diff options
Diffstat (limited to 'lib')
-rw-r--r-- | lib/cluster.h | 22 | ||||
-rw-r--r-- | lib/dihedral.h | 112 | ||||
-rw-r--r-- | lib/dihedral_inf.h | 97 | ||||
-rw-r--r-- | lib/finite_states.h | 9 | ||||
-rw-r--r-- | lib/orthogonal.h | 200 | ||||
-rw-r--r-- | lib/state.h | 7 | ||||
-rw-r--r-- | lib/symmetric.h | 98 | ||||
-rw-r--r-- | lib/wolff.h | 1 | ||||
-rw-r--r-- | lib/z2.h | 60 |
9 files changed, 230 insertions, 376 deletions
diff --git a/lib/cluster.h b/lib/cluster.h index 427c3c8..c5f2be7 100644 --- a/lib/cluster.h +++ b/lib/cluster.h @@ -32,13 +32,13 @@ void flip_cluster(state_t <R_t, X_t> *state, v_t v0, R_t r, gsl_rng *rand) { marks[v] = true; if (v == state->nv) { - R_new = act (r, R_old); + R_new = r.act(R_old); } else { si_old = state->spins[v]; - si_new = act (r, si_old); + si_new = r.act(si_old); } - for (v_t vn : state->g.v_adj[v]) { + for (const v_t &vn : state->g.v_adj[v]) { X_t sj; if (vn != state->nv) { @@ -53,17 +53,17 @@ void flip_cluster(state_t <R_t, X_t> *state, v_t v0, R_t r, gsl_rng *rand) { X_t rs_old, rs_new; v_t non_ghost; if (vn == state->nv) { - rs_old = act_inverse (R_old, si_old); - rs_new = act_inverse (R_old, si_new); + rs_old = R_old.act_inverse(si_old); + rs_new = R_old.act_inverse(si_new); non_ghost = v; } else { - rs_old = act_inverse (R_old, sj); - rs_new = act_inverse (R_new, sj); + rs_old = R_old.act_inverse(sj); + rs_new = R_new.act_inverse(sj); non_ghost = vn; } double dE = state->H(rs_old) - state->H(rs_new); #ifdef FINITE_STATES - prob = H_probs[N_STATES * state_to_ind(rs_old) + state_to_ind(rs_new)]; + prob = H_probs[state_to_ind(rs_old)][state_to_ind(rs_new)]; #else prob = 1.0 - exp(-dE / state->T); #endif @@ -85,7 +85,7 @@ void flip_cluster(state_t <R_t, X_t> *state, v_t v0, R_t r, gsl_rng *rand) { } else { double dE = state->J(si_old, sj) - state->J(si_new, sj); #ifdef FINITE_STATES - prob = J_probs[N_STATES * N_STATES * state_to_ind(si_old) + N_STATES * state_to_ind(si_new) + state_to_ind(sj)]; + prob = J_probs[state_to_ind(si_old)][state_to_ind(si_new)][state_to_ind(sj)]; #else prob = 1.0 - exp(-dE / state->T); #endif @@ -98,13 +98,9 @@ void flip_cluster(state_t <R_t, X_t> *state, v_t v0, R_t r, gsl_rng *rand) { } if (v == state->nv) { - free_spin(state->R); state->R = R_new; } else { state->spins[v] = si_new; - } - - if (v != state->nv) { // count the number of non-external sites that flip nv++; } } diff --git a/lib/dihedral.h b/lib/dihedral.h index 8238835..5588afa 100644 --- a/lib/dihedral.h +++ b/lib/dihedral.h @@ -4,76 +4,46 @@ #include "types.h" #include "potts.h" -template <class T, q_t q> -struct dihedral_t { bool is_reflection; T x; }; - -template <class T, q_t q> -void init(dihedral_t<T, q> *ptr) { - ptr->is_reflection = false; - ptr->x = (T)0; -} - -template <class T, q_t q> -dihedral_t<T, q> copy(dihedral_t<T, q> r) { - return r; -} - -template <class T, q_t q> -void free_spin(dihedral_t<T, q> r) { - // do nothing! -} - -template <q_t q> -potts_t<q> act(dihedral_t<q_t, q> r, potts_t<q> s) { - potts_t<q> s2; - if (r.is_reflection) { - s2.x = ((q + r.x) - s.x) % q; - } else { - s2.x = (r.x + s.x) % q; - } - - return s2; -} - -template <q_t q> -dihedral_t<q_t,q> act(dihedral_t<q_t,q> r1, dihedral_t<q_t,q> r2) { - dihedral_t<q_t,q> r3; - - if (r1.is_reflection) { - r3.is_reflection = !