►
From YouTube: SimPEG meeting August 19, 2020
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Cool
thanks,
everyone
perfect,
who
wants
to
start
us
off
dieter.
Do
you
wanna
tell
us
about
your
vacation.
B
Yes,
I
was
on
vacation,
so
nothing
to
report
from
my
side
back
since
monday
and
since
browsing
taking
through
all
the
email,
backlog
and
issues
I
feel
like
I
need
holiday
again
so
but
yeah
just
getting
started.
It
was
nice,
it
was
sunny
and
no
computer
at
all
for
two
weeks.
So
that's
nice.
C
D
Expand
on
some
of
my
stuff,
so
yeah
mumps
I've
got
the
every.
I
got
mumps
and
troll
mod
hooked
up
as
pimet
or
no
the
not
pymat
solver,
as
in
the
solver
utils,
so
you
can
just
import
them.
The
thing
is,
though,
is
mumps
without
mpi?
Is
it's
really
nothing
special
yeah
having
an
mpi
implementation
would
be
a
little
bit.
Tough.
I've
been
trying
to
go
through
it,
but
it's
yeah.
It's
not
really
driving
very
well
with
sympeg,
but
the
troll
mod.
D
When
I
first
tested
it
out,
it
was
really
nice.
It
was
working.
Well,
I
didn't
get
to
do
it
on
the
full-scale
128
core
computer,
yet
to
test
the
scaling.
I
was
testing
it
on
my
workstation
here
and
I
was
playing
around
with
the
omp
threads
and
I
don't
know
what
I
did.
But
chilmod
is
just
not
not
responding.
Now
it's
nothing.
What
I
like
what
I
was
showing
you
there
or
what
I
was
talking
about
with
you
on
friday,
joe
yeah
yeah.
It
just
slowed
right
down.
D
So
then
I
went
to
my
linux
machine
installed.
It
and
things
were
things
are
working
really
good,
except
for
now.
I'm
getting
an
error,
sometimes
when
I
put
it
through
the
whole
inversion,
I'm
getting
a,
not
pause,
an
a
positive
matrices
or
something
like
that.
I'll
have
to
grab
the
thing
and
post
it,
but.
C
Yeah,
I
was
what
kind
of
size
system
were
you
working
around,
because
I
was
playing
around
with
it
too.
Oh,
what.
D
Actually
I
tried
both.
I
tried
a
mt
problem
and
a
dc
problem,
the
mt
one
I
thought.
Oh,
maybe
it's
not
working
or
it's
not
responding
well
to
to
complex
numbers,
but
then
I
had
it.
I
was
able
to
make
it
work
a
little
bit
with
the
dc
problem,
but
then,
when
I
put
it
into
the
whole
inversion
yeah
after,
like
the
first
iteration
it
just
yeah,
it
would
give
me
this
error
about.
C
D
I
was,
I
was
just
waiting
to
hopefully
get
onto
the
big
computer
here
soon,
so
I
could
just
see
how
it's
working
with
the
scaling
and
then
dive
into
that,
but
I
I
think
troll
mod
is
the
way
to
go
just
like
doing
like
the
normal
fun
like
jt,
vec
or
jvec,
or
anything
like
that.
It's
it's
fast!
It's
just
it's
just
not
quite
working
for
me
and
yeah.
C
When
I
was
when
I
was
playing
around
with
it
myself
over
the
last
couple
days,
I
was.
C
I
I
you
couldn't
just
use
like
the
you
know,
like
the
automatic
method
you
had
to
specify,
which
one
explicitly
see
if
there
is
like
a
supernodal
or
a
certificial,
simple
simplifical,
decomposition
yeah,
so
you
had
to
specify
that
type
and
then
yeah
I
mean
one
of
the
some
of
the
mt
stuff.
I
just
I
don't
think
the
matrices
works
out
for
what
we
need
to
do
with
it.
Okay,
but.
