►
From YouTube: SimPEG Meeting March 31st
Description
Weekly SimPEG meeting from March 31st, 2021
A
A
Oh,
I
just
had
like
a
I'm
just
putting
that
on
the
radar
again
like
I
was
looking
like
looking
at
discourse.
We
had
like
we
had
again
another
question
about
gravity
and
the
sign
convention,
and
I
still
could
not
find
a
page
with
like
where
we
have
the
convention
for
simply,
and
especially
that's
that
positive
app.
So
I'm
just
putting
that
on
the
table
again
that
something
we
we
should
probably
like,
keep
put
make
a
record
off
and
put
somewhere
that
we
can
point
people
to.
B
Yeah,
there's
probably
just
be
a
nice
page
in
the
documentation.
Clearly,
it's
like
coordinate
system.
This
is
the
synthetic
convention
for
all
of
our
stuff
and
then
just
kind
of
go
through
and
see
how
it
a
nice
discussion
a
nice
little
nice
few
words
I
just
kind
of
mentioned
like
oh,
you
know
traditionally
gravity
is
this
way,
but
it's
impeccable
like
this
is
the
effect
that
it
has
on
the
synthetic
coordinate
system,
how
you
would
transform
it
into
the
synthetic
it's
kind
of
be
nice
and
some
other
things,
probably
good
for
mt
as
well.
B
B
B
Multiple
receivers
of
the
same
type,
it
allows
you
to
attach
one
receiver
to
multiple
sources
when
you
have
those
kind
of
receiver
as
well
as
there's
also
another
small
little
bug
from
the
into
the
ip
that
got
faced
there.
I
just
pulled
that
in
and
then
it
also.
Let
me
look
at
there's
a
lot
of
other
places
that
we
could
still
simplify
the
iq
code
just
kind
of
repeated
code
in
there.
B
The
other
thing
is,
I
just
went
through
and
added
some
tests
to
my
height,
my
boundary
condition
branch
convergence
tests
as
far
as
approximating
those
integrals
that
are
showing
you
guys
so
using
the
things
approximately
those
integrals
in
turn
that
they
converge.
That
second
order
convergence
like
we
would
expect.
B
Like
the
2d
curvilinear
mesh,
integral
like
volume,
integral
same
kind
of
thing,
that
happens
on
the
3d
curvilinear
mesh
on
a
face
like
on
a
surface,
because
it's
kind
of
like
just
a
an
integral
like
it's
just
a
2d
mesh
around
the
boundary,
but
I'm
still
having
my
head
having
trouble
wrecking
my
head
around
how
to
do
that:
coordinate
system
transformation
where
we've
got
like
three
vectors,
we're
going
in
two
of
them
and
then
they're
still
in
3d.
So
it's
like
a
3d
surface
and
then
there's
edge
vectors
that
live
on
certain
places.
B
It's
not
just
a
single
quantities
and
it's
kind
of
I
don't
know
it's
been
wrapping
my
head
around
it.
They
know
it
needs
to
happen,
but
either
way
the
way
that
it
already
works,
that
it's
already
written
works
for
everything,
except
that,
like
edge
vector
boundary
integral
when
it's
not
a
flat
surface
on
the
boundary,
but
so
all
the
other
type
of
intervals
already
work,
because
there's
not
vectors
you're,
only
integrating
things
from
nodes
to
faces
or
things
that
already
live
on
faces.
So
it
doesn't
matter
it's
only
when
you're.
C
D
D
If
we
want
to
continue
down
that
route
or
another
thing
is
to
keep
the
ball
rolling
on
the
documentation
on
the
discretized
website,
so
yeah
we've
had
a
lot
of
plans
and
comments
and
input
for
how
we
want
to
organize
everything
from
tutorials
to
api,
but
we're
not
really
following
well.
I
guess
I
haven't
been
able
to
follow
up
on
that
as
much
as
I'd
like
to
so
I
I
think
it's
kind
of
important-
and
so
maybe
I'll
put
some
energy
there
when
I'm
done.
B
E
Sure
so
I
post
it
in
slack
and
we
can
maybe
circle
back
to
this
at
the
end,
if
you
want,
I
sketched
out
a
document
with
some
suggested
next
steps
for
the
governance
conversation,
so
these
are
kind
of
some
of
my
thoughts
on
on
this.
