►
From YouTube: SimPEG meeting April 18
Description
Seogi discusses the theory and implementation of the line-source formulation, and some of the challenges and opportunities.
B
A
C
A
A
So
if
a
bottom
here
there's
a
test
test
updating
multipliers.
So,
for
example,
if
you
try
and
give
it
a
string
or
a
list
of
strings
you'll
fail
because
we
don't
have.
A
B
A
A
B
A
B
A
B
A
A
B
C
C
C
C
B
A
Don't
all
those
wife
I
think
in
a
lot
of
these
white
Devils,
we
have
to
really
carry
with
them.
It
is
very
easy
to
sort
of
get
carried
away
and
Showtime
code
I
because
were
familiar
fitting.
It's
easy
for
us
to
get
you
but
otherwise
actually
create
our
dating
breath
right.
Oh.
C
C
B
A
A
A
B
C
Yeah
yeah
I'm,
open-minded
to
whatever
example
like
if
you
can
build
a
common
model
and
then
show
probably
exam.
D
B
A
B
C
C
B
Few
people
do
diamonds
in
vancouver's.
It
is
true
right
right
and
support
your
honor.
A
C
B
C
B
C
B
A
B
B
B
C
B
We
have
yeah
exactly
if
we
have
something
nice
we
could,
we
could
share
it
with
their
people
and
then
what
the
hell
I
forget,
university,
whatever
there's
more
people,
are
gonna,
be
interested
with
CM
cool.
C
A
A
So
it's
synthetic
/
geo
models
and
like
the
goal
there
is
to
have
some
of
these
models
where
we've
got
surfaces
and
like
physical
properties
assigned
to
rock
unit.
So
we
could
perhaps
start
by
putting
like
the
surfaces
and
even
just
the
maps
and
a
few
things
in
there
and
just
cuz
right
now,
there's
nothing
in
this.
So
we
can
sort
of
starts
with
define
what
it
should
look
like
in
terms
of
actually
storing
models
that
people
can
then
go
ahead
and
use.
B
A
A
C
Okay,
yeah
right
so.
A
C
C
C
C
Well,
like
higher
step
is
actually
energy.
Sorry,
but
like
your
electric
field
and
magnetic
field
have
m
and
P
plus
1
times,
and
that's
set
up
the
result
e.
So
we
can
rearrange
that
equation
in
this
form,
so
we
can
solve
for
it.
So
you
got
like
yet
end
step
and
it's
that
n,
minus
1,
so
n
minus
1
is
no,
so
you
can
move
that
to
a
right
hand,
side
and
at
the
end
of
the
day
and
here's
what's
important
is
like
that
we
compute
the
derivative
of
time
derivative
of
SC.
So
that's.
C
C
C
C
So
that's
what
that's
the
system,
what
we're
solving
we're
not
going
to
solve
this
full
system
by
a
inverse
of
right-hand
sidebar
but
like
this
is
sort
of
conceptual
thing
would
rebuild
and
then
and
the
four
problem.
It's
like
it's
a
for
substitution
right
because,
like
this
guy
is
a
block
block
matrix,
but
you
know
all
of
those
terms
you
can
solve
that.
But
after
that
is
actually
like
this
for
substitution
to
solve
this
system.
C
So
in
that
case
you
should
be
careful
like
so
depends
upon
what
your
model
is
and
then
tendency
on
your
initial
fields
at
zero
time.
You
need
to
be,
it
will
be
careful.
So
previously
we
assume
our
initial
condition
is
not
depend
on
our
model,
but
if
you
are
putting
a
galvanic
stores
or
grounder
source
now
your
initial
condition
is
a
DC
problem
or
our
problem.
C
A
C
Know
so
I'll
get
there.
Yes,
so
there
is
another
implementation
is
required,
is
like
okay,
how
do
you
compute
the
right-hand
side?
I
put
this
kids
yeah
and
the
important
thing
if
you
go?
If
you
go
back
to
hear
like
you
not
here
is
a
electrical
field
and
then
but
we're
not
actually
solving
electrical
field
for
DC
problem.
We
solve
the
electric
potential
and
then
take
the
gradient
to
get
the
electrical
field.
C
C
What
I
get
the
question
was
like
the
in
reality,
our
signal
is
like
its
periodic
function
so,
but
we're
assuming
like
11
period
of
time.