C
For
all
the
the
dc
and
the
time
domain
stuff,
it
is
very
fast
because
those
are
symmetric,
positive
definite
systems
and
if
some
of
the
unfortunately,
though,
like
I
was
talking
with
you
about,
we
were
the
way
that
the
boundary
conditions
are
handled
in
the
dc
case.
D
C
Are
kind
of
hacky
and
not
great
because
they
ruin
all
the
symmetric,
positive
definiteness
of
the
system
itself
yeah.
So
we
should
go
back
and
re
like
implement
them
a
little
nicer.
Okay,.
D
C
D
Okay,
yeah
I'll
dig
a
little
bit
more
around
today.
Hopefully,
I
think
by
tomorrow
I
should
be
able
to
test
the
scaling
too
yeah
yeah
yeah
I'll
get
back
to
you
on
that
joe
I
just
I
wanted
to
make.
C
D
I
got
the
fields
calculation
to
work.
A
couple
yeah
went
on
on
saturday.
There
I
think
yeah,
I
I
can
send
you
the
example
that
I
was
using.
If
you
want.
D
And
then
yeah
I've
been
running
some,
I
guess
2d
inversions
of
mt.
In
with
the
mt
code
yeah,
it
seems
to
be
working
out
great.
I
got
a
nepal
data
set
here.
I'm
just
kind
of
firming.
B
D
And
yeah
some
seems
to
be
working,
fine,
it's
cool
other
than
that
yeah.
Nothing
too
simple
about
it.
A
Fair
enough
thanks,
john
joe,
do
you
want
to
go
next
sure.
C
C
So,
obviously,
that
room
that
ruins
any
symmetric,
symmetric-ness
of
it
by
just
modifying
one
row,
so
you
can't
take
advantage
of
all
that
kind
of
stuff,
and
it
can
also
lead
to
some
weird
scaling
issues
when
you're
solving
it
and
decomposing
it,
the
yeah.
What
I
think
we
need
to
do
is
go
back
and
really
implement
them
properly
in
the
system.
C
One
of
the
things
that
we
could
do
is
just
for
when
it's
like
a
the
whole
one
of
their
own
neumann
boundary
conditions.
We
could
just
ask
to
play
around
with
just
asking
for
the
minimum
norm.
Solution
to
that
system.
C
But
I
what
I
looked
at
in
the
dc
case
was
that
we
would
need
to,
like,
I
said,
actually
go
in
and
modify
the
gradient
and
divergence
matrices.
The
face
averaging
matrices
things
like
that.
That
need
to
be
set
up
specifically
for
that
system,
and
everything
has
to
still
agree
with
dimension
wise,
but
it
shouldn't
be
too.
C
A
C
A
Yeah
revisiting
boundary
conditions
also
for
the
mt
problem
seems
quite
fitting
is
because
yeah,
I
think
overall,
like
it
would
be
nice
if
most
of
the
boundary
condition
work
is
handled
by
discretize
rather
than
peg,
which
is
the
case
like
everywhere
right.
C
C
G
F
Yeah
sure
so
I
mean
I
don't
have
anything
directly
simpek
that
I've
been
working
on,
but
I
did
just
finish
the
comprehensive
workflow
for
dcip,
so
that's
more
within
the
jif
tools
framework,
but
I
did
want
to
end
up
adapting
these
to
simpeg
and
eventually
putting
them
on
our
website.
F
So
I
think
for
anyone,
who's
not
really
familiar
with
this.
This
sort
of
was
a
way
of
of
training.
F
Our
sponsors
to
use
the
jif
tools
framework,
to
start
with
a
real
data
set
and
topography
and
go
through
all
the
steps
to
ultimately
get
a
physical
property
model
and
really
just
explain
the
approach
for
each
step.
You
know
how
do
we
design,
you
know
a
good
mesh?
What
are
the
sort
of
general
rules
for
that?
What
is
a
good?
Should
I
be
assigning?
You
know
floor
or
percent
uncertainties
for
for
my
inversion,
and
why
would
I
be?
