|
| 1 | +from __future__ import division |
| 2 | + |
| 3 | +import numpy as np |
| 4 | +from scipy.constants import c, e |
| 5 | + |
| 6 | +from PyHEADTAIL.general.element import Element |
| 7 | +import PyHEADTAIL.particles.generators as gen |
| 8 | + |
| 9 | +try: |
| 10 | + from PyHEADTAIL.trackers.transverse_tracking_cython import TransverseMap |
| 11 | + from PyHEADTAIL.trackers.detuners_cython import (Chromaticity, |
| 12 | + AmplitudeDetuning) |
| 13 | +except ImportError as e: |
| 14 | + print ("*** Warning: could not import cython variants of trackers, " |
| 15 | + "did you cythonize (use the following command)?\n" |
| 16 | + "$ ./install \n" |
| 17 | + "Falling back to (slower) python version.") |
| 18 | + from PyHEADTAIL.trackers.transverse_tracking import TransverseMap |
| 19 | + from PyHEADTAIL.trackers.detuners import Chromaticity, AmplitudeDetuning |
| 20 | + |
| 21 | +from PyHEADTAIL.trackers.simple_long_tracking import LinearMap, RFSystems |
| 22 | + |
| 23 | +class BasicSynchrotron(Element): |
| 24 | + def __init__(self, optics_mode, circumference=None, n_segments=None, s=None, name=None, |
| 25 | + alpha_x=None, beta_x=None, D_x=None, alpha_y=None, beta_y=None, D_y=None, |
| 26 | + accQ_x=None, accQ_y=None, Qp_x=0, Qp_y=0, app_x=0, app_y=0, app_xy=0, |
| 27 | + alpha_mom_compaction=None, longitudinal_mode=None, Q_s=None, |
| 28 | + h_RF=None, V_RF=None, dphi_RF=None, p0=None, p_increment=None, |
| 29 | + charge=None, mass=None, **kwargs): |
| 30 | + |
| 31 | + |
| 32 | + self.optics_mode = optics_mode |
| 33 | + self.longitudinal_mode = longitudinal_mode |
| 34 | + self.charge = charge |
| 35 | + self.mass = mass |
| 36 | + self.p0 = p0 |
| 37 | + |
| 38 | + self.one_turn_map = [] |
| 39 | + |
| 40 | + detuners = [] |
| 41 | + if Qp_x != 0 or Qp_y != 0: |
| 42 | + detuners.append(Chromaticity(Qp_x, Qp_y)) |
| 43 | + if app_x != 0 or app_y != 0 or app_xy != 0: |
| 44 | + detuners.append(AmplitudeDetuning(app_x, app_y, app_xy)) |
| 45 | + |
| 46 | + # construct transverse map |
| 47 | + self._contruct_transverse_map(optics_mode=optics_mode, circumference=circumference, n_segments=n_segments, s=s, name=name, |
| 48 | + alpha_x=alpha_x, beta_x=beta_x, D_x=D_x, alpha_y=alpha_y, beta_y=beta_y, D_y=D_y, |
| 49 | + accQ_x=accQ_x, accQ_y=accQ_y, detuners=detuners) |
| 50 | + |
| 51 | + # construct longitudinal map |
| 52 | + self._contruct_longitudinal_map(alpha_mom_compaction=alpha_mom_compaction, longitudinal_mode=longitudinal_mode, Q_s=Q_s, |
| 53 | + h_RF=h_RF, V_RF=V_RF, dphi_RF=dphi_RF, p_increment=p_increment) |
| 54 | + |
| 55 | + |
| 56 | + |
| 57 | + @property |
| 58 | + def gamma(self): |
| 59 | + return self._gamma |
| 60 | + @gamma.setter |
| 61 | + def gamma(self, value): |
| 62 | + self._gamma = value |
| 63 | + self._beta = np.sqrt(1 - self.gamma**-2) |
| 64 | + self._betagamma = np.sqrt(self.gamma**2 - 1) |
| 65 | + self._p0 = self.betagamma * self.mass * c |
| 66 | + |
| 67 | + @property |
| 68 | + def beta(self): |
| 69 | + return self._beta |
| 70 | + @beta.setter |
| 71 | + def beta(self, value): |
| 72 | + self.gamma = 1. / np.sqrt(1-value**2) |
| 73 | + |
| 74 | + @property |
| 75 | + def betagamma(self): |
| 76 | + return self._betagamma |
| 77 | + @betagamma.setter |