So
that's
not
quite
true,
so
there's
a
way
that
we
could
like
implement
in
that
way.
So
we
could
assume
the
period
periodic
function
and
then
solve
a
periodic
pop
up
your
attic
boundary
condition.
Then
you
could
get
the
periodic
response
clear
there
might
be.
There
is
a
some
there's,
a
way,
I,
not
exactly
sure,
but
that's
probably
something
like
this.
C
So
there's
yeah,
there's
some
practical
consideration
that
so
now
jmac
is
like.
We
need
to
compute
the
derivative
of
my
our
fields
at
ends,
time
channel
regard
to
a
conductivity
or
whatever
inversion
model,
and
at
the
end
of
day
you
could.
You
could
get
this
form
so
basically,
it's
the
same
same
way
to
solve
the
system.
So
we
need
to
solve
this
forward
problem
or
substitution.
We
could
we
could
just
for
substitute
this
quam
matrixes
to
solve
this
sensitivity
function
and
then
you
do
time
stepping
so
here
that's
a
reason.
C
C
Yes
and
then
like
when
we
are
solving
the
adjoining
system.
It's
similar,
but
not
like
right
so
like
now,
not
not
like
your
first
solve,
is
that
here
too
backward
substitution.
X
is
starting
from
that
look
at
the
latest
time
and
then
backward
propagate
to
a
first
time,
but
still
like.
We
need
to
be
some
like
consideration
on.
Do
this
first
time
channel
because
of
different
size
and
different
okay
yeah.
So
that's
how
it
works
and
yeah
there's
a
little
bit
of
tails.
C
C
So
great
figured,
but
the
geometry
is
like
so
you
got
say.
For
instance,
this
is
a
synthetic
model
and
then
my
wire
path
is
like
goes
like
this,
for
instance,
like
that's
a
horseshoe,
so
how
you
set
that
the
wire
pad
is
like
this,
so
you
need
for
coins
like
to
define
that
horseshoe,
and
then
you
pass
that
point.
So
you
can
stack
it
like
at
source,
location,
X,
Y,
Z
location,
then
here's
a
source
that
line
current,
that's
how
you
build,
and
then
you
need
to
pass
the
white
ones
here.
C
I'm
using
step
off,
wait
one!
So
now
this
line
current
function
is
a
smart
enough
to
integrate
over
the
path.
So
I'll
show
you
how
that
looks
like
okay,
so
here's
a
line,
current
function
and
then
I
have
a
source
type
codes
dog
at
it
because
like
this,
this
this
function
because
they
have
a
multiple
years.
So
if,
if
the
point
is
not
closed,
then
that's
an
organic
source,
but
the
if
the
point
is
that
you're
completely
continuous
and
enclose
it
could
get
inductive
source.
So
that
was
a
reason
why
men
in
that
pace?
A
C
C
I
think
I
like
to
galvanic,
I'm
not
sure
ground
is
yeah
well,
yeah
Kelly
guns
could
work
whether
people
use
like
sometimes
ground
loop,
where
or
maybe
they
may
not
use
crumbed
its
route,
but
so
so
there's
a
difference,
grounded
and
grabbed
from
big
and
ground
0.
As.
C
C
I'm,
open-minded,
like
people,
use
rounded
I'm,
okay
with
that,
but
but
that's
anyway,
I
read
some
distinction
of
the
source
type
here
to
pretend
commendation
and
so
I
have
a
like
Vincent
get
right
inside
and
then
here's
what
I'm
on
computing
the
right
hand
side.
So,
if
you
think
about
like
a
and
now
it's
a
divergence
of
j
s
at
your
right
hand
side,
so
once
you
form
a
jay
s,
then
you
can
just
simply
take
the
gradient
transpose.
A
C
C
So
that's
actually
like
what,
if
holly
words-
and
here
I
need
to
compute
the
e
initial,
so
I
compute
the
so
I
can
get
the
solution,
DC
and
then
compute
that
take
the
nola
gradient
and
get
that
electrical
field.
And
then
here
is
that
derivative
because,
like
that
derivative
is
taken
care
of.
So
that's
that
I
haven't
implemented
a
magnetic
field
one.
So
if
you
want
to
compute
the
magnetic
field
on
the
offline,
so
you
need
to
take.
