-
Notifications
You must be signed in to change notification settings - Fork 0
/
matrix-comparison.tex
275 lines (258 loc) · 9.51 KB
/
matrix-comparison.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
\documentclass{article}
\usepackage{fullpage}
\usepackage{amsmath}
\usepackage{mathtools}
\usepackage{tikz}
\usepackage{bm}
\usepackage[colorlinks,linkcolor=blue,citecolor=blue,urlcolor=blue]{hyperref}
\usepackage{siunitx}
\usepackage{eufrak}
\usepackage{filecontents}
\usepackage{natbib}
\usepackage{doi}
\begin{filecontents}{matrix-comparison.bib}
@article{devendran2017,
title={A fourth-order {Cartesian} grid embedded boundary method for {Poisson's} equation},
author={Devendran, Dharshi and Graves, Daniel and Johansen, Hans and Ligocki, Terry},
journal={Comm.\ App.\ Math.\ Comp.\ Sci.},
volume={12},
number={1},
pages={51--79},
year={2017},
publisher={Mathematical Sciences Publishers},
doi={10.2140/camcos.2017.12.51}
}
@article{shaw2017,
title={Multidimensional method-of-lines transport for atmospheric flows over steep terrain using arbitrary meshes},
author={Shaw, James and Weller, Hilary and Methven, John and Davies, Terry},
journal={J.\ Comp.\ Phys.},
volume={344},
pages={86--107},
year={2017},
publisher={Elsevier},
doi={10.1016/j.jcp.2017.04.061}
}
\end{filecontents}
\newcommand{\vect}{\mathbf}
\newcommand{\moment}{\mathfrak{m}}
\title{Comparison of one-dimensional cubicFit and highOrderFit}
\author{James Shaw}
\begin{document}
\nocite{*}
\maketitle
\begin{figure}
\centering
\begin{tikzpicture}[
scale=0.53,
cpnt/.style={fill=gray},
]
\draw [->] (-3,0) -- (5.5,0) node [right] {$x$};
\draw (-1,-0.5) -- (-1,0.5);
\draw (1,-0.5) -- (1,0.5);
\draw (3,-0.5) -- (3,0.5) node [above] {$f$};
\path [cpnt] (-2,0) circle [radius=0.1] node [below] {$\phi_{uuu}$};
\path [cpnt] (0,0) circle [radius=0.1] node [below] {$\phi_{uu}$};
\path [cpnt] (2,0) circle [radius=0.1] node [below] {$\phi_u$};
\path [cpnt] (4,0) circle [radius=0.1] node [below] {$\phi_d$};
\draw [thick] (3,0) circle [radius=0.15];
\end{tikzpicture}
%
\caption{One-dimensional four-point upwind-biased stencil used to approximate the flux at face $f$.}
\label{fig:stencil}
\end{figure}
Here we make a comparison between two transport schemes: cubicFit \citep{shaw2017} and highOrderFit, which is based on the high-order formulation by \citet{devendran2017}. Both schemes form a matrix equation that is solved to find coefficients used to calculate the flux. We define a one-dimensional, four-point, upwind-biased stencil (figure~\ref{fig:stencil}) with equispaced cell centres.
For cubicFit we approximate a field $\phi$ using a cubic polynomial
\begin{align}
\phi = a_1 + a_2 x + a_3 x^2 + a_4 x^3
\end{align}
that interpolates the four stencil points. A matrix equation is formed in order to calculate the unknown coefficients $a_1 \ldots a_4$,
\begin{align}
\begin{bmatrix}
1 & x_{uuu} & x_{uuu}^2 & x_{uuu}^3 \\
1 & x_{uu} & x_{uu}^2 & x_{uu}^3 \\
1 & x_u & x_u^2 & x_u^3 \\
1 & x_d & x_d^2 & x_d^3
\end{bmatrix}
\begin{bmatrix}
a_1 \\
a_2 \\
a_3 \\
a_4
\end{bmatrix}
=
\begin{bmatrix}
\phi_{uuu} \\
\phi_{uu} \\
\phi_u \\
\phi_d
\end{bmatrix}
\text{.}
\end{align}
If the equispaced cell centres are positioned at $x_{uuu} = -2.5$, $x_{uu} = -1.5$, $x_u = -0.5$, $x_d = 0.5$ then
\begin{align}
\begin{bmatrix}
1 & -2.5 & 6.25 & -15.625 \\
1 & -1.5 & 2.25 & -3.375 \\
1 & -0.5 & 0.25 & -0.125 \\
1 & 0.5 & 0.25 & 0.125
\end{bmatrix}
\begin{bmatrix}
a_1 \\
a_2 \\
a_3 \\
a_4
\end{bmatrix}
=
\begin{bmatrix}
\phi_{uuu} \\
\phi_{uu} \\
\phi_u \\
\phi_d
\end{bmatrix}
\text{.}
\label{eq:cubicFitMatrix}
\end{align}
For highOrderFit, we solve the matrix equation
\begin{align}
\begin{bmatrix}
