►
From YouTube: SimPEG meeting September 9, 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).
C
Yeah,
hello,
everyone
yeah
not
much
on
the
syntax
side,
but
I
just
want
to
mention
that
like
and
it's
just
like
up
to
up
to
you,
but
I
succeeded
to
make
the
request
for
an
audience
to
my
defense.
C
But
it's
like
it's
limited
to
like
something
like
10
people
and
it's
on
invitation
only
so
if,
if
you
wish
to
be
here
at
the
actual
defense,
let
me
know-
and
I
I
will
put
you
on
the
invitation
list,
but
I
just
want
yeah,
but
it's
completely
like
voluntary
and
up
to
you,
but
I
just
wanted
to
let
you
know
that
that
was
a
possibility.
If
you
really
wanted
to
do
that,
but
yeah
it's.
D
C
It's
only
on
sorry
when
is
it
it's
like
thursday,
the
so
it's
like
it's
like
you
know,
eight
days
and
what.
C
A
A
Readings
excellent
soggy.
E
Okay,
I'm
pretty
much
saying
this
dumb.
I
just
gotta,
listen,
yeah,
nothing
still
quite
busy
here,
just
amazing
yeah.
I
really
want
to
do
this
simulation
update
with
1d
and
3d
codes
yeah.
Well,
hopefully,
hopefully
I
can
do
it
sometime
next
week
or
something
like
that
sure
it's
just
a
bit
of
like
it's
a
bit
busy
was
just
finalizing
this
project,
so
yeah
I
still.
E
Luckily,
I
I
had
a
chance
to
work
on
three
different
pilot
studies,
so
I
I
like
I
had
a
chance
to
sort
of
like
act
as
an
end
user
of
all
the
acquired
data
and
see
if
what
we
can
do
so
it's
kind
of
exciting.
But
it's
a
lot
of
work
to
go
through
all
the
data
and
I
think
it
has
different
structures
and
but
I'm
looking
forward
to
packaging,
that
up
for
the
next
month
or
so
and
yeah.
E
But
nothing
really
particular
relevance
today.
A
I
don't
know
thanks
again
actually
dom
one
thing
that
popped
in
my
mind.
When
is
the
seg
workshop?
I
wanted
to
make
sure
it
doesn't
overlap
with
the
machine
learning
when
I'm
supposed
to
be
at.
F
F
So
it's
it's.
They
have
one
line
of
source
and
the
many
receivers
on
the
bottom
and
then
they
bin
it
into
many
source
positions.
I
guess
that's
similar.
You
do
in
airborne
just
in
the
marine
case,
so
you
have
awfully
a
lot
of
shot
positions
actually,
which
is
which
is
sort
of
interesting,
and
it's
interesting.
F
For
many
reasons,
these
are
data
from
2007,
which
is
still
kind
of
very
well
earlyish
for
marine
csm
boom
in
in
the
oil,
but
even
though
it's
kind
of
old
data,
I'm
surprised
that
I
that
they
look
quite
nice,
so
we're
very
interesting
to
see
how
how
the
inversions
will
get
out
of
that
and
other
than
that.
The
paper
that
is
finally
shaping
up
and
lindsay
handed
in
all
the
things
as
well,
so
we
might
get
there.
E
A
I
stopped
messing
with
it
because
it's
one
of
these
things
that
you
can
a.
E
F
A
F
So
we
cannot
actually
exactly
reproduce
it
specifically
where
the
receivers
are
very
close
to
the
stepped
pathology
of
the
model,
so
I'm
pretty
sure
they
used
their
the
the
actual
bath
imagery
and
did
some
finer
meshing
internally
without
the
user,
knowing
the
ones
who
publish
the
data
they.
They
are
not
aware
of
that
and
I
think
yeah,
so
the
fun
of
reproducing
commercial
software
results.
I
guess.
A
A
It
would
be
cool
to
get
some
utilities
around
that,
because
what
I
did
is
basically
from
the
computational
mesh
that
they
used,
which
is
like
nine
million
cells
computed
derivatives
and
to
figure
out
like
where
the
surfaces
are
and
then
through
those
into
the
refinement
to
sort
of
build
up,
build
up
the
mesh.
So
it
actually
was
reasonably
straightforward,
which
is
kind
of
cool.
B
That's
smart.
