►
From YouTube: SimPEG Meeting June 2nd
Description
Weekly SimPEG meeting from June 2nd, 2021
A
So
hello
again,
everyone
welcome
back
nice
to
see
y'all
just
hopefully
go
through
some
quick
things.
Today.
A
A
A
Just
if
that
works
for
everyone
we
need
to.
You
know,
do
it
today,
because
it's
like,
oh
man,
nobody
planned
for
it,
but
let's
look
at
doing
it
next
week.
So
next
wednesday,
after
the
meeting
I'll,
be
doing
that.
A
On
my
end,
I've
gone
through
and
kind
of
reorganized
it
a
bit
internally,
meaning
that,
like
I've
gone
through
a
bunch
of
the
improvements
that
I
was
that
I
can
see
as
far
as
run
time
for
runtime
improvements,
so
just
going
through
there
work
my
way
through
frequency
domain
stuff.
A
It
definitely
works
a
little
bit
quicker
at
least
twice
as
fast
as
the
frequency
domain.
Stuff
was
for
like
a
lot
of
layers,
and
if
it's
just
like
a
single
layer
stuff,
it
goes
really
fast.
It
just
the
more
layers
less
of
the
speed
up,
but
it's
still
faster.
A
C
If
I
would
be
happy
to
test
that
when
once
it's
ready
to
get
tested
and
to
compare
it,
maybe
to
some
of
the
empire
mode
result
that
should
be
fairly
straightforward
to
do
some
dummy
checks.
A
D
D
B
C
Yeah,
I'm
I'm
happy
to
talk
a
bit
about
that
you've
seen
I
finally
push
it
out
and
I
guess
it's
the
same
as
you
after
discretize,
so
which
number
was
it?
It
just
convinced
me
again
that
release
often
and
release
fast,
because
this
one
I
was
building
up
since
february,
and
it's
it's
in
the
end,
just
more
painful
to
have
big
big
big
commits
than
just
every
couple
of
weeks
one.
C
So
I'm
really
happy
it's
out
and
I'm
sure,
I'm
sure
very
soon.
I
have
to
make
a
bug
fix
release
but
yeah.
I
don't
care
anymore
at
that
point.
So
it's
it's
all
out.
It's
hopefully
stable.
There
is
a
problem
with
the
gradient
that
soggy
discovered.
That's
really
cool,
but
I
mean
the
main
point
at
this
stage,
for
both
codes
is
just
forward
modeling.
So
the
gradient
is
just
something
beta
at
best
as
you've
seen
these
projects
are
finished,
but
I
hope
to
improve
on
them
part
of
my
new
jobs.
C
C
C
C
A
C
B
A
Well,
I
think
discretize
is
probably
closer
at
having
1.0.
The
only
other
thing
I
could
see
adding
to
discretized
for
1.0
release
would
be
like
tetrahedral
meshes.
E
I'm
going
to
try
and
do
some
of
that
today
I
got
caught
up
in
some
other
stuff
in
the
last
week,
but
yeah.
I
should
move
forward
with
that.
Okay.
A
C
Tony
hallam,
from
segway
sack
he
has
quite
a
nice
setup,
so
he
uses
read
the
docs
but
then
uses
github
actions
for
the
gallery,
at
least
for
the
pull
requests.
I
think
and
the
releases.
And
then
you
can
build
your
gallery
on.
Read
the
docs
and
upload
the
artifacts
and
re-trigger.
I
read
the
docs
build
and
then
read
the
docs
will
include
that
or
however,
your
setup
is,
I
think,
you're
not
on
redox,
but
yeah.
C
C
At
the
moment,
rita
docks
has
a
15
minute,
build
limit
and
I
think
a
a
400
megabyte,
so
it's
quite
low
and
funnily
enough
segwisek
would
easily
build
with
that,
but
he
was
just
interested
in
how
to
do
it,
and
so
it
can
potentially
grow,
whereas
github
actions,
I
think,
has
eight
gigs
and
two
hours
or
it's
it's
quite
a
large
limit
that
github
action
has.
You
know.
A
Yeah
and
I'm
just
not
even
sure
if
two
hours
would
get
us
there
for
the
gallery,
all
the
tutorials
and
gallery
examples
that
we
have
like.
I
think
I
had,
I
think
two
hours
actually
two
hours
should
work,
because
I
think
that's
the
limit
that
I
have
on
the
azure
pipelines
right
now.
You
can
go
beyond
that,
but
right
now
it's
like
it
defaults
to
being
like
45
minutes,
but
you
can
push
it
like.
I
said
it
said
it's
two
hours.
C
E
No,
no,
not
really.
It
was
mostly
jif
tools
related
week
for
for
me,
so
I'm
actually
kind
of
looking
where
to
jump
back
into
some
stuff.
E
A
E
Yeah
yeah,
I
mean
I'm
open
to
a
new,
a
new
project.
I
think
I
have
some
time
kind
of
opening
up,
I'm
kind
of
curious
about
the
the
natural
source
stuff
and
the
state
of
that,
and
I'm
following
up
on
the
discretized
tutorials
section.
