source: branches/nemo_v3_3_beta/DOC/TexFiles/Chapters/Annex_ISO.tex @ 2298

% ================================================================
% Iso-neutral diffusion : 
% ================================================================
\chapter{Griffies's iso-neutral diffusion}
\label{Apdx_C}
\minitoc

\section{Griffies's formulation of iso-neutral diffusion}

\subsection{Introduction}
 We define a scheme that get its inspiration from the scheme of
\citet{Griffies_al_JPO98}, but is formulated within the \NEMO
framework, using scale factors rather than grid-sizes.

The off-diagonal terms of the small angle diffusion tensor
\eqref{Eq_PE_iso_tensor} produce skew-fluxes along the
i- and j-directions resulting from the vertical tracer gradient:
\begin{align}
  \label{eq:i13c}
  &+\kappa r_1\frac{1}{e_3}\frac{\partial T}{\partial k},\qquad+\kappa r_2\frac{1}{e_3}\frac{\partial T}{\partial k}\\
\intertext{and in the k-direction resulting from the lateral tracer gradients}
  \label{eq:i31c}
 & \kappa r_1\frac{1}{e_1}\frac{\partial T}{\partial i}+\kappa r_2\frac{1}{e_1}\frac{\partial T}{\partial i}
\end{align}
where \eqref{Eq_PE_iso_slopes}
\begin{align*}
  r_1 &=-\frac{e_3 }{e_1 } \left( \frac{\partial \rho }{\partial i}
  \right)
  \left( {\frac{\partial \rho }{\partial k}} \right)^{-1} \\
  &=-\frac{e_3 }{e_1 } \left( -\alpha\frac{\partial T }{\partial i} +
    \beta\frac{\partial S }{\partial i} \right) \left(
    -\alpha\frac{\partial T }{\partial k} + \beta\frac{\partial S
    }{\partial k} \right)^{-1}
\end{align*}
is the i-component of the slope of the isoneutral surface relative to the computational
surface, and $r_2$ is the j-component.

The extra vertical diffusive flux associated with the $_{33}$
component of the small angle diffusion tensor is
\begin{equation}
  \label{eq:i33c}
  -\kappa(r_1^2 +r_2^2) \frac{1}{e_3}\frac{\partial T}{\partial k}.
\end{equation}

Since there are no cross terms involving $r_1$ and $r_2$ in the above, we can
consider the isoneutral diffusive fluxes separately in the i-k and j-k
planes, just adding together the vertical components from each
plane. The following description will describe the fluxes on the i-k
plane.

There is no natural discretization for the i-component of the
skew-flux, \eqref{eq:i13c}, as
although it must be evaluated at u-points, it involves vertical
gradients (both for the tracer and the slope $r_1$), defined at
w-points. Similarly, the vertical skew flux, \eqref{eq:i31c}, is evaluated at
w-points but involves horizontal gradients defined at u-points.

\subsection{The standard discretization}
The straightforward approach to discretize the lateral skew flux
\eqref{eq:i13c} from tracer cell $i,k$ to $i+1,k$, introduced in 1995
into OPA, \eqref{Eq_tra_ldf_iso}, is to calculate a mean vertical
gradient at the u-point from the average of the four surrounding
vertical tracer gradients, and multiply this by a mean slope at the
u-point, calculated from the averaged surrounding vertical density
gradients. The total area-integrated skew-flux from tracer cell $i,k$
to $i+1,k$, noting that the $e_{{3u}_{i+1/2}^k}$ in the area
$e_{{3u}_{i+1/2}^k}e_{{2u}_{i+1/2}^k}$ at the u-point cancels out with
the $1/e_{{3u}_{i+1/2}^k}$ associated with the vertical tracer
gradient, is then \eqref{Eq_tra_ldf_iso}
\begin{equation*}
  \left(F_u^{\mathrm{skew}} \right)_{i+\hhalf}^k = \kappa _{i+\hhalf}^k
  e_{{2u}_{i+1/2}^k} \overline{\overline
    r_1} ^{\,i,k}\,\overline{\overline{\delta_k T}}^{\,i,k},
\end{equation*}
where
\begin{equation*}
  \overline{\overline
    r_1} ^{\,i,k} = -\frac{e_{{3u}_{i+1/2}^k}}{e_{{1u}_{i+1/2}^k}}
  \frac{\delta_{i+1/2} [\rho]}{\overline{\overline{\delta_k \rho}}^{\,i,k}}.
\end{equation*}