\moment_{uuu}^0/\moment_{uuu}^0 & \moment_{uuu}^1/\moment_{uuu}^0 & \moment_{uuu}^2/\moment_{uuu}^0 & \moment_{uuu}^3/\moment_{uuu}^0 \\
\moment_{uu}^0/\moment_{uu}^0 & \moment_{uu}^1/\moment_{uu}^0 & \moment_{uu}^2/\moment_{uu}^0 & \moment_{uu}^3/\moment_{uu}^0 \\
\moment_u^0/\moment_u^0 & \moment_u^1/\moment_u^0 & \moment_u^2/\moment_u^0 & \moment_u^3/\moment_u^0 \\
\moment_d^0/\moment_d^0 & \moment_d^1/\moment_d^0 & \moment_d^2/\moment_d^0 & \moment_d^3/\moment_d^0
\end{bmatrix}
\begin{bmatrix}
a_1 \\
a_2 \\
a_3 \\
a_4
\end{bmatrix}
=
\begin{bmatrix}
\phi_{uuu} \\
\phi_{uu} \\
\phi_u \\
\phi_d
\end{bmatrix}
\end{align}
where $\moment_V^p = \int_V x^p dV$ is the $p$th moment of volume $V$, and the zeroth moment $\moment_V^0$ is equal to the volume.
If the equispaced cells each have $\moment^0 = 1$ with the cell centres positioned as before, then
\begin{align}
\begin{bmatrix}
1 & -2.5 & 6.\dot{3} & -16.25 \\
1 & -1.5 & 2.\dot{3} & -3.75 \\
1 & -0.5 & 0.\dot{3} & -0.25 \\
1 & 0.5 & 0.\dot{3} & 0.25
\end{bmatrix}
\begin{bmatrix}
a_1 \\
a_2 \\
a_3 \\
a_4
\end{bmatrix}
=
\begin{bmatrix}
\phi_{uuu} \\
\phi_{uu} \\
\phi_u \\
\phi_d
\end{bmatrix}
\text{.}
\label{eq:highOrderMatrix}
\end{align}
Notice how the matrix in equation~\eqref{eq:highOrderMatrix} is similar to, but not equal to, the matrix in equation~\eqref{eq:cubicFitMatrix}.
\section*{Taylor series analysis}
\begin{figure}
\centering
\begin{tikzpicture}[
scale=0.53,
cpnt/.style={fill=gray},
]
\draw [->] (-5,0) -- (5.5,0) node [right] {$x$};
\draw (-3,-0.5) -- (-3,0.5);
\draw (-1,-0.5) -- (-1,0.5);
\draw (1,-0.5) -- (1,0.5) node [above] {$\phi_L$};
\draw (3,-0.5) -- (3,0.5) node [above] {$\phi_R$};
\path [cpnt] (-4,0) circle [radius=0.1] node [below] {$\phi_{j-2}$};
\path [cpnt] (-2,0) circle [radius=0.1] node [below] {$\phi_{j-1}$};
\path [cpnt] (0,0) circle [radius=0.1] node [below] {$\phi_j$};
\path [cpnt] (2,0) circle [radius=0.1] node [below] {$\phi_{j+1}$};
\path [cpnt] (4,0) circle [radius=0.1] node [below] {$\phi_{j+2}$};
\draw [thick] (1,0) circle [radius=0.15];
\draw [thick] (3,0) circle [radius=0.15];
\end{tikzpicture}
\caption{One-dimensional fluxes through a cell $\phi_{j+1}$ using four-point upwind-biased stencils to approximate fluxes $\phi_L$ and $\phi_R$.}
\label{fig:flux}
\end{figure}
To find a polynomial that interpolates the 4-point upwind-biased stencil we construct Taylor series approximations centred at $\phi_L$ for the stencil points $\phi_{j-2} \ldots \phi_{j+1}$
\begin{align}
\begin{bmatrix}
1 & -\frac{5}{2} & \frac{1}{2!} \left(-\frac{5}{2}\right)^2 & \frac{1}{3!} \left(-\frac{5}{2}\right)^3 \\[4pt]
1 & -\frac{3}{2} & \frac{1}{2!} \left(-\frac{3}{2}\right)^2 & \frac{1}{3!} \left(-\frac{3}{2}\right)^3 \\[4pt]
1 & -\frac{1}{2} & \frac{1}{2!} \left(-\frac{1}{2}\right)^2 & \frac{1}{3!} \left(-\frac{1}{2}\right)^3 \\[4pt]
1 & \frac{1}{2} & \frac{1}{2!} \left(\frac{1}{2}\right)^2 & \frac{1}{3!} \left(\frac{1}{2}\right)^3
\end{bmatrix}
\begin{bmatrix}
\phi_L \\
\phi'_L \\
\phi''_L \\
\phi'''_L
\end{bmatrix}
&=
\begin{bmatrix}
\phi_{j-2} \\
\phi_{j-1} \\
\phi_j \\
\phi_{j+1}
\end{bmatrix}
\text{.}
\end{align}
Solving this system results in the fourth-order accurate approximation,
\begin{align}
\phi_L = \frac{1}{16} \phi_{j-2} - \frac{5}{16} \phi_{j-1} + \frac{15}{16} \phi_j + \frac{5}{16} \phi_{j+1} + \mathcal{O}(\Delta x^5)
\end{align}
The approximate flux divergence (figure~\ref{fig:flux}) is then
\begin{align}
\phi_R - \phi_L = -\frac{1}{16} \phi_{j-2} + \frac{6}{16} \phi_{j-1} - \frac{20}{16} \phi_j + \frac{20}{16} \phi_{j+1} + \frac{5}{16} \phi_{j+2} \label{eq:cubicFitCoeffs}
\end{align}
but this not a fourth-order accurate flux divergence approximation as we will see next.