| 78 | + def betagamma(self, value): |
| 79 | + self.gamma = np.sqrt(value**2 + 1) |
| 80 | + |
| 81 | + @property |
| 82 | + def p0(self): |
| 83 | + return self._p0 |
| 84 | + @p0.setter |
| 85 | + def p0(self, value): |
| 86 | + self.gamma = 1 / (c * self.mass) * np.sqrt(value**2+self.mass**2*c**2) |
| 87 | + |
| 88 | + @property |
| 89 | + def Q_x(self): |
| 90 | + return np.atleast_1d(self.transverse_map.accQ_x)[-1] |
| 91 | + |
| 92 | + @property |
| 93 | + def Q_y(self): |
| 94 | + return np.atleast_1d(self.transverse_map.accQ_y)[-1] |
| 95 | + |
| 96 | + def track(self, bunch, verbose=False): |
| 97 | + for m in self.one_turn_map: |
| 98 | + if verbose: |
| 99 | + self.prints('Tracking through:\n' + str(m)) |
| 100 | + m.track(bunch) |
| 101 | + |
| 102 | + def install_after_each_transverse_segment(self, element_to_add): |
| 103 | + '''Attention: Do not add any elements which update the dispersion!''' |
| 104 | + one_turn_map_new = [] |
| 105 | + for element in self.one_turn_map: |
| 106 | + one_turn_map_new.append(element) |
| 107 | + if element in self.transverse_map: |
| 108 | + one_turn_map_new.append(element_to_add) |
| 109 | + self.one_turn_map = one_turn_map_new |
| 110 | + |
| 111 | + def generate_6D_Gaussian_bunch(self, n_macroparticles, intensity, |
| 112 | + epsn_x, epsn_y, sigma_z): |
| 113 | + '''Generate a 6D Gaussian distribution of particles which is |
| 114 | + transversely matched to the Synchrotron. Longitudinally, the |
| 115 | + distribution is matched only in terms of linear focusing. |
| 116 | + For a non-linear bucket, the Gaussian distribution is cut along |
| 117 | + the separatrix (with some margin). It will gradually filament |
| 118 | + into the bucket. This will change the specified bunch length. |
| 119 | + ''' |
| 120 | + if self.longitudinal_mode == 'linear': |
| 121 | + check_inside_bucket = lambda z,dp : np.array(len(z)*[True]) |
| 122 | + elif self.longitudinal_mode == 'non-linear': |
| 123 | + check_inside_bucket = self.longitudinal_map.get_bucket( |
| 124 | + gamma=self.gamma).make_is_accepted(margin=0.05) |
| 125 | + else: |
| 126 | + raise NotImplementedError( |
| 127 | + 'Something wrong with self.longitudinal_mode') |
| 128 | + |
| 129 | + eta = self.longitudinal_map.alpha_array[0] - self.gamma**-2 |
| 130 | + beta_z = np.abs(eta)*self.circumference/2./np.pi/self.longitudinal_map.Qs |
| 131 | + sigma_dp = sigma_z/beta_z |
| 132 | + epsx_geo = epsn_x/self.betagamma |
| 133 | + epsy_geo = epsn_y/self.betagamma |
| 134 | + |
| 135 | + injection_optics = self.transverse_map.get_injection_optics() |
| 136 | + |
| 137 | + bunch = gen.ParticleGenerator(macroparticlenumber=n_macroparticles, |
| 138 | + intensity=intensity, charge=self.charge, mass=self.mass, |
| 139 | + circumference=self.circumference, gamma=self.gamma, |
| 140 | + distribution_x = gen.gaussian2D(epsx_geo), alpha_x=injection_optics['alpha_x'], beta_x=injection_optics['beta_x'], D_x=injection_optics['D_x'], |
| 141 | + distribution_y = gen.gaussian2D(epsy_geo), alpha_y=injection_optics['alpha_y'], beta_y=injection_optics['beta_y'], D_y=injection_optics['D_y'], |
| 142 | + distribution_z = gen.cut_distribution(gen.gaussian2D_asymmetrical(sigma_u=sigma_z, sigma_up=sigma_dp),is_accepted=check_inside_bucket), |
