►
From YouTube: SimPEG Meeting January 13th
Description
Weekly SimPEG Meeting
A
So,
first
of
all,
obviously
you
guys
can
tell
that
we're
using
a
different
service
to
do
our
note-taking.
Now
we've
been
we're
going
to
test
out
the
opportunity
here.
The
link
was
shared
in
the
slack
for
the
notes
for
no
document.
If
you
have
issues
signing
in
signing
up
like
that,
just
forward
them
on
and
we'll
get
them
to
the
developers.
A
Somehow
just
a
quick
thing,
we
talked
last
week
we're
going
to
be
doing
an
after
coding
session
today.
So
if
you
guys
want
to
stick
around
after
the
meeting,
we'll
just
go
through
and
code
away,
whatever
we
want
to
we'll
just
all
be
here
together
coding,
probably
going
through
some
pr's
doing
some
other
things
just
chatting
that'll
be
after
the
meeting
once
we're
done.
A
B
B
I've
created
a
good
test
example
for
fdem
development,
so
that's
both
the
rings
that
you
get
when
you
do
em
inversion
and
some
of
the
tiling
stuff
that
dom
has
been
working
on
and
working
on,
finalizing
and
publishing
these
code
validations.
So
I
compared
simpeg
with
analytic
solutions
and
other
coding
packages
and
I'd
like
to
make
that
available
to
people,
so
that's
sort
of
what
I've
been
working
on
and
I'll
give
a
quick
update
for
soggy
because
he
can't
be
here
today.
C
A
B
A
B
Yeah
yeah,
I
think,
there's
still
work
to
be
done
on
on
making
it
fit
with
our
our
class
structure
and
kind
of
ironing
out
those
details,
but
I
think
it's
a
good
time
to
bring
it
in
anyway.
Otherwise,
you're
just
fine-tuning
it
over
there
and
not
really
seeing
how
it
fits
with
with
the
package
it
wants
to
end
up
in
so
well,
there's
a
pr
for
it.
We
still
have
some
work
to
do.
A
I
guess
my
concern
is
like
there's,
probably
a
decent
amount
of
things
that
look,
or
at
least
things
that
looked
like
they
would
have
been
reproduced
by
that.
I
reproduce
that
come
from
like
geoana,
though
there's
a
bunch
of
analytic
functions
in
there,
that
should
probably
just
either
be
imported
from
joanna
or
just
not
necessarily
imported
at
all.
B
Yeah,
well
I
mean
one.
One
thing
that
I
want
to
go
through
is
take
that
pull
request,
make
a
list
of
remaining
items,
big
items
that
we
that
we
want
to
complete
and
then
actually
have
a
game
plan
for
it.
So
right
now
we
want
to
see
if
it's
it
would
actually
run,
and
it
seems
like
it's
doing
a
pretty
good
job,
except
for
the
vrm
problem.
B
F
Yeah
continuing
the
expanding
the
desk
simulation,
you
know,
testing
on
the
on
the
frequency
domain,
3d
frequency
domain
and
the
speed
up
is
quite
amazing.
You
know
it's
almost
like
a
10x
speed
up
over
the
current
implementation
and
really
the
the
major
speed
up
is
not
really.
The
tiling
right
now
is
is
just
to
be
able
to
use
j
during
the
inversion
instead
of
having
to
do
all
the
matrix
and
factorization
over
and
over
again
right
and
that
they're
just
that.
That
saves
us
a
ton
of
a
ton
of
time.
F
It's
exactly
the
same.
I
mean
I
haven't
done
formal
ones,
but
I
tested
the
two
vectors
right.
I
did
not
store
j
versus
doing
the
usual
way
and
then
the
two
vectors
are
identical
right.
There's,
no
reason
why
they
shouldn't
be
because
all
we're
doing
basically
is
one
jvec
and
then
sending
this
well,
not
the
not
the
vector
but
just
j
j
back
call
and
then
sending
it
to
zar
right.
So
it's
the
exact
same
exact,
same
mechanics,
we're
just
storing
it.
Instead
of
keeping
only
the
factorization.
F
No,
so
that's
that's
a
good
point
right
now.
I
only
I
only
did
tiling
in
frequency,
so
it's
basically
splitting
we
have
two.
In
this
example,
we
have
three
frequencies
right,
so
I'm
basically
doing
three
simulations
and
the
three
frequencies
are
run
in
parallel
on
the
same
mesh.
So
it's
not
even
an
issue
for
this
because
they
all
have.
F
And
which
is
a
nice
part
with
the
way
it's
implemented
is
that
we
can
split
spatially
or
in
frequency
as
well
or
both
at
the
same
time
right
so
yeah.
Once
we
get
to
the
tiling
spatial,
then
we'll
have
to
worry
about
this.
But
if
you
don't
tile
spatially,
then
it's
it's
the
same.