It
reminds
me
something
that
I
I
wanted
to
work
on
right
to
be
able
to
use
the
basically
do
a
course
and
version
and
do
exactly
what
you
said:
right,
company
gradients
and
then
refine
the
the
tree
and
then
keep
iterating
like
this.
So
we
get
basically
boundaries
really
well
well
defined.
D
Yeah
sure
so,
let's
see
we
had
that
little
bug
thing
on
friday,
but
at
the
time
focusing
on
discretize
for
those
who
weren't
here
just
kind
of
going
over
some
stuff
trying
to
reorganize,
don't
do
some
old
stuff.
As
part
of
that,
we
decided
we're
gonna,
look
at
restructuring
it
a
little
bit
like
improving
some
of
the
pet
bait
names
of
the
modules
moving
around
where
they
live.
I've
got
like
a
kind
of
working
document
of
what
I
think
would
work
I'll
drop
that
in
here
into
the
notes.
D
D
D
You
know
away
from
the
camel
case,
no
more
type
8
friendly
names,
I'm
not
I'm
in
favor
of
that,
I'm
not
sure.
What
else
is
I
mean
it's
a
big,
it's
a
kind
of
a
fundamental
change
way
to
code
works.
We'll
keep
obviously
we'd
keep
the
other
stuff
around
for
a
while,
but
start
moving
towards
those
in
the
future.
D
D
So
I
would
like
to
restructure,
like
we'd
like
to
restructure,
where
the
modules
live
as
well
as
their
names,
so
it
wouldn't
it
shouldn't
really
affect
how
most
people
import
like
a
tensor
mesh,
so
discretize.tensor
mesh
would
still
be
like
the
name
of
that
and
where
it
would
live
as
far
as
somebody
importing
it
or
doing
stuff
to
simplify
the
structure.
The
code
like
the
source
code
itself,
I
think
at
least
reorganize
it
a
little
bit.
So
it's
not
as
all
on
the
front
page
of
the
thing
as
possible.
D
A
D
D
Right,
I
completely
understand
that
I
get
that,
so
there
are
certainly
cases
to
be
made
for
each
that's
why
I
think
we
should
look
at
some
of
these
properties.
I
you
know
case
by
case
and
see
it.
Okay
does
this?
Should
this
be
renamed
like
okay,
the
like
cell
grad
operator
right
now,
I
could
simply
see
rename,
so
it's
so
underscore
like,
although
something
like
that
right,
it's
still
short
sweet.
You
know
exactly
what
it
means.
D
Other
than
that,
I've
still
been
trying
to
figure
out
why
the
tests
are
failing
on
travis,
and
I
can't
for
the
life
of
me
figure
it
out.
So
I
don't
know
what's
causing
that
psych
fault.
I
can't
I
tried
to
like
reproduce
it
on
my
own
system
in
the
same
sort
of
environment.
F
D
D
D
D
E
D
A
B
D
D
A
Cool
thanks
joe
devin.
H
Yeah,
well,
it's
good
to
see
what
you
guys
have
been
up
to
just
kind
of
looking
to
jump
in
figure
out
where
I
can.
I
can
best
jump
in
and
since
I've
kind
of
finished
off
all
the
jif
tools,
tasks
that
I
had
I'm
pretty
ready
to
devote
some
time
to
simpek.
H
So
I
guess
one
thing
I
was
thinking
of
continuing
is
with
the
em1d
stuff,
when
I
left
I'd
been
going
through
the
the
tests
and
I'd
like
to
continue
on
and
finish
those
tests,
and
I
think
once
that's
complete.
H
After
a
little
bit
of
cleanup,
we
may
be
able
to
merge
that
into
simpeg
and
then
start
to
put
kind
of
the
finishing
touches
on
that.
So
we
can't
we
can
do
all
the
the
normal
stuff
right
we
can.
We
can
do
forward
modeling
and
inversion
for
a
single
sounding,
and
we
can
also
do
the
stitched,
1d
inversion
frequency
domain
time
domainium.
H
We
just
need
to
do
all
the
all
the
travis
tests.
Then
I
think
the
other
project
I
was
interested
in
was
the
mt
stuff
and
I
was
wondering:
have
we
been
able
to
do
any
simulations
of
the
3d
mt
problem
on
on
a
tree
mesh?
Have
we
done
that
or
is
it
all
tensor
right
now.
H
Okay,
I
mean
that's
still
something
I
can
I
can
start
with,
but
yeah.
