# Aerosol Physics

$$
\renewcommand{\d}[1]{\mathrm{d} #1}
\renewcommand{\vec}[1]{\mathbf{#1}}
\newcommand{\ddt}[1]{\frac{d #1}{dt}}
\newcommand{\partialt}[1]{\frac{\partial #1}{\partial t}}
\renewcommand{\div}[1]{\nabla \cdot #1}
\newcommand{\dsub}[1]{_{_{#1}}}
\newcommand{\specidx}{L}
\newcommand{\amass}[1]{q\dsub{m,#1,i}}
\newcommand{\aitmass}[1]{q_{_{m,#1,2}}}
\newcommand{\vmass}[1]{q\dsub{v,#1}}
\newcommand{\mw}[1]{M\dsub{w,#1}}
\newcommand{\rsag}{\rm H_{_2}SO_{_4}}
\newcommand{\rasf}{\rm SO_{_4}}
\newcommand{\rwater}{\rm H_{_2}O}
\newcommand{\sag}{H_{_2}SO_{_4}}
\newcommand{\asf}{SO_{_4}}
\newcommand{\water}{H_{_2}O}
\newcommand{\numc}{N_{_{t,i}}}
\newcommand{\pluseq}{\mathrel{+}=}
\newcommand{\minuseq}{\mathrel{-}=}
\DeclareMathOperator{\sgn}{sgn}
\newcommand{\dsub}[1]{_{_{#1}}}
\newcommand{\specidx}{L}
\newcommand{\amass}[1]{q\dsub{m,#1,i}}
\newcommand{\vmass}[1]{q\dsub{v,#1}}
\newcommand{\mw}[1]{M\dsub{w,#1}}
\newcommand{\rsag}{\rm H_2SO_4}
\newcommand{\sag}{H_2SO_4}
\newcommand{\rasf}{\rm SO_4}
\newcommand{\asf}{SO_4}
$$

Here we describe the physics of aerosols as represented by the modal aerosol
model. First, we review the modal aerosol approximation, which defines the
equations and quantites of interest and describes how various aerosol processes
are parameterized. Then we describe in detail these various parameterized
aerosol processes, their physical significance, and their mathematical
represention. Along the way, we take care to highlight any physical,
statistical, and mathematical assumptions, to make clear the circumstances under
which Haero's aerosol model is valid.

## The Modal Aerosol Approximation

Here we offer an extremely abbreviated "fly-over" of the evolution equations
that quantify the dynamics of aerosols. Here, we emphasize the mathematical
representation of aerosols. For a more detailed explanation of the underlying
ideas, we refer the reader to~\cite{Whitby1991} and~\cite{Friedlander1977}.

All quantities are in SI units unless specifically mentioned. To keep the
discussion simple and focused, we avoid referencing specific coordinates.

Many of the assumptions here are inherited from Dick Easter's Modal Aerosol
Model 4-mode (MAM4) model.

## Size Matters

The modal approach to modeling aerosols is based on the observations that

* the sizes of aerosol particles greatly influence their dynamical behavior
* these sizes span several orders of magnitude (from 0.001 to 100 microns)

Here, a "particle" is an individual aerosol molecule with some volume $V_p$.
Aerosol particles consist of polymers, and a particle consisting of a chain of
$i$ monomers is an $i$-mer.

How do we represent a set of aerosol particles in space, given the importance
of their size? In the simplest case (for small $i$-mers), we can denote
$N_i(\vec{x}, t)$ as the number of $i$-mers per cubic meter in the vicinity of
the point $\vec{x}$ at time $t$. In this language, the evolution equation is a
set of coupled advection-diffusion-reaction (ADR) equations acting under the
influence of a bulk velocity $\vec{v}$:

$$
\begin{equation}
  \partialt{N_i} + \nabla\cdot{(N_i\vec{v})} = \mathcal{D}(\nabla N_i) +
                                        \mathcal{R}(\{N_j\}) +
                                        \mathcal{S}(N_i, t)
  \label{small_imer_dNdt}
\end{equation}
$$

Here we represent terms on the right hand side by

* $\mathcal{D}$: terms related to {\it diffusion processes} for $N_i$
* $\mathcal{R}$: terms related to {\it reaction processes}, in which
  various $j$-mers combine, react, and dissociate to form particles of
  other types
* $\mathcal{S}$: terms related to {\it source and sink processes},
  including external and prescribed sources of $N_i$.

We describe the forms of these terms in [Processes](processes.md). The bulk
velocity $\vec{v}$ is supplied by some dynamical atmospheric model.

For larger $i$-mers ($i > 100$), the above description becomes inadequate, and
we must adopt a description for particle numbers that admits a continuous
number distribution in particle size space.

Let $n(V_p, \vec{x}, t)$ be the number of particles of volume $V_p$ occupying
the point $\vec{x}$ at time $t$. Then the evolution of all $i$-mers is described
by a single ADR equation:

\begin{equation}
  \partialt{n} + \div{(n\vec{v})} = \mathcal{D}(\nabla n) +
                                    \mathcal{R}(n) +
                                    \mathcal{S}(n, t)
  \label{continuous_dndt}
\end{equation}

where $\mathcal{D}$, $\mathcal{R}$, and $\mathcal{S}$ have been expressed
in terms of $n$ instead of $N_i$. This is a tidy equation, but $n$ is a
number distribution function that introduces a new dimension---the particle volume
$V_p$---to the solution space. This makes it inconvenient for doing numerical
calculations. In this description, $n$ can assume any shape in particle volume
space, which raises the question of how to constrain the solution in $V_p$.

## Moment Equations and the Closure Problem
We can reduce the size of our solution space by making a few assumptions:

!!! info "Assumption"
    The detailed structure of the number distribution function
    $n(V_p, \vec{x}, t)$ is unimportant to aerosol dynamics.

If we don't need to obtain the full solution for $n$, we can select a specific
functional form $n(D_p, \vec{x}, t) = n(\vec{x}, t; D_p)$ for it. Then, taking cues from
methods in particle kinetics and turbulence theory, we can integrate the product
of \refeq{continuous_dndt} with powers of $D_p$ to obtain the {\bf moment
equations}

\begin{equation}\label{moments}
  \partialt{\mathcal{M}_k} + \div{(\mathcal{M}_k\vec{v})} = \int_0^{\infty} D_p^k (\mathcal{D} + \mathcal{R} + \mathcal{S}) \d{D_p}
\end{equation}

where $\mathcal{M}_k(\vec{x}, t) = \int_0^{\infty} D_p^k n(\vec{x}, t; D_p) \d{D_p}$
is the ``$k$th moment'' of $n$.

The moment equations are solved by picking a specific form of $n$'s functional
dependence on $D_p$ and playing tricks to avoid actually evaluating the
above integrals. However, the moment equations aren't closed---the advection,
diffusion, reaction, and source terms can all involve $n$ and its spatial
derivatives, and in general, the evolution of $\mathcal{M}_k$ is coupled to higher
moments. To make further progress, we must solve this {\bf closure problem}.

## Modal Equations

Having already given up on obtaining a general solution for $n(\vec{x}, t)$,
we allow ourselves to make another assumption:

!!! info "Modal assumption"
    The number distribution function $n$ is the sum of a set of
    specific number distribution functions $n_i$, each representing a
    {\bf mode} with a specific functional form for a sub-population of
    aerosol particles occupying a certain range in particle size space.

In other words,

\begin{equation}
  n(\vec{x}, t; D_p) = \sum_{i=1}^M n_i(\vec{x}, t; D_p)
  \label{modal_n}
\end{equation}

where $n_i$ represents aerosol particles with sizes falling within the range
of mode $i$ and $M$ is the number of modes. Each mode assumes a specific
functional form for $n_i$ in terms of its relevant size as given by $D_p$.
We include the arguments $\vec{x}$ and $t$ to emphasize that this equation holds
at each point in space and at each instant in time for an aerosol system.

