►
From YouTube: SimPEG Meeting June 16th
Description
Weekly SimPEG Meeting from June 16th, 2021
A
Again,
I
see
all
your
faces
every
week
just
to
see
hope
everyone
had
a
good
weekend.
You
guys
all
got
whatever
you
needed
to
get
done
done,
always
good
to
have
something
accomplished.
Add
your
names
to
the
attending.
Add
some
quick
reports.
If
you'd
like
to
update
us
on
some
stuff
as
far
as
agenda
items,
we're
gonna
again
discuss
with
last
week
we're
talking
about
the
deference
related
docs.
A
I
know
I
put
in
a
read
through
at
least
some
point
before
the
weekend
of
the
week.
It
looked
good
to
me.
I
was
hoping
other
people's
read
it
as
well
and
added.
You
know
thumbs
up
on
some
comments
and
things
have
you
seen
anything
come
through
lindsay
outside
of
that
or
outside
of.
B
I
think
we
got
comments
from
most
folks.
I
realized
that
the
pull
request
we
had
was
just
for
one
minor
fix
that
soggy
made.
It
wasn't
for
the
full
thing,
so
I
started
a
pull
request
for
the
full
thing,
but
if
you
commented
on
either
that's
that's
fine
yeah,
and
so
I
haven't
seen
any
suggestions.
Does
anyone
want
more
time
to
review
this.
B
B
Okay,
so
then
this
next
steps
I
would
suggest
is
we
can
go
ahead
and
merge
this
and
then
do
actually
a
set
of
release.
Notes
with
it.
I
don't
think
it
needs
to
be
anything
long,
I'll,
just
kind
of
point
to
the
point
of
the
changes,
but
so
that
we
kind
of
have
a
check
mark
in
at
a
point
in
time
that
this
got
done
yeah.
So
I
can
do
that
after
the
meeting.
A
A
B
I
added
one
items
there
and
posted
this
on
slack
as
well
so
soggy,
and
I
were
chatting
last
week
about
the
idea
of
maybe
having
like
simple
sort
of
science
talks
so
more
like
a
it
could
be
research,
talk
or
talk
about
a
project
you're
working
on
and
having
that
something
like
maybe
a
monthly
cadence
I
feel
like
weekly
is
too
much
for
that,
and
even
every
other
week
is
is
a
fair
bit,
but
I
would
see
it
as
an
opportunity
to
a
kind
of
keep
a
pulse
on
what
people
are
working
on
and
then
also,
I
think
it's
a
great
chance
to
engage
folks
who
aren't
necessarily,
as
involved
like
in
actually
changing
the
code,
but
use
use
the
code
regularly
like
I
would
love
to
hear
from
craig
or
sean
walker,
others
who
are
using
simpeg
somewhat
regularly,
but
wouldn't
necessarily
come
to
these
meetings
or
sort
of
be
touching
stuff
in
the
code
base.
D
B
Yes,
we
can
come
up
with
something
we'll
leave
that
open
for
ideas.
Does
anyone
have
a
preference
on
day
or
sort
of
timing
in
the
month?
B
What
I
would
suggest
is
we
maybe
try
and
pick
a
day
that
we
try
and
do
this
regularly
on,
like
the
I
don't
know
the
last
thursday
of
the
month
or
something
like
that
and
then
be
flexible,
though
on
the
actual
time,
because
we
do
have
people
in
a
bunch
of
different
time
zones,
and
so
we
can
try
and
try
and
fit
a
time
that'll
be
reasonable
for
both
the
speaker
and
most
of
the
people
who
want
to
attend
but
yeah.
D
And
this
may
open
up
some
sort
of
invitation
because
I,
I
think
not
necessarily
all
syntax
users
are
attending
this
meeting.
So,
for
instance,
some
of
my
call
is
actually
actively
using
syntax
but
they're
not
necessarily
interested
in
coming
here
and
talk
about
symptom
itself,
but
they
might
be
loved
to
talk
about
how
they
use
disinfect
for
their
research
and
what
they
want
to
see
so
yeah
this
probably
opened
up.
Maybe
I
think
not
us
talking,
maybe
other
people
talking
about.