I
guess
I
was
hoping
to
maybe
put
some
work
in
there
once
the
1d
stuff
was
in,
do
like
the
stitched
1d
for
the
mt
problem
and
then
also
try
and
think
about
those
comprehensive
workflows,
but
for
with
with
simpeg.
H
So
how?
How
is
somebody
gonna
come
in
for
a
from?
I
don't
know
an
organization
they've
got
some
data
set,
they'd
really
like
to
go
and
invert
it,
and
we
give
them
just
kind
of
the
the
tools
and
the
work
flow
explanations
for
each
that
they
have
a
successful
be
doing,
and
I
guess
I
wanted
to
get
the
input
from
the
group
before
I
really
get
into
something.
So
I
guess
out
of
those
is
there
anything
that
you
prefer
or
that
you
think
is
maybe
low-hanging
fruit.
A
One
item
that
I
actually
think
is
not
on
this
list,
but
would
be
great
to
have
and
I'll
be
curious.
What
others
think
is
in
the
documentation
to
get
a
publications
page
both
that
includes
like
all
of
our
publications
they
use,
but
then
something
that
other
people
can
easily
contribute
to
and
have
an
easy
structure
for
folks
to
come
in
basically
put
in
the
citation
a
link,
maybe
a
link
to
a
github
repository.
A
So
I
think,
like
sketching
out
what
that
could
look
like
and
maybe
taking
a
look
at
like,
for
example,
pygamily
has
a
really
nice
page
and
just
taking
a
look
at
what
some
of
the
other
pages
are
out
there.
And
what
are
the
things
that
we
should
include
in
that,
because
I
think
you
know
showing
where
simpeg
is
used,
would
definitely
be
a
powerful
way
to
kind
of
show
yeah
what
kind
of
problems
people
are
looking
at
who's
who's
using
this.
A
If
he
says
just
this
past
week
like
it
would
be
great
to
have
to
have
a
central
place
where
people
can
point
to
and
say
like
yep,
I
used
some
tech
here
and
to
kind
of
get
that
work
out
and
known.
So
that
would
be
one
thing
that
I
think
would
be
great
to
have.
I
don't
know
what
others
think.
H
Yeah
I
mean
it
sounds
similar
to
like
the
case
histories
that
we
had
on
em
geoscience,
maybe
not
as
elaborate
as
that,
but
a
section
where
people
could
post
what
they
were
or
yeah
put
published
materials
show
how
they
used
it
for
a
real
problem.
I,
like
it
yeah.
A
Of
okay,
I
guess
from
my
perspective-
and
I
can
I'd
be
curious-
also
what
other
people
think
I
think
the
comprehensive
workflows
might
be
a
touch
early.
It
might
be
worth
more
focusing
on
like.
Are
there
other
issues
that
we
can
go
back
and
sort
of
tackle,
because
I
think
we
want
to
do
maybe,
like
a
high
level
api
and.
D
H
Like
some
kind
of
workflow
there's
there's
a
lot
of
work
that
needs
to
be
done
to
make
sure
we
have
the
tools
to
do
it.
Yeah,
I
I
think
maybe
the
the
1d
stuff
is,
as
is
closest,
to
being
completed
and
is
probably
the
most
impactful
right
now
and
yeah.
Before
I
did
any
kind
of
comprehensive
workflow
thing
like
I
still
need
to
actually
sketch
out
how
the
material
would
even
be
presented
to
somebody
looking
for
it
so
yeah.
I
guess
I
was
gonna.
H
I
was
gonna,
do
the
1d
stuff
and
once
that's
finished
then
maybe
work
with
john
with
mt
related
things
and
see
if
you'd
be
interested
in
that.
D
The
1d
simulation
to
come
up
with
a
background
model
because
it
uses
the
primary
secondary
uses.
They.
It
has
a
1d
background
model
as
the
very
as
the
primary
source
right
figure
out
the
right
hand
side,
and
then
you
know
it
goes
back
and
solves
so
close
to
the
system.
So
it's
not
straightforward.
How
that
translates
to
a
3d
doctrine,
mesh
right,
I'd,
imagine
we
might
have
to
do
some
sort
of
game,
local
volume
averaging
itself,
possibly
or
just
getting
better
about
actually
using
boundary
conditions
in
there.
H
Didn't
you
I
mean,
I
guess,
when
you're
doing
this,
couldn't
you
solve
at
the
boundary
using
the
underlying
tensor
mesh
of
the
octree
and
then
you're
going
to
interpolate
that
solution
onto
the
actual
boundaries
of
that
octree
mesh.
