Skip to content

Audit a respiratory-droplet illustration before using its predictions

Download Python script

Audit a respiratory-droplet illustration before using its predictions

Section titled “Audit a respiratory-droplet illustration before using its predictions”

An evaporation calculation can conserve mass and still represent the wrong materials. This example follows a multicomponent droplet population in initially 30% RH air, while exposing a necessary earlier step: checking that requested chemical names resolve to the intended species.

The retained property lookup fails that identity check. In the saved output, “Lysozyme” resolves to hexachlorophene and “IgA” to isoguanine; the requested calcium and magnesium bicarbonates resolve to carbonates. Those are different substances, not acceptable protein or salt surrogates established by this example. Consequently the results illustrate the workflow and its input-audit failure; they must not be interpreted as quantitative respiratory-droplet or exposure predictions.

Read condensation and evaporation for the mass-transfer mechanism. The useful question here is what additional checks are required before adapting that mechanism to a material mixture.

Python · section 1
# 🚧 Install Particula if you're running this in Google Colab
# Remove the comment below to enable installation
# Install from the repository root: `python -m pip install -e ".[extra]"`
# 📦 Import necessary libraries
import copy # to safely duplicate Python objects
import matplotlib.pyplot as plt # for plotting
import numpy as np # numerical tools
import particula as par # the main Particula package
from particula.dynamics.condensation.species_mapping import (
CondensationSpeciesMap,
)
from particula.execution.gpu_resident_condensation import (
GPUResidentCondensation,
)
from tqdm import tqdm # optional: for progress bars during simulation
# 🎨 Set default plot styling using Tailwind-inspired colors from Particula
TAILWIND = par.util.colors.TAILWIND
base_color = TAILWIND["gray"]["600"]
plt.rcParams.update(
{
"text.color": base_color,
"axes.labelcolor": base_color,
"figure.figsize": (5, 4),
"font.size": 14,
"axes.edgecolor": base_color,
"xtick.color": base_color,
"ytick.color": base_color,
"pdf.fonttype": 42,
"ps.fonttype": 42,
}
)

Compare requested labels with resolved identities

Section titled “Compare requested labels with resolved identities”

The twelve requested labels span water, salts, small organics, and biological materials. The utility first attempts an exact lookup and can fall back to a similar name. Printing the returned name, CAS identifier, and structure is therefore a scientific input check, not merely a status message. The mismatches identified above invalidate the claim that these arrays describe the requested respiratory-fluid composition.

The independently sampled mass ranges are illustrative and lack a cited composition dataset here. Molar masses are kg/mol, densities kg/m³, surface tensions N/m, and vapor pressures Pa. The NaCl surface-tension override is an assumed constant; it is not a concentration-dependent electrolyte model. A verified identity and a numerical property value are separate requirements.

