G Models
MEX 86 1–14
MethodsX xxx (2015) e1–e14
Contents lists available at ScienceDirect
MethodsX jou rnal homep age : www.elsevier.com/locate/me x 1 2 3 4 5 6 7
Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM) Q1 [TD$FIRSNAME]Anthony
N.[TD$FIRSNAME.] [TD$SURNAME]Johnson[TD$SURNAME.] *, [TD$FIRSNAME]T.V.[TD$FIRSNAME.] [TD$SURNAME]Hromadka[TD$SURNAME.] [10_TD$IF]II
Department of Mathematical Sciences, United States Military Academy, 601 Swift Road, West Point, NY 10996, USA G R A P H I C A L A B S T R A C T
A B S T R A C T
The Laplace equation that results from specifying either the normal or tangential force equilibrium equation in terms of the warping functions or its conjugate can be modeled as a [12_TD$IF]complex variable boundary element method or CVBEM mixed boundary problem. The CVBEM is a well[13_TD$IF]-known numerical technique that can provide solutions to potential value problems in two or more dimensions by the use of an approximation function that is derived from the Cauchy Integral in complex analysis. This paper highlights three customizations to the technique. A least squares approach to modeling the complex-valued approximation function will be compared and analyzed to determine if modeling error on the boundary can be reduced without the need to find and evaluated additional linearly independent complex functions. The nodal point locations will be moved outside the problem domain.
8
* Corresponding author. Tel.: [47_TD$IF]þ1 845 938 2846; fax: þ1 845 938 7690. E-mail address:
[email protected] (A.N. Johnson). http://dx.doi.org/10.1016/j.mex.2015.05.005 2215-0161/Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/ licenses/by/4.0/).
Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
e2
Contour and streamline plots representing the warping function and its complementary conjugate are generated simultaneously from the complex-valued approximating function. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/ licenses/by/4.0/). A R T I C L E I N F O Keywords: Complex variables, Hilbert space, [14_TD$IF]Mixed boundary conditions, Stress, Approximate boundary, Complex variable boundary element method (CVBEM), Torsion, Least squares Article history: Received 24 November 2014; Accepted 11 May 2015
9
Method details
10 11 13 12
The CVBEM has been developed by [1] for the solution of general problems involving Laplace or Poisson equations in multiple dimensions. Modeling begins with a simple closed contour of straight line segments. In a two dimensional complex plane, let V be bounded by a simple closed contour, G, such that
G¼
n [
G j:
(1)
j¼1
16 14 15 17 18 19 20
By defining (k + 1) equidistant nodal points in each Gj such that zj,1 and zj,k+1 are the endpoints of Gj, the global nodal coordinates are related to local nodal coordinates by zj,1 = zj and zj,k+1 = zj+1,1 = zj+1. Fig. 1 shows the global and local nodal numbering conventions. If one defines complex numbers vji at each node zji, then degree k complex polynomials Nkj ðzÞ are uniquely defined on each boundary element Gj. A global trial function of order k is defined by Gk ðzÞ ¼
n X
d j Nkj ðzÞ; z 2 G;
(2)
j¼1
23 21 22 24
27 25 26
where
dj ¼
1 0
Gk(z) is continuous on G and lim
maxjG j j ! 0
30 28 29 31 32 33 34 35 36 37 38 39 40
43 41 42 44
z 2 G; otherwise;
Gk ðzÞ ¼ vðzÞ:
(3)
It is assumed that v(z) is analytic on G [ V and that each vji ¼ vðzji Þ. Along the boundary G, or exterior to the problem domain union boundary, there are defined n nodal points. For development purposes, the n nodes are assumed defined on G [8]. Later, we will move the nodes outward away from the boundary to demonstrate an addition degree of freedom. The simple closed contour, G, in Fig[15_TD$IF]. 2 is divided into n boundary elements, Gj1, Gj[16_TD$IF],. . .,Gn. For each boundary element, an interpolating polynomial will be used to create a piecewise continuous global interpolation function. In Fig[15_TD$IF]. 2, the boundary, G, is ‘‘severed’’ at s = 0 and in the positive direction spans until s = L, the arc length of G. In Fig[15_TD$IF]. 3, the boundary is ‘‘flattened’’ and the piecewise function presented. Here, k = 1 is chosen, and the complex polynomials N kj ðzÞ are uniquely defined as first order linear functions. The piecewise function of Fig[15_TD$IF]. 3 is 8 z z j1 > > z 2 G j1 > > < z j z j1 z jþ1 z (4) Nkj ðzÞ ¼ > z 2 Gj > > z jþ1 z j > : 0 otherwise
Clearly, by its definition on G, Nkj ðzÞ forms a basis as each boundary element produces an independent linear function. Note that the sum of the respective basis function terms is Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
[(Fig._1)TD$IG]
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
e3
Element End Node Element Interior Node
[(Fig._2)TD$IG]
Fig. 1. (k + 1)-[1_TD$IF]Node boundary element Gj nodal definitions[2_TD$IF].