(r2.is_reflection); - r3.x = ((q + r1.x) - r2.x) % q; - } else { - r3.is_reflection = r2.is_reflection; - r3.x = (r1.x + r2.x) % q; - } - - return r3; -} - -template <q_t q> -potts_t<q> act_inverse(dihedral_t<q_t,q> r, potts_t<q> s) { - potts_t<q> s2; - if (r.is_reflection) { - s2.x = ((r.x + q) - s.x) % q; - } else { - s2.x = ((s.x + q) - r.x) % q; - } - - return s2; -} - template <q_t q> -dihedral_t<q_t, q> act_inverse(dihedral_t<q_t,q> r1, dihedral_t<q_t,q> r2) { - dihedral_t<q_t,q> r3; - - if (r1.is_reflection) { - r3.is_reflection = !(r2.is_reflection); - r3.x = ((r1.x + q) - r2.x) % q; - } else { - r3.is_reflection = r2.is_reflection; - r3.x = ((r2.x + q) - r1.x) % q; - } - - return r3; -} +class dihedral_t { + public: + bool is_reflection; + q_t x; + + dihedral_t() : is_reflection(false), x(0) {} + dihedral_t(bool x, q_t y) : is_reflection(x), x(y) {} + + potts_t<q> act(const potts_t<q>& s) const { + if (this->is_reflection) { + return potts_t<q>(((q + this->x) - s.x) % q); + } else { + return potts_t<q>((this->x + s.x) % q); + } + } + + + dihedral_t<q> act(dihedral_t<q> r) const { + if (this->is_reflection) { + return dihedral_t<q>(!(r.is_reflection), ((q + this->x) - r.x) % q); + } else { + return dihedral_t<q>(r.is_reflection, (this->x + r.x) % q); + } + } + + potts_t<q> act_inverse(potts_t<q> s) const { + if (this->is_reflection) { + return this->act(s); + } else { + return potts_t<q>(((s.x + q) - this->x) % q); + } + } + + dihedral_t<q> act_inverse(dihedral_t<q> r) const { + if (this->is_reflection) { + return this->act(r); + } else { + return dihedral_t<q>(r.is_reflection, ((r.x + q) - this->x) % q); + } + } +}; diff --git a/lib/dihedral_inf.h b/lib/dihedral_inf.h index f7c4297..a064b48 100644 --- a/lib/dihedral_inf.h +++ b/lib/dihedral_inf.h @@ -4,75 +4,44 @@ #include "height.h" template <class T> -struct dihedral_inf_t { bool is_reflection; T x; }; - -template <class T> -void init(dihedral_inf_t<T> *ptr) { - ptr->is_reflection = false; - ptr->x = (T)0; -} - -template <class T> -dihedral_inf_t<T> copy(dihedral_inf_t<T> r) { - return r; -} - -template <class T> -void free_spin(dihedral_inf_t<T> r) { - // do nothing! -} - -template <class T> -height_t<T> act(dihedral_inf_t<T> r, height_t<T> h) { - height_t<T> h2; - if (r.is_reflection) { - h2.x = r.x - h.x; - } else { - h2.x = r.x + h.x; +class dihedral_inf_t { + public: + bool is_reflection; + T x; + + dihedral_inf_t() : is_reflection(false), x(0) {} + dihedral_inf_t(bool x, T y) : is_reflection(x), x(y) {} + + height_t<T> act(const height_t<T>& h) const { + if (this->is_reflection) { + return height_t(this->x - h.x); + } else { + return height_t(this->x + h.x); + } } - return h2; -} - -template <class T> -dihedral_inf_t<T> act(dihedral_inf_t<T> r1, dihedral_inf_t<T> r2) { - dihedral_inf_t<T> r3; - - if (r1.is_reflection) { - r3.is_reflection = !