Python · section 2
list_of_chemicals = [
"Water",
"NaCl",
"potassium bicarbonate (KHCO3)",
"potassium dihydrogen phosphate (KH2PO4)", # K+ + H2PO4–
"calcium bicarbonate [Ca(HCO3)2]",
"magnesium bicarbonate [Mg(HCO3)2]",
"Urea",
"Lactate",
"Glucose",
"Lysozyme",
"IgA",
"cholesterol",
]
initial_mass_fraction_bounds = [
(0.985, 0.995), # Water
(0.0006, 0.0023), # NaCl
(0.0007, 0.0030), # KHCO3
(0.0002, 0.0027), # KH2PO4
(0.00002, 0.00045), # Ca(HCO3)2
(0.000006, 0.00003), # Mg(HCO3)2
(0.0002, 0.0015), # Urea
(0.00005, 0.0004), # Lactate
(0.00001, 0.0002), # Glucose
(0.000005, 0.00002), # Lysozyme
(0.00002, 0.0002), # IgA
(0.00005, 0.0005), # Cholesterol
]
density_array = np.array([])
molar_mass_array = np.array([])
surface_tension_array = np.array([])
short_name = []
chemical_dict = {}
# Get the CAS numbers for each chemical
for chem in list_of_chemicals:
cas = par.util.get_chemical_search(chem)
print(f"{chem}: {cas}")
chemical_dict[chem] = par.util.get_chemical_stp_properties(cas)
# Store the molar mass and density
molar_mass_array = np.append(
molar_mass_array, chemical_dict[chem]["molar_mass"]
)
density_array = np.append(density_array, chemical_dict[chem]["density"])
surface_tension_array = np.append(
surface_tension_array, chemical_dict[chem]["surface_tension"]
)
short_name.append(chemical_dict[chem]["name"])
# correct NaCl surface tension
surface_tension_array[1] = 0.091 # N/m, typical value
# Print the chemical properties
for chem, props in chemical_dict.items():
print(f"{chem}:")
print(f" Name: {props['name']}")
print(f" CAS: {props['cas_number']}")
print(f" Smiles: {props['smiles']}")
print(f" Molar Mass: {props['molar_mass']:.4f} kg/mol")
print(f" Density: {props['density']:.2f} kg/m³")
print(f" Surface Tension: {props['surface_tension']} N/m")
print(f" Vapor Pressure: {props['pure_vapor_pressure']} Pa\n")
Water: Water
NaCl: NaCl
potassium bicarbonate (KHCO3): potassium bicarbonate (KHCO3)
potassium dihydrogen phosphate (KH2PO4): potassium dihydrogen phosphate (KH2PO4)
calcium bicarbonate [Ca(HCO3)2]: calcium carbonate [usan]
magnesium bicarbonate [Mg(HCO3)2]: magnesium carbonate (mgco3)
Urea: Urea
Lactate: Lactate
Glucose: Glucose
Lysozyme: soy-dome
IgA: IgA
cholesterol: cholesterol
Water:
  Name: water
  CAS: 7732-18-5
  Smiles: O
  Molar Mass: 0.0180 kg/mol
  Density: 997.06 kg/m³
  Surface Tension: 0.07197220523022962 N/m
  Vapor Pressure: 3169.9293388738784 Pa
NaCl:
  Name: sodium chloride
  CAS: 7647-14-5
  Smiles: [Na+].[Cl-]
  Molar Mass: 0.0584 kg/mol
  Density: 2169.99 kg/m³
  Surface Tension: 0.35290821232462 N/m
  Vapor Pressure: 2.245741871175006e-30 Pa
potassium bicarbonate (KHCO3):
  Name: potassium bicarbonate
  CAS: 298-14-6
  Smiles: C(=O)(O)[O-].[K+]
  Molar Mass: 0.1001 kg/mol
  Density: 2170.00 kg/m³
  Surface Tension: None N/m
  Vapor Pressure: None Pa
potassium dihydrogen phosphate (KH2PO4):
  Name: potassium dihydrogen phosphate
  CAS: 7778-77-0
  Smiles: OP(=O)(O)[O-].[K+]
  Molar Mass: 0.1361 kg/mol
  Density: 2340.01 kg/m³
  Surface Tension: None N/m
  Vapor Pressure: None Pa
calcium bicarbonate [Ca(HCO3)2]:
  Name: calcium carbonate
  CAS: 471-34-1
  Smiles: C(=O)([O-])[O-].[Ca+2]
  Molar Mass: 0.1001 kg/mol
  Density: 2710.00 kg/m³
  Surface Tension: None N/m
  Vapor Pressure: None Pa
magnesium bicarbonate [Mg(HCO3)2]:
  Name: magnesium carbonate
  CAS: 546-93-0
  Smiles: C(=O)([O-])[O-].[Mg+2]
  Molar Mass: 0.0843 kg/mol
  Density: 3010.00 kg/m³
  Surface Tension: None N/m
  Vapor Pressure: None Pa
Urea:
  Name: urea
  CAS: 57-13-6
  Smiles: C(=O)(N)N
  Molar Mass: 0.0601 kg/mol
  Density: 932.80 kg/m³
  Surface Tension: 0.06630868366869905 N/m
  Vapor Pressure: 9.647479358454168 Pa
Lactate:
  Name: lactic acid
  CAS: 50-21-5
  Smiles: CC(C(=O)O)O
  Molar Mass: 0.0901 kg/mol
  Density: 1146.11 kg/m³
  Surface Tension: 0.06892887098077884 N/m
  Vapor Pressure: 0.42214021514470007 Pa
Glucose:
  Name: glucose
  CAS: 50-99-7
  Smiles: O=C[C@H](O)[C@@H](O)[C@H](O)[C@H](O)CO
  Molar Mass: 0.1802 kg/mol
  Density: 1618.57 kg/m³
  Surface Tension: 0.10078381855423499 N/m
  Vapor Pressure: 3.472643152420593e-11 Pa
Lysozyme:
  Name: hexachlorophene
  CAS: 70-30-4
  Smiles: C1=C(C(=C(C(=C1Cl)Cl)CC2=C(C(=CC(=C2Cl)Cl)Cl)O)O)Cl
  Molar Mass: 0.4069 kg/mol
  Density: 2009.99 kg/m³
  Surface Tension: 0.07787535981784431 N/m
  Vapor Pressure: 9.468715811988936e-14 Pa
IgA:
  Name: isoguanine
  CAS: 3373-53-3
  Smiles: C1=NC2=NC(=O)NC(=C2N1)N
  Molar Mass: 0.1511 kg/mol
  Density: 815.71 kg/m³
  Surface Tension: 0.1373263034398279 N/m
  Vapor Pressure: 0.00012951501534944908 Pa
cholesterol:
  Name: cholesterol
  CAS: 57-88-5
  Smiles: CC(C)CCC[C@@H](C)[C@H]1CC[C@H]2[C@@H]3CC=C4C[C@@H](O)CC[C@]4(C)[C@H]3CC[C@]12C
  Molar Mass: 0.3867 kg/mol
  Density: 1064.37 kg/m³
  Surface Tension: 0.03942801914463943 N/m
  Vapor Pressure: 8.548725890318368e-07 Pa

Distinguish reproducible inputs from validated inputs

Section titled “Distinguish reproducible inputs from validated inputs”

The sample contains 1000 particles at 10⁻⁴ cm⁻³ (100 m⁻³), corresponding to 10 m³ represented volume, at 298.15 K. The fixed sampling seed controls this realization, not the accuracy of the material model.