G
H
D
Well,
the
idea
is,
we
need
to
come
up
with
the
upscale,
it's
the
kind
of
upscaling
problem.
Usually
the
cells
are
really
fine
near
the
edges
or
like
at
the
surface
right.
The
cells
are
usually
fine
on
the
surface
they
get
bigger,
but
as
you
go
this
way
with
an
octree
mesh
generally,
like
the
cells
get
wider
at
this
surface,
still
like
the
cells,
get
bigger
and
bigger,
as
you
go
out,.
H
D
D
H
Yeah,
so
you
basically
you've
solved
this.
This
suite
of
1d
problems
on
the
underlying
mesh,
but
to
actually
go
in
and
transfer
that
solution
to
the
octree
mesh
in
a
meaningful
way
is
actually
a
little
bit
more
complicated
than
just
kind
of.
H
That,
honestly,
have
you
talked
to
roman,
I
mean
you
might
want
to
connect
with.
So
actually
what
do
you?
What
is
your
schedule
like
tomorrow,
doug
sorry
joe,
do
you
think
maybe
joe
would
benefit
from
sitting
in
on
our
meeting
with
roman
and
maybe
asking
some
of
these
questions.
G
Yeah,
I
think
that's
I.
I
think
that
would
be
good.
I
guess
the
other
thing
joe.
I
was
one
of
our
the
difficulties
with
respect
to
the
near
surface
cells.
Is
that
coming
from
topography?
Or
is
that
just
the
way
in
which
you
know
the
background
earth
with
no
topography
is
being
discretized.
D
I
mean
I
feel
like
it's
mostly
just
the
way,
even
with
the
background
earth,
with
no
topography
being
discretized
so
like.
If
you
think
about
how
the
solution
behaves
at
depth
rate,
it's
exponentially
became
roughly
from
the
surface,
so
it
still
needs
a
lot
of
there's
a
lot
of
variability
near
the
surface
in
the
solution.
D
H
E
Yeah,
my
feeling,
I
I
thought
like
as
even
simply
like
doing
a
linear
interpolation.
I
thought
that
might
be
fine
like
assuming
we
don't
have
like
big
topography.
I
thought
like
because
I
mean
that's
sort
of
literally
what
you're
doing.
If
you
like
that,
the
primary
secondary
field
solution
you
solve
1d
problem
once
and
basically
use
that
single
solution
everywhere.
E
In
your
main,
the
problem
of
octree,
as
you
go
further
away
from
the
core
cells,
you
got
much
coarser
cells,
but
the
impact
of
that
course
ourselves
like,
although
we
may
have
some
errors
in
there,
but
I
I
guess
that
impact
wouldn't
be
that
large
in
in
where
you
got
the
receiver
location.
That's
my
just
guess
like
but
yeah.
So
I
thought
like
as
a
first
shot,
just
do
whatever
simple
the
simplest
interpolation
and
see
if
you
can
get
the
reasonable
result
right,
if
not.
G
You
know
what
we
want
to
what
we
think
these
other.
You
know
examples,
and
you
know,
runs
with
different
averaging
and
just
sort
of
see
how
things
are
going.
H
Good,
I
mean
we
could
I
mean
if
we
solve
the
problem
and
then
we
put
receivers
on
the
surface
even
out
towards
where
the
edge
of
them
yeah
yeah
out
toward
the
edge
of
the
meshes
and
then
oh,
we
could
just
kind
of
compute
the
apparent
resistivity
and
see
how
close
we're
getting
and
see
we
can
test
it.
That
way,.
A
I
guess
I
suspect,
like
what
soggy
said,
apparently
secondary
approach.
I
think
we're
gonna
be
okay
on
an
austrian
mesh.
I
think
joe's
concerns.
I
completely
resonate
with.
If
we're
talking
about
using
boundary
conditions,
to
drive
to
drive
this
solution,
because
there
then
we
actually
the
discretization
of
the
boundary
matters.
But
if
we
compare
directly
between
the
two,
the
two
approaches
I
mean.
A
G
And
not
particularly,
but
starting
off
with
even
just
a
simple.
You
know
like
the
block
in
a
in
a
half
space
and
just
pushing
things
in
a
number
of
different
directions,
and
seeing
at
what
point
you
know
the
numerics
start
to
break
down,
I
mean
I
think
that
would
be.
