44 const int max_iter = 50;
46 double e_previous_step = 0.0;
47 const double tolerance = 5.e-4;
49 for (iter = 0; iter < max_iter; iter++) {
52 if (std::abs(
e_ - e_previous_step) < tolerance) {
55 const double gamma_inv = std::sqrt(1.0 -
v_.
sqr());
61 T_ = T_mub_mus_muq[0];
62 mub_ = T_mub_mus_muq[1];
63 mus_ = T_mub_mus_muq[2];
64 muq_ = T_mub_mus_muq[3];
75 if (iter == max_iter) {
76 std::cout <<
"Warning from solver: max iterations exceeded."
77 <<
" Accuracy: " << std::abs(
e_ - e_previous_step)
78 <<
" is less than tolerance " << tolerance << std::endl;
83 double mus0,
double muq0,
98 return out <<
"T[mu,0]: " << node.
Tmu0() <<
", nb: " << node.
nb()
99 <<
", ns: " << node.
ns() <<
", v: " << node.
v()
100 <<
", e: " << node.
e() <<
", p: " << node.
p()
101 <<
", T: " << node.
T() <<
", mub: " << node.
mub()
102 <<
", mus: " << node.
mus() <<
", muq: " << node.
muq();
106 const std::array<int, 3> n_cells,
107 const std::array<double, 3> origin,
108 bool periodicity,
double e_critical,
109 double t_start,
double delta_t,
111 bool BF_microcanonical)
112 : eos_typelist_(list_eos_particles()),
113 N_sorts_(eos_typelist_.size()),
118 BF_enforce_microcanonical_(BF_microcanonical) {
120 lat_ = std::make_unique<RectangularLattice<ThermLatticeNode>>(
121 lat_sizes, n_cells, origin, periodicity, upd);
122 const std::array<double, 3> abc =
lat_->cell_sizes();
131 bool ignore_cells_under_treshold) {
135 for (
auto &node : *
lat_) {
139 if (!ignore_cells_under_treshold ||
140 node.Tmu0().x0() + std::abs(node.Tmu0().x1()) +
141 std::abs(node.Tmu0().x2()) + std::abs(node.Tmu0().x3()) >=
143 node.compute_rest_frame_quantities(
eos_);
152 +0.5 *
lat_->cell_sizes()[0]),
154 +0.5 *
lat_->cell_sizes()[1]),
156 +0.5 *
lat_->cell_sizes()[2]));
160 ParticleList &plist,
const FourVector required_total_momentum) {
164 required_total_momentum);
170 for (
auto &particle : plist) {
171 particle.set_4momentum(particle.type().mass(),
172 particle.momentum().threevec() + mom_to_add);
181 double E_expected = required_total_momentum.
abs();
182 for (
auto &particle : plist) {
183 particle.boost_momentum(beta_CM_generated);
184 E += particle.momentum().x0();
188 double a, a_min, a_max, er;
189 const int max_iter = 100;
191 if (E_expected >= E) {
199 a = 0.5 * (a_min + a_max);
201 for (
const auto &particle : plist) {
202 const double p2 = particle.momentum().threevec().sqr();
203 const double E2 = particle.momentum().x0() * particle.momentum().x0();
204 E += std::sqrt(E2 + a * (a + 2.0) * p2);
215 }
while (std::abs(er) > tolerance && iter < max_iter);
219 for (
auto &particle : plist) {
220 particle.set_4momentum(particle.type().mass(),
221 (1 + a) * particle.momentum().threevec());
222 particle.boost_momentum(-beta_CM_required);
233 for (
size_t i_type = 0; (i_type <
N_sorts_) && (N_to_sample > 0); i_type++) {
234 if (
get_class(i_type) != particle_class) {
258 const double gamma = 1.0 / std::sqrt(1.0 - cell.
v().
sqr());
259 const double N_this_cell =
267 for (
int i = 0; i <
mult_int_[type_index]; i++) {
270 double partial_sum = 0.0;
271 int index_only_thermalized = -1;
272 while (partial_sum < r) {
273 index_only_thermalized++;
274 partial_sum +=
N_in_cells_[index_only_thermalized];
293 plist.push_back(particle);
298 double time,
int ntest) {
302 const double gamma = 1.0 / std::sqrt(1.0 - cell.
v().
sqr());
303 for (
size_t i = 0; i <
N_sorts_; i++) {
313 for (
size_t i = 0; i <
N_sorts_; i++) {
324 const auto Nbar_antibar = bessel_sampler_B.
sample();
331 for (
size_t i = 0; i <
N_sorts_; i++) {
335 std::pair<int, int> NS_antiS;
341 NS_antiS = bessel_sampler_S.
sample();
343 NS_antiS = std::make_pair(
346 if (NS_antiS.first - NS_antiS.second !=
356 for (
size_t i = 0; i <
N_sorts_; i++) {
360 std::pair<int, int> NC_antiC;
365 conserved_initial.
charge() - ch_sampled);
366 NC_antiC = bessel_sampler_C.
sample();
368 NC_antiC = std::make_pair(
371 if (NC_antiC.first - NC_antiC.second !=
372 conserved_initial.
charge() - ch_sampled) {
383 for (
size_t itype = 0; itype <
N_sorts_; itype++) {
388 const double e_init = conserved_initial.
momentum().
x0();
391 e_tot += particle.momentum().x0();
393 if (std::abs(e_tot - e_init) > 0.01 * e_init) {
395 " too far from ", e_init);
406 int S_plus = 0, S_minus = 0, B_plus = 0, B_minus = 0, E_plus = 0, E_minus = 0;
408 auto condition1 = [](int, int, int) {
return true; };
410 while (conserved_initial.
momentum().
x0() > energy ||
413 energy +=
p.momentum().x0();
414 if (
p.pdgcode().strangeness() > 0) {
416 S_plus +=
p.pdgcode().strangeness();
421 auto condition2 = [](
int S, int, int) {
return (
S < 0); };
423 while (S_plus + S_minus > conserved_initial.
strangeness()) {
425 const int s_part =
p.pdgcode().strangeness();
427 if (S_plus + S_minus + s_part >= conserved_initial.
strangeness()) {
434 auto condition3 = [](
int S, int, int) {
return (
S == 0); };
439 while (conserved_remaining.
momentum().
x0() > energy ||
442 energy +=
p.momentum().x0();
443 if (
p.pdgcode().baryon_number() > 0) {
445 B_plus +=
p.pdgcode().baryon_number();
450 auto condition4 = [](
int S,
int B, int) {
return (
S == 0) && (B < 0); };
452 while (B_plus + B_minus > conserved_remaining.
baryon_number()) {
454 const int bar =
p.pdgcode().baryon_number();
455 if (B_plus + B_minus + bar >= conserved_remaining.
baryon_number()) {
462 auto condition5 = [](
int S,
int B, int) {
return (
S == 0) && (B == 0); };
466 while (conserved_remaining.
momentum().
x0() > energy ||
467 E_plus < conserved_remaining.
charge()) {
469 energy +=
p.momentum().x0();
470 if (
p.pdgcode().charge() > 0) {
472 E_plus +=
p.pdgcode().charge();
477 auto condition6 = [](
int S,
int B,
int C) {
478 return (
S == 0) && (B == 0) && (C < 0);
481 while (E_plus + E_minus > conserved_remaining.
charge()) {
483 const int charge =
p.pdgcode().charge();
484 if (E_plus + E_minus + charge >= conserved_remaining.
charge()) {
491 auto condition7 = [](
int S,
int B,
int C) {
492 return (
S == 0) && (B == 0) && (C == 0);
497 while (conserved_remaining.
momentum().
x0() > energy) {
500 energy +=
p.momentum().x0();
514 for (
auto &particle : particles) {
515 const bool is_on_lattice =
516 lat_->value_at(particle.position().threevec(), node);
517 if (is_on_lattice && node.