Missing vapor pressures are replaced by 10⁻³⁰ Pa, making those lanes nearly nonvolatile. Missing surface tensions receive 0.072 N/m. These substitutions allow execution; they do not resolve missing measurements or incorrect identities. Record them as assumptions before interpreting any partitioning.

Python · section 4
# 🎲 Ensure reproducibility
np.random.seed(100)
# 🔢 Simulation settings
number_of_samples = 1_000 # Number of particles to simulate
total_number_per_cm3 = 1e-4 # Particle number concentration [#/cm³]
simulation_volume = (
number_of_samples / total_number_per_cm3 * 1e-6
) # Simulation volume in m³ (converted from cm³)
temperature = 298.15 # System temperature in Kelvin (25°C)
# 🌡 Build vapor pressure strategies for each chemical
vapor_pressure_strategy_list = []
for chem_i, props in chemical_dict.items():
vapor_pressure_temp = props["pure_vapor_pressure"]
if vapor_pressure_temp is None:
print(f"⚠️ Warning: No vapor pressure data for {chem_i}, using default.")
vapor_pressure_temp = 1e-30 # fallback for non-volatile species
vapor_pressure_i = (
par.gas.ConstantVaporPressureBuilder()
.set_vapor_pressure(vapor_pressure_temp, "Pa")
.build()
)
vapor_pressure_strategy_list.append(vapor_pressure_i)
# 💧 Ensure surface tension values are set
surface_tension_array = np.array(
[0.072 if x is None else x for x in surface_tension_array]
)
⚠️ Warning: No vapor pressure data for potassium bicarbonate (KHCO3), using default.
⚠️ Warning: No vapor pressure data for potassium dihydrogen phosphate (KH2PO4), using default.
⚠️ Warning: No vapor pressure data for calcium bicarbonate [Ca(HCO3)2], using default.
⚠️ Warning: No vapor pressure data for magnesium bicarbonate [Mg(HCO3)2], using default.

Initialize a finite gas reservoir at 30% saturation

Section titled “Initialize a finite gas reservoir at 30% saturation”

Water gas concentration is 0.3 times its modeled saturation concentration; other vapor lanes start near zero. Gas names retain the requested labels even when property lookup returned another substance, so names in a container do not certify the underlying material identity.

Temperature is fixed at 298.15 K and pressure at 101325 Pa. Humidity is only initialized here. The active loop does not reset it to 30% at each step.

Python · section 5
# 💨 Initialize gas-phase concentrations near zero
concentration_gas = np.ones(len(list_of_chemicals)) * 1e-30 # kg/m³
# 💧 Set water vapor to 30% RH
isat_water = vapor_pressure_strategy_list[0].saturation_concentration(
molar_mass_array[0], temperature
)
concentration_gas[0] = isat_water * 0.3 # 30% RH
# 🧪 Build gas-phase species with vapor strategies and concentrations
gas_species = par.gas.GasData(
name=list_of_chemicals,
molar_mass=molar_mass_array.copy(),
partitioning=np.ones(len(list_of_chemicals), dtype=bool),
concentration=concentration_gas[None, :].copy(),
)
environment = par.gas.EnvironmentData(
temperature=np.array([temperature]),
pressure=np.array([101325.0]),
saturation_ratio=np.zeros((1, len(list_of_chemicals))),
)

Track what the composition sampling actually constructs

Section titled “Track what the composition sampling actually constructs”

The three supplied radius modes are 0.3, 4, and 40 µm with relative weights 15, 2, and 0.2. They define an illustrative sample, not a validated cough-size distribution. Water-equivalent spherical mass is multiplied independently by each sampled composition factor. Those factors are not jointly normalized, so the final total mass and reconstructed radius need not exactly equal the water-only sample from which they were built.

The κ array is assumed, with water in lane zero. Each species receives its own surface-phase label, so this configuration retains species-specific surface tensions rather than averaging them across one shared liquid phase. It does not calculate physical phase separation. The inventory helper counts weighted particle mass plus gas mass in the represented volume.