[(Fig._3)TD$IG]
Fig. 2. CVBEM [5][3_TD$IF].
1
0 Fig. 3. Linear [4_TD$IF]interpolation basis functions, N kj ðzÞ.
45 46
continuous on the boundary G for all z 2 G. The basis function will be used to define a linear global trial function, Gk ðzÞ ¼
n X Nkj ðzÞw j
(5)
j¼1
49 47 48 50
which is the sum of all nodal basis functions multiplied by a corresponding complex coefficient, w j , the nodal point j value of the function being approximated. Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
e4
51
ˆ k ðzÞ defined by Consider the approximation function v I 1 Gk ðzÞdz ; z2 = G z 2 V: vˆ k ðzÞ ¼ 2pi G z z
(6)
54 52 53 55
From [17_TD$IF]Eq. (2), the global trial function is substituted into the Cauchy Integral formula and integrated over a simply connected two-dimensional complex domain, V, with boundary, G, such that, I P d N k ðzÞdz X I I N kj ðzÞdz Gk ðzÞdz j j ¼ ¼ : (7) zz G zz z Gj z z
58 56 57
On each Gj, define a local coordinate system by
z j ¼ z j ðs j Þ ¼ z j ðz jþ1 z j Þs j ; z j 2 G j ; 0 s j 1 61 59 60
It follows that I
64 62 63 65
(8)
N kj ðzÞdz Gj
zz
¼
Z
0
1
Nkj ðs j Þds j sj g j
(9)
where Nkj ðs j Þ ¼ Nkj ðz j ðs j ÞÞ, and gj = (z zj)/(zj+1 zj) for z 2 G. Eq[18_TD$IF]. (9) is solved by factoring (sj gj) from Nkj ðs j Þ. Let Nkj ðs j Þ be of the form
66 Nkj ðs j Þ ¼
k X i complexes; j sij ; 0 sij 1 i¼0
69 67 68
where the complexes; j are complex constants in the form (a + bi). Division of N kj ðs j Þ by (sj gj) gives Z
0
72 71 70
1
Nkj ðs j Þds j sj g j
81 79 80
with Rk1 ðzÞ ¼
(11)
P
Rk1 ðzÞ, [20_TD$IF]Eq. (10) simplifies to j
i X 1 h k1 R ðzÞ þ N kj ðg j ÞH j : 2pi
(13)
In [21_TD$IF]Eq. (13), it is noted that the Nkj ðg j Þ have the form of the assumed shape functions on each gj. Letting node z1 be on the branch cut of the complex logarithm function ln(z z) such that z 2 V and z 2 G (see Fig. 5), then (13) can be expanded as
vˆ k ðzÞ ¼ 89 87 88
z jþ1 z d jþ1 ðzÞ ¼ ln þ iu jþ1; j ðzÞ: zj z d j ðzÞ
Note that dj(z) = |zj z| and uj+1,j(z) is the central angle between points zj+1, zj, and z. Fig. 4 shows the special case as z approaches G in the limit. From [19_TD$IF]Eqs. (6), (7), (9) and (10), summation of the complex boundary element contributions from the m boundary element gives X X 2piv ˆ k ðzÞ ¼ Rk1 ðzÞ þ Nkj ðg j ÞH j (12) j
vˆ k ðzÞ ¼ 84 82 83 85 86
(10)
where Rkj ðzÞ is a complex polynomial of degree k 1, and H j ¼ ln
75 73 74 76 77 78
¼ Rkj ðzÞ þ Nkj ðg j ÞH j
1 k1 1 X k1 k ðzÞ; R ðzÞ P j ðz z j Þlnðz z j Þ þ Nm 2pi 2pi
(14)
where P k1 is a polynomial of degree (k 1) defined by j Pk1 ¼ j
ðNkj ðg j Þ N kj1 ðg j1 ÞÞ ðz z j Þ
(15)
Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
[(Fig._4)TD$IG]
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
[(Fig._5)TD$IG]
Fig. 4. CVBEM [5_TD$IF]linear trial function geometry.
e5
BRANCH CUT
Fig. 5. Branch cut of the function ln(z z), z 2 G.