B
Yeah,
that
would
be
great
well
soggy,
if
you're
a
game
to
kick
us
off.
Is
there
a
time
a
day
in
the
near
future,
you'd
like
to
aim
for.
D
I'm
pretty
flexible
so
hopefully
if
he
set
a
relatively
regular
date.
I
don't
know
like
fourth
week
of
thursday
or
something
like
that
I
could
I
could.
I
could
do
a
kickoff
of
some
sort
of
something
related
to
groundwater.
B
Do
we
want
to
try
like
the
first
thursday
of
the
month,
so
that
would
be
like?
I
said
it's
candidate
day
this
time
around,
maybe
the
second
second
thursday
of
the
month.
So
the
eighth.
B
Cool
okay:
well,
we
can
get
that
kind
of
organized
and
we'll
pick
a
time
and,
and
we
can
record
those-
and
I
think
it
would
be
great
to
actually
you
know
we
can
make
a
habit
of
kind
of
highlighting
them,
maybe
even
providing
like
a
little
abstract
too,
like
we
could
start
a
curved
note
project.
That's
kind
of
on
these
talks
and
link
to
the
slides
have
a
short
abstract,
a
short
bio.
D
What
is
the
relevant
time?
It's
like
a
15-day
talk
or
45-minute
talk,
or
what's
what
it
could
be
a
desired
15
minute.
Is
that
better.
B
I
could
go
a
little
longer,
I
would
say
maybe
like
plan
for
kind
of
a
20
minute
half
an
hour
and
then
have
half
an
hour
of
discussion.
Kind
of.
C
And
I
always
suggest
that
we
do
it
in
the
morning
so
that
we
can
get
the
east
coast
europe
right,
because
the
end
of
the
day,
otherwise
it's
just
australia
and
the
west
coast.
That
would
be
my
suggestion
and
thursdays
is
bcgs,
sometimes
also
4,
30,
so
yeah
right.
A
A
And
then
we
can
move
on
to
some
quick
reports,
so
I've
I've
basically
gone
through
the
time
domain,
em1d
stuff
and
I've
got
it
into
something
that
I'm
happy
with
and
meaning
like
just
the
simulation
classes.
I
still
haven't
like
gone
through
and
looked
at
soggy
how
you're
implementing
all
like
the
tiled
or,
like
the
sketch
conversions
like
that,
how
like
how
to
best
implement
that.
A
But
I
don't
I've
seen
it
and
I
I
can
imagine
how
I
would
extend
what
I've
done
to
that
pretty
quickly,
but
it
I
was
able
to
generalize
it
to
like
arbitrary
waveforms.
I
I
switched.
I
added
a
new
waveform
class
that
was
kind
of
more
similar.
I
think
it's,
I
added
a
piecewise
linear
waveform,
so
I
believe
that
was
more
like
what
was
the
general
waveform
in
the
original
em1b
code.
A
It
was
just
treated
as
a
piecewise
linear
function.
We
can't,
I
don't
think
it's
necessary
for
us
to
support
like
the
raw
source
waveform,
because
it
could
just
be
anything
and
it's
hard
for
us
to
know
what
it
is
like
ahead
of
time
like
that,
overall
waveform
is
just
any
function
right.
So
it's
hard
for
us
to
really
support
anything
with
it
as
far
as
the
m1d
integration,
but
everything
that
is
well
defined
in
there.
A
A
Property
on
the
waveforms
that
knows
like
okay,
these,
it's
piecewise,
they're
piecewise
functions
right,
so
it
just
adds
the
nodes
for
the
piecewise
and
then
integrates
between
them.
So
just
I
added
like
very
minimal
amounts
to
the
original
waveforms
to
get
them
to
work,
and
then
I've
kind
of
gone
through
and
just
it's
it's
pretty.
I
don't
know-
I
think
it's
like
it
uses
like.
A
I
just
had
to
do
a
really
high
order,
calcium
blogger
on
it
for
the
convolution
integral,
because
the
way
I
was
doing
it,
it
didn't
matter
how
high
how
high
of
an
order
it
didn't
change
like
it
doesn't
change
the
time
to
actually
forward
model
it.
It
just
changes
it
a
little
bit
it
takes,
makes
it
take
a
little
bit
longer
to
set
it
up,
but
every
time
you
write
it,
it's
the
run.