F
What
would
what
would
be
the
impact
of
these
choices,
and
so
I
have
them
for
mt,
zed,
tem
and
now
dcip,
but
I
like
to
then
reproduce
these
results
with
simpeg
and
then
do
something
similar,
and
so
the
idea
would
be
if,
if
someone
was
actually
going
to
use
simpeg
to
go
invert,
real
data
sets
somebody
from
a
company
wanted
to
to
use
simpeg.
Then
they
would
get
this
comprehensive
workflow
to
to
carry
out
that
and
I
think
we'd
gain
some
traction
by
doing
it.
F
So
that's
sort
of
a
thought
for
something
I'd
like
to
add
in
the
future,
but
yeah.
Now
we
have
these.
We
have
these
three
comprehensive
workflows
and
really
it's
about
trying
to
reproduce
them
with
simpeg
and
it
would
validate
our
codes
against
the
the
ubc
gift
codes
as
well.
F
A
One
thought
there
devin,
I
don't
know
if
this
would
be
of
interest
to
you,
but
I
could
see
actually
that
sort
of
line
of
comparison
or
a
series
of
comparisons
making
a
really
nice
short
paper
that
just
if
we
go
through
like
a
series
of
different
examples,
some
synthetic
and
some
field
based
and
just
do
that
comparison
as
a
bit
of
a
benchmark.
I
could
see
that
being
a
really
valuable,
a
really
valuable
paper
to
be
able
to
point
to
and
say
like
yep
these.
A
These
codes
have
been
validated
and
here's
or
in
some
examples,
so
just
a
thought
to
maybe
put
on
your
radar.
F
The
challenge
for
for
me
has
been
to
try
to
provide
teaching
material
and
really
minimize
the
amount
of
one-on-one
time
with
people
who
are
getting
used
to
using
the
code
because
it
it
takes
up
a
lot
of
time
to
to
have
like
a
one-on-one
meeting
with
somebody
to
train
them,
and
usually
people
have
the
same
questions
and
so
far
I've
found
that
these
tutorials
have
done
a
really
good
job
of
getting
people
up
and
running
and
they
might
have
one
or
two
little
details,
but
the
the
major
portion
of
teaching
I
don't
have
to
do
hands-on
anymore,
so
yeah.
F
I
think
this
would
be
really
good
to
I
guess,
to
reproduce
with
simpeg
and
hopefully
put
on
our
website.
We
don't
have
to
go
and
run
the.
I
guess
we
could
provide
notebooks
and
then
maybe
you
know
not
have
them
actually
be
run
and
just
sort
of
print.
Some
of
the
final
results,
because
the
problems
are
a
little
bit
bigger,
but
yeah.
I've
just
been
thinking
about
doing
that.
F
Yeah,
that's
that's
it
for
me.
A
Excellent
thanks
steven
tebow.
E
Hey
nothing
really
new.
On
the
simply
side,
I
was
wondering
how
the
jupiter
day
went
for
you.
A
Oh
yeah,
that
was,
it
was
a
lot
of
fun.
We
I
put
together
an
example
actually
based
on
the
teaching
demo,
that
I
did
at
ubc
with
widgets
and
dashboards.
I
can
send
the
link
around.
It
was
a
really
cool.
It
was
a
really
fun
day
to
get
a
sense.
Well,
actually,
three
days,
I
attended
two
of
them
just
to
get
a
sense
of
what's
going
on
on
campus
with
folks
who
are
using
jupiter.
E
A
So
there's
a
there's,
all
the
or
most
of
the
materials
are
up
there,
but
it
was
nice
getting
an
overview
of
some
of
the
jupiter
book
stuff.
I
really
appreciated
that.
I
mean
I've
played
around
with
it
a
bit,
but
it
was
nice
getting
a
bit
of
a
demo
on
that
yeah
and
then
you
can
check
out.
I
did
a
demo
with
widgets
and
showed
off
voila
dashboards,
and
I
believe
that
it
was
all
recorded.
A
I
I
don't
think
they've
been
posted
yet,
but
I
think
that
they
were
planning
to
follow
up.