Python · section 6
# 💠 3a. Lognormal size distributions for cough droplet modes
# Modes: bronchiolar, laryngeal, oropharyngeal
modes = np.array([0.6, 8.0, 80.0]) / 2 * 1e-6 # radii in meters
geometric_standard_deviations = np.array([1.65, 2.4, 1.6])
relative_number = np.array([15, 2, 0.2]) # relative weights
# Generate sampled radii
radii_cough = par.particles.get_lognormal_sample_distribution(
mode=modes,
geometric_standard_deviation=geometric_standard_deviations,
number_of_particles=relative_number,
number_of_samples=number_of_samples,
)
# 💧 3b. Assign initial water mass and scale by chemical mass fractions
mass_water = 4 / 3 * np.pi * (radii_cough**3) * density_array[0]
mass_speciation = np.repeat(
mass_water[:, np.newaxis], len(list_of_chemicals), axis=1
)
# Apply random mass fractions from defined bounds
for i, (_chem, bounds) in enumerate(
zip(list_of_chemicals, initial_mass_fraction_bounds)
):
mass_fraction = np.random.uniform(bounds[0], bounds[1], number_of_samples)
mass_speciation[:, i] *= mass_fraction
# ⚗️ 3c. Define activity strategy using kappa parameters
kappa_parameter = np.array(
[0.0, 1.1, 0.6, 0.6, 0.6, 0.6, 0.3, 0.2, 0.1, 0.1, 0.1, 0.01]
)
activity_strategy = (
par.particles.ActivityKappaParameterBuilder()
.set_molar_mass(molar_mass_array, "kg/mol")
.set_density(density_array, "kg/m^3")
.set_kappa(kappa_parameter)
.set_water_index(0)
.build()
)
# Define surface tension strategy
surface_strategy = (
par.particles.SurfaceStrategyMassBuilder()
.set_surface_tension(surface_tension_array, "N/m")
.set_density(density_array, "kg/m^3")
.set_phase_index(np.arange(len(list_of_chemicals)))
.build()
)
# 🧱 3d. Build resolved particle object
resolved_masses = par.particles.create_resolved_particle_data(
masses=mass_speciation,
density=density_array,
volume=simulation_volume,
)
aerosol = par.Aerosol(
particles=resolved_masses, gas=gas_species, environment=environment
)
def species_inventory(aero):
"""Independent species inventory in kg, including represented volume."""
return np.sum(
aero.particles.masses * aero.particles.concentration[..., None], axis=1
) + (aero.gas.concentration * aero.particles.volume[:, None])
initial_inventory = species_inventory(aerosol)
# Optional: print summary of aerosol system
print("Particle slots:", aerosol.particles.masses.shape[1])
print("Gas species:", ", ".join(aerosol.gas.name))
Particle slots: 1000
Gas species: Water, NaCl, potassium bicarbonate (KHCO3), potassium dihydrogen phosphate (KH2PO4), calcium bicarbonate [Ca(HCO3)2], magnesium bicarbonate [Mg(HCO3)2], Urea, Lactate, Glucose, Lysozyme, IgA, cholesterol

Inspect the initialized distribution on its own terms

Section titled “Inspect the initialized distribution on its own terms”

The histogram counts sampled radii per logarithmic bin. It is not yet a number concentration or a diameter spectrum. The native ParticleData stores the species masses and represented counts; its derived radii include the sampled composition and densities. This is the appropriate initialized state to compare with later sizes.

Python · section 7
fig, ax = plt.subplots()
# Log-scale histogram of particle radii
ax.hist(
np.log10(aerosol.particles.radii[0]), # convert to log10
bins=50,
density=False,
alpha=0.5,
)
ax.set_xlabel("log₁₀(Radius [m])")
ax.set_ylabel("Bin Counts")
ax.set_title("Initial Particle Size Distribution")
plt.tight_layout()
plt.show()
Audit a respiratory-droplet illustration before using its predictions: saved scientific figure
Audit a respiratory-droplet illustration before using its predictions · saved execution output

Separate a configured process from an executed process

Section titled “Separate a configured process from an executed process”

Condensation/evaporation maps all twelve lanes, with a common diffusivity 2×10⁻⁵ m²/s and accommodation one. These are simplifying assumptions across a chemically diverse set of requested labels. A Brownian process is also constructed, but its execution is commented out in the loop below. The saved calculation therefore contains mass transfer alone, without coagulation, ventilation, sedimentation removal, or temperature feedback.

