►
From YouTube: SimPEG Meeting December 1st
Description
SimPEG weekly meeting from December 1st, 2021
A
Hello,
nice
to
see
you
guys,
I
see
you
all
for
another
week
of
sempre.
Well,
we
got
devin
joining
us
too.
In
here.
A
Obviously
you
can
see,
we've
got
some
people
in
the
gift
room.
We've
got
a
new
face
joining
us
for
today,
hello,
introducing.
B
B
I
am
a
canadian
guy
living
in
norway
with
I
work
with
a
company
called
emerald,
jew,
modeling,
we're
doing
airborne,
em
specializing
in
geotechnical
applications,
and
since
I'm
on
the
long,
homecoming
tour
visiting
friends
and
family
academy,
I
thought
I'd
pop
in
to
say
hello
to
somebody
that's
working
on
in
jeff
and
on
simpeg
we're
looking
at
ways
of
streamlining
our
whole
workflow
to
make
conversion
simpler
and
easier,
because
what
we
use
today,
what
is
in
this
kind
of
hard
to
interface
with
so
I'm
going
to
be
applying
them
all
today,
just
listening
in
seeing
what
kind
of
stuff
you're
working
on
in
simpek
and
hopefully
getting
a
better
sense
of
what
you
guys
are
working
on.
A
Sure
my
name
is
joe
nice
to
meet
you
hello,
again,
kind
of
in
charge
of
making
sure
everything
comes
into
syntax.
It's
been
thoroughly
reviewed
and
works
properly
kind
of
like
the
last
goal
post
to
get
through.
It's
not
you're
the
gatekeeper,
but
now
other
than
that.
I'm
just
focusing
on
developing.
C
And
so
I'm
mostly
focused
in
simply
developing
the
first
multiphysics
inversion.
Now
we
are
more
coming
in
with
postgradient
etc
from
other
groups
in
new
zealand.
So
that's
amazing,
but
that's
been
my
main
contribution
to
the
package
and
yeah
I'm
actually
getting
back
to
industry
starting
in
in
january.
So.
E
And
then
I'm
devon,
I
have
various
jobs
with
ubc
jiff,
but
a
lot
of
what
I
do
for
simpeg
is,
I
guess,
I'm
the
guy
responsible
for
most
of
the
tutorials
and
working
on
getting
the
api
set
up
and
that
kind
of
stuff
so
less
geared
towards
research,
but
more
about
kind
of
making.
Sure
people
understand
the
functionality
of
trying
to
make
it
more
accessible.
F
G
Egg,
I'm
john.
I
work
mostly
in
industry
with
ds
geophysical
and
adidas
airborne,
I'm
mostly
working
on
frequency
domain,
stuff,
mt
and
dc
resistivity,
and
trying
to
make
conversions
go
faster.
So.
D
D
Yeah,
I'm
sorry
I
did
my
phd
with
with
doug
and
lindsey
together
and
yeah
this
day,
I'm
mostly
working
on
airborne
electromagnetics
and
applying
that
into
california
and
groundwater
issues
so
yeah.
I
think
and
sort
of
the
major
developer
of
the
r1d
em
code,
which
is
sort
of
the
major
tool
that
we're
using
for
inverting
the
airborne
em
data
yeah.
So
let's
see
you
here,
cool.
H
And
I'm
dom
step.
I
Yeah,
I
just
wanted
to
welcome
you,
craig,
really
nice,
having
I'm
sorry,
I
wasn't
able
to
to
get
in
this
morning,
but
I'm
looking
really
forward
to
hearing
what
you're,
what
you're
doing
and
yeah
these
days.
Geotechnical
work
is
pretty
it's
pretty
front
and
center
with
all
of
these
floods
that
are
happening
here
so
really
interesting.
So
yeah
look.
H
And
I'm
don,
I
did
the
I
also
did
my
phd
with
all
the
all
this
crew
and
also
in
the
industry,
marriage
of
science
and
then
kind
of
working
parallel
with
john
right
now
for
optimizing,
the
em
codes
and
some
potential
field
stuff.
That's
me,
nice
to
meet
y'all.
A
F
Yeah,
so
if
you
don't
mind
adding
any
items
published
papers
talks
coming
up
tebow,
we
should
put
a
plug
in
there
for
your
m
r,
yes,
and
then
I
think
we
need
to
decide.