Unfortunately the resulting combination $\overline{\overline{\delta_k
    \bullet}}^{\,i,k}$ of a $k$ average and a $k$ difference %of the tracer
reduces to $\bullet_{k+1}-\bullet_{k-1}$, so two-grid-point oscillations are
invisible to this discretization of the iso-neutral operator. These
\emph{computational modes} will not be damped by this operator, and
may even possibly be amplified by it.  Consequently, applying this
operator to a tracer does not guarantee the decrease of its
global-average variance. To correct this, we introduced a smoothing of
the slopes of the iso-neutral surfaces (see \S\ref{LDF}). This
technique works fine for $T$ and $S$ as they are active tracers
($i.e.$ they enter the computation of density), but it does not work
for a passive tracer.
\subsection{Expression of the skew-flux in terms of triad slopes}
\citep{Griffies_al_JPO98} introduce a different discretization of the
off-diagonal terms that nicely solves the problem.
% Instead of multiplying the mean slope calculated at the u-point by
% the mean vertical gradient at the u-point,
% >>>>>>>>>>>>>>>>>>>>>>>>>>>>
\begin{figure}[h] \begin{center}
    \includegraphics[width=0.90\textwidth]{./TexFiles/Figures/Fig_triad_fluxes}
    \caption{(a) Arrangement of triads $S_i$ and tracer gradients to
      give lateral tracer flux from box $i,k$ to $i+1,k$ (b) Triads
      $S'_i$ and tracer gradients to give vertical tracer flux from
      box $i,k$ to $i,k+1$.}
    \label{Fig_triad}
  \end{center} \end{figure}
% >>>>>>>>>>>>>>>>>>>>>>>>>>>>
They get the skew flux from the products of the vertical gradients at
each w-point surrounding the u-point with the corresponding `triad'
slope calculated from the lateral density gradient across the u-point
divided by the vertical density gradient at the same w-point as the
tracer gradient. See Fig.~\ref{Fig_triad}a, where the thick lines
denote the tracer gradients, and the thin lines the corresponding
triads, with slopes $s_1, \dotsc s_4$. The total area-integrated
skew-flux from tracer cell $i,k$ to $i+1,k$
\begin{multline}
  \label{eq:i13}
  \left( F_u^{13}  \right)_{i+\frac{1}{2}}^k = \kappa _{i+1}^k a_1 s_1
  \delta _{k+\frac{1}{2}} \left[ T^{i+1}
  \right]/e_{{3w}_{i+1}}^{k+\frac{1}{2}}  + \kappa _i^k a_2 s_2 \delta
  _{k+\frac{1}{2}} \left[ T^i
  \right]/e_{{3w}_{i+1}}^{k+\frac{1}{2}} \\
   +\kappa _{i+1}^k a_3 s_3 \delta _{k-\frac{1}{2}} \left[ T^{i+1}
  \right]/e_{{3w}_{i+1}}^{k+\frac{1}{2}}  +\kappa _i^k a_4 s_4 \delta
  _{k-\frac{1}{2}} \left[ T^i \right]/e_{{3w}_{i+1}}^{k+\frac{1}{2}},
\end{multline}
where the contributions of the triad fluxes are weighted by areas
$a_1, \dotsc a_4$, and $\kappa$ is now defined at the tracer points
rather than the u-points. This discretization gives a much closer
stencil, and disallows the two-point computational modes.

 The vertical skew flux \eqref{eq:i31c} from tracer cell $i,k$ to $i,k+1$ at the
w-point $i,k+\hhalf$ is constructed similarly (Fig.~\ref{Fig_triad}b)
by multiplying lateral tracer gradients from each of the four
surrounding u-points by the appropriate triad slope:
\begin{multline}
  \label{eq:i31}
  \left( F_w^{31} \right) _i ^{k+\frac{1}{2}} =  \kappa_i^{k+1} a_{1}'
  s_{1}' \delta _{i-\frac{1}{2}} \left[ T^{k+1} \right]/{e_{3u}}_{i-\frac{1}{2}}^{k+1}
   +\kappa_i^{k+1} a_{2}' s_{2}' \delta _{i+\frac{1}{2}} \left[ T^{k+1} \right]/{e_{3u}}_{i+\frac{1}{2}}^{k+1}\\
  + \kappa_i^k a_{3}' s_{3}' \delta _{i-\frac{1}{2}} \left[ T^k\right]/{e_{3u}}_{i-\frac{1}{2}}^k
  +\kappa_i^k a_{4}' s_{4}' \delta _{i+\frac{1}{2}} \left[ T^k \right]/{e_{3u}}_{i+\frac{1}{2}}^k.
\end{multline}
 
We notate the triad slopes in terms of the `anchor point' $i,k$
(appearing in both the vertical and lateral gradient), and the u- and
w-points $(i+i_p,k)$, $(i,k+k_p)$ at the centres of the `arms' of the
triad as follows (see also Fig.~\ref{Fig_triad}):
\begin{equation}
  \label{Gf_slopes}
  _i^k \mathbb{R}_{i_p}^{k_p} 
  =\frac{ {e_{3w}}_{\,i}^{\,k+k_p}} { {e_{1u}}_{\,i+i_p}^{\,k}}
  \
  \frac 
  {\left(\alpha / \beta \right)_i^k  \ \delta_{i + i_p}[T^k] - \delta_{i + i_p}[S^k] }
  {\left(\alpha / \beta \right)_i^k  \ \delta_{k+k_p}[T^i ] - \delta_{k+k_p}[S^i ] }.
\end{equation}
In calculating the slopes of the local neutral
surfaces, the expansion coefficients $\alpha$ and $\beta$ are
evaluated at the anchor points of the triad \footnote{Note that in \eqref{Gf_slopes} we use the ratio $\alpha / \beta$
instead of multiplying the temperature derivative by $\alpha$ and the
salinity derivative by $\beta$. This is more efficient as the ratio
$\alpha / \beta$ can to be evaluated directly}, while the metrics are
calculated at the u- and w-points on the arms.

% >>>>>>>>>>>>>>>>>>>>>>>>>>>>
\begin{figure}[h] \begin{center}
    \includegraphics[width=0.60\textwidth]{./TexFiles/Figures/Fig_qcells}
    \caption{Triad notation for quarter cells.T-cells are inside
      boxes, while the  $i+\half,k$ u-cell is shaded in green and the
      $i,k+\half$ w-cell is shaded in pink.}
    \label{qcells}
  \end{center} \end{figure}
% >>>>>>>>>>>>>>>>>>>>>>>>>>>>

