The LANL chamber is in the first phase of development with control of both temperature and humidity to be added in future work. The 906 L (0.906 m³, internal volume) chamber is made of 6 stainless steel walls which are inert and reduce the effects of electrostatic charge. The rectangular body and dimensions are shown in Fig. 1. The chamber has an internal surface area of 6 m² with surface-to-volume ratio of 6.6 m⁻¹. The chamber's joints are sealed with a fast cure marine adhesive caulk (Sika, Sikaflex 291) and the outside junctions where the walls intersect are sealed using ZIP System™ Strech Tape (Huber Engineered Woods). Portholes were made for the top and bottom plates where wires and probes be placed for measurements in the chamber and for aerosols to flow in and out of the chamber. Unused portholes are sealed with Swagelok caps and those used for probes and wires are sealed with a rubber gasket or a Teflon ferrule. A leak test was performed by pressurizing the chamber by feeding clean air in and sealing every outlet then seal any leak detected. The chamber is designed to operate at ambient pressure.

Copper tubing lines (3/8′′) are used to supply aerosols to the chamber and deliver outflow sampling to instrumentation. Zero-air generators (T701 Teledyne Inc., USA) provide clean dry air to push aerosol to the chamber and additional dilution air using Teflon tubing (1/4′′). Push flow enters at the bottom of the chamber, creating an upwards direction of flow. Aerosols are sampled from an outlet at the top of the chamber. A dilution flow is connected to the outlet line (88.9 mm from the outlet) to control aerosol concentrations and prevent overwhelming the sampling instruments. A minimum sampling flow rate of 1.5 L min⁻¹ was needed to supply the instruments and we used a 1:5 ratio of push to dilution for the experiments presented here. This infers a residence timescale within the chamber of 604 min (10 h) and half-life of 418 min (6.9 h). The flow rates are controlled with mass flow controllers (MFC; Alicat). Prior to each experiment the chamber was flushed by pushing clean air with a flow of ∼ 10 L min⁻¹ for at least 3 h to reach background (∼ 0–10 cm⁻³).

Two aqueous solutions and controlled combustion of dried biomaterial were used as the sources of aerosols. Sodium chloride (NaCl; Sigma-Aldrich) was chosen because it is a well-understood compound in aerosol studies. Sucrose (Sigma-Aldrich) was used to act as a secondary organic aerosol surrogate, and it is also a well-studied aerosol. Each were dissolved in deionized water (Milli-Q, 18.2 MΩ) in separate solutions and were put on an Atomizer Aerosol Generator (3079, TSI Inc., USA). The particles coming out of the atomizer passed through a silica gel diffusion drier at a generation flow rate of ∼ 2.4 L min⁻¹. The duration of aerosol injection varied based on the desired number concentration. To generate smoke, 0.1–0.5 g samples of dried biomaterial Poa pratensis (Kentucky bluegrass) were weighed out, placed on a quartz boat and into a quartz-tube furnace (Carbolite Gero, TS1-1200, Verder Scientific, UK) that was set to 1000 °C for a flaming combustion condition. This identical setup was used in Benedict et al. (2024) which showed that at 1000 °C burn the black carbon mass fraction averaged 17 % for biomass fuels with a single scattering albedo of 0.35 (at 523 nm). We expect a similar smoke profile for the experiments presented here thus the smoke injected is a combination of soot, inorganic, and organic mass along with volatile vapors. Smoke particles were pushed to the chamber by zero-air at 4 L min⁻¹ for 5 min, a time window used to ensure complete combustion of the sample.

Aerosol size and number distributions downstream of the chamber were measured with a scanning mobility particle sizer (SMPS) that consists of a Differential Mobility Analyzer (3081 DMA, TSI Inc., USA) and a Condensation Particle Counter (3752, TSI Inc., USA). Measurement settings were set to continuously scan for 3 min per scan; 160 s recording with 20 s of purging, measuring sizes 15.7–764.5 nm. Our experimental matrix consisted of 8 repeats of NaCl, 4 repeats of sucrose and 6 smoke experiments with varying biomaterial mass, they are outlined in Supplement Information Table 1. In all experiments the first 6 h of data were used to analyze results.