Are
we
going
ahead
with
the
poster
session?
Anything?
Are
we
just
going
to
let
that
pass
for
december.
F
Now
I
don't
have
particularly
strong
opinions.
There's
like
enough
going
on
in
december.
I'd
be
happy
enough
to
like
announce,
tebow's
seminar
and
kind
of
have
that
be
what
we're
pointing
folks
to,
but
if
anyone's
got
a
desire
to
share
what
they've
been
up
to
we
can
we
can
make
this
happen,
so
I
don't
know.
Does
anybody
really
want
to
have
a
synthetic
poster
session?
I'm.
F
A
That's
it.
Okay,
I
don't
know
the
other
announcements
or
anything
so
move
on
to
quick
reports
as
we're
going
through
here.
I
since
last
week,
just
a
few
little
small
updates,
the
cross-grading
stuff
checking
this
stuff
around.
As
part
of
that,
I'm
also
going
through
and
working
on
getting
rid
of
all
of
the
deprecated
items
like
we
talked
about
really.
They
are
now
throwing
errors
parting
people
to
the
right
thing,
so
force
people
to
change.
A
It
it's
taking
a
while,
because
all
the
tests
need
a
lot
of
the
tests
need
to
be
updated
that
we're
using
old
calling
conventions.
So
it's
the
long
process
of
going
through
all
the
test,
folders
make
sure
that
it's
using
updated,
whereas
before
it
was
not
because
it
was,
we
just
left
them
as
they
were
so
they've
tested
both
things
at
the
same
time.
E
A
So
yeah.
E
Stand
of
making
a
couple
of
those
tests,
it
seemed
like
the
so
this
was
for
the
electric
source
or
for
the
line
current
source
in
simpeg
doing
time,
domain
dm
and
it
seemed
like
the
if
we
just
did
a
whole
space
and
put
an
electric
dipole.
The
electric
field
produced
by
this
seemed
to
work
pretty
well,
but
the
magnetic
field
being
produced
by
that
wasn't
accurate.
E
So
anyway,
both
both
did
not
look
right.
Yeah.
F
E
There's,
I
think,
there's
still
some
some
questions
I
have
in
there,
but
there's
there's
a
whole
host
of
things
that
you
could.
You
could
run
and
play
around
with
that.
We
just
need
to
implement
it
and
do
some
more
testing,
but
at
least
it's
all
set
up
so
that
you
don't
have
to
start
from
scratch.
F
E
Through
after
the
meeting
or
whenever
you
have
time,
but
I
I
already
put
it
in
the
tests
for
simpeg-
and
there
is
a
plot-
it
function,
okay,
so
you'd
be
able
to
just
run
it
and
then
see
the
comparison
between
the
analytic
and
the
numerical
solution.
So
yeah
it's
set
up
pretty
good
to
troubleshoot
all
right.
A
That's
a
that's
about
it.
Yeah
I
mean
also
with
devon
we're
going
through
stuff.
I
know
we
talked
a
while
ago
about
the
properties
over
switching
out
from
that
talking
about
just
generally
what
it
would
look
like.
We
talked
about
looking
at
a
few
different
things,
one
being
one
way
we're
talking
about
the
initialization
of
classes.
D
A
D
F
I
A
A
C
F
Uses
that
base
class
one
thing
I
think
I've
seen
I
don't
know
others,
but
I
think
the
required
properties
we
still
potentially
want
to
list
like
have
explicitly
in
each.
If
there
aren't
like
straightforward
defaults.
C
Yeah,
I
was
gonna
go
for
the
same
thing
like
when
I
ever
like.
When
I
look
at
this
and
you
look
at
okay,
that's
the
thing
that
sometimes
like
you
have
to
really
go
up
the
tree
to
see.
What's
the
actual
argument
that
you
need
to
put
in
and
like
I
was
gonna
have
to
say
like
I,
I
understand
it's
like
kind
of
like
a
between.
C
A
E
E
Yeah
something
else
joe
brought
up
was
the
the
fact
that
upon
instantiation,
you
should
be
able
to
use
that
like
use
that
object
right
away
so
say
time,
domain
pm,
you
don't
have
to
set
the
time
steps
when
you
instantiate
it,
but
that
is
something
that's
required
in
order
to
actually
use
the
simulation
so
kind
of
an
idea
of
forcing
it
to
be
a
valid
object
upon
instantiation.