e() >
e_crit_) {
539 const size_t lattice_total_cells =
lat_->size();
540 for (
size_t i = 0; i < lattice_total_cells; i++) {
546 "Number of cells in the thermalization region = ",
549 ", in % of lattice: ",
561 throw std::invalid_argument(
562 "This thermalization algorithm is"
563 " not yet implemented");
582 struct to_average on_lattice = {0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0};
583 struct to_average in_therm_reg = {0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0};
584 double e_sum_on_lattice = 0.0, e_sum_in_therm_reg = 0.0;
585 int node_counter = 0;
586 for (
const auto &node : *
lat_) {
587 const double e = node.e();
588 on_lattice.T += node.T() * e;
589 on_lattice.mub += node.mub() * e;
590 on_lattice.mus += node.mus() * e;
591 on_lattice.muq += node.muq() * e;
592 on_lattice.nb += node.nb() * e;
593 on_lattice.ns += node.ns() * e;
594 on_lattice.nq += node.nq() * e;
595 e_sum_on_lattice += e;
597 in_therm_reg.T += node.T() * e;
598 in_therm_reg.mub += node.mub() * e;
599 in_therm_reg.mus += node.mus() * e;
600 in_therm_reg.muq += node.muq() * e;
601 in_therm_reg.nb += node.nb() * e;
602 in_therm_reg.ns += node.ns() * e;
603 in_therm_reg.nq += node.nq() * e;
604 e_sum_in_therm_reg += e;
609 on_lattice.T /= e_sum_on_lattice;
610 on_lattice.mub /= e_sum_on_lattice;
611 on_lattice.mus /= e_sum_on_lattice;
612 on_lattice.muq /= e_sum_on_lattice;
613 on_lattice.nb /= e_sum_on_lattice;
614 on_lattice.ns /= e_sum_on_lattice;
615 on_lattice.nq /= e_sum_on_lattice;
618 in_therm_reg.T /= e_sum_in_therm_reg;
619 in_therm_reg.mub /= e_sum_in_therm_reg;
620 in_therm_reg.mus /= e_sum_in_therm_reg;
621 in_therm_reg.muq /= e_sum_in_therm_reg;
622 in_therm_reg.nb /= e_sum_in_therm_reg;
623 in_therm_reg.ns /= e_sum_in_therm_reg;
624 in_therm_reg.nq /= e_sum_in_therm_reg;
627 std::cout <<
"Current time [fm]: " << clock.
current_time() << std::endl;
628 std::cout <<
"Averages on the lattice - T[GeV], mub[GeV], mus[GeV], muq[GeV] "
629 <<
"nb[fm^-3], ns[fm^-3], nq[fm^-3]: " << on_lattice.T <<
" "
630 << on_lattice.mub <<
" " << on_lattice.mus <<
" " << on_lattice.muq
631 <<
" " << on_lattice.nb <<
" " << on_lattice.ns <<
" "
632 << on_lattice.nq << std::endl;
634 <<
"Averages in therm. region - T[GeV], mub[GeV], mus[GeV], muq[GeV] "
635 <<
"nb[fm^-3], ns[fm^-3], nq[fm^-3]: " << in_therm_reg.T <<
" "
636 << in_therm_reg.mub <<
" " << in_therm_reg.mus <<
" " << in_therm_reg.muq
637 <<
" " << in_therm_reg.nb <<
" " << in_therm_reg.ns <<
" "
638 << in_therm_reg.nq << std::endl;
639 std::cout <<
"Volume with e > e_crit [fm^3]: "
Angles provides a common interface for generating directions: i.e., two angles that should be interpr...
ThreeVector threevec() const
void distribute_isotropically()
Populate the object with a new direction.
Clock tracks the time in the simulation.
virtual double current_time() const =0
A class to pre-calculate and store parameters relevant for density calculation.
The FourVector class holds relevant values in Minkowski spacetime with (+, −, −, −) metric signature.
double abs() const
calculate the lorentz invariant absolute value
ThreeVector threevec() const
ThreeVector velocity() const
Get the velocity (3-vector divided by zero component).
void thermalize_BF_algo(QuantumNumbers &conserved_initial, double time, int ntest)
Samples particles according to the BF algorithm by making use of the.
void compute_N_in_cells_mode_algo(F &&condition)
Computes average number of particles in each cell for the mode algorithm.
void print_statistics(const Clock &clock) const
Generates standard output with information about the thermodynamic properties of the lattice,...
std::vector< double > mult_sort_
Real number multiplicity for each particle type.
std::vector< int > mult_int_
Integer multiplicity for each particle type.
HadronGasEos eos_
Hadron gas equation of state.