So
there
should
be
a
series
of
videos
that
come
out
related
to
that
nice.
E
So
I
guess
I
guess
next
week,
if
you're,
all
okay
with
that,
like
next
week
meeting,
I
will
be
happy
to
give
the
presentation
that
will
be
a
good,
a
good
deadline
for
me
to
get
the
presentation
ready
so
and
then
I
will
have
like
plenty
of
time
to
edit
if
necessary,.
G
D
Yeah
I
haven't
gotten
into
actually
talking
to
soggy,
even
about
doing
a
2d
mt
yeah,
it's
a
little
bit
a
little
bit
over
my
head
right
now,
so
I
just
it
seems
to
be
working
in
3d
and
yeah.
Just
went
with
it.
G
Okay,
that's
good
yeah
other
than
that.
I
guess
we're
starting
to
get
ready
for
seg
coming
up,
so
there's
a
number
of
talks
that
are
being
prepared.
So
I
think
that's
been
a
little
bit
the
time
consuming
part
here,
at
least
for
some
of
us.
A
That's
it
all
right,
thanks
doug
yeah
in
terms
of
updates
from
me,
I
need
to
stop
playing
with
the
time
domain
or
the
the
csem,
the
frequency
domain
model
and
just
send
you
some
results.
A
It's
one
of
these.
I
found
it
yeah
one
where
I've
spent
a
ton
of
time.
I've
learned
a
lot
about
meshing
and
things
like
that,
but
soggy
graciously
took
a
look
at
the
layer
model
and
provided
a
couple
suggestions.
So
actually
what
I'm
kind
of
thinking
to
do
and
I'd
appreciate
thoughts
on
this
is
provide
one
example
where
we
actually
run
it
on
the
same
mesh
as
what
you
ran
yours
on
dieter.
A
It's
very
big,
then
for
simpeg,
but
it
can
be
run.
It
takes
about
100
gigs
of
memory,
but
we
we
can
access
that.
So
it's
it's
huge,
they're,
reasonably
comparable
results,
but
sort
of
shows.
You
know
an
iterative
versus
or
the
multi-grid
solver
versus
a
direct
solver
which
might
be
kind
of
nice.
A
The
second
thing
I
was
thinking
to
do
is
actually
show
then
an
octree
like
submit
an
octree
with
them.
That
would
be
less
accurate,
so
the
accuracy
is
on
the
order
of
five
or
so
percent
for
the
five
to
ten
percent
for
the
layer
model,
but
runs
very
quickly,
because
it's
only
two
hundred
thousand
cells
and
for
the
block
model.
It's
actually
pretty
comparable,
because
as
soon
as
you
sort
of
introduce
3d
geometry,
you
know
all
the
codes.
We
are
we're
introducing
more
errors.
A
And
so,
even
though,
on
the
layered
model,
it
was
maybe
a
five
percent
error
with
the
block
model,
it's
a
little,
a
little
less.
So
so
that's
kind
of
what
I
was
thinking
for
that.
B
Okay
sounds
good
yeah,
I
guess
you
could
send
both
and
then
we
see,
probably
we
will
just
put
one
with
all
the
online
results
and
one
into
the
paper
or
something
like
or
mention
it
sure.
B
A
A
So,
in
this
case,
I
think
the
the
marine
csem
problem,
I
think,
is
actually
particularly
challenging
with
this
geometry.
A
So
the
changing
aspect
ratios,
I
think,
causes
issues
yeah
because
we
do
have
like
we're
coarsening
the
cells,
and
so
that
does
it's
not
going
to
be
quite
as
accurate,
and
I
noticed
with
with
your
example
like
the
padding,
the
the
mesh
you
designed,
the
padding
was
quite
gentle,
and
so
it
was
like
a
factor
of
1.1
something
very
gentle
compared
to
sort
of
the
factor
of
1.3
that
we.
