►
From YouTube: SimPEG meeting July 29, 2020
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
Perfect,
well
thanks
everyone.
I
know
this
today
is
agu,
abstract
deadlines,
and
so
I'm
sure
a
few
folks
have
that
on
their
plate.
I've
got
a
few
too
many,
but
that's
all
right
yeah.
So
who
wants
to
start
us
off.
B
B
That
would
work
with
the
latest
simulation
stuff,
so
yeah,
I'm
pretty
much
working
on
that
dieter
noticed
that
he
had
some
import
errors
that
I
didn't
have,
and
I
guess
he
believes
that
it
has
something
to
do
with
him,
working
maybe
on
a
linux
machine
and
me
on
windows.
B
So
essentially
the
game
plan
is
to
for
me
to
go
in
and
fix
those
tests
and
make
a
pull
request
and
then
travis
might
catch
the
import
error
or
catch
any
of
those
errors.
So
some
progress
is
being
made.
I'm
doing
this
agu
abstract,
but
it's
it's
high
on
the
list
for
me.
There
dieter
so
hopefully.
C
D
Worries
I've
I'm
on
a
holiday
from
the
next
two
weeks,
and
I
don't
get
much
done
before
that,
so
it
will
be
mid
august,
the
earliest.
I
can
have
a
look
at
it,
okay,
but
it's
an
interesting
issue
because
with
devin,
because
it's
import
case
sensitive,
which
is
not
case
sensitive
on
windows
but
everywhere
else.
B
A
A
It's
a
feature
of
git,
so.
A
B
A
That
and
then
rename
it
to
the
lowercase
thing
that
you
actually
want,
and
so
that
will
trigger
that.
But
that
actually
could
be
an
issue
right
now
that,
even
if
you
see
things
as
lowercase
on
your
machine
and
you've
pushed
changes,
it
might
not
actually
be
lowercase
machine.
So
there
could
be
a
sensitivity
issue
if
you
didn't
actually
like
ensure
that
so
check
on
github,
if
it's
capital
or
lowercase
on
the
branch
that
you're
on.
B
A
You
might
be
able
to,
but
you
might
actually
have
to
rename
it
to
something
temporary.
I
don't
know.
B
D
A
A
Cool
well
then
peter,
do
you
have
any
updates
you'd
like
to
share.
D
Probably
two
things
I
wrote
in
the
dev
channel
that
about
the
license,
which
might
be
an
idea
and
the
warnings,
but
I
think
joy's
on
that
with
the
warnings
and
also
the
discretize
there's
this
matplotlib.
If
you
update
the
matplotlib
to
333,
you
might
get
warnings
with
blood
slice
and
the
slight
3d
slicer,
because
we're
not.
I
guess
it's
something
from
before.
We
have
to
see
a
limb,
but
then
you
can
also
pass
in
v
min
and
v
max
in
the
p
opts.
D
How
is
it
called
yeah
in
the
optional
arguments
for
the
slicer
and
and
the
other
things?
So
we
probably
have
to
think
about
that.
Maybe
if,
if
we
want
to
deprecate
the
ceiling
all
together
to
just
have
it
in
one
place
or
another
because
mad
blood
since
a
while
since
a
year
or
a
bit
more,
they
become
more
pushy
to
clean
up
and
remove
old
stuff
and
be
more
streamlined.
I
guess
it's
a
painful
process,
but
down
the
road,
it's
good
for
them,
but
so
we
have
to
maybe
particularly
with
discretize
and
under
plotting
functionalities.
D
A
Yeah,
so
I've
been
working
on
that
I
was
hoping
to
get
you
something
by
monday,
but
of
course
that
didn't
happen,
but
it's
getting
closer,
so
I've.
Basically,
I've
been
running
simulations
on
google
compute,
which.
A
Running
with
that,
so
what
I
was
actually
hoping
to
do
is
get
the
layered
model,
basically
cleaned
up,
either
today
or
tomorrow,
and
then
send
that
to
got
soggy
to
showed
up.
I
was
going
to
send
it
to
sagi
and
to
to
joe
and
just
if
they
have
any
other
input
or
ideas.
The
thing
that's
really
been
interesting
to
sort
of
get
a
sense
of
is
really
how
sensitive
you
are.
