Professional Documents
Culture Documents
...and several original papers that will be indicated during the lectures
Plan of the Lectures
• Introduction: Standard Monte Carlo
W (x j xi ) P( x j ) W ( xi x j ) P( xi )
no
H≤0? R uniform RN Є [0,1]
yes
no
General layout of a MC study:
Z N (S , M ) exp( KS hM )
S ,M
Notice that K ,h (S , M ) PK ,h (S , M )
S,M): histogram calculated by a on MC simulation at a given
point of the parameter space K,h.
Multi-histogram MC!
Convergence problems at second order
phase transitions: critical slowing down
2
N
2 1
( A) (A A )
N 1
Which can be written:
N N N
2 1 2 2 2
( A) (A A ) ( A 1A A )
N2 1 N2 1 1 2 1 1
2
N 1 N
2 1 2 2 2
( A) A A 1 A
N t
• Wolff
Cluster flipping algorithms
A. The Fortuin-Kasteleyn transformation: (Physica 57, 536 (1972), J.Phys
Soc. Japan Supp. 26A, 11 (1969))
Let’s define Hlm: the restricted hamiltonian where the sites l,m
no
yes
j nn of i K i j r U [0,1]
i= j ?
p 1 e
no all the
bonds draw a bond linking i p>r?
considered and j
?
yes no
yes
Identify clusters sort a rn
(Hoshen-Kopelman) Choose a cluster
R=1,2,...q
no
All the
clusters set i =R i in the
erase all the bonds cluster
yes done?
Notice that:
Principle: build only one cluster and flip it with a given probability
•These bonds represent the spins that are oriented in the same
direction as those of the cluster that haven’t been chosen for
the physical cluster definition of the cluster
P( ) g ( ) A( ) P( ) g ( ) A( )
with: E
P( ) e
m n A( ) (E E )
(1 Padd ) e
A( )
each one of the m broken bonds :+2J
each one of the n broken bonds : +2J
E=E -E =2J(m-n)
A( ) 2 J m n
e (1 Padd )
A( )
It’s enough to choose Padd =1-e-2 J to have an acceptance ratio
independent of the move!
cluster identification
By a single passage through the
lattice determination of the cluster sizes
label: 0 if molecule B
Notice:
stands for the name of the cluster, a given cluster may have
different labels {m1 m2 ...mp ms .....), ms is the smallest:
proper cluster label
In order to assign the label, look for the previously assigned
nn of each site (sweep the lattice in an ordered way) :
one nn found give to the new site the same label of that neighbour
m41=4, N(4)=1
m42=4, N(4)=2 N(m 21)=N(2)=1
m43=4, N(4)=3 N(2) N(2)+ N(7)+1=4
N(7)=-2
m61=6, N(6)=1 m 76 =2
m62=6, N(6)=2 N(2) N(2)+N(3)+1=9
m63=6, N(6)=3 N(3)=-2
m 77 =2 , N(2)=10
nothing to do with
cluster 3 as N(3)<0
Lattice size : 29x29
General idea:
H ( E) g ( E) exp( E)
If one considers a different equilibrium distribution in the
detailed balance :
(E E ' ) exp( E w( E )) (E' E ) exp( E ' w( E ' ))
If one chooses w( E ) E ln g E
Ei exp Ei ln g ( Ei )
i
Once we have g(E) : E (T )
exp Ei ln g ( Ei )
i
2
Cv (T ) k B E2 E 2
Averages of any other physical quantity A:
A( Ei ) exp Ei ln g ( Ei )
i
A(T )
exp Ei ln g ( Ei )
i
Eˆ NT 2C , N=Ld
Imposing the same level of error to all the sizes, there’s still a
dependence on L affecting Ê. Near the critical temperature
one can use the finite size scaling law for CT=Tc (L) L /2
The MES scales with L as: Eˆ
d L2
L2
Performing a WL simulation in the whole range for
a small system, allows to estimate the restricted
interval for larger systems.
Problems:
~
p xk p k ( xk )
k
Conventional update in replica k
Two types of updates: (detailed balance is verified)
pk ( ~
xk ) pk 1 ( ~
xk 1 ) pk ( xk 1 ) pk 1 ( xk )
r
pk ( xk ) pk 1 ( xk 1 ) pk ( xk ) pk 1 ( xk 1 )
1 for 0
W (X , m X ', n )
exp( ) for 0