A
Yeah,
this
would
be
kind
of
like
the
default
way
to
do
this
kind
of
script
here
to
go
just
be
very
explicit
that
they're
passed
to
that
class
and
link
to
them.
Obviously,.
A
A
C
C
A
F
D
F
A
D
C
Analytic
solution,
so
that's
a
very
useful
package
to
check
that
out.
Our
like
simulation
code,
work
very
well,
etc.
So,
but
it's
also
a
very
simpler
one,
but
it
has.
E
Yeah
that
that
could
work.
I
guess
I'm
wondering
what
is
the
level
of
urgency
for
getting
the
simpek
api
online,
because
that's
sort
of
where
this
whole
thing
started,
and
then
we
realized
in
order
to
do
that,
we
had
to
exercise
properties.
E
So
then
we
started
working
on
that
and
then
we
want
to
go
and
say:
okay.
Well,
let's
test
things
out
on
on
geoanna.
So
now
it's
like
we're
adding
steps
earlier
which
are
taking
us
further
away
from
our
initial
goal.
I
mean
if,
if
we
think
that
this
is
the
ultimately
the
best
way
to
proceed
that
I'm
all
for
it,
but
it's
just
something
to
keep
in
mind
yeah.
I
think.
E
F
F
C
A
E
E
F
Yeah
I
can
create
that
on
google,
so
just
let
me
know
get
our
pages
up
and
we'll
connect
the
wires.
A
G
Yeah
just
been
playing
around
with
a
little
bit
of
part
of
the
simpeg
side
of
things,
was
just
doing
the
worker
just
a
custom
worker
to
sit
on
the
nodes
that
do
all
the
heavy
lifting
yeah.
I
integrated
pi
diesel.
I
found
that
yeah.
If
you
got
any
old
branches
that
you're
still
using
pi
mkl
through
primatte
solver,
something
happened
on
anaconda
and
they
updated
some
of
their
back
end
and
it's
now.
The
mkl
file
is
lib,
lib
mkl,
underscore
rt.1.so,
but
yeah.
E
E
G
Yeah
paid
so
just
yeah,
it
runs
as
well,
so
I've
incorporated
that
now
and
then
yeah.
I
have
it
writing
the
factorizations,
and
now
the
custom
workers
will
find
there.
So
if
they
have
to
run
the
jvec
or
jt
vect
loops
they'll
find
that
factorization
and
use
it.
So
you
get
like
a
two-time
speed
up
and
loops
there
as
well
by
writing
the
factorization
to
disk.
F
Lindsay
yeah
one
more
quick.
I
got
an
email
yesterday
from
someone
at
mira
about
using
the
simpek
logo
in
geoscience
analyst
to
indicate
sort
of
ui
for
when
simpeg
is
being
connected,
which.
F
It's
good
to
sort
of
illustrate.
You
know
where,
where
simpeg
is
used
and
have
that
branding
with
it,
I
think
what
that
does
bring
up
is
that
we
should
have
just
a
bit
of
guidance
on
where
the
simpeg
logo
can
be
used,
how
it
should
be
used,
all
of
that
sort
of
stuff
actually
documented,
and
so
the
jupiter
rowan
pointed
out
the
jupiter
docks
governance
docks
have
a
very
comprehensive
guide,
and
so
we
can
use
that
as
a
bit
of
a
starting
point
and
kind
of
tailor
it
to
our
community.
G
F
That's
awesome:
we
need
like
a
powered
by
some
type
of.
We
do
have
that
badge.
We
do
have
a
powered
by
simpeg
badge,
but
I
think,
like
it's
great
to
have
the
logo
used
when
simpeg
is,
is
used
and
so
yeah
just
having
some
guidance
to
sort
of
encourage
folks
to
do
that.
I
think
would
be
good.
D
I
ended
up
updating
that
em-1d
branch,
so
I
made
a
pull
request
to
main
seems
like
there
are
some
errors
in
the
existing
main
branch.
So
I
guess,
although
I
think
that
I
have
some
problems,
I
actually
I
got
a
quick
question
to
joe
because
there
was
an
error
for
the
source
class
and
then
previously
we
used
the
src
list
and
then
now
it's
changed
it
to
source
list
and
that
kind
of
triggered
the
errors.
Is
that,
like
a
is
that
expected.
D
Yeah,
I
don't
know
why,
but
it's
that
yeah.
That
seems
that
was
the
pattern,
so
I
had
to
change
that
src
list
to
source
less
to
psd.
So
I
thought,
like
that's,
really
some
sort
of
updates
that
we've
made
but-
and
the
other
thing
was
that
the
version
of
joanna,
I
think
for
a
certain
system.
It
has
to
use
the
lower
version,
so
that
kind
of
sometimes
triggers
the
error
in
the
in
the
travis
like
that
that
azure
test,
so
it
would
be
nice.
Is
there
any
like
a
reason?
A
A
D
Yeah
yeah,
that's
actually
that's
what
I
did:
okay,
so
yeah
as
long
as
that's
fine.
Okay,
that
sounds
good
yeah,
so
if
either
joe
or
death,
if
you
can
take
a
kind
of
look
yeah
because
I
think
it
most
of
like
all
the
passes
related
to
that
part
of
addition
is
passing
so
yeah.
If
any
of
you
can
take
a
look,
it
would
be
great
and
converge
that
within
this
year
would
be
would
be
perfect.
Now.
E
Do
we
have
the
electric
dipole
solution
coded
up
in
that,
or
are
we
just
doing
inductive
sources
for
now.
D
Yeah,
I
think,
for
generalizing
kind
of
that
code.
I
would
basically
use
what
the
theater's
code
and
just
find
a
difference
for
sensitivity
unless,
like
somebody's
interested
deriving
all
this
propagation.
E
And
then
two
at
least
as
a
self-contained
example,
because
I've
been
doing
that
simple
code,
validations
project
and
so
with
lindsay
adding
galvanic
sources
to
the
em
simulations
wanna,
compare
1d,
analytic
and
and
like
a
3d,
octree
yeah.
That
would
be
useful
for
that
that
project,
but
I'm
not
sure
if
anyone
else
has
something.
That's
very
tailored
to
a
particular
research
thing
that
they're
doing.
J
J
Hey
sean,
how
are
you
doing
so?
In
the
last
weekend
I
tested
the
cross
gradient
method.
I
use
a
two
different
synthetic
model.
One
is
the
three
simple
blockies
and
another
one
is
the
more
complicated
it's
a
house
shoe
shaped
synthetic
models.
I
use
that
in
my
paper,
so
it
all
works.
Well.
J
So,
and
another
thing
is
a
question
for
you
guys,
so
how
do
I
implement
objective
functions?
Consists
of
the
three
part.
One
data
misfit
terms
and
second
is
regularization
terms,
and
the
third
part
is
also
is
another
terms.
So
just
wondering
is
any
easiest
way
or
the
fast
way
to
implement
such
objective
functions.
So,
based
on
my
understanding,
I
just
I
needed
to
generate
a
new
file.
J
A
J
H
And
then
your
class
will
need
a
derivative
and
a
derivative
method
inside
right
and
then
that
will
that
way
when,
when
each
of
the
components
are
called
for
the
first
and
second
error,
then
they
have
a
function
to
to
pull
from.
Basically.
H
Well,
shout
out
to
to
you
joe
sag
and
dell
for
helping
out
debugging
mt,
so
it's
getting
better
results
turns
out
that
we
cannot
have
data
right
on
the
node.
It's
a
bad
idea,
don't
do
node
and
then
yeah
draping
is
very
important
for
mt.
As
you
were
saying,
john.
So
lesson
lesson
learned.
D
Yes,
speaking
of
that,
because
the
joe,
I
did
not
realize
that
the
inaugural
code
for
the
edge
interpolation,
it's
basically
the
nearest
neighbor.
So
this
is
something
that
potentially,
is
that
right,
like
that's
what
we've
recognized.
A
That's
what
it's
been,
so
it's
it's!
What
we
can
do
and
what
I've
been
meaning
to
get
to
at
some
point
is
update
actually
interpolation
to
at
least
do
like
tri-linear.
When
the
adjacent
cells
are
all
the
same
level,
it
gets
interpolation.
A
A
A
In
but
it
does
do
a
bilinear
interpolation,
so
all
four
of
the
x
edges
right
are
at
the
four
corners
of
the
cube.
Therefore,
for
the
edges
like
they're
all
like
this,
so
one
two
three
four
within
that
cell,
it
does
a
bilinear
interpolation
from
there
from
the
for
those
components.
