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
|
#include <stack>
#include <vector>
#include "pcg-cpp/include/pcg_random.hpp"
#include "randutils/randutils.hpp"
#include "blossom5-v2.05.src/PerfectMatching.h"
using Rng = randutils::random_generator<pcg32>;
using Real = long double;
class AztecDiamond {
public:
using Coordinate = std::array<int, 2>;
typedef struct Vertex {
unsigned index;
Coordinate coordinate;
} Vertex;
typedef struct Edge {
Vertex* tail;
Vertex* head;
Real weight;
std::stack<Real> weights;
Real probability = 0;
} Edge;
private:
std::tuple<Edge&, Edge&, Edge&, Edge&> face(unsigned i, unsigned j) {
unsigned x0 = n - i;
unsigned x = x0 + 2 * (j % i);
unsigned y = x0 + 2 * (j / i);
Edge& e1 = edges[2 * n * y + x];
Edge& e2 = edges[2 * n * y + x + 1];
Edge& e3 = edges[2 * n * (y + 1) + x];
Edge& e4 = edges[2 * n * (y + 1) + x + 1];
return {e1, e2, e3, e4};
}
public:
unsigned n;
std::vector<Vertex> vertices;
std::vector<Edge> edges;
AztecDiamond(int n) : n(n), vertices(2 * n * (n + 1)), edges(pow(2 * n, 2)) {
unsigned M = vertices.size() / 2;
for (int i = 0; i < M; i++) {
vertices[i].index = i;
vertices[M + i].index = M + i;
vertices[i].coordinate = {2 * (i % (n + 1)), 2 * (i / (n + 1)) + 1};
vertices[M + i].coordinate = {2 * (i % n) + 1, 2 * (i / n)};
}
for (unsigned i = 0; i < edges.size(); i++) {
edges[i].tail = &vertices[(1 + (i % (2 * n))) / 2 + (n + 1) * ((i / 4) / n)];
edges[i].head = &vertices[M + (i % (2 * n)) / 2 + n * (((i + 2 * n) / 4) / n)];
}
}
void setWeights(Rng& r) {
for (Edge& e : edges) {
e.weight = r.variate<Real, std::exponential_distribution>(1);
}
}
void computeWeights(Real T) {
for (Edge& e : edges) {
e.weights.push(exp(-e.weight / T));
}
for (unsigned i = n; i > 0; i--) {
#pragma omp parallel for
for (unsigned j = 0; j < pow(i, 2); j++) {
auto [e1, e2, e3, e4] = face(i, j);
Real w = e1.weights.top();
Real x = e2.weights.top();
Real y = e3.weights.top();
Real z = e4.weights.top();
Real cellFactor = std::max(std::numeric_limits<Real>::min(), w * z + x * y);
e1.weights.push(z / cellFactor);
e2.weights.push(y / cellFactor);
e3.weights.push(x / cellFactor);
e4.weights.push(w / cellFactor);
}
}
// This process computes one extra weight per edge.
for (Edge& e : edges) {
e.weights.pop();
}
}
Real computeProbabilities() { // destroys *all* weights
for (Edge& e : edges) {
e.probability = 0;
}
Real logPartitionFunction = 0;
for (unsigned i = 1; i <= n; i++) {
#pragma omp parallel for reduction(+ : logPartitionFunction)
for (unsigned j = 0; j < pow(i, 2); j++) {
auto [e1, e2, e3, e4] = face(i, j);
Real p = e1.probability;
Real q = e2.probability;
Real r = e3.probability;
Real s = e4.probability;
Real w = e1.weights.top();
Real x = e2.weights.top();
Real y = e3.weights.top();
Real z = e4.weights.top();
Real cellFactor = w * z + x * y;
Real deficit = 1 - p - q - r - s;
e1.probability = s + deficit * w * z / cellFactor;
e2.probability = r + deficit * x * y / cellFactor;
e3.probability = q + deficit * x * y / cellFactor;
e4.probability = p + deficit * w * z / cellFactor;
e1.weights.pop();
e2.weights.pop();
e3.weights.pop();
e4.weights.pop();
logPartitionFunction += log(cellFactor);
}
}
return logPartitionFunction;
}
};
bool edgeMatched(PerfectMatching& pm, const AztecDiamond::Edge& e);
PerfectMatching findGroundState(const AztecDiamond& a);
|
