Source code for pysph.sph.surface_tension

"""
Implementation of the equations used for surface tension modelling,
for example in KHI simulations. The references are as under:

 - M. Shadloo, M. Yildiz, "Numerical modelling of Kelvin-Helmholtz
   isntability using smoothed particle hydrodynamics", IJNME, 2011,
   87, pp 988--1006 [SY11]

 - Joseph P. Morris "Simulating surface tension with smoothed particle
   hydrodynamics", JCP, 2000, 33, pp 333--353 [JM00]

 - Adami et al. "A new surface-tension formulation for multi-phase SPH
   using a reproducing divergence approximation", JCP 2010, 229, pp
   5011--5021 [A10]

"""
from pysph.sph.equation import Equation

from math import sqrt

[docs]class SmoothedColor(Equation): r"""Smoothed color function. Eq. (17) in [JM00] .. math:: c_a = \sum_b \frac{m_b}{\rho_b} c_b^i \nabla_a W_{ab}\,, where, :math:`c_b^i` is the color index associated with a particle. """ def __init__(self, dest, sources, smooth=False): self.smooth = smooth super(SmoothedColor, self).__init__(dest, sources)
[docs] def initialize(self, d_idx, d_scolor): d_scolor[d_idx] = 0.0
[docs] def loop(self, d_idx, s_idx, d_rho, s_rho, s_m, s_color, d_scolor, WIJ): # Smoothed color Eq. (17) in [JM00] d_scolor[d_idx] += s_m[s_idx]/s_rho[s_idx] * s_color[s_idx] * WIJ
[docs] def post_loop(self, d_idx, d_color, d_scolor): # overwrite the smoothed color if not needed if not self.smooth: d_scolor[d_idx] = d_color[d_idx]
[docs]class ColorGradientUsingNumberDensity(Equation): r"""Gradient of the color function using Eq. (13) of [SY11]: .. math:: \nabla C_a = \sum_b \frac{2 C_b - C_a}{\psi_a + \psi_a} \nabla_{a} W_{ab} Using the gradient of the color function, the normal and discretized dirac delta is calculated in the post loop. Singularities are avoided as per the recommendation by [JM00] (see eqs 20 & 21) using the parameter :math:`\epsilon` """ def __init__(self, dest, sources, epsilon=1e-6): self.epsilon2 = epsilon*epsilon super(ColorGradientUsingNumberDensity, self).__init__(dest, sources)
[docs] def initialize(self, d_idx, d_cx, d_cy, d_cz, d_nx, d_ny, d_nz, d_ddelta, d_N): # color gradient d_cx[d_idx] = 0.0 d_cy[d_idx] = 0.0 d_cz[d_idx] = 0.0 # interface normals d_nx[d_idx] = 0.0 d_ny[d_idx] = 0.0 d_nz[d_idx] = 0.0 # discretized dirac delta d_ddelta[d_idx] = 0.0 # reliability indicator for normals d_N[d_idx] = 0.0
[docs] def loop(self, d_idx, s_idx, d_scolor, s_scolor, d_cx, d_cy, d_cz, d_V, s_V, DWIJ): # average particle volume psiab1 = 2.0/( d_V[d_idx] + s_V[s_idx] ) # difference in color divided by psiab. Eq. (13) in [SY11] Cba = (s_scolor[s_idx] - d_scolor[d_idx]) * psiab1 # color gradient d_cx[d_idx] += Cba * DWIJ[0] d_cy[d_idx] += Cba * DWIJ[1] d_cz[d_idx] += Cba * DWIJ[2]