I
I
don't
know
if
you
ended
up
finishing
off
all
the
boundary
conditions
stuff
and
if
it's
I'm
kind
of
ready
to
pick
up
there.
A
Yeah,
I
think
it
should
be
good
soon.
I
was
actually
going
through
the
documentation,
all
right,
like
the
documentation
again
and
looking
at
the
old
examples
from
the
galleries
and
like
the
old,
tutorials
and
stuff
I
was
like
I
was,
I
reminded
myself,
there
was
actually
a
decent
amount
of
stuff
in
there
from
the
I
mean,
not
as
in-depth
as
what
you
had
done
with
the
tutorials
that
you
made,
but
there
was
actually
some
decent
stuff
in
there
already
about
like
finite
volume
and
like
how
we're
applying
it
and
stuff,
which
is
good.
E
A
E
So
I
mean
yeah,
I
I
read
through
those
when
I
was
yeah
making
the
stuff
that
that
I
made
yeah
the
best.
A
Be
to
go
through
the
like
write,
the
documentation
for
some
of
those
like
some
of
the
methods
functions,
things
like
that
and
then,
when
you're
doing
like
your
tutorials,
hopefully
it
should
make
them
maybe
shorter.
So
you
can
just
link
to
functions
instead
of
like
talking
about
them
or
anything.
If
it
happens,.
E
E
One
thing
I
have
been
trying
to
do
with
the
the
docstring
stuff
is
wherever
I
can.
I
have
been
putting
in
examples
that
will
demonstrate
that
function.
Oh,
hopefully,
that
in
itself
will
will
make
some
of
the
tutorials
shorter
but
yeah
unless
we
get
a
clear
idea
of
what
we
want.
It's
a
lot
of
time
to
put
in
to
find
out.
Somebody
wants
it
to
look
different.
It's
it's
a
very
big
time
waste
for
me.
Oh.
E
Yeah
yeah
so
I'll
finish
up
with
the
dog
strings,
we'll
get
that
happy
and
I'll
try
and
put
in
all
the
you
know,
examples
to
to
show
how
you
would
execute
those
functions.
I
think
that
will
that
will
be
really
useful
and
then,
when
we're
happy
with
that
revisit
the
the
tutorials
yeah
so
yeah,
that's
pretty
much.
My
plan
for
today
actually.
B
Yeah,
no,
not
much
update
for
on
on
the
coding
part
for
me
this
week.
I
might
just
mention
that
I
just
posted
on
seminars
that
there
is
there
will.
There
will
be
a
bcgs
talk
next
week
and
I
could
not
find
like
an
advertisement
on
the
web
etc.
Just
by
the
email
I
received
so
maybe
worth
mentioning
it.
That's
about
it!
For
me,.
D
D
Yeah
the
offshoot
company
coming
out
of
albert
collins
group
at
memorial
tetramishes,
if
you
guys,
are
interested.
D
Let's
call
it
that
way,
just
thinking
of
a
more
general
directive
to
be
able
to
handle
all
the
nonlinear,
you
know
maps
that
we
have
because
right
now
we
can
have
implementing
one
directive
per
kind
of
you
know
of
regularized,
non-linear
rag
or
I
guess
the
destinations
are
doing
it
themselves,
but
I
think
it
would
be
nice
to
just
have
a
general
one
that
just
cleans
you
know
the
rev
only
at
the
end
of
a
gas
newton
and
just
that's
the
that's
the
map
or
whatever
function
regenerated
in
between
iterations
instead
of
having
to
duplicate
code
lock
but
yeah.
A
Okay,
then,
well
I
mean
I,
I
think
the
amplitude
regularization
is
helpful.
It's
like,
as
I
was
going
through
your
code,
to
get
like
the
formulation.
Again,
I
had
a
thought
about
it
and
that,
like
basically,
I
was
like
had
it
had
a
thought
being
like
oh
yeah,
that
was
gonna
be
like
if
I
was
thinking
about
directly
translating
the
magnitude
of
a
vector
into
like
from
cartesian
to
spherical,
like
I
was
like
thinking.
A
A
D
Yeah,
but
you
know
they
kind
of
hacked
it,
they
they're
using
the
amplitude
and
the
in
the
weight
term,
but
they're
they
haven't
explicitly
said.
Like
you
know,
this
is
what
we're
minimizing.
D
D
Maybe
my
derivatives
are
not
right.
I
need
to
double
a
little
check
that
but
and
then
you
know
in
the
mapping
we
also
have
technically.
We
could
also
have
wires
right
because
you
could
have
like
a
vector
model
with
a
scalar
model,
and
then
you
need
to
be
able
to
have
both
the
mapping
so
yeah.
It
all
needs
to
line
up.
G
D
E
Yeah
I
mean
I
don't
know
how
much
extra
work
it's
going
to
be,
but
while
you're
poking
away
at
that,
would
you
be
able
to
add
that
equivalent
source
layer?
You
know
right
how
we
did
with
the
mdru
project.
I
had
to
go
in
and
and
sort
of
fudge
around
with
things.
So
one
thing
I
noticed
was
that
a
constant
we
use
for
stabilization
of
the
forward
problem
depended
on
dz
which,
if
we
don't
supply
will
will
break.