A
If
you
really
want
to
drive
down
air
to
padding
cells,
not
just
with
respect
to
boundary
conditions,
but
also,
if
you
pad
too
aggressively
it
does,
it
does
reduce
your
accuracy
and
so
trying
to
find
that
sweet
spot
where
we're.
Basically,
you
know
we're
padding
sufficiently
to
satisfy
boundary
conditions,
but
not
so
aggressively
that
we're
degrading
the
accuracy
of
the
solution.
It's
a
bit,
it's
more
fiddly
than
I
thought
it
would
be,
and
so
I'm
to
the
point
of
sort.
F
A
Three
percent
median
error
on
the
offset
line.
I
I
know
we
can
get
it
lower,
because
what
I've
seen
is,
if
I
violate
the
boundary
conditions
a
bit.
The
center
of
the
solution
gets
below
one
percent
error,
but
then
of
course,
I'm
filing
boundary
conditions,
so
the
tails
of
the
line
are
bad
yeah,
so
a
bit
of
fiddling,
but
hopefully
I
think
that
in
a
week
or
so
we
should
have
that
basically
good
enough.
I
mean
I
might
still
be
polishing
it
a
bit
but
like
something
that
you
can.
A
Basically,
you
can
sort
of
consider
it
working.
The
modern
model
is
probably
going
to
be
a
few
more
weeks
out.
At
this
point.
D
A
G
A
And
thanks
for
bringing
up
those
licensing
things,
I
think
that's
a
good
discussion
point
and
we
can
maybe
move
back
in
with
joe
when
he
is.
D
It's
it's
on
github
up
here
that
emg
3d
is
used
by
one
project
which
surprised
me
yeah,
so
I
looked
at
it
and
it's
it's
a
brazilian
who
who
makes
in
amg,
3d
and
simpag
and
and
and
it's
it's
it's
it's
a
funny
reaper.
I
think
he's
new
to
programming,
there's
a
pi
cache
in
it
and
and
all
the
other
files,
but
they're
in,
but
then
often
and
the
things
have
that's
no
problem.
D
I
don't
care,
but
what
often
didn't
happen
then
the
files
are
in,
but
the
license
is
not
in
so
as
soon
as
that
happens,
you
actually
don't
know.
I
mean
it
can
happen
purposefully
and
that
we
can't
avoid
you.
Someone
can
always
rip
it
off,
but
I
think
in
this
case
it
was
accidentally
because
you
take
a
file,
but
that
doesn't
mean
the
license.
D
Is
there
so
I
think
it's
it
would
be
good
to,
and
with
these
identifiers
which
are
pushed
by
the
linux
foundation,
this
would
be
just
one
line
or
maybe
a
two
line
with
the
website.
And
then
it's
not
a
lot
of
code
or
not
a
lot
of
space
that
it
would
use
up
and
if
somebody
copies
a
file,
it
would
retain
the
minimum
information
of
where
it
comes
from
and
what's
the
license.
D
A
Yeah-
and
I
think
that's
a
really
good
idea,
especially
for
projects
like
simpeg
and
dmg
3d
because
like
if
you
are
sort
of
in
a
research
project-
and
you
really
want
to
play
with
the
methodology
like
sometimes
copy
and
pasting
the
file
is-
is
your
starting
point?
Yeah,
that's
not
an
unreasonable
thing
to
do,
but
we
don't
yeah.
We
want
the
provenance
and
the
licensing.
D
G
H
H
So
I
just
I
jumped
back
into
that
doing
2d
and
3d
kind
of
sense
and
just
started
playing
around
with
how
much
how
many
padding
cells
I'm
going
to
need
in
the
one
direction
just
to
make
it
look
like
that
yeah
I
can
extract
something
2d,
that's
reasonable
and
yeah.
It's
looking
really
good.
I
can
do
about
at
about
five
frequencies.
I
can
do
about
100
stations
in
six
hours,
so
yeah
not
too
bad,
and
it
converges.
F
H
It
was
looking
pretty
good,
I
didn't
play
with
the
x-ray
stuff.
Yet
again,
I
was
just
kind
of
busy
with
that
and
that
the
mum
solver.
So
I
got
that
all
built
and
it's
been
really
nice
being
in
linux,
I'm
not
going
back
it's
once.
