mirror of
https://gitlab.rlp.net/mobitar/ReCo.jl.git
synced 2024-11-12 22:40:44 +00:00
241 lines
No EOL
6.3 KiB
Julia
241 lines
No EOL
6.3 KiB
Julia
using CairoMakie
|
|
using LaTeXStrings: @L_str
|
|
using StaticArrays: SVector
|
|
using JLD2: JLD2
|
|
using Dates: Dates
|
|
using CSV: CSV
|
|
using DataFrames: DataFrames
|
|
using Random: Random
|
|
|
|
using ReCo: ReCo
|
|
|
|
includet("../../src/Visualization/common_CairoMakie.jl")
|
|
|
|
function radial_distribution_simulation(;
|
|
n_particles::Int64, v₀s::NTuple{N,Float64}, T::Float64, packing_ratio::Float64
|
|
) where {N}
|
|
Random.seed!(42)
|
|
|
|
n_v₀s = length(v₀s)
|
|
|
|
sim_dirs = Vector{String}(undef, n_v₀s)
|
|
|
|
parent_dir = "radial_distribution_$(Dates.now())"
|
|
|
|
Threads.@threads for v₀_ind in 1:n_v₀s
|
|
v₀ = v₀s[v₀_ind]
|
|
|
|
dir = ReCo.init_sim(;
|
|
n_particles=n_particles,
|
|
v₀=v₀,
|
|
packing_ratio=packing_ratio,
|
|
parent_dir=parent_dir,
|
|
comment="$v₀",
|
|
)
|
|
|
|
ReCo.run_sim(dir; duration=T, seed=v₀_ind)
|
|
|
|
sim_dirs[v₀_ind] = dir
|
|
end
|
|
|
|
return sim_dirs
|
|
end
|
|
|
|
function circular_shell_volume(lower_radius, Δradius)
|
|
return π * (2 * lower_radius * Δradius + Δradius^2)
|
|
end
|
|
|
|
function radial_distribution(;
|
|
sim_dirs::Vector{String}, n_radii::Int64, n_last_snapshots::Int64, n_particles::Int64
|
|
)
|
|
sim_consts = ReCo.load_sim_consts(sim_dirs[1])
|
|
|
|
particle_radius = sim_consts.particle_radius
|
|
|
|
min_lower_radius = 0.0
|
|
max_lower_radius = 5 * (2 * particle_radius)
|
|
Δradius = (max_lower_radius - min_lower_radius) / n_radii
|
|
|
|
lower_radii = LinRange(min_lower_radius, max_lower_radius, n_radii)
|
|
|
|
box_volume = (2 * sim_consts.half_box_len)^2
|
|
volume_per_particle = box_volume / n_particles
|
|
|
|
n_sim_dirs = length(sim_dirs)
|
|
|
|
gs = Vector{Vector{Float64}}(undef, n_sim_dirs)
|
|
|
|
Threads.@threads for sim_dir_ind in 1:n_sim_dirs
|
|
sim_dir = sim_dirs[sim_dir_ind]
|
|
|
|
cs = Matrix{SVector{2,Float64}}(undef, (n_particles, n_last_snapshots))
|
|
|
|
bundle_paths = ReCo.sorted_bundle_paths(sim_dir; rev=true)
|
|
|
|
snapshot_conunter = 0
|
|
break_bundle_path_loop = false
|
|
|
|
for bundle_path in bundle_paths
|
|
bundle::ReCo.Bundle = JLD2.load_object(bundle_path)
|
|
|
|
for snapshot_ind in (bundle.n_snapshots):-1:1
|
|
snapshot_conunter += 1
|
|
|
|
@simd for particle_ind in 1:n_particles
|
|
cs[particle_ind, snapshot_conunter] = bundle.c[
|
|
particle_ind, snapshot_ind
|
|
]
|
|
end
|
|
|
|
if snapshot_conunter == n_last_snapshots
|
|
break_bundle_path_loop = true
|
|
break
|
|
end
|
|
end
|
|
|
|
if break_bundle_path_loop
|
|
break
|
|
end
|
|
end
|
|
|
|
if snapshot_conunter != n_last_snapshots
|
|
error("snapshot_conunter != n_last_snapshots")
|
|
end
|
|
|
|
g = zeros(n_radii)
|
|
|
|
for snapshot_ind in 1:n_last_snapshots
|
|
for p1_ind in 1:n_particles
|
|
for p2_ind in (p1_ind + 1):n_particles
|
|
c1 = cs[p1_ind, snapshot_ind]
|
|
c2 = cs[p2_ind, snapshot_ind]