Time
is
in
the
significant
point.
D
So
did
you
expand
that
to
handle
on
time?
Let's
say
because
the
assumption
that
I
made
was
the
I'm
only
handling
the
off
time.
Did
you
expand
that
assumption?
Like
can?
Can
that
handle
yeah.
A
I
think
I
haven't
tested
it,
but
it
should
it
just
basically
added
in
like
okay,
if
I
try
to
do
integrals.
If
I
try
to
calculate
the
step
response
before
the
step
happens,
it's
zero
all
right.
That's
about
what
we
needed.
D
Here,
yeah-
and
I
also-
I
also
made
an
assumption
that
that
that
doesn't
matter
what
time
you
put
it's
always
like
the
definition
of
time
gates
it's
after
I'm
like
when
the
current
is
zero.
Did
you
so
did
you
kind
of
like
expand
that
assumption
such
that
that
it
can
kind
of
align
with
the
the
current
definition
of
time
domain
code
because,
like
I
think,
the
synthetic
time
domain
it
it
just
follows
the
waveform
the
time
the
time
defined
in
the
waveform,
and
then
it
is
relative
to
to
those
waveform
time?
D
Yes,
I'm
happy
to
review
and
it's
actually
quite
timely.
So
one
of
my
colleagues
who
just
finished
his
phd
is
working
with
californian
department,
world
water
resources
and
they
are
doing
this
large
scale
am
project
in
these
days
and
he
seems
keen
to
use
that
code
and
he
was
actually
asking
me
do
I
need
to
use
that
the
old
branch
of
1d
code
or
the
latest
one
so
yeah.
If
you
can
kind
of
push
this
and
then
merge
as
soon
as
possible,
you.
A
A
D
A
It
already
handles
all
this
stuff
where
it's
like.
Okay,
just
I
hit
soggy,
I
went
through
and
handled
that
you
know,
let's
try
to
group
them
up
by
sounding.
A
I
handled
that
in
a
little
bit
simpler
way,
whereas,
like
I
go
through
it,
you
compute
all
of
like
the
wave,
the
wave
numbers
that
you
would
need
and
then
just
go
through
like
okay,
I'm
only
actually
going
to
compute
it
on
the
unique
wave
numbers,
and
not
all
of
them
like
repeatedly
for
everything
so
that
set
that
will
kind
of
that
little
transform
handle
is
what
you're
trying
to
do
a
little
bit
more
elegantly.
A
Put
forth
okay!
Well,
that's
what
I've
got.
D
Speaking
of
that
joe,
I
had
a
chat
with
theater.
Maybe
this
is
not
urgent,
but
something
that
we
could
potentially
consider
actually
expanding
that
1d
code,
not
only
the
specific
to
a
ground
loop
or
airborne
em.
We
can
actually
switch
that
1d
kernel
with
em
pi
mode.
So
for
other
types
of
survey.
We
can
basically
use
the
empire
mode
to
expand
that,
so
we
could
have
a
little
bit
more
general
em-1b
that
could
handle
various
types
of
em
when
the
inversion,
but
just
a
thought,
yeah.
But
it's
probably
worthwhile.
C
Okay,
just
to
check
joe
confirm
are
we
just
gonna
still
be
able
to
run
general
web
forms
where
it's
like
a
piecewise.
That's
why
it's.
A
D
Joe
one
more
check
like
have
you
seen
how
I
handled
the
the
dual
moment
and
kind
of
did
you
like?
Did
you
keep
that
or
did
you
kind
of
extend
it.
A
It
I
didn't
worry
about
handling
the
dual
moment
information,
but
the
way
it
works
right
now
is
that
it
like
it
still
just
computes
a
single
step
response,
and
then
it
involves
it
with
everything
right.
So
if
you
just
had
multiple
like,
if
you
just
split
it
up
into
two
receivers
like
two
sets
of
sources
and
receivers,
it
shouldn't
make
it
take
any
longer
to
simulate,
it
would
automatically
reuse
that
waveform,
like
it
already
radically
reused
the
step,
responsive
calculator.