The processing of data from the LANL chamber experiments involved two key steps to analyze the underlying aerosol processes of coagulation, wall loss, and dilution (chamber push line). First, we determined the observed size-dependent particle rates: dN(Dp)/dt. The measured size distributions were fitted to a two-mode lognormal distribution. The lognormal distribution parameters were optimized using the Python library SciPy's optimization routines, with the mean squared error as the cost function. We used multiple minimization methods and selected the best fit for each timestep based on the highest Pearson R-squared value with a minimum threshold of 0.85. The methods included Nelder-Mead (Simplex algorithm), Powell's method (Powell's conjugate direction method), L-BFGS-B (Limited-memory Broyden-Fletcher-Goldfarb-Shanno with Box constraints), TNC (Truncated Newton Conjugate-Gradient method), SLSQP (Sequential Least Squares Programming), and trust-constr (Trust Region Constrained method). We took this approach since the best fit varied with concentration and shape of the distribution. L-BFGS-B was typically the best for a lognormal distribution, but as the mode became broader (lower concentrations) then TNC, SLSQP or trust-constr would have a higher Pearson R-squared value. The transition of when this would occur was not an obvious concentration threshold. Therefore, we used all optimization routines for each lognormal distribution and selected the best fit based on the highest Pearson R-squared value.

Second, we fitted these observed rates to theoretical rates calculated from Particula (Particula, 2025), a python-based aerosol microphysics package. The first step was to generate a new time series at a higher size resolution (log-spaced 250 bins), starting at 20 nm and extrapolating the 746 nm SMPS upper limit to 4 µm. The size-dependent particle rate was then computed as the linear slope of 21 point moving window (10 before and 10 after). The time window (60 min) was chosen through iteration, as shorter than 20 min had too much noise to have self-consistent results and longer than 90 min had increasing fit residuals. Our 60 min window results in a smoothed time evolution, which Mahfouz and Donahue (2020a) showed to be effective in coagulation analysis. Our moving window approach is different from smog chamber wall-loss experiments where the full 5 h of the wall-loss experiment would be used to fit an apparent size-dependent, time-invariant wall-loss correction (Wang et al., 2018).

The resulting size-dependent rate was subsequently used to fit the underlying aerosol processes in Eq. (1) where NDp represents the number concentration of particles of diameter, Dp, K12 is the coagulation kernel, WC-1 is the coagulation correction factor, N1 and N2 are the concentrations of particles in the bins for K12, kflow is the chamber flow coefficient, and β is the wall-loss rate. 1dNDpdt=WC-1K12N1N2-kflowNDp-βNDp

The coagulation term is governed by a Brownian Coagulation kernel, K12, that captures the collision frequency between bin number concentrations (N1 and N2). This kernel is described in Seinfeld and Pandis (Seinfeld and Pandis, 2016; Sect. 13 – Fuchs form with alpha efficiency 13.56), and calculated with Particula. Since K12 does not account for other interactions (e.g. Coulomb interactions) that may lead to coagulation, WC-1, the coagulation correction factor, was determined. In our analysis, WC-1 is a free fit parameter to allow for un-modeled behaviors to be represented. The chamber flow coefficient, kflow=Q/V, characterizes how the clean air flow rate (Q) is used to push sample flow out of the chamber volume (V). Finally, the wall-loss term, βNDp, accounts for the size-dependent removal of particles to the chamber walls. 2β=1LWH(4HL+WkeDπ+vpLW×coth⁡πvp4keD)

Equation (2) shows the wall-loss rate (β) varies with particle size, derived from a rectangular-chamber formulation adapted from Crump and Seinfeld (1981) and Crump et al. (1982). It incorporates both diffusion-driven transport and gravitational settling. In this formulation, L, W, and H denote the chamber's length, width, and height, respectively; ke is the eddy wall diffusivity (a free fit parameter); D is the particle diffusion coefficient; and vp is the particle gravitational settling velocity. This physics-based wall-loss coefficient is different from Wang et al.'s (2018) method of apparent size-dependent wall-loss fit. In the apparent size-dependent wall-loss fit the rate equation is a two-term first-order rate equation, where there are no physical terms for the size of the chamber or particle settling velocity, in contrast to what we use in Eq. (2). The apparent size-dependent wall-loss approach is common for smog chamber experiments (Doussin et al., 2023; Keywood et al., 2004; Loza et al., 2012; Nah et al., 2017; Ng et al., 2007) but would not work here since one of our goals is to specifically determine coagulation. In our case, we need a physics-based wall-loss rate equation, so we can determine if there are any coagulation corrections that could be applied. If we had used the apparent size-dependent wall-loss fit, then there would be little to no residuals for a coagulation correction analysis.

If the corrected aerosol volumes remain consistent, within the noise measurement, we can infer that the processes described in Equation 1 accurately represent chamber behavior. Measured volume concentrations in our experiments were corrected by accounting for volume losses due to both wall loss and chamber flow. We calculated the cumulative lost volume and added it back to the measured values at each time point. In smog chamber experiments involving secondary organic aerosol formation, this volume conservation analysis provides a constraint on organic aerosol yields. Supplemental Fig. S8 shows an example of our volume conservation plot from a smoke injection experiment. From this analysis, we conclude that volume is conserved and that no measurable condensation of biomass burning organic vapors occurs under our experimental conditions.

We show, in Fig. 3, the L-BFGS-B optimization routine that was used on Eg. (1) for experimental data from the smoke aerosol generated by combusting Kentucky bluegrass. Figure 3a shows the lognormal-fitted size distribution for the entire experiment, where particle growth is evident as the mode diameter shifts to larger sizes over the six-hour period. Figure 3b breaks down the observed rates after 1.5 h into three calculated, time-varying, size-dependent components: coagulation, dilution, and wall-loss. At that time, coagulation dominates, reducing particles around 100 nm (∼ 0.16 cm⁻³ s⁻¹) and forming larger particles around 200 nm. From these fits we are specifically interested in the kernel correction factor to better understand the importance of agglomeration of freshly emitted BC fractal-like particles and how it changes in time.

In Fig. 4a we show the average wall-loss rates for our three different aerosol types based on 4–8 experiments each. Only results with valid optimizations and an R-squared above 0.85 were included. In the sucrose experiments, this filter led to data gaps during the later time periods (2–6 h) for inclusion in the analysis. To better compare with smoke, we conducted additional NaCl experiments to have a more complete time series for one of the comparisons. The wall-loss rates during the first hour (< 1 h) follow a similar trend apart from NaCl starting at a low wall-loss rate then rising close to a rate of 2 s⁻¹. These initial wall-loss rates are consistent with the general observation in chamber studies that early mixing processes and injection conditions can dominate particle loss. Typical ranges reported in smog chamber experiments span from < 1 s⁻¹ to tens of s⁻¹ depending on injection flow and the use of a fan (Zong et al., 2023), particle species (Li et al., 2017; Wang et al., 2011), and chamber geometry (Wang et al., 2011). Over longer times (> 1 h), all three aerosol types converge toward similar wall-loss rates (0.1 s⁻¹), in agreement with the literature indicating that chamber turbulence diminishes over time as mixing subsides.

Figure 4b shows the statistical distribution of the wall-loss rates for each aerosol type during the first hour and the subsequent five hours. NaCl and sucrose do not exhibit a large variance in diffusivity for the first hour compared to smoke which is 1.12 ± 1.55 s⁻¹. NaCl, sucrose, and smoke show mean wall-loss rates of 0.562 ± 0.975, 0.233 ± 0.286, and 0.201 ± 0.267 s⁻¹, respectively. This convergence to relatively similar values is consistent with past observations in smog chamber experiments, where turbulent mixing dissipates, and the system approaches a quasi-steady loss rate such as the CMU Teflon chamber (Mahfouz and Donahue, 2020b; Wang et al., 2018), the CESAM chamber (Wang et al., 2011), and the AIR chamber (Zong et al., 2023). However, NaCl and sucrose experiments display greater variability than smoke, likely due to residual chamber turbulence stemming from their distinct generation methods (aerosolization vs. combustion).

To investigate the influence of interparticle forces on aerosol coagulation, we fitted a coagulation correction factor that would account for van der Waals forces, shape, and/or Coulomb interactions in the coagulation rate. When WC-1=1, collisions are effectively “elastic,” with no net enhancement or inhibition. In contrast, WC-1>1 indicates that coagulation is enhanced (e.g. due to attractive forces, favorable particle morphology, or turbulence), whereas WC-1<1 implies reduced coagulation (e.g. electrostatic repulsion or other inhibiting effects).

In Fig. 5a, the smoke experiments show an initial period where WC-1>1, which may be explained by the fractal nature of soot aggregates that can promote sticking or chain formation upon collision. By the third hour in all experiments, accounting for the variation the average coagulation corrections extend above and below 1. During this later phase, particle concentrations (< 10⁴cm⁻³) no longer sustain significant coagulation losses, consistent with prior studies showing that coagulation becomes negligible under lower concentration conditions (Hussein et al., 2009; Mahfouz and Donahue, 2020b; Yu et al., 2022).

Figure 5b shows the distribution of coagulation corrections for these time periods. All three aerosols show a mean WC-1 value around 1 (0.969 ± 0.524 for NaCl, 1.16 ± 1.38 for sucrose, and 1.23 ± 0.312 for smoke), suggesting a slight repulsion or negligible net sticking among particles. However, the standard deviations do encompass WC-1=1. Smoke exhibits a slightly higher coagulation corrections initially followed by reduced values (0.941 ± 0.307) in later periods. These observations align with the notion that both particle morphology (e.g., fractal soot structures) and injection-induced turbulence can transiently enhance coagulation, but the effect diminishes as particles coagulate.

To understand the large standard deviations that emerged from our fits of wall loss and coagulation correction, we performed a Monte Carlo error analysis. We began by constructing three number-size distributions, each formed by the sum of two log-normal modes with equal particle numbers. In the first case both modes were centered at 100 nm but differed in geometric standard deviation (GSD), i.e., the distribution is summation of a 100 nm mode with a GSD of 1.4 plus a 100 nm mode with a GSD of 1.8. The resulting distribution reflects the broad distributions we observe in our measurements. The second case repeated this structure at 200 nm. The third case was a hypothetical experiment that combined narrow 100 and 300 nm modes (both GSD = 1.2) to test the response to a bimodal aerosol distribution.

For every distribution we calculated Eq. (1) assuming a wall-eddy diffusivity of 0.1 s⁻¹ and a coagulation correction factor (WC-1) of 1.0. This is a null case in which no additional correction to the Brownian coagulation kernel is required. We then superimposed random noise of ± 20 % on both the size spectrum and the rate. This noise mirrors uncertainties reported in instrument intercomparisons of ± 10 % error between 20 and 200 nm and up to ± 30 % above 200 nm (Wiedensohler et al., 2012). Thus, ± 20 % is a middle point across the range we measured. Applying the same noise to the rate represents the best-case scenario for our analysis pipeline.

With these noisy data sets created, we refit the wall-eddy diffusivity and coagulation correction 80 times at each total number concentration shown in Fig. 6. From the ensemble of fits we calculated the percent error in each retrieved parameter and averaged the results (Fig. 6). The results in Fig. 6 reveal a clear trend in percent error. When total number concentration exceeds roughly 10⁴ cm⁻³, the uncertainty in the coagulation correction begins to fall. This is consistent with the fact that Brownian coagulation scales with particle number squared and can be distinguished from measurement noise only at higher concentrations. Conversely, the error in the wall-eddy diffusivity grows with concentration. Once coagulation dominates the particle loss budget, the data contains too little information to constrain the comparatively low wall loss sink, increasing the relative uncertainty. In other words, when coagulation governs the system dynamics, the wall-loss term becomes a minor, poorly resolved correction.

The three analyzed distributions exhibit similar percent errors in the coagulation correction. The slightly lower error for the 100 nm mode compared to the 200 nm mode is consistent with the behavior of the Brownian coagulation kernel, where smaller particles have higher coagulation coefficients and therefore undergo more frequent collisions. This leads to a greater rate of change in the distribution for a given number concentration, resulting in better signal-to-noise. The hypothetical bimodal distribution generally shows the lowest uncertainty among the three cases (in our experimental range), although the improvement is modest.

Annotations in Fig. 6 mark the concentration ranges for the three chamber campaigns, sucrose, NaCl, and smoke aerosols. They also indicate the measured coefficient of variation in the mean coagulation correction for each case. The agreement between these annotated uncertainties and the Monte Carlo error analysis confirms that the observed variability is consistent with the measurement noise.