|
|
|
|
r⃗₁₂ = c2 - c1
|
|
r⃗₁₂ = ReCo.restrict_coordinates(r⃗₁₂, sim_consts.half_box_len)
|
|
|
|
distance = ReCo.norm2d(r⃗₁₂)
|
|
|
|
lower_radius_ind = ceil(Int64, distance / Δradius)
|
|
|
|
if lower_radius_ind <= n_radii
|
|
g[lower_radius_ind] += 2
|
|
end
|
|
end
|
|
end
|
|
end
|
|
|
|
for (lower_radius_ind, lower_radius) in enumerate(lower_radii)
|
|
g[lower_radius_ind] *=
|
|
volume_per_particle / (
|
|
n_last_snapshots *
|
|
n_particles *
|
|
circular_shell_volume(lower_radius, Δradius)
|
|
)
|
|
end
|
|
|
|
gs[sim_dir_ind] = g
|
|
end
|
|
|
|
return (lower_radii, gs, particle_radius)
|
|
end
|
|
|
|
function plot_radial_distributions(;
|
|
v₀s::NTuple{N,Float64},
|
|
lower_radii::AbstractVector{Float64},
|
|
gs::Vector{Vector{Float64}},
|
|
particle_radius::Float64,
|
|
include_comparison::Bool,
|
|
filename_addition::String="",
|
|
) where {N}
|
|
println("Plotting the radial distributions")
|
|
|
|
init_cairomakie!()
|
|
|
|
fig = gen_figure()
|
|
|
|
min_lower_radius = 0.0
|
|
max_lower_radius = maximum(lower_radii)
|
|
|
|
min_g = 0.0
|
|
max_g = maximum(maximum.(gs))
|
|
|
|
ax = Axis(
|
|
fig[1:2, 1:2];
|
|
xticks=0:(2 * particle_radius):floor(Int64, max_lower_radius),
|
|
yticks=0:ceil(Int64, max_g),
|
|
xlabel=L"r / d",
|
|
ylabel=L"g",
|
|
limits=(min_lower_radius - 0.03, max_lower_radius + 0.03, min_g, max_g * 1.03),
|
|
)
|
|
|
|
lines!(
|
|
ax,
|
|
SVector(min_lower_radius, max_lower_radius),
|
|
SVector(1.0, 1.0);
|
|
linestyle=:dash,
|
|
color=:red,
|
|
linewidth=1,
|
|
)
|
|
|
|
for (g_ind, g) in enumerate(gs)
|
|
v₀ = round(Int64, v₀s[g_ind])
|
|
scatterlines!(
|
|
ax,
|
|
lower_radii,
|
|
g;
|
|
markersize=3,
|
|
linewidth=1,
|
|
label=L"Simulation with $v_0 = %$(v₀)$",
|
|
)
|
|
end
|
|
|
|
if include_comparison
|
|
comparison_curve = CSV.read(
|
|
"analysis/radial_distribution_function/g_of_r_d_ratio_with_0_37_packing_ratio.csv",
|
|
DataFrames.DataFrame;
|
|
header=3,
|
|
)
|
|
lines!(
|
|
ax,
|
|
comparison_curve.r_d_ratio,
|
|
comparison_curve.g_of_r_d_ratio;
|
|
label=L"Reference with $v_0 = 0$",
|
|
color=:orange,
|
|
)
|
|
end
|
|
|
|
axislegend(ax; position=:rt, padding=3, rowgap=-3)
|
|
|
|
save_fig("radial_distribution$filename_addition.pdf", fig)
|
|
|
|
return nothing
|
|
end
|
|
|
|
function run_radial_distribution_analysis()
|
|
v₀s = (0.0, 40.0, 80.0)
|
|
|
|
n_particles = 1000
|
|
|
|
sim_dirs = radial_distribution_simulation(;
|
|
n_particles=n_particles, v₀s=v₀s, T=100.0, packing_ratio=0.37
|
|
)
|
|
|
|
lower_radii, gs, particle_radius = radial_distribution(;
|
|
sim_dirs=sim_dirs, n_radii=58, n_last_snapshots=200, n_particles=n_particles
|
|
)
|
|
|
|
plot_radial_distributions(;
|
|
v₀s=v₀s[1:1], lower_radii, gs=gs[1:1], particle_radius, include_comparison=true
|
|
)
|
|
|
|
plot_radial_distributions(;
|
|
v₀s=v₀s[1:end],
|
|
lower_radii,
|
|
gs=gs[1:end],
|
|
particle_radius,
|
|
include_comparison=false,
|
|
filename_addition="_all_vs",
|
|
)
|
|
|
|
return nothing
|
|
end |