A
Because
I
think
it
was
I'll
have
to
look
again
at
that,
but
I
thought
it
was
a
bit
more
gentle
than
that,
but
anyways
getting
those
aspect
ratios
make
make
a
bit
of
a
difference
so
yeah
and
then
also
getting
out
to
the
getting
out
to
the
boundaries,
because
I
just
picked
a
smaller
mesh.
Is
that
I'm
allowing
some
boundary
effects
in?
But
overall
I
don't
think
it's
it's
too
substantial.
B
In
the
end,
it's
it
is
quite
even
though
it's
a
layered
model,
it's
in
a
way
challenging
also
for
the
fe
codes,
because
it's
a
very
we
have
this
thin
shallow
conductor.
D
B
That
then,
the
fe
codes
they
cut
that
off
at
a
certain
limit.
That's
why
they
get
less
precision
further
away
yeah,
but
it's
a
one
we
can
compare
to
1d
code.
So
it's
interesting
to
say
the
marine
is
challenged,
because
my
experience
at
least
for
my
code
marine,
is
often
easier
because
you
dump
the
signal
before
it
hits
the
air.
A
A
For
sure
for
sure
what
are
actually
sort
of
like
target
uncertainty
values
when
you're
running
marine
cscm
inversion,
I
mean
because
we're
showing
examples
where
we're
fitting
or
we're
trying
to
match
models
to
one
percent.
But
I
imagine
that
that's
much
lower
than
what
we
would
actually
expect.
B
B
So,
besides
the
rotational
issue,
I
think
the
one
percent
error
is
kind
of
a
standard
standard
measure
of
goodness,
with
my
limited
experience,
let's
say
but
yeah
yeah,
I
like
yeah,
you
know
kerry
key
and
he
would
always
mention
the
one
percent
error
should
be
the
target.
A
B
A
Yeah
yeah
cool
yeah.
So
that's
I
guess
the
main
sim
peg
thing
I've
been
working
on
yeah,
then
other
than
that
sdg
getting
stuff
ready
september.
1St
is
coming
soon.
So
yes,
we'll
have
to
get
those
things.
Those
things
put
together,
yeah,
but
that's
that's
about
it.
Does
anyone
else?
Have
any
other
questions?
Discussion
topics
you'd
like
to
bring
up.
B
No,
I
would
have
one,
but
soak
is
not
around,
so
I
wait
for
next
time.
I
had
some
questions
about
a
guy
who
you
see
in
pi
mod
for
loop,
loop,
land,
so
more.
What
sophie
is
always
doing
and
he
used
the
filters,
and
I
know
that
he
had
sometimes
accuracy
issues,
particularly
for
very
high
resistive
subsurfaces,
which
you
sometimes
have,
and
I
never
have
so
I
never
come
across
those
issues.
So
I
looked
into
new
filters
and
I
think
we
could
get
much
shorter
and
faster
and
more
precise
filters.
B
C
B
B
B
B
G
So
you've
been
using
these
for
for
data
sets
or
for
filter
problems
for
digital
things,.
B
E
B
No,
the
filter
designing,
is
something
we
published
with
kerry,
key
and
ever
and
me
last
year.
That's
something
for
me
in
pi
mod,
so
it
has
functionality
to
design
a
filter
as
long
as
you
have
either
an
analytical
transform
pair
or
a
very
precise
numerical,
transform
parent,
and
you
can
use
the
empire
mod
to
design
digital
linear
filters.
B
So
that's
what
I
used
to
be.
We
managed
in
that
pipe.
We
managed
even
to
get
a
filter
for
a
ground,
penetrating
radar,
which
was
quite
a
challenge,
because
then
it's
more
wave
phenomena
and
not
not
just
diffusion,
but
it
even
worked
for
that
and
in
the
meantime
yeah
we
got
filters
for
many
different
applications.
A
One
okay:
well,
I
guess
we
can
wrap
up
the
meeting
portion
of
the
meeting
I'll
end
the
recording
and
we
can
move
on
to
whoever
wants
to
stick
around
for
tackling
some
issues
or
pull
requests
and
such.