Professional Documents
Culture Documents
Hye oR
7a),
= iegtd
, = - kge og Ee
which is independent of Eo.
In other words, fluctuations produce Cy, and Eg has nothing to do with
fluctuations.
() Sree *
where energies ©, and e, are relative toe, , our choice for the
1 2 O°
zero of energy.
(elec) _
cy *
~ yep a1, 2,
= kyr { ty (g,e7e Be58
-Be Be
2
“5 (g,e,e 1s gene ¥ }
e1ec
(0) @* rans Verec’Snuc ‘ot “Wvin/ Tan * 8°
— pttrans ) + clelec) (nuc) , Q(rot) , (vib)
cy = Cy Cy rey Cy cy .Now
(elec) BB. ase
cy = kB Lee - ) + $ (sup?
a°anP(N)
=)
al N= = 3 GR) = 5 Bs
cosh(h) + ¥sinh“{h) + e
inne (n) + ok
Lim = lim sinh(h) + ( L* sosh@n)/ve inh (i) +2 a
he neo cosh(n) + Ysinn@(n) + eK
=0
5.22 (a)
Q(K,h,N) =
, (hea De Fag dinents (23) . ftv?) bef ytnen te #934 woe
{odd s,|
. 9h 2)leyat syleneK lay. 9¢8y] olay. pty Emr Cay. 2794),
= Ceckiny 37? cx" jn! na)
Thus,
_ 2) (0, #95) #neK(9, +85) . 2318485) > [h+K(s,+85)]
(h 72)(s\+83) +K'sis
= £(K,n) le 3]
For all the possible states of Sy and 83:
8, * 8371 e(Kshy eft K's g2tneK) , 72K a
81 = 7-33 ftc,nye"® eh a oh (2)
sp sgectt e(Kynje tt KR" . go 2K, geht 2K @)
By multiplying Eqs (1),(3) with Eq(2) twice quickly yields
£ (Kh) = 2feosh@(h) cosh (2K+h)eosh(-2ken)]'/"Division of Eq(1) by Eq(3) quickly gives
nal cosh(2k+h)
phen + 5 tn C cosn(=aKeny J
Division of Eq(1) times Eq(3) by Bq(2) twice easily leads to
K' = gnfcosh(2k+h )eosh(-2k+h)/oosh*(n)] 74.
(b) For Kh >O: [2K +h] >|-2K +Al, so that AGRE). > 1,
implying 4! > A. Further,
Ak. cosh(2K +h) cosh(—2K +h)
cosh" (h)
= ake + e7tkK— Ay(entK th 4 em *) <
(eh Fem) + e-F) “
Hence, K' < K.
So, the flows as written cause h to increase and K to decrease as
illustrated in the picture,(eo)
AnQ
zZl=
(nh, K) =
Fancet + eo ®)"
g(h,K=0)
&n(2cosh(h))
As written in part (a) and (b), flow from g(h,K = 0.01) will be
towards the h-axis, away from (h,k) = (1,1). We would like then to
invert the parameter flow, i.e., find nh = h(h',k') ;
K = K(n',K") .
Using these inverses, our errors will diminish with each iteration,
since then g(h,K) = 1/2 g(h',K') + 1/2 f(h',K')
One method to get reversed flows (from low K to high K) is to first
flow forward f, om (1,1) (to lower K values), and then retrace the
steps taken back up to (1,1):On the reverse trip, d-e-f, we can use our low K "accurate"
approximation for g, the error halving with each iteration.
aE. a
2.00285 1.00 1.00
2.41 1.94 0.469
2.84 2.81 2.23 x 10%
2.86 2.86 7.09 x 1076
2.86 2.86 7.5 x 10712
Generate starting at (h,K) = (1,1)
and moving down according to flows.
Start at bottom with
-12
g(h = 2.86, K= 7.5 x10 “) = (2.86, 0)
= &n[2cosh(2.86)) ,
and iterate upwards using
gine) = Seth", K+ dan rene).
The convergence to h = 2.86, K = 0 exhibits a stable fixed point,
and in fact any h = hg, K = 0 is a fixed point in flow.
The exact answer is (from Exercise 5.22)
gna(Ket, hel) 2 x + gn(eosh(h) + Ysinh@(h) + eK ) = 2.00285
N
which matches the RG value of 2.00285 above.
2h
An analytic inverse in closed form can be found by using e@" and
Wk
e"K as variables, which reduces the problem to the finding of the
roots of a 4th degree polynomial.5.24
Parts (a) and (b): The rules for construction of legal configurations
admit only three energy levels:
(4) == © if any partition has other than two atoms. Now, given that
there are two atoms per partition, we can have
% oO oO oO oO oO oO oO
aay BL Of gf tere 3 ig | 1B sane)
Having two atoms on the same side of a partition means that the
two atoms in the next cell must belong to the next partition.