ParticleList to_remove_
Particles to be removed after this thermalization step.
const bool BF_enforce_microcanonical_
Enforce energy conservation as part of BF sampling algorithm or not.
ThreeVector uniform_in_cell() const
std::unique_ptr< RectangularLattice< ThermLatticeNode > > lat_
The lattice on which the thermodynamic quantities are calculated.
double mult_class(const HadronClass cl) const
HadronClass get_class(size_t typelist_index) const
Defines the class of hadrons by quantum numbers.
GrandCanThermalizer(const std::array< double, 3 > lat_sizes, const std::array< int, 3 > n_cells, const std::array< double, 3 > origin, bool periodicity, double e_critical, double t_start, double delta_t, ThermalizationAlgorithm algo, bool BF_microcanonical)
Default constructor for the GranCanThermalizer to allocate the lattice.
const double e_crit_
Critical energy density above which cells are thermalized.
ParticleList sampled_list_
Newly generated particles by thermalizer.
std::array< double, 7 > mult_classes_
The different hadron species according to the enum defined in.
void sample_in_random_cell_BF_algo(ParticleList &plist, const double time, size_t type_index)
The total number of particles of species type_index is defined by mult_int_ array that is returned by...
void update_thermalizer_lattice(const std::vector< Particles > &ensembles, const DensityParameters &par, bool ignore_cells_under_threshold=true)
Compute all the thermodynamical quantities on the lattice from particles.
std::vector< size_t > cells_to_sample_
Cells above critical energy density.
const ThermalizationAlgorithm algorithm_
Algorithm to choose for sampling of particles obeying conservation laws.
const ParticleTypePtrList eos_typelist_
List of particle types from which equation of state is computed.
std::vector< double > N_in_cells_
Number of particles to be sampled in one cell.
ParticleData sample_in_random_cell_mode_algo(const double time, F &&condition)
Samples one particle and the species, cell, momentum and coordinate are chosen from the corresponding...
double lat_cell_volume_
Volume of a single lattice cell, necessary to convert thermal densities to actual particle numbers.
const size_t N_sorts_
Number of different species to be sampled.
double N_total_in_cells_
Total number of particles over all cells in thermalization region.
void thermalize(const Particles &particles, double time, int ntest)
Main thermalize function, that chooses the algorithm to follow (BF or mode sampling).
void renormalize_momenta(ParticleList &plist, const FourVector required_total_momentum)
Changes energy and momenta of the particles in plist to match the required_total_momentum.
void sample_multinomial(HadronClass particle_class, int N)
The sample_multinomial function samples integer numbers n_i distributed according to the multinomial ...
void thermalize_mode_algo(QuantumNumbers &conserved_initial, double time)
Samples particles to the according to the mode algorithm.
Class to handle the equation of state (EoS) of the hadron gas, consisting of all hadrons included in ...
static double net_charge_density(double T, double mub, double mus, double muq, bool account_for_resonance_widths=false)
Compute net charge density.
static double partial_density(const ParticleType &ptype, double T, double mub, double mus, double muq, bool account_for_resonance_widths=false)
Compute partial density of one hadron sort.
std::array< double, 4 > solve_eos(double e, double nb, double ns, double nq, std::array< double, 4 > initial_approximation)
Compute temperature and chemical potentials given energy-, net baryon-, net strangeness- and net char...
bool is_tabulated() const
Create an EoS table or not?
void from_table(EosTable::table_element &res, double e, double nb, double nq) const
Get the element of eos table.
static double net_baryon_density(double T, double mub, double mus, double muq, bool account_for_resonance_widths=false)
Compute net baryon density.
static double energy_density(double T, double mub, double mus, double muq)
Compute energy density.
static double net_strange_density(double T, double mub, double mus, double muq, bool account_for_resonance_widths=false)
Compute net strangeness density.
static double pressure(double T, double mub, double mus, double muq, bool account_for_resonance_widths=false)
Compute pressure .