C
Perfect,
that's
good
to
hear,
because
you
know
it
was
nice
side
that
you
coded
up
where
it
was
so
hard
on
the
user
side
right
depending
on
the
system.
Then
you
need
to
build
a
different
type
of
survey.
If
you
can
just
do
it
in
sequence,
you
know
just
like
okay,
we
have
a
high
moment
low
moments,
basically,
two
different
simulations.
A
Okay,
yeah:
that's
what
I
have
move
on
to
peter.
E
Yes,
yes,
our
paper
got
this
morning
finally
accepted
the
one
I
was
working
at
lindsay
and
raphael
from
custom
and
octavio
from
petcham,
so
that
comes
to
an
end
soon
we're
happy
about
that
and
other
than
that.
I'm
working
at
the
moment
a
bit
more
intensely
with
sauki
to
get
this
emg
3d
as
a
forward
modeling
kernel
for
simpac
going.
So
we
hope
to
have
there
something
concrete
generalized
in
the
near
future.
E
E
Maybe
even
sogi
has
some
csamt
data,
so
maybe
even
direction
empty.
We
will
see
how
that
goes.
D
Got
the
the
data
from
beric
and
also
from
doc,
we're
just
collecting
the
data
and
not
working,
but
we're
going
to
work,
but
yeah
it'll
be
kind
of
interesting
to
invert.
Both
data
and
see
kind
of.
A
F
F
Mosquito
yeah,
I
haven't
gotten
through
trying
to
document
a
lot
of
the
the
differential
operators
and
stuff
right
now,
it's
more
about
finalizing
the
the
properties
and
then
the
methods
for
all
of
the
mesh
classes.
So
I
think
I'm
making
pretty
good
progress
on
that.
I
want
to
be
done
it
by
the
end
of
the
week
and
then
once
I've
gone
through
that
I'll
start
thinking
of
of
documenting
the
operators,
because
we
should
come
up
with
a
plan
of
how
much
math
we
want
to
put
in
those
doc
strings.
F
I
would
like
to
have
an
actual
definition
of
of
how
we
approximate
the
quantities
we
don't
have
to
go
through
like
details
of
mesh
construction,
but
it
would
be
nice
to
know
like
this
is
what
our
definition
of
a
kind
of
a
numerical
derivative
is,
and
you
know
how
what
order
accurate
it
is.
I
think
we
could
put
something
in
there.
That
is
sufficient.
A
Yeah,
I
think
the
level
of
math
I
was
like
trying
so
when
I
was
writing
up
the
doc
strings
for,
like
the
boundary
condition
stuff.
I
at
least
I
wouldn't
didn't,
really
give
it
too
much
thought,
but
what
I
put
in
there
was
just
kind
of
like
this
is
the
operation.
This
represents
like
not
too
much
about
how
like
order.
It
was
just
like
okay,
it's
this
is
the
numerical
form
of
this
part
of
it,
like
of.
G
F
Integral
or
something
yeah
yeah,
I
think
we're
on
the
same
page
for
what
level
we
want
it
to
to
be.
At
I
mean
where
the
the
thing
is.
We
want
them
to
know
what
it
is
and
be
able
to
use
it,
but
we
already
took
the
time
of
constructing
it.
So
we
don't
really
need
to
know
nitty
gritty
details
like
I'm
yeah.
F
But
yeah
it's
going
well,
it's
going,
but
yeah.
I
want
to
be
through
the
the
properties
and
and
methods
for
all
of
those
classes
pretty
quickly
and
yeah.
I
need
to
make
a
a
pr
just
so
we
can
keep
track
and
maybe
have
better
feedback
on
that.
So
I'll
probably
do
that
when
I'm
done
this
this
stage
of
of
documenting,
but
that's
basically
it.
H
John
how's,
the
mt
stuff,
going
it's
finally
moving
again
yeah.
I
found
that
I
had
two
tests
that
were
failing
and
I
figured
out
the
one
and
it
was
too
much
ram.
I
guess
what
we've
changed
in
the
mt
side
on
the
receivers
and
storing
matrices
there
yeah
it's
blowing
it
up
so
yeah.
I
think
it
was
like
30
or
so
frequencies
were
being
made
in
this
one
test,
yeah.