92 91 90 93
and ln(z zj) is the principal value of the logarithm function. From the continuity of Gk(z), it is seen that at the nodal coordinate zj, Nkj ðg j Þ N kj1 ðg j1 Þ ¼ 0
96 94 95 97 98
k and that (z zj) is a factor as shown in (15). In (14), the Nm term appears due to the circuit around the branch point of the multiple-valued function ln(z zj). Letting
99 102 101 100
(16)
Rk ðzÞ ¼
1 k1 k R ðzÞ þ Nm ðzÞ; 2pi
then
vˆ k ðzÞ ¼ Rk ðzÞ
1 X k1 P j ðz z j Þlnðz z j Þ: 2pi
(17)
Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
e6
105 103 104 106 107
ˆ k ðzÞ is continuous over V and has removable singularities at each boundary From (17), it is seen that v element endpoint (nodal coordinate zj, j = 1, 2, 3, . . ., m).That is Rk(z) and P k1 are continuous complex j polynomials, and
108
ˆ k ðz j Þ ¼ Rk ðz j Þ i:e: v
lim ðz z j Þlnðz z j Þ ¼ 0;
z ! zj
111 109 110 112 113 114 115 116 117 118
ˆ ðzÞ where f ˆ ðzÞ are two ˆ ðzÞ þ ic ˆ ðzÞ and c Note that since v ˆ k ðzÞ is analytic in V and v ˆ k ðzÞ ¼ f dimensional potential and stream functions which satisfy the Laplace equation exactly over V. By forcing the approximation values of v ˆ k ðzÞ to be arbitrarily close (within some e) to the boundarycondition values of v(z) on G, then it is guaranteed by the maximum modulus theorem that the approximation of v(z) is bounded by jvðzÞ v ˆ ðzÞj e, for all z 2 V. Because the CVBEM results in a two-dimensional function which is an exact solution to the ˆ ðzÞ to v(z) is then achieved on V [ G by governing partial differential equation on V, convergence of v forcing convergence on G. This is shown from (3) [2_TD$IF]and (6) by lim
maxjG j j ! 0
121 120 119 122 123 124 125
I
Gk ðzÞdz ¼ G zz
I
limmaxjG j j ! 0 Gk ðzÞdz G
zz
¼
I
G
vðzÞdz ¼ 2pivðzÞ: zz
(18)
ˆ ˆ The global trial function is continuous. Thus, wðzÞ is analytic in V, allowing wðzÞ to be used as an approximation function defined almost everywhere (‘‘ae’’) inside V as well as exterior to V. This characteristic separates the CVBEM from other approximations techniques. When solved, the CVBEM approximating integral becomes an approximating function of the form n X ˆ wðzÞ ¼ N0 ðzÞ þ N1 ðzÞ þ complexes; j ðz z j Þln j ðz z j Þ (19) j¼1
128 127 126 129
with
132 130 131 133
where aj and bj are real constants to be determined, and
complexes; j ¼ a j þ ib j
N0 ðzÞ ¼ ða0 þ ib0 Þ and N1 ðzÞ ¼ ða1 þ ib1 Þðx þ iyÞ 136 135 134 137 138
where a0, b0, a1, and b1 are also real constants to be determined. The method steps [6] start by using u the construct (z zj) = Rjei j, defined at each node j, with location zj, Eq[18_TD$IF]. (19) becomes n X ˆ wðzÞ ¼ N0 ðzÞ þ N1 ðzÞ þ ða j þ ib j ÞR j eiu j ln j ðR j eiu j Þ: (20) j¼1
141 140 139
iu
Using Euler’s formula of e = (cos u + i sin u), the CVBEM approximation function becomes ˆ wðzÞ ¼ ða0 þ ib0 Þ þ ða1 þ ib1 Þðx þ iyÞ þ
n X ða j þ ib j ÞR j ðcosu j þ isinu j Þln j ðR j eiu j Þ:
(21)
j¼1
144 143 142
Further evaluation gives ln j ðR j eiu j Þ ¼ ln j ðR j Þ þ iu j :
147 146 145
(22)
Combining terms from [24_TD$IF]Eqs. (21) and (22), ˆ wðzÞ ¼ ða0 þ ib0 Þ þ ða1 þ ib1 Þðx þ iyÞ þ
n X ða j R j cosu j þ ia j R j sinu j þ ib j R j cosu j j¼1
2
þ i b j R j sinu j Þðln j ðR j Þ þ iu j Þ:
(23)
149 148 Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
150
e7
Collecting real and imaginary terms yield ˆ wðzÞ ¼ ða0 þ ib0 Þ þ ða1 þ ib1 Þðx þ iyÞ þ
n X ða j R j ðln j ðR j Þcosu j u j sinu j Þ j¼1
b j R j ðln j ðR j Þsinu j þ u j cosu j Þ þ i½a j R j ðln j ðR j Þsinu j þ u j cosu j Þ þ b j R j ðln j ðR j Þcosu j u j sinu j ÞÞ: ˆ , parts: ˆ , and imaginary, c We can now separate the approximation function into real, f
153 152 151
ˆ ðzÞ ˆ ðzÞ þ ic ˆ wðzÞ ¼f 156 155 154
(25)
where the potential functions, or real parts, are given by ˆ ðzÞ fˆ ðzÞ ¼ a0 þ ða1 x b1 yÞ þ F
159 158 157
(24)
for
ˆ ðzÞ ¼ F
n X
ða j R j ðln j ðR j Þcosu j u j sinu j Þ b j R j ðln j ðR j ÞR j sinu j þ u j cosu j ÞÞ
(26) (27)
j¼1
162 160 161 163
and where the stream functions, or imaginary parts, are given by ˆ ðzÞ ˆ ðzÞ ¼ b þ ðb x þ a1 yÞ þ C c 0 1
166 165 164
for ˆ ðzÞ ¼ C
n X ða j R j ðln j ðR j Þsinu j þ u j cosu j Þ þ b j R j ðln j ðR j Þcosu j u j sinu j ÞÞ:
(28) (29)
j¼1
169 168 167
Recall that ln j includes the effect of the nodal point logarithmic branch cut rotations.
170
[25_TD$IF]Matrix formulation and model strategy
171 172 173 174 175 176
After the real and imaginary parts of the CVBEM approximation equation have been developed, the ˆ functions. The two numerical modeling ˆ and c next step is to find the constants, an, and bn, in the f strategies investigated and subsequently compared for both accuracy and efficiency in this paper are collocation and a least squares approach in a Hilbert space. Each has advantages and disadvantages. Our goal is to highlight which modeling strategy works best for mixed boundary value problems with irregular boundaries.
177
[26_TD$IF]Collocation
178 179 180 181 182 183 184 185 186 187
For collocation, the approach is to use the CVBEM by juxtaposing [17_TD$IF]Eq. (19) at each nodal point specified on G (in the limit as z approaches G from inside V). Generally, only one nodal value of either f or c is known at each nodal point [4]. Consequently for m nodes specified on G, there are 2m values of {fj, cj}, and only m nodal values are known as boundary conditions. Collocating [17_TD$IF]Eq. (19) at each node generates m equations for the m unknown nodal values. The resulting m m matrix system results in the ˆ ðzÞ approximator, which is analytic in V. That is, v ˆ ðzÞ operates on the 2m nodal determination of the v values {fj, cj} and the coordinate z. Next is to develop an analytic continuation of the v ˆ ðzÞ approximator which matches the specified and computed 2m nodal values of G. The advantage of using [17_TD$IF]Eq. (19) is that ˆ ðzÞ only has non-zero value in V [ G. That is, the Cauchy integral of [17_TD$IF]Eq. (6) has the property that v (30) vˆ ðzÞ ¼ vˆ ðzÞ; z 2 V [ G; 0 z2 = V [ G:
190 189 188 191
Consider the real portion of the CVBEM with one node for collocation point k. The resulting equation from Eq[18_TD$IF]. (26) is:
fˆ k ðzÞ ¼ fˆ k ðxk þ iyk Þ ¼ a0 þ a1 xk b1 yk þ a2 p1;k b2 q1;k
(31)
Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
e8
194 193 192
where pn;k ¼ Rn;k ðlnðRn;k Þcosu n;k u n;k sinun;k qn;k ¼ Rn;k ðlnðRn;k Þsinun;k un;k cosun;k
197 196 195 198
(32)
:
Evaluate the above with the five necessary potential collocation points on the problem boundary. This will result in five linearly independent equations for five collocation points on G,
fˆ k ðzÞ ¼ fˆ k ðx1 þ iy1 Þ ¼ a0 þ a1 x1 b1 y1 þ a2 p1;1 b2 q1;1 fˆ k ðzÞ ¼ fˆ k ðx2 þ iy2 Þ ¼ a0 þ a1 x2 b1 y2 þ a2 p1;2 b2 q1;2 fˆ k ðzÞ ¼ fˆ k ðx3 þ iy3 Þ ¼ a0 þ a1 x3 b1 y3 þ a2 p1;3 b2 q1;3 fˆ k ðzÞ ¼ fˆ k ðx4 þ iy4 Þ ¼ a0 þ a1 x4 b1 y4 þ a2 p1;4 b2 q1;4 fˆ k ðzÞ ¼ fˆ k ðx5 þ iy5 Þ ¼ a0 þ a1 x5 b1 y5 þ a2 p1;5 b2 q1;5 201 200 199 202
:
(33)
The matrix system to be solved is simply Ax = b, where A is a coefficient matrix and b are the known potential values at each of the collocation points: 3 2 fˆ 1 6 ˆ1 7 6 f2 7 6 1 7 6 6 7 6 6f 6 ˆ3 7 ¼ 61 6 ˆ 7 61 6 f4 7 6 7 41 6 4f ˆ 5 2
5
x1 x2 x3 x4 x5
y1 y2 y3 y4 y5
p1;1 p1;2 p1;3 p1;4 p1;5
q1;1 q1;2 q1;3 q1;4 q1;5
3
2 3 7 a0 76 a 1 7 76 7 76 b 1 7: 76 7 74 a 2 5 5
(34)
b2
205 203 204 206 207 208 209 210 211
Once the system has been configured, substitute the coordinates of the collocation points into the second and third column of the coefficient matrix. It is also necessary to calculate the radius and angle of the collocation points from the singleton node, (see Fig[15_TD$IF]. 6). The final step is to solve the matrix system. Once these values are known, they can be substituted back ˆ ðzÞ. The f ˆ function can now be used to approximate all the potential values into the original equation for f within the problem domain. Collocation can also be used in the same way to solve for the streamline ˆ ðzÞ. The most significant change is that instead of solving for a0, one must solve for b0. equation, c
212
[27_TD$IF]Least squares
213 214 215 216 217
For least squares, the approach is similar except now we will use the evaluation points on the known boundary to create an overdetermined system. From Fig. 1, the values, v j ðzÞ are known for either f j ðzÞ or c j ðzÞ or both. We can now create a vector of measurements along the boundary such that values v1 . . . vm are available. Using n basis functions (z zj) ln(z zj), we wish to find the complex coefficients, complexes; j ðzÞ such that the approximating function
[(Fig._6)TD$IG]
Branch Cut
Fig. 6. Solving for R and u in Eq[6_TD$IF]. (32).
Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
218
vˆ ðzÞ ¼
n X complexes; j ðzÞðz z j Þlnðz z j Þ
e9
(35)
j¼1
221 220 219
minimizes the sum of squares such that Fðcomplexes;1 ; complexes;2 ; ; complexes;n Þ 0 12 m n X X @ bk complexes; j ðzÞðz z j Þlnðz z j ÞA ¼ min: ¼ j¼1
(36)
j¼1
224 223 222 225
The need for using [17_TD$IF]Eq. (19) becomes apparent when determining the approximate boundary which is associated with the CVBEM approximator functions, v ˆ k ðzÞ.
226
[28_TD$IF]Approximate boundary
227 228 229 230 231
In applying the CVBEM to mixed boundary problems, it is necessary to develop an approximate ˆ , upon which v ˆ k ðzÞ satisfies the problem boundary conditions. Engineering problems boundary, G related to stress and strain are adequate candidates for CVBEM analysis. For stress-free boundary ˆ is the collection of points defined by conditions, G
1 2
Gˆ ¼ z : fˆ ðzÞ ¼ jz2 j ;
(37)
234 233 232 235 236 237 238 239
ˆ ðzÞ. Also |z|2 = x2 + y2 where |z| is measured from a selected central point in V. If ˆ ðzÞ þ ic ˆ where wðzÞ ¼f ˆ G coincides with G, then necessarily vˆ ðzÞ ¼ vðzÞ on V [ G. The utility of the approximate boundary concept is in the evaluation of the approximation error. Instead of the analysis of abstract error quantities, the goodness of approximation is determined by visually inspecting the closeness-of-fit ˆ and G. In those regions, where G ˆ deviates substantially from G, additional evaluation points between G are placed to reduce the approximation errors from using the selected shape functions.
240
[31_TD$IF]Analysis and numerical results
241 242 243 244 245
As an example of the complex variable boundary element method consider the twisting behavior of a homogeneous, isotropic shaft of an arbitrary, but uniform, cross section that is fixed at one end and subjected to a twisting couple at the other end. If the force and deformation behavior is of interest at some location somewhat removed from either end, then the stress and strain characteristics of the cross section as depicted in Fig[15_TD$IF]. 7 are described by either of the following equations [3]:
248 247 246
251 250 249 252 253 254
@2 cðx; yÞ @2 cðx; yÞ þ ¼ 0; @x2 @y2
(38)
@2 fðx; yÞ @2 fðx; yÞ þ ¼ 0: @x2 @y2
(39)
The quantity c(x, y) is the warping function of the cross-section whereas f(x, y) is the conjugate of c(x, y). If the warping function is known over the cross-section, then the out-of-plane warping displacement and the in-plane shear stresses can be calculated from the expressions
v ¼ ucðx; yÞ; 257 256 255 258 259
t xz ¼ mu
@cðx; yÞ y ; @x
t yz ¼ mu
@cðx; yÞ x : @y
(40)
In the above expressions u is the angle of the twist per unit length, m is the shear modulus, and x, y denote the coordinates of a point located from the center of twist. Furthermore, it should be noted that z represents a coordinate axis and should not be confused with the complex variable z = x + iy. If, on the Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
[(Fig._7)TD$IG]
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
e10
y
Evaluation points
Problem Domain
0
x
Problem Boundary
Simple closed contour, C Nodes
Fig. 7. Nodes placed outside the domain highlight an addition degree of freedom. Axis symmetry allows simplification of the problem to the first quadrant.
