1

Sairam Tangirala
PHYS 8601, set-6

Study of Reweighting Methods Using Single Histogram Method

Requirement: Here D ( Ei ) is density of states at energy E i and is

independent of T. We have
To employ single histogram reweighting method to study
the phase transition of the 2D Ising Lattice. Estimate the
D ( Ei )α Z ( β ∗ ) H i ( β ∗ ) exp( β ∗ Ei ) [2]
Specific Heat using reweighting method and compare it
with the result from direct importance sampling MC
simulation. Let P(E i , β ) denote the probability density of energy at
β ≠ β∗.
Introduction: Critical Slowing Down

One of the drawbacks of Monte Carlo simulations near 1

P( Ei , β ) = D( Ei )esp (− β .Ei ) [3]
phase transitions is the necessity to perform many MC Z (β )
runs to carefully study the specific heat, susceptibility
curves. The number of MC runs increase the cpu time In the above substitute for D ( Ei ) from Eq. (2), we get
needed for simulations with higher precisions. There are
techniques that can reduce the time consumption in such
Z (β ∗ )
cases. Histogram reweighting method is one of such P( Ei , β ) α
Z (β )
[(
H i ( β ∗ ). exp − β − β ∗ Ei ) ] [4]
methods. It can be used to predict the behavior of the
system at a temperature other than that at which the
simulation was performed. After imposing the normalization condition, we get

Consider a canonical ensemble of spin configurations

P( Ei , β ) =
[(
H i ( β ∗ ). exp − β − β ∗ Ei ) ] [5]
generated at temperature β ∗ . The idea behind histogram [(
∑ H j ( β ∗ ). exp − β − β ∗ E j) ]
technique is to calculate from this ensemble the j

macroscopic properties of the system at β ≠ β ∗ . This is

accomplished as follows. Thus from a single Monte Carlo run at β ∗ we can obtain
P(E i , β ) for various values of β ≠ β ∗ . Any macroscopic
Let H i (β ∗ ) denote the histogram of the energy obtained quantity as a function of β can now be obtained from
from a canonical ensemble of spin configurations the knowledge of P(E i , β ).
generated at β ∗ . In other words H i (β ∗ ) is the number
of spin configurations with energy in the interval ∆E
around E i . Let P(E i , β ∗ ) denote the probability density In the single histogram technique, the estimated P(E i , β )
that the system takes an energy E i . Now P(E i , β ∗ ) is is accurate only for β close to the reference
∗
proportional to H i (β ∗ ) . We have temperature factor β . By generating many histograms
that overlap we can widen the range of β . This is called
1 the multi histogram technique.
P( Ei , β ∗ ) = D ( Ei )esp (−β ∗ Ei ) [1]
Z (β ∗ )
 2

Conclusions: After obtaining the energy histogram curve the specific

heat curve was obtained using the reweighting procedure
Infinite temperature Simulations: explained before. Specific heat was calculated using the
fluctuations in energy. The energy was calculated using
Infinite temperature simulations are equivalent to doing the reweighting method described before ref. Eqn.[5].
simple random sampling MC. At infinite temperature the
ensemble contains all the possible random configurations After first reweighting, the specific heat curve shows a
with canonical distribution for energy. The energy of the peak at ‘T = 3.30’ and this temperature indicates the
configurations was used to calculate the histogram of “effective transition temperature”.
energy. I used production runs of 10 6 MC steps.

Energy Histogram

The following plot.[1] shows comparison of the energy

histograms obtained at T = ‘infinity’ and also at the
‘effective transition temperature ( T = 3.3)’ (obtained
after reweighting, explained later)

Plot. [2] Specific Heat at T = infinity

Simulation at ‘effective transition temperature’

( T = 3.30)

After getting an idea of the’ effective transition

Plot. [1] Energy Histogram temperature’ from the specific heat curve at infinite
temperature; Metropolis algorithm was used to obtain the
energy histogram at ‘T = 3.30’ and the plot[1] shows the
It can be seen that at ‘T = infinity’ the average energy of resulting energy histogram.
the configuration is ‘0’. This is expected because of the
symmetry of the head-tail Ising Spins and simulations Using the energy histogram from above, specific heat
being performed at ‘T = infinity’. was calculated using the fluctuations and is shown in the
following plot [3].
Also at a lower “effective transition temperature ( T =
3.3)” the histogram is shifted towards a lower energy
average suggesting that the mean-energy of the system
decreases as the temperature is reduced.

Specific Heat at T = Infinity

 3

~END~

After the first reweighting iteration the ’effective

transition temperature’ is observed to shift to a lower
value from T = 3.30 to T = 2.35 .

This new value of ’effective transition temperature’ (T =

2.35) can be used to perform further reweighting
iteration to obtain a more accurate estimate of the actual
transition temperature for 2D Ising model.