Each triad $\{_i^k\:_{i_p}^{k_p}\}$ is associated (Fig.~\ref{qcells}) with the quarter
cell that is the intersection of the $i,k$ T-cell, the $i+i_p,k$
u-cell and the $i,k+k_p$ w-cell. Expressing the slopes $s_i$ and
$s'_i$ in \eqref{eq:i13} and \eqref{eq:i31} in this notation, we have
e.g.\ $s_1=s'_1={\:}_i^k \mathbb{R}_{1/2}^{1/2}$. Each triad slope $_i^k
\mathbb{R}_{i_p}^{k_p}$ is used once (as an $s$) to calculate the
lateral flux along its u-arm, at $(i+i_p,k)$, and then again as an
$s'$ to calculate the vertical flux along its w-arm at
$(i,k+k_p)$. Each vertical area $a_i$ used to calculate the lateral
flux and horizontal area $a'_i$ used to calculate the vertical flux
can also be identified as the area across the u- and w-arms of a
unique triad, and we can notate these areas, similarly to the triad
slopes, as $_i^k{\mathbb{A}_u}_{i_p}^{k_p}$,
$_i^k{\mathbb{A}_w}_{i_p}^{k_p}$, where e.g. in \eqref{eq:i13}
$a_{1}={\:}_i^k{\mathbb{A}_u}_{1/2}^{1/2}$, and in \eqref{eq:i31}
$a'_{1}={\:}_i^k{\mathbb{A}_w}_{1/2}^{1/2}$.

\subsection{The full triad fluxes}
A key property of isoneutral diffusion is that it should not affect
the (locally referenced) density. In particular there should be no
lateral or vertical density flux. The lateral density flux disappears so long as the
area-integrated lateral diffusive flux from tracer cell $i,k$ to
$i+1,k$ coming from the $_{11}$ term of the diffusion tensor takes the
form
\begin{equation}
  \label{eq:i11}
  \left( F_u^{11} \right) _{i+\frac{1}{2}} ^{k} =
  - \left( \kappa_i^{k+1} a_{1} + \kappa_i^{k+1} a_{2} + \kappa_i^k
    a_{3} + \kappa_i^k a_{4} \right)
  \frac{\delta _{i+1/2} \left[ T^k\right]}{{e_{1u}}_{\,i+1/2}^{\,k}},
\end{equation}
where the areas $a_i$ are as in \eqref{eq:i13}. In this case,
separating the total lateral flux, the sum of \eqref{eq:i13} and
\eqref{eq:i11}, into triad components, a lateral tracer
flux
\begin{equation}
  \label{eq:latflux-triad}
  _i^k {\mathbb{F}_u}_{i_p}^{k_p} (T) = -A_i^k{ \:}_i^k{\mathbb{A}_u}_{i_p}^{k_p}
  \left(
    \frac{ \delta_{i+ i_p}[T^k] }{ {e_{1u}}_{\,i + i_p}^{\,k} }
    -\ {_i^k\mathbb{R}_{i_p}^{k_p}} \
    \frac{ \delta_{k+k_p} [T^i] }{{e_{3w}}_{\,i}^{\,k+k_p} }
  \right)
\end{equation}
can be identified with each triad. Then, because the
same metric factors ${e_{3w}}_{\,i}^{\,k+k_p}$ and
${e_{1u}}_{\,i+i_p}^{\,k}$ are employed for both the density gradients
in $ _i^k \mathbb{R}_{i_p}^{k_p}$ and the tracer gradients, the lateral
density flux associated with each triad separately disappears.
\begin{equation}
  \label{eq:latflux-rho}
  {\mathbb{F}_u}_{i_p}^{k_p} (\rho)=-\alpha _i^k {\:}_i^k {\mathbb{F}_u}_{i_p}^{k_p} (T) + \beta_i^k {\:}_i^k {\mathbb{F}_u}_{i_p}^{k_p} (S)=0
\end{equation}
Thus the total flux $\left( F_u^{31} \right) ^i _{i,k+\frac{1}{2}} +
\left( F_u^{11} \right) ^i _{i,k+\frac{1}{2}}$ from tracer cell $i,k$
to $i+1,k$ must also vanish since it is a sum of four such triad fluxes.

The squared slope $r_1^2$ in the expression \eqref{eq:i33c} for the
$_{33}$ component is also expressed in terms of area-weighted
squared triad slopes, so the area-integrated vertical flux from tracer
cell $i,k$ to $i,k+1$ resulting from the $r_1^2$ term is
\begin{equation}
  \label{eq:i33}
  \left( F_w^{33} \right) _i^{k+\frac{1}{2}} =
    - \left( \kappa_i^{k+1} a_{1}' s_{1}'^2
    + \kappa_i^{k+1} a_{2}' s_{2}'^2
    + \kappa_i^k a_{3}' s_{3}'^2
    + \kappa_i^k a_{4}' s_{4}'^2 \right)\delta_{k+\frac{1}{2}} \left[ T^{i+1} \right],
\end{equation}
where the areas $a'$ and slopes $s'$ are the same as in
\eqref{eq:i31}.
Then, separating the total vertical flux, the sum of \eqref{eq:i31} and
\eqref{eq:i33}, into triad components,  a vertical flux 
\begin{align}
  \label{eq:vertflux-triad}
  _i^k {\mathbb{F}_w}_{i_p}^{k_p} (T)
  &= A_i^k{\: }_i^k{\mathbb{A}_w}_{i_p}^{k_p}
  \left(
    {_i^k\mathbb{R}_{i_p}^{k_p}}\frac{ \delta_{i+ i_p}[T^k] }{ {e_{1u}}_{\,i + i_p}^{\,k} }
    -\ \left({_i^k\mathbb{R}_{i_p}^{k_p}}\right)^2 \
    \frac{ \delta_{k+k_p} [T^i] }{{e_{3w}}_{\,i}^{\,k+k_p} }
  \right) \\
  &= - \left(\left.{ }_i^k{\mathbb{A}_w}_{i_p}^{k_p}\right/{ }_i^k{\mathbb{A}_u}_{i_p}^{k_p}\right)
   {_i^k\mathbb{R}_{i_p}^{k_p}}{\: }_i^k{\mathbb{F}_u}_{i_p}^{k_p} (T) \label{eq:vertflux-triad2}
\end{align}
may be associated with each triad. Each vertical density flux $_i^k {\mathbb{F}_w}_{i_p}^{k_p} (\rho)$
associated with a triad then separately disappears (because the
lateral flux $_i^k{\mathbb{F}_u}_{i_p}^{k_p} (\rho)$
disappears). Consequently the total vertical density flux $\left( F_w^{31} \right)_i ^{k+\frac{1}{2}} +
\left( F_w^{33} \right)_i^{k+\frac{1}{2}}$ from tracer cell $i,k$
to $i,k+1$ must also vanish since it is a sum of four such triad
fluxes.