H
It
was
going
up
to
like
20
gigs
or
something
like
that,
and
I
guess
the
virtual
machine
couldn't
do
it
so
finally
figured
that
out.
I
just
pulled
back
to
about
three
frequencies
and
yeah
everything
runs
on
that
test.
Now
I
just
got
yeah
the
seismic
one.
That's
still
popping
up
so
yeah.
Looking
at
that,
and
then
after
that,
I
can
finally
move
on
to
the
tiling
in
getting
that
one
brought
in
and
all
the
test
passing
so
one
step.
A
A
A
Dom
how
you
refer
to
like,
oh,
we
need
to
like
discard
the
mesh
after
we're
done
with
it
like
for
your
like
for
the
tiling
inversions
and
like
other
things,
but
it
happens
on
receivers.
It
happens
on
a
lot
of
things.
We
end
up.
Caching,
a
lot
of
quantities
that
really
blow
up
the
wrong
memory
usage.
H
Yeah,
I'm
going
to
start
taking
a
look
at
that
because
we're
we've,
we
finally
got
all
our
backlogged
resources,
so
we've
got
quite
a
bit
to
throw
the
tiling
on
to
and
yeah
with
the
mt
we're
getting
like.
We
put
it
in
production
like
we're
getting
20
iterations
in
12
hours
and
really
nice
models.
It's
yeah!
It's
working
really
well
now,
yeah.
H
E
It
is
probably
also
something
where
sog
and
I
can
give
some
feedback
soon,
because
if,
if
it's
working
the
integration
in
half,
then
we
might
run
inversions
with
10
million
cells.
That
is
possible
from
the
solver.
But
then
it
might
accumulate
things
on
the
simpac
side
in
the
inversion,
and
then
we
will
probably
quickly
find
out
and
have
to
track
down
where
it's
getting
stuck
so
yeah.
H
We're
at
with
like
a
half
a
million
cells,
nine
frequencies
about
a
dozen
stations
yeah
we're
pushing
like
400
gigs.
E
E
H
Direct
solvers,
they
just
just
eat
up
the
memory
cho
mod.
I
can't
even
get
it
to
run
on
big
problems
either
it
just
spits
out
an
error,
just
saying
it
too
big
and
I
thought
maybe
there's
like
you-
can
look
at
the
errors
and
how
they're
throwing
but
that's
on
the
c
side,
and
I
can't
quite
find
that
code
to
see
what's
triggering
that.
Unfortunately,.
H
But
yeah
I've
been
doing
a
lot
of
stuff
in
the
parallelization
yeah.
I
would
like
to
give
a
talk
in
the
near
future,
then
do
with
this
through
simpag
here
and
show
you
guys
what
we've
been
up
to
yeah.
Getting
some
pretty
good
speeds,
it's
awesome,
but
we're
using
mpi.
Unfortunately,
so
it's
not
as
not
windows
friendly
per
se.
D
H
H
B
Anyone
I
had
one
quick
agenda
item
to
add.
If
that's
all
right,
one
thing
and
I
think
joe,
we
talked
a
little
bit
about
this
yesterday,
but
with
new
releases
I
had
to
potentially
do
like
a
mail
out
or
something
like
that
to
reach
kind
of
a
broader
community
and
to
let
folks
know,
what's
changed,
especially
people
who
wouldn't
be
necessarily
in
in
slack
but
are
perhaps
using
simpeg
and
not
like
of
the
release
notes.
It
could
have
a
link
to
the
release,
notes,
but
more
sort
of
with,
like
the
quick
highlights.
B
Of
what
what's
changed
and
like
what
the
user
would
sort
of
see
is
as
impacts
so
speed,
ups
new
functionality,
these
sorts
of
things
that
are
just
kind
of
like
the
quick
hey.
This
is
what
we've
achieved
and
what
we've
been
up
to,
and
also
to
give
credit
to
the
folks
who
have
been
you
know,
making
contributions
to
to
those
releases.
B
We
have
a
mailchimp
account
that
we
haven't
used
in
quite
a
while.
That's
one
option.
I
don't
know
folks
well
a
what
folks
think
of
this
idea
and
b
yeah.
