• Non ci sono risultati.

Monte Carlo Tools for the LHC

N/A
N/A
Protected

Academic year: 2022

Condividi "Monte Carlo Tools for the LHC"

Copied!
58
0
0

Testo completo

(1)

Monte Carlo Tools for the LHC

Gennaro Corcella

CERN, Department of Physics Theory Division

(2)

Outline

First lecture:

Introduction to MC methods Parton shower algorithms

General-purpose event generators Second lecture:

Matrix-element corrections to parton showers MC@NLO and matrix-element generators

Comparisons and concluding remarks

(3)

Introduction to Monte Carlo methods

(A. Pellissetto, Lectures at the the 2nd Seminar of Theoretical Physics, Parma, 1992; S. Weinzierl, hep-ph/0006269, M.H.

Seymour, Lectures at CTEQ Summer School, 2000)

Monte Carlo algorithms are based on the generation of random numbers

A random number generator is a program which produces numbers which approach the properties of purely random numbers

Example: a sample of instable nuclei and a periodic counter which gives 0 if the number of decays is odd, 1 if the number of decays is even

Problem: storage of a huge amount of bits is not manageable

Practical random number generators try to approximate random number sets Example: with a computer able to store words with n bits, generate integer numbers x1, . . . xn, with 0 ≤ xn ≤ 2n, and xn+1 = f (xn)

‘Random’ numbers Un, 0 ≤ Un ≤ 1, can be obtained setting Un = xn/2n As xn+1 and xn are related, they cannot be random numbers!

However, properly choosing the function f , we can get a good approximation of random numbers, in the sense that they appear random to somebody who does not know the applied algorithm

(4)

Many random number generators follow the linear congruential method (LCG), by P. L’Ecuyer, Comp. Phys. Comm. 60 (1990) 329:

xn+1 = (axn + c) mod m 0 ≤ a < m ; 0 ≤ c < m

a is called multiplier, m modulus, c increment

Typically: c = 0 (multiplicative linear congruential generator), and m = 2w, with w available number of bits

The best approximation of a truly random number generator is achieved if xn has the maximum possible period, which is m

A necessary condition is that a and c are relatively prime to m

Preliminary applications of random numbers: generation of a probability distribution and Monte Carlo integration

(5)

Generation of probability distributions: inverse transform method

Problem: given a generator of uniform random numbers u ∈ [0, 1], generate the variable x according to the probability distribution f (x), which is known analitycally

Z +∞

−∞

f (x)dx = 1 ; Prob(x1 < x < x2) =

Z x2 x1

f (x)dx The cumulative distribution F (a) gives the probability that x ≤ a:

F (a) =

Z a

−∞ f (x)dx ; 0 ≤ F (a) ≤ 1

F (a) is a random variable, uniformly distributed in [0, 1], for any x in the domain of f (x). We can set it equal to the generated random number and obtain x inverting the function F (x):

u = F (x) ; x = F−1(u) Example: generate f (x) = 2/x2 for x ∈ [1, 2] , R 2

1 f (x)dx = 1

F (x) =

Z x 1

2

t2 dt = 2 − 2

x ; 2 − 2

x = u ⇒ x = 2

2 − u ; 1 ≤ x ≤ 2

(6)

Acceptance-rejection method

It is used if the function f (x) is known only numerically

Enclose f (x) in a function g(x) = Ch(x), with f (x) and h(x) normalized to one, and C > 1

Compare f (x) with ug(x), with u uniform random number ug(x) ≤ f(x): accept x ; ug(x) ≥ f(x): reject x

Better results if g(x) is fairly close to f (x) and C > 1 a ≤ x ≤ b; 0 ≤ y ≤ c

(7)

Monte Carlo integration:

I = Z b

a g(x)dx = (b − a)hg(x)i

The integral I is computed as the average value of a random variable

One generates u1,. . . uN random numbers in [0, 1], evaluates xi = a + ui(b − a) and g(xi)

The integral is approximated to the following sample mean:

I ' IN = (b − a) 1 N

N

X

i=1

g(xi) The variance reads:

σ2 = (b − a) Z b

a

[g(x)]2dx −

"