You
need
to
be
taken
care
of
the
magnetic
field,
no
matter
what
you
solve
be
or
be.
C
C
Okay
yeah,
so
at
least
I
put
a
errors.
If
you
try
that,
but
portunities
right
and
the
at
the
end
of
day,
what
like
what
we're
calling
it
as
e.
Yet
so
that's
how
it
looks
like
mej,
yes,
its
space,
be,
you
is
still
need
a
distinction
of
SC
plus
some
pine
function,
so
this
time
function
is
actually
provides
a
country.
So
in
it
like
it
like
this,
what
cuz
ik
you
call
MJS
that,
like
integration
of
jas
as
an
SE
yeah,
so
it
just
inside.
A
C
Just
calling
that
part
and
then
multiply
but
waveform
that
you
Val
fair
enough,
so
that's
that's
how
its
structured!
It's
only
works
for
ED
formulation,.
C
A
C
Once
you
call
they
get
initial
fields
that
calls
the
DC
problem
and
yeah
so
that
extra
computes,
the
initial
field
for
for
problem,
I
didn't
have
any
problem
I
had
to
change.
I
think
jay
bag
was
fine.
I
didn't
have
any
problem,
so
the
current
structure
actually
did
worked.
Well,
so
I
think
change
anything
on
gym
bag
and.
C
Poor
jakey
bag.
I
had
to
change
so
I
I,
don't
know,
I,
think
I,
yeah
I
didn't
want
to
break
it
to
him.
So
I
put
it
on
like
that.
Ed
formulation
so,
but
I
had
to
build
like
this
is
the
probably
salt
some
some
part
I
had
to
change,
but
I
think
I
didn't
change
in
a
general
problem
class
yeah,
another
class.
I
think
I
did
change
problem.
C
C
A
C
C
A
C
But
we
have
a
minus
one
step
actually,
which
is
an
initial
condition
yeah,
so
I
did
put
that
p
is
equal
to
minus
13.
Put
it
in
right
that
then
we
need
to
generalize
here,
like
I,
have
a
mu,
sigma
drove
and
that
types
of
things
yeah.
So
then,
like
you
need
to
be
taking
care
of
your
hg
formulation.
So
there
was
like
enough
complex
it
that
I
didn't
want
to
handle
it.
So
that
was
the
reason
why
I
put
it
in
here
as
a
separate
class
yeah.
A
C
Is
sort
of
the
reserve
like
it's
a
it's
a
fairly
simple
example
like
you
put
five-time
champ
yeah,
so
my
time
said
is:
please
live
only
a
16
campus
and
I
one
millisecond
26,
millisecond
yeah
and
then
I
can
invert
it.
It
probably
takes
several
hours,
so
it's
just
like
pretty
big
time
compared
to
TC
I.
Think
hurt,
like
my
TC
inversion.
The
same
case
probably
take
like
a
couple
minutes
so
yeah
its
intensive
good
timing,
but
I
think
they're.
C
Not
there
are
a
lot
of
improvements
could
be
made
like,
for
instance,
we
can
store
the
factorizations
yeah,
that's
better,
like
a
boost
up
the
walk
for
a
small
small
problem,
and
this
is
the
reserve
life.
So
yeah.
That's
actually
works
pretty
well,
so
here
I'm
using
a
3d
TV
to
generate
synthetic
data
and
then
inverting
using
synthetic
oak
mastery.
It's
not
in
Bush,
crying
I!
Think.
A
Into
the
right,
yes,.
C
A
C
Works
pretty
fun
yeah
and
then
there
could
be
multiple
use.
I'm
using
the
step-off
wait
for
now,
but
if
you're
using
general
waveform
like
all
the
problems,
you're
super
work
yeah.
So
you
could
do
what
be
like.
You
could
be
EV
formulation,
but
you
could
to
be
and
also
eat
in
that
case,
like
all
the
stress
activities,
and
it
should
be
like
it's
tested,
so
you
should
just
work
so.
A
C
A
Of
the
things
I
wanted
to
bring
up
with
the
waveforms
to
like,
you
can
go
in
and
add
waveforms
to
the
time
domain
class,
there's
only
like
a
v10
waveform,
you
don't
necessarily
have
to
give
it
like
the
rahway
form,
plus
the
interpolation
function,
handle
your
go
in
and
add
that
under
the
hood,
if
that's
something
that
would
be
valuable
right.
C
Yeah,
that's
something
that
we
could
through
probably
using
the
proper
reason
yeah.
What.
D
C
Because,
like
they're,
changing
yes
they're
changing
relatively
fast
yeah,
so
every
years
like
if
they
have
a
new
system
and
then
they
have
a
different
time
dates
and
also
different
survey,
use
a
different
town
gates
and
different
wait
long,
so
I
I
wasn't
sure.
What's
because
I
sort
of
dangerous,
then
yeah
cuz
I
people
like
because
I,
I
think
I
would
it
for
our
code.
I
think
I
would
like
their
kind
of
make
a
trade
off
like
we're.
C
Not
gonna,
make
it
really
easy
yeah,
but
just
need
to
think
about
what
the
waveform
ease
yeah
and
then
was
there
con
charities.
I
don't
know
I
may
be.
Some
people
may
just
give
bob
at
that
point.
I
think
that's
long
yeah,
but
the
usually
really
like
need
to
force
them
to
think
about
what
that
is,
and
then
I
try.
Okay,
I
need
to
discretize
this
I'm,
pretty
van
yeah
I
think
that's
a
valuable
spam.
Okay.
D
C
C
C
A
A
D
C
So
for
simulation
I
like
cyclamate
the
initial
condition
not
for
any
more
problems,
I'm
amar
has
not
been
implemented,
so
it
should
be
done
so
Jay
bag,
gucci
bag
is
done
and
to
do
is
like
we
need
to
solve
the
DB
formulations.
Lp
h,
HJ
formulation,
I
hasn't
so
it
stopped
working
for
a
step
off
function.
That's
he'll
be
up
to
dude
and
I
won't
like
if
yeah
I
don't
know
like.
C
If
someone
has
a
bigger
example
and
then
they're
having
our
time
like
I
might
because
it
can
make
a
little
bit
faster,
yeah
problem,
sorta,
factorizations
and
in
the
44
problem,
we're
storing
the
fields
but
for
problem
we
don't
quite
me
to
sort
of
fields.
You
know,
and
especially
for
you,
the
Impala
is
pretty
big
yeah.
We
got
that
yeah.
D
C
A
Need
some
of
them,
so
that
would
be
a
good
thing
to
to.
C
C
So
what
we're
getting
is
not
a
point
failure,
so
they
have
a
time
date
and
then
integrate
over
the
time
dates
in
both
like
any
any
any
types
of
the
m3
way.
That
does
it.
So
that's
sometimes
we
take
good
care
of
it.
I
can
you
that's
probably
important
for
our
warm
case,
especially
when
you're
DK's
really
fast.
C
So
receiver,
like
that,
the
line
integration
code
can
be
used
for
receiver
as
well,
so
we
can
build
up
receiver
and
the
source
when
when
source
is
inductive,
and
then
you
have
oh,
it's
like
what
whatever
water
path
you
could
post
like,
they
actually
use
that
line
card,
but
we
can
actually
use
the
vector
potential
to
integrate
over
that.
So
those
are
some
practical
codes
that
needs
to
be
done
to
be
used
for
actual
practical
cases.
Yeah,
so
I
think
I
would
focus
on,
like
probably
true
first,
which
is
easy.
C
A
D
C
So
yeah,
if
you
can
merge
this
cuz
I
have
a
colleague
I
met
in
Indonesia
yeah
room
I'm
near
hollering:
si
will
he
say
he
wants
to
use
yeah
but
like
yeah
I.
A
Too,
okay,
we
can
also
start
deploying
pre
versions,
so
we
can
deploy
like
beta
releases
and
things
like
that.
So
if
there
are
instances
like
that
that
come
off
that
you
want
to
install
a
a
free
version,
so
yeah
yeah
yeah,
we
can
do
that.
Yeah.
B
Inverting,
that's
good
well,.
C
B
A
B
C
A
A
Hearing
problems
is
sketchy
because
it's
unclear
like
what
you
should
be
sharing
in
between
them,
but
I
think
the
global
problem
via
kids
approach
and
then
as
big
like
specifically
say
that
we
want
to
share
sensitivities.
Do
you
have
an
example,
I.