(r2.is_reflection); - r3.x = r1.x - r2.x; - } else { - r3.is_reflection = r2.is_reflection; - r3.x = r1.x + r2.x; + dihedral_inf_t<T> act(const dihedral_inf_t<T>& r) const { + if (this->is_reflection) { + return dihedral_inf_t<T>(!r.is_reflection, this->x - r.x); + } else { + return dihedral_inf_t<T>(r.is_reflection, this->x + r.x); + } } - return r3; -} - -template <class T> -height_t<T> act_inverse(dihedral_inf_t<T> r, height_t<T> h) { - height_t<T> h2; - if (r.is_reflection) { - h2.x = r.x - h.x; - } else { - h2.x = h.x - r.x; + height_t<T> act_inverse(const height_t<T>& h) const { + if (this->is_reflection) { + return this->act(h); + } else { + return height_t(h.x - this->x); + } } - return h2; -} - -template <class T> -dihedral_inf_t<T> act_inverse(dihedral_inf_t<T> r1, dihedral_inf_t<T> r2) { - dihedral_inf_t<T> r3; - - if (r1.is_reflection) { - r3.is_reflection = !(r2.is_reflection); - r3.x = r1.x - r2.x; - } else { - r3.is_reflection = r2.is_reflection; - r3.x = r2.x - r1.x; + dihedral_inf_t<T> act_inverse(const dihedral_inf_t<T>& r) const { + if (this->is_reflection) { + return this->act(r); + } else { + return dihedral_inf_t<T>(r.is_reflection, r.x - this->x); + } } - - return r3; -} +}; diff --git a/lib/finite_states.h b/lib/finite_states.h index 1e8800c..08eff30 100644 --- a/lib/finite_states.h +++ b/lib/finite_states.h @@ -3,27 +3,28 @@ #include <cmath> #include <functional> +#include <array> #define FINITE_STATES // must have N_STATES, states[N_STATES], and state_to_ind defined before // invoking header -double J_probs[N_STATES * N_STATES * N_STATES]; -double H_probs[N_STATES * N_STATES]; +std::array<std::array<std::array<double, N_STATES>, N_STATES>, N_STATES> J_probs; +std::array<std::array<double, N_STATES>, N_STATES> H_probs; template <class X_t> void initialize_probs(std::function <double(X_t, X_t)> J, std::function <double(X_t)> H, double T) { for (q_t i = 0; i < N_STATES; i++) { for (q_t j = 0; j < N_STATES; j++) { for (q_t k = 0; k < N_STATES; k++) { - J_probs[i * N_STATES * N_STATES + j * N_STATES +k] = 1.0 - exp(-(J(states[i], states[k]) - J(states[j], states[k])) / T); + J_probs[i][j][k] = 1.0 - exp(-(J(states[i], states[k]) - J(states[j], states[k])) / T); } } } for (q_t i = 0; i < N_STATES; i++) { for (q_t j = 0; j < N_STATES; j++) { - H_probs[i * N_STATES + j] = 1.0 - exp(-(H(states[i]) - H(states[j])) / T); + H_probs[i][j] = 1.0 - exp(-(H(states[i]) - H(states[j])) / T); } } } diff --git a/lib/orthogonal.h b/lib/orthogonal.h index c0958f2..34cd44e 100644 --- a/lib/orthogonal.h +++ b/lib/orthogonal.h @@ -11,164 +11,134 @@ #include "vector.h" template <q_t q, class T> -struct orthogonal_t { bool is_reflection; T *x; }; +class orthogonal_t : public std::array<std::array<T, q>, q> { + public : + bool is_reflection; -template <q_t q, class T> -void init(orthogonal_t <q, T> *ptr) { - ptr->is_reflection = false; - ptr->x = (T *)calloc(q * q, sizeof(T)); - - for (q_t i = 0; i < q; i++) { - ptr->x[q * i + i] = (T)1; - } -} - -template <q_t q, class T> -orthogonal_t <q, T> copy (orthogonal_t <q, T> m) { - orthogonal_t <q, T> m_copy; - m_copy.is_reflection = m.is_reflection; - - q_t size; - - if (m.is_reflection) { - size = q; - } else { - size = q * q; - } - - m_copy.x = (T *)calloc(size, sizeof(T)); - - for (q_t i = 0; i < size; i++) { - m_copy.x[i] = m.x[i]; + orthogonal_t() : is_reflection(false) { + for (q_t i = 0; i < q; i++) { + (*this)[i].fill(0); + (*this)[i][i] = (T)1; + } } - return m_copy; -} - -template <q_t q, class T> -void