Z b a

g(x)dx

#2

Central limit theorem: IN is a statistic variable with mean I and variance σ/√ N The Monte Carlo estimate for the variance σN is obtained using the same sample x1, . . . xN :

σN2 = b − a N

N

X

i=1

[g(xi)]2 − IN2 ; I ' IN ± σN/√ N

(8)

Importance sampling: change of integration variables ([a, b] = [0, 1]) Z

dxf (x) =

Z f (x)

p(x)p(x)dx =

Z f (x)

p(x)dP (x) ; p(x) = ∂d

∂x1 . . . ∂xdP (x)

If p(x) ≥ 0 and R p(x)dx = 1, we can generate a sample of random numbers x1 . . . xN according to P (x) and get the following estimates:

IN = 1 N

N

X

i=1

f (xi)

p(xi) ; σ2  f p



= 1 N

N

X

i=1

 f (xi) p(xi)

2

− IN2

Example (M.H. Seymour):

R 1

0 dx cos π2x = 0.637 ± 0.307/√ N R 1

0 dx(1 − x2)cos(π2x)

1−x2 = Rρ2

ρ1cos(π2x(ρ))

1−x(ρ)2

= 0.637 ± 0.031/√ N

Better efficiency after change of variables

(9)

VEGAS: learn where the function is large and approximate it to a step function The integration region is divided into a rectangular grid; an integral is performed in each region to understand where the integrand is largest and approximate the optimal probability pbest = |f(x)|/R |f(x)|dx to a step function

The grid is optimized and frozen; each j-th integration gives the estimate:

Ij = 1 Nj

Nj

X

n=1

f (xn)

p(xn) ; σj2 = 1 Nj

Nj

X

n=1

 f (xn) p(xn)

2

− Ij2

The final result reads: I =

m

X

j=1

Nj σj2

−1

m

X

j=1

NjIj σj2

Example with f (x) = cos π2x

(10)

Multi-channel integration

If the integrand has several peaks in different regions, we need a variable transformation for each peak domain

Each channel will have probability pi(x), according to which we generate x:

Channel i is selected with probability αi, with Pm

i=1 αi = 1, p(x) = P

iαipi(x) The integral and its estimate read (Ni = αiN ):

I = Z

dxf (x) =

m

X

i=1

αi

Z f (x)

p(x)dPi(x) ; pi(x) = ∂Pi

∂x ; IN = 1 N

m

X

i=1 Ni

X

ni=1

f (xni) p(xni)

The error is given by:

σ2 = 1

N[W (α) − I2] ; W (α) =

m

X

i=1

αi

Z  f (x) p(x)

2

dPi(x)

(11)

Phase-space integration (ab → 1 . . . n)

Phase-space factor for n particles of momenta p1 . . . pn and masses m1 . . . mn:

n(P, p1, . . . pn) =

n

Y

i=1

d4pi

(2π)32Ei(2π)4δ4 P −

n

X

i=1

pi

!

Factorization: Q = Pj

k=1 pk, Q → p1 . . . pj, P → Qpj . . . pn Φn(P, p1, . . . pn) = 1

2πdQ2j(Q, p1, . . . , pj)dΦn−j+1(P, Q, pj+1, . . . , pn) ;

Sequential approach: n-body phase space as a sequence of 1 → 2 decays dΦn = 1

(2π)n−2dMn−12 . . . dM222(n) . . . dΦ2(2)

with qi = Pi

j=1 pj, Mi2 = qi2, (m1 + . . . mi)2 ≤ Mi2 ≤ (Mi+1 − mi+1)2 M22 = (p1 + p2)2, . . . Mn−12 = (p1 + . . . pn−1)2

(12)

In the rest frame of qi, the phase space for qi(Mi) → pi(mi)qi−1(Mi−1) reads:

2(qi, qi−1, pi) =

qλ(qi2, qi−12 , m2i)

(2π)2 8qi2i d cos θi

λ(x, y, z) = x2 + y2 + z2 − 2xy − 2yz − 2xz

Algorithm for phase space generation: (ui random numbers)

1) Rest frame of qi: qi = P = p1 + . . . pn, Mi = pqi2; ϕi = 2πu1, cos θi = u2; Mi−1 = (m1 + . . . mi−1) + u3(Mi − mi)