That
would
be
interesting
and
then
you'll
gradually
build
it
up,
so
that
we
somehow
are
computing
topography
well
enough
and
then
try
to
see
how
that
is
affected
when
we
use
the
the
tree
just
just
to
get
a
feeling
of.
D
I
mean
I
I
guess.
Fortunately,
it
wouldn't
be
hard
to
overlap
like
the
edges,
so
where
the
you
know,
the
solution
lives
on
the
edges
right
in
the
empty
case
for
the
curl
problem.
So
it
wouldn't
be
hard
to
find
overlapping
edges
from
the
base.
Tensor
1d
and
just
put
it
at
the
cells,
because
all
of
the
edges
of
the
octree
cells
will
still
overlap
with
edges
that
exist
in
the
base.
H
Yeah,
I
have
this
feeling
that,
once
you
talk
to
roman
you'll
find
out
that
the
solution
is
not
going
to
be
as
like,
elaborates
and
and
crazy,
as
you
thought
it
might
have
to
be,
and
that
in
practice
something
that's
it's
a
little
less
complicated
actually
works
pretty
well
in
practice,
and
at
this
point
we've
used
the
the
ubc
gif,
the
fortran
codes
for
a
number
of
real
projects,
and
the
feedback
has
been
quite
good.
H
But
yeah,
I
wouldn't
be
surprised
if
yeah,
the
underlying
mesh
is
used
to
solve
a
host
of
1d
problems
and
then,
since
the
electric
field
is
defined
on
the
edges,
you
just
go
and
find
the
edges
that
match
on
the
tree
mesh
and
then
you
just
define
them
that
way.
D
Yeah,
I
see
how
I
definitely
see
how
it
could
program
and
I
definitely
see
how
it
would
be
decent
results
near
the
edges.
Obviously
I
just
have.
No,
if
anything
is
not
living
on
the
edges,
especially
near
the
outside
it,
I
would
there's
no,
I
don't
think,
there's
any
hope
of
getting
anything
breathable
between
edges
at
the
core
cells
that
are
really
far
out,
which
is,
I
guess,
is
reasonable.
D
Hey
I
had
one
other
thought
recently.
D
D
Yeah,
I
mean
it
tells
it
it
looks
at
this.
You
give
it
nodes
right,
you
give
it
locations
of
nodes
and,
like
a
and
it's
easy
to
think
about
in
a
2d
mesh,
you
basically
give
it
a
grid
of
right.
So
in
a
structured
array
and
those
xy
values
can
be
whatever
you
want
and
it'll
kind
of,
and
then
that'll
that's
what
defines
the
structure
that
meshes
the
structure
of
that
array
you
give
it,
and
so
it
allows
for
arbitrary
points.
So
the
same
kind
of
thing
happens
in
in
the
3d
right.
D
So
like
looking
at
things
how
the
how
the
normals
are
defined,
it
looks
like
it's
like
they
average
or
computes
like
triangulated,
faces
and
then
averages
those
four
normals
of
a
face
to
get
one
normal
for
each
face.
Okay,.
E
Yeah
that
detail-
I'm
not
I
thought,
like.
I
haven't
actually
thought
about
that
you're
right
so
yeah.
You
cannot
define
a
single
plane
for
four
points.
D
To
get
second
order
convergence,
so
whatever
we're
doing
is
probably
sufficient,
but
it
did,
like
I
said
just
didn't,
seem
completely
concrete
to
me
looking
through
it.
So
I
wasn't
sure
if
anybody
else
looked
at
that
or
thought
about
that-
I
know
in
like
so
like
reservoir
simulators
often
have
a
similar
kind
of
mesh
structure
that
they
can
allow
for
arbitrary
grid
points.
Curvilinear
stuff
like
that
and
their
surfaces
are
explicitly
bilinear
curves
on
each
of
the
surface
faces
right.
A
I
think
this
is
wonderful.
I
know
that
he
did.
D
A
Presentation
ages
ago,
where
I
think
he
actually
had
like
a
3d
thing,
you
could
interact
with
to
actually
see
the
structure
of
the
face
so
yeah.
I
think
this
is
this
is
one
we
can.
We
should
bring
him
on.
G
Gator
well
lindsay
is
rowan.
I
I
didn't
quite
get
your
your
problem
that
you
were
talking
about,
but
it
sounded
like
you're
in
the
situation
where
you've
got
a
few
receivers
and
lots
of
sources
and
you're
trying
to
contend
with
that.