free_spin (orthogonal_t <q, T> m) { - free(m.x); -} - -template <q_t q, class T> -vector_t <q, T> act (orthogonal_t <q, T> m, vector_t <q, T> v) { - vector_t <q, T> v_rot; + vector_t<q, T> act(const vector_t <q, T>& v) const { + vector_t <q, T> v_rot; - if (m.is_reflection) { - double prod = 0; - for (q_t i = 0; i < q; i++) { - prod += v[i] * m.x[i]; - } - for (q_t i = 0; i < q; i++) { - v_rot[i] = v[i] - 2 * prod * m.x[i]; - } - } else { - for (q_t i = 0; i < q; i++) { - for (q_t j = 0; j < q; j++) { - v_rot[i] += m.x[q * i + j] * v[j]; + if (is_reflection) { + double prod = 0; + for (q_t i = 0; i < q; i++) { + prod += v[i] * (*this)[0][i]; + } + for (q_t i = 0; i < q; i++) { + v_rot[i] = v[i] - 2 * prod * (*this)[0][i]; + } + } else { + for (q_t i = 0; i < q; i++) { + for (q_t j = 0; j < q; j++) { + v_rot[i] += (*this)[i][j] * v[j]; + } } } - } - return v_rot; -} + return v_rot; + } -template <q_t q, class T> -orthogonal_t <q, T> act (orthogonal_t <q, T> m1, orthogonal_t <q, T> m2) { - orthogonal_t <q, T> m2_rot; + orthogonal_t<q, T> act(const orthogonal_t <q, T>& m) const { + orthogonal_t <q, T> m_rot; - m2_rot.is_reflection = false; - m2_rot.x = (T *)calloc(q * q, sizeof(T)); + m_rot.is_reflection = false; - if (m1.is_reflection) { - for (q_t i = 0; i < q; i++) { - double akOki = 0; + if (is_reflection) { + for (q_t i = 0; i < q; i++) { + double akOki = 0; - for (q_t k = 0; k < q; k++) { - akOki += m1.x[k] * m2.x[q * k + i]; - } + for (q_t k = 0; k < q; k++) { + akOki += (*this)[0][k] * m[k][i]; + } - for (q_t j = 0; j < q; j++) { - m2_rot.x[q * j + i] = m2.x[q * j + i] - 2 * akOki * m1.x[j]; + for (q_t j = 0; j < q; j++) { + m_rot[j][i] = m[j][i] - 2 * akOki * (*this)[0][j]; + } } - } - } else { - for (q_t i = 0; i < q; i++) { - for (q_t j = 0; j < q; j++) { - for (q_t k = 0; k < q; k++) { - m2_rot.x[i * q + j] += m1.x[i * q + j] * m2.x[j * q + k]; + } else { + for (q_t i = 0; i < q; i++) { + m_rot[i].fill(0); + for (q_t j = 0; j < q; j++) { + for (q_t k = 0; k < q; k++) { + m_rot[i][j] += (*this)[i][j] * m[j][k]; + } } } } - } - return m2_rot; -} + return m_rot; + } -template <q_t q, class T> -vector_t <q, T> act_inverse (orthogonal_t <q, T> m, vector_t <q, T> v) { - if (m.is_reflection) { - return act(m, v); // reflections are their own inverse - } else { - vector_t <q, T> v_rot; + vector_t <q, T> act_inverse(const vector_t <q, T>& v) const { + if (is_reflection) { + return this->act(v); // reflections are their own inverse + } else { + vector_t <q, T> v_rot; - for (q_t i = 0; i < q; i++) { - for (q_t j = 0; j < q; j++) { - v_rot[i] += m.x[q * j + i] * v[j]; + for (q_t i = 0; i < q; i++) { + for (q_t j = 0; j < q; j++) { + v_rot[i] += (*this)[j][i] * v[j]; + } } - } - return v_rot; + return v_rot; + } } -} - -template <q_t q, class T> -orthogonal_t <q, T> act_inverse (orthogonal_t <q, T> m1, orthogonal_t <q, T> m2) { - if (m1.is_reflection) { - return act(m1, m2); // reflections are their own inverse - } else { - orthogonal_t <q, T> m2_rot; - m2_rot.x = (T *)calloc(q * q, sizeof(T)); - for (q_t i = 0; i < q; i++) { - for (q_t j = 0; j < q; j++) { - for (q_t k = 0; k < q; k++) { - m2_rot.x[i * q + j] += m1.x[j * q + i] * m2.x[j * q + k]; + vector_t <q, T> act_inverse(const orthogonal_t <q, T>& m) const { + if (is_reflection) { + return this->act(m); // reflections are their own inverse + } else { + orthogonal_t <q, T> m_rot; + m_rot.is_reflection = false; + + for (q_t i = 0; i < q; i++) { + m_rot[i].fill(0); + for (q_t j = 0; j < q; j++) { + for (q_t k = 0; k < q; k++) { + m_rot[i][j] += (*this)[j][i] * m[j][k]; + } } } - } - return m2_rot; + return m_rot; + } } -} + +}; + template <q_t q> -orthogonal_t <q, double> generate_rotation_uniform (gsl_rng *r, vector_t <q, double> v) { +orthogonal_t <q, double> generate_rotation_uniform (gsl_rng *r, const vector_t <q, double>& v) { orthogonal_t <q, double> ptr; ptr.is_reflection = true; - ptr.x = (double *)calloc(q, sizeof(double)); double v2 = 0; for (q_t i = 0; i < q; i++) { - ptr.x[i] = gsl_ran_ugaussian(r); - v2 += ptr.x[i] * ptr.x[i]; + ptr[0][i] = gsl_ran_ugaussian(r); + v2 += ptr[0][i] * ptr[0][i]; } double mag_v = sqrt(v2); for (q_t i = 0; i < q; i++) { - ptr.x[i] /= mag_v; + ptr[0][i] /= mag_v; } return ptr; } template <q_t q> -orthogonal_t <q, double> generate_rotation_perturbation (gsl_rng *r, vector_t <q, double> v0, double epsilon, unsigned int n) { +orthogonal_t <q, double> generate_rotation_perturbation (gsl_rng *r, const vector_t <q, double>& v0, double epsilon, unsigned int n) { orthogonal_t <q, double> m; m.is_reflection = true; - m.x = (double *)malloc(q * sizeof(double)); vector_t <q, double> v; @@ -191,22 +161,22 @@ orthogonal_t <q, double> generate_rotation_perturbation (gsl_rng *r, vector_t <q double m_dot_v = 0; for (q_t i = 0; i < q; i++) { - m.x[i] = gsl_ran_ugaussian(r); - m_dot_v += m.x[i] * v[i]; + m[0][i] = gsl_ran_ugaussian(r); + m_dot_v += m[0][i] * v[i]; } double v2 = 0; double factor = epsilon * gsl_ran_ugaussian(r); for (q_t i = 0; i < q; i++) { - m.x[i] = m.x[i] - m_dot_v * v[i] + factor * v[i]; - v2 += pow(m.x[i], 2); + m[0][i] = m[0][i] - m_dot_v * v[i] + factor * v[i]; + v2 += pow(m[0][i], 2); } double mag_v = sqrt(v2); for (q_t i = 0; i < q; i++) { - m.x[i] /= mag_v; + m[0][i] /= mag_v; } return m; diff --git a/lib/state.h b/lib/state.h index b57a9e2..eefa0cb 100644 --- a/lib/state.h +++ b/lib/state.h @@ -30,12 +30,11 @@ class state_t { std::function <double(X_t, X_t)> J; std::function <double(X_t)> H; - state_t(D_t D, L_t L, double T, std::function <double(X_t, X_t)> J, std::function <double(X_t)> H) : D(D), L(L), g(D, L), T(T), J(J), H(H) { + state_t(D_t D, L_t L, double T, std::function <double(X_t, X_t)> J, std::function <double(X_t)> H) : D(D), L(L), g(D, L), T(T), R(), J(J), H(H) { nv = g.nv; ne = g.ne; g.add_ext(); spins.resize(nv); - init (&R); E = - (double)ne * J(spins[0], spins[0]) - (double)nv * H(spins[0]); M = spins[0] * nv; last_cluster_size = 0; @@ -52,10 +51,6 @@ class state_t { precomputed_sin[i] = sin(2 * M_PI * (double)i / (double)L); } } - - ~state_t() { - free_spin(R); - } }; diff --git a/lib/symmetric.h b/lib/symmetric.h index 41c8fb6..9e9b6e4 100644 --- a/lib/symmetric.h +++ b/lib/symmetric.h @@ -2,84 +2,50 @@ #pragma once #include <stdlib.h> +#include <array> #include "types.h" #include "potts.h" template <q_t q> -class symmetric_t { +class symmetric_t : public std::array<q_t, q> { public: - q_t *perm; -}; - -template <q_t q> -void init(symmetric_t<q> *p) { - p->perm = (q_t *)malloc(q * sizeof(q_t)); - - for (q_t i = 0; i < q; i++) { - p->perm[i] = i; - } -} - -template <q_t q> -void free_spin(symmetric_t<q> p) { - free(p.perm); -} -template <q_t q> -symmetric_t<q> copy(symmetric_t<q> x) { - symmetric_t<q> x2; - x2.perm = (q_t *)malloc(q * sizeof(q_t)); - - for (q_t i = 0; i < q; i++) { - x2.perm[i] = x.perm[i]; - } - - return x2; -} - -template <q_t q> -potts_t<q> act(symmetric_t<q> r, potts_t<q> s) { - potts_t<q> s2; - s2.x = r.perm[s.x]; - return s2; -} - -template <q_t q> -symmetric_t<q> act(symmetric_t<q> r1, symmetric_t<q> r2) { - symmetric_t<q> r3; - r3.perm = (q_t *)malloc(q * sizeof(q_t)); - for (q_t i = 0; i < q; i++) { - r3.perm[i] = r1.perm[r2.perm[i]]; - } - - return r3; -} + symmetric_t() { + for (q_t i = 0; i < q; i++) { + (*this)[i] = i; + } + } -template <q_t q> -potts_t<q> act_inverse(symmetric_t<q> r, potts_t<q> s) { - potts_t<q> s2; + potts_t<q> act(const potts_t<q> &s) const { + return potts_t<q>((*this)[s.x]); + } - q_t i; + symmetric_t<q> act(const symmetric_t<q>& r) const { + symmetric_t<q> r_rot; + for (q_t i = 0; i < q; i++) { + r_rot[i] = (*this)[r[i]]; + } - for (i = 0; i < q; i++) { - if (r.perm[i] == s.x) { - break; + return r_rot; } - } - s2.x = i; + potts_t<q> act_inverse(const potts_t<q>& s) const { + for (q_t i = 0; i < q; i++) { + if ((*this)[i] == s.x) { + return potts_t<q>(i); + } + } - return s2; -} + exit(EXIT_FAILURE); + } -template <q_t q> -symmetric_t<q> act_inverse(symmetric_t<q> r1, symmetric_t<q> r2) { - symmetric_t<q> r3; - r3.perm = (q_t *)malloc(q * sizeof(q_t)); - for (q_t i = 0; i < q; i++) { - r3.perm[r1.perm[i]] = r2.perm[i]; - } + symmetric_t<q> act_inverse(const symmetric_t<q>& r) const { + symmetric_t<q> r_rot; + for (q_t i = 0; i < q; i++) { + r_rot[(*this)[i]] = r[i]; + } - return r3; -} + return r_rot; + } +}; diff --git a/lib/wolff.h b/lib/wolff.h index da4b7b6..a4a663c 100644 --- a/lib/wolff.h +++ b/lib/wolff.h @@ -16,7 +16,6 @@ void wolff(count_t N, state_t <R_t, X_t> *s, std::function <R_t(gsl_rng *, X_t s v_t v0 = gsl_rng_uniform_int(r, s->nv); R_t step = gen_R(r, s->spins[v0]); flip_cluster <R_t, X_t> (s, v0, step, r); - free_spin(step); measurements(s); } @@ -17,49 +17,37 @@ * */ -struct z2_t { bool x; }; +class z2_t { + public: + bool x; -void init(z2_t *p) { - p->x = false; -} + z2_t() : x(false) {} -void free_spin(z2_t p) { - // do nothing! -} + z2_t(bool x) : x(x) {} -z2_t copy(z2_t x) { - return x; -} - -ising_t act(z2_t r, ising_t s) { - ising_t rs; - - if (r.x) { - rs.x = !s.x; - return rs; - } else { - rs.x = s.x; - return rs; + ising_t act(const ising_t& s) { + if (x) { + return ising_t(!s.x); + } else { + return ising_t(s.x); + } } -} -z2_t act(z2_t r1, z2_t r2) { - z2_t r3; + z2_t act(const z2_t& r) { + if (x) { + return z2_t(!r.x); + } else { + return z2_t(r.x); + } + } - if (r1.x) { - r3.x = !r2.x; - return r3; - } else { - r3.x = r2.x; - return r3; + ising_t act_inverse(const ising_t& s) { + return this->act(s); } -} -ising_t act_inverse(z2_t r, ising_t s) { - return act(r, s); -} + z2_t act_inverse(const z2_t& r) { + return this->act(r); + } +}; -z2_t act_inverse(z2_t r1, z2_t r2) { - return act(r1, r2); -} |