Python · section 9
# 💧 Isothermal Condensation
condensation_strategy = par.dynamics.CondensationIsothermal(
molar_mass=np.array(molar_mass_array),
activity_strategy=activity_strategy,
surface_strategy=surface_strategy,
vapor_pressure_strategy=vapor_pressure_strategy_list,
species_map=CondensationSpeciesMap(
tuple(gas_species.name),
tuple((i, i) for i in range(len(list_of_chemicals))),
),
diffusion_coefficient=2e-5, # typical air value [m²/s]
accommodation_coefficient=1, # full sticking efficiency
# skip_partitioning_indices=[] # optional: exclude species
)
condensation_process = par.dynamics.MassCondensation(condensation_strategy)
# ⚪ Brownian Coagulation
coagulation_strategy = (
par.dynamics.BrownianCoagulationBuilder()
.set_distribution_type("particle_resolved") # use individual particles
.build()
)
coagulation_process = par.dynamics.Coagulation(coagulation_strategy)

Record 20 seconds of adaptive closed mass transfer

Section titled “Record 20 seconds of adaptive closed mass transfer”

There are 201 stored states, including both endpoints of the full 20 s run. ROS3 adapts condensation steps inside each 0.1 s output interval, with rtol 10⁻⁵, particle atol 10⁻²⁵ kg and gas atol 10⁻¹² kg/m³. Water exchanges with a finite gas reservoir. Better integration does not repair uncertain chemical properties or validate respiratory-fluid predictions.

At every stored state, the assertion compares each gas-plus-particle inventory with its initial value at rtol=10⁻¹² and atol=10⁻³⁰ kg. This checks material accounting for the supplied arrays; it cannot detect that a property array belongs to the wrong chemical. Radius-bin counts, bulk particle species masses, and vapor saturation ratios support distinct diagnostics below.

All twelve materials exchange on CUDA. The adaptive clock, acceptance and step-size decisions execute inside a native CUDA conditional graph. Snapshots explicitly download the 201 requested observations. Internal work counts are printed by the run; inventory checks alone do not establish convergence.

Python · section 10
print("Initial gas concentration (kg/m^3):", aerosol.gas.concentration)
# 🔁 Deep copy for simulation
aerosol_process = copy.deepcopy(aerosol)
# Simulation parameters
total_time = 20 # seconds
total_steps = 200
time_step = total_time / total_steps
time_array = np.linspace(0.0, total_time, total_steps + 1)
time_len = len(time_array)
# 📊 Initialize arrays for tracking outputs
bins_lognormal = np.logspace(-8, -4, 100)
distribution_counts = np.zeros((time_len, len(bins_lognormal) - 1))
species_mass = np.zeros((time_len, len(list_of_chemicals)))
vapor_saturation = np.zeros((time_len, len(list_of_chemicals)))
simulation = GPUResidentCondensation(
aerosol_process,
condensation_strategy,
device="cuda:0",
method="ros3",
rtol=1e-5,
particle_atol=1e-25,
gas_atol=1e-12,
first_step=1e-6,
max_step=time_step,
)
# 🌀 Time integration loop
for step, _t in enumerate(
tqdm(time_array, desc="Running Sim", mininterval=0.5)
):
if step > 0:
while simulation.time < float(_t):
simulation.step(float(_t))
aerosol_process = simulation.snapshot()
# Record size distribution
distribution_counts[step, :], edges = np.histogram(
aerosol_process.particles.radii[0].copy(),
bins=bins_lognormal,
density=False,
)
# Record species mass in particles
species_temp = aerosol_process.particles.masses[0].copy()
species_mass[step, :] = np.sum(
species_temp * aerosol_process.particles.concentration[0, :, None],
axis=0,
)
np.testing.assert_allclose(
species_inventory(aerosol_process),
initial_inventory,
rtol=1e-12,
atol=1e-30,
)
# Record vapor saturation ratios
vapor_saturation[step, :] = np.array(
[
aerosol_process.gas.concentration[0, i]
/ strategy.saturation_concentration(
molar_mass_array[i], aerosol_process.environment.temperature[0]
)
for i, strategy in enumerate(vapor_pressure_strategy_list)
]
)
# Normalize counts to number concentration (#/m³)
concentrations = distribution_counts / simulation_volume
# Final system state and actual adaptive work
print("Final gas concentration (kg/m^3):", aerosol_process.gas.concentration)
print("Adaptive condensation work:", simulation.statistics)
Initial gas concentration (kg/m^3): [[6.9110312e-03 1.0000000e-30 1.0000000e-30 1.0000000e-30 1.0000000e-30
  1.0000000e-30 1.0000000e-30 1.0000000e-30 1.0000000e-30 1.0000000e-30
  1.0000000e-30 1.0000000e-30]]
Warp 1.11.1 initialized:
   CUDA Toolkit 12.9, Driver 13.4
   Devices:
     "cpu"      : "CPU"
     "cuda:0"   : "NVIDIA GeForce RTX 5060" (8 GiB, sm_120, mempool enabled)
   Kernel cache:
     ~/.cache/warp/1.11.1
Module particula.execution._condensation_ode_kernels 9425cdf load on device 'cuda:0' took 3.52 ms  (cached)
Module particula.gpu.kernels.thermodynamics 10b2126 load on device 'cuda:0' took 5.94 ms  (cached)
Module particula.gpu.kernels.condensation ad8087d load on device 'cuda:0' took 5.99 ms  (cached)
Module particula.execution._adaptive_condensation_physics a3e6636 load on device 'cuda:0' took 5.43 ms  (cached)
Module particula.execution._condensation_reservoirs b231d81 load on device 'cuda:0' took 5.95 ms  (cached)
Module particula.execution._condensation_device_control 053a506 load on device 'cuda:0' took 3.92 ms  (cached)
Module particula.execution._condensation_ros3_kernels d88df3a load on device 'cuda:0' took 5.17 ms  (cached)
Module particula.execution._condensation_block_kernels d2447a5 load on device 'cuda:0' took 5.08 ms  (cached)
Final gas concentration (kg/m^3): [[6.91103209e-03 9.99999850e-31 9.99999848e-31 9.99999847e-31
  9.99999848e-31 9.99999849e-31 4.18285543e-13 1.90984831e-14
  1.55679389e-24 5.12376933e-28 6.26522875e-18 9.74225658e-20]]
Adaptive condensation work: StepStatistics(accepted=8921, rejected=1, domain_rejected=0, rate_calls=446100, jacobian_calls=8922, time_derivative_calls=0, factorizations=8922, linear_solves=26766)
Execution progress
Running Sim:   0%|          | 0/201 [00:00<?, ?it/s]
Running Sim:   1%|          | 2/201 [03:03<5:03:58, 91.65s/it]
Running Sim:   1%|▏         | 3/201 [03:18<3:16:40, 59.60s/it]
Running Sim:   2%|▏         | 4/201 [03:27<2:13:54, 40.79s/it]
Running Sim:   2%|▏         | 5/201 [03:33<1:34:05, 28.81s/it]
Running Sim:   3%|▎         | 6/201 [03:36<1:05:49, 20.25s/it]
Running Sim:   3%|▎         | 7/201 [03:40<48:21, 14.95s/it]
Running Sim:   4%|▍         | 8/201 [03:42<35:28, 11.03s/it]
Running Sim:   4%|▍         | 9/201 [03:44<25:32,  7.98s/it]
Running Sim:   5%|▍         | 10/201 [03:46<20:29,  6.44s/it]
Running Sim:   5%|▌         | 11/201 [03:49<16:42,  5.28s/it]
Running Sim:   6%|▌         | 12/201 [03:52<14:12,  4.51s/it]
Running Sim:   6%|▋         | 13/201 [03:54<12:00,  3.83s/it]
Running Sim:   7%|▋         | 14/201 [03:55<08:52,  2.85s/it]
Running Sim:   7%|▋         | 15/201 [03:56<07:25,  2.39s/it]
Running Sim:   8%|▊         | 16/201 [03:58<06:30,  2.11s/it]
Running Sim:   9%|▉         | 18/201 [03:58<04:01,  1.32s/it]
Running Sim:  10%|▉         | 20/201 [03:59<02:49,  1.07it/s]
Running Sim:  10%|█         | 21/201 [04:00<02:33,  1.17it/s]
Running Sim:  11%|█         | 22/201 [04:01<02:48,  1.06it/s]
Running Sim:  11%|█▏        | 23/201 [04:02<03:12,  1.08s/it]
Running Sim:  12%|█▏        | 25/201 [04:03<02:18,  1.27it/s]
Running Sim:  13%|█▎        | 27/201 [04:04<02:10,  1.34it/s]
Running Sim:  14%|█▍        | 28/201 [04:06<02:31,  1.14it/s]
Running Sim:  14%|█▍        | 29/201 [04:06<02:15,  1.27it/s]
Running Sim:  16%|█▌        | 32/201 [04:07<01:26,  1.96it/s]
Running Sim:  17%|█▋        | 35/201 [04:08<01:04,  2.55it/s]
Running Sim:  18%|█▊        | 37/201 [04:08<01:00,  2.69it/s]
Running Sim:  19%|█▉        | 39/201 [04:10<01:15,  2.13it/s]
Running Sim:  20%|██        | 41/201 [04:11<01:21,  1.96it/s]
Running Sim:  21%|██▏       | 43/201 [04:12<01:25,  1.85it/s]
Running Sim:  22%|██▏       | 45/201 [04:13<01:12,  2.15it/s]
Running Sim:  23%|██▎       | 47/201 [04:13<01:06,  2.32it/s]
Running Sim:  24%|██▍       | 49/201 [04:15<01:13,  2.06it/s]
Running Sim:  25%|██▌       | 51/201 [04:16<01:19,  1.89it/s]
Running Sim:  26%|██▋       | 53/201 [04:17<01:17,  1.92it/s]
Running Sim:  28%|██▊       | 56/201 [04:17<00:56,  2.57it/s]
Running Sim:  30%|██▉       | 60/201 [04:18<00:39,  3.56it/s]
Running Sim:  32%|███▏      | 64/201 [04:19<00:33,  4.08it/s]
Running Sim:  33%|███▎      | 67/201 [04:20<00:44,  2.99it/s]
Running Sim:  34%|███▍      | 69/201 [04:21<00:47,  2.76it/s]
Running Sim:  36%|███▌      | 72/201 [04:22<00:40,  3.17it/s]
Running Sim:  37%|███▋      | 74/201 [04:23<00:41,  3.06it/s]
Running Sim:  38%|███▊      | 76/201 [04:24<00:44,  2.78it/s]
Running Sim:  39%|███▉      | 78/201 [04:24<00:42,  2.90it/s]
Running Sim:  41%|████▏     | 83/201 [04:25<00:27,  4.23it/s]
Running Sim:  44%|████▍     | 88/201 [04:25<00:20,  5.52it/s]
Running Sim:  46%|████▋     | 93/201 [04:26<00:16,  6.59it/s]
Running Sim:  49%|████▉     | 98/201 [04:26<00:13,  7.45it/s]
Running Sim:  51%|█████     | 103/201 [04:27<00:12,  8.10it/s]
Running Sim:  54%|█████▎    | 108/201 [04:27<00:11,  8.44it/s]
Running Sim:  56%|█████▌    | 113/201 [04:28<00:12,  6.84it/s]
Running Sim:  58%|█████▊    | 117/201 [04:30<00:17,  4.90it/s]
Running Sim:  60%|██████    | 121/201 [04:31<00:15,  5.12it/s]
Running Sim:  64%|██████▎   | 128/201 [04:31<00:10,  6.85it/s]
Running Sim:  67%|██████▋   | 135/201 [04:32<00:07,  8.36it/s]
Running Sim:  71%|███████   | 142/201 [04:32<00:06,  9.62it/s]
Running Sim:  74%|███████▍  | 149/201 [04:33<00:04, 10.63it/s]
Running Sim:  78%|███████▊  | 156/201 [04:33<00:03, 11.41it/s]
Running Sim:  81%|████████  | 163/201 [04:34<00:03, 11.99it/s]
Running Sim:  85%|████████▍ | 170/201 [04:34<00:02, 12.42it/s]
Running Sim:  88%|████████▊ | 177/201 [04:35<00:01, 12.73it/s]
Running Sim:  92%|█████████▏| 184/201 [04:35<00:01, 12.96it/s]
Running Sim:  95%|█████████▌| 191/201 [04:36<00:00, 13.12it/s]
Running Sim:  99%|█████████▊| 198/201 [04:36<00:00, 13.24it/s]
Running Sim: 100%|██████████| 201/201 [04:36<00:00,  1.38s/it]

The dashed and solid curves use the same radius bins. Their vertical values are number concentration per bin in m⁻³, not dN/dr. A shift toward smaller radii is consistent with net evaporation under the modeled inputs; it does not by itself prove equilibrium or a validated droplet lifetime. Particles moving outside the histogram’s limits would also change the plotted bin sum.

Python · section 11
fig, ax = plt.subplots(figsize=(7, 5))
# Get radii before and after simulation
initial_distribution = aerosol.particles.radii[0].copy()
final_distribution = aerosol_process.particles.radii[0].copy()
# Compute binned counts
counts_initial, _ = np.histogram(
initial_distribution, bins=bins_lognormal, density=False
)
counts_final, _ = np.histogram(
final_distribution, bins=bins_lognormal, density=False
)
# Plot initial distribution
ax.plot(
bins_lognormal[:-1],
counts_initial / simulation_volume,
label="Initial Size Distribution",
color=TAILWIND["sky"]["800"],
linestyle="--",
)
# Plot final distribution
ax.plot(
bins_lognormal[:-1],
counts_final / simulation_volume,
label="Final Size Distribution",
color=TAILWIND["rose"]["400"],
linestyle="-",
)
# Set axis scales and labels
ax.set_xscale("log")
ax.set_xlabel("Particle Radius (m)")
ax.set_ylabel("Concentration (#/m⁻³)")
ax.legend()
plt.tight_layout()
plt.show()
Audit a respiratory-droplet illustration before using its predictions: saved scientific figure
Audit a respiratory-droplet illustration before using its predictions · saved execution output

Read gas saturation separately from particle equilibrium

Section titled “Read gas saturation separately from particle equilibrium”

The water saturation ratio divides gas concentration by pure-water saturation concentration at the fixed temperature. Unity means saturation relative to that reference, not equilibrium for every curved, multicomponent particle.

The small reservoir change in the saved run reflects this population’s low loading relative to gas water. It is an outcome of the chosen volume and concentrations, not a general claim that humidity stays constant near a cough. The lower axis limit is 0.3, so inspect the actual vertical scale when judging the change.