[docs] def post_loop(self, d_idx, d_cx, d_cy, d_cz, d_nx, d_ny, d_nz, d_N, d_ddelta): # absolute value of the color gradient mod_gradc2 = d_cx[d_idx]*d_cx[d_idx] + \ d_cy[d_idx]*d_cy[d_idx] + \ d_cz[d_idx]*d_cz[d_idx] # avoid sqrt computations on non-interface particles # (particles for which the color gradient is zero) Eq. (19, # 20) in [JM00] if mod_gradc2 > self.epsilon2: # this normal is reliable in the sense of [JM00] d_N[d_idx] = 1.0 # compute the normals mod_gradc = 1./sqrt( mod_gradc2 ) d_nx[d_idx] = d_cx[d_idx] * mod_gradc d_ny[d_idx] = d_cy[d_idx] * mod_gradc d_nz[d_idx] = d_cz[d_idx] * mod_gradc # discretized Dirac Delta function d_ddelta[d_idx] = 1./mod_gradc
[docs]class MorrisColorGradient(Equation): r"""Gradient of the color function using Eq. (17) of [JM00]: .. math:: \nabla c_a = \sum_b \frac{m_b}{\rho_b}(c_b - c_a) \nabla_{a} W_{ab}\,, where a smoothed representation of the color is used in the equation. Using the gradient of the color function, the normal and discretized dirac delta is calculated in the post loop. Singularities are avoided as per the recommendation by [JM00] (see eqs 20 & 21) using the parameter :math:`\epsilon` """ def __init__(self, dest, sources, epsilon=1e-6): self.epsilon2 = epsilon*epsilon super(MorrisColorGradient, self).__init__(dest, sources)
[docs] def initialize(self, d_idx, d_cx, d_cy, d_cz, d_nx, d_ny, d_nz, d_ddelta, d_N): # color gradient d_cx[d_idx] = 0.0 d_cy[d_idx] = 0.0 d_cz[d_idx] = 0.0 # interface normals d_nx[d_idx] = 0.0 d_ny[d_idx] = 0.0 d_nz[d_idx] = 0.0 # reliability indicator for normals and dirac delta d_N[d_idx] = 0.0 d_ddelta[d_idx] = 0.0
[docs] def loop(self, d_idx, s_idx, d_scolor, s_scolor, d_cx, d_cy, d_cz, s_m, s_rho, DWIJ): # Eq. (17) in [JM00] Cba = (s_scolor[s_idx] - d_scolor[d_idx]) * s_m[s_idx]/s_rho[s_idx] # color gradient d_cx[d_idx] += Cba * DWIJ[0] d_cy[d_idx] += Cba * DWIJ[1] d_cz[d_idx] += Cba * DWIJ[2]
[docs] def post_loop(self, d_idx, d_cx, d_cy, d_cz, d_nx, d_ny, d_nz, d_N, d_ddelta): # absolute value of the color gradient mod_gradc2 = d_cx[d_idx]*d_cx[d_idx] + \ d_cy[d_idx]*d_cy[d_idx] + \ d_cz[d_idx]*d_cz[d_idx] # avoid sqrt computations on non-interface particles # (particles for which the color gradient is zero) Eq. (19, # 20) in [JM00] if mod_gradc2 > self.epsilon2: # this normal is reliable in the sense of [JM00] d_N[d_idx] = 1.0 # compute the normals mod_gradc = 1./sqrt( mod_gradc2 ) d_nx[d_idx] = d_cx[d_idx] * mod_gradc d_ny[d_idx] = d_cy[d_idx] * mod_gradc d_nz[d_idx] = d_cz[d_idx] * mod_gradc # discretized Dirac Delta function d_ddelta[d_idx] = 1./mod_gradc
[docs]class SY11ColorGradient(Equation): r"""Gradient of the color function using Eq. (13) of [SY11]: .. math:: \nabla C_a = \sum_b \frac{2 C_b - C_a}{\psi_a + \psi_a} \nabla_{a} W_{ab} Using the gradient of the color function, the normal and discretized dirac delta is calculated in the post loop. """ def __init__(self, dest, sources, epsilon=1e-6): self.epsilon2 = epsilon*epsilon super(SY11ColorGradient, self).__init__(dest, sources)
[docs] def initialize(self, d_idx, d_cx, d_cy, d_cz, d_nx, d_ny, d_nz, d_ddelta, d_N): # color gradient d_cx[d_idx] = 0.0 d_cy[d_idx] = 0.0 d_cz[d_idx] = 0.0 # interface normals d_nx[d_idx] = 0.0 d_ny[d_idx] = 0.0 d_nz[d_idx] = 0.0 # discretized dirac delta d_ddelta[d_idx] = 0.0 # reliability indicator for normals d_N[d_idx] = 0.0
