summaryrefslogtreecommitdiff
path: root/hadamard_pt.hpp
blob: b73afe093eeda0d7b096e88ab0fab3813cc3dff9 (plain)
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
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180

#pragma once
#include "hadamard_mcmc.hpp"
#include <execution>
#include <list>

void swap(MCMC& s1, MCMC& s2) {
  std::swap(s1.M, s2.M);
  std::swap(s1.E, s2.E);
}

class ParallelMeasurement {
public:
  virtual void after_step(bool, unsigned, double, double, const MCMC&, const MCMC&){};
};

class PT {
private:
  Rng rng;

public:
  std::vector<MCMC> Ms;
  ParallelMeasurement& B;
  std::vector<Measurement*>& As;

  PT(double β₀, double β₁, unsigned N, unsigned n, ParallelMeasurement& B,
     std::vector<Measurement*>& As)
      : B(B), As(As) {
    Ms.reserve(N);
    for (unsigned i = 0; i < N; i++) {
      double β = β₀ + i * (β₁ - β₀) / (N - 1);
      Ms.push_back(MCMC(n, β, *As[i]));
    }
  }

  double T(unsigned i) { return 1 / Ms[Ms.size() - i - 1].β; }

  bool step(unsigned i, unsigned j, bool dry = false) {
    double Δβ = Ms[i].β - Ms[j].β;
    double ΔE = Ms[i].E - Ms[j].E;

    bool accepted = Δβ * ΔE > 0 || exp(Δβ * ΔE) > rng.uniform((double)0.0, 1.0);

    if (accepted)
      swap(Ms[i], Ms[j]);

    if (!dry) {
      B.after_step(accepted, i, Δβ, ΔE, Ms[i], Ms[j]);
    }
    return accepted;
  }

  std::vector<double> tune(unsigned n0, unsigned m, double ε, double ε2) {
    unsigned n = n0;

    while (true) {
      std::vector<color> colors(Ms.size(), none);
      colors.front() = down;
      colors.back() = up;

      std::vector<unsigned> nu(Ms.size(), 0);
      std::vector<unsigned> nd(Ms.size(), 0);

      for (unsigned i = 0; i < n; i++) {
        std::for_each(std::execution::par_unseq, Ms.begin(), Ms.end(),
                      [m, ε](MCMC& M) { M.tune(m, ε); });

        for (unsigned k = 0; k < m * Ms.size(); k++) {
          unsigned j = rng.uniform((unsigned)0, (unsigned)(Ms.size() - 2));

          if (this->step(j, j + 1, true)) {
            std::swap(colors[j], colors[j + 1]);
            colors.front() = down;
            colors.back() = up;
          }
        }

        for (unsigned j = 0; j < Ms.size(); j++) {
          if (colors[j] == up) {
            nu[j]++;
          } else if (colors[j] == down) {
            nd[j]++;
          }
        }
      }

      std::vector<double> f(Ms.size());
      std::vector<double> δf(Ms.size());

      std::vector<unsigned> f_keep;
      double f_last = 0;
      for (unsigned i = 0; i < Ms.size(); i++) {
        f[i] = nu[i] / (double)(nu[i] + nd[i]);
        δf[i] = f[i] * sqrt(1.0/nu[i] + 1.0/(nu[i]+nd[i]));
        if (f[i] >= f_last) {
          f_keep.push_back(i);
          f_last = f[i];
        }
      }

      for (signed i = 0; i < f_keep.size() - 1; i++) {
        for (unsigned j = f_keep[i]; j < f_keep[i + 1]; j++) {
          f[j] = f[f_keep[i]] + (f[f_keep[i + 1]] - f[f_keep[i]]) /
                                    (Ms[f_keep[i + 1]].β - Ms[f_keep[i]].β) *
                                    (Ms[j].β - Ms[f_keep[i]].β);
          δf[j] = sqrt(pow(δf[f_keep[i]], 2) + pow(sqrt(pow(δf[f_keep[i + 1]], 2) + pow(δf[f_keep[i]], 2)) /
                                    (Ms[f_keep[i + 1]].β - Ms[f_keep[i]].β) *
                                    (Ms[j].β - Ms[f_keep[i]].β), 2));
        }
      }

      // measure the difference between f and a straight line
      double Δf² = 0;
      double δΔf⁴ = 0;

      for (unsigned j = 1; j < f.size(); j++) {
        Δf² += pow(f[j] - (j + 1.0) / (Ms.size() + 1.0), 2);
        δΔf⁴ += pow(2 * δf[j] * f[j], 2);
      }

      std::vector<double> η(Ms.size() - 1);

      for (unsigned i = 0; i < Ms.size() - 1; i++) {
        η[η.size() - i - 1] = sqrt((f[i + 1] - f[i])) / (1 / Ms[i].β - 1 / Ms[i + 1].β);
      }

      double C = 0;
      for (unsigned i = 0; i < η.size(); i++) {
        C += η[i] * (T(i + 1) - T(i));
      }

      std::vector<double> T1(Ms.size() - 2);

      double x = 0;
      unsigned j = 1;

      for (unsigned i = 0; i < η.size(); i++) {
        double xnew = x + η[i] * (T(i + 1) - T(i)) / C;
        while (j < xnew * η.size() && j < η.size()) {
          T1[j - 1] = T(i) + (j / (double)η.size() - x) / η[i] * C;
          j++;
        }
        x = xnew;
      }

      double err = 0;
      for (unsigned i = 0; i < T1.size(); i++) {
        err += fabs(T1[i] - 1 / Ms[Ms.size() - i - 2].β);
        Ms[Ms.size() - i - 2].β = 1.0 / T1[i];
      }

      double Δf = sqrt(Δf²);
      double δΔf = sqrt(δΔf⁴) / 2 / Δf;
      double relErr = Δf / Ms.size();
      // double relErr = err / T1.size() * Ms.size() / (1 / Ms.front().β - 1 / Ms.back().β);

      std::cout << "RMS difference from ideal transit flow is " << relErr << " ± " << δΔf / Ms.size() << ".\n";
      if (relErr < ε2 && !(δΔf != δΔf)) {
        return f;
      } else {
        if (5 * δΔf > Δf || δΔf != δΔf) {
          n *= 2;
          std::cout << "Error in RMS difference too close to difference, increasing tuning to " << n << ".\n";
        }
      }
    }
  }

  void run(unsigned n, unsigned m, bool dry = false) {
    for (unsigned i = 0; i < n; i++) {
      std::for_each(std::execution::par_unseq, Ms.begin(), Ms.end(),
                    [m, dry](MCMC& M) { M.run(m, dry); });
      for (unsigned j = 0; j < Ms.size(); j++) {
        unsigned k = rng.uniform((unsigned)0, (unsigned)(Ms.size() - 2));
        this->step(k, k + 1, dry);
      }
    }
  }
};