| 143 | + ).generate() |
| 144 | + |
| 145 | + return bunch |
| 146 | + |
| 147 | + def generate_6D_Gaussian_bunch_matched( |
| 148 | + self, n_macroparticles, intensity, epsn_x, epsn_y, |
| 149 | + sigma_z=None, epsn_z=None): |
| 150 | + '''Generate a 6D Gaussian distribution of particles which is |
| 151 | + transversely as well as longitudinally matched. |
| 152 | + The distribution is found iteratively to exactly yield the |
| 153 | + given bunch length while at the same time being stationary in |
| 154 | + the non-linear bucket. Thus, the bunch length should amount |
| 155 | + to the one specificed and should not change significantly |
| 156 | + during the synchrotron motion. |
| 157 | + |
| 158 | + Requires self.longitudinal_mode == 'non-linear' |
| 159 | + for the bucket. |
| 160 | + ''' |
| 161 | + assert self.longitudinal_mode == 'non-linear' |
| 162 | + epsx_geo = epsn_x/self.betagamma |
| 163 | + epsy_geo = epsn_y/self.betagamma |
| 164 | + |
| 165 | + injection_optics = self.transverse_map.get_injection_optics() |
| 166 | + |
| 167 | + bunch = gen.ParticleGenerator(macroparticlenumber=n_macroparticles, |
| 168 | + intensity=intensity, charge=self.charge, mass=self.mass, |
| 169 | + circumference=self.circumference, gamma=self.gamma, |
| 170 | + distribution_x = gen.gaussian2D(epsx_geo), alpha_x=injection_optics['alpha_x'], beta_x=injection_optics['beta_x'], D_x=injection_optics['D_x'], |
| 171 | + distribution_y = gen.gaussian2D(epsy_geo), alpha_y=injection_optics['alpha_y'], beta_y=injection_optics['beta_y'], D_y=injection_optics['D_y'], |
| 172 | + distribution_z = gen.RF_bucket_distribution(self.longitudinal_map.get_bucket(gamma=self.gamma), sigma_z=sigma_z, epsn_z=epsn_z), |
| 173 | + ).generate() |
| 174 | + |
| 175 | + return bunch |
| 176 | + |
| 177 | + def _contruct_transverse_map(self, optics_mode=None, circumference=None, n_segments=None, s=None, name=None, |
| 178 | + alpha_x=None, beta_x=None, D_x=None, alpha_y=None, beta_y=None, D_y=None, |
| 179 | + accQ_x=None, accQ_y=None, detuners=[]): |
| 180 | + |
| 181 | + if optics_mode == 'smooth': |
| 182 | + if circumference is None: |
| 183 | + raise ValueError('circumference has to be specified if optics_mode = "smooth"') |
| 184 | + |
| 185 | + if n_segments is None: |
| 186 | + raise ValueError('n_segments has to be specified if optics_mode = "smooth"') |
| 187 | + |
| 188 | + if s is not None: |
| 189 | + raise ValueError('s vector cannot be provided if optics_mode = "smooth"') |
| 190 | + |
| 191 | + |
| 192 | + s = (np.arange(0, n_segments + 1) |
| 193 | + * circumference / n_segments) |
| 194 | + |
| 195 | + alpha_x=0.*s |
| 196 | + beta_x=0.*s+beta_x |
| 197 | + D_x=0.*s+D_x |
| 198 | + alpha_y=0.*s |
| 199 | + beta_y=0.*s+beta_y |
| 200 | + D_y=0.*s+D_y |
| 201 | + |
| 202 | + elif optics_mode == 'non-smooth': |
| 203 | + if circumference is not None: |
| 204 | + raise ValueError('circumference cannot be provided if optics_mode = "non-smooth"') |
| 205 | + |
| 206 | + if n_segments is not None: |
| 207 | + raise ValueError('n_segments cannot be provided if optics_mode = "non-smooth"') |
| 208 | + |
| 209 | + if s is None: |
| 210 | + raise ValueError('s has to be specified if optics_mode = "smooth"') |