[docs] def loop(self, d_idx, s_idx, d_color, s_color, d_cx, d_cy, d_cz, d_V, s_V, DWIJ): # average particle volume psiab1 = 2.0/( d_V[d_idx] + s_V[s_idx] ) # difference in color divided by psiab. Eq. (13) in [SY11] Cba = (s_color[s_idx] - d_color[d_idx]) * psiab1 # color gradient d_cx[d_idx] += Cba * DWIJ[0] d_cy[d_idx] += Cba * DWIJ[1] d_cz[d_idx] += Cba * DWIJ[2]
[docs] def post_loop(self, d_idx, d_cx, d_cy, d_cz, d_nx, d_ny, d_nz, d_N, d_ddelta): # absolute value of the color gradient mod_gradc2 = d_cx[d_idx]*d_cx[d_idx] + \ d_cy[d_idx]*d_cy[d_idx] + \ d_cz[d_idx]*d_cz[d_idx] # avoid sqrt computations on non-interface particles if mod_gradc2 > self.epsilon2: # this normal is reliable in the sense of [JM00] d_N[d_idx] = 1.0 # compute the normals mod_gradc = 1./sqrt( mod_gradc2 ) d_nx[d_idx] = d_cx[d_idx] * mod_gradc d_ny[d_idx] = d_cy[d_idx] * mod_gradc d_nz[d_idx] = d_cz[d_idx] * mod_gradc # discretized Dirac Delta function d_ddelta[d_idx] = 1./mod_gradc
[docs]class SY11DiracDelta(Equation): r"""Discretized dirac-delta for the SY11 formulation Eq. (14) in [SY11] This is essentially the same as computing the color gradient, the only difference being that this might be called with a reduced smoothing length. Note that the normals should be computed using the SY11ColorGradient equation. This function will effectively overwrite the color gradient. """ def __init__(self, dest, sources, epsilon=1e-6): self.epsilon2 = epsilon*epsilon super(SY11DiracDelta, self).__init__(dest, sources)
[docs] def initialize(self, d_idx, d_cx, d_cy, d_cz, d_ddelta): # color gradient d_cx[d_idx] = 0.0 d_cy[d_idx] = 0.0 d_cz[d_idx] = 0.0 # discretized dirac delta d_ddelta[d_idx] = 0.0
[docs] def loop(self, d_idx, s_idx, d_color, s_color, d_cx, d_cy, d_cz, d_V, s_V, DWIJ): # average particle volume psiab1 = 2.0/( d_V[d_idx] + s_V[s_idx] ) # difference in color divided by psiab. Eq. (13) in [SY11] Cba = (s_color[s_idx] - d_color[d_idx]) * psiab1 # color gradient d_cx[d_idx] += Cba * DWIJ[0] d_cy[d_idx] += Cba * DWIJ[1] d_cz[d_idx] += Cba * DWIJ[2]
[docs] def post_loop(self, d_idx, d_cx, d_cy, d_cz, d_nx, d_ny, d_nz, d_N, d_ddelta): # absolute value of the color gradient mod_gradc2 = d_cx[d_idx]*d_cx[d_idx] + \ d_cy[d_idx]*d_cy[d_idx] + \ d_cz[d_idx]*d_cz[d_idx] # avoid sqrt computations on non-interface particles if mod_gradc2 > self.epsilon2: mod_gradc = sqrt( mod_gradc2 ) # discretized Dirac Delta function d_ddelta[d_idx] = mod_gradc
[docs]class InterfaceCurvatureFromNumberDensity(Equation): r"""Interface curvature using number density. Eq. (15) in [SY11]: .. math:: \kappa_a = \sum_b \frac{2.0}{\psi_a + \psi_b} \left(\boldsymbol{n_a} - \boldsymbol{n_b}\right) \cdot \nabla_a W_{ab} """ def __init__(self, dest, sources, with_morris_correction=True): self.with_morris_correction = with_morris_correction super(InterfaceCurvatureFromNumberDensity,self).__init__(dest, sources)
