1: \documentclass[fleqn,11pt]{ices05}
2: \textheight 20cm
3:
4: \usepackage{graphicx}
5:
6: \begin{document}
7: {\small
8:
9: \title{Computation of Electrostatic Field near Three-Dimensional Corners and
10: Edges}
11:
12: \author{N.Majumdar$^1$, S.Mukhopadhyay\footnote{Saha Institute of Nuclear
13: Physics, 1/AF, Sector 1, Bidhannagar, Kolkata 700064, West Bengal,
14: India, e-mail: nayana.majumdar@saha.ac.in}
15: }
16: }
17:
18: \abstract{Theoretically, the electric field becomes infinite at corners of two
19: and three dimensions and edges of three dimensions. Conventional finite-element
20: and boundary element methods do not yield satisfactory results at close
21: proximity to these singular locations. In this
22: paper, we describe the application of a fast and accurate BEM solver (which uses
23: exact analytic expressions to compute the effect of source distributions on flat
24: surfaces) to compute the electric field near three-dimensional corners and
25: edges.
26: Results have been obtained for distances as close as 1$\mu m$ near the corner /
27: edge and good agreement has been observed between the present results and
28: existing analytical solutions.}
29:
30: \section{Introduction}
31: Accurate computation of electric field near corners and edges is important in
32: many applications related to science and engineering. While the generic problem
33: is important even in branches like fluid and solid mechanics, the specific
34: electrostatic problem is very important in the modeling and simulation of
35: micro electro-mechanical systems (MEMS), electric probes and antennas, to name a
36: few. While it is true that the ideal corner / edge does not exist in nature, the
37: singularity being smoothed by rounding corners, sharp increase in charge density
38: does occur near these geometries. Since the electric field is proportional to
39: the square of the charge density, it is very important to estimate the charge
40: density and the resulting variation of electric field in their vicinity.
41:
42: There have been many attempts at solving the above problem using the
43: finite-element (FEM) and the boundary element method (BEM). However, because of
44: the nature of the
45: problem, significant modifications to the basic method needed to be carried out
46: to handle the boundary singularities. On the simpler side, special mesh
47: refinement schemes have been used while on the more sophisticated side, the
48: form of local asymptotic expansions (which may often be found) have been used.
49: Very effective FEM solvers have been developed \cite{Babuska}
50: which calculate singular coefficients by post-processing the
51: numerical solution. These solvers improve the solution by refining the mesh and
52: changing the degree of piecewise polynomials. Similarly, several accurate
53: BEM solvers have been developed \cite{Igarashi}, \cite{Elliotis}.
54: For example, in the singular function boundary integral method,
55: the singular coefficients are calculated directly and the solution is
56: approximated by the leading terms of the local asymptotic solution.
57:
58: In this paper, we present a solution to the corner / edge problem by using
59: a recently developed three-dimensional BEM solver \cite{Mukhopadhyay}. This
60: solver uses exact
61: analytic expressions for computing the influence of charge evenly distributed
62: on discretized flat elements on the boundary. Through the use of these
63: closed-form expressions, the solver avoids one of the most important
64: approximations
65: of the BEM, namely, the assumption that the charge on a boundary element is
66: concentrated at one point (normally the centroid) of the element. As a result,
67: the computed capacitance matrix is very accurate and the solver is able to
68: handle difficult
69: geometries and charge distributions with relative ease. The solver has been
70: used to solve the classic problem of two planes intersecting at various angles.
71: Exact solution to this problem exists \cite{Jackson} and our results have been
72: compared with the exact results. The comparison is very good even as close as
73: 1 $\mu m$ to the corner / edge. Especially important is the fact that the
74: solver produces quite accurate results even for the case of an edge. It has also
75: been possible to reproduce the dependence of the strength of the electric field
76: as a function of the distance from the geometric singularity. All the
77: calculations have been carried out in three dimensions and, here, we also
78: present the variation of the electric field along the corner or edge. It is
79: observed that the electric field retains its mid-plane value for much of the
80: distance along its length, but increases significantly within 20\% of the axial
81: ends. It may be noted here
82: that for these calculations, only algebraic mesh refinement near the edge was
83: applied and no other sophisticated techniques such as those mentioned above
84: were applied. This fact made the development of the solver and its application
85: free of mathematical and computational complications. Since corners and
86: edges play an important role in many electro-mechanical structures, the solver
87: can be used to study the electrostatic properties of such geometries.