It's
absolutely
a
sketch,
and
so,
if
there's
stuff
in
there
that
you
disagree
with
or
have
concerns
about
or
want
to
see
changed
or
suggestions
for
improvement
like
please
share,
share
all
those
thoughts.
This
is
meant
to
be
a
conversation
starter.
E
So
what
I
was
kind
of
thinking
is.
We
can
give
folks
a
bit
of
time
to
read
through
this
kind
of
collect
your
own
thoughts.
Add
notes
to
the
document.
You
can
have
sort
of
a
bit
of
an
asynchronous
conversation
first
and
then
pick
a
time
for
a
meeting,
so
we
can
figure
out.
I
don't
know
how
long
folks
want
sort
of
in
between.
E
E
B
G
E
Sure,
well
and
john,
if
you
want
to
read
through
and
also
like,
I'm
happy
to
jump
on
a
quick
call
like
one-on-one
to
catch
your
thoughts,
if
you're
not
able
to
make
it
too
and
that
that
goes
for
anyone
who's.
Not
if
we
pick
a
time
that
you're
not
able
to
make
I'm
happy
to
carve
out
time
to
to
get
thoughts.
One-On-One,
I
think
monday's
a
holiday.
F
B
H
B
B
F
E
So
watch
for
that
on
the
governance,
channel
and
yeah
so
feel
free
to
add
notes
directly
to
the
google
doc.
If
there's
sort
of
discussion
items
you
want
to
ping
on
slack,
go
for
it.
If
there's
anything,
you
want
to
talk
about
one
on
one
or
anything
like
that
feel
free
to
ping.
Me
directly,
I'm
more
than
happy
to
chat
chat
on
this,
the
other
one
I
wanted
to
just
flag
or
the
other
item.
E
E
G
E
F
F
F
I
will
be
the
same
behavior.
The
issue
is
that
it's
cheaper
in
terms
of
you
know,
minimization,
to
put
to
put
anomalies
between
the
positive
and
the
negative
on
the
under
the
data.
Then
I've
been
putting
it
in
the
middle
right,
so
you
end
up
having
you
know,
basically
mag
anomalies
that
are
at
the
wrong
place.
F
J
B
F
I
F
Yeah,
but
they
they
know
that
they
have
different
phases
of
volcanism
and
they
have
some
parts
of
the
remnants
some
but
they're
not,
and
it's
a
it's
at
the
equator
right.
So
it's
really
hard
to
understand.
Just
looking
at
the
data,
we
can't
start
playing
those
games
like
it's,
so
you
know
covering
like
large
areas
and
things
are
changing
constantly.
We
need.
We
need
a
more
foolproof
strategy.
F
It's
trying
to
yeah
that's
why
the
amplitude
inversion
is
interesting,
because
technically
everything
is
an
envelope,
so
you
could
basically
do
a
joint
inversion
of
amplitude
and
tmi
and
then
having
the
amplitude
version.
Try
to
just
fix
the
you
know:
let's
see
what
we
did
for
in
our
you
know
our
abstract,
but
do
it
as
a
joint,
both
an
nvi
and
sorry
do
like
an
mdi
with
two
data
sets.
B
B
F
F
It
can
fit
everything
by
just
putting
magnetic
magnetic
like
basically
cells
all
around
it
with
you
know
the
right,
magnetization
and
so
again,
that's
something
we
need
to
prevent
from
happening,
because
then
we
can't
trust
the
angles
of
magnetization,
and
then
we
just
get
the
shell
of
it
all
right.
So
we
don't
get
like
a
bulb.
B
It's
I
think
it's
like
a
similar
problem
as
far
as
like
so
we're
more
sensitive,
like
we're
actually
sensitive
to
like
a
susceptibility
contrast
like
that.
The
quantity
that
we're
actually
inverting
for
could
be
a
sensitive
sense.
You
know
susceptibility
contrast,
so
we
could
think
so,
wherever
it's
like
constant
in
the
direction,
you
know
kind
of
there's
no
sensitivity
to
something.
That's
not
changing.
F
F
Yeah
you
can
indefinitely
yeah,
you
can
remove.
You
can't
remove
the
wrap
yeah,
but
then,
after
this
you,
if
you
remove
the
raft,
then
you
can't
really
do
sparsity
or
I
guess
you
need
to
do
it
in
two
steps,
like
you
were
doing
suck
right,
do
like
no
ref
and
then
do
another
run
sparsity
with
the
ref
back
in
somehow,
like
the
previous
model,.
G
Yeah,
oh
yeah
yeah.
I
did
a
little
bit
of
stuff
last
week.
Besides
moving
so
yeah,
I
just
worked
a
little
bit
more
on
the
mpi
das
stuff
that
we're
doing
with
our
cluster
and
theis
and
then
yeah
I'm
putting
together
a
little
document.
I'll,
probably
just
share
it
on
curved
note
how
to
use
kubernetes
with
the
existing
build
out
of
desk
and
yeah.
Hopefully
have
that
done
here
soon
before
I
leave,
but
I
I'll
see
I'll
still
be
doing
a
lot
of
coding
at
night
when
I'm
gone.
B
On
hearing
no
no
objections,
so
the
other
thing
I
wanted
to
talk
about
today
is
devin
stephen
posted,
a
an
issue,
something
that
him
and
I
have
been
chatting
about
is
like
just
the
tutorials
in
general
and
one
where
the
data
should
live.
We
kind
of
brought
this
up
last
week
a
couple
weeks
ago,
maybe
about
where
they
like
how
people
can
get
access
to
them
and
as
well
as
just
where
the
tutorials
themselves
should
live.
B
So
I
suggested
that
we
just
kind
of
at
this
point
split
them
off
into
a
separate
repository.
That
is
just
the
template
tutorials,
it's
something
that
we
can.
What
we
can
do
is
we
can
something
we
can
trigger
a
build
on
that.
Every
time
like
a
commit
is
is
pushed
to
the
main
branch.
We
could
trigger
a
commit
to
the
tutorials
to
make
sure
to
have
them
run
at
least.
B
E
Not
triggering
or
only
triggering
builds
on
a
pull
request
so
not
having
those
those
run
every
single
time.
But
if
there's
like,
if
it's,
if
it's
a
pull
request,
then
then
trigger
it.
Only
on
the
one
direction,
not
both
both
ways,
because
in
some
ways
it
would
be
nice
to
keep
them
sort
of
in
in
sync
and
like
in
in
one
place
so
that
we
don't
kind
of
lose
track
or
start
to
neglect
and
be
repository.
D
I
guess
one
thing
I
was
thinking
is
it's
important
that,
whatever
whatever
is
on
our
main
branch
or
whatever,
whatever
has
been,
I
guess
as
a
release?
The
tutorials
have
to
work
with
those.
So,
whatever
tools,
we're
saying
this
is
our
latest
official
version
has
to
be
working,
but
it
would
be.
I
guess
you
just
you
need
something
also
if
you
were
developing
stuff
to
to
test
it
as
well.
E
D
I
think
I
think
that's
a
big
part
of
it
that
really
kind
of
prompted
this
I
mean
even
a
more
pressing
thing
than
tutorials
is
finishing
off
the
api
documentation,
we're
just
kind
of
lagging
behind
in
in
some
of
that.
I'd
really
like
to
finalize
just
the
auto
generation
of
the
api
docs
and
then
come
up
with
a
plan
for
where
we
want
to
put
these
tutorials,
and
I'm
I'm
totally
happy
with
that.
D
F
And
it
kind
of
takes
the
simpek
project
itself.
Out
of
you
know,
on
attention
right,
like
yeah,
we
want
to
focus
on
the
package
itself,
not
the
use
cases.
That's
what
I
was
trying
to
say
right
and
then
my
comment
was
that
use
cases
should
live
outside.
Like
you
know,
in
a
simple
package,
it
should
be
should
be
light.
Look
at,
for
instance,
sci-fi
right.
F
I
downloaded
it
yet
last
night,
it's
eight
megs
right
sci-fi
is
tiny
and
it's
way
bigger
project
than
than
synthetic
right
and
our
our
repo
is
getting
almost
the
same
size
as
sci-fi.
So
I
think
we
need
to
focus
focus
more
try
to
like
you
know
what
is
senpaig.
What
are
we
trying
to
do
in
synthetic
and
then
use
cases
can
live
like
outside?
You
know
be
like
parallel
projects,
basically.
E
I'm
a
bit
concerned
with
discoverability
of
the
tutorials.
Then
I
mean
because
right
now
it
is
nice
that
they
are
in
the
docks
and
at
least
like
if
you
install
some
paper
from
pi-fi
or
condo,
forge
the
tutorials.
I
don't
believe
come
come
with
it,
so
that
shouldn't
impact
the
size
of
the
install
and
unless
you're
a
developer
and
then
install
from.
F
Yeah,
but
we
can
link
right,
we
can
just
do
external
projects
and
then
link
it
to
to
like
a
separate,
separate
pages.
What
type
is
that?
Does
you
know?
There
is
like
here's,
all
the
other
people
who've
done
the
work
with
privacy
they're
exposed,
but
take
them
out
right,
you
don't
need,
doesn't
need
to
be
lived
inside.
D
Yeah-
and
I
guess
I
thought
that
the
we
have
the
simpeg.xyz
landing
page,
which
kind
of
says
this
is
all
the
stuff
we're
working
on.
We
have
a
link
to
the
simpeg
documentation,
which
I
think
we
want.
We
want
to
make
that
leaner.
We
have
a
link
to
the
discretized
documentation
and
then
I
guess
there
was
pi
matt,
solver
and
something
else,
and
then
I
guess
we
would
just
have
a
place
there.
B
D
E
I
guess
I
don't
know,
and
this
might
be
a
question
we
should
have
been
asking.
The
slack
is:
what
site
do
people
interact
with
first,
because
I
wouldn't
be
surprised
if
most
people
are
interacting
with
docs.simpeg.xyc
first
and
that's
where
they're
going
to
learn
about
simply,
and
if
that's
the
case
then
it's
like,
I
would
suggest
that
we
make
it
very
highly
discoverable
from
from
there
I
mean
we
should
absolutely
have
examples,
sort
of
from
the
site
too,
but
I
I
would
suspect
that
a
lot
of
people
are
finding
it
through
the
dogs.
C
B
B
E
B
B
J
D
D
Okay,
and
I
guess
by
by
calling
the
tutorials
test
somehow
from
the
simpeg
test
sort
of
ensuring
that
works.
But
there
needs
to
be,
I
guess,
a
process
of
saying
I've.
We've
just
released
something
a
new
version
of
simpeg.
We
need
to
now
let
the
tutorials
know
that
they're
they're
calling
a
newer
version
of
simpek,
am
I
making
sense.
D
H
H
But
what
I
try
to
do
now,
with
the
finishing
of
my
project,
is
have
a
smaller
set
still
in
the
docks
that
is
run
all
the
time
to
catch
things
early,
because
sometimes
it's
annoying
when
everything
is
ready
and
then
you
run
the
gallery
and
you
find
out
that
you
messed
up
something
that
is
not
in
your
tests
yet
because
your
tests
are
never
complete
and
then
it
fails
so,
but
simpek
has
that
anyway.
So
I
don't
think
it
should
be
a
big
problem.
F
F
H
I
would
actually
agree
with
you
davin,
because
I
often
go
to
the
to
the
api,
also
on
discretize
and
simpek,
because
to
find
something
in
a
tutorial.
If
you're
not
doing
exactly
the
same
is
sometimes
difficult,
but
then
the
api
often
still
lacks
the
information
or
a
small
example
how
you
would
use
it
yeah.
I
actually
think
the
tutorials
are
probably
more
entry
level
really
good,
but
then
everyone
who
is
using
it
more
frequently,
I
think
good
apis-
would
help
a
lot.
D
B
B
D
B
D
D
B
B
B
D
D
B
H
F
D
Yeah,
some
of
them
there's
some
of
them
have
octree
models,
I'm
not
storing
any
octree
meshes.
I
don't
want
to
store
any
sensitivities,
but
it's
compressed.
That's
compressed,
no,
no
uncompressed,
okay,
compress
them
yeah.
First,
I
took
a
sort
of
a
cursory
look
at
kind
of
the
size
of
the
simpeg
repository
versus
the
size
of
data
files
and
stuff.
I
was
keeping
in
the
tutorials
and
went.
Oh
that's
a
pretty
significant
percentage.
D
C
B
Inversion
scripts
because
it's
like
it's
almost
it's
like
I've,
said
it
before,
but
it's
the
kind
of
the
same
logic
that
oh
yeah,
we
don't
know
the
model
or
we
don't
know
the
true
model
for
the
inversion.
Well,
I
mean
we're
loading
in
the
true
model
to
display
it
anyway,
so
we
might
as
well
just
recreate
it
functionally.
Instead.
D
Yeah-
and
I
I
did
that
because
I
didn't-
I
didn't-
want
the
tutorials
to
be
crazy
long
or
have
this
really
lengthy
step
that
would
kind
of
distract
you
from
the
flow
of
what
you're
supposed
to
be
learning.
But
if
the
user
does
anything
to
to
change
the
mesh,
that's
going
to
be
used
in
the
inversion,
then
the
size
of
the
models
don't
match
anymore,
so
yeah.
I
kind
of
see
that
maybe
recreating
the
the
models.
D
D
F
F
D
Yeah
I
mean
I
do
like
the
idea
of
people
actually
being
able
to
go
and
like
physically
look
at
these
files,
instead
of
kind
of
something
that's
taking
a
thing
from
the
internet
and
then
loading
it
into
a
script.
I
know
I
would
actually
want
to
go
and
have
the
option
to
open
the
data
file
and
and
see
it.
B
E
D
Yeah,
it's
another
another
part
of
actually
like
unzipping
it
and
having
this
full
data
file.
Is
that
it
lets
you
teach
the
users
how
to
use
the.
I
o
utilities
that
I've
been
developing
and
that's
really
like
a
place
where
somebody
would
start
is:
oh
I've
got
this
formatted
dcip
file
and,
and
then
you
just
don't,
do
the
downloading
step.
Basically.
E
E
Actually,
I
think
the
century
one
we
kept
the
data
in
the
repository
for
the
transform,
but
one
of
the
things
that
I
did
actually
sort
of
early
in
that
first
notebook
is
actually
print
out
the
contents
of
the
file
or
the
first
10
lines
of
the
file,
or
something
like
that
in
the
in
the
tutorial
itself.
So
then
folks
can
actually
take
a
quick
look
and
if
they
wanted
to
drill
a
level
deeper,
they
can
go
and
open
up
the
file
in
their
own
text.
Editor.
D
Okay,
yeah,
I
guess
I'm
just
hoping
it
doesn't
go
from
a
file
someplace
directly
into
the
framework.
I
would
I
want
something
where
it's
download
and
obtain
the
file
and
then
unpack
it
and
then
have
a
step
that
says.
Okay,
you
have
a
data
file,
you're
loading
it
with
a
standard
I
o
or
whatever
I
o
functionality.
We.
D
E
E
B
So
do
we
have
it
we're
going
to
keep?
Do
we
have
like
a
kind
of
decision
on
what
we
want
to
do
with
the
tutorials?
Then
I
mean
often,
if
you
I
mean
it'll,
be
a
few
steps
on
that.
That
seems
like
we're
kind
of
generally
leaning
towards
trying
to
split
them
up
and
see
how
like
see
how
it
goes
see
what
we
can
automate,
what
we
can
do
that
way
just
in
the
future.
It's
not
super
relevant
to
this
second,
but.
B
B
D
There's
only
one
one
short
thing:
we
did
a
bunch
of
development
with
ubc
codes
for
frequency
domain
on
octree
and
compared
that
against
simpag
we
had
the
ability
to
put
in
a
susceptibility
model
and
the
simpag,
the
analytic
and
the
e3d
codes.
All
of
them
show
quite
a
lot
of
agreement,
good
agreement
for
layered
earth
and
for
a
sphere
in
a
vacuum.
So
yeah
in
terms
of
I
think
maybe
some
improvements
could
be
done
to
the
the
3d
fem
in
in
simpeg.
But
everything
is
is
agreeing
quite
well.
H
Now
that
brings
me
up
there
that
reminds
me
lindsay.
You
were
asking
once
about
the
error
level
we
were
trying
to
achieve
in.
You
know
comparison,
but,
as
you
saw
now,
rooney's
comment
that
yeah
he
really
thinks
in
maureen.
You
should
keep
below
one
one
percent
relative
error,
so
otherwise
he
says
well
to
to
to
detect
something
three
four
kilometers
below
the
seafloor.
You
need.