[docs] def initialize(self, d_idx, d_kappa, d_wij_sum): d_kappa[d_idx] = 0.0 d_wij_sum[d_idx] = 0.0
[docs] def loop(self, d_idx, s_idx, d_kappa, d_nx, d_ny, d_nz, s_nx, s_ny, s_nz, d_V, s_V, d_N, s_N, d_wij_sum, s_rho, s_m, WIJ, DWIJ): nijdotdwij = (d_nx[d_idx] - s_nx[s_idx]) * DWIJ[0] + \ (d_ny[d_idx] - s_ny[s_idx]) * DWIJ[1] + \ (d_nz[d_idx] - s_nz[s_idx]) * DWIJ[2] # averaged particle number density psiij1 = 2.0/(d_V[d_idx] + s_V[s_idx]) # local number density with reliable normals Eq. (24) in [JM00] tmp = 1.0 if self.with_morris_correction: tmp = min(d_N[d_idx], s_N[s_idx]) d_wij_sum[d_idx] += tmp * s_m[s_idx]/s_rho[s_idx] * WIJ # Eq. (15) in [SY11] with correction Eq. (22) in [JM00] d_kappa[d_idx] += tmp * psiij1 * nijdotdwij
[docs] def post_loop(self, d_idx, d_wij_sum, d_nx, d_kappa): # correct the curvature estimate. Eq. (23) in [JM00] if self.with_morris_correction: if d_wij_sum[d_idx] > 1e-12: d_kappa[d_idx] /= d_wij_sum[d_idx]
[docs]class ShadlooYildizSurfaceTensionForce(Equation): r"""Acceleration due to surface tension force Eq. (7,9) in [SY11]: .. math: \frac{d\boldsymbol{v}_a} = \frac{1}{m_a} \sigma \kappa_a \boldsymbol{n}_a \delta_a^s\,, where, :math:`\delta^s` is the discretized dirac delta function, :math:`\boldsymbol{n}` is the interface normal, :math:`\kappa` is the discretized interface curvature and :math:`\sigma` is the surface tension force constant. """ def __init__(self, dest, sources, sigma=0.1): self.sigma = sigma # base class initialization super(ShadlooYildizSurfaceTensionForce, self).__init__(dest, sources)
[docs] def initialize(self, d_idx, d_au, d_av, d_aw): d_au[d_idx] = 0.0 d_av[d_idx] = 0.0 d_aw[d_idx] = 0.0
[docs] def loop(self, d_idx, d_au, d_av, d_aw, d_kappa, d_nx, d_ny, d_nz, d_m, d_rho, d_ddelta): mi = 1./d_m[d_idx] rhoi = 1./d_rho[d_idx] # acceleration per uint mass term Eq. (7) in [SY11] tmp = self.sigma * d_kappa[d_idx] * d_ddelta[d_idx] * rhoi d_au[d_idx] += tmp * d_nx[d_idx] d_av[d_idx] += tmp * d_ny[d_idx] d_aw[d_idx] += tmp * d_nz[d_idx]
[docs]class CSFSurfaceTensionForce(Equation): r"""Acceleration due to surface tension force Eq. (25) in [JM00]: .. math: \frac{d\boldsymbol{v}_a}{dt} = \frac{1}{rho_a} \sigma_b \kappa_a \boldsymbol{n}_a Note that as per Eq. (17) in [JM00], the un-normalized normal is basically the gradient of the color function. The acceleration term therefore depends on the gradient of the color field. """ def __init__(self, dest, sources, sigma=0.1): self.sigma = sigma # base class initialization super(CSFSurfaceTensionForce, self).__init__(dest, sources)
[docs] def initialize(self, d_idx, d_au, d_av, d_aw): d_au[d_idx] = 0.0 d_av[d_idx] = 0.0 d_aw[d_idx] = 0.0
[docs] def loop(self, d_idx, d_au, d_av, d_aw, d_kappa, d_cx, d_cy, d_cz, d_rho): rhoi = 1./d_rho[d_idx] # acceleration per uint mass term Eq. (25) in [JM00] tmp = self.sigma * d_kappa[d_idx] * rhoi d_au[d_idx] += tmp * d_cx[d_idx] d_av[d_idx] += tmp * d_cy[d_idx] d_aw[d_idx] += tmp * d_cz[d_idx]
