[915] | 1 | % $Id: numerics_bc.tex 915 2012-05-30 15:11:11Z hoffmann $ |
---|
| 2 | \input{header_tmp.tex} |
---|
| 3 | %\input{header_lectures.tex} |
---|
| 4 | |
---|
| 5 | \usepackage[utf8]{inputenc} |
---|
| 6 | \usepackage{ngerman} |
---|
| 7 | \usepackage{pgf} |
---|
| 8 | \usetheme{Dresden} |
---|
| 9 | \usepackage{subfigure} |
---|
| 10 | \usepackage{units} |
---|
| 11 | \usepackage{multimedia} |
---|
| 12 | \usepackage{hyperref} |
---|
| 13 | \newcommand{\event}[1]{\newcommand{\eventname}{#1}} |
---|
| 14 | \usepackage{xmpmulti} |
---|
| 15 | \usepackage{tikz} |
---|
| 16 | \usetikzlibrary{shapes,arrows,positioning} |
---|
| 17 | \def\Tiny{\fontsize{4pt}{4pt}\selectfont} |
---|
| 18 | \usepackage{amsmath} |
---|
| 19 | \usepackage{amssymb} |
---|
| 20 | \usepackage{multicol} |
---|
| 21 | \usepackage{pdfcomment} |
---|
| 22 | |
---|
| 23 | \institute{Institut fÌr Meteorologie und Klimatologie, Leibniz UniversitÀt Hannover} |
---|
| 24 | \date{last update: \today} |
---|
| 25 | \event{PALM Seminar} |
---|
| 26 | \setbeamertemplate{navigation symbols}{} |
---|
| 27 | |
---|
| 28 | \setbeamertemplate{footline} |
---|
| 29 | { |
---|
| 30 | \begin{beamercolorbox}[rightskip=-0.1cm]& |
---|
| 31 | {\includegraphics[height=0.65cm]{imuk_logo.pdf}\hfill \includegraphics[height=0.65cm]{luh_logo.pdf}} |
---|
| 32 | \end{beamercolorbox} |
---|
| 33 | \begin{beamercolorbox}[ht=2.5ex,dp=1.125ex, |
---|
| 34 | leftskip=.3cm,rightskip=0.3cm plus1fil]{title in head/foot} |
---|
| 35 | {\leavevmode{\usebeamerfont{author in head/foot}\insertshortauthor} \hfill \eventname \hfill \insertframenumber \; / \inserttotalframenumber} |
---|
| 36 | \end{beamercolorbox} |
---|
| 37 | \begin{beamercolorbox}[colsep=1.5pt]{lower separation line foot} |
---|
| 38 | \end{beamercolorbox} |
---|
| 39 | } |
---|
| 40 | %\logo{\includegraphics[width=0.3\textwidth]{luhimuk_logo.pdf}} |
---|
| 41 | |
---|
| 42 | \title[Numerics and Boundary Conditions]{Numerics and Boundary Conditions\\ |
---|
| 43 | (used in PALM) |
---|
| 44 | } |
---|
| 45 | \author{Siegfried Raasch} |
---|
| 46 | |
---|
| 47 | \begin{document} |
---|
| 48 | |
---|
| 49 | % Folie 1 |
---|
| 50 | \begin{frame} |
---|
| 51 | \titlepage |
---|
| 52 | \end{frame} |
---|
| 53 | |
---|
| 54 | % Folie 2 |
---|
| 55 | \begin{frame} |
---|
| 56 | \frametitle{Overview} |
---|
| 57 | \scriptsize PALM is (almost) using simple, standard and fast numerical schemes: |
---|
| 58 | \begin{itemize} |
---|
| 59 | \scriptsize |
---|
| 60 | \item<2-> \textbf{Spatial and temporal discretization by finite differences}\\ |
---|
| 61 | \item<3-> \textbf{Explicit timestep methods:}\\ |
---|
| 62 | - Euler\\ |
---|
| 63 | - Leapfrog\\ |
---|
| 64 | - \underline{Runge-Kutta}, second or \underline{third order} |
---|
| 65 | \item<4-> \textbf{Advection method}\\ |
---|
| 66 | - Upstream\\ |
---|
| 67 | - Piacsek-Williams (second order central finite differences)\\ |
---|
| 68 | - Upstream-spline\\ |
---|
| 69 | - Bott-Chlond-scheme (monotone, positiv definit, for scalars only)\\ |
---|
| 70 | - \underline{5th-order scheme of Wicker and Skamarock}, (as used in WRF model) |
---|
| 71 | \item<5-> \textbf{Poisson-equation for pressure}\\ |
---|
| 72 | - \underline{Direct FFT-method}\\ |
---|
| 73 | - Multigrid-method |
---|
| 74 | \item<6-> \textbf{Lagrangian particle model included} |
---|
| 75 | \item<7-> \textbf{Boundary conditions:}\\ |
---|
| 76 | - \underline{Cyclic} and non-cyclic horizontal boundary conditions\\ |
---|
| 77 | - Surface layer with Monin-Obukhov similarity\\ |
---|
| 78 | - Topography\\ |
---|
| 79 | - Turbulent inflow (for non-cyclic horizontal boundary conditions) |
---|
| 80 | |
---|
| 81 | \end{itemize} |
---|
| 82 | \end{frame} |
---|
| 83 | |
---|
| 84 | |
---|
| 85 | \section{Numerics} |
---|
| 86 | \subsection{Numerics} |
---|
| 87 | |
---|
| 88 | % Folie 3 |
---|
| 89 | \begin{frame} |
---|
| 90 | \frametitle{Numerical Grid} |
---|
| 91 | \footnotesize |
---|
| 92 | \vspace{2mm} |
---|
| 93 | \includegraphics[width=\textwidth]{numerics_bc_figures/numerical_grid.png} |
---|
| 94 | \begin{itemize} |
---|
| 95 | \item<1->Equations are spatially discretized on an Arakawa-C grid. |
---|
| 96 | \item<2->All scalar variables s (e.g. , $p^*$, $e$, $K_{\mathrm{m}}$, |