We can explicitly identify (Fig.~\ref{qcells}) the triads associated with the $s_i$, $a_i$, and $s'_i$, $a'_i$ used in the definition of
the u-fluxes and w-fluxes in
\eqref{eq:i31}, \eqref{eq:i13}, \eqref{eq:i11} \eqref{eq:i33} and
Fig.~\ref{Fig_triad} to  write out the iso-neutral fluxes at $u$- and
$w$-points as sums of the triad fluxes that cross the $u$- and $w$-faces:
%(Fig.~\ref{Fig_ISO_triad}):
\begin{flalign} \label{Eq_iso_flux} \textbf{F}_{iso}(T) &\equiv
  \sum_{\substack{i_p,\,k_p}}
  \begin{pmatrix}
    {_{i+1/2-i_p}^k {\mathbb{F}_u}_{i_p}^{k_p} } (T)      \\
    \\
    {_i^{k+1/2-k_p} {\mathbb{F}_w}_{i_p}^{k_p} } (T)      \\
  \end{pmatrix}.
\end{flalign}
\subsection{Ensuring the scheme cannot increase tracer variance}
\label{sec:variance}

We now require that this operator cannot increase the
globally-integrated tracer variance.
%This changes according to
% \begin{align*}
% &\int_D  D_l^T \; T \;dv \equiv  \sum_{i,k} \left\{ T \ D_l^T \ b_T \right\}    \\
% &\equiv + \sum_{i,k} \sum_{\substack{i_p,\,k_p}} \left\{  
%     \delta_{i} \left[{_{i+1/2-i_p}^k {\mathbb{F}_u }_{i_p}^{k_p}} \right] 
%       + \delta_{k} \left[ {_i^{k+1/2-k_p} {\mathbb{F}_w}_{i_p}^{k_p}} \right]  \ T \right\}    \\
% &\equiv  - \sum_{i,k} \sum_{\substack{i_p,\,k_p}} \left\{  
%                 {_{i+1/2-i_p}^k {\mathbb{F}_u }_{i_p}^{k_p}} \ \delta_{i+1/2} [T]
%              + {_i^{k+1/2-k_p} {\mathbb{F}_w}_{i_p}^{k_p}}  \ \delta_{k+1/2} [T]   \right\}      \\
% \end{align*}
Each triad slope $_i^k\mathbb{R}_{i_p}^{k_p}$ drives a lateral flux
$_i^k{\mathbb{F}_u}_{i_p}^{k_p} (T)$ across the u-point $i+i_p,k$ and
a vertical flux $_i^k{\mathbb{F}_w}_{i_p}^{k_p} (T)$ across the
w-point $i,k+k_p$.  The lateral flux drives a net rate of change of
variance at points $i+i_p-\half,k$ and $i+i_p+\half,k$ of
\begin{multline}
  {b_T}_{i+i_p-1/2}^k\left(\frac{\partial T}{\partial t}T\right)_{i+i_p-1/2}^k+
  \quad {b_T}_{i+i_p+1/2}^k\left(\frac{\partial T}{\partial
      t}T\right)_{i+i_p+1/2}^k \\
 \begin{split}
  &= -T_{i+i_p-1/2}^k{\;} _i^k{\mathbb{F}_u}_{i_p}^{k_p} (T) \quad + \quad  T_{i+i_p+1/2}^k
  {\;}_i^k{\mathbb{F}_u}_{i_p}^{k_p} (T) \\
  &={\;} _i^k{\mathbb{F}_u}_{i_p}^{k_p} (T)\,\delta_{i+ i_p}[T^k], \label{locFdtdx}
 \end{split}
\end{multline}
while the vertical flux similarly drives a net rate of change of variance at points $i,k+k_p-\half$ and
$i,k+k_p+\half$ of
\begin{equation}
\label{locFdtdz}
  _i^k{\mathbb{F}_w}_{i_p}^{k_p} (T) \,\delta_{k+ k_p}[T^i].
\end{equation}
The total variance tendency driven by the triad is the sum of these
two. Expanding $_i^k{\mathbb{F}_u}_{i_p}^{k_p} (T)$ and
$_i^k{\mathbb{F}_w}_{i_p}^{k_p} (T)$ with \eqref{eq:latflux-triad} and
\eqref{eq:vertflux-triad}, it is
\begin{multline*}
-A_i^k\left \{
{ } _i^k{\mathbb{A}_u}_{i_p}^{k_p}
  \left(
    \frac{ \delta_{i+ i_p}[T^k] }{ {e_{1u}}_{\,i + i_p}^{\,k} }
    - {_i^k\mathbb{R}_{i_p}^{k_p}} \
    \frac{ \delta_{k+k_p} [T^i] }{{e_{3w}}_{\,i}^{\,k+k_p} }\right)\,\delta_{i+ i_p}[T^k] \right.\\
- \left. { } _i^k{\mathbb{A}_w}_{i_p}^{k_p}
  \left(
    \frac{ \delta_{i+ i_p}[T^k] }{ {e_{1u}}_{\,i + i_p}^{\,k} }
    -{\:}_i^k\mathbb{R}_{i_p}^{k_p}
    \frac{ \delta_{k+k_p} [T^i] }{{e_{3w}}_{\,i}^{\,k+k_p} }
  \right) {\,}_i^k\mathbb{R}_{i_p}^{k_p}\delta_{k+ k_p}[T^i]
\right \}.
\end{multline*}
The key point is then that if we require
$_i^k{\mathbb{A}_u}_{i_p}^{k_p}$ and $_i^k{\mathbb{A}_w}_{i_p}^{k_p}$
to be related to a triad volume $_i^k\mathbb{V}_{i_p}^{k_p}$ by
\begin{equation}
  \label{eq:V-A}
  _i^k\mathbb{V}_{i_p}^{k_p}
  ={\;}_i^k{\mathbb{A}_u}_{i_p}^{k_p}\,{e_{1u}}_{\,i + i_p}^{\,k}
  ={\;}_i^k{\mathbb{A}_w}_{i_p}^{k_p}\,{e_{3w}}_{\,i}^{\,k + k_p},
\end{equation}
the variance tendency reduces to the perfect square
\begin{equation}
  \label{eq:perfect-square}
  -A_i^k{\:} _i^k\mathbb{V}_{i_p}^{k_p}
  \left(
    \frac{ \delta_{i+ i_p}[T^k] }{ {e_{1u}}_{\,i + i_p}^{\,k} }
    -{\:}_i^k\mathbb{R}_{i_p}^{k_p}
    \frac{ \delta_{k+k_p} [T^i] }{{e_{3w}}_{\,i}^{\,k+k_p} }
  \right)^2\leq 0.
\end{equation}
Thus, the constraint \eqref{eq:V-A} ensures that the fluxes associated
with a given slope triad $_i^k\mathbb{R}_{i_p}^{k_p}$ do not increase
the net variance. Since the total fluxes are sums of such fluxes from
the various triads, this constraint, applied to all triads, is
sufficient to ensure that the globally integrated variance does not
increase.