88:
89: \section{Theory}
90: We have considered the geometry as presented in \cite{Jackson} in which two
91: conducting planes intersect each other at an angle $\beta$. The planes are
92: assumed to be held at a given potential. A circular cylinder is also included that just
93: encloses the two intersecting plane, has its center at the intersection point
94: and is held at zero potential. The general solution in the polar coordinate
95: system ($\rho$, $\phi$) for the potential ($\Phi$) close to
96: the origin in this problem has been shown to be
97: \begin{equation}
98: \label{eq:Potential}
99: \Phi(\rho, \phi) = V\,
100: +\, {\sum_{m=1}^{\infty}}\, a_m\, \rho^{m \pi / \beta}\,sin(m \pi \phi/\beta)
101: \end{equation}
102: where the coefficients $a_m$ depend on the potential remote from the corner at
103: $\rho = 0$ and $V$ represents the boundary condition for $\Phi$ for all $\rho
104: \geq 0$ when $\phi = 0$ and $\phi = \beta$. In the present case where a circular
105: cylinder just encloses the two plates, the problem of finding out $a_m$
106: reduces to a basic fourier series problem with a well known solution
107: \begin{equation}
108: \label{eq:Coeff}
109: a_m\, =\, \frac{4}{m \pi}\,\,\, for\, odd\, m
110: \end{equation}
111: It may be noted here that the series in (\ref{eq:Potential}) involves
112: positive powers of $\rho^{\pi/\beta}$, and, thus, close to the origin (i.e., for
113: small $\rho$), only the first term in the series will be important. The electric
114: field components ($E_{\rho}, E_{\phi}$) are
115: \begin{equation}
116: \label{eq:Efield_rho}
117: E_{\rho}(\rho,\phi) =
118: -\frac{\pi}{\beta}
119: {\sum_{m=1}^{\infty}}\, a_m\, \rho^{(m \pi/\beta) - 1} sin(m \pi \phi / \beta)
120: \end{equation}
121: \begin{equation}
122: \label{eq:Efield_phi}
123: E_{\phi}(\rho,\phi) =
124: -\frac{\pi}{\beta}
125: {\sum_{m=1}^{\infty}}\, a_m\, \rho^{(m \pi/\beta) - 1} cos(m \pi \phi / \beta)
126: \end{equation}
127: Thus, the field components near $\rho = 0$ vary with
128: distance as $\rho^{(\pi/\beta) - 1}$ and this fact is expected to be reflected
129: in a correct numerical solution as well.
130:
131: \section{Numerical Solution}
132: While the above theoretical solution is a two-dimensional one, we have used the
133: BEM solver to compute a three-dimensional version of the above problem.
134: In order to reproduce the two-dimensional behavior at the mid-plane, we have
135: made the axial length of the system sufficiently long, viz., 10 times the
136: radius of the cylinder. The radius of the cylinder has been fixed at one meter,
137: while the length of the intersecting flat plates has been made a micron shorter
138: than the radius. The length of the plate has been kept smaller than the radius
139: of the cylinder to avoid the absurd situation of having two values of the
140: voltage at the same spatial point. We believe that it has been possible to
141: maintain
142: such a small gap as 1 $\mu m$ between the circular cylinder and the flat plates
143: without facing numerical problems because the BEM solver we have used computes
144: the capacity matrix using analytic expressions which calculate accurate values
145: of potential and electric field at extremely close distances from the singular
146: surfaces \cite{Mukhopadhyay}.
147:
148: The cylinder has been discretized uniformly in the angular and axial directions.
149: The flat plates have also been uniformly discretized in the axial direction.
150: In the radial direction, however, the flat plate elements have been made
151: successively smaller towards the edges using a simple algebraic equation in
152: order to take care of the fact that the surface charge density increases
153: considerably near the edges. The
154: electric field has been computed for various values of $\beta$ ($1.25 \pi,\,
155: 1.5 \pi,\, 1.75 \pi\, and\, 2 \pi$),
156: referred to as cases 1, 2, 3 and 4 respectively in the following section.
157:
158: \section{Results}
159: In Figure\ref{fig:ef1}, we have presented a comparison of the variation of the
160: electric field as a function of the distance from the edge as found from the
161: analytical solution (\ref{eq:Efield_rho}) and the BEM solver. Computations have
162: been
163: carried out upto a distance of 1 $\mu m$ from the edge and to properly
164: represent the sharp increase in field, logarithmic scales have been
165: used. The computed electric field is found to be in remarkable agreement for
166: all values of $\beta$. There is a small disagreement between the two results
167: only at the point closest
168: to the corner / edge ($\approx 1.3\%$). At present, we are in the process of
169: improving the BEM solver so that this error can be minimized.
170: \begin{figure}
171: \vspace{-0.5in}
172: \begin{center}
173: \includegraphics[height=2in,width=4in]{fig1c.eps}
174: \caption{\label{fig:ef1} Comparison of electric field values}
175: \end{center}
176: \end{figure}
177:
178: Although, the distance dependence of the electric field is expected to
179: match with the theoretical prediction because the computed and the theoretical
180: estimates are found to match closely, in Figure \ref{fig:ef2}, we have plotted
181: curves corresponding to $\rho^{\pi / \beta\, -\, 1}$ and compared the computed
182: values as points against the curves. Thus, we can confirm that the computed
183: electric field obeys the $\rho^{(\pi/\beta)\, -\, 1}$ relation quite accurately
184: near the corner / edge. Here, to emphasize and visualize the sharpness of the
185: increase of electric field near the singular geometry, only the distance has
186: been drawn using the logscale.