Python · section 12
fig, ax = plt.subplots(figsize=(7, 5))
# Plot water vapor saturation ratio (species index 0)
ax.plot(
time_array,
vapor_saturation[:, 0],
color=TAILWIND["sky"]["800"],
label="Vapor Saturation Ratio (Water)",
)
ax.set_ylim(bottom=0.3)
ax.set_xlabel("Time (s)")
ax.set_ylabel("Vapor Saturation Ratio")
ax.set_title("Vapor Saturation Ratio for Water Over Time")
ax.legend()
plt.tight_layout()
plt.show()
Audit a respiratory-droplet illustration before using its predictions: saved scientific figure
Audit a respiratory-droplet illustration before using its predictions · saved execution output

Identify the weighting behind a composition plot

Section titled “Identify the weighting behind a composition plot”

The stack plot divides summed particle species mass by summed total particle mass. It is a bulk mass fraction, not an arithmetic average of each particle’s composition, despite the retained “Mean” axis label. Larger particles can dominate this diagnostic. It cannot on its own identify which size class retains water longer.

Legend names come from the resolved property entries. Their disagreement with the requested labels is further evidence of the identity problem, not a change of species caused by evaporation.

Python · section 13
# Compute total mass fraction over time
mass_fraction = species_mass / np.sum(species_mass, axis=1, keepdims=True)
# Set color palette (one color per species)
color_list = [
TAILWIND["sky"]["800"], # Water
TAILWIND["rose"]["400"], # NaCl
TAILWIND["red"]["800"], # KHCO3
TAILWIND["red"]["600"], # KH2PO4
TAILWIND["amber"]["500"], # Ca(HCO3)2
TAILWIND["orange"]["400"], # Mg(HCO3)2
TAILWIND["teal"]["900"], # Urea
TAILWIND["teal"]["700"], # Lactate
TAILWIND["green"]["500"], # Glucose
TAILWIND["indigo"]["600"], # Lysozyme
TAILWIND["indigo"]["400"], # IgA
TAILWIND["cyan"]["500"], # Cholesterol
]
# Create stackplot of species mass fraction
fig, ax = plt.subplots(figsize=(8, 5))
ax.stackplot(
time_array,
mass_fraction.T, # Each row = one species
labels=short_name, # Species names
colors=color_list,
alpha=1.0,
)
ax.set_xlabel("Time (s)")
ax.set_ylabel("Mean Mass Fraction in Particles")
ax.set_ylim(0, 1)
ax.grid(True)
ax.legend(
title="Species",
loc="lower left",
framealpha=0.9,
fontsize=10,
)
plt.tight_layout()
plt.show()
Audit a respiratory-droplet illustration before using its predictions: saved scientific figure
Audit a respiratory-droplet illustration before using its predictions · saved execution output

Follow size redistribution without overinterpreting a plateau

Section titled “Follow size redistribution without overinterpreting a plateau”

The contour displays log₁₀ of number concentration per radius bin through time. It complements the bulk mass-fraction plot by retaining size information. A slowly changing band is not sufficient evidence of thermodynamic equilibrium; that would require examining mass-transfer rates and convergence as well. The plotted range omits radii below 0.1 µm even though the stored histogram extends lower.

Python · section 14
fig, ax = plt.subplots(figsize=(7, 5))
# Create 2D mesh grid for time and particle radius
X, Y = np.meshgrid(time_array, edges[:-1])
# Convert number concentration to log10 scale (avoid log(0) issues)
log_conc = np.log10(
concentrations,
where=concentrations > 0,
out=np.full_like(concentrations, np.nan), # mask zeros
)
# Contour plot of log concentration
cont = ax.contourf(X, Y, log_conc.T)
# Axis settings
ax.set_yscale("log")
ax.set_ylim(1e-7, 1e-4)
ax.set_xlabel("Time (s)")
ax.set_ylabel("Particle Radius (m)")
# Add colorbar
fig.colorbar(cont, label="Log₁₀ Number Concentration")
plt.tight_layout()
plt.show()
Audit a respiratory-droplet illustration before using its predictions: saved scientific figure
Audit a respiratory-droplet illustration before using its predictions · saved execution output

Carry forward the audit, not an unsupported material prediction

Section titled “Carry forward the audit, not an unsupported material prediction”

The workflow links sampled composition, finite-reservoir evaporation, and size and mass diagnostics. Its inventory checks pass for the supplied numerical species, but the identity mismatches prevent interpreting those results as respiratory-fluid behavior. Successful execution is not material validation.

Before a research adaptation, replace fuzzy matches with verified identifiers and documented properties or explicitly justified macromolecular surrogates. Normalize the intended starting composition, choose a defensible activity and surface model, then rerun the full calculation. Compare size-resolved water loss across humidity and timestep, using measured evaporation data to test the material assumptions. None of that validation is supplied by the retained run.

Explore this topic: Emissions and exposure