The expression \eqref{eq:V-A} can be interpreted as a discretization
of the global integral
\begin{equation}
  \label{eq:cts-var}
  \frac{\partial}{\partial t}\int\half T^2\, dV =
  \int\mathbf{F}\cdot\nabla T\, dV,
\end{equation}
where, within each triad volume $_i^k\mathbb{V}_{i_p}^{k_p}$, the
lateral and vertical fluxes/unit area
\[
\mathbf{F}=\left(
_i^k{\mathbb{F}_u}_{i_p}^{k_p} (T)/_i^k{\mathbb{A}_u}_{i_p}^{k_p},
{\:}_i^k{\mathbb{F}_w}_{i_p}^{k_p} (T)/_i^k{\mathbb{A}_w}_{i_p}^{k_p}
 \right)
\]
and the gradient
 \[\nabla T = \left(
\delta_{i+ i_p}[T^k] / {e_{1u}}_{\,i + i_p}^{\,k},
\delta_{k+ k_p}[T^i] / {e_{3w}}_{\,i}^{\,k + k_p}
\right)
\]
\subsection{Triad volumes in Griffes's scheme and in \NEMO}
To complete the discretization we now need only specify the triad
volumes $_i^k\mathbb{V}_{i_p}^{k_p}$. \citet{Griffies_al_JPO98} identify
these $_i^k\mathbb{V}_{i_p}^{k_p}$ as the volumes of the quarter
cells, defined in terms of the distances between T, u,f and
w-points. This is the natural discretization of
\eqref{eq:cts-var}. The \NEMO model, however, operates with scale
factors instead of grid sizes, and scale factors for the quarter
cells are not defined. Instead, therefore we simply choose
\begin{equation}
  \label{eq:V-NEMO}
  _i^k\mathbb{V}_{i_p}^{k_p}=\quarter {b_u}_{i+i_p}^k,
\end{equation}
as a quarter of the volume of the u-cell inside which the triad
quarter-cell lies. This has the nice property that when the slopes
$\mathbb{R}$ vanish, the lateral flux from tracer cell $i,k$ to
$i+1,k$ reduces to the classical form
\begin{equation}
  \label{eq:lat-normal}
-\overline{A}_{\,i+1/2}^k\;
\frac{{b_u}_{i+1/2}^k}{{e_{1u}}_{\,i + i_p}^{\,k}}
\;\frac{\delta_{i+ 1/2}[T^k] }{{e_{1u}}_{\,i + i_p}^{\,k}}
 = -\overline{A}_{\,i+1/2}^k\;\frac{{e_{1w}}_{\,i + 1/2}^{\,k}\:{e_{1v}}_{\,i + 1/2}^{\,k}}{{e_{1u}}_{\,i + 1/2}^{\,k}}.
\end{equation}
In fact if the diffusive coefficient is defined at u-points, so that
we employ $A_{i+i_p}^k$ instead of $A_i^k$ in the definitions of the
triad fluxes \eqref{eq:latflux-triad} and \eqref{eq:vertflux-triad},
we can replace $\overline{A}_{\,i+1/2}^k$ by $A_{i+1/2}^k$ in the above.

\subsection{Summary of the scheme}
The divergence of the expression \eqref{Eq_iso_flux} for the fluxes gives the iso-neutral diffusion tendency at
each tracer point:
\begin{equation} \label{Gf_operator} D_l^T = \frac{1}{b_T}
  \sum_{\substack{i_p,\,k_p}} \left\{ \delta_{i} \left[{_{i+1/2-i_p}^k
        {\mathbb{F}_u }_{i_p}^{k_p}} \right] + \delta_{k} \left[
      {_i^{k+1/2-k_p} {\mathbb{F}_w}_{i_p}^{k_p}} \right] \right\}
\end{equation}
where $b_T= e_{1T}\,e_{2T}\,e_{3T}$ is the volume of $T$-cells.
The diffusion scheme satisfies the following six properties:
\begin{description}
\item[$\bullet$ horizontal diffusion] The discretization of the
  diffusion operator recovers \eqref{eq:lat-normal} the traditional five-point Laplacian in
  the limit of flat iso-neutral direction :
  \begin{equation} \label{Gf_property1a} D_l^T = \frac{1}{b_T} \
    \delta_{i} \left[ \frac{e_{2u}\,e_{3u}}{e_{1u}} \;
      \overline{A}^{\,i} \; \delta_{i+1/2}[T] \right] \qquad
    \text{when} \quad { _i^k \mathbb{R}_{i_p}^{k_p} }=0
  \end{equation}

\item[$\bullet$ implicit treatment in the vertical] Only tracer values
  associated with a single water column appear in the expression
  \eqref{eq:i33} for the $_{33}$ fluxes, vertical fluxes driven by
  vertical gradients. This is of paramount importance since it means
  that an implicit in time algorithm can be used to solve the vertical
  diffusion equation. This is a necessity since the vertical eddy
  diffusivity associated with this term,
  \begin{equation}
    \frac{1}{b_w}\sum_{\substack{i_p, \,k_p}} \left\{  
      {\:}_i^k\mathbb{V}_{i_p}^{k_p} \:A_i^k \: \left(_i^k \mathbb{R}_{i_p}^{k_p}\right)^2
    \right\}  = 
    \frac{1}{4b_w}\sum_{\substack{i_p, \,k_p}} \left\{  
      {b_u}_{i+i_p}^k\:A_i^k \: \left(_i^k \mathbb{R}_{i_p}^{k_p}\right)^2
    \right\},
 \end{equation}
  (where $b_w= e_{1w}\,e_{2w}\,e_{3w}$ is the volume of $w$-cells) can be quite large.

\item[$\bullet$ pure iso-neutral operator] The iso-neutral flux of
  locally referenced potential density is zero. See
  \eqref{eq:latflux-rho} and \eqref{eq:vertflux-triad2}.

\item[$\bullet$ conservation of tracer] The iso-neutral diffusion
  conserves tracer content, $i.e.$
  \begin{equation} \label{Gf_property1} \sum_{i,j,k} \left\{ D_l^T \
      b_T \right\} = 0
  \end{equation}
  This property is trivially satisfied since the iso-neutral diffusive
  operator is written in flux form.

\item[$\bullet$ no increase of tracer variance] The iso-neutral diffusion
  does not increase the tracer variance, $i.e.$
  \begin{equation} \label{Gf_property1} \sum_{i,j,k} \left\{ T \ D_l^T
      \ b_T \right\} \leq 0
  \end{equation}
  The property is demonstrated in
  \ref{sec:variance} above. It is a key property for a diffusion
  term. It means that it is also a dissipation term, $i.e.$ it is a
  diffusion of the square of the quantity on which it is applied.  It
  therefore ensures that, when the diffusivity coefficient is large
  enough, the field on which it is applied become free of grid-point
  noise.

\item[$\bullet$ self-adjoint operator] The iso-neutral diffusion
  operator is self-adjoint, $i.e.$
  \begin{equation} \label{Gf_property1} \sum_{i,j,k} \left\{ S \ D_l^T
      \ b_T \right\} = \sum_{i,j,k} \left\{ D_l^S \ T \ b_T \right\}
  \end{equation}
  In other word, there is no need to develop a specific routine from
  the adjoint of this operator. We just have to apply the same
  routine. This property can be demonstrated similarly to the proof of
  the `no increase of tracer variance' property. The contribution by a
  single triad towards the left hand side of \eqref{Gf_property1}, can
  be found by replacing $\delta[T]$ by $\delta[S]$ in \eqref{locFdtdx}
  and \eqref{locFdtdx}. This results in a term similar to
  \eqref{eq:perfect-square},
\begin{equation}
  \label{eq:TScovar}
  -A_i^k{\:} _i^k\mathbb{V}_{i_p}^{k_p}
  \left(
    \frac{ \delta_{i+ i_p}[T^k] }{ {e_{1u}}_{\,i + i_p}^{\,k} }
    -{\:}_i^k\mathbb{R}_{i_p}^{k_p}
    \frac{ \delta_{k+k_p} [T^i] }{{e_{3w}}_{\,i}^{\,k+k_p} }
  \right)
  \left(
    \frac{ \delta_{i+ i_p}[S^k] }{ {e_{1u}}_{\,i + i_p}^{\,k} }
    -{\:}_i^k\mathbb{R}_{i_p}^{k_p}
    \frac{ \delta_{k+k_p} [S^i] }{{e_{3w}}_{\,i}^{\,k+k_p} }
  \right).
\end{equation}
This is symmetrical in $T $ and $S$, so exactly the same term arises
from the discretization of this triad's contribution towards the
RHS of \eqref{Gf_property1}.
\end{description}


$\ $\newline %force an empty line
% ================================================================
% Skew flux formulation for Eddy Induced Velocity :
% ================================================================
\section{ Eddy induced velocity and Skew flux formulation}

When Gent and McWilliams's [1990] diffusion is used (\key{traldf\_eiv} defined), 
an additional advection term is added. The associated velocity is the so called 
eddy induced velocity, the formulation of which depends on the slopes of iso-
neutral surfaces. Contrary to the case of iso-neutral mixing, the slopes used 
here are referenced to the geopotential surfaces, $i.e.$ \eqref{Eq_ldfslp_geo} 
is used in $z$-coordinate, and the sum \eqref{Eq_ldfslp_geo}
+ \eqref{Eq_ldfslp_iso} in $z^*$ or $s$-coordinates. 

The eddy induced velocity is given by: 
\begin{equation} \label{Eq_eiv_v}
\begin{split}
 u^* & = - \frac{1}{e_{2}e_{3}}\;          \partial_k \left( e_{2} \, A_{e} \; r_i \right)   \\
 v^* & = - \frac{1}{e_{1}e_{3}}\;          \partial_k \left( e_{1} \, A_{e} \; r_j \right)   \\
w^* & =    \frac{1}{e_{1}e_{2}}\; \left\{ \partial_i  \left( e_{2} \, A_{e} \; r_i \right) 
                         + \partial_j  \left( e_{1} \, A_{e} \;r_j \right) \right\} \\
\end{split}
\end{equation}
where $A_{e}$ is the eddy induced velocity coefficient, and $r_i$ and $r_j$ the slopes between the iso-neutral and the geopotential surfaces. 

The traditional way to implement this additional advection is to add it to the Eulerian 
velocity prior to computing the tracer advection. This allows us to take advantage of 
all the advection schemes offered for the tracers (see \S\ref{TRA_adv}) and not just 
a $2^{nd}$ order advection scheme. This is particularly useful for passive tracers 
where \emph{positivity} of the advection scheme is of paramount importance. 

\citet{Griffies_JPO98} introduces another way to implement the eddy induced advection, 
the so-called skew form. It is based on a transformation of the advective fluxes 
using the non-divergent nature of the eddy induced velocity. 
For example in the (\textbf{i},\textbf{k}) plane, the tracer advective fluxes can be 
transformed as follows:
\begin{flalign*}
\begin{split}
\textbf{F}_{eiv}^T = 
\begin{pmatrix} 
           {e_{2}\,e_{3}\;  u^*}       \\
      {e_{1}\,e_{2}\; w^*}  \\
\end{pmatrix}   \;   T
&=
\begin{pmatrix} 
           { - \partial_k \left( e_{2} \, A_{e} \; r_i \right) \; T \;}       \\
      {+ \partial_i  \left( e_{2} \, A_{e} \; r_i \right) \; T \;}    \\
\end{pmatrix}        \\
&=       
\begin{pmatrix} 
           { - \partial_k \left( e_{2} \, A_{e} \; r_i  \; T \right) \;}  \\
      {+ \partial_i  \left( e_{2} \, A_{e} \; r_i  \; T \right) \;}   \\
\end{pmatrix}        
 + 
\begin{pmatrix} 
           {+ e_{2} \, A_{e} \; r_i  \; \partial_k T}  \\
      { - e_{2} \, A_{e} \; r_i  \; \partial_i  T}  \\
\end{pmatrix}   
\end{split}
\end{flalign*}
and since the eddy induces velocity field is no-divergent, we end up with the skew 
form of the eddy induced advective fluxes:
\begin{equation} \label{Eq_eiv_skew_continuous}
\textbf{F}_{eiv}^T = \begin{pmatrix} 
           {+ e_{2} \, A_{e} \; r_i  \; \partial_k T}   \\
      { - e_{2} \, A_{e} \; r_i  \; \partial_i  T}  \\
                                 \end{pmatrix}
\end{equation}
The tendency associated with eddy induced velocity is then simply the divergence 
of the \eqref{Eq_eiv_skew_continuous} fluxes. It naturally conserves the tracer 
content, as it is expressed in flux form. It also preserve the tracer variance.

The discrete form of  \eqref{Eq_eiv_skew_continuous} using the slopes \eqref{Gf_slopes} and defining $A_e$ at $T$-point is then given by:
\begin{flalign} \label{Eq_eiv_skew}   \begin{split}
\textbf{F}_{eiv}^T   \equiv                                   
                        \sum_{\substack{i_p,\,k_p}} \begin{pmatrix} 
+{e_{2u}}_{i+1/2-i_p}^{k}                                    \ {A_{e}}_{i+1/2-i_p}^{k} 
\ \ { _{i+1/2-i_p}^k \mathbb{R}_{i_p}^{k_p} }    \ \ \delta_{k+k_p}[T_{i+1/2-i_p}]      \\
    \\
- {e_{2u}}_i^{k+1/2-k_p}                                      \ {A_{e}}_i^{k+1/2-k_p} 
\ \ { _i^{k+1/2-k_p} \mathbb{R}_{i_p}^{k_p} }    \ \delta_{i+i_p}[T^{k+1/2-k_p}]    \\   
                                                                       \end{pmatrix}   
\end{split}   \end{flalign}
Note that \eqref{Eq_eiv_skew} is valid in $z$-coordinate with or without partial cells. In $z^*$ or $s$-coordinate, the the slope between the level and the geopotential surfaces must be added to $\mathbb{R}$ for the discret form to be exact. 

Such a choice of discretisation is consistent with the iso-neutral operator as it uses the same definition for the slopes. It also ensures the conservation of the tracer variance (see Appendix \ref{Apdx_eiv_skew}), $i.e.$ it does not include a diffusive component but is a `pure' advection term.




\newpage      %force an empty line
% ================================================================
% Discrete Invariants of the skew flux formulation
% ================================================================
\subsection{Discrete Invariants of the skew flux formulation}
\label{Apdx_eiv_skew}


Demonstration for the conservation of the tracer variance in the (\textbf{i},\textbf{j}) plane. 

This have to be moved in an Appendix.