---|
| 97 | $K_{\mathrm{h}}$) are defined at the cell centers. |
---|
| 98 | \item<3->Velocity components ($u$, $v$, $w$) are shifted by half of the grid spacing. |
---|
| 99 | \item<4->Spacings are equidistant, stretching along $z$ is possible. |
---|
| 100 | \end{itemize} |
---|
| 101 | |
---|
| 102 | \tikzstyle{plain} = [rectangle, draw, fill=white!20, text width=0.33\textwidth, font=\small] |
---|
| 103 | \begin{tikzpicture}[remember picture, overlay] |
---|
| 104 | \node at (current page.north west){% |
---|
| 105 | \begin{tikzpicture}[overlay] |
---|
| 106 | \node[plain, draw,anchor=west] at (88mm,-55mm) { |
---|
| 107 | \noindent \scriptsize general definition (cylic):\\ |
---|
| 108 | $\Psi$(0:nz+1,-1:ny+1,-1:nx+1)\\ |
---|
| 109 | $\Psi$(:,-1,:) $=\Psi$(:,ny,:)\\ |
---|
| 110 | $\Psi$(:,ny+1,:) $=\Psi$(:,0,:) |
---|
| 111 | |
---|
| 112 | }; |
---|
| 113 | \end{tikzpicture} |
---|
| 114 | }; |
---|
| 115 | \end{tikzpicture} |
---|
| 116 | \end{frame} |
---|
| 117 | |
---|
| 118 | % Folie 4 |
---|
| 119 | \begin{frame} |
---|
| 120 | \frametitle{Timestep Methods (I)} |
---|
| 121 | \footnotesize |
---|
| 122 | \begin{itemize} |
---|
| 123 | \item<1->\textbf{Euler}\\ |
---|
| 124 | \vspace{1mm} |
---|
| 125 | $\dfrac{\partial \psi(t)}{\partial t} = F (\psi(t)) \rightarrow |
---|
| 126 | \dfrac{\psi(t + \Delta t) - \psi(t)}{\Delta t} |
---|
| 127 | \approx F (\psi(t))$ \hspace{8mm} |
---|
| 128 | \onslide<2-> $u\dfrac{\Delta t}{\Delta x}=C<1$\\ |
---|
| 129 | \begin{flushright} |
---|
| 130 | for stability |
---|
| 131 | \end{flushright} |
---|
| 132 | |
---|
| 133 | \onslide<1->$\psi (t+\Delta t) = \psi(t) |
---|
| 134 | + \Delta t \cdot F(\psi(t)) \hspace{28mm} \mathcal{O}(\Delta t)$\\ |
---|
| 135 | (used for SGS-TKE in special cases) |
---|
| 136 | |
---|
| 137 | \vspace{3mm} |
---|
| 138 | \item<3-> \textbf{Leapfrog}\\ |
---|
| 139 | $\psi(t + \Delta t) = \psi(t - \Delta t)+2 \Delta t \cdot F(\psi(t))$ \hspace{16mm} $ |
---|
| 140 | \mathcal{O}(\Delta t^2)$ \hspace{3mm} $C \le 0.1$\\ |
---|
| 141 | Time-splitting requires a weak time filter (Asselin filter) |
---|
| 142 | |
---|
| 143 | \vspace{3mm} |
---|
| 144 | \item<4-> \textbf{Runge-Kutta, third-order}\\ |
---|
| 145 | $k_1=F(\psi(t))$\\ |
---|
| 146 | \vspace{1mm} |
---|
| 147 | $k_2=F \left( \psi(t) + \frac{1}{3} \Delta t \cdot k_1 \right)$\\ |
---|
| 148 | \vspace{1mm} |
---|
| 149 | $k_3=F \left( \psi(t) - \frac{3}{16} \Delta t \cdot k_1 |
---|
| 150 | + \frac{15}{16} \Delta t \cdot k_2 \right)$\\ |
---|
| 151 | \vspace{1mm} |
---|
| 152 | $\psi(t + \Delta t) = \psi(t) + \frac{1}{30}\Delta t (5 k_1 + 9 k_2 + 16 k_3)$ |
---|
| 153 | \hspace{12mm} $\mathcal{O}(\Delta t^2)$ \hspace{3mm} $C \le 0.9$\\ |
---|
| 154 | \end{itemize} |
---|
| 155 | \end{frame} |
---|
| 156 | |
---|
| 157 | % Folie 5 |
---|
| 158 | \begin{frame} |
---|
| 159 | \frametitle{Timestep Methods (II)} |
---|
| 160 | \footnotesize |
---|
| 161 | \onslide<1->In the PALM code, the different timestep schemes are treated by one |
---|
| 162 | equation using switches: |
---|
| 163 | $\psi (t + \Delta t ) = (1 - c_1) \cdot \psi (t - \Delta t ) + c_1 \cdot \psi (t) |
---|
| 164 | + \Delta t \cdot \left[ c_2 \cdot F (\psi (t) ) + c_3 \cdot F (\psi (t - |
---|
| 165 | \Delta t ) ) \right]$ |
---|
| 166 | \vspace{1mm} |
---|
| 167 | |
---|
| 168 | \onslide<2-> |
---|
| 169 | \begin{centering} |
---|
| 170 | \begin{table} |
---|
| 171 | \begin{tabular}{cccc} |
---|
| 172 | \bf{Scheme} & \bf{c$_1$} & \bf{c$_2$} & \bf{c$_3$}\\ |
---|
| 173 | Euler & 1 & 1 & 0\\ |
---|
| 174 | Leapfrog & 0 & 2 & 0\\ |
---|
| 175 | RK (1st step) & 1 & 1/3 & 0\\ |
---|
| 176 | RK (2nd step) & 1 & 15/16 & -25/48\\ |
---|
| 177 | RK (3rd step) & 1 & 8/15 & 1/15\\ |
---|
| 178 | \end{tabular} |
---|
| 179 | \end{table} |
---|
| 180 | \end{centering} |
---|
| 181 | |
---|
| 182 | \onslide<3-> |
---|
| 183 | \begin{align*} |
---|
| 184 | \psi (t - \Delta t) &= \psi (t) \hspace{15mm} \textbf{after each RK substep}\\ |
---|
| 185 | \psi (t) &= \psi (t + \Delta t) |
---|
| 186 | \end{align*} |
---|
| 187 | \end{frame} |
---|
| 188 | |
---|
| 189 | % Folie 6 |
---|
| 190 | \begin{frame} |
---|
| 191 | \frametitle{Advection Methods (I)} |
---|
| 192 | \small |
---|
| 193 | \begin{itemize} |
---|
| 194 | \item<1-> Piacsek Williams C3 (1970, J. Comput. Phy., 6, 392). |
---|
| 195 | \item<2-> Scheme of 2nd order accuracy. |
---|
| 196 | \item<3-> Conserves integrals of linear and quadratic quantities. |
---|
| 197 | \item<4-> Requires incompressibility $\rightarrow$ flux form of advection term. |
---|
| 198 | \onslide<4-> \includegraphics[width=0.8\textwidth]{numerics_bc_figures/advection_methods.png} |
---|
| 199 | \end{itemize} |
---|
| 200 | $$\left.\frac{\partial (u \psi)}{\partial x}\right\vert_i = \frac{1}{2 \Delta x} |
---|
| 201 | \left( u_{i+\frac{1}{2}} \psi_{i+1} - u_{i-\frac{1}{2}} \psi_{i-1} \right)$$ |
---|
| 202 | \begin{itemize} |
---|
| 203 | \item<5-> In case of momentum advection (e.g. $\psi=u$), $u_{i-1}$ and |
---|
| 204 | $u_{i+1}$ have to be obtained by linear interpolation. |
---|
| 205 | \item<5-> May cause $2 \Delta x$ wiggles in case of sharp gradients. |
---|
| 206 | \end{itemize} |
---|
| 207 | \end{frame} |
---|
| 208 | |
---|
| 209 | % Folie 7 |
---|
| 210 | \begin{frame} |
---|
| 211 | \frametitle{Advection Methods (II)} |
---|
| 212 | \begin{itemize} |
---|
| 213 | \item<1-> \small Upstream-spline\\ \scriptsize |
---|
| 214 | \onslide<1-> - Requires Euler timestep\\ |
---|
| 215 | \onslide<2-> - Nonlocal scheme: produces heavy load on communication network\\ |
---|
| 216 | \onslide<3-> - Numerically unstable under stable stratification |
---|
| 217 | \item<4-> \small Bott-Chlond\\ \scriptsize |
---|
| 218 | \onslide<4-> - Chlond (1994)\\ |
---|
| 219 | \onslide<5-> - Monotone, positive definit. Can only be used for scalars\\ |
---|
| 220 | \onslide<6-> - Conserves sharp gradients\\ |
---|
| 221 | \onslide<7-> - Numerically expensive\\ |
---|
| 222 | \onslide<8-> - Not optimized for use on cache-based machines. |
---|
| 223 | \item<9-> \small Default: Wicker and Skamarock scheme (5th order)\\ \scriptsize |
---|
| 224 | \onslide<9-> - Much better accuracy than Piacsek Williams\\ |
---|
| 225 | \onslide<10-> - Much simpler algorithm than Bott-Chlond\\ |
---|
| 226 | \onslide<11-> - Requires additional ghost layers\\ |
---|
| 227 | \onslide<12-> - Adds additional numerical dissipation |
---|
| 228 | |
---|
| 229 | \end{itemize} |
---|
| 230 | \end{frame} |
---|
| 231 | |
---|
| 232 | % Folie 8 |
---|
| 233 | \begin{frame} |
---|
| 234 | \frametitle{Advection Methods â Wicker/Skamarock (I)} |
---|
| 235 | \footnotesize |
---|
| 236 | \begin{itemize} |
---|
| 237 | \item Wicker and Skamarock (2002, Mon. Wea. Rev. 130, 2088 â 2097). |
---|
| 238 | \item Based on flux form of advection term |
---|
| 239 | \item Difference of fluxes at the edge of the grid cell is used to |
---|
| 240 | discretise advection term |
---|
| 241 | \end{itemize} |
---|
| 242 | |
---|
| 243 | \begin{columns}[T] |
---|
| 244 | \begin{column}{0.55\textwidth} |
---|
| 245 | \hspace{8mm}\includegraphics[width=0.8\textwidth]{numerics_bc_figures/numerical_grid_small.png} |
---|
| 246 | \end{column} |
---|
| 247 | \begin{column}{0.45\textwidth} |
---|
| 248 | $\frac{ \partial \psi}{\partial t} = - \nabla (u_i \psi) \approx |
---|
| 249 | - \frac{F_{i+\frac{1}{2}} - F_{i-\frac{1}{2}}}{\Delta x}$ |
---|
| 250 | \end{column} |
---|
| 251 | \end{columns} |
---|
| 252 | |
---|
| 253 | \tikzstyle{plain} = [rectangle, text width=0.1\textwidth, font=\small] |
---|
| 254 | \begin{tikzpicture}[remember picture, overlay] |
---|
| 255 | \node at (current page.north west){% |
---|
| 256 | |
---|
| 257 | \begin{tikzpicture}[overlay] |
---|
| 258 | \node[plain, anchor=west] at (2mm,-68mm) { |
---|
| 259 | \tikz |
---|
| 260 | { |
---|
| 261 | \draw[blue, -latex', line width=5pt] (1,0) -- (2,0); |
---|
| 262 | } |
---|
| 263 | |
---|
| 264 | $F_{i-\frac{1}{2}}$ |
---|
| 265 | }; |
---|
| 266 | \end{tikzpicture} |
---|
| 267 | |
---|
| 268 | \begin{tikzpicture}[overlay] |
---|
| 269 | \node[plain, anchor=west] at (62mm,-68mm) { |
---|
| 270 | \tikz |
---|
| 271 | { |
---|
| 272 | \draw[blue, -latex', line width=5pt] (,0) -- (2,0); |
---|
| 273 | } |
---|
| 274 | $F_{i+\frac{1}{2}}$ |
---|
| 275 | }; |
---|
| 276 | \end{tikzpicture} |
---|
| 277 | |
---|
| 278 | }; |
---|
| 279 | \end{tikzpicture} |
---|
| 280 | \end{frame} |
---|
| 281 | |
---|
