of
eater,
get
the
edge
plasma
conditions
from
salt
ps
model,
the
sheath,
the
implantation
and
then
subsequently
the
erosion
and
redeposition,
and
pass
information
to
shalodl
to
evolve
this,
and
this
is
something
that
we've
developed
in
coordination
with
other
side
x,
in
particular
the
atom
side
x.
So
we've
developed
a
framework.
E
That's
basically,
a
file
based
integration
of
sequentially
run
codes
where
we
need
to
go
in
the
future
and
able
to
better
enable
and
use
these
tools
to
help
evaluate
designs.
Is
we've
got
to
close
this
loop
going
all
the
way
back
and
we
need
to
push
for
dynamic
coupling.
So
with
that,
I
want
to
stop
talking
and
not
spend
too
much
time
and
pass
it
over
to
dave
humphries.
E
E
G
G
As
brian
mentioned,
it
was
a
joint
oscar
fes
workshop
that
brought
together
the
two
communities
in
pretty
much
equal
balance,
as
well
as
participants
from
the
public
sector
and
other
areas
to
identify
priority
research
opportunities
in
the
machine
learning
ai
space,
and
I'm
going
to
use
this
mlai
and
I'll
say
it
that
way.
So,
hopefully
that
will
be
interpretable.
G
G
So
the
first
pro
was
science
discovery,
which
means
approaches
to
bridging
gaps
in
our
theoretical
understanding,
by
identifying
what's
missing,
using
data-driven
approaches.
So
we
can
accelerate
hypothesis
generation
and
testing.
We
can
optimize
experimental
planning
to
help
us
fill
those
gaps
in
in
a
an
extremely
fast
way
and
we're
thinking
about
things
like
theory,
data,
hybrid
models
for
confinement
and
some
of
our
major
theory
gaps
right
now.
G
One
one
version
of
this
actually
is
the
the
success
with
nif
in
bridging
their
theoretical
predictions,
using
data
and
improving
their
their
pulse
scenarios,
and
that's
a
hybrid
approach
to
to
expanding
your
science
machine
learning.
Boosted
diagnostics
means
application
of
ml
methods
to
maximize
how
much
information
you
get
out
of
your
measurements,
so
enhanced
interpretability,
systematic
fusion
of
multiple
data
sources
and
generation
of
synthetic
diagnostics
that
allow
you
to
amplify
your
inference,
and
this
is
a
space
again
for
advanced
computational
resources.
G
So
again,
I
think
the
the
key
role
for
computational
resources
is
in
accelerating
computational
algorithms,
with
model
reduction
in
multi-scale
multiphysics
simulations
and
supporting
hierarchies
of
fidelity
in
computer
codes
for
whole
device
modeling.
In
fact,
beyond
that
in
the
kinds
of
of
integrated
multi-physics
platforms,
that
brian
was
talking
about,
for
example,
control
augmentation
is
an
area,
that's
really
rich
for
machine
learning
applications,
but
I
think
less
so
for
exascale
and
large-scale
cpu
gpu
hybrid
systems.
G
Prediction
is
another
area
that
people
frequently
associate
with
machine
learning,
so
algorithms
that
predict
key
plasma,
states
and
indeed
system
states
for
a
functioning
experiment
and
eventually
a
power
reactor
are
critical
and
machine
learning
can
provide
sometimes
the
only
solutions
to
these
kinds
of
things
and
disruption.
Prediction
is
one
example
of
where
it's
been
very,
very
successful
to
date,
another
area
and
the
last,
and
perhaps
the
most
important
in
the
the
the
sense
of
enabling
all
of
the
above
is
the
fusion
data
machine
learning
platform.
G
That'll
support
our
ability
to
apply
all
these
methods
in
a
scalable
way
to
fusion,
so
you've
got
to
create
a
novel
system
from
managing
formatting
curating
accessing
marshalling,
using
the
data,
both
experimental
and
simulation
in
an
optimal
way
for
machine
learning,
so
the
foundational
resources
and
activities
I've
mentioned
at
a
high
level,
of
course
are
experiments
both
u.s
and
international,
and
the
greater
your
ability
to
integrate
these
together
in
common
platforms,
the
greater
your
ability
to
apply
powerful
machine
learning
approaches
to
extract.
You
know
unified
information
from
all
of
that
data.
G
So
this
is
an
observation
that
I
think
is
relevant
for
the
fusion
data
platform
of
the
supporting
infrastructure.
In
addition
to
the
kind
of
memory
data
you
know,
curation.
You
also
need
lots
of
infrastructure
to
manage
that
data
and
to
make
it
useful
to
to
applying
machine
learning
analysis.
So
automated
metadata
extraction
is
something
that
is
is
woefully
lacking
in
our
field.
G
You
know
for
leadership,
scale,
computing
and
indeed
more
more
machine
learning
tuned
systems,
so
surrogate
model
training
will
be
one
thing:
that's
immediately
made
possible
by
making
these
large
scale
simulation
databases,
an
acceleration
of
simulations
and
other
algorithms
that
are
enabled
by
that.
But,
of
course,
when
you
have
enormous
simulation
databases,
you
can
couple
them
with
experiments.
G
You
know
so
things
like
you
know,
adas
and
and
international
databases
and
the
meta
metadata,
if
you
will
that
governs
all
of
these
things,
constraining
data
on
designs,
etc,
etc,
and
so,
finally,
I
alluded
to
this
for
the
fusion
data
platform,
the
infrastructure
for
curation
and
preparation
of
the
metadata
and
marshalling
your
data
for
an
application,
the
algorithm
integrated,
visualization
and
data
space
analysis.
So
one
of
the
things
that
people
aren't
usually
thinking
of
when
they
think
about
doing
machine
learning.
G
Is
you
spend
a
lot
of
your
time,
maybe
most
of
your
time,
analyzing
the
data
space
itself
to
prepare
for
the
appropriate
kind
of
of
algorithmic
solution
and
an
optimized
workflow,
and
you
need
enormous
amounts
of
infrastructure
for
that
which
we
really
don't
have.
People
are
fond
of
making
their
own
little
things
out
of
python
or
matlab,
and
you
know
and
that's
kind
of
how
we
do
it,
but
we
need
to
make
a
large-scale
system
to
support
this.
E
A
I
see
eli
has
his
hands
up,
but
I'm
gonna,
I'm
gonna,
take
my
prerogative
and
ask
one
one.
First
before
you
lie,
and
that
is
you
you
presented
a
you
know
a
very,
very
clear.
You
know
vision
which
I
think
is
great
with
the
data
pilot
plan
and
other
things,
there's
a
lot
of
talk
about
infrastructure
that
needs
to
be
built.
That
doesn't
exist,
and
I
don't
I
don't
just
mean
you
know.
B
B
Now,
if
you
just
do
trial
and
cut
you
know
it's
not
going
to
get
us
to
where
we
need
to
be.
If
you
just
do
theory
and
computation
without
doing
the
experimental
work,
it's
not
going
to
give
you
the
confidence
that
you
need.
So
it's
going
to
be
a
combination
of
all
it
is
it's
absolutely
critical
to
close
the
gaps
vote
in
the
science
side
and
technology
side.
A
Yeah,
you
know
there
are
a
lot
of
things
that
need
to
be
done,
a
lot
of
tasks
and
things
like
this,
and
so
you
can
almost
imagine
putting
some
of
this
into
kind
of
a
doe
project
framework
with
milestones
and
this
sort
of
thing.
I
was
wondering
if
there
is
some
organizational
or
structure
or
plan
plan
to
do
that
and
where
a
nurse
might
fit
into
that
kind
of
thing.
A
C
Yeah
so
chris
you
had,
you
talked
about
a
bunch
of
sort
of
different
types
of
of
modeling.
That
needed
to
be
done
right,
both
in
terms
of
you
know
very
large
scale,
high
resolution
simulations
and
also
just
parameter
space
searches
to
narrow
things
down,
so
that
you
can
be
more
efficient
about
the
experiment,
part
of
things.
C
So
I
guess
I've
kind
of
a
two-prong
question
about
this
one
is:
is
that
what
is
referred
to
as
whole
device
modeling
in
in
other
sort
of
doe
fes
language,
or
is
that
an
expansion
or
augmentation,
or
a
separate
effort
from
that?
And
then
related
to
that?
How
much
of
that
is
just
how
much
of
that
computing
is
just
straight
simulation
and
how
much
of
that
takes
much
more
advantage
of
the
data
platform
that
that
dave
humphries
was
talking
about.
C
F
Thanks
yeah,
so
the
the
I
mean,
I
would
say,
like
the
whole
device
modeling,
we
have
different
frameworks,
but
you
know
I
I
mean
the
analogy:
we've
used
sometimes
like,
if
you
think
about
like
climate
modeling
or
something
like.
If
I
want
to
do,
you
know
just
the
climate
model
like
I
need
to
couple
my
models
of
the
atmosphere,
the
ocean
and
all
these
different
phenomena
going
together.
So
that's
what
we
mean
in
that
sense,
and
so
then
you
know
what
I
was
saying
was
like.
F
F
You
know
you,
depending
on
what
you
need
and
what
your
applications
are.
You
need
to
get
all
those
physics
down
to
operate
very
quickly
so
that
you
know
you're,
not
your
overall,
like
you
know,
climate
model
isn't
being
held
up
by
needing
to
resolve
a
cloud,
but
you've
got
20
other
things
running
so
like
you.
If
you
have
to
do
that,
many
many
different
things
going
on.
You
need
all
those
to
be
fast,
and
yet
we
have
to
find
that
right.
F
So
you
know
in
that
sense
that's
sort
of
where
you
I
think,
and
that
you
know
that
historically,
we've
done
a
lot
of
that
already
on
the
the
plasma
physics
side,
so
sort
of
like
this
picture
behind
me
like
okay,
that's
you
get
the
glowing
plasma,
but
you
know
the
next
step.
I
think
there
is
sort
of
what
do
we.
You
know.
We
need
to
understand
that
better.
But
what
do
we
do
with
all
the
heat
that
starts
impinging
on
the
walls?
The
stuff
brian
was
talking
about
the
fluxes,
the
neutrons?
F
How
do
they
irradiate
the
materials?
How
do
we
build
the
materials
that
stand
up
to
that
environment?
Things
like
that?
So
you
know
all
these
kinds
of
things
might
talk
about,
could
go
there,
so
I
mean
in
that
sense
I'm
not
as
sure
about
you
know.
Maybe
dave
may
also
jump
in,
but
I
could
certainly
see
in
terms
like
the
data
platform
and
all
like
there's
things
that
we
know
how
to
simulate.
F
F
You
know:
we've
historically,
we've
kind
of
focused
either
on
totally
empirical
or
totally
analytic
models,
and
I
think
you
know
historically
times
we've
done
things
that
we
call
semi-empirical
models
where
we
have
some
insight,
but
you
know
maybe
we
can
have
some
more
calibration
from
experiments,
so
I
think
that's
a
whole
vast
opportunity
now,
where
you
know
we
take
some
of
the
results
from
our
experiments.
You
use
these
data
tools.
The
the
data
platforms
combine
that
with
what
can
we
get
from
simulations?
F
What
can
we
get
from
experiment
and
knowing
where
we
need
to
project
to
combine
that
all
to
build
those
reduced
models
and
let
us
get
to
where
we
want
to
be
on
a
timely
fashion,
and
I
was
just
going
to
have
just
one
other
kind
of
you
know
from
richard.
You
know
the
previous
question.
This
question,
you
know,
came
up
a
lot
in
the
cpp
and
one
of
the
things
we
talked
about
was
having
things
like
the
need
for
what
we
called
computing.
F
I
mean
we
talked
about
computing
infrastructure
by
engineering
sports.
We
often
think
about
like
if
we
have
a
large
experimental
facility
with
you.
You
have
the
scientists,
but
you
also
have
a
whole
lot
of
technicians
and
people
that
help
you
keep
you
all.
Just
the
systems
running
and
a
lot
of
the
codes,
especially
these
higher
performance
ones,
are
now
so
complicated
that
you
know
that
it's
not
a
one-man.
It's
not
a
one-person
effort
anymore,
that
you
know,
if
especially
like,
if
you
there's
a
tension
there.
F
You
kind
of
need
more
support,
so
getting
sort
of
like
that
kind
of
equivalent
support
of
how
do
we,
you
know,
keep
these
things
portable,
so
you
don't
have
someone
that's
really
focused
on
physics
and
using
that
code
is
trying
to
learn.
You
know
how
do
I
get
this
onto
gpus,
so
so
getting
all
those
kinds
of
that
equivalent.
You
know
kind
of
engineering,
support
and
portability,
and
things
like
that
is
is
definitely
a
need
and
that's
true.
You
know,
I
think,
across
the
whole
field,
yeah.
A
Yeah,
thank
you
and
I
know
we're
we're
over
already,
and
so
I
want
to
thank
you
again
and
you
know,
there's
there's
there's
a
lot
of
there
that
a
lot
in
the
presentation.
There
are
just
more
touch
points
I
think,
between
quite
kind
of
we're
thinking
about
and
what
you've
been
thinking
about
too,
which
I
think
is
great,
and
hopefully
you
know
we
can
continue
to
work
with
community
in
the
future,
especially
around
the
need
for
simulation.
We
don't
want
to.
A
C
A
G
Yes,
we're
talking
about
the
fusion
data
platform
in
our
vision
of
the
machine
learning
enabling
environment
that
it's
important
to.
We
were
identifying
this
as
an
essential,
absolutely
essential
piece
to
make
large-scale
integrated
machine
learning
possible.
In
the
us
I
mean
as
long
as
you
have
balkanized
data,
you
will
be
limited
in
your
ability
to
extract.
What's
there
right
and-
and
that
is
I
think,
very
much
in
active
use.
G
G
You
know
full
duplex
kind
of
system,
and
it
it
does
depend
on.
You
know
eli's
magic
for
achieving
high
throughput
data
because
one
of
the
big
throttles
for
this
kind
of
activity,
if
we
want
to
pull
in
if,
for
example,
international
data
from
europe
from
china
from
korea
is
that
and
it's
it's
the
same
for
real-time
engagement
when
we're
trying
to
stream
data
between
shots,
as
it
is
for
a
workflow
in
machine
learning,
analysis
or
any
scientific
analysis.
I
would
argue
actually
so
yeah.