I've
gone
back,
it's
it's
hard
to
go
back
to
windows,
but
mumps
is
really
interesting.
I'm
finding
it's
about
an
order
of
magnitude
faster
than
parties.
H
So
just
as
is,
if
I
think
I
did
a
ten
thousand
by
forty,
seven
thousand
matrix-
inverted
it
and
it
took
about
0.1
milliseconds
of
cardiso
and
the
point
zero
one
sec
with
monks,
and
I
think
that's
because
the
joe
was
saying
that
the
the
factorization
is
parallelized
now
too
or
in
months
so
yeah.
It
really
makes
quick
work
of
it
and
it
scales
yeah,
as
I
called
it
a
little
bit
of
a
plot
there
on
the
dev
channel,
showing
that
yeah.
H
As
you
add
the
cores,
you're
seeing
scaling
and
then
I
did
it
a
couple
of
size
and
matrices
as
well
and
as
the
matrices
gets
bigger,
the
more
cores
are
helping
out
as
well.
So
you
get
a
little
bit
more
it
actually
the
curve,
as
I'm
upping
the
matrix
matrices,
it's
almost
getting
more
linear
than
kind
of
logarithmic.
H
So
that's
kind
of
interesting.
It
will
the.
If
you
go
from
one
to
two
chords,
when
you
got
a
really
big
problem,
you
don't
see
much
of
a
game,
but
then,
as
you
add
them
the
it
just
just.
I
guess
once
it
breaks
up
the
problem
into
the
parallelization,
then
you
can
just
start
adding
cores
and
it
it
actually
benefit
yeah,
whereas
smaller
problems
won't
need
yeah.
I
guess
that
many
force
to
actually
get
more
speed.
H
Yeah,
so
I
went
into
the
documentation
and
actually
a
lot
of
it
was
through
petsy
the
documentation
through
there
they
had
a
lot
of
recommendations
on
what
to
use
to
build
monks
in
a
repair
fashion,
so
the
main
ones
was
one
called
scala
pack
and
blancs
and
openmpi
or
mpi.
Those
are
the
four
main
ones
to
get
some
really
good
speed
and
then
for
like
the
parallelization
itself,
I
use
scotch
yeah
it's.
H
I
can't
really
remember
how
they
explained
it,
but
it
just
it
kind
of
like
distributes
the
threading
and
it
kind
of
manages
all
that
more
and
you
can
also
team
that
up
with
another
package
called
memphis
or
paramedus,
and
I
haven't
got
that
one
to
build
properly.
H
So
I
think
we
can
even
get
more
gains
if
we
can
incorporate
that
library
too
yeah.
I
built
everything
from
source
in
the
most
optimal
way
to
get
scaling
and
parallelization
and
seems
to
work
it's
just
now.
My
problem
is
getting
pie,
monks
to
compile
the
c
side
of
things.
The
fortran
compiler
seems
to
be
working
okay,
but
when
I
try
to
do
the
c
compiler,
it
is
having
problems
finding
the
the
mumps
headers
so
yeah,
I'm
just
fighting
with
that.
H
A
A
That
would
be
great,
I
mean
and
either
I
can
try
running
it
on
google
and
like
sort
of
try
and
replicate
the
build
process,
or
I
can
also
like
iterate
and
set
it
up
on
on
one
of
your
machines.
If
that's
okay
and
see
if
we
can
get
both
going,
that
would
be
kind
of
cool.
D
Would
be
cool,
and
I
mean
we
can
also
update
that
in
the
paper
in
in
the
review
stage
later,
if
that
will
come
also
for
the
block
model,
I
guess
you
have
to
be
careful
with
language,
though
you
can't
say:
mumps
is
10
times
faster
than
part
d,
so
it's
10
times
faster
than
the
par
diesel
version.
We
are
using
yeah.
H
D
H
D
H
It's
looking
pretty
promising
and
I
think
we'll
get
some
really
good
speed.
Ups
in
our
versions,
if
we
can
get
months
going,
I've
like
now
that
I've
went
through
it
and
when
I
look
back,
it's
actually
not
that
hard
to
build,
but
it
was
kind
of
okay.
G
F
H
Yeah,
I'm
just
trying
to
use
psython
to
just
compile
some,
I
guess
just
a
wrapper,
but
that's
what
I'm
having
a
problem
with
right
now
is
trying
to
get
pi
mumps
to
find
or
when
it
calls
the
c
compiler
to
find
the
headers.
And
it's.
F
D
Don't
know
I
might,
I
might
contact.
H
My
moms
and
just
ask
them:
if
what
they're
sorry
whoever's,
I
think
it's
pretty
up
there
kept
update.
So
hopefully
I'd
get
a
reply
from
them
and
get
older
or
I'll
figure
it
out
by
the
way.
I
I
think
initially
we
used
the
maps
and
we
come.
I
think
at
that
point
lars
compiled.
I
I'm
not
sure,
like
I
think
lars
come
like
in
lars
at
that
point,
that
he
was
a
postdoc
in
ubc
he
compiled
and
then
rowan
wrapped
it
up
with
pi
man,
solver
and
yeah.
The
installation
was
a
headache,
but
it
worked
actually
well
at
that
point
and
it
was
a
serial
mums
at
that
point.
G
A
H
I
But
like
let's
say:
if,
if
you
write
up
a
rapper
in
python
and
like
a
can,
you
put
some
sort
of
like
open
license
into
it
like
in?
If
you
build
a
package
that
wraps
it
up
a
mums
and
then,
if
you
write
a
plate,
let's
say
pie
pie
package,
for
instance,
can.
C
I
Just
assign
like
a
mit
license
or
like
yeah,
I
was
just
curious
what
like
what
we
could
do
with
that,
that.
H
To
look
into
that
a
little
bit
more
in
the
details,
but
yeah,
I
can't
remember
the
I'm
not
sure
if
we
can
just
apply
like
an
mit
license.
Yeah
I'd
have
to
look
in
I'll,
have
to
look
at
that
a
little
bit
more
get
back.
A
The
other
thing
I
was
going
to
jump
in
and
say
too,
it
also
depends
on
what
you're
distributing
like
odds.
Are
we're
not
actually
going
to
redistribute
mumps
in
a
package
that
we
wrap
right.
So,
if
we're
just
writing
code
that
imports
pumps,
that
should
be,
that
should
be
fine,
and
you
should
be
able
to
license
like
that
piece
that
connects
mumps
to
python.
However,
as
long
as
you're,
not
basically
repackaging
and
redistributing.
H
That's,
I
think,
pretty
much
what
pie
moms
does
like
it
says
in
their
installation.
It's
like!
Oh,
you
have
to
have
monks
built
already
for
to
use
like
we're,
not
doing
all
that
for
you,
but
I
guess
the
condo
recipe
builds
the
sequential
mumps.
So
I
don't
know
how
that
exactly
works.
Maybe
we
would
end
up
doing
something
like
that.
H
Yeah,
when
I,
when
I'll
post
some
stuff
about
it
here
on
slack
once
I
get
it
here,.
I
H
Pie,
mumps
is
just
like
kind
of
like
what
you
were
saying
like
if
we
were
to
write
just
like
a
wrapper
to
use
it
that
that's
all
that
they
do,
they
just
have
the
wrapper
for
it.
You
have
to
have
moms
already
pre-built.
I
H
Yeah,
I
I
was
thinking
of
going
that
way,
first,
just
to
make
it
easy,
because
if
the
wrapper
is
already
built
and
then
looking
at
their
syntax,
it's
pretty
easy
to
use
pretty
straightforward.
But
if
it
comes
down
to
it,
yeah
I'll
start
writing
our
own
rappers
and
it
might.
H
I
Yeah
I
just
looked
around
and
I
think
the
listing
was
actually
quite
diverse.
I
was
from
atmospheric
sciences
to
crustals
and
like
some
of
your
circumstances,
that's
pretty.
G
I
F
F
G
I
Developed
for
a
like
while
ago,
but
I
think
they
it
is
written
in
c,
but
they're
wrapping
up
in
matlab
and
python,
so
yeah
it's
yeah
we
could
like
readily
use.
I
guess
I
haven't.
I
haven't
tried,
but
yeah.
G
G
Could
yeah
any
anybody
could
use
it?
So
that's
that's
an
interesting
yeah.
I
I
Other
than
that,
oh,
I
wrote
an
abstract
also
for
agu
about
the
sparse
inversion.
Well,
it's
not
necessarily
the
sparse
inversion,
but
I
use
the
mixed
norm
inversion
to
answer
some
hydrogeology
question
like
imaging
that
thin
clay
bed,
it's
very
extensive
clay
bath
in
the
central
valley,
so
yeah
people
living
in
central
valley
know
there
are
like
a
very
extensive
clay
layer
which
act
as
an
aquitard.
I
So
simply
their
groundwater
model
is
three
layer
model.
You
have
upper
aquifer
clay
and
lower
aquifer,
so
actually
imaging
that
clay
layer
very
well
is
a
very
important
task.
So
once
you
can
image
that
very
well,
you
can
readily
update
that
their
groundwater
model
as
a
first
shot
so
yeah,
and
then
you
can
expect.
Oh
what
what
kind
of
impact
that
that
made
to
your
model
so
yeah.
It's.
C
A
C
Yeah
not
much
update
on
the
simplex
side,
like
I
worked
a
bit
on
merging
my
bench
with
with
the
new
master
benefit
simulation,
but
with
all
the
name
changes
a
bit
of
a
tedious
process.
So
I'm
doing
one
bit
at
the
time
when
I
feel
for
it
and
otherwise
yeah
I've
just
been
looking
around
like
at
like
what
people
were
doing
with
genpai
or
python
shows,
or
things
like
that,
just
like
trying
to
define
the
next
problem
for
for
me.
What's
the
next
thing
I
want
to
solve.
E
Well,
we're
thick
in
htu
at
this
point
right
I
I
did
have
one
for
john
and
that
is:
do
you
have
a
like
sort
of
a
standard
test
that
you're
using
by
which
you
you
know
you're,
comparing
mumps
and
paradiso,
and
you
know
just
how
long
it's
taking
for
factoring
and
how
long
it
you
know
how
efficient
it
is
with
respect
to
solving
multiple
right-hand
sides
or
just
just
a
standard
test
that
you
might
say.
E
H
Yeah,
I
I
don't
rate
now,
but
that's
kind
of
what
I'm
putting
together
right
kind
of
as
as
we
speak,
I
just
wanted
to
get
it
built,
and
then
we
were
going
to
start
yeah
doing
some
comparisons
pulling
out
the
information
yeah,
how
much
time
it
spends
in
yeah,
factorization
yeah,
nothing
right
now.
I
I
got
set
to
do
for
comparison,
but
just
like
the
stuff
that
I've
been
running
right
now.
H
Is
I've
just
been
yeah
doing
a
fields
problem
for
a
dc
and
running
those
through
each
each
solver
party.
So
or
I.
H
Guess
really
is
what
simpag
is
using
and
yeah
in
the
moms
fortran
in
interface.
Okay,
that
sounds
good.
Well,
yeah,
not
nothing
right
now,
but
I
don't
know
if
you've
got
any
idea
like
if
you
got
a
problem
set
that
might
be
interesting,
a
couple
to
throw
through
there
once
I
get
it
built,
we
can
set
you
up
and
or
we
can
set
up
and
try
for
sure.
A
Have
you
tried
anything
from
frequency
domain
with
it's
a
complex
system
yet.
H
D
G
H
Compiled
the
the
z
library,
which
is
the
double
complex
and
yeah,
everything
worked
out,
pretty
well
same
same
results.
Actually
I
didn't
compare
the
accuracies
yet
but
yeah.
They
generally
did
the
same
thing.
A
Yeah
because
I
I
tried
playing
around
as
a
part
of
sort
of
working
in
this
paper
or
working
on
the
simulations
with
dieter
playing
around
with
trying
to
get
even
cereal,
mumps
and
sort
of
pie
mops
installed.
So
I
went
down
like
a
little
bit
of
a
rabbit,
hole
and
decided
to
stop,
because
I
couldn't
actually
I
tried
passing
it
a
complex,
matrix
and
pie.
Mumps
was
not
a
fan
of
that,
and
so
I
think
it's
actually
something
in
the
wrapper.
A
I
don't
think
it
was
with
the
build,
but
anyways
all
of
these
fun
things.
So
that's
just
that's
what
flagged
for
me
making
sure
with
these
comparisons,
it's
worth,
I
think,
always
testing
something.
That's
real
and
and
something
that's
complex.
H
Yeah
yeah,
that's
the
thing
with
like
bumps.
I
would
assume
that,
like
the
pre-built
package
would
or
install
them
all
but
yeah
doing
it
manually
yeah,
you
literally
have
to
tell
it
like.
I
want
these.
These
kind
of,
like
yeah,
float
or
double
complex
and
yeah.
I
built
both
of
those
to
test
them
and.
A
I
guess
maybe
any
other
updates,
so
yeah
I've
also
been
working
on
agu
abstracts,
there's
one
that
we're
putting
in
on
mt.
If
there's
anyone
else
who
wants
to
join
in
on
that.
If
that's
something
you'd
like
to
contribute
to,
please
feel
free
to
add
your
name
and
please
just
ping
me,
because
I
have
already
set
up
the
submission,
so
I
would
just
need
to
update
it,
but
that's
not
a
problem.
A
If
you
are
interested
in
contributing
to
bots
or
if
you're
you
know,
if
you
don't
add
yourself
and
later
on,
you
can
you
contribute?
We'll
absolutely
add
you
to
the
presentation
poster
things
like
that,
so
so
there's
that
the
other
thing
so
with
that
abstract,
one
of
the
things
that
I
put
in
there
is
sort
of
these
ideas
around
starting
to
monitor
or
have
visualizations
that
you
can
monitor,
inversion
progress
or
perhaps
simulation
progress.
A
So
the
first
thing
I
just
started
playing
around
and
getting
a
sense
of
bokeh
the
plotting
library,
because
you
can
stream
data
to
bokeh
and
actually
like
you
can
set
up
a
server
that
is
sort
of
independent
from
your
jupiter
server,
because,
like
matplotlib
right,
you
just
get
a
static
figure,
whereas
if
we
want
to
be
monitoring
something
we
want
to
be,
we
want
that
data
to.
We
want
the
plot
to
update,
as
new
data
comes
in,
so
I've
started
playing
around
with
it
a
bit.
A
The
api
is
quite
different
than
matplotlib,
so
it
definitely
is
just
it
right
now.
I
it's
extremely
time
consuming,
because
it's
just
like
how
do
I
plot
a
line,
and
I
it
took
me
a
little
while,
but
their
examples,
thankfully,
are
really
pretty
good.
The
other
thing
I'm
I
want
to
sort
of
get
a
better
mental
model
of
is
there's
bouquet,
but
then
there's
also
hollow
views
and
so
hollow
views
aims
at
being
basically
like
a
higher
level.
Well,
this
is
my
understanding
at
this
point.
A
Basically,
it
aims
to
be
an
api
for
all
of
the
flying
libraries,
so
you
can
set
the
back
end
to
hollow
views
to
matplotlib
or
you
can
set
it
to
4k
or
you
can
set
it
to
plotly,
and
so
that's
one
that
I
would
like
to
get
much
more
familiar
with,
because
I
think
that
would
be
a
really
nice
option
down
the
road
is
you
know
you
can
just
choose
whatever
plotting
library
you
like
and
set
that
and
so
thinking
through
some
of
the
plotting
utilities
and
things
like
that
that
we
have
in
discretize
and
probably
in
simpeg.
A
It
would
be
nice
if
we
can
get
to
a
point
where
we're
perhaps
using
that
and
then
like.
If
you're
writing
a
paper,
you
want
to
be
using
matplotlib,
that's
great,
but
if
you
want
to,
you
know,
create
something,
that's
interactive
and
has
more
of
these
html
javascript
style
plots.
Then
you
can
pick.
You
can
pick
your
favorite
library
yeah,
so
I'm
sort
of
in
the
learning
phase.
A
So
if
anyone
else
is
playing
around
with
those,
I'm
just
curious,
I'm
happy
to
sort
of
iterate
and
ask
questions
of
each
other
yeah.
I
think
that's
basically
it
from
my
end.
D
They
make
a
git
stiff
and
see
what
was
changed
like
in
simpek,
what
which
files
were
changed
and
then
they
decide
from
that
which
examples
would
be
affected
and
only
run.
Those
examples
so
sounds
good.
I
guess
it
would
be
quite
a
lot
of
work
to
implement,
but
maybe
in
the
future.
So
in
this
way
we
could
run
the
examples
as
part
of
the
ci
yeah
here
he
is.