| 282 | % Folie 9 |
---|
| 283 | \begin{frame} |
---|
| 284 | \frametitle{Advection Methods â Wicker/Skamarock (II)} |
---|
| 285 | |
---|
| 286 | |
---|
| 287 | |
---|
| 288 | \textbf{Finite difference approxiamtion of 6$^{\text{th}}$ order} |
---|
| 289 | \begin{tikzpicture}[scale=2] |
---|
| 290 | \tikzstyle{ann} = [draw=none,fill=none,right] |
---|
| 291 | \matrix[nodes={draw, thick, fill=blue!20}, row sep=0.3cm,column sep=0.5cm]{ |
---|
| 292 | \node[rectangle, rounded corners]{ |
---|
| 293 | $F^{\text{6th}}_{i-\frac{1}{2}} = \frac{1}{60} u_{i-\frac{1}{2}} \left( 37 |
---|
| 294 | (\Psi_i + \Psi_{i-1}) - 8 (\Psi_{i+1} + \Psi_{i-2}) + (\Psi_{i+2} |
---|
| 295 | + \Psi_{i-3}) \right)$ |
---|
| 296 | };\\ |
---|
| 297 | }; |
---|
| 298 | \end{tikzpicture} |
---|
| 299 | |
---|
| 300 | \vspace{5mm} |
---|
| 301 | |
---|
| 302 | \textbf{Artificially added numerical dissipation term} |
---|
| 303 | \begin{tikzpicture}[scale=2] |
---|
| 304 | \tikzstyle{ann} = [draw=none,fill=none,right] |
---|
| 305 | \matrix[nodes={draw, thick, fill=blue!40}, row sep=0.3cm,column sep=0.5cm]{ |
---|
| 306 | \node[rectangle, rounded corners]{ |
---|
| 307 | $\frac{-1}{60} \left| u_{i-\frac{1}{2}} \right| \left( 10 (\Psi_i - |
---|
| 308 | \Psi_{i-1}) - 5 (\Psi_{i+1} - \Psi_{i-2}) + (\Psi_{i+2} - \Psi_{i-3}) \right)$ |
---|
| 309 | };\\ |
---|
| 310 | }; |
---|
| 311 | \end{tikzpicture} |
---|
| 312 | \end{frame} |
---|
| 313 | |
---|
| 314 | % Folie 10 |
---|
| 315 | \begin{frame} |
---|
| 316 | \frametitle{Advection Methods â Wicker/Skamarock (III)} |
---|
| 317 | |
---|
| 318 | \begin{tikzpicture}[scale=2] |
---|
| 319 | \tikzstyle{ann} = [draw=none,fill=none,right] |
---|
| 320 | \matrix[nodes={draw, thick, fill=blue!20}, row sep=0.3cm,column sep=0.5cm]{ |
---|
| 321 | \node[rectangle, rounded corners]{ |
---|
| 322 | $F^{\text{6th}}_{i-\frac{1}{2}} = \frac{1}{60} u_{i-\frac{1}{2}} |
---|
| 323 | \left( 37 (\Psi_i + \Psi_{i-1}) - 8 (\Psi_{i+1} + \Psi_{i-2}) + (\Psi_{i+2} + |
---|
| 324 | \Psi_{i-3}) \right)$ |
---|
| 325 | };\\ |
---|
| 326 | }; |
---|
| 327 | \end{tikzpicture} |
---|
| 328 | |
---|
| 329 | \begin{columns}[T] |
---|
| 330 | \begin{column}{0.7\textwidth} |
---|
| 331 | \includegraphics[width=1\textwidth]{numerics_bc_figures/numerical_oscillations.png} |
---|
| 332 | \end{column} |
---|
| 333 | \begin{column}{0.3\textwidth} |
---|
| 334 | Centered Finite Differences produces numerical oscillations (''wiggles'') |
---|
| 335 | near sharp gradients. |
---|
| 336 | \end{column} |
---|
| 337 | \end{columns} |
---|
| 338 | \end{frame} |
---|
| 339 | |
---|
| 340 | % Folie 11 |
---|
| 341 | \begin{frame} |
---|
| 342 | \frametitle{Advection Methods â Wicker/Skamarock (IV)} |
---|
| 343 | \footnotesize |
---|
| 344 | |
---|
| 345 | \begin{columns}[T] |
---|
| 346 | \begin{column}{0.1\textwidth} |
---|
| 347 | \begin{tikzpicture}[scale=2] |
---|
| 348 | \tikzstyle{ann} = [draw=none,fill=none,right] |
---|
| 349 | \matrix[nodes={draw, thick, fill=blue!40}, row sep=0.3cm,column sep=0.5cm]{ |
---|
| 350 | \node[draw=none,fill=none]{ |
---|
| 351 | $F^{\text{5th}}_{i-\frac{1}{2}} = F^{\text{6th}}_{i-\frac{1}{2}}$ |
---|
| 352 | };\\ |
---|
| 353 | }; |
---|
| 354 | \end{tikzpicture} |
---|
| 355 | \end{column} |
---|
| 356 | \begin{column}{0.8\textwidth} |
---|
| 357 | \begin{tikzpicture}[scale=2] |
---|
| 358 | \tikzstyle{ann} = [draw=none,fill=none,right] |
---|
| 359 | \matrix[nodes={draw, thick, fill=blue!40}, row sep=0.3cm,column sep=0.5cm]{ |
---|
| 360 | \node[rectangle, rounded corners]{ |
---|
| 361 | $- \frac{1}{60} \left| u_{i-\frac{1}{2}} \right| \left( 10 (\Psi_i - \Psi_{i-1}) - |
---|
| 362 | 5 (\Psi_{i+1} - \Psi_{i-2}) + (\Psi_{i+2} - \Psi_{i-3}) \right)$ |
---|
| 363 | };\\ |
---|
| 364 | }; |
---|
| 365 | \end{tikzpicture} |
---|
| 366 | \end{column} |
---|
| 367 | \end{columns} |
---|
| 368 | |
---|
| 369 | \begin{columns}[T] |
---|
| 370 | \begin{column}{0.7\textwidth} |
---|
| 371 | \includegraphics[width=1\textwidth]{numerics_bc_figures/numerical_oscillations_2.png} |
---|
| 372 | \end{column} |
---|
| 373 | \begin{column}{0.3\textwidth} |
---|
| 374 | \vspace{3mm} |
---|
| 375 | \underline{Advantage}\\ |
---|
| 376 | Numerical Dissipation damps small scale oscillations.\\ |