The degeneracy is 2, and the energy is E- N-O-0.
(iii) The only other noninfinite energy states are those where each
partition has an atom on each side:
means that the next partition is either lo or ol ,
and the choice between these two possibilities for each partition
is unrelated to other partitions, e.g.,
0 0 0
slo lo o owt
These states have energy E = Ne, and the degeneracy is 2,
(ec) AS a result of the preceding analysis,
Q=24 28 evONe
(a) Ae 209, an(2 + (207BE)N
Ni” =pn =BN
Therefore,
via’ x { vk,T an2 + €, 2eF®> 1,
Nee N 0 » 20 Fe 4Hence
~Be
<1 ang >4
<> a> 38) 0, 208 4
(e) A phase transformation is associated with 8 = i an 2, or
€
kgin2
This problem illustrates how phase transitions can arise due to
competition between entropic and energetic effects.
P ° Pp
5.26 (a) Q= oe cea)P/? 5 expt J x u,u,,,Jf dB expl-se’/20 + (ey Su,de) .
7 \ oye Matar | a
The integral can be evaluated to be
e
S Pp a -8/20LE-o/B(eu ) ap?
exp{e/2al$ (en } uy r1°} [ae © et
ist eo
2 P Pp
= expt 88 ¢ Jue Lupa Ae: va :
2P’ i det
So,
ge Sees lim ((eayP/2 5 gtel
B Poo
(b) From the text, we know
¥
Bik uu
+ hu]
P/2 gi-1
lim y fel
ate) = oe
(ea)
ut
i
= 2 cosntp/a* + y7E*)
Hence,ae erBEr/20 (BE)
| is significantly greater than zero. This is an indication that
we are frozen into either the up state or the down state. One result of
this broken symmetry is that as our runs apparently show + 1 a8 well, we will observe o,, + 0asT+0.
1.00 =
oes Spontaneous
magnetization
0.60
\|
0.40
0.20
0.00
kaT/J
0.60
Correlations in the
0.48
Ising Magnet
036
0.24
<8; Sj > - and <8;s;> on the distance from the
biasing field is averaged over. Correlations shown in the graph are
$9842 - S0g80g» .
0.8
Correlations with spins in the interface
06
0.4
02
0.0
1 2 3 4 5
Neighbor distance
These representative results were obtained on a 40 x 20 lattice at
k,T/J = 1.2 using 100 passes to equilibrate and 10,000 passes to
mBasure correlations. The biasing field keeps spins 1-20 up and 21-40
down in the top row. Spins in columns 1, 20, 21 and 40 are considered
to be in the interface, and columns 10, 11, 30 and 31 were taken for
bulk measurements. The upper line shows correlations along the
interface, and the lower line shows the correlation of a spin at thg
intgrface into the bulk. Correlations between bulk spins are - 10°”
10~" and are not discernible on this graph.
r6.12 A straightforward algorithm is to choose a large enough, or
equivalently, to scale the range of the uniform random numbers [0,1] to
(-Xnax+ XmaxJ SO that most of the Gaussian will fit in this range. Then
trial x's will be accepted with a probability = p(x) , normalized so
that for all x, prob(x) ¢ 1. The trouble is that the fraction of
accepted trial x's tends to be small. This fraction is just that of the
area under the curve to the rectangle:
1.0
0.0
-xX x
max max
In other words, if we pick x uniformly from the interval, we will then
reject most choices. This inefficient method does not exploit any form
of importance sampling. There are a couple of possibilities for
improvement.
ax?
One is to imagine that the distribution p(x) = va/a e' is due
to an energy B(x), where pE(x) = ax? , and to employ the usual
Metropolis algorithm with diffusive motion. Then trial samples will
tend to come from the regions of lower E(x) and higher probability. The
drawback is that one random number x;,, is correlated with the previous
one x, from which it made its move to x,,, = x, + 6x , with 6x
small. Nonetheless, the calculated moments should be correct.
Another method, equivalent to the "force bias" Monte Carlo of
problem 6.13, is to change + , to reflect the more likely regions of
vw
p(x), thereby increasing the A ,'s .
vw
62The first algorithm described above corresponds to m , = constant
xx
12
and A ,«e-aX " , as we said, that scheme is not very efficient. If
xx
instead we make a change of variables, such as x = f(y), with y picked
uniformly from [0,1] then the probability of picking x in [xgsxq + dx]
is
df”
vax «
KeKy
ay,
ax
'
where y = f-'(x), So if we want to achieve 100% acceptance
A, = 1, then our trials must be chosen witha , = Yasn e
xx xx
and thus
x
2 : 0 _ 2
- Jae ™* which implies f Neg) = Vat fo dx = ert (xy) ,
oO
i.e., f is the inverse of the error function, Note this scheme only
generates positive x; the sign should be assigned randomly.
