►
From YouTube: SimPEG Meeting April 10
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
Hey
Devin
has
kindly
agreed
to
present
what
he
is
up
to
so
we'll
go
through
his
presentation
and
then
there's
just
a
couple
things
to
chat
about
with
respect
to
pull
requests
where
things
are
at
and
I
would
also
like
to
give
you
guys
a
bit
of
an
update
on
what
we're
up
to
with
starting
open
geophysics,
which
is
a
github
organization
for
some
more
general
projects
that
are
useful
beyond
just
simple,
so
pulling
out
some
of
the
pieces
that
we
can
share
across
different
packages
like
endo.
So
we
will
do
that
at
the
end.
B
Hear
me:
ok,
so
I
guess,
thanks
for
taking
the
time,
I
think
the
way
I'm
going
to
do.
This
is
maybe
just
a
really
brief
overview
of
the
background
theory,
because
it's
maybe
unfamiliar
for
everybody
and
then
I
was
just
going
to
go
through
the
two
examples
that
I
made
kind
of
shows
what
the
new
coding
can
do.
So
the
the
core,
the
RM
stuff,
is
actually
been
merged
to
master
and
right
now,
still
don't
get
any
examples
and
the
doc
strings
sort
of
get
the
supporting
documentation,
so
other
people
can
actually
use
it.
B
But
the
best
way
to
sort
of
think
about
this
at
first
is
if
you've
got
some
kind
of
inductive
time
domain
survey,
you're
going
to
excite
the
earth
and
you're
going
to
get
an
inductive
response,
but
you're
also
going
to
charge
the
earth
then
in
sort
of
a
magnetic
way
so
similar
or
analogous
to
IP.
You
get
this
effect
at
the
late
time
due
to
this
viscous
remnant
magnetization.
So
at
the
early
time
we
have
the
inductive
response.
A
B
C
B
Cool,
so
so,
at
the
early
time
you
tend
to
have
this
enough
response
and
then
at
the
late
time
you
have
the
vrm
response.
When
you
see
it,
and
the
theory
tells
us
and
experimentation
has
told
us
that
if
you
are
to
foreign
model
the
inductive
response
separately,
so
you
could
use
any
SIL
mesh
codes
or
analytic
solutions,
and
then
you
modeled
the
vrm
response
independently,
which
is
the
problem
that
I
worked
on.
If
you
add
them
together,
it
very
well
approximated
the
response
from
a
conductive
and
viscous
curve.
B
B
B
So
the
way
I
go
about
doing.
It
is
really
just
the
integral
equation
that
we
used
for
mag.
You
can
formulate
it
using
potential
fields
and
effectively
we
just
parameterised
the
off
time
magnetization.
So
that's
that's!
Pretty
much
the
name
of
the
game
here:
I
won't
spend
too
much
time
talking
about
mathematical
models
for
this,
but
the
one
that
seems
to
do
quite
well
is
this
log
uniform
distribution
of
time
relaxation
constants,
so
we've
got
a
mathematical
model
that
describes
the
magnetic
susceptibility.
It
depends
on
these
three
parameters:
Delta
Chi,
tau,
1
and
tau
2.
B
B
We
can
define
this
model
parameter.
So
if
you
remember
the
the
susceptibility
equation
that
I
had
depended
on
these
three
parameters,
define
a
model
parameter
is
kind
of
an
amalgamation
of
those.
You
can
parameterize
the
off
time,
magnetization
in
each
cell
by
basically
this
time
dependent
function,
the
inducing
field
in
each
cell
and
this
small
parameter
in
each
cell.
So
if
I
want
the
magnetization
at
any
time,
I
can
use
this
this
function
here,
and
this
depends
on
the
wave
form.
B
So
you
guys
might
have
some
questions
about
that
and
I
couldn't
talk
about
it
if
you're
interested,
but
ultimately
we're
able
to
formulate
a
problem.
That
is
a
linear
problem.
So
whether
or
not
I'm
interested
in
predicting
B
or
DB
DT
at
a
whole
bunch
of
times,
I
basically
take
a
matrix
vector
product
where
a
is
a
really
dense
matrix
and
holds
all
the
geometry
of
the
problem,
and
then
T
is
a
sparse,
very
tall
matrix.
That
gives
me
the
response
at
whatever
time.
I
want
so
I'm.