But
in
some
situations
like
that,
then
you
can
just
use
reciprocity
and
have
your
basically
your
receivers
act
as
the
sources.
F
F
Which,
which
acquired
slightly
different
than
emgs
and
had
less
source
position
but
with
longer
source
signals,
because
it
was
kind
of
a
much
a
finite
source
waveform
that
they
got
more
frequency
out
of
it.
But
then
you
had
lesser
points,
at
least
in
the
early
days,
so
yeah,
it's
just
the
first
time
I
looked
at
the
mgs
and
it's
more
about.
Then
they
do
this
source
spinning,
because
it's
continuous
the
way
that
they
have
to
source
signal
and
then
that's
why
I
think
that
it's
more
like
airborne.
F
G
There
are
respective
I
mean
if
you,
if
you
thought
about
you,
know
looking
at
really
like
a
plane,
that's
flying
really
very
slowly
right
and
then
you
know
really.
You
should
be
taking
into
account
the
fact
that
this
plane
is
actually
moving
while
things
are
happening
and
that
that
has
definitely
has
potential,
but
it
would
be
that's
another
level
of
problems.
Yeah.
E
Even
I
think
I
did
some
competition
at
one
second,
I
think
so.
The
final
horizontal
resolution
is
about
20
to
40
meter.
So
I
think
you
got
like
50
stacks
or
something
like
that
more
than
like
5200
stacks
yeah.
So
if
you
can,
if
you
can
get
really
clean
data,
then
I
think
actually
doing
the
same
idea
is
actually
sar.
E
Like
intraparametric
like
the
satellite
images,
it
actually
used
that
same
idea
of
your
moving,
and
by
doing
so
you
get
better
resolution,
but
I
wasn't
sure
whether
you
can
actually
get
that
for
kind
of
more
geometrics,
interesting
problems.
I
Yeah
a
few
things
I
guess
I've
been
playing
with
the
mt
still
yeah,
that
1d
inversion.
There
is
still
doing
good
we're
using
it
just
to
qcr
stations
as
we
go
along
collecting
them.
We've
got
about
20
stations
going
doing.
I
did
another
2d
test
with
the
3d
inversion
on
a
nepal
data
set.
So
there's
like
4000
meters
of
elevation
change.
This
is
massive
and
yeah.
It
seemed
to
line
up.
The
surface
was
doing
a
little
bit
of
things
differently
than.
E
I
Inversion
that
the
colleague
there
that
I
was
confirming
with
but
yeah,
I
think
that
was
just
purely
my
discretization
and
yeah
nothing
to
be
concerned
about
he
said,
but
it
looked
pretty
much
similar,
otherwise
that
and
I've
been
playing
with
the
solvers
a
little
bit
again,
I
updated
my
mt
simulation
branch.
There
yeah
something
happened
in
the
survey
when
you
define
the
survey
and
it
just
wasn't
yeah
it
wasn't
initiating
properly
so
now
everything's
good
to
go.
I
I
don't
know
if
that
was
with
an
update
after
I
merged
in
the
new
master
branch,
but
yeah
everything's
pretty
rocking
there.
I
actually
found
something.
I
was
looking
at
the
desk
a
little
bit
more
and
so
to
ask
when
you
parallelize
like
when
you
set
your
processes
and
get
on
a
linux
machine,
you
get
a
little
bit
more
options.
You
can
actually
choose
to
fork
and
spawn
and
then
there's
another
one
and
windows
only
has
the
option
to
do.
I
I
think
just
the
spawn,
and
that
might
be
why
the
solvers
are
kind
of
interrupting
each
other,
because
when
I
run
the
like
the
partiso
solver
in
parallel
with
das-
and
I
use
it
on
linux
and
fork
it-
everything
seems
like
everything
seems
to
work
pretty
well,
except
for
when
you
get
to
really
small
values
like
10
to
the
minus
41,
then
the
values
don't
seem
to
match.
If
I
just
run
it
in
cereal,
but
all
like
anything.
I
Anything
above
like
negative,
like
10
to
the
negative
20,
seem
to
be
like
pretty
spot
on
they're,
not
changing
randomly
like
they
were
when
we
were
testing
it
on
windows.
So
I'm
going
to
look
into
that
a
little
bit
more
see
what's
going
on
there,
because,
even
if
you
run
into
cereal,
those
really
small
numbers,
I
still
get
a
different
value
but
they're.