| 211 | + |
| 212 | + else: |
| 213 | + raise ValueError('optics_mode not recognized') |
| 214 | + |
| 215 | + self.transverse_map = TransverseMap(s=s, |
| 216 | + alpha_x=alpha_x, |
| 217 | + beta_x=beta_x, |
| 218 | + D_x=D_x, |
| 219 | + alpha_y=alpha_y, |
| 220 | + beta_y=beta_y, |
| 221 | + D_y=D_y, |
| 222 | + accQ_x=accQ_x, accQ_y=accQ_y, detuners=detuners) |
| 223 | + |
| 224 | + self.circumference = s[-1] |
| 225 | + self.transverse_map.n_segments = len(s)-1 |
| 226 | + |
| 227 | + if name is None: |
| 228 | + self.transverse_map.name = ['P_%d'%ip for ip in xrange(len(s)-1)] |
| 229 | + self.transverse_map.name.append('end_ring') |
| 230 | + else: |
| 231 | + self.transverse_map.name = name |
| 232 | + |
| 233 | + for i_seg, m in enumerate(self.transverse_map): |
| 234 | + m.i0 = i_seg |
| 235 | + m.i1 = i_seg+1 |
| 236 | + m.s0 = self.transverse_map.s[i_seg] |
| 237 | + m.s1 = self.transverse_map.s[i_seg+1] |
| 238 | + m.name0 = self.transverse_map.name[i_seg] |
| 239 | + m.name1 = self.transverse_map.name[i_seg+1] |
| 240 | + m.beta_x0 = self.transverse_map.beta_x[i_seg] |
| 241 | + m.beta_x1 = self.transverse_map.beta_x[i_seg+1] |
| 242 | + m.beta_y0 = self.transverse_map.beta_y[i_seg] |
| 243 | + m.beta_y1 = self.transverse_map.beta_y[i_seg+1] |
| 244 | + |
| 245 | + # insert transverse map in the ring |
| 246 | + for m in self.transverse_map: |
| 247 | + self.one_turn_map.append(m) |
| 248 | + |
| 249 | + def _contruct_longitudinal_map(self, alpha_mom_compaction=None, longitudinal_mode=None, Q_s=None, |
| 250 | + h_RF=None, V_RF=None, dphi_RF=None, p_increment=None): |
| 251 | + |
| 252 | + # compute the index of the element before which to insert |
| 253 | + # the longitudinal map |
| 254 | + if longitudinal_mode is not None: |
| 255 | + for insert_before, si in enumerate(self.transverse_map.s): |
| 256 | + if si > 0.5 * self.circumference: |
| 257 | + break |
| 258 | + |
| 259 | + if longitudinal_mode == 'linear': |
| 260 | + |
| 261 | + eta = alpha_mom_compaction - self.gamma**-2 |
| 262 | + |
| 263 | + if Q_s == None: |
| 264 | + if p_increment!=0 or dphi_RF!=0: |
| 265 | + raise ValueError('Formula not valid in this case!!!!') |
| 266 | + else: |
| 267 | + Q_s = np.sqrt( e*np.abs(eta)*(h_RF*V_RF) |
| 268 | + / (2*np.pi*self.p0*self.beta*c) ) |
| 269 | + |
| 270 | + self.longitudinal_map = LinearMap( |
| 271 | + np.atleast_1d(alpha_mom_compaction), |
| 272 | + self.circumference, Q_s, |
| 273 | + D_x=self.transverse_map.D_x[insert_before], |
| 274 | + D_y=self.transverse_map.D_y[insert_before]) |
| 275 | + |
| 276 | + |
| 277 | + elif longitudinal_mode == 'non-linear': |
| 278 | + self.longitudinal_map = RFSystems( |
| 279 | + self.circumference, np.atleast_1d(h_RF), |
| 280 | + np.atleast_1d(V_RF), np.atleast_1d(dphi_RF), |
| 281 | + np.atleast_1d(alpha_mom_compaction), self.gamma, p_increment, |
| 282 | + D_x=self.transverse_map.D_x[insert_before], |
| 283 | + D_y=self.transverse_map.D_y[insert_before], |
| 284 | + mass=self.mass, charge=self.charge |
| 285 | + ) |
| 286 | + else: |
| 287 | + raise NotImplementedError( |
| 288 | + 'Something wrong with longitudinal_mode') |
| 289 | + |
| 290 | + self.one_turn_map.insert(insert_before, self.longitudinal_map) |
0 commit comments