It's
the
exact
same
same
thing,
right,
yeah,
just
saving
time,
because
we
run
stuff
in
in
parallel.
D
D
H
H
H
H
G
G
C
E
It's
just
at
the
moment.
The
gradient
that
I
have
in
the
simulation
class
can
only
handle
electric
receivers
and
that
should
move
soon
to
handle
electric
and
magnetic
receivers
and
that's
kind
of
the
first
thing,
because
then
for
the
joint,
I
need
magnetic
sources
for
the
back
propagation
yeah
other
than
that
I
had
to.
I
think
I
mentioned
last
time
that
I
have
this
automatic
meshing
that
they
can
throw
any
model
at
it
and
survey,
and
then
it
does
the
gridding
internally
and
of
course
it
was
abused.
E
So
I
had
to
make
the
all
the
estimation
of
the
parameters
much
more
conservative,
which
leads
potentially
to
bigger
meshes.
But
if
yeah
it's
it's
the
danger
of
automatic
things
that
the
users
can
just
input,
something
that
they
didn't
anticipate
and
then
it
turned
out
that
the
computation
mesh
was
way
too
small
and
the
receivers
are
always
placed
at
the
edge
of
the
mesh
and
then
it
was
wrong
and
then
it
says
cmg3d
is
wrong
and
I
said
well
yeah
in
a
way.
Yes,
I
have
to
make
it
more
conservative.
I
E
E
So
how
it
does
now
it
if
you
give
it
points,
you
give
it
two
points.
It
looks
in
which
cell
these
are
and
then
it
doesn't
a
joint
interpolation.
It
propagates
this
point
to
the
edges
and
then,
if
you
do
an
arbitrary
shaped
source,
you
just
give
all
the
edges
everywhere.
There
is
a
kink.
You
give
these
x
y
set
points,
and
then
it
just
places
all
these
cells
that
are
affected
and
distributes
the
energy
to
the
edges
that
are
next
to
it.
E
If
you
make
a
perfect
loop
with
four
elements
or
more,
then
then
you
have
a
zero
moment
and
you
have
basically
a
magnetic
source
and
if
it's
just
somehow
shaped
it's
an
electric
source
yeah,
but
it
works,
it
should
work.
I
haven't
tested
extensively,
but
it
should
work
in
any
way.
Okay,
and
I
hope
it
should
also
work
for
magneto
to
lyrics.
Actually,
if
I
put
in
random
a
layer
of
random
sources
somewhere
high
up
in
the
sky,
it
should
work,
but
I
have
never
tried
it.
F
F
E
Yes
and
when
I
use
sim
peg
and
the
mg3
for
the
comparisons,
I
can
create
a
vector
with
emg
3d
and
put
it
into
simpac,
or
vice
versa,
that
they
are
the
same,
but
I
can
only
create
source
fields
for
tensor
meshes
because
that's
all
emg
3d
can
do
right,
but
then
I
don't
think
it's
it's.
I
mean
if
you
have
one
dipole,
if
you
can
create
a
finite
length,
dipole,
it's
just
an
a
matter
of
looping
over
all
the
segments
and
adding
them
up.
A
B
A
B
C
A
A
C
B
C
A
J
I
don't
have
tutu,
I
should
feel
like.
I
resonate
with
dieter's
statement
that
I
thought
january
would
be
calm.
That's
not
the
case,
but
I
did
put
some
of
sean
walker's
comments.
He
sent
a
thoughtful
email
after
watching
some
of
the
governance
discussion,
and
so
he
compiled
some
just
thoughtful
comments
and
I
asked
if
he
was
okay
with
me
posting
that
in
our
governance
dock,
so
I've
added
that
at
the
bottom.
So
folks
want
to
just
take
a
look.
J
A
As
far
as
me,
I've
been
working
more
on
some
implementing
the
boundary
condition,
matrices
that
are
in
some
of
the
theory,
sections
just
kind
of
getting
them
generalized
for
some
of
those
other
weak
forms
for
the
ones
that
are
not
just
yeah
working
on
those.
For
generally,
my
my
my
goal,
for
that
is
have
to
be
able
to
implement
the
robin
boundary
conditions
on
all
of
them.
So
it's
just
one
thing
that
we
have
to
call
for
you
know
any
mix
of
women
virtually
or
robin
it
can't
be
made
with
the
one
call.
A
I
F
No,
I
mean
the
other
than
the
obvious
that
you
know
if
you
carpet,
if
you
carpet
your
surface
with
more
you
know,
distribution
distributed
sources,
receiver,
you
get
more
sensitivities,
but
then
he
didn't
like
the
results
that
vcip3d
was
giving
him.
So
he
started
cranking
down
the
chi
factor
and
then
started
to
get
tons
of
garbage,
and
he
was
just
showing
slices
over
slices
of
like
models
that
are
totally
overfitting
the
data
without
giving
any
explanation
about
anything.