H
A
H
Basically,
you
don't
want
to
be
on
the
node
ever
because
that's
that's,
like
you
know,
50
50
chances
that
you're
going
to
be
on
the
wrong
side.
Yes,.
A
A
H
Well,
first
of
all,
you
know
make
sure
that
I
move
all
my
my
data
points
out
of
the
out
of
node.
If
I'm
right
on
a
node,
I
mean
in
real
cases
in
real
life
scenarios,
that's
probably
never
going
to
happen,
but
for
the
synthetic
case
it
would
right.
So
you
need
to
have
a
check
that
you're
not
on
the
node
and
second,
you
need
to
be
draped
onto
the
active
cells.
H
You
cannot
have
data
points
in
the
air
because
the
field
is
discontinuous,
so
you
need
to
have
all
your
points
draped
on.
Basically
those
are
the
two
checks
that
need
to
do
under
the
hood
and
also
the
1d
seems
like
the
1d
values
that
we
use
quite
important
right.
The
kind
of
that's
kind
of
like
follow
devin's
defense
lead
on
this
and
kind
of
have
to
put
a
connectivity
that
makes
sense
for
the
for
the
1d
boundary
boundary
condition.
H
I
was
not
being
careful
enough
and
yeah
yeah.
It's
quite
fickle.
H
Yeah
impedances
yeah,
have
you
you've
done
a
very
received
phase.
John
yeah.
H
Yeah,
that's
what
I
was
going
to
ask
the
group
so
the
whole
like
bottom.
You
know
top
bottom
division,
that's
happening
in
the
impedance
calculation,
so
we're
not
protecting
against
those
crazy
small
e
fields.
Eh
like
we're.
Just
assuming
that
h
is
gonna.
It's
gonna
decrease
as
well.
C
That
I'm
following
well.
H
E
H
H
A
E
D
Yeah,
I
I
don't.
I
don't
really
see
the
case
because
d
you're
not
just
inverting
the
the
single
value,
you're
inverting,
the
matrix,
so
as
long
as
the
determinant
of
the
matrix
is
not
zero
so,
but
I
think
that
would
not
happen
because
the
once
you
got
an
orthogonal
component,
there
shouldn't
be
like
a
zero
determinant
of
that
matrix.
I
guess
so.
I
think
it
it
shouldn't
be
okay,
it
shouldn't
happen.
D
G
Joel
and
I
were
running
a
little
few
like
the
unit
test
there
and
yeah,
we
are
finding.
If
you
move
the
model,
you
could
have
it
fail.
If
you
just
like
position
the
mesh
a
little
differently
yeah,
it
would
fail,
but
again
it
was
if
it
lined
it
up
that
it
wasn't
that
perfect
scenario.
I
guess
it
kind
of
happens,
but
there
I
still
either
agree
with
dom.
There
should
probably
be
a
little
bit
of
protection
just
in
case
because
it
is,
it
seems
like
it
is
a
possibility.
D
H
H
E
H
Just
the
fact
that
you
and
I
devon,
ran
exactly
the
same
same
code.
You
know
same
technically
same
versions
and
we're
getting
different.
Spurious
data
points
makes
me
think,
there's
some
instabilities
on
somewhere.
Obviously
there
was
like
the
whole
node
issue,
but
yeah.
I
don't.
E
Know-
and
there
was
the
fact
that
the
fact
that
I,
like
I
hacked
my
my
mkl,
so
that
pimax
solver
would
find
the
mkl
library
instead
of
using
pi
diesel.
So
I
did
some
hacky
thing
on
my
computer
to
get
that
to
work.
I
need
to
upgrade
to
the
latest
the
latest
solver
that
we're
implementing
in
cipe,
so
that
that
one's
on
me,
but
there's
definitely
other
stuff.
H
Yeah
I
tried
both
installation
and
the
solution
was
the
same.
So
that's
that's
not
a
yeah.
I
tried
the
cg
solve
also
right
just
as
a
as
a
test,
because
I
wasn't
sure
if
it
was
parties
or
something
else
and
man.
I
don't
know
if
you
guys
have
done
iterative
solvers
on
this,
but
not
so
stable
eh
need
to
solve.
You
need
to
solve
a
lot
and
even
then
the
simulations
are
not
are
not
very
consistent,
so
I'll
stick
to
parties,
though
till
further
notice,
all
right.
I
John
I've
got
a
question,
so
you
said
you
inverted
both
impedances
and
also
amplitude
and
phase.
I
Did
you
notice
a
like
a
substantial
difference
in
the
amount
of
time
that
was
taken
on
one
versus
the
other?
You
know
I
nothing
substantial.
No.
I
Okay,
it's
interesting
yeah
because
I
I
remember
in
previous
you
know
years
when
we
did
this.
There
was
actually
a
very
substantial
change
in
the
amount
of
time
you
know,
because
you've
got
that
extra
mapping.
That's
that's
him
anyway!
That's
that
was
one
question.
I
The
other
was
when
you're
testing
do
you
have
an
example
for
how
the
system
works
if
the
phase
actually
goes
out
of
quadrant.
G
That
I
don't,
but
that
is
something
that
I've
been
wanting
to
do.
G
I
think
that's
kind
of
what
I'm
still
waiting
on.
If
I
go
back
to
that,
we
do
have
something
around,
but
I
don't
know
if
people
got
it
to
me
or
my
mt
guy,
but
we
should
have
one
yes.
I
Oh
okay,
because
that
would
be
really
that
would
be
really
interesting,
not
only
from
a
numerical
point
of
view,
because
now
you
have
to
worry
about.
You
know
zeros
in
phase,
but
also
yeah,
just
actually
from
a
scientific
modeling
viewpoint,
because
you
know
many
years
ago
when
we're
acquiring
empty
data
through
the
nelson
batholith.
I
Then
there
was
some
questions
about
the
phase
which,
for
a
number
of
stations,
was
very
compatible
between
stations,
but
it
was
just
blasting
out
of
phase
and
then
the
question
is,
you
know
what's
causing
that?
Are
these?
You
know
just
currents
that
are
kind
of
sweeping
in
from
the
mantle
or
whatever,
which
then
brings
us
back
to
the
boundary
conditions,
and
it
seems
to
me
that
if
we're
testing
the
mt
problem,
that
would
it
would
be
really
nice
to
get
high
quality
data.
I
That
really
shows
that
consistent
out
of
quadrant
behavior
for
the
for
the
phase
and
just
see
how
the
programs
go
and
see.
If
there's
interesting
anything
interesting
scientifically
to
be
learned
from
that
nice.
G
Yeah,
the
data
sets
right
now
have
just
been
2d,
so
we
just
haven't
really
had
anything
to
test
it
out
with
on
senpai
yet,
but
and
also
I
need
to
get
that
data
set,
but
there
is
certainly
some
data
floating
around
I'll
have
to
get
on
to
it.
G
We'll
I'll
try
to
serve
that
up
for
you,
okay,.
E
Thank
you,
john.
If
you're,
if
you're
inverting,
for
what
is
it
the
resistivity
and
phase,
do
you
sort
of
have
like
a
row
x,
x
and
a
row
y
y,
okay,
yeah,
it's
all
four.
E
Question
about
which
inversion
would
be
quicker
and
if
you're
doing
resistivity
in
phase,
but
you
only
had
values
based
off
the
off
diagonal
components.
It
would
probably
be
faster
compared
to
the
x
x,
x,
y
y
x
and
y
y
for
impedances,
because
you
just
you,
have
less
data
to
fit.
G
H
G
Really
yeah
yeah-
I
haven't
found
the
time
to
get
on
to
that,
but
yeah
it's
it's
failing
really
bad.
H
No,
but
I
mean
it
like
it
runs
just
the
for
the
simulations
are
pretty
crummy.
You
can
kind
of
see
a
lot
of
you.
I
think
you
will
need
to
pad
out
a
lot
each
of
the
tiles
and
yeah
I'm
not
sure
how
much,
how
much
savings
we're
gonna
be
able
to
make
out
of
this,
but
that's
a
good
call.
H
A
A
The
way,
I
think,
the
way
that
I
was
I
was
trying
to
do
it
was
I
had
a
like
one
mesh
that
worked
very
well,
and
I
just
moved
that
mesh
around
for
each
for
each
different
station
and
if
it
didn't
work
bother
about
trying
to
overlap.
I
just
did
the
whole
volume
averaging,
like
the
volume
averaging
operation
between
them.