Imagine a population of atmospheric aerosol particles containing two distinct chemical species: black carbon (which absorbs light and warms the atmosphere) and sulfate (which reflects light and absorbs water). How are these chemicals distributed across the population?
Historically, climate models relied on binary classifications to answer this:
- External Mixing: Every particle is chemically pure. Some particles are 100% black carbon, while others are 100% sulfate.
- Internal Mixing: Every particle is an identical, uniform blend of the entire population's chemical makeup.
In reality, the atmosphere is messy. Processes like coagulation, condensation, and chemical aging create a continuous spectrum of intermediate states. As shown by Gao et al., treating these states simplistically introduces massive errors into climate simulations because a particle's exact chemical arrangement dictates how it interacts with sunlight and water vapor. To build better models, researchers needed a way to turn "mixing state" from a qualitative description into a precise, continuous number.
The Intuition of Diversity Measures
To solve this problem, Riemer introduced an elegant framework adapted from ecological diversity and information theory. Instead of counting biological species in an ecosystem, this framework measures the "diversity" of chemical components within a population of particles.
The framework relies on two core metrics calculated using Shannon entropy:
- Average Per-Particle Diversity ($D_\alpha$): The effective number of chemical species found inside a single particle, averaged across the population.
- Bulk Population Diversity ($D_\gamma$): The effective number of chemical species present across the entire population combined.
To understand how these two numbers define a mixing state, imagine a box of colored markers:
- The Externally Mixed Box: Every marker is a single solid color, but the box contains five different colors. Here, the individual diversity $D_\alpha$ is 1 (each marker has only one color), while the bulk diversity $D_\gamma$ is 5.
- The Internally Mixed Box: Every single marker is identical, manufactured with five multi-colored stripes. Here, the individual diversity $D_\alpha$ is 5, which perfectly matches the bulk diversity $D_\gamma$ of 5.
The key insight established by Riemer is that a population's mixing state is fundamentally determined by the relationship between these two diversities. By calculating an affine ratio of these measures, the author defined the mixing state index ($\chi$):
$$\chi = \frac{D_\alpha - 1}{D_\gamma - 1}$$
This index maps complex, high-dimensional chemical distributions onto a simple, intuitive scale from 0% to 100%:
- $\chi = 0\%$ (Fully External): Each particle is 100% pure ($D_\alpha = 1$), meaning all diversity exists between particles rather than within them.
- $\chi = 100\%$ (Fully Internal): Every particle has an identical composition ($D_\alpha = D_\gamma$).
Validating with Real-World Data
While theoretically elegant, calculating $\chi$ requires detailed knowledge of individual particle compositions. Healy et al. demonstrated that this information-theoretic framework could be applied directly to real-world data collected via single-particle mass spectrometry. By converting high-dimensional mass spectra into Shannon entropy metrics, the authors successfully mapped real-world aerosol populations onto the 0% to 100% mixing state scale, providing a standardized tool for atmospheric field measurements.
Why Mixing State Matters for Physical Predictions
Why go through the mathematical trouble of calculating $\chi$? Because assuming a population is perfectly internally mixed when it is actually externally mixed leads to severe predictive failures in environmental models.
For example, Ching et al. investigated how mixing state affects predictions of cloud condensation nuclei (CCN)—the particles around which cloud droplets form. They found that for highly internally mixed populations (high $\chi$), ignoring the exact mixing state has negligible consequences. However, for externally mixed populations, ignoring the mixing state causes errors in CCN predictions ranging from -40% to +150%.
Similarly, Ching et al. used particle-resolved models to study the "nucleation-scavenging" of black carbon—the process by which black carbon is washed out of the atmosphere by rain. They discovered that assuming a simplified, fully internal mixing state leads to massive overestimations of how quickly black carbon is removed, which could cause climate models to underestimate its atmospheric warming effects.
Scaling Up via Machine Learning
Despite its utility, tracking individual particles in global climate models (GCMs) is computationally prohibitive. GCMs typically rely on simplified "modal" representations that group aerosols into broad categories. As Zheng et al. demonstrated, these modal models suffer from significant structural uncertainty and struggle to accurately capture the co-occurrence of absorbing and hygroscopic species.
To bridge this gap, Hughes et al. developed machine learning models trained on high-fidelity, particle-resolved simulations. These models can accurately predict the global distribution of the mixing state index ($\chi$) using standard macro-scale variables already present in climate models. This allows global simulations to benefit from the accuracy of particle-resolved metrics without the prohibitive computational cost.
Parallel Frameworks in Fluid Dynamics
The core challenge of moving past binary definitions of homogeneity is not unique to atmospheric science. Other fields have developed complementary mathematical tools to quantify mixing states along a continuous spectrum.
For instance, in fluid mechanics, simple stirring can stretch a solute into fine, complex filaments without actually achieving molecular-level blending. To differentiate between these states, Kabacaoğlu et al. utilized the negative index $H^{-1}$ Sobolev norm. This metric quantifies mixing by measuring how concentration fluctuations decay across different spatial scales, providing a rigorous way to evaluate how structures like vesicle suspensions alter solute transport.
In industrial engineering contexts, such as nuclear reactor venting systems, Choi et al. showed that the standard deviation of species concentrations across a cross-section can serve as a practical threshold metric for homogeneity.
Whether through Shannon entropy, Sobolev norms, or spatial variance, the underlying philosophy remains identical: replacing binary assumptions of "mixed" versus "unmixed" with rigorous, scale-aware metrics is essential for predicting the behavior of complex, multi-component systems.
Key Foundational Work
Quantifying aerosol mixing state with entropy and diversity measures (2013) — contributed 7 concept elements to this direction.
Single particle diversity and mixing state measurements (2014) — contributed 3 concept elements to this direction.