If
you
have
any
suggestions
on
how
you
might
like
to
see
that
done
so
that
it
does
reach
a
wider,
a
wider
group,
it
would
be
great
if
we
could
get
some
folks
who
are
using
simpeg
regularly
in
in
industry
to
subscribe
to
a
list
like
this,
so
that
they
could
just
be
aware
of.
What's
what's
changing.
G
How
does
the
mailchimp
work
and
yeah,
how
do
we
keep
track
of
who's
actually
using
any
of
the
sempeg
routines.
B
So
we
can't
really
keep
track
of
who's
using
simpeg
unless
they're
asking
us
questions,
but
what
so?
What
mailchimp
does
is
you
can
it's
an
opt-in
process,
and
so
what
we
could
do
is
sort
of
like
send
a
first
email
to
a
bunch
of
people
like
a
personal
email
to
a
bunch
of
people
that
we
think
might
be
interested
in
subscribing
and
then
they
can
subscribe
and
then,
basically,
what
it
is,
is
we
create,
like
a
little
web
form,
so
they're
kind
of
like
nicely
formatted
emails?
B
Basically,
so
you
could
create
it
in
the
mailchimp
like
template
place
and
then,
when
you
hit
send
you
can
pick
which
list
it
goes
out
to
which
list
of
your
subscribers
kind
of
based
on
their
their
interests
that
they've
indicated,
but
the
simplest
is
we
just
email
it
to
everybody?
Who's
who's
opted
into
this
list.
B
Yeah,
that's
that's
a
good
idea.
We
can
make
maybe
a
little
bit
of
a
list
of
where
we
want
it
to
go.
I
mean
I
don't
want
to
turn
this
into
something
that
like
is
kind
of
a
huge
lift
after
a
release,
but
if
there's
you
know
twitter,
two
slack
channels
that
we
hit
and
sort
of
a
mail
out
that
that
seems
feasible
and
not
not
too
much
work.
B
Okay,
well
maybe
what
I
can
do
is
I'll
draft
up
like
just
a
little
try
and
write
this
down
and
actually
put
it
in
that
community,
repo
of
a
bit
of
just
a
like
simple
releases
guide,
and
we
can
kind
of
iterate
on
where
we
would
like
to
send
things
out
to
and
then
maybe
brainstorm
just
a
an
email
list
of
people.
We
can
maybe
send
a
first
personal
email
to
to
see
if
they'd
be
interested
in
subscribing.
D
B
I
mean
we
could
like
it
would
be
nice
to
perhaps
include
some
of
those,
and
we
certainly
could
sort
of
include
a
small
section
in
there
on
sort
of
like
new
science
or
papers,
or
things
like
that
that
have
been
published
that
use
simpeg
or
presentations
that
are
upcoming.
We
could.
We
could
definitely
use
it
for
for
something
a
bit
like
that.
A
B
We
we
can
in
some
ways
I
would
be
tempted
to
just
keep
it
pretty
simple
and
just
make
sure
that
we're
not
sending
these
out
too
often
that
people
feel
spammed.
But
it's
like
once
a
month.
E
E
You
could
say,
at
least
for
the
email,
maybe
only
to
minor
releases
and
major
releases,
but
not
bug
fixes
something
like
this.
I
don't
know
if
you
wanna.
E
I
don't
know
I
find
it
difficult.
I
I
had
the
example
with
curve
note
and
then
you
were
in
the
chat
of
simpek
and
it
is
lack
of
swung
and
in
the
curve
note
slack
and
you
turned
up
to
get
the
same
info
several
times
and
yeah.
That's
probably
the
same
with
release
note
if
you
put
them
on
twitter
and
on
the
simpex
slack
and
an
email
and
your
potential
in
all
of
them.
B
Okay,
well,
we
can
try
like
sketching
this
out
a
little
bit
and
then
maybe
at
the
next
meeting
kind
of
take
a
look
and
see
if
it
seems
like
a
reasonable
approach
and
perhaps
identify
somebody
who
might
take
kind
of
a
leap
role
in
just
making
sure
that
this
info
gets
out
at
the
right
points.
In
time.
D
Joe,
what
is
the
status
of
2dmt
code?
Is
that
sort
of
available
in
main
branch
or
still
somewhere
in
the
local
branch?
It's
it's
still.