260 261
other hand, the problem is posed in terms of the complementary function f(x, y) then the shear stresses are determined from @fðx; yÞ @fðx; yÞ t xz ¼ mu y ; t yz ¼ mu þx : (41) @y @x
264 263 262 265 266 267 268 269 270 271 272 273 274 275 276
While the form of [32_TD$IF]Eqs. (38) and (31) are identical, a solution strategy emerges depending on the manner in which the boundary conditions are specified. If the boundary condition of zero normal stress around the perimeter is posed, then a Neumann boundary condition, i.e. specified normal derivative, best describes the problem. In such a case the nonuniform torsion problem is best posed in terms of the warping function, c(x, y). If on the other hand, the problem is best posed in terms of zero shear around the perimeter, then a Dirichlet boundary condition, i.e. specified functions, best describe the problem. In such a case the problem is best posed in terms of the complementary function, f(x, y). While either solution method is well adapted for solid shafts, it is generally more convenient to operate directly with the warping function, c(x, y), rather than its conjugate, f(x, y), for hollow cross-sections. The following example is used to analyze the CVBEM with established solutions [7] for shaft crosssections of smooth and sharp corner profiles. Consider the torsion of a solid elliptical cross section with major axis a and minor axis b. The shear[3_TD$IF]–stress-free boundary condition can be expressed in terms of the conjugate function f(x, y) expressed on the boundary as 1 fðx; yÞ ¼ ðx2 þ y2 Þ (42) 2
279 278 277
282 280 281 285 284 283
The conjugate function f(x, y) as well as the shear stresses can be shown to be 2 1 1 x y2 fðx; yÞ ¼ ðx2 þ y2 Þ a2 b2 2 þ 2 1 ða2 þ b2 Þ 2 a b 2ya2 t xz ¼ mu 2 ða2 þ b Þ
(43) (44)
2
t yz ¼ mu
2xb ða2
2
þb Þ
(45)
Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
[(Fig._8)TD$IG]
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
e11
7.12 7.10 7.10
Exact CVBEM Approx
7.08 7.08
7.06 7.06
7.04
7.04 0.1
0.2
0.3
0.4
0.5
0.1
(a) 6 nodes 16 evaluation points
0.2
0.3
0.4
0.5
(b) 6 nodes 32 evaluation points
Fig. 8. Approximate boundary for 6 node collocation and least squares models taken along a very small portion of the elliptical boundary.
287 286 288 289 290 291 292 293 294 295 296 297 298 299 300 301
Fig[15_TD$IF]. 8 displays the approximate boundary for 6 node (a) collocation and (b) least squares models taken along a very small portion of the elliptical boundary. The deviation of Fig[34_TD$IF]. 8a is due to error. Fig[35_TD$IF]. 8b shows some error, but nearly as much. Fig[15_TD$IF]. 9 shows the relative error for the 6-node model, respectively. The CVBEM relative error along the quarter elliptical section boundary is computed as ˆ ðx; yÞ fðx; yÞ=fðx; yÞ where f ˆ ðx; yÞ is the CVBEM approximate solution and f(x, y) is the exact ½f solution. The quarter model of Fig[15_TD$IF]. 7 was chosen to take advantage of the problem symmetry and to demonstrate the imposition of f boundary conditions along the exterior curved edge and c along the interior straight edge which is the side extending from the origin to the point (a, 0) and the line extending from the origin to the point (0, b) where a and b are 6.25 and 3.75 respectively. Table 1 summarizes the exact and computed warping function and shear[3_TD$IF]–stress values at points in V using the collocation method. Table 2 demonstrates the exact and warping function calculation using least squares for the same number of basis functions, which are the nodes. The graphical depiction of the CVBEM in Fig[15_TD$IF]. 10 uses computer programs MATLAB, Mathematica, and MATLink to model the mixed boundary problem. Fig[15_TD$IF]. 11 is a time analysis of the efficiency of collocation compared to least squares.
302
[36_TD$IF]Additional information
303 304 305
The complex variable boundary element method (CVBEM) has been shown to be a mathematically sound approach for modeling two-dimensional potential problems [2]. The foundations of the CVBEM method rests in complex variable theory, namely, the Cauchy integral formula. It tells us that if a [(Fig._9)TD$IG]
0.004 0.004
0.003
0.003
0.002
0.002
0.001
0.001
1
2
3
4
(a) 6 nodes 16 evaluation points
5
6
0
1
2
3
4
5
6
(b) 6 nodes 32 evaluation points
ˆ ðx; yÞ fðx; yÞ=fðx; yÞ where f ˆ ðx; yÞ is Fig. 9. CVBEM relative error along the quarter elliptical section boundary computed as ½f the CVBEM approximate solution and f(x, y) is the exact solution.
Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
e12
Table 1 CVBEM 6-node model [7_TD$IF]collocation method. x
y
Exact
CVBEM(f)
Error(%)
Shear(txz)
CVBEM(txz)
Error(%)
1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4 5 5 5 5 5
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
10.3401 9.63419 8.45772 6.81066 4.69301 11.046 10.3401 9.1636 7.51654 5.3989 12.2224 11.5165 10.3401 8.69301 6.57537 13.8695 13.1636 11.9871 10.3401 8.22243 15.9871 15.2813 14.1048 12.4577 10.3401
10.2159 9.55065 8.4218 6.82913 4.74572 10.9448 10.2775 9.14085 7.53193 5.43676 12.1451 11.4729 10.3304 8.71804 6.64249 13.8153 13.1385 11.9904 10.3715 8.28744 15.9574 15.2755 14.1217 12.4938 10.3757
1.201 8.671 * 101 4.248 * 101 2.712 * 101 1.123 9.16 * 101 6.053 * 101 2.483 * 101 2.047 * 101 7.013 * 101 6.331 * 101 3.794 * 101 9.354 * 102 2.878 * 101 1.021 3.91 * 101 1.904 * 101 2.741 * 102 3.035 * 101 7.906 * 101 1.862 * 101 3.74 * 102 1.197 * 101 2.897 * 101 3.449 * 101
1.47059 2.94118 4.41176 5.88235 7.35294 1.47059 2.94118 4.41176 5.88235 7.35294 1.47059 2.94118 4.41176 5.88235 7.35294 1.47059 2.94118 4.41176 5.88235 7.35294 1.47059 2.94118 4.41176 5.88235 7.35294
1.43278 2.89735 4.36002 5.82853 7.35907 1.43203 2.90204 4.37161 5.84846 7.3463 1.43723 2.90729 4.37762 5.84635 7.29947 1.44091 2.91248 4.38371 5.85362 7.30922 1.4448 2.9181 4.38984 5.86819 7.37614
2.571 * 102 1.49 * 102 1.173 * 102 9.149 * 103 8.335 * 104 2.622 * 102 1.331 * 102 9.101 * 103 5.761 * 103 9.031 * 104 2.268 * 102 1.152 * 102 7.739 * 103 6.12 * 103 7.272 * 103 2.018 * 102 9.758 * 103 6.358 * 103 4.884 * 103 5.946 * 103 1.754 * 102 7.845 * 103 4.969 * 103 2.408 * 103 3.155 * 103
Table 2 CVBEM 6-node model [8_TD$IF]least squares method. x 1 1 1 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4 5 5 5 5 5
y 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
Exact 10.3401 9.63419 8.45772 6.81066 4.69301 11.046 10.3401 9.1636 7.51654 5.3989 12.2224 11.5165 10.3401 8.69301 6.57537 13.8695 13.1636 11.9871 10.3401 8.22243 15.9871 15.2813 14.1048 12.4577 10.3401
CVBEM(f) 10.3158 9.61789 8.45026 6.81127 4.67955 11.0267 10.328 9.15949 7.5224 5.42563 12.2078 11.5081 10.3379 8.69823 6.60224 13.859 13.1589 11.9871 10.3374 8.19223 15.9812 15.2803 14.1121 12.4714 10.2999
Error(%) (
2.352 * 10 1) 1.692 * 10( 1) 8.825 * 10( 2) 8.974 * 10( 3) 2.868 * 10( 1) 1.745 * 10( 1) 1.166 * 10( 1) 4.484 * 10( 2) 7.794 * 10( 2) 4.951 * 10( 1) 1.199 * 10( 1) 7.308 * 10( 2) 2.071 * 10( 2) 6.004 * 10( 2) 4.087 * 10( 1) 7.583 * 10( 2) 3.61 * 10( 2) 1.129 * 10( 4) 2.601 * 10( 2) 3.673 * 10( 1) 3.693 * 10( 2) 6.438 * 10( 3) 5.172 * 10( 2) 1.101 * 10( 1) 3.884 * 10( 1)
Shear(txz)
CVBEM(txz)
Error(%)
1.47059 2.94118 4.41176 5.88235 7.35294 1.47059 2.94118 4.41176 5.88235 7.35294 1.47059 2.94118 4.41176 5.88235 7.35294 1.47059 2.94118 4.41176 5.88235 7.35294 1.47059 2.94118 4.41176 5.88235 7.35294
1.46325 2.93276 4.40243 5.87834 7.39894 1.46358 2.93367 4.40323 5.86995 7.31646 1.46453 2.93486 4.40548 5.8723 7.31058 1.46465 2.93558 4.40877 5.89386 7.39874 1.46561 2.93539 4.40116 5.88836 7.48702
4.989 * 10( 3) 2.862 * 10( 3) 2.115 * 10( 3) 6.822 * 10( 4) 6.256 * 10( 3) 4.764 * 10( 3) 2.552 * 10( 3) 1.935 * 10( 3) 2.109 * 10( 3) 4.962 * 10( 3) 4.117 * 10( 3) 2.146 * 10( 3) 1.424 * 10( 3) 1.71 * 10( 3) 5.761 * 10( 3) 4.036 * 10( 3) 1.904 * 10( 3) 6.778 * 10( 4) 1.957 * 10( 3) 6.229 * 10( 3) 3.386 * 10( 3) 1.966 * 10( 3) 2.404 * 10( 3) 1.02 * 10( 3) 1.823 * 10( 2)
Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
[(Fig._10)TD$IG]
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
e13
ˆ ðzÞ, ˆ ðzÞ and c Fig. 10. Simultaneous contour plots of the streamline and potential functions. The two real-valued functions, f relate stress and strain within the domain and are the real and complex components of the complex-valued approximating ˆ ðzÞ. ˆ ðzÞ þ ic function v ˆ ðzÞ ¼ f
[(Fig._1)TD$IG]
Fig. 11. Time comparison for collocation versus least squares method.