The continuous property to be demonstrated is :
\begin{align*}
\int_D \nabla \cdot \textbf{F}_{eiv}(T) \; T \;dv  \equiv 0
\end{align*}
The discrete form of its left hand side is obtained using \eqref{Eq_eiv_skew}
\begin{align*}
 \sum\limits_{i,k} \sum_{\substack{i_p,\,k_p}}  \Biggl\{   \;\;
 \delta_i  &\left[                                                    
{e_{2u}}_{i+i_p+1/2}^{k}                                  \;\ \ {A_{e}}_{i+i_p+1/2}^{k} 
\ \ \ { _{i+i_p+1/2}^k \mathbb{R}_{-i_p}^{k_p} }   \quad \delta_{k+k_p}[T_{i+i_p+1/2}]         
   \right] \; T_i^k      \\
- \delta_k &\left[ 
{e_{2u}}_i^{k+k_p+1/2}                                     \ \ {A_{e}}_i^{k+k_p+1/2} 
\ \ { _i^{k+k_p+1/2} \mathbb{R}_{i_p}^{-k_p} }   \ \ \delta_{i+i_p}[T^{k+k_p+1/2}]   
   \right] \; T_i^k      \         \Biggr\}   
\end{align*}
apply the adjoint of delta operator, it becomes
\begin{align*}
 \sum\limits_{i,k} \sum_{\substack{i_p,\,k_p}}  \Biggl\{   \;\;
  &\left(                                                    
{e_{2u}}_{i+i_p+1/2}^{k}                                  \;\ \ {A_{e}}_{i+i_p+1/2}^{k} 
\ \ \ { _{i+i_p+1/2}^k \mathbb{R}_{-i_p}^{k_p} }   \quad \delta_{k+k_p}[T_{i+i_p+1/2}]         
   \right) \; \delta_{i+1/2}[T^{k}]      \\
- &\left( 
{e_{2u}}_i^{k+k_p+1/2}                                     \ \ {A_{e}}_i^{k+k_p+1/2} 
\ \ { _i^{k+k_p+1/2} \mathbb{R}_{i_p}^{-k_p} }   \ \ \delta_{i+i_p}[T^{k+k_p+1/2}]   
     \right) \; \delta_{k+1/2}[T_{i}]       \         \Biggr\}       
\end{align*}
Expending the summation on $i_p$ and $k_p$, it becomes:
\begin{align*}
 \begin{matrix}  
&\sum\limits_{i,k}   \Bigl\{ 
  &+{e_{2u}}_{i+1}^{k}                             &{A_{e}}_{i+1    }^{k} 
  &\ {_{i+1}^k \mathbb{R}_{- 1/2}^{-1/2}} &\delta_{k-1/2}[T_{i+1}]    &\delta_{i+1/2}[T^{k}]   &\\
&&+{e_{2u}}_i^{k\ \ \ \:}                            &{A_{e}}_{i}^{k\ \ \ \:}      
  &\ {\ \ \;_i^k \mathbb{R}_{+1/2}^{-1/2}}  &\delta_{k-1/2}[T_{i\ \ \ \;}]  &\delta_{i+1/2}[T^{k}] &\\
&&+{e_{2u}}_{i+1}^{k}                             &{A_{e}}_{i+1    }^{k} 
  &\ {_{i+1}^k \mathbb{R}_{- 1/2}^{+1/2}} &\delta_{k+1/2}[T_{i+1}]     &\delta_{i+1/2}[T^{k}] &\\
&&+{e_{2u}}_i^{k\ \ \ \:}                            &{A_{e}}_{i}^{k\ \ \ \:}       
    &\ {\ \ \;_i^k \mathbb{R}_{+1/2}^{+1/2}} &\delta_{k+1/2}[T_{i\ \ \ \;}] &\delta_{i+1/2}[T^{k}] &\\
%
&&-{e_{2u}}_i^{k+1}                                &{A_{e}}_i^{k+1} 
  &{_i^{k+1} \mathbb{R}_{-1/2}^{- 1/2}}   &\delta_{i-1/2}[T^{k+1}]      &\delta_{k+1/2}[T_{i}] &\\   
&&-{e_{2u}}_i^{k\ \ \ \:}                             &{A_{e}}_i^{k\ \ \ \:} 
  &{\ \ \;_i^k  \mathbb{R}_{-1/2}^{+1/2}}   &\delta_{i-1/2}[T^{k\ \ \ \:}]  &\delta_{k+1/2}[T_{i}] &\\
&&-{e_{2u}}_i^{k+1    }                             &{A_{e}}_i^{k+1} 
  &{_i^{k+1} \mathbb{R}_{+1/2}^{- 1/2}}   &\delta_{i+1/2}[T^{k+1}]      &\delta_{k+1/2}[T_{i}] &\\   
&&-{e_{2u}}_i^{k\ \ \ \:}                             &{A_{e}}_i^{k\ \ \ \:} 
  &{\ \ \;_i^k  \mathbb{R}_{+1/2}^{+1/2}}   &\delta_{i+1/2}[T^{k\ \ \ \:}]  &\delta_{k+1/2}[T_{i}] 
&\Bigr\}  \\
\end{matrix}   
\end{align*}
The two terms associated with the triad ${_i^k \mathbb{R}_{+1/2}^{+1/2}}$ are the 
same but of opposite signs, they cancel out. 
Exactly the same thing occurs for the triad ${_i^k \mathbb{R}_{-1/2}^{-1/2}}$. 
The two terms associated with the triad ${_i^k \mathbb{R}_{+1/2}^{-1/2}}$ are the 
same but both of opposite signs and shifted by 1 in $k$ direction. When summing over $k$ 
they cancel out with the neighbouring grid points. 
Exactly the same thing occurs for the triad ${_i^k \mathbb{R}_{-1/2}^{+1/2}}$ in the 
$i$ direction. Therefore the sum over the domain is zero, $i.e.$ the variance of the 
tracer is preserved by the discretisation of the skew fluxes.

%%% Local Variables: 
%%% TeX-master: "../../NEMO_book.tex"
%%% End: