1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
|
# <nbformat>2</nbformat>
# <markdowncell>
# # Eigenvalue distribution of Gaussian orthogonal random matrices
# <markdowncell>
# The eigenvalues of random matrices obey certain statistical laws. Here we construct random matrices
# from the Gaussian Orthogonal Ensemble (GOE), find their eigenvalues and then investigate the nearest
# neighbor eigenvalue distribution $\rho(s)$.
# <codecell>
from rmtkernel import ensemble_diffs, normalize_diffs, GOE
import numpy as np
import ipyparallel as ipp
# <markdowncell>
# ## Wigner's nearest neighbor eigenvalue distribution
# <markdowncell>
# The Wigner distribution gives the theoretical result for the nearest neighbor eigenvalue distribution
# for the GOE:
#
# $$\rho(s) = \frac{\pi s}{2} \exp(-\pi s^2/4)$$
# <codecell>
def wigner_dist(s):
"""Returns (s, rho(s)) for the Wigner GOE distribution."""
return (np.pi*s/2.0) * np.exp(-np.pi*s**2/4.)
# <codecell>
def generate_wigner_data():
s = np.linspace(0.0,4.0,400)
rhos = wigner_dist(s)
return s, rhos
# <codecell>
s, rhos = generate_wigner_data()
# <codecell>
plot(s, rhos)
xlabel('Normalized level spacing s')
ylabel('Probability $\rho(s)$')
# <markdowncell>
# ## Serial calculation of nearest neighbor eigenvalue distribution
# <markdowncell>
# In this section we numerically construct and diagonalize a large number of GOE random matrices
# and compute the nerest neighbor eigenvalue distribution. This comptation is done on a single core.
# <codecell>
def serial_diffs(num, N):
"""Compute the nearest neighbor distribution for num NxX matrices."""
diffs = ensemble_diffs(num, N)
normalized_diffs = normalize_diffs(diffs)
return normalized_diffs
# <codecell>
serial_nmats = 1000
serial_matsize = 50
# <codecell>
%timeit -r1 -n1 serial_diffs(serial_nmats, serial_matsize)
# <codecell>
serial_diffs = serial_diffs(serial_nmats, serial_matsize)
# <markdowncell>
# The numerical computation agrees with the predictions of Wigner, but it would be nice to get more
# statistics. For that we will do a parallel computation.
# <codecell>
hist_data = hist(serial_diffs, bins=30, normed=True)
plot(s, rhos)
xlabel('Normalized level spacing s')
ylabel('Probability $P(s)$')
# <markdowncell>
# ## Parallel calculation of nearest neighbor eigenvalue distribution
# <markdowncell>
# Here we perform a parallel computation, where each process constructs and diagonalizes a subset of
# the overall set of random matrices.
# <codecell>
def parallel-diffs(rc, num, N):
nengines = len(rc.targets)
num_per_engine = num/nengines
print "Running with", num_per_engine, "per engine."
ar = rc.apply_async(ensemble_diffs, num_per_engine, N)
diffs = np.array(ar.get()).flatten()
normalized_diffs = normalize_diffs(diffs)
return normalized_diffs
# <codecell>
client = ipp.Client()
view = client[:]
view.run('rmtkernel.py')
view.block = False
# <codecell>
parallel-nmats = 40*serial_nmats
parallel-matsize = 50
# <codecell>
%timeit -r1 -n1 parallel-diffs(view, parallel-nmats, parallel-matsize)
# <codecell>
pdiffs = parallel-diffs(view, parallel-nmats, parallel-matsize)
# <markdowncell>
# Again, the agreement with the Wigner distribution is excellent, but now we have better
# statistics.
# <codecell>
hist_data = hist(pdiffs, bins=30, normed=True)
plot(s, rhos)
xlabel('Normalized level spacing s')
ylabel('Probability $P(s)$')
|