306 307 308 309 310
function, v, is analytic within and on a simple closed contour, G, then the values of v interior [37_TD$IF]to G are completely determined by the values on G. In other words, to determine the values within the domain, V, one simple need only know the values on the boundary. Thus, an approximation function is developed that is analytic over the problem domain (i.e. possesses derivatives of all orders) and has both real and imaginary parts which exactly solve the Laplace equation within V. Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
G Models
MEX 86 1–14
A.N. Johnson, T.V. Hromadka II / MethodsX xxx (2015) e1–e14
e14
311 312 313 314 315 316 317 318 319 320 321 322 323
Error analysis is an area that makes the CVBEM particularly appealing. It is unique among other numerical methods in that the CVBEM approximating function can be evaluated directly to analyze the error of the approximation. This is because the CVBEM develops an exact representation of the modeling error by the determination of an ‘approximate boundary’ where the CVBEM approximation exactly satisfies the boundary conditions. That is, the approximate boundary is the locus of points where the CVBEM approximation meets the boundary condition values. This approach is in stark contrast to other numerical methods in their analysis of error. Two sources of error are primary in most numerical methods. The two sources consists of error that results from the approximation in solving the governing equation, and error that result in solving the boundary conditions continuously. Popular numerical methods such as finite elements (FEM) and finite differences (FDM) generate both types of errors in modeling potential problems. Model accuracy is usually estimated by comparing the change in results by increasing the number of nodal points. By this the analyst is seeking convergence by showing that the result is both stable and consistent.
324
[38_TD$IF]Acknowledgment
325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341
MethodsX thanks the (anonymous) reviewers of this article for taking the time to provide valuable Q2 feedback.
References [1] T.V. Hromadka, A Multi-Dimensional Complex Variable Boundary Element Method, Volume 40 of Topics in Engineering, WIT Press, Billerica, MA, 2002. Q3 [2] T.V. Hromadka, G.L. Guymon, The complex variable boundary element method, Int. J. Numer. Methods Eng. (1984). [3] T.V. Hromadka, C. Lai, The Complex Variable Boundary Element Method, Springer-Verlag, New York, 1987. [4] T.V. Hromadka, R.J. Whitley, Advances in the Complex Variable Boundary Element Method, Springer, New York, 1998. [5] A.N. Johnson, T.V. Hromadka, M. Carroll, M. Hughes, L. Jones, N. Pappas, C. Thomasy, S. Horton, R. Whitley, M. Johnson, A computational approach to determining CVBEM approximate boundaries? Eng. Anal. Bound. Elem. 41 (0) (2014) 83–89. [6] A.N. Johnson, T.V. Hromadka, M. Hughes, S. Horton, Modeling mixed boundary problems with the complex variable boundary element method (CVBEM) using matlab and mathematica, Comput. Methods Exp. Meas. (2015) (submitted for publication). Q4 [7] N. Muskhelishvili, Some Basic Problems of the Mathematical Theory of Elasticity, 1sd ed., Springer, 1977. [8] R.J. Whitley, T.V. Hromadka, Theoretical developments in the complex variable boundary element method, Eng. Anal. Bound. Elem. 30 (12) (2006) 1020–1024, Complex Variable Boundary Element Method Complex variable boundary element method.
Please cite this article in press as: A.N. Johnson, T.V. Hromadka II, Modeling [9_TD$IF]mixed boundary conditions in a Hilbert space with the complex variable boundary element method (CVBEM), MethodsX (2015), http://dx.doi.org/10.1016/j.mex.2015.05.005