[docs]class AdamiColorGradient(Equation): r"""Gradient of color Eq. (14) in [A10] .. math:: \nabla c_a = \frac{1}{V_a}\sum_b \left[V_a^2 + V_b^2 \right]\tilde{c}_{ab}\nabla_a W_{ab}\,, where, the average :math:`\tilde{c}_{ab}` is defined as .. math:: \tilde{c}_{ab} = \frac{\rho_b}{\rho_a + \rho_b}c_a + \frac{\rho_a}{\rho_a + \rho_b}c_b """
[docs] def initialize(self, d_idx, d_cx, d_cy, d_cz, d_nx, d_ny, d_nz, d_ddelta, d_N): d_cx[d_idx] = 0.0 d_cy[d_idx] = 0.0 d_cz[d_idx] = 0.0 d_nx[d_idx] = 0.0 d_ny[d_idx] = 0.0 d_nz[d_idx] = 0.0 # reliability indicator for normals d_N[d_idx] = 0.0 # Discretized dirac-delta d_ddelta[d_idx] = 0.0
[docs] def loop(self, d_idx, s_idx, d_V, s_V, d_rho, s_rho, d_cx, d_cy, d_cz, d_color, s_color, DWIJ): # particle volumes Vi = 1./d_V[d_idx]; Vj = 1./s_V[s_idx] Vi2 = Vi * Vi; Vj2 = Vj * Vj # averaged particle color rhoi = d_rho[d_idx] rhoj = s_rho[s_idx] rhoij1 = 1./(rhoi + rhoj) # Eq. (15) in [A10] cij = rhoj*rhoij1*d_color[d_idx] + rhoi*rhoij1*s_color[s_idx] # comute the gradient tmp = cij * (Vi2 + Vj2)/Vi d_cx[d_idx] += tmp * DWIJ[0] d_cy[d_idx] += tmp * DWIJ[1] d_cz[d_idx] += tmp * DWIJ[2]
[docs] def post_loop(self, d_idx, d_cx, d_cy, d_cz, d_nx, d_ny, d_nz, d_ddelta, d_N): # absolute value of the color gradient mod_gradc2 = d_cx[d_idx]*d_cx[d_idx] + \ d_cy[d_idx]*d_cy[d_idx] + \ d_cz[d_idx]*d_cz[d_idx] # avoid sqrt computations on non-interface particles if mod_gradc2 > 1e-6: # this normal is reliable in the sense of [JM00] d_N[d_idx] = 1.0 # compute the normals mod_gradc = 1./sqrt( mod_gradc2 ) d_nx[d_idx] = d_cx[d_idx] * mod_gradc d_ny[d_idx] = d_cy[d_idx] * mod_gradc d_nz[d_idx] = d_cz[d_idx] * mod_gradc # discretized dirac delta d_ddelta[d_idx] = 1./mod_gradc
# FIXME: The implementation based on the formulation presented in # [A10] seems to be incorrect.
[docs]class AdamiReproducingDivergence(Equation): r"""Reproducing divergence approximation Eq. (20) in [A10] to compute the curvature .. math:: \nabla \cdot \boldsymbol{\phi}_a = d\frac{\sum_b \boldsymbol{\phi}_{ab}\cdot \nabla_a W_{ab}V_b}{\sum_b\boldsymbol{x}_{ab}\cdot \nabla_a W_{ab} V_b} """ def __init__(self, dest, sources, dim): self.dim = dim super(AdamiReproducingDivergence,self).__init__(dest, sources)
[docs] def initialize(self, d_idx, d_kappa, d_wij_sum): d_kappa[d_idx] = 0.0 d_wij_sum[d_idx] = 0.0
[docs] def loop(self, d_idx, s_idx, d_kappa, d_wij_sum, d_nx, d_ny, d_nz, s_nx, s_ny, s_nz, d_V, s_V, DWIJ, XIJ, RIJ, EPS): # particle volumes Vi = 1./d_V[d_idx]; Vj = 1./s_V[s_idx] # dot product in the numerator of Eq. (20) nijdotdwij = (d_nx[d_idx] - s_nx[s_idx]) * DWIJ[0] + \ (d_ny[d_idx] - s_ny[s_idx]) * DWIJ[1] + \ (d_nz[d_idx] - s_nz[s_idx]) * DWIJ[2] # dot product in the denominator of Eq. (20) xijdotdwij = XIJ[0]*DWIJ[0] + XIJ[1]*DWIJ[1] + XIJ[2]*DWIJ[2] xijdotdwij /= (RIJ + EPS) # accumulate the contributions d_kappa[d_idx] += nijdotdwij * Vj d_wij_sum[d_idx] += RIJ * xijdotdwij * Vj
[docs] def post_loop(self, d_idx, d_kappa, d_wij_sum): # normalize the curvature estimate d_kappa[d_idx] /= d_wij_sum[d_idx] d_kappa[d_idx] *= self.dim