ParticleData contains the dynamic information of a certain particle.
void set_4momentum(const FourVector &momentum_vector)
Set the particle's 4-momentum directly.
const ParticleType & type() const
Get the type of the particle.
void set_4position(const FourVector &pos)
Set the particle's 4-position directly.
const FourVector & momentum() const
Get the particle's 4-momentum.
void set_formation_time(const double &form_time)
Set the absolute formation time.
void boost_momentum(const ThreeVector &v)
Apply a Lorentz-boost to only the momentum.
int32_t charge() const
The charge of the particle.
int baryon_number() const
The Particles class abstracts the storage and manipulation of particles.
A container for storing conserved values.
void add_values(const ParticleData &p)
Add the quantum numbers of a single particle to the collection.
int baryon_number() const
FourVector momentum() const
The ThermLatticeNode class is intended to compute thermodynamical quantities in a cell given a set of...
void compute_rest_frame_quantities(HadronGasEos &eos)
Temperature, chemical potentials and rest frame velocity are calculated given the hadron gas equation...
double muq() const
Get the net charge chemical potential.
FourVector Tmu0() const
Get Four-momentum flow of the cell.
double ns_
Net strangeness density of the cell in the computational frame.
void set_rest_frame_quantities(double T0, double mub0, double mus0, double muq0, const ThreeVector v0)
Set all the rest frame quantities to some values, this is useful for testing.
double p() const
Get pressure in the rest frame.
double p_
Pressure in the rest frame.
double mus_
Net strangeness chemical potential.
ThreeVector v() const
Get 3-velocity of the rest frame.
double ns() const
Get net strangeness density of the cell in the computational frame.
double e_
Energy density in the rest frame.
double nb_
Net baryon density of the cell in the computational frame.
void add_particle(const ParticleData &p, double factor)
Add particle contribution to Tmu0, nb, ns and nq May look like unused at first glance,...
double muq_
Net charge chemical potential.
double mub_
Net baryon chemical potential.
FourVector Tmu0_
Four-momentum flow of the cell.
double mus() const
Get the net strangeness chemical potential.
ThreeVector v_
Velocity of the rest frame.
ThermLatticeNode()
Default constructor of thermal quantities on the lattice returning thermodynamic quantities in comput...
double mub() const
Get the net baryon chemical potential.
double nb() const
Get net baryon density of the cell in the computational frame.
double nq_
Net charge density of the cell in the computational frame.
double e() const
Get energy density in the rest frame.
double T() const
Get the temperature.
The ThreeVector class represents a physical three-vector with the components .
The intention of this class is to efficiently sample from the Bessel distribution ,...
std::pair< int, int > sample()
Sample two numbers from given Poissonians with a fixed difference.
ThermalizationAlgorithm
Defines the algorithm used for the forced thermalization.
std::ostream & operator<<(std::ostream &out, const ActionPtr &action)
Convenience: dereferences the ActionPtr to Action.
std::array< einhard::Logger<>, std::tuple_size< LogArea::AreaTuple >::value > logg
An array that stores all pre-configured Logger objects.
int poisson(const T &lam)
Returns a Poisson distributed random number.
int binomial(const int N, const T &p)
Returns a binomially distributed random number.
double sample_momenta_from_thermal(const double temperature, const double mass)
Samples a momentum from the Maxwell-Boltzmann (thermal) distribution in a faster way,...
void update_lattice(RectangularLattice< T > *lat, const LatticeUpdate update, const DensityType dens_type, const DensityParameters &par, const std::vector< Particles > &ensembles, const bool compute_gradient)
Updates the contents on the lattice.
static constexpr int LGrandcanThermalizer
LatticeUpdate
Enumerator option for lattice updates.
constexpr double really_small
Numerical error tolerance.
DensityType
Allows to choose which kind of density to calculate.
HadronClass
Specifier to classify the different hadron species according to their quantum numbers.
@ Antibaryon
All anti-baryons.
@ ZeroQZeroSMeson
Neutral non-strange mesons.
@ NegativeSMeson
Mesons with strangeness S < 0.
@ NegativeQZeroSMeson
Non-strange mesons (S = 0) with electric cherge Q < 0.
@ PositiveSMeson
Mesons with strangeness S > 0.
@ PositiveQZeroSMeson
Non-strange mesons (S = 0) with electric cherge Q > 0.
Define the data structure for one element of the table.
double mub
Net baryochemical potential.
double muq
Net charge chemical potential.
double mus
Net strangeness potential.