In some problems, such a perfect solution is not possible. But as
long as A, increases and begins to approach unity, we have made good
progress. In the present context, any change of variables favoring
sampling smaller x is better. For example, let
x= e where y € [0,1] is uniformly picked, This implies6.13
So the x are picked with probability
Ty Sst and so we must accept with probability
* + ex)
2 2
(_+ ex) ~ax
Ay « 7 ‘t/a @
to preserve w, . The proportionalities are chosen so that m, <1 and
a, $1. Again only positive x are generated and the signs should be
x
randomly chosen. Note only one index is retained because any given
choice is completely independent of previous choices.
(a) The condition of detailed balance, w ,/w, = exp(-BsE ,),
w' “vly vy
with a6 ,=E ,~E , gives
wi VN
mA, ~ my Ay
{S = exp(-BsE ,), or A ,=—%*™ exp(-pae ,
= 3 ' w vw iy vw
viveviy vw
Since A , is a probability, A ,< 1
vw vw
* "BE, Tsliy “BOE ,
ip e WwW ¢1,weletA = e WW and
' vw Fe
vw
A , =1 3 otherwise, we can switch indices, v#v', L.e.,
viv
“BAE ,
e vv and A jad
vy
let A ,=aA, =1 So then
w viv
Py voo= vel
woyem ye
vw w' I-p, v' = v-1
This form, which is independent of v, is possible because of the
symmetry of the problem — each pair of adjacent levels isequivalent to any other pair. With detailed balance, designating
>E , v' = vtl , we therefore have,
~Bhw po.
2 = >
aL
Note: in the limits Ww>0, p+1/2,
and fw>e, po
° BCH” ME + BL )) enya
6.14 (a) Q=f agltre pple °
o-BE 2 BH
{age tr me
aang 1 39
= are eo
oCBE. DD) Q (SEL) Q
Evaluating the trace gives
oo 2
Q= f age FF 0 2 cosntala® + ute + B)*)!/71
BE aL i
= 2s) J dBe cosh(BE) , and so
3g op g BE /20 ap El elog.e
3G) - 2f ae sinh(BE) = 5 2B + Bo) (u/a)
= a:
= aru) J dbo PE 24 sinnGe) y (E+ app /E
from which the desired formula follows directly.
6.16 The modification to include windows is simple. Since we are basically
adding a potential that is infinite outside the window (i.e., not under
the umbrella) and zero otherwise, we need only check if we've steppedoutside the window, This is done at the time we also check
2 BAE 5 pnd . Moves outside the window are rejected. we can scan
through each window for the minimum in A(M). Once one is found for a
given T, we can assume for T' near T, that Moi, is near Mains 80 We
need only scan a few neighboring windows.
However, as T+ T, , it will be hard to find the minimum since a wide
range of M are equally probable. We find that the barrier drops faster
than the minimum moves in as illustrated in the schematic figure.
Notice that the criteria ofa vanishing bistability can be used
unambiguously to define a critical temperature for finite systems, of
course, such a definition leads to system size dependent T, which
becomes the true critical temperature only as N= Lx L> =. So, our
Main
2
several values of L. The L + @ behavior of Mgin/L° is the exact M(T),
results for the "phase diagram," (S") vs T is shown in the figure for
and we've plotted M(T) on the same graph for comparison.
1.0
N
=)
& 08 Magnetization for
—
2 Different Lattice Sizes
06
kg
Fig. These results were obtained with 5000 passes/window, and each window had
the width L®/20 except for the L = 40 case where we used L°/40 window
widths.As described above, we determined T, for each value of L by
extrapolating to Maj, = 0. The results of those extrapolations are
given in the table. We then extrapolated T, to the L + @ by using a
procedure known as finite size scaling. In particular, we assumed
b
(tT, - TCL)" «@ 1/L
where T, = T,(*) . We determined the exponent b from the slope of
[T, - Ty(L)] vs. nL. This procedure gave b= 1,0 +0. Then a plot
of T,(L) vs. 1/L gave
Ty = 2.26 + 0.02
as the 1/L + 0 intercept. This estimate can be compared with Onsager's
Ty = 2.269.
L Tg(L)
10 2.60
20 2.43
30 2.38
4o 2.34
A discussion of the finite size scaling method is given in Sec. 1.2 of
Binder and Stauffer's Chapter in Applications of the Monte Carlo Method
in Statistical Physics, ed. K. Binder (Springer-Verlag, N.Y., 1984).
Also see Part 2 of Gould and Tobochnik's Computer Simulation Methods
referred to at the beginning of this manual chapter.Chapter 7. Classical Fluids
7.10 The kinetic energy and pair potential parts of