---|
| 377 | \vspace{3mm} |
---|
| 378 | \underline{Disadvantage}\\ |
---|
| 379 | In a turbulent flow numerical dissipation removes energy from small scales. |
---|
| 380 | |
---|
| 381 | \end{column} |
---|
| 382 | \end{columns} |
---|
| 383 | \end{frame} |
---|
| 384 | |
---|
| 385 | % Folie 12 |
---|
| 386 | \begin{frame} |
---|
| 387 | \frametitle{Advection Methods â Wicker/Skamarock (V)} |
---|
| 388 | |
---|
| 389 | \begin{columns}[T] |
---|
| 390 | \begin{column}{0.6\textwidth} |
---|
| 391 | \includegraphics[width=1\textwidth]{numerics_bc_figures/numerical_properties.png} |
---|
| 392 | \end{column} |
---|
| 393 | \begin{column}{0.4\textwidth} |
---|
| 394 | \includegraphics[width=1\textwidth]{numerics_bc_figures/pw_ws.png} |
---|
| 395 | \end{column} |
---|
| 396 | \end{columns} |
---|
| 397 | |
---|
| 398 | \begin{itemize} |
---|
| 399 | \item Better resolution of larger scales $(> 8\,\Delta x)$ and hence less |
---|
| 400 | numerical energy transfer from larger to smaller scales compared to lower |
---|
| 401 | order schemes. |
---|
| 402 | \item Less spectral energy at smaller scales. |
---|
| 403 | \end{itemize} |
---|
| 404 | |
---|
| 405 | \end{frame} |
---|
| 406 | |
---|
| 407 | % Folie 13 |
---|
| 408 | \begin{frame} |
---|
| 409 | \frametitle{Pressure Solver (I)} |
---|
| 410 | \footnotesize |
---|
| 411 | \begin{itemize} |
---|
| 412 | \item<1-> Governing equations of PALM require incompressibility |
---|
| 413 | \item<2-> Incompressibility is reached by a predictor-corrector method\\ |
---|
| 414 | \scriptsize |
---|
| 415 | 1. Momentum equations are solved without the pressure term giving a |
---|
| 416 | provisional velocity field which is not free of divergence.\\ |
---|
| 417 | $\overline{u}^{t+\Delta t}_{i_{\mathrm{prov}}} = \overline{u}^t_i + |
---|
| 418 | \Delta t \left( - \frac{\partial}{\partial x_k} \overline{u}^t_k |
---|
| 419 | \overline{u}^t_i - (\varepsilon_{ijk} f_j \overline{u}^t_k |
---|
| 420 | - \varepsilon_{i3k} f_3 u_{\mathrm{g}_k}) |
---|
| 421 | + g \frac{\overline{\theta^*}^t}{\theta_0} \delta_{i3} |
---|
| 422 | - \frac{\partial}{\partial x_k} \overline{u'_k u'_i}^t \right)$\\ |
---|
| 423 | \vspace{3mm} |
---|
| 424 | \onslide<3-> 2. Assign all remaining divergences to the (perturbation) |
---|
| 425 | pressure $p^*$ so that the new corrected velocity field is the sum of the |
---|
| 426 | provisional, divergent field and the perturbation pressure term.\\ |
---|
| 427 | $\overline{u}^{t+\Delta t}_{i} = |
---|
| 428 | \overline{u}^{t+\Delta t}_{i_{\mathrm{prov}}} - |
---|
| 429 | \frac{\Delta t}{\rho_0} \frac{\partial \overline{p^*}^t}{\partial x_i}$\\ |
---|
| 430 | \vspace{3mm} |
---|
| 431 | \onslide<4-> 3. The divergence operator is applied to this equation. |
---|
| 432 | Demanding a corrected velocity field free of divergence, this leads to a |
---|
| 433 | Poisson equation for the perturbation pressure.\\ |
---|
| 434 | $\frac{\partial^2 \overline{p^*}^t}{\partial x_i^2} = \frac{\rho_0} |
---|
| 435 | {\Delta t} \frac{\partial \overline{u}_{i_{\mathrm{prov}}}^{t + \Delta t}} |
---|
| 436 | {\partial x_i}$\\ |
---|
| 437 | \vspace{3mm} |
---|
| 438 | \onslide<5-> 4. After solving the Poisson equation, the final velocity field |
---|
| 439 | is \\ |
---|
| 440 | calculated as given in step 2.\\ |
---|
| 441 | |
---|
| 442 | \end{itemize} |
---|
| 443 | \end{frame} |
---|
| 444 | |
---|
| 445 | % Folie 14 |
---|
| 446 | \begin{frame} |
---|
| 447 | \frametitle{Pressure Solver (II)} |
---|
| 448 | \small |
---|
| 449 | |
---|
| 450 | \begin{itemize} |
---|
| 451 | \item FFT-solver\\ |
---|
| 452 | \onslide<1-> 1. Discretization of the Poisson-equation by central differences\\ |
---|
| 453 | \onslide<2-> 2. 2D discrete FFT in both horizontal directions\\ |
---|
| 454 | \onslide<3-> 3. Solving the resulting tridiagonal set of linear equations\\ |
---|
| 455 | \onslide<4-> 4. Inverse 2D discrete FFT in both horizontal directions leading |
---|
| 456 | to the perturbation pressure |
---|
| 457 | |
---|
| 458 | \begin{itemize} |
---|
| 459 | \item<5-> Very fast and accurate, $\mathcal{O}(n \log n)$, $n$: |
---|
| 460 | number of gridpoints |
---|
| 461 | \item<6-> CPU requirement $<$ 50\% of total CPU time |
---|
| 462 | \item<7-> Due to non-locality of the FFT, transpositions are required |
---|
| 463 | on parallel computers |
---|
| 464 | \item<8-> Usage requires periodic boundary conditions and uniform grids |
---|
| 465 | along $x$ and $y$ |
---|
| 466 | \end{itemize} |
---|
| 467 | \end{itemize} |
---|
| 468 | \end{frame} |
---|
| 469 | |
---|
| 470 | % Folie 15 |
---|
| 471 | \begin{frame} |
---|
| 472 | \frametitle{Pressure Solver (III)} |
---|
| 473 | \scriptsize |
---|
| 474 | \begin{columns} |
---|
| 475 | \begin{column}{1.03\textwidth} |
---|
| 476 | \begin{itemize} |
---|
| 477 | \item<1-> Multigrid-method\\ |
---|
| 478 | \begin{itemize}\scriptsize |
---|
| 479 | \item Iterative solver\\ |
---|
| 480 | \scriptsize |
---|
| 481 | basic idea: Poisson equation is transformed to a fixed point problem:\\ |
---|
| 482 | $\vec{p}^{k+1} = T \cdot \vec{p}^k + \vec{c}^k$\\ |
---|
| 483 | \vspace{1mm} |
---|
| 484 | \onslide<2-> starting from a first guess, the solution will be |
---|
| 485 | improved by repeated execution of the fixed point problem:\\ |
---|
| 486 | $\begin{array}{rcl} |
---|
| 487 | \vec{p}^{1} &=&T \cdot \vec{p}^0 + \vec{c}^0\\ |
---|
| 488 | \vec{p}^{2} &=&T \cdot \vec{p}^1 + \vec{c}^1 |
---|
| 489 | \vspace{-2mm}\\ |
---|
| 490 | |
---|
| 491 | &\vdots& |
---|
| 492 | \vspace{-1.5mm}\\ |
---|
| 493 | |
---|
| 494 | \vec{p}^{k} &=&T \cdot \vec{p}^{k-1} + \vec{c}^{k-1}\\ |
---|
| 495 | \vec{p}^{k+1} &=&T \cdot \vec{p}^k + \vec{c}^k\\ |
---|
| 496 | \end{array}$\\ |
---|
| 497 | \vspace{1mm} |
---|
| 498 | \onslide<3-> Depending on the structure of the matrix $T$ and vector |
---|
| 499 | $c$ different iterative solvers can be defined, e.g.: Jacobi-scheme |
---|
| 500 | (here on 2D-uniform grid):\\ |
---|
| 501 | $p^{k+1}_{i,j} = \frac{1}{4} \cdot \left( p^k_{i-1,j} + p^k_{i+1,j} |
---|
| 502 | + p^k_{i,j-1} + p^k_{i,j+1} - \Delta x^2 f(i,j,k) \right)$\\ |
---|
| 503 | \scriptsize \vspace{2mm} |
---|
| 504 | \item<4-> With each iteration step $k$ the improved solution converges towards |
---|
| 505 | the exact solution. |
---|
| 506 | \item<5-> Iterative schemes are 'local schemes' $\rightarrow$ information |
---|
| 507 | is needed \\ only from neighboring grid-points. |
---|
| 508 | \item<6-> Very low convergence: $\mathcal{O}(n^2)$. |
---|
| 509 | \end{itemize} |
---|
| 510 | \end{itemize} |
---|
| 511 | \end{column} |
---|
| 512 | \end{columns} |
---|
| 513 | \end{frame} |
---|
| 514 | |
---|
| 515 | % Folie 16 |
---|
| 516 | \begin{frame} |
---|
| 517 | \frametitle{Pressure Solver (IV)} |
---|
| 518 | \begin{itemize} |
---|
| 519 | \item<1-> Multigrid-method\\ |
---|
| 520 | \begin{itemize} |
---|
| 521 | \item Due to their locality, iterative solvers show a frequency-dependent |
---|
| 522 | reduction of the residual: low frequencies are reduced slower than high |
---|
| 523 | frequencies. |
---|
| 524 | \item<2-> The main idea of the multigrid method is to reduce errors of different |
---|
| 525 | frequencies on grids with different grid spacing: |
---|
| 526 | \begin{itemize} |
---|
| 527 | \item errors of high frequency are reduced on fine grids |
---|
| 528 | \item errors of low frequency are reduced on coarse grids. |
---|
| 529 | \end{itemize} |
---|
| 530 | \end{itemize} |
---|
| 531 | \end{itemize} |
---|
| 532 | \onslide<2-> |
---|
| 533 | \begin{figure}[htp] |
---|
| 534 | \centering |
---|
| 535 | \includegraphics[scale=0.35]{numerics_bc_figures/errors.png} |
---|
| 536 | \end{figure} |
---|
| 537 | \end{frame} |
---|
| 538 | |
---|
| 539 | % Folie 17 |
---|
| 540 | \begin{frame} |
---|
| 541 | \frametitle{Pressure Solver (V)} |
---|
| 542 | |
---|
| 543 | \begin{columns}[T] |
---|
| 544 | \begin{column}{0.65\textwidth} |
---|
| 545 | \begin{itemize} |
---|
| 546 | \item<1-> Multigrid-method\\ |
---|
| 547 | \begin{itemize} |
---|
| 548 | \footnotesize |
---|
| 549 | \item On each grid-level an approximate solution of the fixed point |
---|
| 550 | equation is obtained performing a few iterations. |
---|
| 551 | \item<2-> The solution is transmitted to the next coarser grid-level |
---|
| 552 | where it is used as the first guess to solve the fixed point problem. |
---|
| 553 | \item<3-> This procedure is performed up to the coarsest grid-level |
---|
| 554 | containing two grid-points in each direction. |
---|
| 555 | \item<4-> From the coarsest grid-level the procedure is passed in |
---|
| 556 | backward order to get the final solution. |
---|
| 557 | \item<5-> For large grids faster than FFT method. |
---|
| 558 | \item<6-> V- and W-cycles are implemented. |
---|
| 559 | \end{itemize} |
---|
| 560 | \end{itemize} |
---|
| 561 | \end{column} |
---|
| 562 | \begin{column}{0.5\textwidth} |
---|
| 563 | \onslide<2-> |
---|
| 564 | \includegraphics[width=1\textwidth]{numerics_bc_figures/multigrid.png} |
---|
| 565 | \end{column} |
---|
| 566 | \end{columns} |
---|
| 567 | \end{frame} |
---|
| 568 | |
---|
| 569 | |
---|
| 570 | |
---|
| 571 | \section{Boundary Conditions} |
---|
| 572 | \subsection{Boundary Conditions} |
---|
| 573 | |
---|
| 574 | % Folie 18 |
---|
| 575 | \begin{frame} |
---|
| 576 | \frametitle{Boundary Conditions (I)} |
---|
| 577 | |
---|
| 578 | \begin{itemize} |
---|
| 579 | \item<1-> Lateral $(xy)$ boundary conditions:\\ |
---|
| 580 | \begin{itemize} |
---|
| 581 | \item Cyclic by default, allowing undisturbed evolution / advection of turbulence. |
---|
| 582 | \begin{columns}[T] |
---|
| 583 | \begin{column}{0.2\textwidth} |
---|
| 584 | %leer |
---|
| 585 | \end{column} |
---|
| 586 | \begin{column}{0.5\textwidth} |
---|
| 587 | \includegraphics[width=1\textwidth]{numerics_bc_figures/lateral_bc.png}\\ |
---|
| 588 | \vspace{2mm} |
---|
| 589 | \end{column} |
---|
| 590 | \begin{column}{0.4\textwidth} |
---|
| 591 | $\begin{array}{rcl} |
---|
| 592 | \Psi(-1) &=& \Psi(n)\\ |
---|
| 593 | \Psi(n+1) &=& \Psi(0)\\ |
---|
| 594 | \end{array}$ |
---|
| 595 | \end{column} |
---|
| 596 | \end{columns} |
---|
| 597 | |
---|
| 598 | \item<2-> Dirichlet (inflow) and radiation (outflow) conditions are allowed along |
---|
| 599 | either $x$- or $y$-direction. |
---|
| 600 | |
---|
| 601 | \item<3-> In case of a Dirichlet condition, the inflow is laminar (by default) and |
---|
| 602 | the domain has to be extended to allow for the development of a turbulent |
---|
| 603 | state, if neccessary. |
---|
| 604 | \item<4-> Non-cyclic lateral conditions require the use of the multigrid-method |
---|
| 605 | for solving the Poisson-equation. |
---|
| 606 | |
---|
| 607 | \end{itemize} |
---|
| 608 | \end{itemize} |
---|
| 609 | \end{frame} |
---|
| 610 | |
---|
| 611 | % Folie 19 |
---|
| 612 | \begin{frame} |
---|
| 613 | \frametitle{Boundary Conditions (II)} |
---|
| 614 | |
---|
| 615 | \begin{columns}[T] |
---|
| 616 | \begin{column}{0.7\textwidth} |
---|
| 617 | \scriptsize |
---|
| 618 | \begin{itemize} |
---|
| 619 | \item<1-> Surface boundary condition: |
---|
| 620 | \begin{itemize} |
---|
| 621 | \scriptsize |
---|
| 622 | \item<1-> Monin-Obukhov-similarity is used by default, i.e. a Prandtl-layer is |
---|
| 623 | assumed between the surface and the first grid layer.\\ |
---|
| 624 | $\frac{\partial \overline{u}}{\partial z} = \frac{u_{*}}{\kappa z} |
---|
| 625 | \Phi_{\mathrm{m}}$; \hspace{3mm} $u_{*} = \sqrt{- \overline{w' u'_0}} |
---|
| 626 | = \sqrt{\frac{\tau_0}{\overline{\rho}}}$\\ |
---|
| 627 | $\frac{\partial \overline{\theta}}{\partial z} = \frac{\vartheta_{*}} |
---|
| 628 | {\kappa z} \Phi_{\mathrm{h}}$; \hspace{3mm} $\vartheta_{*} = |
---|
| 629 | \frac{\overline{w' \theta'_0}}{u_{*}}$\\ |
---|
| 630 | \vspace{2mm} |
---|
| 631 | \item<2-> Integration between $z=z_0$ (roughness height) and $z=z_{\mathrm{p}}$ |
---|
| 632 | (top of Prandtl-layer, $k=1$) gives the only unknowns $u_{*}$ and |
---|
| 633 | $\theta_{*}$ which then define the surface momentum and heat flux, |
---|
| 634 | used as the real boundary conditions.\\ |
---|
| 635 | \vspace{2mm} |
---|
| 636 | \item<3-> $\Phi_{\mathrm{m}}$, $\Phi_{\mathrm{h}}$: Dyer-Businger functions\\ |
---|
| 637 | \onslide<4->$\Phi_{\mathrm{m}} = \left\{ \begin{array}{cc} |
---|
| 638 | 1+5\,\mathrm{Rif} & \text{stable}\\ |
---|
| 639 | 1 & \text{neutral}\\ |
---|
| 640 | (1-16\,\mathrm{Rif})^{-1/4} & \text{unstable}\\ |
---|
| 641 | \end{array} \right. $ |
---|
| 642 | |
---|
| 643 | \end{itemize} |
---|
| 644 | \end{itemize} |
---|
| 645 | \end{column} |
---|
| 646 | \begin{column}{0.3\textwidth} |
---|
| 647 | \onslide<1-> |
---|
| 648 | \includegraphics[width=1\textwidth]{numerics_bc_figures/surface_bc.png}\\ |
---|
| 649 | \vspace{-2mm} |
---|
| 650 | Prandtl-layer\\ |
---|
| 651 | \vspace{8mm} |
---|
| 652 | \onslide<5-> |
---|
| 653 | $\mathrm{Rif} = \frac{\frac{g}{\tilde{\theta}} |
---|
| 654 | \overline{w' \theta'_0}}{\overline{w' u'} |
---|
| 655 | \frac{\partial \overline{u}}{\partial z}}$\\ |
---|
| 656 | \scriptsize Richardson flux number |
---|
| 657 | \end{column} |
---|
| 658 | \end{columns} |
---|
| 659 | \end{frame} |
---|
| 660 | |
---|
| 661 | % Folie 20 |
---|
| 662 | \begin{frame} |
---|
| 663 | \frametitle{Boundary Conditions (III)} |
---|
| 664 | |
---|
| 665 | \begin{itemize} |
---|
| 666 | \item<1-> Surface boundary condition: |
---|
| 667 | \begin{itemize} |
---|
| 668 | \footnotesize |
---|
| 669 | \item<1-> Monin-Obukhov-similarity is only valid for a |
---|
| 670 | horizontal surface with homogeneous conditions. |
---|
| 671 | \item<2-> The surface temperature has to be prescribed. |
---|
| 672 | Alternatively, the surface heat flux can be prescribed. |
---|
| 673 | \item<3-> Instead of MO-similarity, no-slip conditions or |
---|
| 674 | free-slip conditions can be used |
---|
| 675 | \begin{equation*} |
---|
| 676 | u(z=0) = 0, \quad v(z=0)=0 \hspace{15mm} |
---|
| 677 | \frac{\partial u}{\partial z} = 0, \quad |
---|
| 678 | \frac{\partial v}{\partial z} = 0 \qquad |
---|
| 679 | \end{equation*} |
---|
| 680 | realized by |
---|
| 681 | \begin{flalign*} |
---|
| 682 | u(k=0)=-u(k=1) \hspace{18mm} u(k=0)=u(k=1)\\ |
---|
| 683 | v(k=0)=-v(k=1) \hspace{18mm} v(k=0)=v(k=1) |
---|
| 684 | \end{flalign*} |
---|
| 685 | \item<4-> Pressure boundary condition: |
---|
| 686 | $\dfrac{\partial p}{\partial z} = 0$ |
---|
| 687 | in order to guarantee $w(z=0)=0$ |
---|
| 688 | \item<5-> SGS-TKE condition $\dfrac{\partial e}{\partial z}=0$ |
---|
| 689 | \end{itemize} |
---|
| 690 | \end{itemize} |
---|
| 691 | \end{frame} |
---|
| 692 | |
---|
| 693 | % Folie 21 |
---|
| 694 | \begin{frame} |
---|
| 695 | \frametitle{Boundary Conditions (IV)} |
---|
| 696 | |
---|
| 697 | \begin{itemize} |
---|
| 698 | \item<1-> Boundary conditions at the top (default) |
---|
| 699 | \begin{itemize} |
---|
| 700 | \item<1-> Dirichlet conditions for velocities: $u=u_{\mathrm{g}}, |
---|
| 701 | \quad v=v_{\mathrm{g}}, \quad w=0$ |
---|
| 702 | \item<2-> Neumann conditions (temporal constant gradients) for scalars: |
---|
| 703 | $$\frac{\partial \theta}{\partial z} = |
---|
| 704 | \left. \frac{\partial \theta}{\partial z} \right\vert_{t=0}$$ |
---|
| 705 | \item<3-> Pressure: Dirichlet $p=0$ |
---|
| 706 | or Neumann $\dfrac{\partial p}{\partial z} = 0$ |
---|
| 707 | \item<4-> SGS-TKE: Neumann $\dfrac{\partial e}{\partial z} = 0$ |
---|
| 708 | \item<5-> A damping layer can be switched on in order to absorb |
---|
| 709 | gravity waves. |
---|
| 710 | \end{itemize} |
---|
| 711 | \end{itemize} |
---|
| 712 | \end{frame} |
---|
| 713 | |
---|
| 714 | % Folie 22 |
---|
| 715 | \begin{frame} |
---|
| 716 | \frametitle{Initial Conditions} |
---|
| 717 | |
---|
| 718 | All 3D-arrays are initialized with vertical profiles (horizontally homogeneous).\\ |
---|
| 719 | \quad \\ |
---|
| 720 | Two different profiles can be chosen: |
---|
| 721 | \begin{itemize} |
---|
| 722 | \item<2-> \textbf{constant (piecewise linear) profiles} |
---|
| 723 | \begin{itemize} |
---|
| 724 | \footnotesize |
---|
| 725 | \item \textbf{e.g.} $u=0, v=0, \dfrac{\partial \theta}{\partial z}=0$ |
---|
| 726 | \textbf{up to} $z=\unit[1000]{m}$, |
---|
| 727 | $\dfrac{\partial \theta}{\partial z}=+1.0$ \textbf{up to top} |
---|
| 728 | \end{itemize} |
---|
| 729 | \item<3-> \textbf{velocity profiles calculated by a 1D-model (which is a part of PALM)} |
---|
| 730 | \begin{itemize} |
---|
| 731 | \footnotesize |
---|
| 732 | \item \textbf{constant (piecewise linear) temperature profile is used |
---|
| 733 | for the 1D-model} |
---|
| 734 | \end{itemize} |
---|
| 735 | \end{itemize} |
---|
| 736 | \onslide<4-> |
---|
| 737 | \underline{Under horizontally homogeneous initial conditions, random}\\ |
---|
| 738 | \underline{fluctuations have to be added in order to generate turbulence!} |
---|
| 739 | \end{frame} |
---|
| 740 | |
---|
| 741 | \end{document} |
---|