Just
so
small
though
they're
like
10
to
the
e
or
either
minus
50,
some
or
40
something
but
yeah.
I
was
something
else.
I
was
going
to
look
into
a
little
bit.
I
B
We're
getting
close,
probably
to
the
arithmetic
point
for
that
again.
I
So
yeah
I'm
going
to
run
it
on
a
few
more
tests.
That
was
all
mt.
I
was
running
it
on
and
it
seemed
to
be
yeah
producing
results
that
we
didn't
see
with
the
windows
machines,
so
that's
kind
of
interesting
yeah
other
than
that
I've
just
been
working
on
non-sim
peg
stuff.
Actually,
I've
been
playing
around
the
g05
file
there.
I
was
trying
to
pass
auctri
meshes
of
big
surveys
and
it
was
just
taking
so
long
because
those
files
are
just
so
large
but
yeah
you
throw
everything
into
a
g05
file.
I
B
D
E
E
Yeah,
I
think
I
did
both
mac
and
linux.
I
haven't
tried
on
windows,
so
yeah,
probably
the
reason
why.
I
I
did
it
on
mac
too
actually,
and
it
was
giving
me
similar
results
as
the
linux,
so
it
might
just
be
windows
related
this.
This
problem
could
be
restricted,
anyways,
maybe
there's
a
way
around
it,
but
yeah.
I
was
going
through
the
das
documentation
and
it
said
that
yeah
only
linux
and
even
mac
had
one
restricted
as
well
as
two,
but
yeah
windows
is
the
one
that
only
you
only
have
the
one
option,
whereas
the
other
os's
give
you
a
little
bit
more.
B
But
so
you
were
spinning
your
your
parties
inside
the
inside
of
the
the
cluster
right
inside
this
or
inside
the
the
worker.
Yes,
yeah.
D
I
Right
actually
that
reminds
me
I
did
do
some
more
work
on
simpeg.
I
forgot
so
I've
broken
it
down
kind
of
like
how
joe
did
like
a
mini
survey.
So
now
I
just
break
it
or
I
have
a
utils
in
the
dc
problem
right
now
that
it'll
break
it
down
by
sources
and
then
you'll
have
a
survey
for
each
source,
and
then
you
can
make
your
simulations
and
it
works.
I
can
I
can
go
and
invert,
but
the
thing
is
I'm
a
little
lost.
I
B
Yeah
exactly
yeah,
that's
let's
work
on
it.
That's
that's
super
and
if
you
have
a
notebook,
let's
just
work
on,
it
sounds
like
you've
already
all
done,
all
the
all
the
hard
work
which
is
spitting.
I
B
B
But
yeah
I'm
sure
if
we
can
create
a
utils,
that's
general
enough.
A
A
Excellent
doug.
G
G
Most
of
them
are
archived
they're
on
the
old
jiff
raid,
and
I
think
the
goal
over
the
next
week
will
be
to
kind
of
get
onto
that
system
and
just
sort
of
see
what
we've
got
and
then
kind
of
bring
everything
over
to
dropbox
and
figure
out,
which
ones
we
actually
can
use
for
testing
and
distribute.
G
Most
of
them
have
got
a
number
of
publications
that
are
associated
with
them,
and
so,
if
we
can
yeah
get
those
out,
then
they'd
be
really
great.
Data
sets
to
use
for
all
this.
You
know
some
of
the
synthetic
testing,
so
that's
kind
of
that's
kind
of
the
goal
we're
going
to
get
together
with
devin
and
roman
tomorrow
and
just
see
if
we
can
dig
back
far
enough.
If
all
the
drives
are
around
and
stuff
is
still
there.
A
G
People
to
you
know,
try
a
potential.
You
know
just
the
fact
that
you
know
here's.
The
data
they've
been
published
before
there's
papers
that
have
been
written
and
here's
the
simple
script
to
go
ahead
and
do
it,
and
here
here's
a
result
that
you'd
get.
I
mean,
I
think,
that'll
between
things
like
that
and
also
the
publication
page
lindsey
you
were
talking
about
those
are
really
kind
of
important
things
from
the
point
of
view
of
getting
traction
and
getting
buy-in
and
for
people
to
easily
see
you
know
what's
available
and
how
to
use
it.
A
Forward,
I
guess,
in
terms
of
my
updates,
I
finally
sent
peter
results.
That's
that's
my
achievement
for
the
week
so
yeah
I
mean.
If
anyone
wants
to
play
around
with
it,
we
can
drop
the
link
into
slack,
so
there's
three
models:
there's
a
layer
of
block
and
then
the
marlin
model,
the
layer
in
the
block.
I
ended
up
using
the
tensor
mesh.
A
Although
I
kept
one
of
the
notebooks
that
uses
an
octree
mesh,
I
might
actually
go
and
update
that
one,
because
I
was
able
to
design
an
octree
mesh
that
was
like
better
accuracy.
Just
so
folks
can
see
that
but
yeah
those
are
there
and
then
the
marlin
model
used
the
octree
mesh
so
and
takes
advantage
of
the
new
volume
averaging
work
that
joe's
done.
So
that's
exciting
that
we'll
have
that
out
in
a
publication
and
then
what
else?
A
Oh
I've
been
going
back
and
forth
a
bit
with
some
of
the
folks
at
lbl,
so
avenue,
michael
wilt
and
dave
allenbaugh
they've
been
looking
at
doing
some
casing
simulations,
but
looking
at
resistive
cases,
which
is
kind
of
interesting,
it's
not
something.
I
looked
at
at
all
in
my
thesis.
A
So
they've
got
places
where
they've
got
fiber
glass
cased
wells
that
they
want
to
run
surveys
in,
and
it's
been
interesting
one
of
the
things
that
was
kind
of
that
they
were
surprised
by
and
trying
to
sort
of
get
a
handle
on.
This
is
that
the
casing
can
actually
be
relatively
resistive,
like
10
to
the
negative
six
humans
per
meter.
A
We
really
don't
see
an
impact
of
it
far
away
from
the
well.
It's
really
only
if
you
crank
up
that
resistivity
to
like
ten
to
the
seven
ten
to
the
eight
that
you
actually
start
to
see
that
so
trying
to
understand
kind
of
why
why
there's
a
pretty
dramatic
shift
in
in
there
is
is
of
interest,
so
we're
kind
of
playing
around
with
with
that
right
now
so
kind
of
interesting
to
see.
A
A
A
Yeah,
pvc
or
I
think
this
one's
fiberglass
is
that
they're
really
that
they're
monitoring
or
no
sorry
it's
a
carbon
capture
and
storage
monitoring
problem.
E
Yeah,
I
think
it's
that
where's,
that
the
calgary
there's
a
field
site.
A
E
Yeah,
I
think
they
actually
used
the
carbon
fiber
as
as
a.
A
A
This
one
is
in
florida
the
site
that
they
cover
because
it
can
be,
they
actually
put
electrodes
on
the
outside
of
the
piecing,
whereas
this
one,
the
well's,
already
been
drilled.
So
they
don't
have
that
option
so
now
they're
trying
to
see
like
if
we
actually
put
a
system
inside
the
well,
can
you
actually
see
anything
inside
of
it.
E
Right,
and
are
you
like
solving
a
dc
problem
as
a
first
shot
said
that
what
you
were
doing
they.
A
Started
with
the
em
problem
and
were
sort
of
surprised
by
some
of
those
results.
So
then
we
flip
back
to
the
dc
problem
and
yeah.
You
can
kind
of
see
that
basically
10
to
the
6,
you
don't
see.
H
A
Difference,
I
suspect,
and
it
would
be
interesting
to
play
around
with
this.
I
suspect
it's
probably
the
ratio
between
either
probably
the
ratio
between
the
resistivity
of
the
casing
and
maybe
the
inside
inside
of
the
world
that
matters
the
most
yeah.
It.
E
Might
be
interesting
to
do
this
a5
decomposition
lindsay,
because
what
I
can
actually
see
is
like
if
you
got
good
galvanic
coupling,
then
you'll
absolutely
see
that
impact
yeah.
A
E
If
the
induction
is
the
majority
of
that
response,
then
you
may
not
see
like
oh
there's,
no
difference
whether
you
put
a
resistive
casing
or
not
so
actually
by
decomposing
that
your
fields
into
two
components,
you
can
actually
see
what
is
a
major
impact
and
make
some
intuition.
I
don't
know
just.
A
A
Cool,
I
think,
that's
all
that
I
basically
have
for
updates
yeah.
Does
anyone
else
have
questions
thoughts.
D
I
I
It's
a
double
ipa,
but
it's
more
like
the
hazy
style
and
I
found
hops
that
tastes
like
pina,
colada,
so
whoa,
it's
it's
gonna,
be
an
interesting
beer.