E
E
I
had
to
fudge
a
couple
things
in
the
code,
but
if,
if
you
were
able
to
do
that,
it
could
be
useful
that
might
open
it
up
to
a
bit
of
a
larger
project,
because
I
don't
know
if
we
want
to
use
the
same
simulation
class
for
that
or
if
we
want
to
say.
E
Okay:
here's
a
here
is
a
like
equivalent
layer
class
so
that
the
the
how
you're,
defining
the
tops
and
bottoms
of
cells
is
much
more
upfront
because
right
now,
you're,
like
kind
of
going
in
there
and
playing
so
it's
it's
something
to
to
think
about.
I
don't
know
if
it's
a
lot
of
extra
work
for
you
or
if
you
don't
have
the
time,
but
it's
something.
That's
that's
pretty
useful.
D
Yeah,
the
equivalent
source
is
sorted
in
in
a
pipeline
that
we
want
to
be
able
to
give
to
our
users
right
the
option
to
do
yeah.
That's
a
good
point.
Yeah.
E
Of
think
it
could
be
its
own
simulation
and
then
you
would
actually
you
could
feed
in
maybe
a
topography
object
which
would
define
the
tops
of
your
cells,
and
then
you
could
feed
in
either
a
constant
or
a
certain
other
surface
object
to
define
the
bottoms
of
your
cells,
because.
G
Yeah
well.
A
G
A
What
I'm,
what
I
meant
was
that
like
right
now,
you
could
use
the
3d
simulation
code
to
do
a
public
source
layer
like
right.
You
could
just
have
it.
You
would
feed
it
a
mesh
that
is
three-dimensional
with
only
one
layer,
with
only
one
cell
that
it
would
have
to
be
flat
right.
It
would
definitely
have
to
be
flat,
but.
D
Now,
the
code,
you
you
can't,
you
can
overwrite
the
the
z
nodes
and
then
your
mesh
can
be
like
this.
Basically,
that's
what
that's
what
we
were
doing
right.
E
Yeah
so
either
way
it's
not
it's
not
straightforward.
Unless
somebody
tells
you
about
it,
it's
useful
and
I
don't
I
don't
know
if
there's
any
kind
of
time
differ
there
really
shouldn't
be,
but
there
is
maybe
there
is
some
kind
of
time
save
if
you're
having
a
2d
mesh
instead
of
a
3d
mesh
with
the
same
number
of
cells.
I
don't
know
if
it
really
saves
you
anything.
A
No,
I
don't
think
so,
not
really,
if
you're
describing
it.
The
only
time
that
you
ever
get
like
on
the
right
to
linear
type
meshes
is
when
you
can,
like
you,
compute,
that
with
that
one
note
nodal
function
right
and
then
you
do
the
integral
and
stuff
on
the
right
to
linear
mesh.
You
can
save
it
by
only
protecting
that
nodal
value
once
again
for
once
for
each
node
and
then
summing
them.
E
That's
something
if
you
have
a
3d
mesh,
that's
one
layer
thick
is:
is
the
regularization
gonna
be
smart
enough
when
you're
taking
the
derivative
in
the
vertical
direction?
E
To,
I
guess
either
not
take
it
into
account
or
recognize
that
it's
you're
not
gonna,
be
able
to
take
a
derivative
in
the
vertical
direction.
I
thought
there
was
maybe
going
to
break
there.
So
one
of
the
things
we
thought
about
is
is
putting
two
layers
so
that
the
derivative
and
the
vertical
would
work.
A
D
There's
we
haven't
covered
the
case
for
a
single
cell.
You
know
how
long
that,
like
I
mentioned,
I'm
not
sure.
A
E
Help
us
create
the
issue
yeah
I'll.
I
can
do
that.
F
Yeah,
just
some
updates
on
I've
got
or
I
merged
maine
into
simulations
mt.
I've
got
it
up
to
where
I
had
it
before
so
there's,
just
like
a
few
tests
that
are
in
the
seismic
and
the
flow,
I
think
they're
on
the
examples
that
are
just
failing.
So
I
just
gotta
pick
that
out
and
yeah
I've
started
to
pick
away
at
the
tiling
branch
getting
that
all
brought
up
to
date
and
or
looking
at
what
us
are
looking
at.
F
F
I
don't
know
if
we
wanted
to
chat
about
that
joe.
What
I
was
telling
you
yesterday,
where
I
had
to
remove
that.
If
you
have
the
imaginary
side
of
the
frequency
domain
receiver,
it
applies
one.
It
seems
like
when
I
merge
it
into
what
we
did
before
it
was
doing
it
twice.
A
A
Sounds
good
anything.
A
C
A
C
A
My
statement
was
basically
like
there's
no
reason
that
you
just
have
to
put
the
boundary
any
further
out
when
you're
using
a
curvilinear
mesh
than
you
would
with
the
tensor
mesh
like
the
boundaries
are
handled
very
similarly
to
each
other
and
they're.
The
curl
linear
mesh
certainly
supports
like
it
supports
the
boundaries
wherever
you
get
it
yeah.