B
Not
storing
these
things
explicitly
because
T
is
very
tall
and
then
I
reproduce
scaled
a's
and
then
it
becomes
prohibitively
large
distorting
so
I
keep
them
separately
but
effectively.
This
is
the
Ford
problem,
so
the
first
example
that
I
had
was
that
I
wanted
to
Ford
Model
a
response
from
a
conductive,
magnetically,
viscous
earth,
and
so.
B
B
So
the
first
thing
that
you
need
to
do
is
specify
the
wave
form,
because
that's
going
to
determine
this,
this
characteristic,
decaying
behavior.
So
if
it's
say
a
step
off
and
you're.
Looking
for
the
B
response,
it's
gonna
decay
as
long
T.
If
it's
db/dt
it'll
be
1
over
T
and
if
there's
a
piecewise,
linear,
piecewise
linear
wave
form,
you
can
get
the
analytic
solution
for
that
convolution
as
well,
or
a
square
pulse.
So
you're
specifying
the
wave
form
you
can
even
put
in
your
own
decay
behaviors.
B
If
you
wanted
to
do
something
else,
so
you
really
do
need
to
specify
the
wave
form.
That's
the
first
thing,
and
then
we
go
through
specifying
the
transmitters
and
receivers
in
the
same
way.
So
you
make
the
receiver
list
and
you
add
that
to
source
lists
in
this
case
I
just
did
this
am
61
survey,
where
I
took
a
dipole
source
lying
about
half
a
meter
above
the
surface
of
the
earth
and
then.
C
B
Yeah
so
then
we
do
the
same
way.
We
attach
the
source
list
and
then
we
define
the
problem.
So
we
put
in
a
mesh
we
specify
which
cells
or
to
Lois
purpose.
And
then
the
extra
thing
that
we
don't
have
to
do
is
add
a
keyword
argument.
But
if
the
the
transmitter
is
really
close
to
the
cells,
then
the
primary
field
is
going
to
be
varying
a
lot
within
that
cell
spatially.
B
So
there's
actually
an
algorithm
to
refine
the
sensitivity
so
that
you're
you're
computing
the
sensitivity
of
the
data
with
respect
to
that
that
cell
properly
that
region.
So
it
doesn't
change
the
size
of
your
model
space,
it
doesn't
further
disparate
eyes.
It
really
waits
the
sensitivity,
so
that's
important
and
then
just
to
do
the
time
domain.
En
response.
I
just
took
the
analytic
solution
for
a
half
space
and
then
I
I
scaled,
it
to
add
some
lateral
variability
and
insert.
B
Getting
what
do
you
mean
like
a
scale
like
it?
So
I'm
just
used
like
a
Gaussian
distribution
again
on
the
surface
kind
of
a
2d
Gaussian
distribution.
So
you
can
even
see
the
the
imprint
of
that
in
the
in
the
bottom
left
corner.
Just
so
that
you
know
you,
you
don't
have
just
this
home.this
homogeneous
dv/dt
response.
It
just
seemed
like
a
bit
more
realistic,
I,
initially
modeled
it
during
using
a
steel
mesh
coat.
B
A
E
B
B
Late
times
we
see
PRM
and
that
if
the
the
vrm
response
is
big,
then
this
sort
of
crossover
happens
earlier
and
if
it's
a
smaller
response
it
happens
later
and
then
we
just
see
the
decays
at
I
started
the
the
anomaly
at
10
to
the
negative
5
10
to
the
negative
4
to
the
negative
three
seconds.
So
if
the
early
time
we
see
this
vrm
or
sorry
at
the
inductive
at
the
late
time,
we
see
VRM
and
then
in
this
middle
plot,
we'll
see
a
combination
of
the
features
of
both.
B
B
So
this
is
our
true
model
at
the
surface,
and
this
is
what
we
get
back
for:
equivalent
source
layer
and
you'll
see
them
in
a
little
bit
yeah.
So
if
the
black
line
is
our
observed
response
and
the
blue
line
is
the
predicted
vrm
at
all
the
times
from
that
model
we
recovered.
If
we
subtract
the
blue
from
the
black,
we
get
this
green
line
and
we
can
see
it
lives,
overtop
of
the
true
inductive
response,
so
eventually
you're
going
to
get
a
point
where
the
you
know.
B
The
noise
is
much
larger
than
a
signal,
but
we
can
see
that
now,
we've
really
isolated
that
that
inductive
response,
and
so
we
could
then
use
all
of
our
time
domain
e/m
codes
to
invert
that
and
get
conductive
structures.
And
if
we
looked
at
the
bottom,
we
see
the
total
observed
response,
so
there's
vr
m
and
e
m
in
it.
B
B
B
C
B
Is
just
the
surface
layer,
yeah
you're,
getting
a
one
in
one
layer
and
I'm,
just
tape
I'm
just
taking
again
I'm
defining
just
cells
at
the
surface,
so
I
did.
What
is
that
my
active?
My
active
set
is
like
all
of
the
cells
below
the
surface
and
above
minus
two
meters,
or
something
on
ice,
but
later
or.
E
A
E
B
E
B
E
B
Could
be
in
that
I
think
it
could
be.
In
summary,
some
depth
I'm,
not
an
expert
on
on
the
genesis
of
these
soils.
My
understanding
is
that
there
are
problem
when
they're
on
the
service
so
that
there
tends
to
be.
You
tend
to
be
able
to
go
into
a
region
and
go
no.
You
do
the
survey.
If
you
see
the
red
soils
and
then
you
know.
E
C
B
D
D
B
B
2
and
for
VR
rabbits
T
to
the
negative
1,
so
it's
very
interrupt
and
it
maintains
that
behavior.
So
if
you
have
enough
time
channels
and
you're
late
times,
aren't
super
noisy,
it's
actually
quite
easy
to
recognize.
But
let's
say
this
block
here
was
cut
off
at
say:
1
in
one
millisecond.
Yes,
you
wouldn't
be
able
to
do
all
this
stuff,
and
then
you
would
think
that
you
have
a
very
conductor.
E
B
B
0.1
Siemens
from
here
a
fairly
conducted
background.
So
if
it's
10
to
the
negative
3,
you
completely
obliterate
that
inductive
signaling
with
just
the
all
the
area,
we
kind
of
choose
the
background,
conductivity
and
vr
m
to
make
it
nice
pretty
easy
for
you
guys.
But
there
is
data
status,
especially
uxo
scenarios,
where
there
is
no
background.
Inductive
response,
it's
just
the
VR
app
you
just
get
that
so
it
is.
It
is
possible
to
have
those
those
data,
but.
E
E
F
B
A
So
you
guys
can
see:
that's
okay,
so
I
just
wanted
to
queue.
I
guess
maybe
actually,
let's
start
with
pull
requests
and
just
see
where
things
are
at
and
then
we'll
jump
over
to
open
and
cube
physics,
so
I
know
Devin's
working
on
strings,
so
this
is
in
progress
right.
This
is
not
ready
for
review.
No
I
just.
D
E
G
E
E
E
A
H
I
A
So,
on
my
end,
like
oh
I'd
like
to
get
the
dtft
spectra
like
Ian
and
then
I'll
update
that
for
the
e/m
sill
stuff
and
then
I'd
like
to
try
and
bring
in
both
the
3d
cell
mesh
and
this
pull
request,
so
that
will
include
the
change
where
the
cell
grad
stencils
are
no
longer
methods
their
property.
So
it's
just
brackets
versus
not
without
a
map
extra
regularization.
So
this
this
florists
will
take
care
of
that.
Even
though
it's
p.m.
it
takes
care
of
that
as
well.
I
mean
I've,
improved
some
stuff
on
the
fields.
A
A
H
H
A
When
there's
updates
we'll
go
through,
but
if
not
no
problem,
okay,
so
the
other
thing
that
I
want
to
talk
to
you
guys
about
and
get
people
to
jump
in
and
where
they're
interested
is
open
geophysics.
So
this
is
a
github
organization
that
we
have
started
to
start
collaborating
with
other
projects.
More
so
I
saw
you
couldn't
in
close
contact
with
Leo
as
well
as
the
theater
on
you
know
how
to
eat.
How
do
we
combine
efforts
on
some
of
these
things
as
well
as
how
do
we
start
writing
a
high-level
API
for
inversions?
A
A
So
a
lot
of
you
know
target
users
or
like
they're,
comfortable
working
with
something
like
scikit-learn,
where
you
know
I
format,
my
inputs
correctly
and
then
does
stuff
under
the
hood.
You
know
sing
to
me.
Crowder's
of
stuff
were
chosen
for
me
and
then
I
can
go
play
with
that.
So
if
some
of
that
conversation
is
happening
in
this
repository
right
now,
which
mania
may
not
get
renamed
just
to
immersion.
A
So
that's
what
that
is
matrix
utils
is
we're
playing
us
from
the
utilities.
Like
mashed
answer.
Those
kinds
of
things
make
the
back
end
agree
to
this
repo
because
we
use
them
everywhere
and,
like
you,
don't
necessarily
the
import
structure.
Right
now
is
weird,
but
they're
they're
useful
beyond
just
discretizing
something,
so
those
will
go
their
test.
I'm
crying.
This
is
something
I'm
kind
of
excited
about.
A
Do
all
the
proper
like
when
you
look
at
it
on
Travis
it'll
catch
like
the
error
that
would
have
been
displayed
in
Jupiter,
and
then
it
actually
prints
it
and
with
all
the
syntax
highlighting
and
everything
that
they
would
used.
So
it's
pretty
easy
to
tell
you
know
if
I
set
up
and
I'm
distributing
and
collection
of
notebooks
and
I
want
to
make
sure
that
my
requirement
is
file
is
well
I'm.
A
So
in
this
case,
I
have
forgotten
to
include
pieman
solar
in
live
requirement
as
well,
and
so
it's
like
very
easy
to
catch,
and
so
there's
one
example
that
you
guys
can
check
out
so
visit
a
little
bit
of
usage
there
and
then
I'm
using
it
for
the
airborne
iam
know
books
that
we've
distributed.
So
if
you
want
to
like,
have
a
look-
and
you
just
see-
you
know
one
use
case
of
it-
you
can
see
how
I
said
up
the
trogs
file
and
my
tests
and
all
of
that,
and
so
it
would.
A
A
That's
great
yeah,
so
it's
kind
exciting
and
some
of
the
Jupiter
folks
are
excited
about
it.
So
that's
cool,
so
I'll
chat
with
them
about
that
one
and
then
the
last
one
right
now
is
the
objective
function
side
of
things,
and
so
this
we're
gonna
try
and
pull
out
some
of
that
core
work
that
we've
done
on
the
objective
functions,
because
that
it
sounded
quite
useful
for
other
than
version
packages.
A
H
A
H
A
H
A
H
A
Was
a
brief
conversation
there's
mostly
just
enthusiasm,
that's
important!
We
just
gave
oh,
but
I
did
tell
him.
We
were
looking
at
you
know,
developing
this
higher
level,
API
and
I.
Think
that
that's
that's
a
really
good
thing
as
well.
Do
you
know
when
did
she
was
potentially
coming
about
seen
in
the
next
few
months.
C
H
H
A
A
A
Yes,
I
am
done
a
test
run,
that's
it
yeah
I
mean
what
you
can
do
and
this
I
might
try
and
do
for
publication.
This
is
including
some
assert
statements
so
like
assert
that
you
know
whatever
I
do
Fran
equals
some
downloaded
the
storage
results,
so
you
can
do
some
things
like
that,
but
actually
sort
of
turn
it
into
a
test,
but
right
now
all
it
does
is
make
sure
it
gets
from
a-z
I've.
H
A
Have
been
manually,
but
one
thing
that
can
help
with
that
is,
if
you
turn
on
Travis
cron
jobs.
So
if
you
go
to
Travis
and
we
can
look
at
this
impact
settings,
for
example,
we
run
a
test
every
month
on
master,
irrespective
of
if
it's
changed.
So
if
we
look
here
at
more
options,
settings
secured
cron
jobs,
so
on
master
it
runs
every
month
and
it's
you
know
you
can
configure
this
setting.
A
So
in
this
case
we
don't
run
it
if
there
was
a
test
within
a
couple
of
days,
but
on
things
like
research
repositories,
it's
nice
because
you
can
just
like
have
your
notebooks
that
are
associated
with
your
publication
and
then
just
turn
on
the
cron
job.
And
basically
you
know
you've
been
an
email
if
in
three
months
know
how
I'll
get
something
that
you
need
to
fix.
H
A
A
H
H
A
I
think
that's
basically
you
know
boom.
Do
you
want
well
anything
else
you
guys
want
to
share
for
today.