2) Set

|~pi0| =

qλ(Mi2, Mi−12 , m2i)

2Mi ; ~pi0 = |~pi0|(sin θisin ϕi, sin θi cos ϕi, cos θi)

p0i = “q

| ~pi0| + m2i, ~pi0

, q0i−1 = “q

| ~pi0| + Mi−12 , − ~pi0

3) Transform back to the original Lorentz system;

4) Set i → n − 1: if n ≥ 2 repeat all steps The weight reads:

w = 1

(2π)3n−422n−1Mn

n

Y

i=2

qλ(Mi2, Mi−12 , m2i) Mi

(13)

Monte Carlo techniques can be used to simulate multi-parton radiation in high-energy processes

















  

 

 

 

 

 

 

 

 









































 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

   

 

 

 

 

 

































 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  









 

 

 

 

  ! !!!

! !!!

! !!!

! !!!

! !!!

"" """

"" """

"" """

"" """

#### #

#### #

#### #

#### # $$$$ $

$$$$ $

$$$$ $

$$$$ $

%%% %%

%%% %%

%%% %%

%%% %%

& &&&&

& &&&&

& &&&&

& &&&&

''''

''''

''''

''''

( (((( (

( (((( (

( (((( (

)))) ))

)))) ))

)))) ))

**** *

**** *

**** *

**** *

+++ ++

+++ ++

+++ ++

+++ ++

, ,,,

, ,,,

, ,,,

, ,,,

----

----

----

---- . ..

. ..

. ..

. ..

. ..

///

///

///

///

0000

0000

0000

0000

0000

11 11

11 11

11 11

11 11 22 2

22 2

22 2

22 2

22 2

333

333

333

333

333

44

44

44

44

44

55

55

55

55

55

66 6

66 6

66 6

66 6

66 6

66 6

777

777

777

777

777 88 88

88 88

88 88

88 88

88 88

9 999

9 999

9 999

9 999:: :::

:: :::

:: :::

:: :::

:: :::

;;; ;;

;;; ;;

;;; ;;

;;; ;;

;;; ;; << <<

<< <<

<< <<

<< <<

<< <<

<< <<

= ==

= ==

= ==

= ==

= ==

= == > >>

> >>

> >>

> >>

> >>

> >>

???

???

???

???

???

???

@@@ @

@@@ @

@@@ @

@@@ @

AA AA

AA AA

AA AA

AA AA

Figure by Frank Krauss

Standard Monte Carlo event generators (HERWIG/PYTHIA):

Hard 2 → 2 subprocess: leading-order (LO) matrix element Parton showers in the soft or collinear approximation

Matrix-element corrections for hard and large-angle parton radiation

(14)

A simpler case: e+e → γ(q) → q(p1)¯q(p2)g(p3) (see also M.Grazzini’s lectures)

____^^^^

____^^^^- γ(q)

@@

@@@

I q(p¯ 2)





 

  g(p3)



q(p1)

____^^^^

____^^^^- γ(q) 

q(p1)

@I@@

 g(p3)

@I@@ q(p¯ 2)

d2σ

dx1dx2 = σ0CF αs

x21 + x22

(1 − x1)(1 − x2) ; xi = 2pi · q

q2 = 2Ei

√s ; x1 + x2 + x3 = 2

x1 → 1: g k ¯q ; x2 → 1: g k q ; x3 → 0: soft gluon radiation x21 + x22

(1 − x1)(1 − x2) = −2 + 1 + (1 − x3)2 x3

 1

1 − x1 + 1 1 − x2



Collinear approximation: θ = θ23 → 0 , x1 → 1

z = E3

E2 + E3 = x3

2 − x1 ' x3 ; θ2 ' 2(1 − cos θ) ' 4(1 − x1) x3(1 − x3)

P (z) = CF1 + (1 − z)2

z : Altarelli − Parisi splitting function x21 + x22

(1 − x1)(1 − x2) ' 1 CF

4

x3(1 − x3)

P (z)

θ2 , J = x3(1 − x3) 4

(15)

x21 + x22

(1 − x1)(1 − x2) ' 1 CF

4

x3(1 − x3)

P (z)

θ2 , J = x3(1 − x3) 4

Differential cross section in terms of z and θ:

d

2

σ = σ

0

α

S

2π P (z)dz dθ

2

θ

2

Universal in the collinear limit

- - @

@@ 1 − z

R



 z

- - @

@@ z

R



 1 − z

Pgq(z) = CF 1+(1−z)z 2 Pqq(z) = CF 1+z1−z2

  

@

@@ 1 − z

R z

  





 

  1 − z



 z

Pqg(z) = CF 1 + (1 − z)2

Pgg(z) = CAz4+1+(1−z)z(1−z) 4

(16)

Soft approximation: γ(q) → q(p1)¯q(p2)g(p3, ) ω = E3  E1,2 ; |~p3|  |~p1,2|

M = −iegSTau(p¯ 1)



γµ i

6 p3+ 6 p1 6 + 6  i

6 p2+ 6 p3γµ



v(p2)

= gSTaν

 pν1

p1 · p3 − pν2 p2 · p3



e¯u(p1µv(p2)

Eikonal factorization: universal in the soft limit

d2σ = σ0CFαS π

2dω ω

d cos θ

(1 − cos θ)(1 + cos θ) = σ0CFαS

2dω ω

 d cos θ

1 − cos θ + d cos θ

1 − cos(π − θ)



Collinear limit : d2σ = σ0 αS

2π P (z)dz dθ2

θ2 P (z)dz = CF 1 + (1 − z)2

z dz ' CF 2 ωdω Soft and/or collinear limit :

d2σ = σ0 αS

2π P (z)dz d cos θ 1 − cos θ

Starting point to simulate multiple radiation: need of ordering variable

(17)

In the collinear approximation, any Q2 ∝ θ2 is feasible to order multiple radiation

- θ

p, E 

 zE

@@

@@

@@ R

(1 − z)E

Gluon transverse momentum: kT2 = z2(1 − z)2E2θ2 Invariant mass: p2 = z(1 − z)E2θ2

Collinear limit: ln kT2 ∼ ln p2 ∼ ln θ22

θ2 = dkT2

kT2 = dp2 p2

Soft gluons can be emitted anywhere, at any angle

-







 

   

 



Angular ordering allows one to implement probabilistically multiple soft emissions

(18)

Angular ordering

__^^

__^^- @

@@

@@

I q(p¯ 2)



  

g(p3)



q(p1)

__^^

__^^- 

q(p1)

@@I @

  

g(p3)

@@

I@ q(p¯ 2)

__^^

__^^- @

@@

@@ I

 -



θ12 θ13

__^^

__^^-

 R@@

-  R@@

θ12

θ23

|M|2 ∼ W = ω2 2

 p1

p1 · p3 − p2 p2 · p3

2

= 1 − cos θ12

(1 − cos θ13)(1 − cos θ23) (soft limit)

W = W1 + W2 ; W1 = 1 2



W + 1

1 − cos θ13 − 1

1 − cos θ23



Z 0

13

2π W1 = 1

1 − cos θ13, if θ13 < θ12

= 0 if θ13 < θ12

Colour coherence: a parton radiates up to its colour partner

(19)

Implementation of angular ordering

dP =

dσ2σ

0

=

αS

P (z)dz

1−cos θd cos θ ____^^^^u θmax

@@ I@



θ < θmax





-

Need to evaluate probability of no branching at larger angles Analogy with nuclear decay:

dP = λdt dN = −N0λdt N (t) = N0 exp(−λt) = N0 exp[−R t

0 dP ] Probability of no decay in [0, t]: N (t)N

0

= exp h

− R

t

0

dP i

Probability of no branching in [θ, θmax]:

Smax, θ) = exp

"

−αS

Z θmax θ

d cos θ0 1 − cos θ0

Z zmax zmin

dzP (z)

#

Sudakov form factor

dP = αS

2π P (z)dz d cos θ

1 − cos θ ∆Smax, θ)

Unitarity: 1=R+V ∆S sums virtual and unresolved emissions

(20)

Angular-ordered parton showers

__^^

__^^- @

@@

@@

@@@ I



   

 

@@

@ R









__^^

__^^- @θ

@I@@



@-

@I@

@@

@ R

@@I@





- @

@@ R

θ1 θ2

θ20 θ01

 

Parton shower ⇒ colour flow ⇒ angular ordering:

θ1 < θ; θ2 < θ1; θ10 < θ; θ20 < θ10

dP1 = αS

2π P (z1)dz1 d cos θ1

1 − cos θ1S(θ, θ1)

dP2 = αS

2π P (z2)dz2 d cos θ2

1 − cos θ2S1, θ2)dP1

Iterating dP one construct the multiple-radiation algorithm

(21)

Evolution variable in general-purpose event generators

__^^

__^^u@

@@

p1(E1)

I



p(E) θ



 k(ω)

- p2(E2), z = ω/E

HERWIG (G.C., I.Knowles,G. Marchesini, S. Moretti, K. Odagiri, P. Richardson, M.H. Seymour, B.R. Webber) :

Q2 = E2(1 − cos θ) ' E2θ2/2 ; Qmax = √p · p1 ; E = Qmax ; θ < π/2 Soft approximation: angular ordering

HERWIG++ (S.Gieseke, A. Ribon, M. Seymour, P. Stevens, B. Webber): Q02 = Q2+max(kz22,p2)+z2(1−z)k2 2 (only e+e at the moment)

Angular ordering, better treatment of soft phase space and heavy quark masses PYTHIA (up to 6.2 version) (T. Sjostrand, L. Lonnblad, S. Mrenna, P. Skands): Q2 = p2

It includes angular ordering only by an additional veto (see CDF PRD 50 (1994) 5562)

PYTHIA 6.3: Q2 = kT2

ARIADNE (L. Lonnblad): Q2 = kT2

The Sudakov form factor will be: ∆S(Q2max, Q2) = ∆S(Q2max, Q20)/∆S(Q2, Q20)

(22)
(23)

Implementation of parton shower algorithm in a Monte Carlo program

Main step: given a generator of random numbers R and branching with evolution variable Q21, generate another branching at Q22

The Sudakov form factor gives the probability of evolution between two values of Q2 with no resolvable emission

S(Q21, Q22) = R

The variable z is to be distributed according to the splitting function P (z) The cumulative distribution will be ( : infrared cutoff)

F (z) =

Rz

 P (z)dz R 1−

 P (z)dz z will be obtained by solving the following equation:

Z z

 dzP (z) = R0

Z 1−



dzP (z)

(24)

Initial-state radiation

Hard-scattering quantities are fixed (e.g. mW for W production, xBj for Deep Inelastic Scattering): better efficiency with backward evolution

~ -

h

Q21, x1

@@

@@

@@

R Q22, x2 = zx1

__^^

__u -^^



 z = Eg/E1

Q21 < Q22 (ordering variable); x1 > x2

Step 1: evolution Q22 → Q21 at fixed x2 with no branching

S(Q22, Q21)fq(x2, Q21)

fq(x2, Q22) = ∆S(Q22, Q20) fq(x2, Q22)

fq(x2, Q21)

S(Q21, Q20) = R Step 2: generate z = x2/x1 at given Q21:

Z z



dz

z P (z)f x2

z , Q21

= R0

Z 1−



dz

z P (z)f x2

z , Q21

(25)

The algorithm is analogous to the final-state radiation, with ∆S → ∆S/f f /∆S satisfies the DGLAP evolution equation

(see Grazzini’s lectures and ‘QCD and Collider Physics’, pp.163–173 for proof) :

Q2

∂Q2

 f

S



= 1

S

Z dz z

αS

2π f (x/z, Q2)P (z)

Solving iteratively for an evolution Q21 → Q2: f (x, Q2)

S(Q21, Q2) = f (x, Q21)

S(Q21, Q21) + αS

Z Q21 Q2

dQ02 Q02

Z dz

z P (z) f (x/z, Q02)

S(Q21, Q02)

f (x, Q2) = ∆S(Q21, Q2)f (x, Q21) + αS

Z Q2 Q21

dQ02 Q02

S(Q21, Q2)

S(Q21, Q02)

Z dz

z P (z)f (x/z, Q02) Interpretation :

S(Q21, Q2)f (x, Q21): no branching in [Q21, Q2]

P (z) f (x/z, Q02) : branching at Q02; ∆S(Q21, Q2)/∆S(Q21, Q02) evolution Q02 → Q2

(26)

Colour coherence in the hard scattering (example: q(p1)¯q(p2) → q0(p3)¯q0(p4))



    

-

@@

@@

@

¯ I

q0 q0



@@

@@

@

R q @@@

@@ R

-





@@

@@@

R

Mandelstam variables: ˆs = (p1 + p2)2, ˆt = (p1 − p3)2, ˆu = (p2 − p3)2

Double-differential cross section (massless appoximation):

d2σˆ

dˆsdˆt = 4

9 α2Sˆt2 + ˆu2 ˆ

s3 δ(ˆs + ˆt + ˆu)

Generate ˆs, ˆt and ˆu according to d2σ/(dˆsdˆt)

In a given frame, determine components of p1, p2, p3 and p4 in terms of ˆs, ˆt, ˆu Boost back to laboratory frame

Subsequent emissions:

q and q0 radiate at θ < θ1; ¯q and ¯q0 at θ0 < θ2

~  -

 θ1

@@ I@@θ2

(27)

Branching algorithm for the final-state radiation (forward evolution)

__^^

__^^u@

@I@



q(p) θ



 g(p1)

- q(p2)

z =

EE1

dP =

αS

P (z)dz

dQQ22

S

(Q

2max

, Q

2

)

Initial-state radiation (backward evolution)

~ -

h

q(p)@@

@@@ R q(p2)

__^^

__u -^^



 g(p1) θ

z =

EE2

dP =

αS

P (z)dz

x/zx fbf(x/z,Q2)

a(x,Q2)

dQ2 Q2

S,a(Q2max,Q2)

S,b(Q2,Q20)

Scale of αS in parton showers: using the transverse momentum of the emitted parton allows one to resum a class of soft/collinear logarithms

(D. Amati et al. NPB (1980) 173 )

αS(kT2) = αS(Q2)

1 + αS(Q2)b0 ln(Q2/kT2) ' αS(Q2)

"

1 − αS(Q2)b0 lnQ2

kT2 + . . .

#

b0 = 33 − 12nf

12π (LO)

(28)

Parton showers and resummation: expansion of the Sudakov form factor

S(Q2max, Q2) = exp

"

−αS

Z Q2max Q2

dQ02 Q02

Z zmax zmin

dzP (z)

#

Example: all q → qg, soft and collinear limit (z → 0, Q2 → 0) P (z) = CF 1 + (1 − z)2

z ' 2CF

z zmin = Q0

Q zmax = 1 − zmin (HERWIG)

S ' exp



−αSCF

4π log2 Qmax Q20

 

1 + αS

4π log2 Q2

Q20 + α2SCF2

32π2 log4 Q2 Q20 . . .



dP ∝ αS π

dQ2 Q2

dz z



1 + αSCF

4π log2 Q2

Q20 + α2SCF2

32π2 log4 Q2

Q20 + . . .



Probability of one emission at Q2:

P1 ∝ Z

dP = αS

4π log2 Q2

Q20 + α2S

256π2 log4 Q2

Q20 + . . . Resummation of double logarithms ∼ αnSL2n

Soft or collinear limit: resummation of single logarithms ∼ αnSLn

Riferimenti

Documenti correlati

The plots display expression levels of miR-149-3p for paired tumor vs normal matched samples (log10 CPM).. After 48 hours, invasion assays were performed. Cells were plated under

Consequently, clinicians can expect their patients to be more attentive to some esthetic factors than to others.’’ However, no systematic review analyzed the perception of

Las galerías, en las que se ha hecho levantamiento de la existencia de las lesiones en el intradós de la bóveda, se encuentran en la planta baja del cuerpo principal y precisamente,

2 - Presence, abundance, size and sex ratio of the mud crab Rhithropanopeus harrisii, a newly introduced invasive species, have been assessed in six sampling stations along

To examine the rigidity dependence of the secondary cosmic rays in detail, the lithium to boron Li=B and beryllium to boron Be=B flux ratios were computed using the data in Tables

Ketorolac can induce clinically relevant renal alterations in elderly patients, whereas its short course is considered safe for young adults with normal preoperative renal

This map synthesized aspects of morphological evolution, sediment budget assessment, and areas of potential sediment recharge as described above, it reports river segments