1. QCD
mul(-‐jet
calcula(on
on
MadGraph
Yoshitaro
Takaesu
(Sokendai)
In
collabora7on
with
K.
Hagiwara
(KEK)
May
10,
2011
1
Y.Takaesu
-‐
PHENO
2011
2. What
is
MadGraph?
テキ
2
Automa7c
Amplitude
Generator
Input
Output
Model
+
process
Diagram
HELAS
Code
MG
May
10,
2011
2
Y.Takaesu
-‐
PHENO
2011
3. MG
is
very
useful
and
popular
May
10,
2011
Y.Takaesu
-‐
PHENO
2011
3
BUT
…
4. Current
Limits
of
MG
Generated
Codes
>
8MB
cannot
be
compiled.
8
MB
10
0
10
1
10
2
10
3
104
105
2 3 4 5 6
log10
FileSize(KB)
# of Final State Particles
gg > ng
ug > ug ng
uu > uu ng
gg → 5g
qg → q5g
qq → qq4g
May
10,
2011
4
Y.Takaesu
-‐
PHENO
2011
The
file
size
of
codes
for
QCD
processes
5. What
makes
the
codes
to
be
LARGE
?
May
10,
2011
Y.Takaesu
-‐
PHENO
2011
5
6. The
structure
of
the
code
May
10,
2011
6
Y.Takaesu
-‐
PHENO
2011
Color
Matrix
Dual
Amp.
M =
i
CiAi
Color
Factor
dual
amp.
ex) Tr[Ta
Tb
Tc
Td
]
The
size
of
Color
Matrix
and
Dual
Amps
get
Large
as
#
of
final
par7cles
grows.
color
|M|2
=
ij
Ai(
color
CiC∗
j )A∗
j
7. Overcome
the
limit
May
10,
2011
7
Y.Takaesu
-‐
PHENO
2011
Dual
Amp.
Color
Matrix
8. Overcome
the
limit
May
10,
2011
8
Y.Takaesu
-‐
PHENO
2011
Dual
Amp.
Remove
Color
Matrix
Calculate
it
on
the
fly
9. Overcome
the
limit
May
10,
2011
9
Y.Takaesu
-‐
PHENO
2011
Divide
the
subrou7ne
part
into
small
pieces
(1
Source
file
for
1
Dual
Amp)
Remove
Color
Matrix
Calculate
it
on
the
fly
10. Overcome
the
limit
May
10,
2011
10
Y.Takaesu
-‐
PHENO
2011
Divide
the
subrou7ne
part
into
small
pieces
(1
Source
file
for
1
Dual
Amp)
Remove
Color
Matrix
Calculate
it
on
the
fly
Compile
those
files
11. Making
the
new
Code
• Modified
MG
to
generate
small
codes
for
each
Dual
Amp.
• Implemented
the
subrou7ne
which
calculate
the
color
matrix
from
the
informa7on
of
Dual
Amps
• Evaluated
by
using
those
codes
and
the
subrou7ne
May
10,
2011
Y.Takaesu
-‐
PHENO
2011
11
color
|M|2
12. May
10,
2011
Y.Takaesu
-‐
PHENO
2011
12
Example:
gg
-‐
ng
Codes
can
be
compiled
and
evaluated.
Run7me
of
the
Code
Preliminary
0.01
0.1
1
10
100
1000
10000
100000
2 3 4 5 6 7
log10
RunTime(s)
# of Final State Particles
gg ng
gg → 5g
gg → 6g
gg → 7g (
OK
but
slow..)
13. Conclusion
•
Generated
Codes
by
MG
for
mul7-‐parton
QCD
process
cannot
be
compiled
for
5
final
states
processes.
•
Proposed
a
way
to
overcome
the
limita7on
by
dividing
and
reducing
a
code
into
small
files.
•
Calculated
mul7-‐gluon
processes
(
5g)
May
10,
2011
Y.
Takaesu
-‐
Recursive
rela7ons
in
MG
13
14. Future
works
May
10,
2011
Y.
Takaesu
-‐
Recursive
rela7ons
in
MG
14
• Apply
our
method
to
processes
with
quarks:
qg
q
ng,
qq
qq
ng
• Phase
space
integra7on
and
event
genera7on
• Speed
up
by
using
GPU
15. May
10,
2011
Y.
Takaesu
-‐
Recursive
rela7ons
in
MG
15