### Log-Normal Distribution Functions

Haero constructs its multi-modal distribution functions from log-normal
probability distribution functions (PDFs) for each mode $i$. The PDF for mode
$i$ expresses the fraction of aerosol particles present per unit size interval,
in terms of a continuous particle diameter $D_p$ within that mode. Such PDFs
have been found to represent measured aerosol size distributions with a level of
accuracy comparable to that of the relevant measurement
techniques~\cite{Whitby1991}.

!!! info "Log-normal PDF"
    The probability distribution function $f_i$ used to construct
    the number distribution function $n_i$ for mode $i$ is a continuous
    function of the particle diameter $D_p$, given as a log-normal distribution,
    \begin{equation}\label{log_normal_pdf}
      f_i(D_p) = \frac{1}{\sqrt{2\pi} D_p \ln \sigma_{g,i}} \ 
      \exp \left [-\frac{(\ln D_p - \ln D_{g,i})^2}{2\ln^2 \sigma_{g,i}} \right]
    \end{equation}
    where we have introduced the geometric mean $D_{g,i}$ and the geometric standard
    deviation $\sigma_{g,i}$ of $D_p$ within mode $i$.

In general, these two parameters are determined for each mode via moment equations and corresponding time-evolution equations.
Haero makes an additional simplifying assumption:

!!! info "Constant standard deviation"
    The geometric standard deviation $\sigma_{g,i}>1$ for each mode $i$ is constant in time.  
    The justification for this assumption in \cite{Easter2004,Wilson2001,Whitby1991} is given in \cite{Whitby1981}.

This PDF can also be expressed in "logarithmic form" in terms of $\ln D_p$
of $D_p$:

\begin{equation}\label{log_normal_pdf_log}
  g_i(\ln D_p) = \frac{1}{\sqrt{2\pi} \ln\sigma_{g,i}} \
      \exp \left [ -\frac{(\ln D_p - \ln D_{g,i})^2}{2\ln^2 \sigma_{g,i}} \right]
\end{equation}

Here, $g_i$ is a normal distribution of $\ln D_p$ with a mean of $\ln D_{g,i}$
and standard deviation of $\ln\sigma_{g,i}$.

These PDFs are referred to as {\bf normalized size distribution functions} in
Haero. We can justify this name by multiplying each PDF by the total particle
number density $N_i$ for mode $i$ to obtain the equivalent number distribution
functions:

\begin{align}\label{log_normal_n}
  n_i(D_p) = N_i f_i(D_p) &= \frac{N_i}{\sqrt{2\pi} D_p \ln \sigma_{g,i}} \ 
  \exp \left [ - \frac{(\ln D_p - \ln D_{g,i})^2}{2\ln^2\sigma_{g,i}} \right ] \\
  \hat{n}_{i}(\ln D_p) = N_i g_i(\ln D_p) &= \frac{N_i}{\sqrt{2\pi}\ln \sigma_{g,i}} \ 
	\exp \left [ - \frac{(\ln D_p - \ln D_{g,i})^2}{2\ln^2\sigma_{g,i}} \right ]
\end{align}

![Normalized size distributions $n_i$ for each mode in the Haero model
  with $N_i$ set by the \texttt{num\_aer} variable.](images/pdf_examples.png)

The logarithmic form $\hat{n}_i$ is useful because of the relationship
$ n_i(D_p) = \hat{n}_i(\ln D_p)/D_p$, which allows us to write

\begin{equation} \label{equ_norm_lognorm}
  n_i(D_p) \d{D_p} = \hat{n}_i(\ln D_p) \d{\ln D_p}
\end{equation}

to simplify integrands involving number distribution functions.
The benefit of this log-normal functional form choice is that the moment integrals can be defined analytically, obviating the need for potentially costly numerical quadrature \cite[eqns.~(3.7)]{Whitby1991}.
The $k$th moment is given in this case by,

\begin{equation}
  M_k^{(i)} = N_i D_{g,i}^k \exp\left(\frac{k^2}{2}\log^2\sigma_{g,i}\right).
\end{equation}

In particular,

\begin{align}
  M_0^{(i)} &= \int_0^\infty n_i(D_p)\,d D_p = N_i, \label{eq:zero_moment}\\
  M_1^{(i)} &= \int_0^\infty D_p n_i(D_p)\,d D_p = N_iD_{g,i}\exp({\frac{1}{2}\log^2\sigma_{g,i}}),\\
  M_2^{(i)} &= \int_0^\infty D_p^2n_i(D_p)\,d D_p = N_iD_{g,i}^2\exp(2\log^2\sigma_{g,i}),\\
  M_3^{(i)} &= \int_0^\infty D_p^3n_i(D_p)\,d D_p = N_iD_{g,i}^3\exp(\frac{9}{2}\log^2\sigma_{g,i}) \label{eq:third_moment}
\end{align}

!!! info "Particle size"
    Particle size $\overline{D_i}$ for the $i$th mode is defined by the mean
    diameter associated with the 3rd moment \cite[eqn.~(1)]{Whitby1981},
    \begin{equation}
      \overline{D_i} = \left(M_3^{(i)}/N_i)\right)^{1/3} = D_{g,i}\exp\left(\frac{3}{2}\log^2\sigma_{g,i}\right)
    \end{equation}

!!! info "Particles have spherical shapes"
    Aerosol particles are spherical, and their size can be parameterized by
    their diameter $\overline{D_i}$. Corollaries for area and volume follow
    ($\overline{A_i} = \pi\overline{D_i}^2$,
     $\overline{V_i} = \frac{\pi}{6}\overline{D_i}^3$).

When we adopt the log-normal assumption, we express our solution for particles in
mode $i$ in terms of its zeroth, first, second, and third moments
$\mathcal{M}_k^{(i)}(\vec{x}, t)$. These are, respectively:

* $\mathcal{M}_0^{(i)} = N_i$, the total number of concentration for particles
  in mode $i$
* $\mathcal{M}_1^{(i)} = N_i\overline{D}_i$, the number-weighted mean diameter of
  particles in mode $i$
* $\mathcal{M}_2^{(i)} = \frac{1}{\pi}N_i\overline{A}_i$, a number-weighted
  surface area of particles in mode $i$
* $\mathcal{M}_3^{(i)} = \frac{6}{\pi}N_i\overline{V}_i$, a number-weighted
  volume of particles in mode $i$

In this language, the aerosol evolution equation for the $k$th moment of the
$i$th mode is

\begin{equation}\label{modal_evolution}
  \partialt{\mathcal{M}_k^{(i)}} = F_k^{(i)}(N_i, \overline{D}_i, \overline{A}_i, \overline{V}_i, T, p, \mathsf{...})
\end{equation}

where the air temperature $T$ and the air pressure $p$ enter the moment
equations, via right hand side terms in \ref{moments}. These are the
**modal equations**.

!!! info "Parmeterization assumption"
    The time evolution of a moment $\mathcal{M}_k^{(i)}$ can be
    expressed in terms of the quantities $N_i$, $\overline{D}_i$,
    $\overline{A}_i$, $\overline{V}_i$, the air temperature $T$, the pressure $p$, and a few
    selected additional quantities.

!!! info "2-moment scheme"
    Aerosol dynamics can be represented well by two moments, $M_0^{(i)}$ and
    $M_3^{(i)}$. We must find equations that describe the time evolution of
    $M_0^{(i)}$ and $M_3^{(i)}$ in terms of known E3SM variables.

    Modal number concentrations $N_m$ are defined and transported for each
    mode [\# / kg air] as prognostic model variables (tracers).
    These are the $\mathcal{M}_0$ moments for each mode, \ref{zero_moment}.

    Mass mixing ratios $q_{s,m}$ are defined for each species $s$ in each mode
    $m$ [ kg aerosol / kg air ] as prognostic model variables (tracers).
    Therefore, the third moments $\mathcal{M}_3$ of each mode are found using
    the density of each aerosols species.

    Volume mixing ratios are diagnosed by $V_{s,m} = q_{s,m}/\rho_{s,m}$
    [m$^3$ aerosol / kg air], where $\rho_{s,m}$ is the constant species
    density [kg aerosol / m$^3$].

## Mode definitions in MAM4

The choices of size range and width for the modes in MAM are based on
measurements of tropospheric aerosols (see~\cite{Easter2004} and references
therein).

Here are the relevant parameters for MAM4, the 4-mode legacy MAM model.

| Mode           | Lower bound of $D\dsub{gn,d,i}$ | Upper bound of $D\dsub{gn,d,i}$ | $\sigma\dsub{g,i}$ | $N_i(0)$ |
| -------------- | ------------------------------- | ------------------------------- | ------------------ | -------- |
| Aitken         | 0.0087 | 0.052 | 1.6 | 7.9031E7 |
| Accumulation   | 0.0535 |  0.44 | 1.8 | 7.9697E7 |
| Coarse         |    1.0 |   4.0 | 1.8 | 8.0361E7 |
| Primary carbon |   0.01 |   0.1 | 1.6 | 8.1022E7 |

Thus, in MAM, all modes have the same mathematical form, but the parameters
are different for each mode.

## Multi-Species Modes

We have derived the modal equations assuming that each mode contains a
population of undifferentiated aerosol particles. We assume that each particle
is composed of several (potentially different) chemical species:

!!! info "Internally mixed"
    All aerosol species within a mode are assumed to be carried uniformly by all
    particles represented by the mode's log-normal distribution function, i.e.,
    within each mode, particles are internally mixed.

!!! info "Additive particle volume"
    Quantitatively, the internal mixing assumption implies that the volume of
    each individual particle is the sum of the volumes of its various
    constituents; in this case, the species within the mode.

These assumptions lead to the following modal averages.

!!! info "Mode aerosol mass mixing ratio"
    The mass mixing ratio of species $s$ in mode $m$ is denoted by $q_{sm}$.
    The mass mixing ration of the mode [kg aerosol / kg air] is then
    \begin{equation}\label{mode_mixing_ratio}
      q_{m} = \sum_{s\in m} q_{sm}.
    \end{equation}

!!! info "Modally averaged aerosol density"
    The density of aerosol species $s$ in mode $m$ is denoted $\rho_{sm}$.
    The average modal density [kg aerosol / m$^3$ aerosol] is
    \begin{equation}\label{mode_density}
      \rho_m = \sum_{s\in m}q_{sm} \rho_{sm} \left(\sum_{s\in m} q_{sm}\right)^{-1} = \frac{\sum_{s\in m}q_{sm}\rho_{sm}}{q_m}.
    \end{equation}

!!! info "Modally averaged hygroscopicity"
    The hygroscopicity of species $s$ in mode $m$ is denoted $b_{sm}$.
    The mode-averaged hygroscopicity [-] is
    \begin{equation}\label{mode_hygro}
      b_m = \sum_{s\in m} \frac{q_{sm} b_{sm}}{\rho_{sm}}\left(\sum_{s\in m} \frac{q_{sm}}{\rho_{sm}}\right)^{-1}.
    \end{equation}

!!! info "Modal mean particle volume"
    The per-particle mean volume for each mode $m$ [m$^3$/particle] is
    \begin{equation}\label{mode_particle_volume}
      {V_{m}} = \frac{1}{N_m}\sum_{s\in m} \frac{q_{sm}}{\rho_{sm}}.
    \end{equation}

!!! info "Modal geometric mean particle diameter"
    The geometric mean diameter of the particles in mode $m$ [m] is
    \begin{equation}\label{mode_particle_diameter}
      {D_{m}} = \sqrt[3]{\frac{\pi}{6}{V_{Pm}}}\exp\left(-\frac{3}{2}\log^2 \sigma_m\right).
    \end{equation}