A fourth-order accurate approximation to the flux divergence is derived from a Taylor series approximation of $\partial \phi / \partial x$ centred at $\phi_{j+1}$ (figure~\ref{fig:flux}),
\begin{align}
\begin{bmatrix}
1 & -3 & \frac{1}{2!} \left(-3\right)^2 & \frac{1}{3!} \left(-3\right)^3 & \frac{1}{4!} \left(-3\right)^4 \\[4pt]
1 & -2 & \frac{1}{2!} \left(-2\right)^2 & \frac{1}{3!} \left(-2\right)^3 & \frac{1}{4!} \left(-2\right)^4 \\[4pt]
1 & -1 & \frac{1}{2!} \left(-1\right)^2 & \frac{1}{3!} \left(-1\right)^3 & \frac{1}{4!} \left(-1\right)^4 \\[4pt]
1 & 0 & \frac{1}{2!} 0^2 & \frac{1}{3!} 0^3 & \frac{1}{4!} 0^4 \\[4pt]
1 & 1 & \frac{1}{2!} 1^2 & \frac{1}{3!} 1^3 & \frac{1}{4!} 1^4
\end{bmatrix}
\begin{bmatrix}
\phi_{j+1} \\[4pt]
\phi'_{j+1} \\[4pt]
\phi''_{j+1} \\[4pt]
\phi'''_{j+1} \\[4pt]
\phi''''_{j+1}
\end{bmatrix}
&=
\begin{bmatrix}
\phi_{j-2} \\
\phi_{j-1} \\
\phi_j \\
\phi_{j+1} \\
\phi_{j+2}
\end{bmatrix} \label{eq:highOrderFluxDiv}
\intertext{which yields}
\left. \frac{\partial \phi}{\partial x} \right|_{j+1} = -\frac{1}{12} \phi_{j-2} + \frac{6}{12} \phi_{j-1} - \frac{18}{12} \phi_j + \frac{10}{12} \phi_{j+1} + \frac{3}{12} \phi_{j+2} + \mathcal{O}(\Delta x^4) \text{.} \label{eq:highOrderFluxDivCoeffs}
\end{align}
To find the fourth-order face flux we invert the matrix in equation~\eqref{eq:highOrderMatrix} and find that
\begin{align}
\phi_L = \frac{1}{12} \phi_{j-2} - \frac{5}{12} \phi_{j-1} + \frac{13}{12} \phi_j + \frac{3}{12} \phi_{j+1} \text{.} \label{eq:highOrderFlux}
\end{align}
Using equation~\eqref{eq:highOrderFluxDiv} and equation~\eqref{eq:highOrderFlux} we can verify that $\left. \partial \phi / \partial x \right|_{j+1} = \phi_R - \phi_L + \mathcal{O}(\Delta x^4)$ as desired (figure~\ref{fig:fluxDivComparison}).
\begin{figure}
\centering
\includegraphics[width=5in]{interim-results/taylor-series.png}
\caption{Flux divergence cofficients using a 5-point Taylor series approximation (equation~\ref{eq:highOrderFluxDivCoeffs}), cubicFit approximation (equation~\ref{eq:cubicFitCoeffs}) and highOrderFit approximation (equation~\ref{eq:highOrderFlux}).}
\label{fig:fluxDivComparison}
\end{figure}
\bibliographystyle{ametsoc2014}
\bibliography{matrix-comparison}
\end{document}