187: \begin{figure}
188: \vspace{-0.5in}
189: \begin{center}
190: \includegraphics[height=2in,width=4in]{fig2c.eps}
191: \caption{\label{fig:ef2} Dependence of electric field as a function of distance}
192: \end{center}
193: \end{figure}
194:
195: In Figure\ref{fig:ef3d}, we have presented the variation of the electric field
196: at various distances from the corner / edge along the length of the geometry.
197: While the field is found to increase considerably towards the end, for most
198: of the length ($\approx 80\%)$, the two-dimensional value seems to be a good
199: approximation for the real value. However, the deviation (as large as
200: $\approx 50\%$) from the analytic solution towards the
201: front and back ends of the geometry should be taken into consideration while
202: designing real-life applications. At close distances near the singular location,
203: oscillation in the electric field value is observed. Initial study indicates
204: that the source of this oscillation is possibly the large size of the boundary
205: elements in the axial direction.
206: \begin{figure}
207: \vspace{-0.5in}
208: \begin{center}
209: \includegraphics[height=2in,width=4in]{fig3c.eps}
210: \caption{\label{fig:ef3d} Variation of electric field along the axis}
211: \end{center}
212: \end{figure}
213:
214: In order to facilitate numerical comparison, we have presented the values of the
215: electric field in the mid-plane as obtained from the exact solution and the BEM
216: solver in Table \ref{table:table1} for the three dimensional edge. We have
217: chosen this case since it is known to be the most difficult among all those
218: considered in this work. It is clear from the table that the present solver
219: does indeed yield very accurate results.
220: \begin{table}
221: \vspace{-0.5in}
222: \centering
223: \begin{tabular}{| l | c | c | c |}
224: \hline
225: Distance & Exact & Computed & Error (\%) \\
226: \hline
227: 0.1 & 1.83015317 & 1.829997 & 0.0085332 \\
228: \hline
229: 0.01 & 6.303166063 & 6.302387 & 0.012359868 \\
230: \hline
231: 0.001 & 20.11157327 & 20.103534 & 0.039973352 \\
232: \hline
233: 0.0001 & 63.65561168 & 63.548305 & 0.168573794 \\
234: \hline
235: 0.00001 & 201.31488353 & 199.998096 & 0.654093481 \\
236: \hline
237: 0.000001 & 636.6191357 & 628.098310 & 1.33844951 \\
238: \hline
239: \end{tabular}
240: \caption{\label{table:table1}Comparison of electric field values}
241: \end{table}
242:
243: \section{Conclusions}
244: A fast and accurate BEM solver has been used to solve the complex problem of
245: finding the electrostatic force field for three dimensional corners and
246: edges. Accurate solutions have been obtained upto distances very close
247: singular geometry. The two dimensional analytic solution has been found to be
248: valid for a large part of the geometry, but near the axial ends, the difference
249: has been found to be significant. The results and the BEM solver is expected
250: to be very useful in solving critical problems associated with the design of
251: MEMS, integrated circuits, electric probes etc. Since these problems are
252: generic in nature, the solver should be important for analysis of problems
253: related to other fields of science and technology.
254:
255: \reference{
256: \item \label{Babuska}
257: I Babuska, B Guo (1986):
258: "The h-p version of the finite element method Part I: the basic
259: approximation results", \textit{Computational Mechanics}, 33, 21-41.
260:
261: \item \label{Igarashi}
262: H Igarashi, T Honma (1996):
263: "A boundary element method for potential fields with corner singularities",
264: \textit{Appl. Math. Modelling}, 20, 847-852.
265:
266: \item \label{Elliotis}
267: M Elliotis, G C Georgiou, C Xenophontos (2002):
268: "The solution of a Laplacian problem over an L-shaped domain with a singular
269: boundary integral method",
270: \textit{Commun. Numer. Meth. Eng.}, 18, 213-222.
271:
272: \item \label{Mukhopadhyay}
273: S Mukhopadhyay, N Majumdar (2005):
274: "Development of a BEM solver using exact expressions for computing the
275: influence of singularity distributions",
276: submitted to the \textit{International Conference on Computational \&
277: Experimental Engineering and Sciences} to be held at Indian Institute of
278: Technology, Chennai from 1 to 6 December, 2005.
279:
280: \item \label{Jackson}
281: J D Jackson (1988):
282: "Classical Electrodynamics", $2^{nd}$ edition,
283: \textit{Wiley Eastern Limited}, (1988), 75-78.
284: }
285: \end{document}
286: