Sunteți pe pagina 1din 264

Fluid Locomotion and Trajectory Planning for Shape-Changing Robots

Thesis by

Richard Mason
In Partial Fulllment of the Requirements for the Degree of Doctor of Philosophy

California Institute of Technology Pasadena, California

2003 (Defended June 17, 2002)

ii

2003

Richard Mason All Rights Reserved

iii

Acknowledgements
The greatest thanks are due to my advisor, Joel Burdick, for his long support. Thanks to Jim Radford for useful discussions on the subject of inertia and for the use of JMath 0.9.11, and to Darrin Edwards for the translation of Russian uids literature. Many thanks to Kristi Morgansen and to the many other people who have worked to develop the robot sh at Caltech, including Scott Kelly, Carl Anhalt, Victor Duindam, Liu Fong, Justin Kao, Bob Keeney, Seth Lacy, Howen Mak, Brian Platt, Wendy Saintval, Nate Senchy, Nathan Schara, Susan Sher, Chris Storey, Jens Taprogge, and Laron Walker. And loving thanks to my patient wife, Maribeth.

iv

Abstract
Motivated by considerations of shape changing propulsion of underwater robotic vehicles, I analyze the mechanics of deformable bodies operating in an ideal uid. I give particular attention to shlike robots which may be considered as one or more exing or oscillating hydrofoils. I then describe methods of planning trajectories for a shlike robot or any other sort of robot whose locomotion has a periodic or quasi-periodic nature. This doctoral work was performed by Richard Mason at the California Institute of Technology. My thesis advisor was Professor Joel W. Burdick.

Contents
Acknowledgements Abstract 1 Introduction 1.1 1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary and Relation to Previous Work . . . . . . . . . . . . . . . . iii iv 1 1 3 7 7 10 10 11 11 13 18 22 24 25 27 28 31 45 54 58 60

2 Smooth Deformable Bodies in an Ideal Fluid 2.1 2.2 Potential Flow Around a Smooth Deformable Body . . . . . . . . . . Ideas from Geometric Mechanics . . . . . . . . . . . . . . . . . . . . . 2.2.1 2.2.2 2.2.3 2.3 2.4 Principal Fiber Bundles . . . . . . . . . . . . . . . . . . . . . Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The Fluid Mechanical Connection . . . . . . . . . . . . . . . . . . . . Planar Amoeba Example . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 2.4.2 2.4.3 Displacement by Periodic Motion . . . . . . . . . . . . . . . . Rectilinear Motion Planning . . . . . . . . . . . . . . . . . . . Optimal Control Analysis for Amoeba . . . . . . . . . . . . .

3 Deformable Joukowski Foils 3.1 3.2 3.3 3.4 3.5 Allowed Deformations of the Foil . . . . . . . . . . . . . . . . . . . . The Potential Function . . . . . . . . . . . . . . . . . . . . . . . . . . The Locked Added Inertia . . . . . . . . . . . . . . . . . . . . . . . . Vortex Shedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Forces on the Foil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.1 Unsteady Flow Force Terms . . . . . . . . . . . . . . . . . . .

vi 3.5.2 3.6 3.7 3.8 Bernoulli Eect Force Terms . . . . . . . . . . . . . . . . . . . 70 72 86 92 93 98 98

Moments on the Foil . . . . . . . . . . . . . . . . . . . . . . . . . . . Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . Self-Propulsion of a Deformable Foil . . . . . . . . . . . . . . . . . . . 3.8.1 Uniform Foil Density . . . . . . . . . . . . . . . . . . . . . . .

4 Robot Carangiform Fish 4.1 4.2 4.3 4.4 Review of Carangiform Swimming . . . . . . . . . . . . . . . . . . . .

Description of the Experiment . . . . . . . . . . . . . . . . . . . . . . 101 Model with Added Mass and Quasi-Steady Lift . . . . . . . . . . . . 105 Miscellaneous Experimental Observations . . . . . . . . . . . . . . . . 118 125 133

5 Optimal Control 6 Trajectory Planning Using Density Functions 6.1 6.2 6.3 6.4 6.5

Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 Summary of the Approach . . . . . . . . . . . . . . . . . . . . . . . . 134 Density Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 Order Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140 Computing the Density of Reachable States . . . . . . . . . . . . . . 143 6.5.1 The Fourier Transform . . . . . . . . . . . . . . . . . . . . . . 144 . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

6.6

Trajectory Planning 6.6.1 6.6.2 6.6.3 6.6.4 6.6.5 6.6.6

Dijkstras Algorithm in Trajectory Planning . . . . . . . . . . 146 The Ebert-Upho Algorithm . . . . . . . . . . . . . . . . . . . 149 Implicit Storage of Paths in Graph . . . . . . . . . . . . . . . 149 Multiscale Densities . . . . . . . . . . . . . . . . . . . . . . . . 151 A Divide-and-Conquer Algorithm . . . . . . . . . . . . . . . . 155 Obstacle Navigation with Density Functions . . . . . . . . . . 156 161 163

7 Conclusions and Future Work Bibliography

vii A Integrals for Unsteady Flow Forces B Integrals for Bernoulli Eect Forces C Coecients in Equation (3.156) D Coecients in Equation (3.160) 169 183 203 218

viii

List of Figures
2.1 2.2 Schematic of deformable swimming body . . . . . . . . . . . . . . . . The variables s1 (t), s2 (t), s3 (t) correspond to 3 deformation modes. The 1st and 2nd modes together yield motion in the x-direction. The 1st and 1st modes together yield motion in the y-direction. . . . . . . 2.3 Computer animation of a planar amoeba whose density equals that of the surrounding uid. There are roughly 15.3 oscillations between snapshots. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 The Joukowski map z = + c + a2 /( + c ) sends a circle to an airfoil contour. Here a = 1, rc = 1.1, and c = (0.0781 + 0.2185i). . . . . . 3.2 Fluid velocity resulting from the unit potentials w1 , w2 , and w3 (top to bottom.) Here a = 1, rc = 1.1, and c = (0.0781 + 0.2185i). Note that I took U to be negative in the rst gure. Negative U , or velocity in the negative x-direction, corresponds to forward motion when x is negative, as here. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Fluid velocity resulting from the shape deformation unit potentials
s s s w1 , w2 , and w3 (top to bottom.) Here a = 1, rc = 1.1, and c =

18

22

27

41

(0.0781 + 0.2185i). . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Plots of locked added inertia coecients for a variety of Joukowski foils. In each plot, rc and arg(c ) are held constant while c is varied. The
2 coecients I11 , I12 , I13 are given in units of rc , while I13 , I23 are given 3 4 in units of rc and I33 has units of rc . . . . . . . . . . . . . . . . . . .

42

50

3.5

More plots of locked added inertia coecients for a variety of Joukowski foils. In each plot, rc and arg(c ) are held constant while c is varied.
2 The coecients I11 , I12 , I13 are given in units of rc , while I13 , I23 are 4 3 given in units of rc and I33 has units of rc . . . . . . . . . . . . . . . .

51

ix 3.6 At each time step a new vortex is placed at the one-third point on a circular arc tangent to the trailing edge and intersecting the last vortex shed. The marks are the discrete vortices while the + marks delimit the wake segments which the vortices represent. . . . . 3.7 Wake from a heaving and exing deformable Joukowski foil. The foil began its motion with its centroid at the origin. The centroid then translated with uniform velocity Ucm = 1 in the x-direction while heaving in the y-direction with velocity Vcm = (0.4 sin t). Meanwhile the foil shape underwent a sinusoidal exing motion with (x , y ) = (0.05, 0.4 sin t). The modelled wake behavior at t = 12 is shown with the time step t between shed vortex points being (top to bottom) t = 0.12, t = 0.06, t = 0.04. . . . . . . . . . . . . . . . . . . . . 3.8 If the direction of integration is counterclockwise in each case, then

55

59

Sx

Sv

Sb

. . . . . . . . . . . . . . . . . . . . . . . . .

63 71

3.9

2 3.10 The contour Cp encloses a pole at = rc /c . There is a branch cut

I consider a small excursion around the pole at = a c . . . . . . . between the origin and = c . If integration is counterclockwise in each case, then
C

2 3.11 The contour Cp encloses a pole at = rc /c . If the direction of inte-

Cx

Cp

Cb2

. . . . . . . . . . . . . . .

75

gration is counterclockwise in each case, then

Cx

Cp

Cv

Cb

. 79

3.12 Flapping motion of a foil with (Ucm , Vcm ) = (1, 0), = 0.3 cos(t), (x , y ) = (0.15, 0.3 sin(t)), rc = 1. The point vortices are being shed at intervals of t = 2/100 = 0.0628. The snapshots shown are at intervals of 10 t = 2/10. . . . . . . . . . . . . . . . . . . . . . . . . 3.13 Continuation of Figure 3.12. Flapping motion of a foil with (Ucm , Vcm ) = (1, 0), = 0.3 cos(t), (x , y ) = (0.15, 0.3 sin(t)), rc = 1. The point vortices are being shed at intervals of t = 2/100 = 0.0628. The snapshots are at intervals of 10t = 2/10. . . . . . . . . . . . . . . . 90 89

x 3.14 Forces and moments on the apping foil of Figures 3.12-3.13, in a stationary frame oriented with the reader. While the oscillating force in the y-direction has an average approaching zero, the oscillating thrust in the x-direction has a positive mean value of 0.23. The moment also has a signicantly positive mean value. . . . . . . . . . . . . . . . . . 3.15 A foil starting from rest executes a turn by making the deformation (x , y ) = (0.15, 0.3 sin(t)). . . . . . . . . . . . . . . . . . . . . . . . . 3.16 The continued progress of the oscillating foil from Figure 3.15. When the foil drives through its own wake, I assume that any vortices which come too close to the foil are annihilated. . . . . . . . . . . . . . . . . 4.1 4.2 Schematic (side and top view) of Carangiform sh propulsion. . . . . A cartoon of carangiform swimming consisting of snapshots of the tail as the sh swims from left to right. The arrows indicate lift forces acting on the tail. The sh body is removed for clarity. . . . . . . . . 100 4.3 A two-jointed sh model. Our experiment replaces the three links with at plates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 4.4 4.5 4.6 4.7 4.8 4.9 Side view schematic and photograph of experiment. . . . . . . . . . . 102 Top view schematic and rear top photo of experiment. . . . . . . . . 103 97 99 96 91

Bottom view of an earlier version of experiment. . . . . . . . . . . . . 104 The idealized model. . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 Drag force acting on the sh in forward motion. . . . . . . . . . . . . 107 Drag force acting on the sh at varied angles of attack. . . . . . . . . 107

4.10 Experiment and simulation for the forward gait with 1 = 0.6 sin(4.0t), 2 = 0.3 sin(4.0t + /3). A PD controller drives the joints to follow the desired trajectory. The simulation slightly overpredicts thrust (or underpredicts drag.) . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

xi 4.11 Forward thrust generated by the gait 1 = 0.6 sin(4.0t), 2 = 0.3 sin(4.0t+ /3), from Figure 4.10. Almost immediately the sh enters a periodic cycle where lift is the dominant source of thrust, but added mass forces on the tailn are signicant at the very start of thrust. . . . . . . . . 113 4.12 Experiment and simulation for the gait 1 = 1.2 sin(2.0t), 2 = 0.6 sin(2.0t+ 5/18). Both lift forces and added mass forces are signicant components of the thrust. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 4.13 A gait which results in stable forward propulsion even when the sh is free to rotate or deviate. Note that added mass forces dominate lift and drag forces. Since is near , the forward direction is along the positive x-axis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 4.14 Lateral forces and moment generated by the gait in Figure 4.13. . . . 117 4.15 A turning gait. Note that added mass eects dominate the moment. Since starts near zero, the forward direction is initially aligned with the negative x-axis. . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

4.16 Forces experienced in the turn of Figure 4.15. The x and y forces are in a world frame, not xed in the body of the sh. . . . . . . . . . . . 120 4.17 Steady-state velocities for single-frequency sinusoidal gaits. In each subplot, frequency and peduncle amplitude are held constant while steady-state velocity is shown as a function of the tailn amplitude and phase. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 4.18 Steady-state velocities of another set of gaits with higher amplitude of peduncle oscillation. . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 4.19 Distance traveled by the sh in eight seconds, for gaits of the form 1 = 1.3 sin(3.5t) and 2 = A sin(3.5t + ). The horizontal axes represent the parameters A and . . . . . . . . . . . . . . . . . . . . . . . . . . 123 4.20 Strouhal numbers, for gaits of the form 1 = 1.3 sin(3.5t) and 2 = A sin(3.5t + ). The horizontal axes represent the parameters A and . 124 4.21 Orientation change over time, for gaits of the form 1 = 0.6 + 1.3 sin(3.5t) and 2 = 1.1 sin(3.5t + ). . . . . . . . . . . . . . . . . . 124

xii 5.1 A bang-bang maneuver, starting from rest, that produces both acceleration and displacement in the x-direction, while producing less acceleration and almost no displacement in the y- and directions. The joint accelerations (1 , 2 ) are (2, 2) from t = 0 to t = 0.29, then (2, 2) to t = 0.97 The values of 1 , 2 oscillate rapidly just before the end of the trajectory. . . . . . . . . . . . . . . . . . . . . . 131 6.1 6.2 6.3 6.4 6.5 6.6 This sh robot can reach any of four positions in one tail stroke. . . . 135 Positions reachable in ve or fewer tail strokes. . . . . . . . . . . . . . 136 States in Figure 6.2 viewed as a density function in conguration space. 137 Density from Figure 6.3 using a coarser discretization. . . . . . . . . . 141 Discretization V0 and coarser discretization V1 . . . . . . . . . . . . . . 142 The density P in the vicinity of a particular goal, as a function of P . The path length P can be chosen so that P is suciently high. Often, P increases with increasing P , so a longer path may yield a better approximation to the goal. . . . . . . . . . . . . . . . . . . . . . . . . 150 6.7 The algorithm nds a 10-step path to the goal. While the nal heading is not quite right, this is roughly the shortest path with possibly acceptable error. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 6.8 Higher accuracy can be obtained by varying P to obtain a more dense P . This 14-step path ends in a better approximation to the goal. . . 152 6.9 A 16-step path reaches the goal conguration almost perfectly. . . . . 153

xiii

List of Tables
3.1 3.2 3.3 6.1 6.2 Solution for the deformable Joukowski foil potential function. . . . . . 43 Derivatives with respect to of the deformable Joukowski foil potentials. 44 Components of the locked added inertia for the Joukowski foil. . . . . 53

Resolution measurements in SE(2). . . . . . . . . . . . . . . . . . . . 144 Time and space bounds, where N is the number of volume elements, P is the number of trajectory steps, K is the number of possible motions at each step, and the constants and depend upon the chosen methods of function maximization and matrix multiplication. Although the Barraquand/Latombe algorithm has the lowest time and space requirements, it does not have all the same utility as the algorithms presented here; notably it does not allow task-by-task trade-os between path length and accuracy. . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

Chapter 1
1.1

Introduction

Motivation

In this dissertation, I attack several problems relating to the general theme of locomotion in a uid by a shape-changing body. The underlying motivation is to better understand the propulsion and control of biomimetic underwater vehicles. Not surprisingly, many aquatic creatures are impressive swimmers, and observers have argued that the locomotion of various biological swimmers, from sh to cetaceans to penguins, must be exceptionally ecient. Such proposals date at least back to 1936 when Gray estimated the muscle power available to a dolphin and how much power should be required to propel a dolphin-shaped object through the water at delphinine speeds [Gra36]. Grays Paradox is that the latter estimate was seven times higher than the former estimate, implying that dolphins should not be able to swim as fast as they do. Over the decades various aws have been exposed in Grays calculations, but the idea persists that dolphins and other biological swimmers may be unexpectedly ecient, having evolved to exploit some unusual uid mechanics. More recently, Triantafyllou et al. have proposed that fast-swimming sh use tail motion as a means of vorticity control, minimizing energy lost in the wake and recovering energy from vortices originating ahead of the tail. A exible body utilizing these energy recovery methods could have an apparent drag less than that of a rigid body of the same size and baseline shape. There is experimental support for this theory: notably, the MIT RoboTuna [Bar96, BGT96] was able to achieve signicant drag reduction with a exible oscillating body, suggesting not only that sh-like propulsion can be ecient but that the eect can be reproduced by macroscopic motions of a mechanical device. Since non-nuclear underwater vehicles are often signicantly constrained by the battery energy available to them, any mechanism for increasing their eciency which could be copied from the biological world would be a signicant

2 benet. A variation on the idea of wake control is that if a biomimetic robot swimmer recaptures energy from its wake, it will make less noise in the water. Also, whatever noise it does make will presumably sound more like a sh (or cetacean, et cetera) and not like an ordinary vehicle. So sh-like robots may be useful in stealthy approach or surveillance applications. Biological swimmers are also impressive in being notably more maneuverable than ordinary watercraft. Water vehicles with conventional propulsors and control surfaces typically require several body lengths to make a turn and have low maneuverability at low speeds. The addition of multiple propulsors along lateral axes can improve low-speed maneuverability but at a serious cost in weight, volume, and complexity. Meanwhile, some sh can make a 180-degree turn in a single body length, while predatory sh are capable of accelerations on the order of 100m2 /s2 from a dead stop. There may be a role for high-manueverability vehicles which work in the same way. For all these reasons, in the last decade there has been wide interest in biomimetic swimming robots, which propel themselves through a uid not by conventional thrusters, but by undulations or quasi-periodic changes in shape. Experimental eorts have included attempts to imitate the action of a disembodied sh tailn [AHB + 91, ACS+ 97, SLD00] and tow tank models of entire sh such as the MIT RoboTuna [Bar96, BGT96]. Free-swimming biomimetic robots include the Vorticity Control Unmanned Undersea Vehicle built at the Charles Stark Draper Laboratory [AK99]; the MIT RoboPike project; an eel-like robot at the University of Pennsylvania [CMOM01]; a sh robot at the Ship Research Institute in Tokyo [HTT00]; a dolphin robot at the Tokyo Institute of Technology [NKO00]; an underwater vehicle with paired mechanical pectoral ns at Tokai University [KWS00]; and ongoing eorts to develop robot lampreys [AWO00] and robot penguins [Ban00]. In this thesis I will suggest some modelling and planning techniques relevant to these and future biomimetic vehicles, and also describe an experimental platform constructed at Caltech.

1.2

Summary and Relation to Previous Work

In Chapter 2, I examine the self-propulsion of a smooth body changing shape in a quiescent irrotational potential ow, a problem previously studied by Saman [Saf67] and by Miloh and Galper [Mil91, MG93, GM94]. I specialize to a body with a nite number of modes of deformation and derive the equations of motion emphasizing the connection from geometric mechanics: this may be considered an extension of Ostrowksi and Burdicks work on undulatory locomotion [OB98, Ost95] to uid mechanical systems. I then focus on the particular example of the squirming circle, which was introduced in the case of one-dimensional motion by Kelly and Murray [KM96], and which bears comparison to the experimental amoebot built by Chen [CLC98]. I extend the example to motion in the plane and provide optimal control results for following a kinematic trajectory in xed time while minimizing control eort. In Chapter 3, I present general closed-form expressions for the force and moment on a deformable Joukowski foil executing arbitrary translation and rotation while also undergoing an arbitrary time-dependent change in its shape parameters. My motivation is to model the forces on a bending exible n or exible streamlined body: however, I leave open the possibility that the foil has not only time-varying camber but also time-varying thickness, chord length, and/or area. I assume inviscid ow with a nite number of point vortices, including wake vortices shed from the trailing edge of the foil over time. These results may be considered a generalization of the closed-form expressions found by Streitlien [Str94, ST95] for a rigid Joukowski foil undergoing arbitrary rigid-body motion. Of course, my results include the rigid-body case as a subset. In particular, I provide closed-form expressions for the added inertia coecients of a rigid Joukowski foil. Previously published closed-form expressions for the added inertia [Str94, ST95, Sed65] suer from typographical and other elementary errors such as dropped terms and dropped factors of two. This thesis corrects the historical record, while extending results to the deformable case.

4 In Chapter 4, I describe an experimental three-link robot sh, built by myself and others, which propels both itself and an attached gantry suspended above the water. The dierences between this experimental platform and previous robot sh eorts are as follows: Our platform is fully mobile in the plane and genuinely self-propelled, as opposed to being statically mounted or towed through the water as in [Bar96, BGT96, AHB+ 91, ACS+ 97]. This makes it a better model of free swimming. In particular, the fact that our sh body is free to sideslip or rotate in the plane has a dramatic impact on swimming motions. Useful thrust provided by a tailstroke can be much greater when the body is xed, towed, or constrained to one dimension than when it yaws freely. Also, we were able to consider problems of turning and yaw stability. On the other hand, the restriction to planar motions simplies analysis of the system to a planar problem. The attached gantry enables us to have much more precise telemetry of the shs motion than is feasible for a truly free-swimming platform such as [AK99], enabling us to make more precise comparisons between theory and experiment. This has also enabled feedback control experiments. I compare the experimental results from the platform to the results from theory and computer simulation. It turns out that a relatively simple model incorporating quasi-static lift and added-mass eects is good enough to make qualitative predictions about the platforms swimming behavior. I show that for swimmers executing certain types of maneuvers, lift forces dominate added mass forces, and vice versa. In Chapter 5, I consider the equations of motion for the three-link swimmer in the limit where added mass forces dominate, and how to nd solutions to these equations which are optimal according to a plausible performance index. Once found, these optimal solutions to short-time-horizon planning problems can be pieced together to form larger trajectories. In Chapter 6, I examine the problem of trajectory planning for a mobile robot restricted to a set of nite motions, which can be iterated and concatenated to form

5 larger trajectories. This is a reasonable model for shape-changing robots, which (a) locomote through repeated periodic undulations; (b) will likely have a nite number of gaits or periodic motions which are understood or deemed optimal; and (c) may not be small-time locally controllable. I suggest three planning methods based on the density of points in conguration space reachable by the robot in a certain number of steps. The rst algorithm proposed is directly inspired by the density-based Ebert-Upho algorithm for inverse kinematics of a discrete manipulator, as outlined by Chirikijian and Kyatkin [CK01]. I adapt the Ebert-Upho algorithm to the domain of mobile robots, notionally replacing the links of a discrete manipulator with segments of a path. One important change is that unlike the number of links in an existing manipulator, the number of steps in a mobile robots path can be freely increased or decreased as necessary on a task-by-task basis. I discuss how the mobile robot can make task-by-task trade-os between path length and the precision with which the goal is reached, and illustrate with an example using a model system. I also enhance the Ebert-Upho algorithm by evaluating each density function on several length scales instead of just one, since different length scales are appropriate depending on whether solutions near the desired goal are dense or sparse. The second density-based planning algorithm uses a divide-and-conquer strategy instead of the essentially linear method of the Ebert-Upho algorithm. The divideand-conquer algorithm requires exponentially less memory than the Ebert-Upho algorithm, and is exponentially faster during the expensive mapping/pre-computation phase. At run time, when a goal is provided and a trajectory is planned using the pre-computed map, the divide-and-conquer strategy may be either faster or slower than the linear strategy, depending on how well-behaved the mobile systems density functions are, and whether the divide-and-conquer algorithm can take advantage of parallel processing. Finally, I suggest a way to use density functions to help search for a path through a eld of static obstacles. When compared to existing trajectory-planning methods based on Dijkstras al-

6 gorithm [Lat90], the density-based methods have the following advantages: Path length is a fully controllable parameter. The method can be used to nd not just the shortest path, but a path with any specied acceptable length. This facilitates trade-os between path length and goal accuracy, as mentioned earlier, and could also be valuable in rendezvous problems. Because the density function represents all feasible paths, instead of recording only the best path to each location like Dijkstra-based methods, the densitybased methods are more robust in some ways. If the best path is unexpectedly unusable, the density-based methods are able to plan an alternative trajectory without starting from scratch. It is possible for Dijkstra-based methods to nd a suboptimal solution, or no solution at all, because the correct solution was lost to round-o error. This problem may be especially acute for robots, such as sh robots, which are not small-time locally controllable. The density-based planning methods can also lose solutions to round-o error, but because the density functions can be found by convolution of sets of large motions, instead of by incrementally adding small motions, the problem is less severe. This makes the densitybased methods better suited for problems involving either long trajectories or a coarsely discretized conguration space.

Chapter 2

Smooth Deformable Bodies in

an Ideal Fluid
2.1 Potential Flow Around a Smooth Deformable Body
Real amoebae are microscopically small. They operate at a very low Reynolds number and the relevant uid equations are those of creeping ow [SW89]. However, if we were to build a macroscopic robot amoeba and expect it to swim through water, the Reynolds number of its ambient ow would be much higher and inertial forces will dominate instead of viscous ones. I thus make the reasonable idealization that the robot amoeba is a connected deformable body swimming through an inviscid and incompressible uid. I also assume that the uid is irrotational everywhere, and that the amoeba cannot generate vorticity in the uid. Unless the amoeba grows sharp ns, this too is a reasonable assumption. Any amoeba robot which we might actually construct would have a nite number of actuators. Indeed, we would like to use as few actuators as possible. Therefore, rather than allow the boundary of the amoeba to be innitely variable, I assume that its shape can be described by a nite number, ns , of shape variables, s. The space of all possible shapes, denoted by S, is a nite-dimensional manifold. I x a frame, F B , to the body of the swimmer and let F W denote a xed reference frame. (See Figure 2.1.) The location of the F B is given by g(t) SE(d), d = 2, 3. In coordinates, elements of SE(d) can be represented by homogeneous matrices, g. R p 0
T

g=

(2.1)

8
S

Body Fluid Z1 Y 1 F(,s) X1 n(,s)

Z0 Y 0

X0

Figure 2.1: Schematic of deformable swimming body The matrix R SO(d) describes the orientation of F B with respect to F W , while

p I d is the position of F B s origin. The velocity of the moving reference frame, as R seen by an observer in F B , is g 1 g: g 1 g =

0T 0

(2.2)

where is a d d skew symmetric matrix and I d . The quantity g 1 g is an R element of the Lie algebra of SE(d), se(d). I shall denote by the identication of se(d) with I R body velocities. The swimmers smooth surface, , is parameterized by a coordinate chart that is a function of (d 1) parameters 1 , . . . , (d1) , or by an atlas of (d 1)-dimensional charts. The surface parameters themselves are functions of the shape variables, s1 , , s ns . Given the assumptions described above, the uid motion around the swimmer is described by potential ow, and its domain, D, is assumed to be unbounded. In the most general case, the ambient uid undergoes non-uniform motion. I will assume that the ambient ow is quiescent. The Kircho principle for potential ow around a rigid body [VVP73] can be extended to show that the general uid potential, , for an irrotational uid surrounding a deformable body will take the form:
3 ns
d(d+1) 2

: (g 1 g) = [ T T ]T , where and are the linear and angular

=
i=1

(i g i

i g ) i+3

+
j=1

s s j . j

(2.3)

9 The terms {g } are the standard Kircho potentials for a rigid bodyin this case, for i to the total potential due to body deformations. Let F (, s) denote the location of a surface point with respect to a body xed frame. The normal vector to the surface at that point is denoted n(, s). Then the instantaneous velocity, in the body frame, of the surface point parameterized by is + F (, s) + conditions. g n(, s) = ni (, s) i i = 1, 2, 3
F (, s)si . si

the deformable body at a xed shape, s. The term s terms represent the contribution

At the body surface, the uid velocity and

surface velocity in the normal direction must match, leading to the following boundary

g n(, s) = (F (, s) n(, s))i i = 4, 5, 6 i s n(, s) = F (, s) n(, s) si i si These form separate Neumann problems for the Laplace equation (
2

(2.4) = 0) for each


1

term g or s . A unique solution (up to a constant) exists for each term [NS82]. i i is 0 || ||2 dV Ttotal = Tbody + Tf luid = 1 q T (q)q + 2 2 D gg gs (q) (q) g 0 = 1 (g T sT ) 2 2 (gs )T (q) ss (q) s

The total kinetic energy of the constant density uid and deformable body system

( n) d, (2.5)

where 0 is the uid density, and (q) is the kinetic energy metric (or mass matrix) of the deformable body (in the absence of surrounding uid). Because the potentials
If the uid external to the swimmer is considered to be an unbounded domain, D, then a unique solution (up to an overall constant) is only guaranteed if the velocity of the uid drops o suciently quickly (at least as fast as r 2 ) at innitybut this requirement is physically reasonable. Alternatively, one could adopt the common-sense position that the uid really occupies a large but bounded domain.
1

10 take the form of Equation (2.3), the total kinetic energy can be put in the form
1 Ttotal = 2 q T M (q)q.

For the moment I will ignore any additional potential forces acting on the deformable body, and hence the systems Lagrangian is equivalent to Ttotal . One could next derive the governing mechanics from the Euler-Lagrange equations. Instead I take a more abstract approach.

2.2

Ideas from Geometric Mechanics

New insight can be obtained by applying methods of geometric mechanics to the system described in the previous section. In particular I wish to nd symmetries that lead to reduction. This section provides a brief summary of relevant ideas. More extensive background can be found in [MR94, MS93a].

2.2.1

Principal Fiber Bundles

Let Q denote the conguration space of the deformable swimmer, which consists of its position, g SE(d), and its shape, s S. Hence, the conguration space is SE(d) S, and a conguration q Q can be given local coordinates q = (g, s). This conguration space has a surprisingly rich structure due to the fact that SE(d) is a Lie group. Recall that every Lie group, G, has an associated Lie algebra, denoted g. In our context, the Lie algebra of SE(d), denoted by se(d), consists of the velocities of F B relative to F W , as seen by a body xed observer. Elements of se(d) can be represented by matrices of the form g 1 g, as seen in Equation (2.2). If the swimmers initial body xed frame position is h G, and it is displaced by an amount g, then its nal position is gh. This left translation can be thought of as a map Lg : G G given by Lg (h) = gh. The left translation induces a left action of G on Q. A left action is a smooth mapping : G Q Q such that: (1) e (q) = q for all q Q; and (2) g (h (q)) = Lg h (q) for all g, h G and q Q. The conguration

11 space Q endowed with such an action is a principal ber bundle. Q is called the total space, S the base space (or shape space), and G the structure group. The canonical projection : Q S = Q/G is a dierentiable projection onto the second coordinate factor: (q) = (g, s) = s. The sets 1 (s) Q for s S are the bers, and Q is the union over S of its bers. The usefulness of this structure will become more apparant in Section 2.2.3.

2.2.2

Symmetries

In Lagrangian mechanics, symmetries result in conservation laws. By a symmetry, we mean an invariance of the Lagrangian with respect to some operation. Denition 1. The lifted action is the map T g : Tq Q T(q) Q : (q, v) (g (q), Tq g (v)) for all g G and q Q. I.e., Tq g is the Jacobian of the group action. For left translation on G, T g has the coordinate form: T g Lh g s hg s

Tq h q = of symmetry, or invariance.

(2.6)

Based on the group action and lifted action, we can introduce the following notions

Denition 2. A Lagrangian function, L : T Q I is said to be G-invariant if it R, is invariant with respect to the lifted action, i.e., if L(q, vq ) = L(h (q), Tq h vq ) for all h G and all vq Tq Q.

2.2.3

Connections

To analyze and control propulsion of a deformable body, one would like to systematically derive an expression that answers the question: If I wiggle the bodys surface,

12 how does the body move? The relationship between shape and position changes can be formalized in terms of a connection, an intrinsic mathematical structure that is associated with a principal ber bundle. I begin with some necessary technical denitions. Denition 3. If g, the vector eld on Q denoted by Q and given by Q (q) = d exp(t) (q) dt
t=0

(2.7)

is called the innitesimal generator of the action corresponding to . The vertical space, Vq Q Tq Q, is dened as Vq Q = ker(Tq ) = {v| v = Q (q) g}. Denition 4. The connection one-form, (q) : Tq Q g, is a Lie-algebra valued one-form having the following properties: (i) (q) is linear in its action on Tq Q. (ii) (q)Q = for g. (iii) (q)q is equivariant, i.e., it transforms as (h (q))Tq h (q)q = Adh (q)q, where the adjoint action Adh : g g is dened as Adh = Th1 Lh (Te Rh1 ) for g. The innitesimal generator of the adjoint representation is given by ad = g = [, ] for , g. (2.8)

The horizontal space is the kernel of the connection one-form, Hq Q = {z | (q)z = 0}, and is complementary to Vq Q. It can be shown that the connection one-form can be expressed in local coordinates q = (g, s) as follows: (q)q = Adg (A(s)s + g 1 g), (2.9)

where A : T S g is termed the local form of the connection. Hence, any q = (g, s) which lies in Hq Q must satisfy a constraint of the following form: g 1 g = A(s)s. (2.10)

13 Note that the local connection plays a central role in the ensuing analysis. For Lagrangian systems with symmetries, such as the one studied here, the conservation laws that arise from symmetries constrain the overall system motion. This constraint can be expressed as a connection, called the mechanical connection [MR94]. It can be shown that the mechanical connection is given by the expression (q)vq = I 1 (q)J(vq ), where the momentum map, J : T Q g , satises: J(vq ); = vq , Q , for all g and vq Tq Q. The expression , denotes inner produce with respect to the kinetic energy metric. The locked inertia tensor is the map, I(q) : g g which satises I(q); = Q , Q for all , g. (2.11)

2.3

The Fluid Mechanical Connection

I now revisit the mechanics of the deformable swimmer in light of the ideas presented in the last section. I rst show that a Lagrangian that is invariant with respect to a group action h induces a reduced Lagrangian. This is a general result that is independent of the particular uid mechanical model that is studied in this paper. Proposition 2.3.1. [Ost95] If L(q, q) is a G-invariant Lagrangian, i.e., it satises Denition 2, then the reduced Lagrangian, l : T Q/G I can be expressed as R, I(s) IA(s) V (s), l(s, s, ) = T sT (2.12) AT I(s) m(s) s

where = g 1 g g, s Tr S, I(s) is the locked inertia tensor, and A(s) is the local form of the mechanical connection.

14 Proof. The Lagrangian of a mechanical system can be written as G 11 1 L(g, s, g, s) = (g T , sT ) 2 GT


12

g G 12 V (g, s). G 22 s

Making use of (2.9) and (2.11), we see that g T G 11 g = (g, 0), (g, 0) = J(g, 0); Adg = I(g, 0); Adg = I Adg ; Adg = Ad I Adg ; g

sT G T g = (0, s), (g, 0) = J(0, s); Adg = I(0, s); Adg 12 = I Adg A(s)s; Adg = Ad I Adg A(s)s; . g Because of the G-invariance of L we can dene m(s) = G 22 (e, s) and V (s) = V (q). If

we also dene I(s) = Ad I Adg then the truth of the proposition follows. g

To apply Proposition 2.3.1 to our particular uid mechanical problem, I must show that the uid mechanical Lagrangian is invariant with respect to a group action. As seen in Section 2.1, the Lagrangian is a function of and based on the invariance of and . Hence, invariance is

. Note that the potential, , is dened with

respect to F B . Straightforward calculations based on this fact and an analysis of the boundary conditions can be used to prove the following fact. Proposition 2.3.2. In the case of quiescent ambient ow, the potential, , dened by Equation (2.3) and subject to boundary conditions (2.4) is SE(d)-invariant. Similarly, is SE(d)-invariant. This proposition cannot in general be extended to the case of a non-quiescent ambient ow. The following is a direct consequence of Proposition 2.3.2. Proposition 2.3.3. The uids kinetic energy, considered as the Lagrangian Tf luid : T Q

15 I and given by R Tf luid = 0 0 || ||2 dV = 2 D 2 ( n) dS,

is invariant with respect to a left SE(d)-action. As a corollary to Prop. 2.3.1 and Prop. 2.3.3, I can state that the kinetic energy of a deformable swimmer in an inviscid irrotational uid takes the following special form. Proposition 2.3.4. The kinetic energy of the deformable swimmer, Equation (2.5), assumes the reduced form: Id 1 Ttotal (s, s, ) = ( T , sT ) 2 A T Id
d

Id A d , md s

(2.13)

where Id is the 6 6 locked added inertia tensor, with entries: (Id )ij = gg (s) ij 0 2 (g ( g n) + g ( g n)) dS, i j j i (2.14)

where gg (s) is the locked inertia tensor of the deformable body system (considered in the absence of the uid) and the second term is the classical added uid mass/inertia. Meanwhile, Id Ad is a (6 nr ) matrix with entries: (Id Ad )ij = gs (s) ij 0 2 (g ( s n) + s ( g n)) dS. j j i i (2.15)

Hence, the local form of the uid mechanical connection, Ad , can be computed as
1 Ad (s) = Id (s)(Id Ad )(s).

(2.16)

When the symmetry principles are taken into account, the governing equations of motion that one derives from the Euler-Lagrange mechanical equations reduce to this

16 form:
1 g 1 g = Ad (s)s + Id (s)

(2.17) (2.18) (2.19)

= ad M (s) = T (s) B(s, s) C(s), s

where is the systems momentum, in body coordinates. The variable represents the shape forces. The rst equation is the connection, modied to include the possibility that the swimmer starts with nonzero momentum. The second equation describes the evolution of the momentum, as seen in body coordinates. In spatial coordinates, this momentum is constant since it is a conserved quantity. We will hereafter assume that the swimmer starts with zero momentum, thereby eliminating the second equation and simplifying the rst to the form of Equation (2.10). The third equation is known as the shape dynamics and is only a function of the shape varibles. For the purposes of control analysis and trajectory generation, I need only focus on the connection. So, the innitesimal relationship between shape changes and body velocity is described by the local form of the connection: g = gA(s)s = gAi (s)si , (2.20)

where the index i implies summation. I would like to nd a solution for this equation that will aid in designing or evaluating motions that arise from shape variations. Because SE(d) is a Lie group, the solution to Equation (2.20) will generally have the form g(t) = g(0)ez(t) , where z se(d). An expansion for the Lie algebra valued function z(t) has been given by Magnus [Mag54]. 1 1 1 z = A + [A, A] + [[A, A], A] + [A, [A, A]] + 2 3 12

17
t

A(t)

A( )s( )d,
0

where [, ] is the Jacobi-Lie bracket on g. To obtain useful results, examine the group displacement resulting from a periodic path : [0, T ] M such that (0) = (T ). Taylor expand Ai about (0) and then judiciously regroup, simplify, apply integration by parts, and use the fact that the path is cyclic [RB98]. 1 z() = Fij (0) dsi dsj 2 1 + (Fij,k [Ai , Fjk ])(0) 3 where Fij Aj,i Ai,j [Ai , Aj ] (2.22) (2.21) dsi dsj dsk + ,

is termed the curvature of the connection, the notation ,j indicates dierentiation with respect to sj , and the following shorthand notation is used
T

dsi dsj dsk

tk 0 0

tj

si (ti )dti sj (tj )dtj sk (tk )dtk .

Summation over indices is implied. The connection, A, and its curvature, F , are evaluated at (0) so that the coecients of the integrals are constants. For a complete discussion of this series, see [RB98]. Thus, our geometric analysis shows that for small boundary deformations, the displacement of the body over one period of shape deformation is proportional to the curvature of the connection. The curvature is an excellent measure of the eectiveness of the swimmer. This result further bolsters the central role of the uid mechanical connection in the analysis of deformable swimmers. For proportionally small deformations, the displacement experienced during one

18

Figure 2.2: The variables s1 (t), s2 (t), s3 (t) correspond to 3 deformation modes. The 1st and 2nd modes together yield motion in the x-direction. The 1st and 1st modes together yield motion in the y-direction. deformation cycle is
1 gdisp = ez() exp 2 Fij

dsi dsj .

(2.23)

The term 1 2

i,j

Fij

dsi dsj is Lie-algebra valued, and therefore it will take the

form of g 1 g in Equation (2.2). The exponential of such a matrix is e


0T

(I e )( ) + ( ) 1

(2.24)

2.4

Planar Amoeba Example

As an example, I consider the propulsive movements of a roughly circular device whose boundary shape is modulated by a small amount. Bearing in mind that a practical robot amoeba should have as few actuators as possible, I restrict the possible deformations of its boundary to a set parameterized by three variables s1 (t), s2 (t), s3 (t) as follows. Fix a frame in the body of the amoeba, and let the shape of the amoeba be described in polar coordinates in the body frame by the equation (see Figure 2.2): F (, s) = r0 [1 + (s1 cos(2) + s2 cos(3) + s3 sin(3))]. (Note that the centroid C of the deformed amoeba is not, in general, located at the origin of the body frame; I will return to this point later.) Under this time-varying

19 deformation the area A of the amoeba is


2 A(t) = r0 [1 +

1 2 (s1 (t)2 + s2 (t)2 + s3 (t)2 )]. 2

The perfectly irrotational uid surrounding the amoeba has density . The potential is determined by the surface boundary conditions, by the requirement that u =

go to zero as r approaches innity, and by the requirement that there be no circulation around the amoeba. The surface boundary condition in polar coordinates is
T

( n)| = x cos + y sin x sin + y cos

+ r0 s1 cos(2) + s2 cos(3) + s3 sin(3) 0 The unit normal to the surface at any point is 1
1 2 2 2 2

n.

n(s) =

( (s1 cos(2 ) + s2 cos(3) + s3 sin(3))

+ O( 3 ),

where = 3 s3 cos(3) + 2 s1 sin(2) + 3 s2 sin(3). Solving Laplaces equation by separation of variables, I nd = 1 x + 2 y + 3 + s s1 + s s2 + s s3 + O( 3 ), 1 2 3 where, using the notation c() = cos(), s() = sin(),
2 2 r0 r3 r4 r5 r0 c() + s1 c() + 0 s2 c(2) 0 s1 c(3) 0 s2 c(4) r r r2 r3 r4 3 5 2 r0 r0 r0 2 + 2 s3 s(2) 4 s3 s(4) + (3s2 + 5s2 + 5s2 )c() 2 3 r r 2r 1 r4 r5 r6 r3 2 0 s1 s2 c(2) + 03 s2 c(3) + 0 s1 s2 c(4) 05 (5s2 3s2 + 3s2 )c(5) 1 1 2 3 r2 4r r4 4r 7r8 2r3 r5 r7 3 0 s1 s2 c(6) 0 (s2 s2 )c(7) 20 s1 s3 s(2) + 0 s1 s3 s(4) 3 r6 4r7 2 r r4 6 7 8 3r0 3r0 7r0 + 5 s2 s3 s(5) 6 s1 s3 s(6) 7 s2 s3 s(7) 2r r 2r

1 =

20
3 r5 r2 r5 r3 r0 s3 c(2) + 0 s3 c(4) 0 s1 s() 0 s2 s(2) 0 s1 s(3) r2 r4 r r2 r3 3 r5 r5 3r6 3r7 2r0 0 s2 s(4) + 2 s1 s3 c(2) + 0 s1 s3 c(4) + 0 s2 s3 c(5) + 60 s1 s3 c(6) r4 r2 r4 2r5 r 2 4 8 3 r r 7r 2r + 0 s2 s3 c(7) 0 (3s2 + 5s2 + 5s2 )s() 20 s1 s2 s(2) 03 s2 s(3) 1 2 3 7 2r 2r r 4r 1 r6 3r7 7r8 r5 0 s1 s2 s(4) 05 (5s2 + 3s2 3s2 )s(5) 60 s1 s2 s(6) 0 (s2 s2 )s(7) 1 2 3 3 r4 4r r 4r7 2

2 =

2 r0 s() + r

3 =

5 r4 r5 r0 s3 c(3) 0 s1 s(2) 0 s2 s(3) r3 r2 r3 3 3r7 7r8 3r3 3r0 s1 s3 c() + 50 s1 s3 c(5) + 0 s2 s3 c(6) 0 s1 s2 s() + 2 r r 2r6 r 5r6 3r7 7r8 0 s2 s(4) 50 s1 s2 s(5) 0 (s2 s2 )s(6) 1 3 4r4 r 4r6 2

s = 1

4 r0 c(2) + 2r2

2 r0 5r6 3r7 3r7 s1 ln(r) 0 s1 c(4) 0 s2 c(5) 0 s3 s(5) 2 8r4 5r5 5r5

s = 2

5 r0 c(3) + 3r3

2 r0 r3 3r7 s2 ln(r) 0 s1 c() 0 s1 c(5) 2 r 5r5 8 7r8 7r0 06 s2 c(6) s3 s(6) 12r 12r6

s = 3

5 r0 s(3) + 3r3

2 8 r0 7r0 r3 s3 ln(r) + s3 c(6) 0 s1 s() 2 12r6 r 8 7r0 3r7 s2 s(6) 0 s1 s(5) 5r5 12r6

From I can readily nd the rightmost terms of equations (2.14) and (2.15); it remains to nd gg and gs . To compute gg and gs , I must make some assumptions regarding the amoebas internal structure. For this example I assume that the amoeba is homogeneous (which

21 implies that the center of mass is located at the centroid) and that it has the kinetic energy of an instantaneously rigid body of mass M with the same center-of-mass velocity and angular velocity. The velocity of the centroid is given by x C = + y s1 s2 + s1 s2 s1 s3 s1 s3 + s1 s3 + s1 s2

r0

+ O( 3 ),

while the moment of inertia around the centroid is M A(t)


2 0 0 r(t,)

= M

r cos() r sin()

2 2

r0 s1 (t)s2 (t)

r0 s1 (t)s3 (t)

1 4 r0 (1 + 3 2 (s2 + s2 + s2 )) + O( 3 ) 1 2 3 2 1 2 r0 (1 + 2 2 (s1 (t)2 + s2 (t)2 + s3 (t)2 ))

r dr d

2 = M r0

1 5 2 2 + (s1 + s2 + s2 ) + O( 3 ), 2 3 2 4
2

so the matrices gg and gs are given to order 1 0 0 1

by r0 s 1 s 3
2 2

gg

M gs

r0 s 1 s 2

Using formulas (2.14) and (2.15), to nd Id and Id Ad , I end with a local connection form Ad = (Id )1 (Id Ad ). r0 (1 )s2 r0 s 1 0
2 r0 2 s3 M

5 2 2 r0 s1 s3 2 r0 s1 s2 r0 ( 1 + 4 2 ( 3 s2 )) i=1 i 2 2 r0 s 2 2 r0 s 1 0 3 2 = M 2 r0 s 3 0 r0 s1 + O( ). 0 0 0

0 r0 s1
2 r 0 2 s2 M

Ad =

r0 (1 )s3 0

+ O( 3 ),

(2.25)

2 2 where = (2r0 )/(M + r0 ). After using the connection to derive the motion of

22
3 3 3 2 2 2

-3

-2

-1

3 -3

-2

-1

3 -3

-2

-1

-1

-1

-1

-2

-2

-2

-3

-3

-3

Figure 2.3: Computer animation of a planar amoeba whose density equals that of the surrounding uid. There are roughly 15.3 oscillations between snapshots. the frame, I may easily derive the velocity of the centroid: s1 s2 s1 s3
3 2r0

C=

r0

s1

s2 s2 3 s3 s2 2

s3 s2 s3 s2 s2 s3

+ O( 3 ).

As seen above, the abstract approach espoused here leads to a surprisingly succinct description of the essential governing equations.

2.4.1

Displacement by Periodic Motion

Example (continued): Consider the case of our idealized planar amoeba. Since the curvature F of the connection is a skew symmetric quantity, there are only three independent nonzero curvature terms, F12 , F13 , and F23 . Let us assume that the rst deformation mode is forced periodically by input s1 = cos(t), while the second mode is periodically forced by input s2 = sin(t). For the chosen input forcing functions, the integral terms associated with the F13 and F23 terms are zero. From Equations

23 (2.25) and (2.22) we see that 0 0 s2 2 M r0 = 2 r0 0 0 s3 + O( 3 ) 2 M + r0 0 0 0 2r0 s3 s1 0 M 2r0 3 2 = r0 M s 3 0 0 + O( ) 0 0 0 0 0 1 3 2r0 2 = 0 0 0 + O( 3 ). 2 M + r0 0 0 0

A1

(2.26)

A2

(2.27)

F12

(2.28)

Therefore, the displacement over one period of forcing is 1 F12 (0) 2


2 2

gdisp exp

s1 ( )d
0

s2 (t)dt

0 t

1 2 F21 (0)

s2 ( )d
0

s1 (t)dt (2.29)

= exp(F12 (0)).

If I discard the high-order terms in F12 and exponentiate only the curvature proportional to
2

, we nd gdisp 0 1 0 0 1 0
3 2 2 2 r0 2 M +r0

0 Thus, to O( 2 ), each oscillation results in a displacement of (2 2 r0 M +r2 ) along 0

(2.30)

r 2

the x-axis. The simplicity with which this result can be obtained is a direct result of the geometric approach outlined in Sections 2.2 and 2.3.

24

2.4.2

Rectilinear Motion Planning

The series expansion outlined in Section 2.4.1 can be used as a basis for developing motion planning algorithms, as it directly relates control inputs to net displacement. Interested readers are referred to [RB98] for more details. Here I take a simplied approach to our example. Consider the two-degree-of-freedom problem where we do not wish the amoeba to rotate, but wish the centroid to follow a path in the plane. Suppose that I choose sinusoidal inputs with time-varying amplitude, as follows: s1 (t) = cos(t), s2 (t) = a(t) sin(t), s3 (t) = b(t) sin(t). Then C=
2 2r0 2 r0 2 r0 + M

a(t) sin2 (t) b(t) sin (t)


2

+ O( 3 ).

Thus, moving the centroid of the amoeba along a given curve in the plane is remarkably easy. (By contrast, moving the body frame origin along a given curve, using this form of input, would be more complicated, since the velocity of the body frame origin depends on the derivatives a(t) and b(t) to leading order.) At any point along the curve, I make the velocity of the centroid tangent to the curve by appropriate choice of a and b (b/a is the slope of the curve). As long as the curve is smooth, a and b are smooth functions of time. This method steers the centroid along a path in the plane, but not necessarily at a desired speed at any given instant. In particular, the velocity of the centroid will periodically vanish (when sin2 (t) vanishes). Figure 2.3 shows snapshots of a computer simulation of this model as it tracks a straight line with unit slope. I must also note that the average velocity of the centroid is disappointingly low, on the order of
2

r0 (which is the norm of the connections curvature!). Each shape


2

change cycle moves the amoeba a distance on the order of

r0 . Thus if

= 0.1,

then 100 oscillations are required to move the amoeba one body radius. Mechanically feasible amoeba will be relatively slow swimmers.

25

2.4.3

Optimal Control Analysis for Amoeba

I now demonstrate that the sinusoidal inputs used in the last two sections are optimal inputs, according to one natural measure of performance. First I restrict the problem to motion along the x-axis. Therefore I have one base variable Cx and two shape variables, s1 and s2 (I can neglect s3 ). Assume that the control inputs are u1 = s1 and u2 = s2 , so Cx = performance index: J(0) = 1 2
T 2

r0 u1 s2 . Suppose I choose a minimum-control-eort

uT u dt =
0

1 2

T 0

(u2 + u2 ) dt 1 2

(2.31)

Assume that Cx (0) = 0 and for simplicity s2 (0) = 0. At time T I require a nal state s1 (T ) = s1 (0), s2 (T ) = s2 (0), and Cx (0) = d. The Hamiltonian and costate equations for this optimal control problem are 1 H = (u2 + u2 ) + 1 u1 + 2 u2 + 3 2 r0 u1 s2 2 2 1 (1 , 2 , 3 ) = (0, 3 2 r0 u1 , 0) From this we see that 1 and 3 are constants, while
T

2 (t) = 2 (T ) + 3 2 r0
t

u1 (t ) dt

The stationarity condition is u1 + 1 + 3 2 r0 s2 u2 + 2 0 = 0

26 A solution to these equations is s1 = a cos(N t/T ) s2 = a sin(N t/T ) N t sin((2N /T )t) Cx = 2 r0 a2 [ ] T 2 (4N /T ) where a = s1 (0) and N =
2d
2 a2 r 0

To summarize, optimal inputs for s1 (t) and s2 (t) are sinusoidal functions of time, 90 degrees out of phase. Optimal inputs are those which cause the amoeba to swim a given distance along the x-axis in a given time, while minimizing control eort as dened in (2.31). By symmetry, sinusoids 90 degrees out of phase in s1 (t) and s3 (t) will cause optimal motion along the y-axis. Further, by making s2 (t) and s3 (t) each oscillate 90 degrees out of phase with s1 (t), I can drive the amoeba in any direction in the plane, as seen in Section 2.4.2.

27

Chapter 3

Deformable Joukowski Foils

I now consider another deformable body: a somewhat shlike swimmer shaped like a Joukowski foil with changeable shape parameters. The boundary of a Joukowski foil is the image in the physical z-plane of a circle C, centered on the origin in the plane, under the transformation z = F () = + c + a2 . + c (3.1)

The size and shape of the foil are determined by the parameters a and c , and the radius of C is rc = a c . The real number a always has the opposite sign of the real part of c . The inverse of the Joukowski map is given by:1 = F 1 (z) =
1

1 z + z 2 4a2 c 2

(3.2)

The square root in Equation (3.2) is a multivalued function and it is important to choose the correct branch, i.e., assuming z lies on or outside the foil, the sign of the square root should be chosen so that rc . Streitlien [Str94] suggests the expanded form = 1 z + z 2a z + 2a c .This 2 will yield the correct answer for most points in the z-plane but can still give the wrong branch for points close to the concave surface of a cambered foil.
1.5 1.5

0.5

0.5

0.5

0.5

1.5 2.5

1.5

0.5

0.5

1.5

2.5

1.5 2.5

1.5

0.5

0.5

1.5

2.5

Figure 3.1: The Joukowski map z = + c + a2 /( + c ) sends a circle to an airfoil contour. Here a = 1, rc = 1.1, and c = (0.0781 + 0.2185i).

28 The area of the foil is given by A = 1 2i 1 = 2i 1 = 2i z dz

a2 a2 1 d ( + c )2 + c C 2 a2 a2 rc 1 d + c + 2 ( + c )2 rc + c C 2 2 1 rc a2 a2 c a4 a2 rc = + c + 2 2 2i C rc + c ( + c )2 ( + c )2 (rc + c )( + c )2 a4 a4 r 2 2 2 = rc 2 c 2 2 = rc 1 2 (rc ) (rc 2 )2 + c +

2 where = c and where I have taken advantage of the fact that = rc / on the

circle C. The geometric center zc of the foil is given by [Str94]: zc A = 1 4i 1 4 z 2 dz

(3.3) a2 + c
2

+ c +

2 = rc c +

a6 c 2 (rc 2 )3

rc 2 2

+ 2

a2 rc 2
rc 2

+ c

(3.4)

(3.5)

2 a6 c + c (rc 2 )3 zc = 2 2 (rc 2 ) ((rc 2 )2 a4 )

(3.6)

3.1

Allowed Deformations of the Foil

Suppose the Joukowski foil is translating with velocities U and V in the x and y directions and rotating about the origin at rate , and furthermore the foil shape parameter c changes at a rate c = x + iy . Since c is varying in time, either rc or a or both must also be time-varying so the relation rc = a c is satised. I can

29 write:
2 2 rc = (a x )2 + y 1 rc = (a x )a (a x )x + y y rc

(3.7) (3.8)

I will choose to regard x , y , and a as the controllable shape parameters, and treat rc as a dependent quantity, although in principle any three of the four variables could be taken as independent. The velocity of the center of area zc is zc = U + iV + izc + rc 2 c c

a6 a4 + rc 2 c c
4 2

a6 c c + c rc 2 c c
2 2 2

2 a 6 c c + c + rc 2 c c
6

a 4 rc 2 c c

rc 2 c 2 c + 3 c i a6

a4 rc 2 c c + i a6 c c c rc 2 c c
2

a4 rc 2 c c

x
2

a4 rc 2 c c
4 2

+ rc 2 c c
2 2 2

a4 rc 2 c c
2

i 2a

c c c + a

rc + c 2 c 3 c a4 rc 2 c c
2

rc c c rc c c

3 a c + 2 c rc c c 2 a rc a c rc c c + 2 2 2 rc 2 c c a4 rc 2 c c 6 a5 c 4a
3 6 2

rc 2 c c

a4 + rc 2 c c

a c + c rc c c

rc
3

rc 2 c c

a4 + rc 2 c c

a 2 2

(3.9)

30 The acceleration of the center of area zc is zc = U + iV + izc + izc zc zc zc zc x + y + a+ rc x y a rc 2 zc 2 2 zc 2 2 zc 2 2 zc 2 + + r a + + 2 x 2 y 2 c x y a2 rc + +2 2 zc 2 zc 2 zc 2 zc 2 zc 2 zc x a + 2 y a + 2 arc x y + 2 x rc + 2 y rc + 2 x y x a y a x rc y rc arc (3.10) A closed-form expression can be obtained by taking the appropriate partial derivatives of Equation (3.6). The derivatives with respect to x and y are facilitated by considering that if c and c are regarded as independent variables, then: = + x c c = i i y c c (3.11) (3.12)

For the time being I will not place any other constraints on what deformations of the foil are allowed. But if the deformations were required to preserve the area of the
2 2 2 2 foil, for example, then since the area A = rc (1 a4 /(rc x y )2 ), I would require 2 2 2 (rc 2 )3 + a4 (rc + 2 ) rc 2a3 rc (rc 2 )a 2a4 rc x x 2a4 rc y y = 0

(3.13)

which, combined with Equation (3.8), yields


2 2 2 2 a (a x ) (rc 2 )3 + a4 (rc + 2 ) 2a3 rc (rc 2 ) 2 2 2 x (a x )(rc 2 )3 + a5 (rc + 2 ) + a4 x (rc 2 ) 2 2 + y (rc 2 ) (rc 2 )2 a4 y = 0 (3.14)

31 If the area A is held constant then the velocity of the center of area simplies to

zc = U + iV + izc rc 2 + +

rc 2 c c rc 2 c c

+ a 6 rc 2 + c 2 c + 3 c
4

A rc 2 c c
4

i rc 2

a 6 rc 2 + 2 c 3 c c
2 4 2 2 4

x y

A rc c c
2

2 r c rc 8 c 4 r c 6 c 2 c a 6 c c + 6 r c 4 c 3 c + c 5 c 2 r c 2 c a6 + 2 c 4 c A rc c c
4

6 a5 r c 2 c A rc 2 c c
3

rc

a (3.15)

A requirement of constant area is not the only reasonable choice of constraint. One alternative, non-area-preserving approach would be to allow c to vary while keeping rc and a xed. This would be equivalent to varying the camber of the foil while keeping its thickness and chord length approximately constant.

3.2

The Potential Function

Suppose the uid around the foil is at rest at innity, and allow for the presence of a central vortex with strength c at the origin and a certain number of free vortices (k = 1, . . . , nv ) with strengths k at locations zk = F (k ) in the ow. Then the uid ow in the z plane is described by the complex potential w = + i made up of generalized Kircho potentials:
s s k s k w5 (; k )+x w1 ()+y w2 ()+aw3 () k

w = U w1 ()+V w2 ()+w3 ()+c w4 ()+

(3.16)
k where w5 is the portion of the potential associated with the kth wake vortex, and s s s w1 , w2 , w3 are the parts of the potential associated with deformations of the shape of

32 the foil. First I nd w1 , w2 , w3 by considering motion of the rigid foil in the absence of deformation. As in Equation (2.4), at the surface of the foil undergoing rigid motion, the ow must match the boundary condition: ( n) = dx dy dx dy | = ( + )| = (V + x) + (U y) x ds y ds ds ds (3.17)

where I use the same notation as above: n is a unit vector normal to the foil boundary, and is a coordinate parameterizing the surface. I use the subscript the surface . From this one can conclude that | = V x x | = U y y (3.18) (3.19)

to mean at

So a sucient condition to satisfy the boundary condition at the surface of the foil is Im {w} = | =

1 U y V x (x2 + y 2 ) 2 i = Im (U iV )z zz 2

(3.20)

By substituting Equation (3.16) into the left-hand side of Equation (3.20), we see that the boundary condition for the w1 component of the potential is Im {w1 } = Im {z}

a2 + c C r2 a2 = Im c + c + + c C 2 a2 r = Im c + c + + c C = Im + c +

(3.21)

33
2 (recalling that = rc / on C) and so an acceptable solution for w1 is: 2 a2 rc w1 = + c + + c

(3.22)

Note that this solution for w1 satises not only the boundary condition at the surface of the foil, but also the requirement that the uid be at rest at innity. The quantity + c +
a2 , +c

for example, would have satised the boundary condition at the foil,

but not at innitythis is why the modication of Equation (3.21) was necessary. Similarly: Im {w2 } = Im {iz}
2 rc

= Im i ic i r2 = Im i c and an acceptable solution for w2 is w2 = i = Im i

a2 + c a2 ic i + c a2 ic i + c

(3.23)
C

2 rc a2 ic i + c

(3.24)

34 Meanwhile: Im {w3 } = Im = Im = Im = Im = Im = Im = Im i zz 2 a2 a2 i ( + c + )( + c + ) 2 + c + c C i + c + c a4 + c + c + 2 + a2 ( + )+ 2 + c + c ( + c )( + c ) + c i a + c + c + 2 + 2a2 ( )+ 2 + c ( + c )( + c ) i 2 a4 2 2 + c rc + 2c + + 2a ( )+ 2 + c ( + c )( + c )


C 4 C

i 2 r2 / + c r2 a4 )+ rc + 2c c + 2 + 2a2 ( c 2 2 + c ( + c )(rc / + c ) r2 / + c r2 a4 i 2 )+ rc + 2c c + 2 + 2a2 ( c 2 2 + c c ( + c )( + rc /c )

(3.25)
C

The expression inside the curly brackets satises the boundary conditions at the foil
2 and at innity, but the last term has a singularity at = rc /c , which is outside

the boundary of C. To cancel this singularity, I add another singularity of opposite


2 strength at rc /c , and use the Milne-Thomson circle theorem [MT68] to preserve

the surface boundary condition. The term to be added is


2 2 i a4 rc a4 rc i 2 2 2 2 2 2 c (rc 2 )( + rc /c ) 2 c (rc 2 )(rc / + rc /c )

By inspection this corrective term is purely real on the circle C, so: Im {w3 }
2 2 rc a4 i 2 2 2 rc / + c = Im rc + 2c + + 2a ( )+ 2 2 + c c ( + c )( + rc /c ) 2 2 a4 rc a4 rc + 2 2 2 2 2 c (rc 2 )( + rc /c ) c (rc 2 )(rc / + rc /c ) C

= Im

r2 r2 / + c a4 ( c ) i 2 rc + 2c c + 2 + 2a2 ( c )+ 2 2 + c ( + c )(rc 2 )

(3.26)

The quantity inside the curly brackets now has no singularities outside the circle

35 C, and still approaches a constant at innity. So we have a solution for w3 which satises all boundary conditions:2 w3 = r2 / + c r2 a4 ( c ) i 2 rc + 2c c + 2 + 2a2 ( c )+ . 2 2 + c ( + c )(rc 2 ) (3.27)

The potential components due to vortices are straightforward: w4 = i log( ) rc


k w5 = i log

(3.28)

rc k 2 k rc /k

(3.29)

k Notice that w5 , the unit potential for a free vortex located at zk = F (k ), represents k both the vortex itself and its image at the inverse point. The w5 terms do not create

any net circulation around a large contour encircling the foil and all vortices; any such circulation is presumed to be folded into the w4 central vortex term. To nd the potential component associated with changes of the foil shape parameters x , recall Equation (2.4) and observe that the boundary condition at the surface of the deforming foil is
s dw1 d d

Re i

= Re
C

F (; c ) (i dz) c

(3.30)

a2 a2 1 d = Re i 1 ( + c )2 ( + c )2 C a2 a2 = Re i 1 2 d 1 ( + c )2 (rc / + c )2
2

Streitlien [Str94, ST95] prescribes this same method of deriving the potential but, apparently in error, arrives at w3 = i
4 2 a 4 c r2 / + c a4 + r c 4 c r c + + a2 c + . 2 2 + c ( + c )(rc + 2 ) 2(rc 2 )

This is not equivalent to the correct answer and does not satisfy the boundary condition at the surface of the foil.

36 Integrating both sides with respect to , we nd


s Re {iw1 }C

= Re (i)

a2 c
2

2 2 4 4 a2 rc (rc + 2rc c c + (a2 c )c ) 2 2 c (rc + c )(rc c c )2 2 3

2 2 2 4 a2 (rc + 2rc c c + c (a2 c )) 2a4 rc c log(( + c )/rc ) 2 2 ( + c )(rc c c )2 (rc c c )3 2 6 4 2 2 2 +2a2 rc (rc 3rc c c + 3rc c c + c (a2 c )c ) 2 3 3

(3.31)

2 2 log((rc + c )/rc ) 2 c (rc c c )3 C

where I have chosen the constant of integration in order to make the arguments of the logarithms dimensionless.
s Im {w1 }C

= Im

2 + 3 2 2 c c (rc + c )(rc 2 )2 4 2 2 2 a2 (rc 2rc 2 c a2 + 4 ) 2a4 rc c log(( + c )/rc ) + 2 2 ( + c )(rc 2 )2 (rc 2 )3 2 6 4 2 +2a2 rc (rc 3rc 2 + 3rc 4 + c a2 c 6 ) 3 3

a2

2 4 4 a2 rc (rc 2rc 2 a2 c + 4 )

2 2 log((rc + c )/rc ) 2 c (rc 2 )3

(3.32)
C

s Im {w1 }C

2 a2 r2 a2 r4 ((r2 2 )2 a2 c ) rc = Im 2 c + 3c c 2 2 c c (rc + c )(rc 2 )2 2 2 2 a2 ((rc 2 )2 c a2 ) 2a4 rc c log(( + c )/rc ) + 2 2 ( + c )(rc 2 )2 (rc 2 )3 2 2 +2a2 rc ((rc 2 )3 + c a2 c ) 3 2 2 log((rc + c )/rc ) 2 c (rc 2 )3 3

(3.33)
C

s Im {w1 }C = Im

2 4 4 2 a2 rc a2 rc a4 rc rc + + 3 2 2 2 2 c c (rc + c ) c (rc + c )(rc 2 )2 2 2 a4 c 2a4 rc c log(( + c )/rc ) a2 + 2 2 ( + c ) ( + c )(rc 2 )2 (rc 2 )3 2 +2a2 rc

c a2 2 (rc 2 )3

2 2 log((rc + c )/rc ) C

(3.34)

37
s The quantity inside the curly brackets is prevented from being a valid solution for w 1 2 by the singular behavior at = rc /c . As before, I will construct a corrective term

which removes the singularity while preserving the other boundary conditions via the Milne-Thomson circle theorem. The corrective term is
4 a2 rc 4 a4 rc 2 c (rc 2 )2

1 2 2a2 rc 2 + rc c

1 c
3

4 4 a4 rc a2 rc 3 2 c c (rc 2 )2

1 2 2a2 rc 2 + (r 2 /) rc c c

1 c a2 + 2 3 c (rc 2 )3

c a2 + 2 (rc 2 )3

2 2 log((rc + c )/rc )

2 2 2 log((rc +(rc /)c )/rc )

Since this expression is purely real on C, it follows that


s Im {w1 }C = Im 2 2 2 a2 rc a4 c rc a2 + + 2 2 c ( + c ) ( + c )(rc 2 )2 2 4 4 2a4 rc c log(( + c )/rc ) a4 rc a2 rc 1 2 2 )3 3 2 2 )2 2 + (r 2 /) (rc c c (rc rc c c 2 1 c a 2 2 2 2 2a2 rc log((rc + (rc /)c )/rc ) (3.35) + 2 3 2 )3 c (rc C

s Im {w1 }C = Im

2 2 2 rc a2 a2 rc a4 c + + 2 2 c ( + c ) ( + c )(rc 2 )2 2 2 a4 rc 2a4 r2 c a2 rc 2 c 2 3 log(( + c )/rc ) 3 2 (rc ) c c (rc 2 )2 1 c a2 + c 2 2a2 rc + 2 ) log( 3 2 )3 c (rc C

( + c ) (3.36)

s If I were now to use the quantity inside the curly brackets as a trial solution for w 1 ,

we would nd that it resulted in a nonzero amount of circulation around a closed contour surrounding the foil, which would vary with x . But Kelvins circulation theorem states that the circulation around a closed material curve lying entirely in the uid is an invariant of the motion, even if the curve is non-reducible [Saf92]. To satisfy this theorem and avoid producing changing circulation around a curve

38 enclosing both the foil and any shed vortices, I add a central vortex term:
2 2a4 rc (iy ) log( ) 2 2 )3 (rc rc

This is plainly real on C so that


s Im {w1 }C = Im 2 2 2 a2 a2 rc a4 c rc + + 2 2 c ( + c ) ( + c )(rc 2 )2 2 2 2a4 r2 c a4 rc a2 rc 2 c 2 3 log(( + c )/rc ) 3 2 2 )2 (rc ) c c (rc ( + c ) 2 4 2 c a 1 2a r (iy ) + c 2 2a2 rc ) + 2 c 2 3 log( ) (3.37) log( + 2 3 2 )3 c (rc (rc ) rc C

s So a valid solution for w1 is: 2 2 2 rc a2 rc a4 c a2 + + 2 2 c ( + c ) ( + c )(rc 2 )2 2 2 a2 rc + c a4 rc 2a4 r2 c ) 2 c 2 3 log( 3 2 2 )2 (rc ) rc c c (rc ( + c ) 1 + c 2a4 r2 (iy ) c a2 2 + 2 log( 2a2 rc ) + 2 c 2 3 log( ) 2 )3 3 c (rc (rc ) rc

s w1 =

(3.38)

By a similar development we nd that


s Im {w2 }C = Im (i) 2 2 4 4 rc a2 rc a2 rc a4 rc + 3 2 2 2 2 c c (rc + c ) c (rc + c )(rc 2 )2 2 2 2a4 rc c log(( + c )/rc ) a2 a4 c + 2 2 ( + c ) ( + c )(rc 2 )2 (rc 2 )3 2 +2a2 rc

c a2 2 (rc 2 )3

2 2 log((rc + c )/rc ) C

(3.39)

To which I add this corrective term, real on C, to remove singular behavior outside

39 the foil boundary:


4 a2 rc 4 a4 rc 2 c (rc 2 )2

i 2 + 2ia2 rc 2 + rc c

1 c
3

4 a2 rc 3 c

4 a4 rc 2 c (rc 2 )2

i 2 2ia2 rc 2 2 rc + (rc /)c

1 c a2 + 2 3 c (rc 2 )3

c a2 2 (rc 2 )3

2 2 log((rc + c )/rc )

2 2 2 log((rc +(rc /)c )/rc )

And another term, also real on C, to cancel any circulation produced around the foil, (i) to nd
s Im {w2 }C = Im (i) 2 2 2 rc a2 rc a4 c a2 + 2 2 c ( + c ) ( + c )(rc 2 )2 2 2 2 a2 rc 2a4 rc c log(( + c )/rc ) a4 rc + 2 2 )3 3 2 2 )2 (rc c c (rc ( + c ) + c 2a4 r2 x c a2 1 2 log( ) + 2 c 2 3 log( ) (3.40) +2a2 rc + 2 3 2 )3 c (rc (rc ) rc C 2 2a4 rc x log( ) 2 2 )3 (rc rc

Thus:
s w2 2 2 2 a2 rc a2 a4 c rc + = (i) 2 2 c ( + c ) ( + c )(rc 2 )2 2 2 2 a2 rc 2a4 rc c log(( + c )/rc ) a4 rc + 2 2 )3 3 2 2 )2 (rc c c (rc ( + c ) 4 2 2 1 + c 2a r x c a 2 + 2 log( ) + 2 c 2 3 log( ) +2a2 rc 3 c (rc 2 )3 (rc ) rc

(3.41)

Finally,
s dw3 d d

Re i

= Re
C

F (; c ) (i dz) a

= Re i = Re i

a2 2a 1 d ( + c )2 + c C a2 2a 1 d 2 ( + c )2 rc / + c

(3.42)
C

40 Integrating both sides with respect to , we nd


s Re {iw3 }C

= Re (i)2a

2 2 2 2 rc ((rc 2 )2 a2 c ) log((rc + c )/rc ) 2 c (rc 2 )2 2

2 a2 c a2 rc log(( + c )/rc ) 2 2 (rc 2 )2 c ( + c )(rc 2 ) 2

(3.43)
C

s Im {w3 }C = Im 2a

2 2 rc ((rc 2 )2 a2 c ) 2 c (rc 2

2 a2 rc log(( + c )/rc ) a2 c 2 2 (rc 2 )2 c ( + c )(rc 2 ) 2

2 )2

2 2 log((rc + c )/rc ) C

(3.44)

s Im {w3 }C = Im 2a

2 2 rc ((rc 2 )2 a2 c ) 2 c (rc 2

2 2 rc a2 c a2 rc log(( + c )/rc ) 2 2 c ( + c )(rc 2 ) (rc 2 )2 2

2 )2

2 2 log((rc + c )/rc ) C

(3.45)

2 The corrective term to remove the logarithmic singularity at = rc /c is 2 2 rc ((rc 2 )2 a2 c ) 2 c (rc 2 )2 2 2 2 2 2 rc ((rc 2 )2 a2 c ) 2 2 2 log((rc + (rc /)c )/rc ) 2 2 c (rc 2 )2

2 2 log((rc + c )/rc ) +

So we nd
s Im {w3 }C = Im 2a 2 2 a2 rc log(( + c )/rc ) rc a2 c 2 2 c ( + c )(rc 2 ) (rc 2 )2 2 ( + c ) r2 ((r2 2 )2 a2 c ) log( ) (3.46) + c c2 2 c (rc 2 )2 C

s And a valid solution for w3 is 2 2 rc a2 rc log(( + c )/rc ) a2 c 2 2 c ( + c )(rc 2 ) (rc 2 )2 2 r2 ((r2 2 )2 a2 c ) ( + c ) + c c2 2 log( ) 2 )2 c (rc

s w3 = 2a

(3.47)

41

U = 1 1.5

0.5

0.5

1.5 2.5 1.5

1.5

0.5

0 V=1

0.5

1.5

2.5

0.5

0.5

1.5 2.5 1.5

1.5

0.5

0 =1

0.5

1.5

2.5

0.5

0.5

1.5 2.5

1.5

0.5

0.5

1.5

2.5

Figure 3.2: Fluid velocity resulting from the unit potentials w1 , w2 , and w3 (top to bottom.) Here a = 1, rc = 1.1, and c = (0.0781 + 0.2185i). Note that I took U to be negative in the rst gure. Negative U , or velocity in the negative x-direction, corresponds to forward motion when x is negative, as here.

42

x = 1 1.5

0.5

0.5

1.5 2.5

1.5

0.5

0 y = 1

0.5

1.5

2.5

1.5

0.5

0.5

1.5 2.5 1.5

1.5

0.5

0 a = 1

0.5

1.5

2.5

0.5

0.5

1.5 2.5

1.5

0.5

0.5

1.5

2.5

s Figure 3.3: Fluid velocity resulting from the shape deformation unit potentials w 1 , s s w2 , and w3 (top to bottom.) Here a = 1, rc = 1.1, and c = (0.0781 + 0.2185i).

43

w = U w1 () + V w2 () + w3 () + c w4 () +
k

k s s k w5 (; k ) + x w1 () + y w2 () + aw3 () s

w1 = w2 = w3 = w4 =
k w5 = s w1 =

s w2 =

s w3 =

a + c 2 a2 r i c ic i + c 2 2 i 2 rc a4 ( c ) 2 2 rc / + c rc + 2c + + 2a ( )+ 2 2 + c ( + c )(rc 2 ) i log( ) rc rc k i log 2 k rc /k 2 2 a4 c a2 r 2 a2 rc + c + 2 2 c ( + c ) ( + c )(rc 2 )2 2 2 a2 rc 2a4 r2 c + c a4 rc 2 c 2 3 log( ) 3 2 2 )2 (rc ) rc c c (rc ( + c ) 2 4 2 1 c a 2a r (iy ) + c 2 2a2 rc + 2 ) + 2 c 2 3 log( ) log( 3 c (rc 2 )3 (rc ) rc 2 2 2 4 2 2 a rc a a c r + (i) c 2 2 c ( + c ) ( + c )(rc 2 )2 2 2 2a4 r2 c + c a4 rc a2 rc 2 c 2 3 log( )+ 3 2 2 )2 (rc ) rc c c (rc ( + c ) 4 2 2 + c 2a r x c a 1 2 log( ) + 2 c 2 3 log( ) +2a2 rc + 2 3 c (rc 2 )3 (rc ) rc 2 2 2 2 r a c ar + c 2a c 2 c 2 2 log( ) 2 2) c ( + c )(rc (rc ) rc 2 ( + c ) r2 ((r2 2 )2 a2 c ) log( ) + c c2 2 2 )2 c (rc + c +

2 rc

Table 3.1: Solution for the deformable Joukowski foil potential function.

44

dw1 d dw2 d dw3 d dw4 d k dw5 d s dw1 d

= = = = = =

s dw2 = d

s dw3 = d

2 rc a2 2 ( + c )2 a2 r2 i c +i 2 ( + c )2 2 2 2 c rc a2 rc a2 c a4 c a2 rc + 2 + i + 2 2 ( + c ) ( + c )2 ( + c )2 ( + c )2 (rc 2 ) i 1 1 i 2 k rc /k 2 2 1 a2 r 2 a2 a4 c 2a4 r2 c rc 2a4 r2 iy 1 2 c + 2 c2 3 + 2 c 2 3 2 2 2 c ( + c )2 ( + c )2 (rc 2 )2 (rc ) + c (rc ) 2 2 a2 rc c c c a2 1 a4 rc 2 + 2a2 rc + 2 3 2 2 )2 2 3 2 )3 c c (rc ( + c ) c (rc ( + c )() 2 2 2 4 2 4 2 2 r ar a c 2a r c 2a4 r2 x 1 a 1 (i) c + 2 c + 2 c2 3 + 2 c2 3 2 2 2 c ( + c )2 ( + c )2 (rc 2 )2 (rc ) + c (rc ) 2 2 c a2 a4 rc c c a2 rc 1 2 2 + 2a rc + 2 3 2 2 )2 2 3 2 )3 c c (rc ( + c ) c (rc ( + c )() 2 2 2 2 2 2 2 2 2 a c a rc r ((r ) a2 c ) r c 2a 2c + c c2 2 2 (r 2 2 ) 2 2 )2 2 )2 c ( + c ) c ( + c )(rc c (rc ( + c )

Table 3.2: Derivatives with respect to of the deformable Joukowski foil potentials.

45

3.3

The Locked Added Inertia

The virtual inertia tensor of a rigid Joukowski foil ought to be well known. However, while the necessary integrals are conceptually straightforward, they are somewhat tedious, and perhaps for this reason, the existing literature [Str94, ST95, Sed65] is plagued with typographical and other elementary errors. So as I develop the equations of motion for a deformable foil, I will also have the incidental goal of producing an error-free treatment of the rigid foils added inertia. As I said in Proposition (2.3.4), the kinetic energy of the system is 1 T = q T q + Tf 2 (3.48)

where the rst term is the kinetic energy of the foil itself, and Tf is the kinetic energy of the uid: Tf = Since = Re{w} ( n) d = Re i dw dz dz it follows that Tf = = = = = = Re{w}Re i dw dz dz 2 Re{w}Re i dw d d 2 C Re C wRe i dw d d 2 Re C w(i dw d + i dw d) d d 4 Re i C w dw d + i C w dw d d d 4 Re i C w dw d i C w dw d d d 4 (3.50) (3.51) 2

( n) d

(3.49)

(3.52)

46 Considering the foil to be rigid, and disregarding any eects of vorticity, we see that Tf will be a quadratic function of body velocity . U I V

Tf =

1 U V 2

(3.53)

where I is the locked added inertia matrix and its components are I11 = Re i C w1 dw1 d i C w1 dw1 d d d 2 4 2 r2 2 r c a2 a4 c = Re i C r2 + a c 2 + (+ )2 (+ )(+ )2 d c c c (+c ) 2 2 2 2 a2 r c r c a2 rc 1 4 Re (i) rc (2i) r2 + r2 (2i) + r2 (2i) + (2i)a4 (r2 2 )2 = c c c c 2 r2 2 (3.54) = rc 2a2 + a4 2 c 2 2 (rc )

I12 =

Re i C w1 dw2 + w2 dw1 d i C w1 dw2 + w2 dw1 d d d d d 4 2 a2 2 a2 2irc = Re (i) C (+ )2 2irc 2 d c (+c ) 4 2 2 Re (i) 2irc a2 ( 2i ) 2irc a2 ( 2i ) = 2 2 rc rc 4 = 0

(3.55)

I22 =

Re i C w2 dw2 d i C w2 dw2 d d d 2 4 2 a2 2 r2 rc a4 = Re i C rc2 + (+ )2 + a c 2 + (+ )(+ )2 d c c c (+c ) 4 2 2 = Re 2rc + 2a2 + 2a2 + 2a4 r2rc 2 c 2 r2 2 = rc + 2a2 + a4 2 c 2 2 (rc )

(3.56)

47

I13 =

Re i 4 = Re 4 +
C

w3 dw1 d i d (

w1 dw3 d d

2 2 a2 c rc a2 rc a2 c a4 ( c ) r2 ) d + + + )( c 2 ( + c ) ( + c ) 2( + c )(rc 2 ) 2 ( + c )2

2 2 2 a2 a2 c a4 c a2 rc a2 rc c r 2 rc )( 2c + 2 + ) d + 2 ( + c ) C ( + c ) ( + c )2 ( + c )2 ( + c )2 (rc 2 ) a4 2 a4 c a6 r 2 c 2 = Re 2i rc c 2 + a2 (c c ) + + 2 c 2 3 2 4 rc 2 c (rc 2 ) (rc ) 2 a4 r 2 c a4 r 2 c a4 c a6 c rc a4 c 2 2 c 2 2 a 2 c + 2 c 2 2 + 2 2 rc c + a2 c 2 rc 2 (rc ) (rc ) rc 2 (rc 2 )3 2 2a4 (c c ) a6 rc 2 2 2a (c c ) + 2 (c c ) Re i rc (c c ) = 2 2 rc 2 (rc 2 )3 a6 r 2 c 2a4 c 2 2a2 c + 2 c 2 3 (3.57) = Re i rc c 2 rc 2 (rc )

4 a4 a4 rc a4 r 2 c a6 r 2 c a4 c 2 Re 2i rc c + a2 c 2 c a2 + 2 c 2 2+ 2 c 2 3 2 4 (rc 2 ) c c (rc 2 )2 (rc ) (rc )

2 2 2 a2 a2 rc rc a2 rc a2 c a4 c c r 2 )( 2c + 2 + + ) d 2 ( + c ) ( + c ) ( + c )2 ( + c )2 ( + c )2 (rc 2 )

where in the nal step I have made use of the fact that Re {ic } = Re ic .

48

I23 =

Re i C w3 dw2 d i C w2 dw3 d d d 4 2 2 c rc a2 rc a2 c a4 ( c ) = Re i + + + 2 4 ( + c ) ( + c ) 2( + c )(rc 2 ) C +i


C

2 rc

a2 ( + c )2

4 2 a4 a4 rc a4 c a6 r 2 c a4 c rc 2 Re 2 c rc + c a2 + + a 2 c + 2 2 2 c 2 3 2 4 c c (rc 2 )2 (rc 2 )2 (rc 2 ) (rc ) 2 2 2 a4 c rc a4 c rc a4 c a6 c rc a4 c 2 2 + a 2 c + 2 2 2 +rc c + a2 c 2 rc 2 (rc 2 )2 (rc 2 )2 rc 2 (rc 2 )3 2 2a4 (c + c ) a6 rc (c + c ) 2 = Re (rc + 2a2 )(c + c ) 2 2 2 rc 2 (rc 2 )3 4 6 2 2a c a r c 2 = Re rc c + 2a2 c 2 2 c 2 3 (3.58) 2 rc (rc )

2 2 2 a2 a2 rc rc a2 rc a2 c a4 c c r 2 + )( 2c + 2 + + ) d 2 ( + c ) ( + c ) ( + c )2 ( + c )2 ( + c )2 (rc 2 )

These expressions for the locked inertia terms are in agreement with the results of Streitlien [Str94, ST95] and with those of Sedov [Sed65], after correcting some typographical and other elementary errors.3
3

In Sedov [Sed65], his expression xy for the arbitrary Joukowski foil in Table 1 should read: xy = a2 sin(2) 2

The expression in the same table should read: = a4 2 2 r R 8r2 R2 cos4 () 2rR sin2 (2) + cos(4) 8

The transformation formula in Sedovs equation (4.11) should read in part: y = (x xy + x ) sin() + (xy y + y ) cos()

In Streitlien, equation (41) in [ST95] and equation (49) in [Str94] should read: w3 z dz
S

2 2 a4 + a2 c + a4

2 4 4 4 4 4 4 4 2rc 2 2a8 r2 2 c 2 a + rc 2 a + rc + a6 2 + 2 c 2 4 rc + a 4 rc 2 2 2 (rc 2 )2 (rc 2 )2 (rc ) 2(rc 2 ) 2(rc 2 )3

Finally, the pure-rotational added inertia, m66 in Streitliens notation, should consist not only of the real part of the above integral, but also of an added term equal to twice the real part of the foils polar moment of inertia. So Streitliens equation (48) in [ST95] and equation (59) in [Str94] should

49

I33 =

Re i 2 = Re i 2
2 c rc 2

w3 dw3 d i d

w3 dw3 d d

2 2 a2 rc a2 c a4 ( c ) c rc + + + 2 ( + c ) ( + c ) 2( + c )(rc 2 ) 2 a2 rc

( + c )

2 2 rc 2 c 2 2 2 2 2 4 4 6 4 Re 2 rc + a c a 2 +a 2 a 2 +a 2 = 2 rc 2 rc 2 rc 2 (rc 2 )2 r4 r2 2 r2 2 +a4 2 c 2 2 a4 2 c 2 2 + a6 2 c c 2 3 (rc ) (rc ) (rc ) 4 r 2 r2 r2 2 2 +a2 c + a4 a4 2 c 2 2 + a4 2 c 2 2 a6 2 c c 2 3 (rc ) (rc ) (rc ) 2 2 2 4 2 a4 c a rc a c a ( c ) c rc i d + + + 2 2 ( + c ) ( + c ) 2( + c )(rc 2 ) ( + c )2 (rc 2 ) C 2 2 2 2 2 = Re 2 rc 2 + a2 (c + c ) + a4 (2 2 ) + a6 2 c 2 2 2 (rc 2 ) (rc ) 4 rc c r 2 c r2 2 c 2 a6 + a6 a6 2 c 2 3 + a8 2 c 2 4 a 2 2 2 (rc 2 ) c (rc 2 ) c (rc 2 )3 (rc ) (rc ) 4 2 2

2 a2 c a4 c a2 rc + 2 ( + c )2 ( + c )2 ( + c )2 (rc 2 )

2 2 = Re 2a4 + rc 2 + a2 (c + c ) 2a4 2

+a

2 c r 4 c r 2 c c 2 2 + 2 2c 2c 2 3 2 (rc 2 )2 (rc 2 ) (rc 2 )3 (rc ) 4 2 rc 2

2 2 (rc 2 )
2

+ a8
2

2 rc 2 2 (rc 2 )4

= Re 2a +

+a

2 (c

2 + c r2 2 2 + c ) 2a 2 + a6 c + a8 2 c 2 4 2 (rc 2 ) (rc 2 )2 (rc )


2 4

2 2 = Re 2a4 + rc 2 + 2a2 c 2a4

2 2 r2 2 + 2a6 2 c 2 2 + a8 2 c 2 4 2 (rc 2 ) (rc ) (rc )

(3.59)

In Figures 3.43.5, I plot the added mass components for a variety of foil shapes
read:
2 m66 = 2Re a4 + a2 c + a4 4 4 4 2 c 2a8 r2 2 4 2rc 2 2 a + rc + a6 2 + 2 c 2 4 rc 2 2 )2 2 )2 2 2) (rc (rc (rc ) 2(rc 2 +a4 rc 4 a4 + r c 4 r2 + c 2 2 )3 2(rc 2 2 rc + 2 2 a8 2

2 rc + 2 2 2 (rc 2 )4

50

Arg = 0
c

(r constant)
c

Arg = 0
c

(r constant)
c

7 12

I11 I12 I22

10

I 13 I 23 I /r
33

r2

r3 6 4 2 0 0 0.1 0.2 0.3 0.4 0.5 0.6

| |
c c

0.1

0.2

0.3

| |
c c

0.4

0.5

0.6

Arg = 0.5
c

(r constant) 7

Arg = 0.5
c

(r constant)

12

I 11 I12 I
22

10

I 13 I23 I /r
33

r2

r3 6 4 2 0 0 0.1 0.2 0.3 0.4 0.5 0.6

| c |

0.1

0.2

0.3

| c |

0.4

0.5

0.6

Figure 3.4: Plots of locked added inertia coecients for a variety of Joukowski foils. In each plot, rc and arg(c ) are held constant while c is varied. The coecients 2 3 I11 , I12 , I13 are given in units of rc , while I13 , I23 are given in units of rc and I33 has 4 units of rc .

51

Arg = 1
c

(r constant)
c

Arg = 1
c

(r constant)
c

7 12

I11 I12 I22

10

I 13 I 23 I /r
33

r2

r3 6 4 2 0 0 0.1 0.2 0.3 0.4 0.5 0.6

| |
c c

0.1

0.2

0.3

| |
c c

0.4

0.5

0.6

Arg = 1.5
c

(r constant) 7

Arg = 1.5
c

(r constant)

12

I 11 I12 I
22

10

I 13 I23 I /r
33

r2

r3 6 4 2 0 0 0.1 0.2 0.3 0.4 0.5 0.6

| c |

0.1

0.2

0.3

| c |

0.4

0.5

0.6

Figure 3.5: More plots of locked added inertia coecients for a variety of Joukowski foils. In each plot, rc and arg(c ) are held constant while c is varied. The coe2 3 cients I11 , I12 , I13 are given in units of rc , while I13 , I23 are given in units of rc and I33 4 has units of rc .

52 sampled from the (c , a) parameter space. Specically, I let c vary along each of several dierent rays emanating from the origin, while always choosing a so as to keep rc = 1. In other words, in each plot, c is the independent variable, while arg(c ) is constant (but a dierent constant in each pair of gures.) The closed form results for the added inertia were compared against results from numerical integration of the potential for these particular foil shapes, and found to be in agreement.

53

2 I11 = rc 2a2 + a4

I12 = 0 I22 = I13 = = I23 = = I33 = =


2 rc 2 4

2 rc 2 (rc 2 )2

2 rc + 2a + a 2 (rc 2 )2 2a4 c a6 r 2 c 2 Re i rc c 2a2 c 2 + 2 c 2 3 rc 2 (rc ) a6 r 2 y 2a4 y 2 2 c 2 3 rc y + 2a2 y + 2 rc 2 (rc ) a6 r 2 c 2a4 c 2 Re rc c + 2a2 c 2 2 c 2 3 rc 2 (rc ) 2a4 x a6 r 2 x 2 2 c 2 3 rc x + 2a2 x 2 rc 2 (rc ) 2 2 r2 2 2 2 Re 2a4 + rc 2 + 2a2 c 2a4 2 + 2a6 2 c 2 2 + a8 2 c 2 4 (rc 2 ) (rc ) (rc ) 2 2 2 (x y ) r2 2 2 2 2 2a4 + rc 2 + 2a2 (x y ) 2a4 2 + 2a6 2 + a8 2 c 2 4 (rc 2 ) (rc 2 )2 (rc )

Table 3.3: Components of the locked added inertia for the Joukowski foil.

54

3.4

Vortex Shedding

The wake from an airfoil moving in a potential ow may be computationally modelled as a trail of discrete point vortices. At each of a series of time steps, a new point vortex is shed from the airfoil, with strength chosen so as to satisfy the Kutta condition at that time step [Str94, ST95, Gie68]. The Kutta condition requires that the uid velocities on the trailing-edge upper surface and lower surface be equal in magnitude but opposite in tangential direction [Gie68]. To satisfy this requirement, the component of dw/d tangential to the surface C must vanish at = (a c ), the point corresponding to the trailing edge. Since the shed vortex generates a uid velocity tangential to C on C, it is always possible to choose a strength k for the vortex to satisfy the Kutta condition. (In fact, we must have dw/d = 0 at = a c in order for the physical velocity dw/dz at the trailing edge to be bounded. But the normal component of dw/d must vanish naturally from the form of the potential: it cannot be made to vanish by choosing the strength of the shed vortex.) Before solving for the strength k , the vortex must be placed at some location near the trailing edge of the airfoil. There is no obvious unique way to choose the exact starting location of the vortex. I adapt a scheme from Streitlien [Str94], interpolating between the trailing edge and the last vortex shed. The idea, illustrated in Figure 3.6, is that each discrete vortex represents a segment of a continuous vortex sheet which departs the trailing edge parallel to it. Close to the trailing edge the sheet can be approximated by a circular arc (or in the degenerate case, a straight line) tangent to the trailing edge and intersecting the discrete vortex shed in the previous time step. The trailing edge is located at z = 2a. If a > 0, the angle of the cusped trailing edge is = 2 where = arctan((y )/(a x ). If a < 0, the angle of the cusped trailing edge is = 2 + . I place the new discrete vortex one-third of the way along this arc, so that each vortex is approximately at the midpoint of the wake segment that it represents. 4
Streitlien [Str94] also seeks to place the vortex at the midpoint of its wake segment, but uses 4 instead of ! This seems to be a miscalculation. However, for small t the behavior of the simulation 3 should not strongly depend on this detail.
4

55

z0

0.5 2 r

/3 0 zk zk1 zk2

0.5 1.5

2.5

3.5

Figure 3.6: At each time step a new vortex is placed at the one-third point on a circular arc tangent to the trailing edge and intersecting the last vortex shed. The marks are the discrete vortices while the + marks delimit the wake segments which the vortices represent.

56 From examination of Figure 3.6, I have the following relations: z0 zk1 = rei(/2++) z0 2a = rei(/2+) z0 zk = rei(/2++/3) from which it follows that zk1 2a = rei(/2+) (1 ei ) zk 2a = rei(/2+) (1 ei/3 ) From the known location of the last shed vortex zk1 , I can solve for : ei = ei2 and then for zk : zk 2a = (zk1 2a) (zk1 2a) 1 ei/3 = 1 ei 1 + ei/3 + ei2/3 (3.66) (zk1 2a)2 zk1 2a 2 (3.65) (3.63) (3.64) (3.60) (3.61) (3.62)

In general, there are two circular arcs which intersect zk1 and are tangent to the trailing edge of the foil. (Together they make up the circle which intersects z k1 and is tangent to the trailing edge.) Note that 2 is also a solution to Equation (3.65). I must make sure to pick the correct arc, i.e., the one that departs the trailing edge in the correct direction (not necessarily the shorter of the two arcs.) What I want is to choose the branch of (i.e., add or subtract 2 from as necessary) so that is positive if arg(zk1 2a) (, + ), and is negative if arg(zk1 2a) ( , ). Importantly, these equations give sensible results in the degenerate case, when the arc is a straight line and = 0. As a consequence of Kelvins theorem, each one of the shed vortices subsequently moves with the same velocity as a uid particle at the same location zk = F (k ).

57 That is, the vortices convect according to Rouths rule: d dzk = [w ik log(z zk )]z=zk dt dz d d = [w ik log( k )] dz d = = = d d [w ik log( k )] dz d d d [w ik log( k )] dz d d d [w ik log( k )] dz d

+ ik
=k

d d [log( k ) log(z zk )] dz d
k

=k

+ ik lim
=k

+ ik
=k

(d2 z/d 2 ) 2(dz/d)2

dz/d d 1 dz k z zk
=k

=k

ik

a2 (c + k ) a2 (c + k )2
2

(3.67)

The velocities k can be recovered using dzk dk dc a2 = + dt dt dt (k + c )2 dk = dt a2 1 (k + c )2


1

dk dc + dt dt dzk dc . dt dt

(3.68)

(3.69)

In each time step the new vortex positions and velocities are found using a fourthand fth-order Runge-Kutta integration method. Since the location of each vortex exerts an inuence in determining the velocity of every other vortex, the diculty of integrating these dierential equations forward in time grows as O (n 2 ) where nv v is the number of vortices. One way to keep the computational burden from growing too large is to coalesce clusters of vortices which are more than a few chord lengths from the foil, and replace each cluster by a single point vortex at the clusters center of vorticity. Sarpkaya [Sar75] employed this procedure in simulating vortex shedding from an inclined at plate, and found that vortex clusters more than eight chord lengths from the plate could be coalesced without a material eect on the simulation. Carrying this process to its logical conclusion, a vortex which is suciently far away may be regarded as innitely distant and removed from the set of wake vortices, while the strength (k ) of its image is added to the central vortex strength c . The size of the time step t between shed vortices should obviously be chosen

58 suciently small that the systems behavior is no longer sensitive to the size of t. For the at plate, Sarpkaya [Sar75] found that a time step of t = 0.04a/U was suciently small (with the results for t = 0.02a/U and t = 0.08a/U not being materially dierent, but with t = 0.16a/U being too large.) For various motions of the deformable foil with U/a about unity and with the critical frequency of the deformations also being of order unity, I found time steps as large as t = 0.12 were adequate to qualitatively capture the wake structure and the forces on the foil, but not to reliably estimate the magnitude of the forces, for which a time step as small as t = 0.04 was required. Figure 3.7 shows a particular deformable foil simulation performed with a range of choices of t.

3.5

Forces on the Foil

In an inertial frame instantaneously aligned with the z coordinate frame, in which the uid at innity is at rest, the pressure p at each point in the uid is given by [MT68]: p = 1 2 (u + v 2 ) t 2 w+w 1 = t 2 2

dw dw dz dz

(3.70)

Using X and Y to represent the forces on the foil along the x- and y-axes respectively, we see that X =

p (dy) p dx

(3.71) (3.72)

Y or equivalently

X + iY = i

p dz

(3.73)

59
4 3 2 1 0 1 2 3 4 14 4 3 2 1 0 1 2 3 4 14 4 3 2 1 0 1 2 3 4 14 12 10 8 6 4 2 0 2 4 12 10 8 6 4 2 0 2 4 12 10 8 6 4 2 0 2 4

Figure 3.7: Wake from a heaving and exing deformable Joukowski foil. The foil began its motion with its centroid at the origin. The centroid then translated with uniform velocity Ucm = 1 in the x-direction while heaving in the y-direction with velocity Vcm = (0.4 sin t). Meanwhile the foil shape underwent a sinusoidal exing motion with (x , y ) = (0.05, 0.4 sin t). The modelled wake behavior at t = 12 is shown with the time step t between shed vortex points being (top to bottom) t = 0.12, t = 0.06, t = 0.04.

60 so: X + iY = i 2 w w dw dw + + t t dz dz dz. (3.74)

This expression captures all the forces on the foil, including the added mass forces. Therefore in evaluating this integral I will, among other things, rediscover the locked inertia coecients already derived in Section 3.3.

3.5.1

Unsteady Flow Force Terms

First I consider the part of integral (3.74) which is due to the time-varying potential.

(w+w) dz = U t + c

(w1 + w1 ) dz +V

(w2 + w2 ) dz +
k w5 dz

(w3 + w3 ) dz

(w4 + w4 dz

+
k

k 2

+ x

s s (w1 + w1 dz

+ y

s s (w2 + w2 dz

+a

s s (w3 + w3 dz

+U

w1 w1 ( + t t + c 2

+V w4 t

w2 w2 ( + t t +
k

+
k w5 dz t

w3 w3 + t t

k 2
s s w2 w2 + t t

+ x

s s w1 w1 + t t

+ y

+a

s s w3 w3 + t t

(3.75)
k where I have used the fact that w4 and w5 are real on by construction.

Some elements of the integral can be evaluated using Cauchys integral formula
2 and the fact that = rc / on C. The results are presented below with more details

in Appendix A.

61

2 w1 dz = 2i(a2 rc )

(3.76)

w1 dz = 2i(a2

2 a4 rc ) 2 (rc 2 )2

(3.77)

2 w2 dz = 2(a2 + rc )

(3.78)

w2 dz = 2(a2 +

2 a4 rc ) 2 (rc 2 )2

(3.79)

2 w3 dz = 2 rc c + a2 c

a4 c 2 (rc 2 )

(3.80)

w3 dz = 2 c a2

2 a6 c rc a4 c 2 2 (rc 2 ) (rc 2 )3

(3.81)

s s s In order to perform the integrals involving logarithmic terms in w4 and w1 , w2 , w3 ,

I do integration by parts, in each case choosing the branch cut associated with the logarithmic potential to pass through the point zcut = 2a, i.e., the trailing edge of the foil. More details are given in Appendix A.

w4 dz = 2(2a c )

(3.82)

62

s w1

2 2 2 2 a4 (c + rc ) 4a5 rc c 2a2 rc 2 2 dz = 2i +a + 2 c (rc 2 )2 (rc 2 )3 2 a2 2a4 r2 (iy ) 1 2 + 2 + 2 c 2 3 (2a c ) 2a2 rc 2 (rc ) c (rc 2 )3 2 rc 2

(3.83)

s w1

dz = 2i a +

a4 c
2

2 2a4 rc c 2 (rc 2 )3

4 a4 r 2 a6 r 4 a6 r 2 c a4 rc 2 c2 2 + 2 c 2 4 2 + 2 c2 4 2 (rc ) (rc ) c (rc 2 )2 (rc )

2a c +
3

2 +2a4 rc

c a2 2 (rc 2 )3

a2 c 2 (rc 2 )
2 (rc

c 2)

2 2a4 rc (iy ) (2a c ) 2 (rc 2 )3

(3.84)

s 2 w2 dz = 2 rc + a2

2 2a2 rc 4a5 r2 c a4 ( 2 r2 ) 2 c 2 c2 2 c 2 3 2 c (rc ) (rc ) 4 2 2 a2 2a r x 1 2 + 2 + 2 c 2 3 (2a c ) + 2a2 rc 2 (rc ) c (rc 2 )3

(3.85)

s w2

dz = 2 a

a4 c
2

2 2a4 rc c 2 (rc 2 )3

4 a6 r 4 a4 r 2 a6 r 2 c a4 rc + 2 c2 4 + 2 c2 2 2 c 2 4 2 2 (rc ) (rc ) c (rc 2 )2 (rc )

2a c +
3

2 +2a4 rc

c a2 2 (rc 2 )3

a2 c 2 (rc 2 )

c 2 (rc 2 )

2 2a4 rc x (2a c ) 2 (rc 2 )3

(3.86)

s w3 dz = 4ia

2 2 2 rc 2a3 rc r2 ((r2 2 )2 a2 c ) a2 c 2 + c c 2 (3.87) 2 c (rc 2 ) (rc 2 )2 c (rc 2 )2

63
1.5 z 1 0.5 0 0.5 1 1.5 S
x k

Sv S
b

Figure 3.8: If the direction of integration is counterclockwise in each case, then Sx Sv Sb . S

s w3 dz = 4ai

a4 r 2 c a2 + 2 c 2 3 (rc ) c 3 2 ar a2 c +4i 2 c 2 2 2a c + 2 (rc ) (rc 2 )


2 2 rc ((rc 2 )2 a2 c ) 2 c (rc 2 )3 2

4ia3

(3.88)

k Instead of evaluating w5 directly on , we consider another contour of integration

illustrated in Figure 3.8. The contour segment Sv encircles the vortex at z = zk . The branch cut between that vortex and its image vortex inside the foil is chosen to pass through the trailing edge of the foil at z = 2a. The contour segment Sb encloses the branch cut between the wake vortex and the trailing edge. Since the contour
k S Sx Sb Sv encloses a space with no singularities or branch cuts of w5 ,

k the integral of w5 over that contour must be zero.

k w5 dz =

k w5 dz

Sx

k w5 dz

Sv

k w5 dz

Sb

k w5 dz

(3.89)

k Since w5 is continuous in the vicinity of Sx , that integral vanishes. The integral

64 around Sv also vanishes in the limit as the radius of Sv approaches zero. The value
k of w5 is 2 higher on the lower side of Sb than on the upper side, so z=zk

Sb

k w5 dz =

2 dz
z=2a

= 2(zk 2a) Finally, in the limit of large ,


k w5 = i log(

(3.90)

rc r2 ) + log( k ) log( c ) k k

rc n 1 ) + i log() i k n = i log( k n n=1 Identifying the terms that go as 1/, we see that
k w5 dz = k w5 1 2 rc k

i log() i

n=1

2 rc k

1 (3.91) n n

a2 ( + c )2

d (3.92)

= 2 k So,
k w5 dz = 2 k

2 rc zk + 2a k

(3.93)

The integrals involving time derivatives of the Kircho potentials can all be evaluated using Cauchys integral formula. w1 d 2 2 dz = 4i ( a a rc rc ) = 2i a rc t dt

(3.94)

3 4 a2 i rc (rc 2 2 ) + a a (rc 5 + rc 3 2 ) + a2 rc 3 c x + i y w1 dz = t rc (rc 2 2 )3 (3.95)

65 d 2 w2 dz = 4 ( a a + rc rc ) = 2 a + rc 2 t dt

(3.96)

3 4a2 aarc 3 (rc 2 2 ) + rc (rc 2 2 ) + a2 rc 3 c x iy w2 dz = t rc (rc 2 2 )3

(3.97)

2a4 rc rc c 4a3 ac w3 dz = 2 2rc rc c + 2aac + + r c 2 x 2 2 t rc 2 (rc 2 2 ) a4 x a4 i y a4 i 2 y a4 2 x (3.98) 2 2 + irc 2 y 2 2 rc 2 rc 2 (rc 2 2 ) (rc 2 2 )

w3 2 a2 dz = t rc (rc 2 2 )4 + rc 2 2
4

4 a3 a rc 3 c rc 2 2 2 a a rc c rc 2 2 2 rc c + rc x i y
2

a2 c rc 2 2

4 rc 2 rc 2 rc 2 + rc c x i y (3.99)

+a4 rc 3 2 rc rc c 2 2 x i y + rc 2 x + i y

w4 dz = 0 t

(3.100)

k w5

dz = 2

rc 2 rc k + rc k k
2

k a + (c + k )2
2 s w1 dz rc

(3.101)

s w1 dz = c t

s w1 dz + a c

s w1 dz + rc a

(3.102)

66

s w1 dz = c

4a

i rc 2 c c 6 a rc 2 c y 3 rc 2 2 y 2 i rc 2 c 2 c + + (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4

2 a i rc 4 2 i rc 4 c 2 i rc 4 c 4 a i rc 2 2 (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4


2

(3.103)

s w1 dz = a

8 a4 i rc 2 c 4 a3 i rc 2 2 2 a3 i rc 2 (rc 2 2 )3 (rc 2 2 )3 (rc 2 2 )2 2 a3 i c 2 8 a 4 rc 2 y 4 a 3 rc 2 c y + (rc 2 2 )2 (rc 2 2 )3 (rc 2 2 )3 4 a i

(3.104)

s w1 dz = rc

2 a2 i rc 8 a5 i rc 3 c 2 a2 i rc 9 + 2 c 2 c (rc 2 2 )4 (rc 2 2 )4 4 a4 i rc 3 2 4 a5 i rc c 2 c 8 a2 i rc 7 c + + + c (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4 4 (i rc ) + 2 a4 i rc c 2 c 8 a2 i rc 3 c c 12 a2 i rc 5 c + + (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4


4 2 2 3

2 a2 i rc c 2 c a4 i r c 3 2 a4 i rc c 2 + + (rc 2 2 )4 (rc 2 2 )3 (rc 2 2 )3 8 a 5 rc 3 y 4 a 4 rc 3 c y a4 i r c 2 + + (rc 2 2 )3 (rc 2 2 )4 (rc 2 2 )4 4 a 5 rc 2 y 2 a 4 rc c 2 c y + (rc 2 2 )4 (rc 2 2 )4

(3.105)

67

s 4a4 w1 2 a rc 6 + 3 a2 i rc 4 c + rc 6 c i rc 6 c dz = 2 2 )5 c (rc i rc 4 c 3 a2 rc 4 c + 2 a2 i rc 4 c 2 a rc 4 2 + 3 a2 i rc 2 c 2 c

+rc 4 c 2 c + 2 i rc 4 c 2 c +2 i rc 2 c 4 c 2 a2 rc 2 c c + a2 i rc 2 c c + 4 a rc 2 c 2 c 2 rc 2 c 3 c i rc 2 c 3 c i c 5 c 6 a rc 4 c y +3 rc 4 c 2 y + 6 a rc 2 c 2 c y 3 rc 2 c 3 c y (3.106)
2 2 2 2 2 2

s w1 4a3 2 2 dz = 2 a2 rc 2 2 c + i rc 2 + 2 2 + c + 8 a r c 2 rc 2 2 2 2 )4 a (rc + rc 2 2 i rc 2 + c 2 rc 2 2 4 r c 2 c c + y

c + y (3.107)

s w1 4a2 a4 i rc 6 i rc 10 + a2 i rc 6 c 2 8 a3 rc 6 c + 5 a4 i rc 4 2 dz = 2 rc rc (rc 2 )5

+4 a2 rc 6 2 + 5 i rc 8 2 3 a2 i rc 4 c 3 c 4 a4 rc 4 c + 2 a4 i rc 4 c
2 2 2 2

2 2

+4 a3 rc 4 c c + 2 a4 i rc 2 c 2 c 2 a2 rc 4 c 2 c 10 i rc 6 c 2 c + 3 a2 i rc 2 c 4 c 2 a4 rc 2 c c + 4 a3 rc 2 c 2 c 2 a2 rc 2 c 3 c
3 3 4 3 3 3

+10 i rc 4 c 3 c a2 i c 5 c 5 i rc 2 c 4 c + i c 5 c 8 a3 rc 6 y + 4 a2 rc 6 c y +4 a3 rc 4 2 y 2 a2 rc 4 c 2 c y + 4 a3 rc 2 c 2 c y 2 a2 rc 2 c 3 c y
2 2

(3.108)

2 s 4a4 rc w2 dz = 2 rc 2 c c + 2 2 c + c 3 i y 2 a rc 2 + 2 2 3 i c y 2 c (rc 2 )4 (3.109)

68

s 32 a4 rc 2 c 16 a3 rc 2 2 w2 dz = 4 a + a (rc 2 2 )3 (rc 2 2 )3 8 a3 c 2 16 a3 i rc 2 (2 a c ) y 8 a3 r c 2 + (3.110) + (rc 2 2 )2 (rc 2 2 )2 (rc 2 2 )3

s 6 a 4 rc 2 2 2 a 4 rc 2 4 a5 c 2 a4 c 2 12 a5 rc 2 c w2 dz = 4rc 1 + + rc (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )3 (rc 2 2 )3 (rc 2 2 )3 2 a4 2 a4 12 a5 i rc 2 y 6 a4 i rc 2 c y + + + (rc 2 2 )3 (rc 2 2 )2 (rc 2 2 )4 (rc 2 2 )4 4 a5 i y 2 a4 i c y (3.111) + (rc 2 2 )3 (rc 2 2 )3

s 4a4 w2 a2 rc 2 2 3 c + c + 2 i c + rc 2 3 c + (2 + i) c dz = 2 c (rc 2 )5 +2 a i rc 2 rc 2 2 rc 2 + 2 2 + 3 c y

c rc 2 2

(1 + i) rc 4 + c 3 c rc 2 c c + c 2 i c 3 i y (3.112)

s w2 4a3 dz = 2 a2 rc 2 rc 2 c 2 c + c + 2 i c 2 a (rc 2 )4 8 a i rc 2 rc 2 2 c + y

+ rc 2 2

rc 4 + c 3 c r c 2 c c + c 4 i c 4 i y (3.113)

69

s a2 a4 rc 5 1 w2 12 a2 rc 5 c 2 2 dz = 4a + rc rc rc c 2 (rc 2 2 )5 (rc 2 2 )5 8 a 2 rc 7 c 5 a 4 rc 3 2 2 a 2 rc 9 + 2 5 (rc 2 2 )5 c (rc 2 2 )5 c (rc 2 2 )

2 a 2 rc c 4 c 2 a4 i rc c c 8 a3 i rc 3 c + + + (rc 2 2 )5 (rc 2 2 )5 (rc 2 2 )4


2

8 a 2 rc 3 c 3 c 4 a 4 i r c 3 c 2 a 4 rc c 2 c + (rc 2 2 )5 (rc 2 2 )5 (rc 2 2 )5


2 3

8 a3 i rc 3 y 2 a 4 rc 3 c 2 2 3 + (rc 2 2 )4 c (rc 2 2 )2 (rc 2 2 ) (rc 2 + 2 ) 4 a2 i rc 3 c y 4 a3 i rc 2 y 2 a2 i rc c 2 c y + (3.114) (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4 a2 rc 3


s 4 a3 i rc 2 3 rc 2 + 4 a c c c w3 dz = 3 c rc 2 c c

2 a2 i rc c 2 c 4 a2 i rc 3 2 4 a3 i rc c c + (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4


2

(3.115)

s 12 a2 i 2 a rc 2 + c 2 rc 2 c c w3 dz = 2 a rc 2 c c s 16 a3 i rc rc 2 c + a rc 2 + c c w3 dz = 3 rc rc 2 c c

(3.116)

(3.117)

s w3 4a3 2 dz = 4 a rc 2 c rc 2 2 + c rc 2 2 2 c (rc 2 )4 a2 rc 2 3 i rc 2 + 2 2 + 4 i 2

(2 + i) rc 2 i 2 (3.118)

70

s 4a2 w3 6 a rc 4 + 3 rc 4 c i rc 4 c 3 a2 rc 2 c + 6 a2 i rc 2 c dz = 2 2 )3 a (rc

+6 a rc 2 2 3 rc 2 c 2 c + 2 i rc 2 c 2 c i c 3 c

(3.119)

s 8a3 w3 a2 rc 2 c (1 + 2 i) rc 2 + (1 + i) 2 dz = 2 2 )4 rc rc (rc

+c rc 2 2

(1 + i) rc 4 + (1 2 i) rc 2 2 + i c 2 c
2

2 a rc 6 rc 2 c 2 c

(3.120)

3.5.2

Bernoulli Eect Force Terms

Now I turn to the last term in Equation (3.74), involving cross-terms between the Kircho potentials. In order to evaluate this integral, I rst consider its complex conjugate. dw dw dz = dz dz dw dw dz dz dz

(3.121)

dw dw dz = dz dz =

=
C

=
C

dw d dw d dz d dz d dz dw dw d d d d dz 2 ( + c )2 rc dw dw d d d ( + c )2 a2 2 2 ( + c )2 rc dw dw d d ( + a + c )( a + c ) 2

d (3.122)

The factors apart from dw/d and dw/d introduce poles at = 0; at = (a c ); and at = (a c ). The former two poles lie inside C, and the third lies exactly on C.

71
1.5

0.5 C

0 a c Sa 0.5

1.5 1.5

0.5

0.5

1.5

Figure 3.9: I consider a small excursion around the pole at = a c . Now consider the contour C+ = C + Sa illustrated in Figure 3.9. The integral around the small semicircle Sa will be equal to the residue of the integrand at = ac times i. Therefore: dw dw dz = dz dz dw dw d d dw dw d d ( + c )2 ( + a + c )( a + c )
2 rc 2 2 rc 2 2 rc 2

d d d

C+

Sa

C+

dw dw d d dw dw d d dw dw d d

( + c )2 ( + a + c )( a + c ) ( + c )2 ( + a + c )
2 rc 2

( + c )2 ( + a + c )( a + c )

i =
C+

=ac 2 rc 2

+i

2 dw dw arc 2 2(a c ) d d

( + c ) ( + a + c )( a + c )
=ac

d (3.123)

As long as the Kutta condition is satised, dw/d will equal zero at = a c , and the nal term in Equation (3.123) vanishes.
2 Through the substitution rc /, I can replace dw/d with a function of

which takes on the same values on C+ . Then we can evaluate all terms of the integral

72 on C+ by the theory of residues. The results are given in Appendix B.

3.6

Moments on the Foil

The moment about the point z = 0 is given by M =

x p dx + y p dy p Re z dz

= = Re

p z dz

(3.124)

So employing Equation (3.70), we have 1 M = Re 2 w z dz + t w z dz + t dw dw z dz dz dz (3.125)

2 2 I have to make the replacements z = F () and dz = (rc / 2 )(1 a2 /((rc /) + c )2 ).

Then the following integrals can be readily evaluated in the -plane using the theory of residues: w1 z dz = 2 a2 i rc 6 c rc 4 2 a2 + 3 c 2 c c 2 a2 + c 2 c +rc 2 c 2 a4 + 3 c a2 + c 2 c
3

/ rc 2 c c

(3.126)

w1 z dz =

w1 z dz
3

= 2 i a6 rc 2 c + a2 c rc 2 c c a4 rc 2 c c c c
2

r c 2 c rc 2 c c

3 3

+ r c 2 c + c

/ rc 2 c c

(3.127)

73

w2 z dz = 2 a2 rc 6 c + rc 4 2 a2 + 3 c 2 c + c 2 a2 + c 2 c +rc 2 c 2 a4 3 c a2 + c 2 c / rc 2 c c

(3.128)

w2 z dz =

w2 z dz = 2 a6 rc 2 c + a 2 c rc 2 c c a4 rc 2 c c
3

+ r c 2 c rc 2 c c
2

3 3

rc 2 c c rc 2 + c

/ rc 2 c c

(3.129)

w3 z dz = a2 2 a4 c

rc 2 c c
2

+ 2 c 2 rc 2 c c
2

+ a 6 rc 4 + 3 r c 2 c c / rc 2 c c
4

+a2 rc 2 c c

rc 4 7 r c 2 c c + 4 c 2 c

(3.130)

w3 z dz =
2

w3 z dz rc 2 c c
2

= 2 a6 c

+ 2 a 2 c 2 rc 2 c c
2

+ 2 r c 2 c c rc 2 c c
2

a 8 rc 4 + r c 2 c c
4

+a4 rc 2 c c

3 rc 4 5 r c 2 c c + 4 c 2 c

/ rc 2 c c

(3.131)

In order to evaluate the terms involving log(/rc ), I make the substitution = rc ei (and d = irc ei d) and integrate with respect to . The integration with respect to must respect the branch cut at the trailing edge of the foil and therefore the limits of integration must be = [, 2 ] where = arctan(y /(a x )).

74

log( ) z dz = rc =

a2 log( ) + c + rc ( + c ) C

rc 2 2

a2 rc 2 rc 2 + c
2

2 2 i a3 r c 2 c + r c 2 c + 2 i a c 2 2 (i a2 rc 2 a2 rc 2 + a2 2 rc 2 ) 2 c c 2 i (a rc 2 + a2 c ) 2 i a2 + c 2 c 2 i ( (a3 rc 2 ) + a4 c ) c c c rc 2 + c c + 2 i a 2 rc 2 2 i a 2 rc 2 c
2

a2

rc 2 c c a2
2

a log( ) c rc 2 + a c c c ) (3.132) rc 2

c 2

rc 2 c c

log(

a2 a2 rc 2 1 + c + 2 log( ) d log( )z dz = rc rc rc + c ( + c )2 C 2 i a3 rc 2 2 i a3 rc 2 2 rc 2 (a + c ) (i a a + a c + c ) + 2 = c 2 (rc + a c ) c c rc 2 + c c 2 i 2 a3 rc 4 + a rc 6 2 a4 rc 2 c + 2 a2 rc 4 c + 2 a3 rc 2 c 2 c (rc 2 + a c ) rc 2 a c + c c c 2


2

+ 2 i a 2 rc 2 2 i a 2 rc 2 c

a2

rc 2 c c a2
2

a log( ) c rc 2 + a c c c ) (3.133) rc 2

rc 2 c c

log(

The integral of log(( + c )/) z dz, which occurs in the shape deformation potentials, can be best evaluated using the contour in Figure 3.10. I must account for the
2 pole at = rc /c , which occurs in the factor replacing dz. The integrand drops

o as O (1/ 2 ) for large , so there is no contribution from C , and the integrand is continuous in the vicinity of Cb2 and Cx , so there is no contribution there. The only contribution comes from Cp , which can be evaluated by taking the residue of the

75

1 C C 0 0 1 C 2
b2 c

4 C 5
p

r2/ c c 1 0 1 2 3 4 5 6 7 8 9

2 Figure 3.10: The contour Cp encloses a pole at = rc /c . There is a branch cut between the origin and = c . If integration is counterclockwise in each case, then = C Cx Cp Cb2 . C

76
2 integrand at = rc /c .

log(

+ c ) z dz =

log(
C

a2 + c ) + c + ( + c ) rc 2 2
2

+ c a2 log( = ) + c + ( + c ) Cp

+ 2

rc 2 2 a2 rc 2
rc 2
2

+ 2

a2 rc 2
rc 2

+ c

= +

2 a2 i c rc 4 2 rc 2 c c + a2 c + c 2 c c rc 2 c c rc 2 c c
2 2

+ c

2 a 2 i r c 2 rc 2 + a c c c

2 rc 2 + a c + c c log (rc 2 c c )/rc

(3.134)

The following integral does have a nonvanishing contribution from C as well as Cp . log(

rc 2 + c + c a2 a2 ) z dz = log( ) 1 + c + 2 rc + c ( + c )2 C 2 2 2 rc a a + c ) 1 + c + 2 log( = d 2 rc + c ( + c ) C a2 + c rc 2 a2 d log( ) 1 + c + 2 rc + c ( + c )2 Cp 2 i a2 rc 2 rc 4 + 2 rc 2 c c + a2 c c 2 c a2 + c 2 2 c c rc 2 c c


2 2

= 2 i c

2 2 log (rc c c )/rc

(3.135)

Finally, I can conclude that log(

+ c ) z dz = rc

log(

+ c ) + log( ) z dz rc

(3.136)

log(

+ c ) z dz = rc

log(

+ c ) + log( ) z dz rc

(3.137)

I now have the tools to evaluate the moments arising from the shape deformations

77 and from the central vortex potential. It immediately follows that 2 i 2 i a3 rc 2 c + rc 2 c + 2 i a c 2 2 i (i a2 rc 2 a2 rc 2 + a2 2 rc 2 ) 2 c c

w4 z dz =

2 (a rc 2 + a2 c ) 2 a2 + c 2 c 2 ( (a3 rc 2 ) + a4 c ) + + + c c c rc 2 + c c + 2 a 2 rc 2 + 2 a 2 rc 2 c
2

a2

rc 2 c c a2
2

a log( ) c rc 2 + a c c c ) (3.138) rc 2

c 2

rc 2 c c

log(

s w1 z dz = 2 a2 i (rc c ) (rc + c ) rc 2 + c 2 c 2 2 5 3

rc 2 c c

+a4 rc 2 c c

i rc 6 c + i c 3 c + rc 4 c i c + 2 (i + ) c 2 (c i y )
2

i rc 2 c 2 c 4 c 3 + 2 c 2 c i y + c i a6 rc 2 c 2 c
2

c 2 i y

rc 4 + 2 c 2 c 3 c + c i y + rc 2 c 2 c + c + 2 i y
2

+2 a5 rc 2 c 2 c

rc 2 2 c c i rc 2 c c

rc 2 c c (i c + y ) rc 3 r c c c
3 2

+2 a3 c 3 c rc 2 + 2 c c +2 a2 rc 2 c c rc 2 c c
2

(i c + y )

+ c 3 c ( + ) rc 2 i c c rc 6 3 r c 4 c c + 3 r c 2 c 2 c rc 2 c c ) rc 2
2

+i c 2 c ( ) rc 2 + i c c y 2 i a2 rc 4 rc 2 + (a c ) c
3

rc 2 (a + c ) c log(

+c 3 c + a2 c + c 2 i a4 rc 4 c (c i y ) + rc 2 + c a c

c 2 rc 2 + (a c ) c c
2

rc 2 c a + c

a rc 2 (a + c ) c log( ) c 2 rc + a c c c log( ) rc 2
2

/ c 3 c

rc 2 + c c

(3.139)

78

s w1 z dz = 2 c i a rc 2 c 2 c

rc 2 c c
3

+ i r c 2 c 2 c

rc 2 c c
6

+i a3 c c rc 2 c c +i a9 rc 2 c c
3

rc 4 c c
3 2

+ i a 2 c rc 2 c c
3 2 3

rc 4 c c
6

rc 4 + c c +i rc 2 c

+ a 6 rc 2 c c

i rc 6 c i c c

c 3 c 2 c + c c + 2 c
6 2

+rc 4 c c i c + 2 (i + ) c + i y +a7 c rc 2 c c
2

i rc 6 c i c c + i rc 2 c
2

c 3 c 2 c + c c + 2 c

+rc 4 c i c 2 + 2 ( ) c c + i y + 2 i c c + i y +i a8 rc 2 c c rc 2 c c
3

rc 4 + c c + 2 r c 2 c c + i y
3

+a4 rc 2 c rc 2 c c +i 2 rc 6 6 rc 4 c c + c 2 c +a5 rc 2 c
2 2

2 ( + ) rc 2 c c
2

c + c + rc 2 c c

7 c + c 2 y + 2 y
2

rc 2 c c
3

2 rc 2 c c ( ) c + c y
2

+i 2 rc 6 6 rc 4 c c + c 2 c

c + c + rc 2 c c
2

9 c + c 2 y + 2 y

+2 a6 rc 4 c 2 c rc 2 + (a c ) c +2 a4 rc 4 rc 2 + c a c
2

rc 2 + (a + c ) c

a i c + y log( ) c i rc 6 3 r c 4 c c

rc 2 + (a c ) c
3

rc 2 c a + c log(

+3 rc 2 c 2 c + c +a2 c
3

c 3 + a2 c + c

rc 2 c c ) rc 2

i c + y log( / c 2 c
3

rc 2 + a c c c ) rc 2 rc 2 c c
5

rc 2 + (a c ) c

(3.140)

The other shape deformation integrals can be evaluated similarly. For the integrals involving the wake vortices:
k w5 z dz = k w5 + c +

a2 ( + c )

2 2 a2 rc rc + 2 2 (rc + c )2

(3.141)

I must consider the contour illustrated in Figure 3.11. Not only do I have to account

79

2 k C 0 a
c

1 C
b

Cv C
x

b2

4 Cp
c

_ r2/

2 Figure 3.11: The contour Cp encloses a pole at = rc /c . If the direction of integration is counterclockwise in each case, then C = C Cx Cp Cv Cb .

80
2 for the pole at = rc /c , which occurs in the factor replacing dz, but I must also

allow for the branch cut between the wake vortex and its image. Since the integral around the entire contour in Figure 3.11 must be zero, I have
k w5 + c +

a2 ( + c )

2 2 rc a2 rc + 2 2 (rc + c )2 k w5

d
2 2 a2 rc rc + 2 2 (rc + c )2

=
C

Cx

Cp

Cv

Cb

a2 + c + ( + c )

d (3.142)

k Recalling the series expansion for w5 in Equation (3.91), 2 rc k n

k w5

rc n 1 = i log( k n + )i k n n=1

i
n=1

1 n n

(3.143)

I isolate the part of the integrand of Equation (3.141) that goes as 1/ and conclude
k w5 + c +

a2 ( + c )

2 2 a2 rc rc + 2 2 (rc + c )2

d r a2 r 2 2 ) rc 2c k c (3.144)

= 2 log(

The integrand is continuous on Cx and Cb , so there is no contribution from those integrals. Since Cv encloses only a logarithmic singularity, it gives no contribution as
k the radius of Cv approaches zero. On Cb , w5 is larger by the amount 2 on the lower

side than on the upper side. So,


k w5 + c +

Cb

a2 ( + c )
k

= 2
ac

2 2 a2 rc rc + 2 2 (rc + c )2 a2 + c + ( + c )

d
2 2 a2 rc rc + 2 2 (rc + c )2

d (3.145)

81 This is an elementary line integral in which equals


k w5 + c +

Cb

a2 ( + c )

2 2 a2 rc rc + 2 2 (rc + c )2

d
2 2

= 2 rc 2

2 a2 rc 4 2 r c 2 c c + a 2 c + c 2 c 2 a c + 2 c c rc 2 + c rc 2 + c c

a2 + c 2 log() + c 2

a2 rc 4 a 4 c 2 2 a 2 rc 2 c c + a 2 c 2 c c 2 rc 2 + c c
2 2 2

log( + c ) =k
=(ac )

(a2 rc 4 ) + 2 a2 rc 2 c c + a4 c a2 c 2 c c
2

log(rc 2 + c )

rc 2 c c

(3.146)

Finally, the contribution from Cp is just 2i times the residue of the integrand at
2 = rc /c .

Cp

k w5 + c +

a2 ( + c )

2 2 a2 rc rc + 2 2 (rc + c )2

d
2

= 2 a2 c rc 2 + c c rc
2

rc 4 2 r c 2 c c + a2 + c 2 c rc + c k
2 2

rc 2 k k rc 2 + c k k rc k c + k (3.147)

rc + (a c ) c

rc (a + c ) c / c
2

c + k log c + k

rc 2 c c

rc 2 + c k

Collecting all the terms, I nd


k w5 z dz =

2 A + B log(

rc k (r2 + ac 2 )k c + k ) + B log( ) + C log( c ) ) + C log( k a c a rc k (c + k )

(3.148)

82 where a + c + 2 (a c ) + c 2 c rc 2 + (a c ) c a2 + c 2 c k
2 2

A = rc 2 +rc 2 B =

a2 rc 4 2 r c 2 c c + a2 + c 2 c
2

a2 rc 4 2 r c 2 c c + a2 + c 2 c
2 a2 rc

rc + c c 2 a2 rc 4 2 r c 2 c c + a2 + c 2 c + 2 c rc 2 + c c rc 2 + c k
2

rc 2 k k

c rc 2 + c c c
2

rc 2 + c k

c + k

2 rc

C =

a2 rc 2 rc 4 2 rc 2 c c + a2 + c 2 c c
2

rc 2 c c

Only the leading term in Equation (3.148) will actually appear after I take the real part to nd the moment in Equation (3.125). To see this, I evaluate this integral on the contour of Figure 3.11:
k w5 z dz = k w5 z dz = k w5 z dz

(3.149)

k w5 z dz =

k w5 1

a2 ( + c )2

rc 2 a2 + c + 2 rc + c

d d (3.150)

=
C

k w5

a2 a4 a2 rc 2 a2 c rc 2 + c + 2 ( + c )2 ( + c )2 rc + c ( + c )2 rc 2 + c

k 2 Since w5 i log(rc /k ) ik / + irc /(k ) for large , I can identify the terms of

83 the integrand that go as 1/ and nd a2 c a2 a2 rc 2 rc 2 a4 + c + 2 ( + c )2 ( + c )2 rc + c ( + c )2 rc 2 + c = 2 a2 + c c k + rc 2 k + rc 2 a2 rc 2 c


2

k w5

log

rc k

(3.151)

The integral cancels around all the linear parts of the contour except for Cb .
k w5 1

Cb

a2 ( + c )2
k

= 2
ac

a2 rc 2 d + c + 2 rc + c a2 a2 rc 2 1 + c + 2 rc + c ( + c )2
2

= 2

a2 + c c +

a2 rc 4 + a 2 c 2 2 r c 2 c c + c 2 c c ( + c ) rc 2 + c c

(a2 rc 2 ) + rc 2 c 2 log() + c 2
2

a2 rc 6 a 4 rc 2 c 2 2 a 2 rc 4 c c + a 2 rc 2 c 2 c c 2 rc 2 + c c
2 2 2

log( + c ) =k
=ac

(a2 rc 6 ) + 2 a2 rc 4 c c + a4 rc 2 c a2 rc 2 c 2 c c
2

log(rc 2 + c )

rc c c

(3.152)

2 Finally, the contribution from Cp is found by taking the residue at = rc /c .

Cp

k w5 1

a2 ( + c )2

a2 rc 2 + c + 2 rc + c

d
2

= 2 a2 rc 2 rc 4 + 2 rc 2 c c + (a c ) (a + c ) c log rc 2 + c k k rc k c + k / c
2

rc 2 c c

(3.153)

84 So collecting all terms and taking the complex conjugate, I nd


k w5 z dz = k w5 z dz =

2 A B log(

rc (r2 + ac 2 )k k c + k ) ) C log( c ) B log( ) C log( a a c k rc k (c + k )

(3.154)

where A = a 2 c c c rc 2 1 + a3 + c rc 2 + c c c rc 2 + c k a4 c k rc 2 + c c c + k
2

a2 rc 2 c c k + c k + c k c 2 + c c + k c c k c + k

The log terms in Equation (3.154) are negative complex conjugates of the log terms in Equation (3.148). This shows that all the log terms taken together are pure imaginary and will be discarded in Equation (3.125). So only the leading terms A + A are actually signicant in determining the moment on the foil. The Bernoulli pressure terms can again be dealt with by using the contours illus-

85 trated in Figure 3.9. dw dw z dz dz dz dw dw ( + c )2 ( + c )2 a2 C d d dw dw ( + c )2 = ( + c )2 a2 C+ d d = =


C+ Sa

+ c +

a2 + c a2 + c + + c a2 + c a2 + c
2

2 rc d 2 r2 c d 2 2 rc 2 2 rc 2

dw dw d d

dw dw d d

( + c )2 ( + c )2 a2

+ c + + c + + c +

d d

i =

dw dw d d
C+

( + c )2 ( + c + a)

( + c )2 ( + c )2 a2

2 rc 2

a2 + c a + c

=ac

dw dw d d

( + c ) ( + c )2 a2

+ c +

r2 c 2

d (3.155)
=ac

2 dw dw a2 rc + i 2 (a c ) d d

where the last term vanishes due to the Kutta condition.

86

3.7

Equations of Motion

Collecting terms, we see that the force on the foil is X + iY = U A1 V A2 A3 + x B 1 + y B 2 + a B 3 + U 2 C 1 + U V C 2 + V 2 C 3 + U C 4 + V C 5 + 2 C 6 + U x D1 + U y D2 + U a D 3 + U rc D4


2 2 + x E 1 + x y E 2 + y E 3 + x a E 4 + y a E 5 + a2 E 6 + x rc E 7 + y rc E 8 + arc E 9

+ V x D5 + V y D6 + V a D 7 + V rc D8 + x D9 + y D10 + a D11 + rc D12 + U c F 1 + V c F 2 + c F 3 + x c F 4 + y c F 5 + ac F 6

+
k

6 k U G k + V G k + G k + x G k + y G k + a G k + rc G k + k G k + k G k + c G k 1 2 3 4 5 7 8 9 10 +
k j=k

k j G k + 11

2 k G k + c H 1 + 12

k Hk + 2

2 arc dw 2 4(a c ) ds

2 =ac

(3.156) where the nal term vanishes as long as the Kutta condition is satised, and where A1 = A2 = i A3 = 2 a2 + rc 2 + a4 rc 2 (rc 2 2 )2 a4 rc 2 2 a2 + r c 2 + (rc 2 2 )2

i a6 rc 2 c + 2 a4 c (rc 2 2 ) rc 2 c (rc 2 2 ) + 2 a2 c (rc 2 + 2 ) (rc 2 2 )3

I can relate some of the coecients in Equation (3.156) to the locked inertia coecients found in Section 3.3 as follows: A1 = I11 + iI12 A2 = I12 + iI22 A3 = I13 + iI23 The other coecients are given in Appendix C. (3.157) (3.158) (3.159)

87 After creation, the wake vortices are assumed to have constant strength k . The central vortex strength c is also normally assumed to be constant. Nevertheless, the terms in Equation (3.156) that are proportional to c and k can be signicant whenever vortices are introduced to or removed from the ow. Whenever the k th vortex is shed from the foil into the wake, its strength may be considered to rise from zero to k in the time step of creation, and the Hk term will 2 exert some force on the foil (with k = k /(t)). Streitlien [Str94] noted that if this eect is neglected in simulation of a rigid foil, the force calculations will still converge to the correct result as the time step t approaches zero, but only at a rate t. On the other hand, in order to reduce the computational burden and avoid tracking the location of all shed wake vortices indenitely, we may wish to remove some vortices distant from the foil, and either combine them into a single vortex at their center of vorticity, or stop tracking them altogether and assume they have moved permanently to innity. In the latter case, the images of the innitely distant vortices move to the origin and the central vortex strength c is changed by an amount (k ), the strength of the images. During the time step in which such simplication of the wake occurs, we should have force contributions from the Hk and possibly the H1 terms, 2 which should substantially cancel if the wake simplication is justied.

88 The moment on the foil is given by M = U J 1 V J 2 J 3 + x K 1 + y K 2 + a K 3 +U 2 L1 +U V L2 +V 2 L3 +U L4 +V L5 +2 L6 +U x M1 +U y M2 +U a M3 +U rc M4


2 2 + x N 1 + x y N 2 + y N 3 + x a N 4 + y a N 5 + a2 N 6 + x rc N 7 + y rc N 8 + arc N 9

+ V x M5 + V y M6 + V a M7 + V rc M8 + x M9 + y M10 + a M11 + rc M12 + U c P 1 + V c P 2 + c P 3 + x c P 4 + y c P 5 + ac P 6 + rc c P 7

+
k

k U Qk + V Qk + Qk + x Qk + y Qk + a Qk + rc Qk + k Qk + k Qk + c Qk 6 9 10 1 2 3 4 5 7 8 k j Qk 11 +
k 2 k

+
k j=k

Qk 12

+ c R1 +

k Rk 2

a2 y (a x ) dw + 2 rc ds

2 =ac

(3.160) where the nal term vanishes as long as the Kutta condition is satised. The coecients are given in Appendix D. As an example, I calculate the forces and moments on a foil executing the following particular motion: the centroid of the foil moves with uniform velocity (U cm , Vcm ) = (1, 0) while the foil pitches with = 0.3 cos(t) and deforms with (x , y ) = (0.15, 0.3 sin(t)), while rc = 1. This trajectory, illustrated in Figures 3.12-3.13, is intended to be suggestive of a swimming sh, since the pitching and exing deformation are countervailing, leaving the nose of the foil pointed in approximately the same direction while the rear part aps back and forth. The forces and moments experienced by the foil are plotted in Figure 3.14. The motion generates a positive forward thrust and also a positive average moment, while the lateral forces settle into a symmetrical oscillation with zero mean. There is a transient spike in the lateral force in the rst few time steps but otherwise the system displays periodic behavior immediately, i.e., there is little noticeable secular change in force on the foil as the wake evolves in time.

89

4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14

4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14

Figure 3.12: Flapping motion of a foil with (Ucm , Vcm ) = (1, 0), = 0.3 cos(t), (x , y ) = (0.15, 0.3 sin(t)), rc = 1. The point vortices are being shed at intervals of t = 2/100 = 0.0628. The snapshots shown are at intervals of 10 t = 2/10.

90

4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14

4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 4 3 2 1 0 1 2 3 4 4 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14 2 0 2 4 6 8 10 12 14

Figure 3.13: Continuation of Figure 3.12. Flapping motion of a foil with (Ucm , Vcm ) = (1, 0), = 0.3 cos(t), (x , y ) = (0.15, 0.3 sin(t)), rc = 1. The point vortices are being shed at intervals of t = 2/100 = 0.0628. The snapshots are at intervals of 10t = 2/10.

91
4 3 2 1 0 1 2 3 4
25 Thrust Lift Moment 20

10

12

14

15

10

10

10

12

14

Figure 3.14: Forces and moments on the apping foil of Figures 3.12-3.13, in a stationary frame oriented with the reader. While the oscillating force in the y-direction has an average approaching zero, the oscillating thrust in the x-direction has a positive mean value of 0.23. The moment also has a signicantly positive mean value.

92

3.8

Self-Propulsion of a Deformable Foil

Equations (3.156) and (3.160) can be used to nd the force and moment on a deformable foil undergoing any prescribed motion and deformation. They are also very broadly applicable without requiring any assumptions about the internal structure of the foil. A prescribed trajectory might arise in a practical sense if the deformable foil is attached to an external body which can move the foil through a given trajectory independent of the uid forces experienced by the foil. In particular, if the apping foil of Figure 3.14 were attached as a tailn propulsor to another body, we can see how the tailn could provide forward thrust to the body while inertial or internal forces provided by the body absorbed the lateral forces. By using stabilizing forward ns, or by using paired propulsors, the body could manage the moments generated by the tailn and maintain the desired orientation. Suppose, on the other hand, that instead of attaching the deformable foil as an appendage to another body, we are interested in its ability to swim as an isolated body in its own right. In order to nd which motions (if any) generate self-propulsion, I have to make some assumption about the mass distribution (z). The linear momentum of the foil, expressed as a complex number, is Px + iPy = (z) U + iV + iz + F F c + a c a dA, (3.161)

while the angular momentum is L = Re i (z) U + iV + iz +


F c c

F a a

z dA .

(3.162)

As a consequence of Stokess theorem, for an arbitrary function f (z, z) on a region R bounded by R:


R

f dz = 2i

f dA z

(3.163)

So if the function (z) is suciently congenial (for example, a polynomial of z and z) I can transform the momentum expressions into contour integrals around the foil

93 boundary, then make the usual transformations and evaluate the integrals using the theory of residues in the -plane.

3.8.1

Uniform Foil Density

As an example, I assume that the foil is homogeneous with uniform density (z) = . (The density may, however, be time-dependent if the foil undergoes area-changing deformations.) Then: Px + iPy = or
2 irc 2 2

1 2i

2 U z + iV z + iz 2 /2 + c F /2 + a F 2 /2 c a

dz

(3.164)

Px + iPy =
C

i a2 + + c + 2 + c

a
rc 2 2

+ c

+ 1 a2 + c

2 a a + c + 2 + c a2 ( + c )2 U +i

a2 +c

+ c +

a2 + c V

c d (3.165)

+ + c +

+ c +

a2 + c

The integral can be evaluated by nding the residues of the integrand at = 0 and = c , yielding Px + iPy = rc 2 a + + 1+ 4 a5 rc 2
3

4 a5 c rc 2 + c c i a6
2

c rc 2 + c c

i c

i a 6 rc 2

3 a 6 rc 4

c rc 2 + c c
4

c 2 rc 2 + c c + 1

6 a 6 rc 2

c rc 2 + c c
3

a4

c 2 rc 2 + c c
2

3 a6
2

c 2 rc 2 + c c i a4 rc 2 + c c
2

c (3.166)

rc 2 + c c

U+

94 If the total mass m = A of the body is conserved, then: 4 a5 a c rc 2 c c a4 + rc 2 c c


3 2 3 4 2

Px + iPy = m U + i V + +

i a6 c + c rc 2 c c
2

rc 2 c c + rc c c
2 2

a4 + rc 2 c c a4 + rc c c
2 2

rc 8 4 r c 6 c c + 3 a6 + 2 r c 4 c 2 c 4 r c 2 c 3 c + c 4 c

. (3.167)

Meanwhile the angular momentum equals 1 L = Re 2


U z + iV z + iz 2 /2 + c c F 2 /2 + a a F 2 /2 z dz .

(3.168)

Evaluating the integral by the method of residues I nd that 2 i a7 a rc 2 + 2 c c a2 + r c 2 c c

L = mRe

rc 2 c c +

a2 r c 2 + c c
2

rc 2 + 2 c c
5

a4 + rc 2 c c

2 rc 2 c c rc 2 c c
3

i c rc 2 c c

+ 3 a 8 rc 2 + c c
2

a4 + rc 2 c c
3

i a6 c + c rc 2 c c rc 2 c c + a2 + rc 2 c c

U
3

a2 + r c 2 c c V a2 + r c 2 c c

a6 c + c rc 2 c c rc 2 c c a2 + rc 2 c c

. (3.169)

95 Taking the real part, I nd that rc 2 + 2 c c a4 + rc 2 + c c


2 5 2

L=m

2 rc 2 + c c rc 2 c c rc 2 c c rc 2 c c + 1 + a2 rc 2 c c
3

+ 3 a 8 rc 2 + c c a4 + rc 2 + c c

y x
2

+ 3 a 8 rc 2 + c c a4 + rc 2 + c c

x y
2

1 y U a2 + rc 2 c c rc 2 + c c 1 1 x V + 1 + a2 2 2+ rc a r c 2 + c c c c +

. (3.170)

By setting d (Px + iPy ) = X + iY dt d L = M dt (3.171) (3.172)

and using Equations (3.156) and (3.160) and (3.167) and (3.170), I can solve for (U , V , ) in terms of (U, V, ) and known quantities such as the prescribed deformation of the foil and the state of the wake vortices. I can then nd the position (x, y, ) of the z-frame attached to the foil, with respect to a stationary world frame, by observing that x cos() sin() y = sin() cos() 0 0 cos() sin() x y = sin() cos() 0 0 U (3.173) 0 V 1 U sin() cos() 0 U 0 0 V cos() sin() 0 V 0 0 0 1 0

96
4 4 4 4 3 3 3 3

4 4 4

4 4 4

4 4 4

4 4 4

4 4

4 4

4 4

4 4

Figure 3.15: A foil starting from rest executes a turn by making the deformation (x , y ) = (0.15, 0.3 sin(t)). and integrating as necessary. It is important to note that our force and moment equations were derived in an inertial frame instantaneously aligned with the rotating z-coordinate frame, not in the rotating frame itself. Eorts to nd useful self-propulsion inputs for an isolated deformable foil are ongoing, but I can make some preliminary observations. For typical motions of the deformable foil, the angular accelerations that are generated by oscillating shape deformations appear large relative to the linear accelerations, so it is easy to discover inputs which yield large turns with a small amount of displacement5 , such as the turn illustrated in Figure 3.15. It is not so easy to generate the converse: steady swimming in a particular direction while keeping the orientation within tight bounds. In further study of the issue, it may be useful to consider nonuniform mass distributions (z, z). The classic vertical silhouette of a sh with wide body, narrow peduncle, and wide tail means that when the planar silhouette is considered, the forward body and the tail have more inertia (both in actual mass and in added mass) that may be fairly represented by the uniform treatment I have tried in this section.
Of course, the possibility of such maneuvers is part of the motivation for studying shlike swimming [ACS+ 97, Wei72].
5

97
10 10 10 8 8 8

10 10

10

10 10

10

10 10

10

Figure 3.16: The continued progress of the oscillating foil from Figure 3.15. When the foil drives through its own wake, I assume that any vortices which come too close to the foil are annihilated. To further stabilize the orientation, it may help to appeal to empirical dissipative or damping forces which do not arise naturally in the ideal uid model. When the foil executes general maneuvers in the uid, it is possible for previously shed point vortices to come near the surface of the foil, as can be seen in Figure 3.15. In principle, if the vortex comes too close to the surface, it can attain very high velocities (because it is under powerful inuence from its own counter-rotating image vortex), causing numerical diculties in the integration of the simulation equations, but in reality such a vortex would probably be destroyed by the action of viscous eects [Sar75]. In simulations of ow over an inclined at plate, Sarpkaya simply removed from the calculation any vortices which came closer than 0.1a to the surface of the plate, justifying the distance 0.1a in part by reference to the likely radius of the vortices in a real ow. I established a similar zone around the deformable foil when carrying out the simulation illustrated in Figure 3.16.

98

Chapter 4
4.1

Robot Carangiform Fish

Review of Carangiform Swimming

There are a wide variety of sh morphologies and at least a few dierent types of sh locomotion. This chapter focuses on a robot model of carangiform1 shes, fastswimming shes which resemble tuna and mackerel. Carangiform shes typically have large, high-aspect-ratio tails, and they swim using primarily motions of the rear and tail, while the forward part of the body remains relatively immobile. 2 To leading order, the geometry of carangiform swimming and the forces related to propulsion can be described as follows. First, I can roughly idealize the main body of the sh as a rigid body. The body is connected to the tail by a pedunclea slender region of generally negligible hydrodynamic inuence. Though three-dimensional eects may be important for some sh maneuvers, for purposes of modelling the gross thrust generation process, I simplify the geometry to the two dimensions of the horizontal plane, and our robot model moves in the plane. As the tail moves, nonzero circulation is generated in the surrounding uid so that the Kutta-Joukowski condition is satised. This vorticity is shed into the shs wake, and the pattern of vortices left behind by the passing of the sh is roughly a reverse Karman vortex street (Figure 4.1). This shed vorticity is a source of energy loss [Lig75]. Biological studies [AHB+ 91] suggest that apping motions of the sh tails are optimized to recapture some of the energy lost in the wake. A vortex is shed near the extremum of the tails sideways motion. The tail motion as it reverses direction can then be roughly interpreted as pushing o of the shed vortex. As a
Latin for mackerel-shaped. Some researchers use the additional category thunniform (Latin for tuna-shaped) to refer to sh in which propulsive force is even more concentrated in a high-aspect-ratio tail, which is often lunate in shape. In their terminology our robot might resemble a thunniform rather than carangiform sh. However, I will use the more general term, since the distinction is primarily one of degree, and our experimental tailn was typically a rectangular at plate rather than lunate.
2 1

99
Peduncle
Plane of 2D approximation

Shed Vortices

Figure 4.1: Schematic (side and top view) of Carangiform sh propulsion. benet, the counterrotating vortex generated by this secondary stroke helps to cancel the primary shed vortex, making the wake less detectable by predators. The shed vorticity may also inuence the stability of the shs motion, though there have not yet been signicant studies of this eect. Finally, a uid boundary layer along the shs body induces drag, and sheds vorticity into the shs wake. There is some evidence that a shs geometry and tails motions may be adapted to recapture some of this lost energy, thereby improving eciency. In general, sh tails have a degree of exibility,3 but the caudal ns of carangiform swimmers are quite sti and therefore it is a reasonable simplication to treat the tail as a rigid lifting surface. Figure 4.2 gives an idea of how carangiform swimming can generate thrust without the benet of a exible tailn. The tail of a sh is shown pitching and heaving up and down as the sh moves from right to left. The basic idea is that the tail maintains a negative angle of attack on the upstroke, and a positive angle of attack on the downstroke, with the result that the lift force on the hydrofoil/tail is always oriented so as to propel the sh forward [Lig75].

100

Figure 4.2: A cartoon of carangiform swimming consisting of snapshots of the tail as the sh swims from left to right. The arrows indicate lift forces acting on the tail. The sh body is removed for clarity.

peduncle

tailfin

Figure 4.3: A two-jointed sh model. Our experiment replaces the three links with at plates.

101

4.2

Description of the Experiment

My colleagues and I built an experimental sh as a test-bed for our models. The experiment is intended to resemble an idealized carangiform sh that consists of only three links: a rigid body in front, a large wing-like tail at the rear, and a slender stem, or peduncle, which connects the two. The three rigid links are connected by rotational joints with joint angles 1 and 2 . See Figure 4.3. I continue to idealize the model by supposing that I can neglect three-dimensional eects and regard the problem as essentially planar. Figures 4.4 and 4.5 show schematic diagrams and photographs of the side and top views of the apparatus, while Figure 4.6 shows a photograph from beneath of an earlier prototype. The tailn is a thin at plate with a chord of 15.2 cm and a typical depth in the water of 38 cm. The peduncle is a thin supporting arm, 13 cm in length, which I believe experiences only negligible hydrodynamic forces. The body is a thick at plate, intended to provide a degree of rotational inertia and rotational damping to help stabilize the sh during planar swimming. The entire sh is suspended in a 4-foot-wide by 4-foot-deep by 36-foot-long water tank from a passive gantry-like multi-degree-of-freedom carriage. The supporting infrastructure consists of two orthogonal sets of rails and a rotating platform, all supported on low friction bearings. By apping its tail, this mechanism allows the sh to propel itself and its supporting carriage around the tank. The frictional drag on the rails is suciently low that this carriage system is a reasonable approximation to untethered swimming. The system can move with three degrees of freedom in the plane: forward, sideways, and rotationally. Furthermore, the gantry suspension allows buoyancy eects to be ignored, thereby keeping the experiment focus on thrust generation and maneuvering. The carriage also simplies the experiment by keeping the motors, electronics, etc. out of the water. The tail and peduncle degrees of freedom are independently driven by two DC motors (Escap model 35 NT2 R82), each of which is capable of 75 W of power and
3

Flexible tails might be considered deformable foils, as in Chapter 3.

102

Polhemus sensor linear bearings

coaxial cable shafts drive motors encoders rail 0.13 m

body

tailfin

0.434 m

0.381 m

0.442 m peduncle

0.152 m

Figure 4.4: Side view schematic and photograph of experiment.

Figure 4.5: Top view schematic and rear top photo of experiment.

           

                  

        

motors

103

rails

peduncle tailfin

104

Figure 4.6: Bottom view of an earlier version of experiment. 110 mNm of torque. Timing belts and two coaxial shafts transmit power from the motors to the submerged joints of the sh. Also mounted on the carriage are optical shaft encoders (Hewlett-Packard HEDS-5500), which record the shs joint angles at each instant and enable feedback control of the tail. The encoders are accurate to within about forty minutes of arc. Finally, a Polhemus position/orientation sensor is rigidly attached to the last stage of the carriage. In this way, the absolute position and orientation of the sh can be determined. Since every aspect of the shs movement is instrumented, I may accurately compare experimental results with theory for the purposes of assessing the validity of simplied models. For purposes of determining the center of mass or center of rotation of the system, the gantry carriage elements, which do not rotate with the submerged links but are connected to the three-link sh by a rotational bearing, can be regarded as a point mass located at the central axis of the rotational bearing. Since this mass is a signicant fraction of the actual mass (if not the added mass) of the system, the location of the rotational bearing has a substantial impact on the location of the systems center of rotation. In an earlier version of the apparatus, the rotational bearing axis was located relatively far back on the body link and close to the peduncle. This setup could have been taken as equivalent to a free-oating swimmer with a short body or unusually massive hindquarters. This conguration had poor yaw stability.

105
vb lb lp vqc lp lb 1 2 lt le

le

Figure 4.7: The idealized model. Subsequently the experiment was modied to move the rotational axis forward relative to the body link. This resulted in dynamics more nearly comparable to those of a biological sh or a plausible free-swimming robot, and also improved the passive weathervane stability of the system with respect to yaw. The mass of the entire robot and gantry system is m = 35 kg, and the moment of inertia of the body with the tail fully extended is I = 0.5038 kg m2 .

4.3

Model with Added Mass and Quasi-Steady Lift

In an eort to model the three-link experimental system in a relatively simple way, I treated the rst link (body) as a source of drag, to be calculated empirically from measurements; the second link (peduncle) as hydrodynamically negligible; and the third link (tailn) as a at plate, generating lift and drag according to quasi-static two-dimensional wing theory. Figure 4.7 shows the geometry obtained by further idealizing the system in Figure 4.1. Let lb be the distance between the bodys center of rotation and the location of the body/peduncle connector. Let the total length of the body be l. The peduncle has length lp . The tailn has chord lf and span d. Let le be a unit vector pointing in the direction of the leading edge of the tailn hydrofoil. cos() sin() 0 cos(2 ) le = sin() cos() 0 sin(2 ) , 0 0 1 0

(4.1)

where is the orientation of the body in an inertial frame and 1 , 2 are joint angles with respect to the bodys longitudinal axis. Using the Kutta-Joukowski theorem and

106 assuming that the tail hydrofoil is in a quasi-steady uniform ow with the velocity implied by the instantaneous velocity of the foils quarter-chord point, I arrive at the following lift force on the hydrofoil: L = f lf d(vqc le ) vqc , (4.2)

where f is the density of the uid, lf is the chord of the tailn plate, d is depth or span, and vqc is the velocity of the quarter-chord point of the tailn.4 From Lanchester-Prandtl wing theory, I estimate the drag on the tailn at [AD59]: Df =
2 2f lf

vqc

vqc

(vqc le )2 vqc

(4.3)

Meanwhile the quasi-static torque generated around the midpoint of the tail is: f = f d
2 lf 4

xm ym cos(2 + 22 ) +

(ym x2 ) 2 m sin(2 + 22 ) , 2

(4.4)

where (xm , ym ) is the velocity of the tails midpoint. The body of the sh has instantaneous translational velocity vb = (x, y, 0); it also has instantaneous rotational velocity . The velocity of the hydrofoils quarter-chord point and midpoint are respectively: x lb sin() lp ( + 1 ) sin( + 1 ) (lf /4)( + 2 ) sin( + 2 )

vqc

x m ym 0
4

= y + lb cos() + lp ( + 1 ) cos( + 1 ) + (lf /4)( + 2 ) cos( + 2 ) 0 x lb sin() lp ( + 1 ) sin( + 1 ) (lf /2)( + 2 ) sin( + 2 ) = y + lb cos() + lp ( + 1 ) cos( + 1 ) + (lf /2)( + 2 ) cos( + 2 ) 0

The drag acting on the body of the sh was empirically measured by towing
This implicitly assigns the wing a coecient of lift CL = 2 sin(). Then the drag estimate 2 CL follows from the relationship CD = (d/lf ) , where d/lf is the aspect ratio of the tailn.

107

=0 18

16
2

14

Drag = 17.81 U 0.16 U + 0.06 C = 0.25


D

12 Drag force (N)

10

0.1

0.2

0.3

0.4

0.5 0.6 Velocity U (m/s)

0.7

0.8

0.9

Figure 4.8: Drag force acting on the sh in forward motion.

10

8 C = 2.5 2 + 1.2 + 0.1


D

6 C (Ns2/m2)

0.2

0.4

0.6

0.8 1 Angle of attack (rad)

1.2

1.4

1.6

1.8

Figure 4.9: Drag force acting on the sh at varied angles of attack.

108 the apparatus through the water with a variety of falling weights and measuring the terminal velocity reached for a given amount of applied force. The drag force was found to be approximately quadratic with velocity. Figure 4.8 shows the drag force on the body moving straight forward at zero angle of attack. The drag, about 17.8 Ns2 /m2 , can be normalized by the water density and the area of the body: 1 D = C D f l d U 2 , 2 (4.5)

where f = 1000 kg/m3 , l = 46 cm, and d = 31 cm, to yield a normalized coecient of drag CD = 0.25 at = 0. However, the measured drag includes not only the various forms of hydrodynamical drag, but also any resistance due to the rail bearings. Figure 4.9 shows the experimentally measured coecient of drag CD for the sh body at a variety of dierent angles of attack . Dissipative forces acting against rotation were not measured explicitly, but they can be estimated if I suppose that the measured coecients of drag CD can be applied to incremental elements of the plate, each of which moves with a dierent velocity as the body rotates. The total drag force acting on the body is 1 Db = f d 2 1 = f d 2 where x sin() s
lb

CD v(s)
lb l lb lb l

v(s) ds v(s) (4.6)

CD v(s) v(s) ds,

v(s) =

y + cos() s

(4.7)

109 The total drag moment acting on the body is b 1 = f d 2 1 = f d 2


lb

CD v(s)
lb l lb lb l

2 (s cos(), s sin())

v(s)

v(s)

ds (4.8)

CD v(s) (cos()y sin()x + s) s ds.

Finally, I consider the added inertia from the uid surrounding the body. The planar added mass coecients for a at plate of length l in a frame aligned with the plate at its midpoint are5 I11 = 0 I12 = 0 I13 = 0 I22 = f d(l/2)2 I23 = 0 I33 = 2f d(l/4)4 . (4.9) (4.10) (4.11) (4.12) (4.13) (4.14)

In another frame whose origin is located at (, ) in the plate frame, and rotated by angle , the transformed added mass coecients are [Sed65] I11 = I11 cos2 () + I22 sin2 () + I12 sin(2) 1 I12 = (I22 I11 ) sin(2) + I12 cos(2) 2 I13 = (I11 I12 + I13 ) cos() + (I12 I22 + I23 ) sin() I22 = I11 sin2 () + I22 cos2 () I12 sin(2) I23 = (I11 I12 + I13 ) sin() + (I12 I22 + I23 ) cos() I33 = I11 2 + I22 2 2I12 + 2(I13 I23 ) + I33 .
5

(4.15)

These can be obtained from the results for the locked added inertia of a Joukowski foil, presented in Section 3.3, with the substitutions c = = 0 and rc = a = (l/4).

110 So at the center of rotation of the body, the added inertia due to the body plate is body I11 = f d(l/2)2 sin2 () 1 body I12 = f d(l/2)2 sin(2) 2 body I13 = 0 body I22 = f d(l/2)2 cos2 () body I23 = 0 body I33 = 2f d(l/4)4 , while the added inertia due to the tailn plate is tail = d(l /2)2 sin2 ( + ) I11 f f 2 tail = 1 d(l /2)2 sin(2 + 2 ) I12 f f 2 2 tail = d(l /2)2 sin( + ) [(x x ) cos( + ) + (y y ) sin( + )] I13 f f 2 m 2 m 2 tail = d(l /2)2 cos2 ( + ) I22 f f 2 tail = d(l /2)2 cos( + ) [(x x ) cos( + ) + (y y ) sin( + )] I23 f f 2 m 2 m 2 +(x xm )(y ym ) sin(2 + 22 )] + 2f d(lf /4)4 , where the tailn midpoint (xm , ym ) is xm ym x + lb cos() + lp cos( + 1 ) + (lf /2) cos( + 2 ) y + lb sin() + lp sin( + 1 ) + (lf /2) sin( + 2 ) (4.17) (4.16)

tail = d(l /2)2 (x x )2 cos2 ( + ) + (y y )2 sin2 ( + ) I33 f f m 2 m 2

(4.18)

These virtual inertias are added to the ordinary inertia of the system. m 0 0 I = 0 m 0 0 0 I

(4.19)

111 I total = I + I body + I tail (4.20)

Finally, the tailn sees an additional added mass force and moment fam and am due to the acceleration component of the tailn which is independent of (, y , ). x lf 2
2

fam = f d

sin ( + 2 ) 1 2 sin(2 + 22 )

1 2

sin(2 + 22 )
2

cos ( + 2 )

xm ym

(4.21)

2 xm = (lb cos() lp cos( + 1 ) (lf /2) cos( + 2 ))2 lp cos( + 1 )1 2 ym = (lb sin() lp sin( + 1 ) (lf /2) sin( + 2 ))2 lp sin( + 1 )1 (lf /2) sin( + 2 )2 + lp cos( + 1 )1 + (lf /2) cos( + 2 )2
2

2 (lf /2) cos( + 2 )2 lp sin( + 1 )1 (lf /2) sin( + 2 )2

(4.22)

(4.23)

am = 2f d(lf /4)4 2 At last, then, the equation of motion for the system is x L + Df + Db + fam I total y = f + b + am + (xm x, ym y) (L + fam )

(4.24)

(4.25)

Some things neglected in this model: By taking a quasi-static approximation to the lift force, I am disregarding any special spatial structure of the wakeroughly speaking, I am treating any vorticity shed from the tailn as if it were swept away and immediately became very distant. The actual, as opposed to virtual, mass of the tailn is considered small enough that I neglect any changes in I due to changes in joint angle. I treat the bodies in isolation and ignore any hydrodynamical interactions between them, as well as any forces whatsoever acting on the peduncle. I also neglect some mundane details of the experimental apparatus: I ignore the presence of the tank walls, surface wave eects, and any stiction or resistance from the gantry bearings, except insofar as they may have inuenced our empirical measurement of drag forces.

112
Experiment: = 0.6 sin(4 t), = 0.3 sin(4 t /3)
1 2

0.8

desired 1 2 desired 1 actual actual


2

0.6

0.4

0.2 radians

0.2

0.4

0.6

0.8

3 t (s)
1 2

Experiment vs. Simulation: = 0.6 sin(4 t), = 0.3 sin(4 t /3) 2

1.8

Experiment Simulation

1.6

1.4

distance (meters)

1.2

0.8

0.6

0.4

0.2

3 t (s)

Figure 4.10: Experiment and simulation for the forward gait with 1 = 0.6 sin(4.0t), 2 = 0.3 sin(4.0t + /3). A PD controller drives the joints to follow the desired trajectory. The simulation slightly overpredicts thrust (or underpredicts drag.)

113
Thrust: 1 = 0.6 sin(4 t), 2 = 0.3 sin(4 t /3) 30 tail lift tail drag body drag tail added mass

25

20

15 Newtons

10

10

3 t
Early Thrust Phase

30 tail lift tail drag body drag tail added mass

25

20

15 Newtons

10

10

0.1

0.2

0.3 t

0.4

0.5

0.6

0.7

Figure 4.11: Forward thrust generated by the gait 1 = 0.6 sin(4.0t), 2 = 0.3 sin(4.0t + /3), from Figure 4.10. Almost immediately the sh enters a periodic cycle where lift is the dominant source of thrust, but added mass forces on the tailn are signicant at the very start of thrust.

114

Experiment: 1 = 1.2 sin(2 t), 2 = 0.6 sin(2 t 5/18) 1.5 1 desired desired 2 actual 1 2 actual

0.5

radians

0.5

1.5

3 t (s)
1 2

Experiment: = 1.2 sin(2 t), = 0.6 sin(2 t 5/18) 2 Experiment Simulation 1.8

1.6

1.4

distance (meters)

1.2

0.8

0.6

0.4

0.2

3 t (s)
2

Thrust: = 1.2 sin(2 t), = 0.6 sin(2 t 5/18)


1

60 tail lift tail drag body drag tail added mass

50

40

30

20

10

10

10

Figure 4.12: Experiment and simulation for the gait 1 = 1.2 sin(2.0t), 2 = 0.6 sin(2.0t + 5/18). Both lift forces and added mass forces are signicant components of the thrust.

115 Despite these omissions, this model does a fairly good job of describing the qualitative behavior of the system. To compare the experiment to the simulation model, the computer simulation received as input the actual joint angle trajectories for each gait as reported by the optical shaft encoders. Based on these inputs, the model equations of motion were integrated by a Runge-Kutta method, and the simulated shs displacement is superimposed on the analogous graph of experimental data. I carried out these comparisons for a large number of forward-swimming gaits with inputs sinusoidal in 1 and 2 . Examples are shown in Figures 4.10 and 4.12. See also Figures 4.17-4.18. For most of these large-amplitude gaits, lift forces acting on the tailn are the predominant source of thrust, except for the very beginning of motion when added mass forces on the tailn are signicant. See Figure 4.11. Occasionally both lift forces and added mass forces will be signicant in the steady state, as in Figure 4.12. These large-amplitude gaits would have generated large angular oscillations, if I had allowed it: for these gaits, the sh was only allowed to swim forward longitudinally. The lateral and yaw degrees of freedom were removed, so stability of the gait with respect to those degrees of freedom was not an issue. The practical utility of this propulsion in one dimension is that it is the limiting case of a biomimetic swimmer with a large moment of inertia, e.g., a boat with a apping propulsor small relative to the boat. These gaits could also be used by a set of paired propulsors, or by a biomimetic swimmer with some other means of stabilization. Our sh robot as currently congured, however, does not have a large enough moment of inertia to use these gaits eectively in full planar motion. Instead, to propel itself in the plane while keeping a stable heading, the robot sh must use small-amplitude high-frequency strokes. Presumably this will also hold true for future small free-swimming robots. A successful gait for forward propulsion of a small robot that is free to rotate or deviate from its heading is illustrated in Figure 4.13. The inputs shown are 1 = 0.4 sin(8t), 2 = 0.4 sin(8t + /3). Note that I have centered the heading around = so that positive motion in the x-direction is forward motion. The achievable speeds are notably lower than for the large-amplitude

116
THRUST (x direction) 70 tail lift tail drag body drag tail added mass

60

50

40

Newtons

30

20

10

10

20

5 t (seconds)

10

1 = 0.4 sin(8 t), 2 = 0.4 sin(8t /3) 3.5

2.5 x y

1.5

0.5

0.5

5 t (seconds)

10

Figure 4.13: A gait which results in stable forward propulsion even when the sh is free to rotate or deviate. Note that added mass forces dominate lift and drag forces. Since is near , the forward direction is along the positive x-axis.

117

LATERAL FORCE (y direction) 200 tail lift tail drag body drag tail added mass

150

100

Newtons

50

50

100

150

5 t (seconds)

10

MOMENTS 60

40

20

Newtons

20

40 tail lift tail drag body drag tail added mass

60

80

5 t (seconds)

10

Figure 4.14: Lateral forces and moment generated by the gait in Figure 4.13.

118 gaits constrained to one dimension. Interestingly, for this free-swimming gait, added mass forces, not lift forces, are the predominant source of thrust. Also, added mass forces play a vital role in turning maneuvers. Morgansen

[MDM+ 01] suggested the turning gait illustrated in Figure 4.15 with 1 = 0.4 sin(3.5t) and 2 = 0.4 sin(7t). The reasonably good validation of the model by experiment suggests that this simple set of ODEs can be used as a design tool in biomimetic swimming without necessarily having to resort to computationally intensive uid models. Perhaps most importantly, Morgansen, Duindam, Vela, and Burdick [MDM+ 01, MVB02] were able to use this platform and model to perform feedback control experiments: to our knowledge the rst instance of a robot sh maintaining a trajectory by feedback control.

4.4

Miscellaneous Experimental Observations

I experimented with a large number of possible joint trajectories, especially sinusoidal inputs of varying amplitude, frequency, and phase, to discover the best gaits, which I usually dened as being those with the highest acceleration and forward speed, although I was also interested to note especially smooth or jerky gaits, and later gaits with turning behavior. Figures 4.174.18 show steady-state velocities for some of the gaits tested. Triantafyllou et al. [TTG93] emphasized the importance of the dimensionless Strouhal number, dened as St =
fA , V

where f is the frequency of the tailn oscillation,

A is the double amplitude of the tail-to-tip excursion, and V is the average forward swimming velocity. Triantafyllou et al. advanced arguments that the Strouhal number of an oscillating foil system should be in the range 0.250.35 for optimum thrust. They also cited biological observations suggesting that a wide range of sh and cetaceans actually do operate in this range. In Figure 4.20, I plot the Strouhal number for the same set of sinusoidal gaits represented in Figure 4.19. It will be seen that the Strouhal number for our system is precisely in the optimum range, particularly for

119
MOMENTS 80 tail lift tail drag body drag tail added mass

60

40

20

Newtons

20

40

60

80

100

5 t (seconds)

10

1 = 0.4 cos(3.5 t), 2 = 0.4 cos(7 t) 1

0.5

0.5 x y 1

1.5

5 t (seconds)

10

Figure 4.15: A turning gait. Note that added mass eects dominate the moment. Since starts near zero, the forward direction is initially aligned with the negative x-axis.

120

THRUST (x direction) 150 tail lift tail drag body drag tail added mass 100

50

Newtons

50

100

150

5 t (seconds)

10

LATERAL FORCE (y direction) 200 tail lift tail drag body drag tail added mass

150

100

50

Newtons

50

100

150

200

250

5 t (seconds)

10

Figure 4.16: Forces experienced in the turn of Figure 4.15. The x and y forces are in a world frame, not xed in the body of the sh.

121 the fastest gaits. (Note that the horizontal axes in Figures 4.20 and 4.19 are reversed, to increase visibility of the surfaces.) This tends to further support Triantafyllous observations and also suggests that our system is operating in a uid mechanical regime fairly representative of biological sh swimming. After identifying successful gaits for forward swimming, we began to experiment with unconstrained maneuvers such as turns. Some early turning gaits were found by trial and error (see Figure 4.21), and later some were derived more formally from the model [MDM+ 01] (see Figure 4.15.) However to date, turning maneuvers are relatively awkward and disappointing relative to the dramatic turning performance of biological sh [Wei72], although the robots ability to turn in place may be competitive relative to other articial vehicles of similar power and complexity. Biological sh curve their whole spines during turns, including signicant motion of the head, and it might be that a three-link model with rigid head and body is insucient to properly capture this aspect of sh swimming. Changes in the strokes range of motion to cause turns should be synchronized to the periodic motion of the tail, so probably the tail should only be re-oriented to a new range of motion in between propulsive tail beats. This suggests a planning method. If a variety of strokes are each known, through simulation or experiment, to produce a given impulse and angular impulse per beat, then the sh could choose individual beats as necessary to add a quantized amount of momentum or angular momentum. Maneuvers could then be built up of series of such individual strokes. We could draw an analogy between these quantized strokes and the impulse bits of thrusters used to orient spacecraft.

122

Figure 4.17: Steady-state velocities for single-frequency sinusoidal gaits. In each subplot, frequency and peduncle amplitude are held constant while steady-state velocity

123

Figure 4.18: Steady-state velocities of another set of gaits with higher amplitude of peduncle oscillation.

3.5 3

Distance (m)

2.5 2 1.5 1 0.5 20 30 40 50 60 70 90 70 80 60 50 40

Tailfin Amplitude (deg)

Phase Lag (deg)

Figure 4.19: Distance traveled by the sh in eight seconds, for gaits of the form 1 = 1.3 sin(3.5t) and 2 = A sin(3.5t + ). The horizontal axes represent the parameters A and .

124

1.4 1.2

Strouhal Number

1 0.8 0.6 0.4 0.2 0 70 60 50 40 30 60 50 20 40 70 80 90

Tailfin Amplitude (deg)

Phase Lag (deg)

Figure 4.20: Strouhal numbers, for gaits of the form 1 = 1.3 sin(3.5t) and 2 = A sin(3.5t + ). The horizontal axes represent the parameters A and .

20 0 20 Heading (deg) 40 60 80 100 120 140 0 2 4 Time (s) 6

=60 deg =70 deg =90 deg

Figure 4.21: Orientation change over time, for gaits of the form 1 = 0.6 + 1.3 sin(3.5t) and 2 = 1.1 sin(3.5t + ).

125

Chapter 5

Optimal Control

As was illustrated in Figures 4.13 and Figures 4.15, our experience with the robot carangiform sh indicates that for orientational stability, small biomimetic swimmers may have to operate in a low-speed, high-frequency regime where added mass forces dominate lift and drag forces. Turns and low-speed maneuvers, the operations where shlike vehicles are most likely to outperform conventional watercraft, also seem to fall into this regime. With this in mind, in this chapter I will look at the equations of motion for the three-link sh if only added mass eects are retained and lift and drag forces are neglected, and briey consider how trajectories under these equations of motion could be designed using optimal control. If lift and drag forces are assumed negligible relative to added mass forces, the carangiform shs equation of motion becomes: x fam , I total y = am + (xm x, ym y) fam where fam , am , xm , ym , and I total are as dened in Equations (4.18)-(4.24). Taking the state of the system to be specied by q = (x, y, , x, y, , 1 , 2 , 1 , 2 ) and regarding the joint accelerations as inputs: u1 = 1 u2 = 2 (5.3) (5.4) (5.2)

(5.1)

126 I nd that the state equation of the system is d (x, y, , x, y, , 1 , 2 , 1 , 2 ) = f T = (x, y, , f4 , f5 , f6 , 1 , 2 , u1 , u2 )T , dt where: f6 = = {lf 2 f 512 lp 2 m2 1 128 l2 lp 2 m f 1 288 lf 2 m2 2 72 l2 lf 2 m f 2 (5.5)

8 lf 4 m f 2 l2 lf 4 2 f 2 2 128 lb lp m 4 m + l2 f 1 cos(1 ) 128 lb lp m 4 m + l2 f 1 cos(1 2 2 ) 512 lf lp m2 1 cos(1 2 ) 128 l2 lf lp m f 1 cos(1 2 ) 512 lf lp m2 2 cos(1 2 ) 128 l2 lf lp m f 2 cos(1 2 ) 512 lp 2 m2 1 cos(2 (1 2 )) 128 l2 lp 2 m f 1 cos(2 (1 2 )) 512 lb lf m2 2 cos(2 ) 128 l2 lb lf m f 2 cos(2 ) + l2 lf 4 2 f 2 2 cos(2 2 ) +512 lb lp m2 2 sin(1 ) + 128 l2 lb lp m 2 f sin(1 ) 1 1 +1024 lb lp m2 1 sin(1 ) + 256 l2 lb lp m 1 f sin(1 ) +512 lb lp m2 2 sin(1 2 2 ) + 128 l2 lb lp m 2 f sin(1 2 2 ) 1 1 +1024 lb lp m2 1 sin(1 2 2 ) + 256 l2 lb lp m 1 f sin(1 2 2 ) +1024 lb lp m2 2 sin(1 2 2 ) + 256 l2 lb lp m f 2 sin(1 2 2 ) +512 lf lp m2 2 sin(1 2 ) + 128 l2 lf lp m 2 f sin(1 2 ) 1 1 +1024 lf lp m2 1 sin(1 2 ) + 256 l2 lf lp m 1 f sin(1 2 ) +512 lf lp m2 2 sin(1 2 ) + 128 l2 lf lp m f 2 sin(1 2 ) +512 lp 2 m2 2 sin(2 (1 2 )) + 128 l2 lp 2 m 2 f sin(2 (1 2 )) 1 1 +1024 lp 2 m2 1 sin(2 (1 2 )) + 256 l2 lp 2 m 1 f sin(2 (1 2 )) +512 lp 2 m2 2 sin(2 (1 2 )) + 128 l2 lp 2 m f 2 sin(2 (1 2 )) 512 lb lf m2 2 sin(2 ) 128 l2 lb lf m f 2 sin(2 ) 512 lb 2 m2 2 sin(2 2 ) 128 l2 lb 2 m f 2 sin(2 2 ) }/{} (5.6)

127 and: f4 = x = {lf 2 f 128 I lf 2 + l4 lf f 2 lf 5 f 2 + lf 5 2 f 2 2 l 4 1 + l f 4 f 1 2 cos(1 2 ) 2 lb lf 4 f 2 cos(2 )

+2 lp 128 I 1 + f

256 I lp 2 sin(1 2 ) 2 l4 lp 2 f sin(1 2 ) 2 lf 4 lp 2 2 f 2 sin(1 2 ) 1 1 1 512 I lp 1 sin(1 2 ) 4 l4 lp 1 f sin(1 2 ) 4 lf 4 lp 1 2 f 2 sin(1 2 ) 256 I lp 2 sin(1 2 ) 2 l4 lp f 2 sin(1 2 ) 2 lf 4 lp 2 f 2 2 sin(1 2 ) +256 I lb 2 sin(2 ) + 2 l4 lb f 2 sin(2 ) + 2 lb lf 4 2 f 2 2 sin(2 ) 2 4 m + l2 f cos() sin(2 ) + 8 m cos(2 ) sin() }/({2 }) (5.7) and: f5 = y = { lf 2 f 128 I lf 2 + l4 lf f 2 lf 5 f 2 + lf 5 2 f 2 2 l 4 1 + l f 4 f 1 2 cos(1 2 )

+2 lp 128 I 1 + f

2 lb lf 4 f 2 cos(2 ) 256 I lp 2 sin(1 2 ) 2 l4 lp 2 f sin(1 2 ) 1 1 2 lf 4 lp 2 2 f 2 sin(1 2 ) 512 I lp 1 sin(1 2 ) 4 l4 lp 1 f sin(1 2 ) 1 4 lf 4 lp 1 2 f 2 sin(1 2 ) 256 I lp 2 sin(1 2 ) 2 l4 lp f 2 sin(1 2 ) 2 lf 4 lp 2 f 2 2 sin(1 2 ) + 256 I lb 2 sin(2 ) +2 l4 lb f 2 sin(2 ) + 2 lb lf 4 2 f 2 2 sin(2 ) 8 m cos(2 ) cos() 2 4 m + l2 f sin(2 ) sin() }/{2 } (5.8)

128 and: = 4096 I m2 + 1024 I l2 m f + 1024 I lf 2 m f + 32 l4 m2 f + 512 lb 2 lf 2 m2 f + 256 lf 4 m2 f + 512 lf 2 lp 2 m2 f + 128 I l2 lf 2 2 f 2 + 8 l6 m 2 f 2 + 8 l4 lf 2 m 2 f 2 + 128 l2 lb 2 lf 2 m 2 f 2 + 64 l2 lf 4 m 2 f 2 + 128 l2 lf 2 lp 2 m 2 f 2 + 32 lf 4 m2 2 f 2 + l6 lf 2 3 f 3 + 8 l2 lf 4 m 3 f 3 + 8 lf 6 m 3 f 3 + l2 lf 6 4 f 4 + 256 lb lf 2 lp m f 4 m + l2 f cos(1 ) + 256 lb lf 2 lp m f 4 m + l2 f cos(1 2 2 ) + 1024 lf 3 lp m2 f cos(1 2 ) + 256 l2 lf 3 lp m 2 f 2 cos(1 2 ) + 512 lf 2 lp 2 m2 f cos(2 (1 2 )) + 128 l2 lf 2 lp 2 m 2 f 2 cos(2 (1 2 )) + 1024 lb lf 3 m2 f cos(2 ) +256 l2 lb lf 3 m 2 f 2 cos(2 )+512 lb 2 lf 2 m2 f cos(2 2 )128 I l2 lf 2 2 f 2 cos(2 2 ) + 128 l2 lb 2 lf 2 m 2 f 2 cos(2 2 ) l6 lf 2 3 f 3 cos(2 2 ) l2 lf 6 4 f 4 cos(2 2 ) (5.9) There are, of course, many ways in which optimality of a trajectory between two points could be dened or quantied by the engineer. For the sake of the current example, I will assume that the trajectories being sought are minimum-time trajectories, subject to the constraint that the joint accelerations are bounded: 1 2 2 2 (5.10) (5.11)

In the language of optimal control, then, I am seeking to minimize the performance index: J=
0 T

1 dt

(5.12)

and the appropriate Hamiltonian for the problem is [LS95]: H = 1 + T f (5.13)

129 The costate equations are given by H = q (5.14)

and these could readily be written in closed (if slightly cumbersome) form. I also have Pontryagins minimum principle: H(q , u , , t) H(q , u, ) (5.15)

which says that the optimal joint accelerations ui = i minimize the Hamiltonian, not just relative to alternative functions for the joint acceleration and the alternative positions and velocities resulting from the state equation, but also relative to other values for the acceleration while keeping the optimal position and velocity functions xed, consistent with the acceleration bounds but not consistent with the state equation. This implies that the joint accelerations must be at their maximum or minimum allowed limits at all times, unless the value of the joint accelerations has no inuence on the Hamiltonian H at all. H 1 = 2 sign( ) 1 x y + 5 + 6 ) 1 1 1 4 = 2 sign 9 + {lf 2 lp f 128 I + f l4 + lf 4 f cos(1 2 ) 2 2 4 m + l f cos() sin(2 ) + 8 m cos(2 ) sin() } 5 + { lf 2 lp f 128 I + f l4 + lf 4 f cos(1 2 ) 8 m cos(2 ) cos() 2 4 m + l2 f sin(2 ) sin() } 6 + {lf 2 f 512 lp 2 m2 128 l2 lp 2 m f 128 lb lp m 4 m + l2 f cos(1 ) 128 lb lp m 4 m + l2 f cos(1 2 2 ) 512 lf lp m2 cos(1 2 ) = 2 sign(9 + 4 128 l2 lf lp m f cos(1 2 ) 512 lp 2 m2 cos(2 (1 2 )) 128 l2 lp 2 m f cos(2 (1 2 )) } (5.16)

130

H ) 2 = 2 sign( 2 = 2 sign(10 + 4 x y + 5 + 6 ) 2 2 2

= 2 sign (10 4 + {lf 2 f 128 I lf + l4 lf f lf 5 f + lf 5 2 f 2 2 lf 4 lp f cos(1 2 ) 2 2 lb lf 4 f cos(2 ) 2 4 m + l2 f cos() sin(2 ) + 8 m cos(2 ) sin() } 5 + { lf 2 f 128 I lf + l4 lf f lf 5 f + lf 5 2 f 2 2 lf 4 lp f cos(1 2 ) 2 2 lb lf 4 f cos(2 ) 8 m cos(2 ) cos() 2 4 m + l2 f sin(2 ) sin() } 6 + {lf 2 f 288 lf 2 m2 72 l2 lf 2 m f 8 lf 4 m f l2 lf 4 2 f 2 512 lf lp m2 cos(1 2 ) 128 l2 lf lp m f cos(1 2 ) 512 lb lf m2 cos(2 ) 128 l2 lb lf m f cos(2 ) + l2 lf 4 2 f 2 cos(2 2 ) } Thus I can explore the space of congurations reachable in a given time by the sh by systematically choosing trial values of at t = 0, and shooting forward using the state and costate equations and Equations (5.16)(5.17) to arrive at some nal state at t = T . After shooting forward, we will have determined the minimum-time trajectory to whatever point I arrive at. In this way I could build up a catalog of reachable points. Of particular interest will be any pair of trajectories, one of which reaches nal values of 1 , 2 , 1 , 2 , , which are the same or approximately the same as the starting values for the other trajectory. These trajectories could be concatenated together to form a larger trajectory, since the equations of motion are invariant with respect to transformations in x, y, , and the values of x and y are not signicant either unless they become large enough to invalidate the assumption of added-mass dominance. Figure 5.1 shows one minimum-time maneuver found by numerical optimization of bang-bang trajectories, which could plausibly be concatenated with itself. Although I could have used a dierent model of the system, a dierent denition (5.17)

131

0.5

1 2 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.01

0.01

0.02

0.03

0.04

0.05 x y 0.06 0.1 dx/dt dy/dt d/dt 0.05 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

0.05

0.1

0.15

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Figure 5.1: A bang-bang maneuver, starting from rest, that produces both acceleration and displacement in the x-direction, while producing less acceleration and almost no displacement in the y- and directions. The joint accelerations (1 , 2 ) are (2, 2) from t = 0 to t = 0.29, then (2, 2) to t = 0.97 The values of 1 , 2 oscillate rapidly just before the end of the trajectory.

132 of optimality, or a dierent scheme for searching through the space of trajectories, the broader point is that in the lab, relatively ample computational power can be used to optimize the feasible trajectories of a swimmer over short time horizons. A suitable set of such feasible trajectories can then be used as motion primitives and pieced together to form larger trajectories. This will be the subject of the next chapter.

133

Chapter 6

Trajectory Planning Using

Density Functions
6.1 Motivation

Many mobile robots operate under nonholonomic or other kinodynamic constraints which complicate the task of planning their motions. In this chapter I present an algorithm for planning the gross motion of many types of mobile robots in the presence of such constraints. There is a large literature on the subject of nonholonomic and kinodynamic motion planning. Many approaches to this problem are based on a decomposition of the planning problem into two parts. In the rst step, the kinematic and dynamic constraints are ignored, and the gross path of the system that avoids obstacles is determined. This holonomic path can be chosen manually, or by a conventional rigid body holonomic motion planner. Next, the actual control inputs that respect the constraints are determined so that the system approximately tracks the holonomic path. Based on dierential geometric concepts, there are many local motion planners that compute the controls inputs which cause the vehicle to approximately track the specied trajectory (e.g., [LS93, MS93b, Gur92, MS91]). Implicitly, this approach assumes that the vehicle is small-time locally controllable (STLC), so that it is capable of locally approximating any trajectory. When this criterion is not satised, the success of these methods is uncertain. The successful probabilistic roadmap motion planning paradigm [KSLO96] has also been adapted to nonholonomic motion planning (e.g., [SO95]). Unfortunately, a local nonholonomic motion planning problem must be solved each time a candidate node is considered for addition to the roadmap. Lavalle and Kuner [LK00] have also developed a probabilistic kinodynamic planning approach that is based on an incre-

134 mental simulation of the systems dynamics. A forward simulation of the systems dynamics is used to expand a search tree. This approach circumvents the small time local controllability assumption of many other techniques, and has produced excellent simulation results. However, its probabilistic convergence may require an exponential number of computations. Here I introduce a dierent type of algorithm for approximately solving this class of motion planning problems. My deterministic approach is based on constructing an approximation to the vehicles reachable set of states. I use techniques of fast Fourier transforms on Lie groups to eciently compute these sets. This approach is motivated by the Ebert-Upho algorithm for solving the inverse kinematics of discretely actuated manipulators [CK01]. A solution path can then be constructed from knowledge of the reachable set. Like [LK00], this technique does not require small-time local controllability, and I show that the algorithm has an attractive computational complexity. Because I use a nite set approximation to the continuous mechanics, my technique can also be used to plan the motions of discrete nonholonomic systems [CMPB00]. Section 6.2 provides an intuitive overview of the approach, focusing on the vehicles reachable set of states. Sections 6.3 and 6.5 describe the density of reachable states concept, and algorithms for its computation. Section 6.6 then describes the planning method, and the algorithm is illustrated by an example. Another variation of the algorithm with tighter time and space bounds is summarized in Section 6.6.5. Finally, a third variation of the algorithm to be used in the presence of static obstacles is presented in Section 6.6.6.

6.2

Summary of the Approach

Most nonholonomically constrained wheeled vehicles move by the inuence of periodic or quasi-periodic controls. Hopping, walking, and swimming robots move by periodic oscillations of their driving actuators. A common theme among these systems is that their gross trajectories can be naturally broken into individual hops, steps, or

135
0.05

0.05

0.05

0.1

0.15

Figure 6.1: This sh robot can reach any of four positions in one tail stroke. undulations. The trajectories of such robots can therefore often be considered as a sequence of quasi-periodic steps, each one of which causes some change in the robots position while returning the robots actuators to approximately the same state at the end of the step as at the beginning. Furthermore, dierent combinations of the periodic inputs can be interpreted as dierent vehicle gaits that move the vehicle in dierent directions [OB98]. Individual steps or gaits can be designed using optimal control techniques. Based on these observations, my approach discretizes the motion group SE(D) (D=2 or 3) in which the robot operates, and plans a trajectory to any given volume element in SE(D) by considering the number of reachable endpoints in that volume element and in intermediate volume elements along the trajectory. My algorithm is largely inspired by the Ebert-Upho algorithm for solving the inverse kinematics of discrete multistage manipulators [CK01]. I adapt this algorithm to motion planning purposes, and improve upon its computational complexity. I also draw on ideas from Dijkstras algorithm, as applied to path planning in a discretized conguration space by Barraquand and Latombe [BL89]. Like those algorithms, my method requires a memory-intensive mapping of the workspace, but generates trajectories to chosen goal points very quickly once the map is constructed. I consider a mobile robot which moves by discrete steps, transforming its location

136
0.4

0.2

0.2

0.4 0.2 0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Figure 6.2: Positions reachable in ve or fewer tail strokes. in SE(D) by any one of K motions in the nite set A SE(D). This discreteness of movement may arise because the robots actuators are actually discrete, or because it is convenient to consider a restricted set of optimal motions at each step (e.g., a carlike robot whose optimal motions are either straight ahead, left arcs, or right arcs [RS90]), or solely for the sake of approximating the reachable set of states. In the case of complex kinodynamic constraints, the nite set of motions can be crafted from a forward simulation of the system for each of a nite set of inputs. This procedure avoids the need for small-time local controllability. For example, Figure 6.1 depicts a hypothetical robot sh which can stroke its tail to make one of four motions: turn left, move forward and sideslip left, move forward and sideslip right, or turn right. These motions are consistent with its complex hydrodynamics. For a sequence of P such steps, the robot can follow K P potential trajectories to K P endpoints in SE(D), with some of the endpoints possibly being duplicates. Figure 6.2 shows how the reachable congurations of the robot sh grow quickly. Figure 6.3 shows the same set of congurations as points in (x, y, ) conguration space.

137

3 1 0 1 y 1 x 1 0

Figure 6.3: States in Figure 6.2 viewed as a density function in conguration space.

138

6.3

Density Functions

I will be concerned with functions expressing the density of a cloud of points in a continuous group. Like any nonuniform density, this density function is not welldened unless I specify the volume elements over which the density is measured. More precisely, let F be a Lie group, and G F be a relevant subset of the group. I begin with the concept of a disjoint discretization V0 of G. This is a collection of disjoint, nite, connected subsets (or volume elements) of G which covers G, together with a function which associates points in G with volume elements. I will call the set of volume elements V0 (G). The elements in V0 (G) are chosen a priori and might, for example, form a regular grid covering G. The associated function V 0 (x) : G V0 (G) maps any element x G to the unique volume element V0 (x) V0 (G) which contains x. For reasons to be made clear below, I also need the more general concept of a non-disjoint discretization V of G. This again involves an arbitrarily chosen set V (G) of nite, connected subsets of G. The elements of V (G) cover G but are not necessarily disjoint. The function V (x) : G V (G) is chosen to associate one volume element V (x) with each element x G. The function V (x) must be chosen so that x V (x) x G. However, since x may lie in more than one element of V (G), this requirement is not enough to wholly determine the function V (x) (as it was in the disjoint case). There is, therefore, an element of freedom in choosing the function V (x). A non-disjoint discretization V can be thought of as built up from a disjoint discretization V0 , if for any x G, V (x) consists of the connected union of V0 (x) with a number of neighboring elements of V0 (G). Again, V may be designed in one way or another by the choice of dierent sets of neighboring elements to form the union. Once I have a set of volume elements, I can proceed to dene a density function. let H R be a measure of the volume of H. The density function (x) : G R is Let n(H) for any subset H G be the number of discrete points that lie in H. And

139 dened as (x) = n(V (x)) . V (x) (6.1)

How large should the volume elements that dene the density function be? The innitesimal limit of the smallest possible volume element is not necessarily the most useful scale to choose. It is not useful to take the volume elements small relative to the spacing of the discrete points and then say that the density is zero almost everywhere. I wish to nd a reachable conguration close to a specied group element (i.e. the goal). Consequently, the volume elements should be suciently large so that each one contains one or more reachable points. Figure 6.4 shows the density function of Figure 6.3, but considered with a coarser discretization so that more volume elements are perceived to have nonzero density. But if the volume elements are both large and disjoint, the discretization of G is coarse, and the resulting paths will be correspondingly imprecise. I can partly avoid this problem by using non-disjoint volume elements. To use a geographic analogy, a census taker might dene the population density at each location, x, as the number of people within a one kilometer radius of x, and then measure and record the population density at locations spaced ten meters apart. In this case the range of V would consist of overlapping one-kilometer-radius circles with centers ten meters apart. The choice of V determines two dierent kinds of spatial resolution. The rst resolution involves the number of volume elements in a particular volume of G. The second resolution is the size of the volume elements. If the volume elements are disjoint, these two resolutions are essentially the same. The main virtue of nondisjoint volume elements is that the rst resolution can remain ne while the second one is made coarse.1
Another advantage of using non-disjoint volume elements is that their shape can be chosen freely. For example, one could make a good case for using volume elements whose projection on R D was spherical. A spherical volume element would treat translational distance isotropically, unlike a rectangular volume element which contains more points in some directions (the corners) than in others. But RD cannot be covered by disjoint spherical volume elements of the same size.
1

140 I start with a discretization V0 where the volume elements are disjoint, and then construct coarser discretizations as necessary. For example a coarser discretization V1 on G can be dened as follows. (See Figure 6.5.) Let V0 (G) be the range of V . Then for any x G, V1 (x) = {H V0 (G)|H V0 (x) = }.

Note that if H borders V0 (x), then the intersection of H with V0 (x) is nonempty. We can iterate this coarsening procedure. For k > 1, Vk (x) = {H Vk1 (G)|H Vk1 (x) = }. (6.2)

If the volume elements in the range of V0 have uniform volume, then the volume of the elements in the range of Vk will increase exponentially with k.

6.4

Order Notation
, then O (g(n)) means the set of all

In measuring the computational cost of various algorithms, I will use the following order notation. If g(n) is a function from N to functions f (n) from N to

such that for some constants c > 0 and n0 0, f (n) c g(n) for all n n0 . (6.3)

So if a function is in O (g(n)), then that function is bounded from above by a xed multiple of g(n) for suciently large values of n. Conversely, to speak of bounds from below I use (g(n)), which is the set of all functions f (n) from N to some constants c > 0 and n0 0, f (n) c g(n) for all n n0 . The set (g(n)) is the intersection of O (g(n)) and (g(n)) [LD91]. (6.4) such that for

141

3 1 0 1 y 1 x 1 0

Figure 6.4: Density from Figure 6.3 using a coarser discretization.

Figure 6.5: Discretization V0 and coarser discretization V1 .


(a) Points x, y, z G.
y

(c) Sets V1 (x), V1 (y), V1 (z) G.

                                                                

142

(b) Sets V0 (x), V0 (y), V0 (z) G.

143

6.5

Computing the Density of Reachable States

Recall the premise that the robot transforms its location in SE(D) by one of K possible motions. There are K P sequences of P steps which will take the robot to as many as K P endpoints in SE(D). I write P to designate the function such that P (g) for a conguration g SE(D) is the number of endpoints in the volume element V (g) centered on g divided by the volume of V (g). (See Figure 6.3.) In other words, P (g) is proportional to the number of dierent ways in which the robot can reach the volume element V (g) in P steps. For small P , I can compute P by simply enumerating the K P points that can be reached. Obviously this method becomes unfeasible as P grows large. For larger P , the density function P can be found by taking the convolution of two known density functions representing smaller numbers of steps.

P (H) = (Q R ) (H) =
SE(D)

(6.5) (6.6)

Q (H) R H1 H d(H),

where H, H SE(D) and Q + R = P . Since k k = 2k , I can produce P with a number of convolutions in O (log P ) operations. Since 1 has at most K nonzero elements, straightforward convolution of 1 with

any other function can be performed in O (N K) time. If K is not large2 , this may actually be the most sensible way in which to proceed. However even if K is not large, this method has the drawback that P can only be computed by serially convolving 1 with 1 , 2 , . . . , (P 1) , requiring O (P ) convolutions. Performing a convolution by numerical integration of Equation (6.6) should require O (N 2 ) operations, where N is the total number of volume elements in the discretization of SE(D). For approximation error to be kept low, N must be quite
2 If K is large, in particular if K is an appreciable fraction of N , then this method may be computationally prohibitive. The more sophisticated methods of convolution described later will 1 denitely be superior if K is in (N ) where > 3 .

144 N Nr NR Np Nu NF Total number of samples on SE(2) O(S 3 ) Number of samples on SO(2) O (S) Number of samples on I 2 R O (S 2 ) Number of samples on p interval O (S) Number of samples on [0, 2) O (S) Total number of harmonics O (S 2 )

Table 6.1: Resolution measurements in SE(2). large, even for D = 2 and especially for D = 3, so naive convolution becomes prohibitively expensive. Fortunately, a faster method of convolution is available.

6.5.1

The Fourier Transform

It is common knowledge that in I n , the Fourier transform of two convolved functions R is simply the product of their individual Fourier transforms. F(f1 f2 ) = F(f1 )F(f2 ) (6.7)

The fast Fourier transform (FFT) algorithm provides an ecient way to numerically approximate a functions Fourier transform. Since the FFT, the multiplication of the resulting Fourier transforms, and the inverse FFT can each be performed in subquadratic time, this is fast way to perform convolution in I n . To generalize these R concepts to SE(D), we apply concepts from the eld of noncommutative harmonic analysis. The remainder of this section summarizes material from [CK01]. I can express the matrix elements of the Fourier transform of a function f (r, ) on SE(2) by [CK01] fmn (p) =
2 rI R
2

2 =0

=0

f (r, ) ein ei(pr) eim() d2 r d d.

(6.8)

In practice I use a band-limited approximation of the Fourier transform and only compute elements with |m|, |n| < S for some S Z+. The group SE(2) is discretized with the number of samples described in Table 6.1. The Fourier transform can be carried out numerically in the following way. First, assuming that f (r, ) is sampled

145 on a Cartesian grid of r I 2 values, the integration over I 2 is R R f1 (p, ) =


2

I R

f (r, )ei(pr) d2 r

(6.9)

using the usual FFT in O (NR Nr log(Nr )) time. Then I interpolate from the Cartesian grid to sample f1 (p, ) on a polar-coordinate grid of p values. This set of f1 (p, , ) values can be found in O (NR Nr ) = O (N ) time assuming that the interpolation of each individual grid point can be done in constant time. (See Table 6.1 for denitions of N , NR , et cetera.) Then the usual one-dimensional FFT can be used to integrate, rst along the dimension in O (Nr NR log(NR )) time f2 (p, ) =
(m)

f1 (p, , )eim d
SO(2)

(6.10)

and along the dimension in O (NR Np Nu log(Nu )) time fmn (p) =


0 2

f2 (p, )eim ein d.

(m)

(6.11)

The entire process takes O (N log(N )) time. The recovery of a function from its band-limited Fourier transform also takes O (N log(N )) time. 1 f (r, ) = 2
S

p dp
0 m,n=S

fmn (p)

ei(mn) eirp eim d


0

(6.12)

The Fourier transform on SE(2) has the crucial convolution property [KC99] F(f g) = F(g)F(f ),
S

(6.13) (6.14)

F(f g)mn (p) =

gmk (p)fkn (p).


k=S

Thus, the convolution of functions f and g on SE(2) can be performed by nding the Fourier transforms in O (N log(N )) time, multiplying the Fourier matrices, and performing the inverse Fourier transform in O (N log(N )) time. The most computa-

146 tionally expensive step is matrix multiplication, which takes O (N NR ) = O N 4/3 time if the obvious method of matrix multiplication is used. This bound can be slightly improved by using more advanced matrix multiplication algorithms. 3 In any case, for large N this method is clearly a drastic improvement over O (N 2 ) straightforward numerical convolution.

6.6
6.6.1

Trajectory Planning
Dijkstras Algorithm in Trajectory Planning

I begin by reviewing a well-known mobile robot planning method discussed in [Lat90], based on Dijsktras algorithm. By comparing this algorithm to the density-based methods to be introduced in this chapter, I will show the advantages (and disadvantages) of density-based planning relative to existing methods. Later, in Section 6.6.6, we will show how ideas from this algorithm can be combined with density-based planning to obtain some of the advantages of both. The Dijkstra-based idea is to divide the conguration space into N cells in a disjoint discretization V0 , then construct a directed graph whose edges are feasible paths between congurations, and whose nodes are congurations with no more than one conguration per cell. Once constructed, the graph serves as a map: a feasible path from the start cell to any cell reached by the graph can be recovered from the graph in O (P ) time, where P is the number of steps in the path. The graph is a tree and only contains one path to one conguration in any given cell. This one path is intended to be the best path according to some cost function (possibly, but not necessarily, distance or number of steps) used to guide construction of the path. The graph is constructed according to the following well-known procedure [Lat90]. Maintain a list OPEN of congurations which have been reached but whose successors have not yet been considered. Also maintain a list CLOSED of congurations whose successors have been considered, and mark each cell in V0 which contains a CLOSED
The matrix multiplication step will require O N (+1)/3 computations, where = 3 for standard matrix multiplication, but = log 2 7 2.81 using Strassens algorithm [CK01].
3

147 conguration. Initially OPEN will only contain the start conguration and CLOSED will be empty. Given a cell in V0 , it can be determined in constant time whether it contains a CLOSED conguration (and if so what the conguration is.) The OPEN list is maintained as a balanced tree ordered by cost and can be accessed in at most logarithmic time.4 Select the node in OPEN with the lowest cost to reach, transfer it to CLOSED, and compute its K successors. For each successor, if it lies in a cell which already contains a CLOSED conguration, discard it. Otherwise, add it to OPEN. Keep a record of the immediate ancestor of every node which is not discarded. The algorithm terminates when every cell that the planner desired to map a path to is marked CLOSED, or when OPEN is empty and no more cells can be reached. Because the algorithm explores the successors of at most one conguration per cell, the time to construct the graph is in O (KN log(N )), or O (KN ) if OPEN can be accessed in constant time. The complete path to a CLOSED cell can be recovered from the graph by nding the reached conguration in that cell and then recursively tracing the ancestry of that conguration backwards. Limitations of the Latombe-Barraquand-Dijkstra Algorithm A drawback of the Latombe-Barraquand-Dijkstra (LBD) algorithm is that, since it prunes what would otherwise be an exponentially growing tree of possible paths by recording at most one conguration per cell, there is a danger that it will prune away a desirable solution. Suppose that the system is capable of reaching congurations g1 and g2 , where g2 = g1 (I +
4

D(D+1) i Xi ) i=1

for some small scalars

and Xi are elements of se(D).

If the cost function used to select nodes for exploration is suciently simple, then OPEN can be accessed in constant time. Three examples of suciently simple cost functions are (a) number of steps along the path; (b) number of steps along the path which require any change of steering from the previous step; (c) number of steps along the path which require a specic change of steering from the previous step, e.g., a change from forward to reverse for a robot car. The relevant facts about these cost functions are that they take integer values and the value changes by at most one between a conguration and it successors. In each of these cases, OPEN should be maintained not as a tree but as a collection of linked lists, each linked list containing OPEN nodes of equal cost.

148 Since g1 and g2 are in the same cell V (g1 ) = V (g2 ), and g2 requires an equal or greater cost to reach, the LBD algorithm will not record g2 or explore its successors. This means that if the precise desired goal is g2 , the LBD algorithm will only return the approximate solution g1 . But more importantly, after a further transformation h, corresponding to one or more additional steps, the successor of g2 would have been:
D(D+1) D(D+1) i Xi ) h i=1

g2 h = g1 (I + Even if

= g1 h +
i=1

i Xi h

(6.15)

is small, for h suciently large the error between successors of g1 and g2

will grow large compared to the size of the cells V0 (g1 h) and V0 (g2 h) and therefore the successors will lie in dierent cells. It is possible, therefore, that the successors of g1 reach V0 (g2 h) only at higher cost than g2 h, or that they never reach V0 (g2 h) at all. The LBD algorithm, which aims to nd the lowest-cost path to each cell, may thus lose the desired solution, particularly if the system is not STLC. This problem is aggravated as h grows large, suggesting that the optimality of solutions found by the LBD algorithm becomes more suspect as the total path length grows larger. The problem also grows worse as the size of the cells of V0 increases, so it may not be possible to use the LBD algorithm on a very coarse scale: past a certain point of coarseness I may get, not just a more approximate solution, but no solution at all. I will now suggest another planning algorithm, using density functions. Part of this algorithms appeal is that (a) the density function represents all solutions and not just the ones nominally with lowest cost; and (b) a density function constructed by convolution can represent long paths formed by the concatenation of two large motions rather than by the incremental addition of many smaller motions; and for these two reasons the density-based algorithm is less likely to lose solutions due to round-o error. The density-based algorithm should also function at a variety of scales, both ne and coarse. Informally speaking, it can be applied not only to the parking problem for a car-like robot, but also to the cross-country trip problem.

149

6.6.2

The Ebert-Upho Algorithm

Let V0j denote the discretization of SE(D) chosen for a j step density function5 . Suppose that the density functions j and their associated discretizations V0j for gdes SE(D) in P steps, one sequentially chooses the j th motion step to maximize the reachable state density function of the remaining P j steps around the goal. That is, the rst step g1 is chosen from the set of K allowable motions in SE(D) to
1 1 1 maximize (P 1) (g1 gdes ). With g1 xed, g2 is chosen to maximize (P 2) (g2 g1 gdes ). 1 1 1 One goes on to choose the jth step gj to maximize (P k) (gj g2 g1 gdes ). After

1 j P are known. To nd a trajectory which ends as closely as possible to the goal

P 1 steps, the nal step gP is chosen in order to minimize a distance measure e = d(gdes , g1 g2 gP ) (6.16)

This scheme for nding a trajectory that approximately reaches the goal in P steps is identical to the Ebert-Upho algorithm for solving the inverse kinematics of a discretely actuated P -link manipulator [CK01]. It returns an approximate solution in O (KP ) time.

6.6.3

Implicit Storage of Paths in Graph

Because this algorithm starts with a goal reachable in P steps, makes a transformation to a point reachable in P 1 steps, then P 2 steps and so on, it is vaguely reminiscent of the BLD algorithm [BL89]. I can increase the resemblance by explicitly constructing a tree connecting reached volume elements. For every conguration g0 where j (g0 ) is nonzero, let v = {g| V j (g) = V j (g0 )}.

then store a pointer between V (g) and V (h1 g). This step requires O 2D KN P

To construct the graph, search for the g v and h A which maximize j1 (h1 g),

time for P density functions. Alternatively, instead of searching over both g and h, one can specify a representative value of g v; in this case, the graph construction
Generally, the discretizations V0 and V0j are the same. However, I allow for the option that one may choose to resample SE(D) for purposes of rening the density function j .
5

150 takes O (KN P ) time. In either case, by following edges of the constructed graph instead of explicitly checking K density values at each step of the trajectory, paths can be found in O (P ) rather than O (KP ).
10
8

10

P at goal (x=0.51, y=0.8, =0)

10

10

10

10

10

10

10

10

10

10

12

14

16

Figure 6.6: The density P in the vicinity of a particular goal, as a function of P . The path length P can be chosen so that P is suciently high. Often, P increases with increasing P , so a longer path may yield a better approximation to the goal. When such an explicit graph is used, the method is still distinct from the traditional application of Dijkstras algorithm in at least two ways. Dijkstras algorithm chooses steps backwards from the goal to the trajectory, whereas this method chooses steps in a forward fashion. This means inter alia that while both methods require O (P ) time to construct the whole path, if only the next step in the path is required then this method can nd it in O (1) time. Also, instead of associating a single pointer with each volume element and implicitly storing only the shortest path to each volume element, up to P pointers and implicitly up to P successful trajectories are stored for every volume element. With this construction, one can either choose the shortest path which approximately reaches the goal, or possibly choose a longer path which reaches the goal more exactly. Example. Figures 6.66.9 show the output from an implementation of this algorithm as applied to the sh robot of Figure 6.1 trying to reach the goal (x, y, ) =

151
0.8

0.6

0.4

0.2

0 0.2 0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Figure 6.7: The algorithm nds a 10-step path to the goal. While the nal heading is not quite right, this is roughly the shortest path with possibly acceptable error. (0.51, 0.8, 0). Figures 6.76.9 illustrate the trade-o between path length and nal goal accuracy.

6.6.4

Multiscale Densities

This section introduces a multiscale extension of the basic algorithm. Recall that j denotes the density function j dened at the coarser scale Vkj . To motivate the k multiscale approach, note that each density function j is associated to a discretization V j . The dierent discretizations V 1 , , V P need not be identical. The algorithm, as

described so far, neglects the possibility that at the j th step gj , the density function

1 1 1 (P j) (gj g2 g1 gdes ) is zero for all K possible choices of gj . In this case the goal

gdes lies in a volume element which is not considered reachable. It might then be said that the resolution of V j was chosen too ne. Since I wish to reach the closest conguration to gdes , I should have chosen the volume elements big enough so that the one which contains gdes also contains some reachable point. Unfortunately, in advance of knowing gdes , it is not necessarily possible to know how

152
0.8

0.6

0.4

0.2

0 0.2 0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Figure 6.8: Higher accuracy can be obtained by varying P to obtain a more dense P . This 14-step path ends in a better approximation to the goal. ne a resolution is too ne. I will now adapt the algorithm to use density functions at a coarser resolution when necessary. Recall that according to the coarsening scheme outlined earlier, for k 1, Vkj (x) =
j j H Vk1 (G)|H Vk1 (x) = . (P j)

(6.17)

I want to nd the smallest k such that k

1 1 1 (gj g2 g1 gdes ) is nonzero for some

valid choice of gj . Then I choose gj to maximize this value. In the worst case, it might be necessary to consider O (log(N )) scales.6 All

O (log(N )) scales can be constructed from the nest scale in O (N log(N )) time, which is dominated by the time required to construct the density function at the nest scale in the rst place. To store the density function at all O (log(N )) scales would require O (N log(N )) space. But not all of this information need be retained.
I have dened a series of scales whose individual volume elements increase geometrically in size. Only O (log(N )) scales can be dened before reaching the coarsest scale, where the density function is uniform across the whole environment. I could instead have coarsened the scale more gradually, say linearly, and dened the density function at O (N ) dierent scales. But the complexity of this multiscale algorithm would rise to O (KN P ) in time, and the time required to construct the dierent density functions in the rst place would be (N 2 ).
6

153
0.8

0.6

0.4

0.2

0 0.3

0.2

0.1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Figure 6.9: A 16-step path reaches the goal conguration almost perfectly. The construction of the density functions at coarser scales only requires knowledge of the next ner scale. I only need to record for each volume element the nest scale at which the density function assumes nonzero value there, and that nonzero value. Consequently, the multiscale algorithm requires O (N ) space. Assume that the multiscale density functions are precomputed, and that the nest scale at which each volume element has a nonzero density is known. Then the algorithm can be employed as before, with the proviso that arguments are evaluated at the nest scale which displays nonzero density. Since this can be done in constant time for each argument, the worst-case performance of the modied algorithm is still O (KP ), or O (P ) if a graph is used to store trajectories. Error Bounds The worst-case error can be bounded at the outset by the size of the volume element VkP (gdes ) for the smallest value of k such that P (gdes ) is nonzero. After j steps k have already been taken, the remaining expected error is bounded by the size of Vk
(P j) 1 1 1 (gj g2 g1 gdes ) for the smallest k such that k (P j) 1 1 1 (gj g2 g1 gdes ) is

154 nonzero. And of course, when the algorithm is complete the actual error is e = d(gdes , g1 g2 gP ). Limitations of Ebert-Upho Methods At this point, I should acknowledge some advantages which the Dijkstra-based BLD algorithm retains over the density-based Ebert-Upho method. For single-processor implementations, the time requirements of the BLD algorithm are somewhat lower. The BLD algorithm has more exibility to assign dierent costs to dierent allowed motions. The density-based method treats all steps equally and cannot easily account for path cost if the cost function cannot be related to number of steps. The Dijkstra-based methods can take account of world-xed constraints like static obstacles and space-dependent currents, provided these are known at the precomputation/mapping stage. In Section 6.6.5, I will present a dierent density-based algorithm with dierent computational bounds, oering the potential for faster performance over long paths if the systems density functions are reasonably well-behaved. Adjusting density functions to account for obstacles and other conditions located at particular points in space is problematic. Both the concept of constructing density functions from convolution, and the Ebert-Upho planning algorithm, rely on the idea that the density of allowed transformations is invariant under transformations of the origin in SE(D). This clearly does not hold for the density of solutions in the presence of obstacles which destroy the symmetry of the conguration space. Nevertheless, the unaltered density functions for the open-space problem, without obstacles, can be used to aid a BLD-like attack on the problem with obstacles. In Section 6.6.6, I describe how to use density functions to navigate around obstacles in this way.

155
Method Barraquand and Latombe Ebert-Upho Multiscale Ebert-Upho Multiscale E-U plus graph Log method Log method plus graph Log method with O (P ) processors Log method, graph, O (P ) processors Method Barraquand and Latombe Ebert-Upho Multiscale Ebert-Upho Multiscale E-U plus graph D&C method D&C method plus graph D&C method with O (P ) processors D&C method, graph, O (P ) processors Time to Construct Map O (N K) O N (+1)/3 P or O (N KP ) O N (+1)/3 P or O (N KP + N P log N ) O N (+1)/3 P + N KP O N (+1)/3 log P O N (+1)/3 log P + N log P O N (+1)/3 (log P )/P O N (+1)/3 (log P )/P + N (log P )/P Space O (N ) O (N P ) O (N P ) O (N P ) O (N log P ) O (N log P ) O (N log P ) O (N log P ) Time to Plan Path O (P ) O (KP ) O (KP ) O (P ) O ( P ) O (P ) O ( log P ) O (log P )

Table 6.2: Time and space bounds, where N is the number of volume elements, P is the number of trajectory steps, K is the number of possible motions at each step, and the constants and depend upon the chosen methods of function maximization and matrix multiplication. Although the Barraquand/Latombe algorithm has the lowest time and space requirements, it does not have all the same utility as the algorithms presented here; notably it does not allow task-by-task trade-os between path length and accuracy.

6.6.5

A Divide-and-Conquer Algorithm

I now present a variation of the algorithm with logarithmic time complexity. This version is based on the observation that if the densities j , j+1 , , 2j are known,

then simple addition yields the function [j,2j1] , which is the density of endpoints reachable by trajectories of at least j but no more than 2j 1 steps. Convolving

[j,2j1] with j yields the function [2j,3j1] , and convolution of [j,2j1] with 2j re-

sults in [3j,4j1] . The functions [2j,3j1] and [3j,4j1] can be added to nd [2j,4j1] . Finally convolution of 2j with itself produces 4j . Thus, three convolutions and one addition extends our knowledge of j , 2j , and [j,2j1] , to 4j and [2j,4j1] . A further addition obtains [2j,4j] . Iteration of this procedure produces the sequence of functions [1,2] , [2,4] , [4,8] , [8,16] , , [P/2,P ] in time and space logarithmic in P . Armed with these functions, a trajectory to gdes SE(D) is planned as follows.

156 Choose a trajectory length P large enough so that the density [P/2,P ] (gdes ) is suciently high; the higher it is, the more closely the goal is likely to be reached. Using a function maximization algorithm, search for an element h SE(D) which maximizes

the product [P/4,P/2] (h)[P/4,P/2] (h1 gdes ). Having found such an element h, search for a path of no more than P/2 steps which approximates h, and another path of

no more than P/2 steps which approximates h1 gdes . In this manner, by recursively nding the midpoint of each unknown path segment, I eventually nd P points along the path, which describe an entire trajectory which approximately reaches gdes . This approach exploits the exponential properties of convolution better than the linear Ebert-Upho algorithm, and therefore only requires the computation and storage of O (log P ) density functions. The main drawback is the use of the functionmaximization search. Depending on the details of the system, it may be dicult to predict the time required by this step. In the theoretical worst case, if it is necessary to check every possible volume element to nd the maximum product of densities, then could be O (N ). But if the density functions of the system are fairly wellbehaved, then in practice will be manageably small. For example, if direction set maximization methods are usable, then should only be quadratic in the dimension of the conguration space. Using a single processor, the path-nding procedure takes O ( P ) time to nd a trajectory. However, the algorithm lends itself to parallelization. If O (P ) processors are available, then the trajectory can be found in O ( log(P )) time.

6.6.6

Obstacle Navigation with Density Functions

Suppose that the density functions P for the system in the absence of obstacles are already computed for a wide range of P . Then the goal gdes is provided but a certain number of the cells in the discretization V0 are now occupied by obstacles and may not be entered. Paths to gdes which would succeed in the absence of obstacles may now be blocked. Here I will describe an algorithma sort of fusion of the BLD algorithm and the linear Ebert-Upho algorithmwhich uses the density functions to inform a

157 search for a path through the obstacles. Begin by nding the lowest value of P such that P (gdes ) is nonzero (or acceptably high.) That is, there is known to be an acceptable path of length P to the goal in the absence of obstacles, but there is no path of length P 1 in either the absence or presence of obstacles. Maintain three lists: A list OPEN of congurations which have been reached, and which lie on a path of length P to the goal in the absence of obstacles, but whose successors have not all yet been considered; A list DEFERRED of congurations which have been reached, and whose successors have not been considered, but which does not lie on a path of length P to the goal, even in the absence of obstacles. (These congurations are deferred because they may still lie on a path of length greater than P which reaches the goal.) A list CLOSED of congurations whose successors have all been considered. For every conguration c = g1 g2 gj which is in OPEN, DEFERRED, or CLOSED, I record the integer j(c), which is the number of steps j by which c was reached,
1 and also the conguration cgj which was the immediate ancestor of c. Every cell

in V0 that contains a conguration in either OPEN, DEFERRED, or CLOSED, has a pointer to that conguration. Thus, given any cell in V0 , it can be determined in constant time if it has been reached, and if so the path to the cell can be recovered in O (j) time. Both the OPEN and DEFERRED lists are maintained as a collection of linked lists ordered by j. Initially OPEN contains only the start conguration (with j = 0) and DEFERRED and CLOSED are empty. Then proceed as follows: 1. Take the conguration c in OPEN with maximum j(c).

158 2. Find the allowable transformation g A such that cg does not intersect an

OPEN or DEFERRED or CLOSED cell, which maximizes7 (P (j(cg)) (g 1 c1 gdes ).

3. If there is no such g (i.e., all K successors of c lie in obstacles or cells that have already been reached) then remove c from OPEN and place it in CLOSED. 4. If (P j1) (g 1 c1 gdes ) is zero, then there is no path of length P from the origin to the goal which passes through c. Remove c from OPEN and place it in DEFERRED. 5. Otherwise, (P j1) (g 1 c1 gdes ) is greater than zero. Add cg to OPEN. Leave c in OPEN. Since j(cg) = j(c) + 1, the conguration cg is now at the top of the OPEN list and will be considered next. 6. Go back to step 1 and repeat until the goal is reached or until OPEN is empty. If the goal is reached, the algorithm terminates (success.) 7. If OPEN is empty, then there is no path of length P from origin to goal in the presence of the obstacles. If DEFERRED is also empty, then the algorithm terminates (failure.) If DEFERRED is not empty, there may be a longer path from the origin to the goal. Increment P by one. Move all the congurations in DEFERRED back into OPEN. Return to step 1. With two caveats, this algorithm will nd the minimum length path to the goal if one exists. Proof: Whenever a given cell is reached for the rst time, and a conguration c in the cell is added to OPEN, the path of length j to c is known to be part of a path of length P that would reach the goal in the absence of obstacles. Furthermore, all paths of length less than P that would reach the goal in the absence of obstacles have been explored until they either collided with an obstacle or began retracing a previously explored path of length less than P (which in turn was explored
Finding the maximizing successor once takes only O (K) time, but if the node c is revisited K times and must produce K unexplored successors in order of decreasing solution density, this could take O K 2 time. For better asymptotic performance, the successors should all be sorted by solution density when c is rst considered, in O (K log K) time. Of course, K may be small enough that this is a niggling distinction.
7

159 until it collided with an obstacle). Therefore, there cannot be a feasible path of length k < j to the cell containing c, because if so the cell containing c would have been previously entered while the algorithm was searching for paths of length P (j k). So in particular, when the goal cell is reached the path to the goal cell will be of minimum length. Since every reachable cell will eventually be reached, the minimum length path to the goal will be found if a path exists. The two caveats or holes in the proof concern the possibility of numerical error. First, the proof assumes that the information in the density functions is reliable, and in particular, that the density function will be nonzero when there is a solution and zero when there is no solution. Obviously, to the extent that there is approximation error or any other kind of error in the density functions, the path search may be misdirected accordingly. Second, and perhaps more importantly, the algorithm assumes that multiple congurations in the same cell do not have to be retained, because further motion starting from those similar congurations will reach substantially similar cells. But as we have seen, over long paths even tiny dierences in starting orientation can lead to large dierences in the nal destination. This algorithm thus inherits the same vulnerability to round-o error as the original BLD algorithm. In the absence of small time local controllability, there is no obvious remedy except to keep the cell resolution as high as computationally feasible. Time Bounds In the worst case, this algorithm does not perform any better than the BLD algorithm, because there is no guarantee that searching where solutions disregarding obstacles are known to be dense will yield a solution which accounts for obstacles. In the best case, however, when the obstacles turn out not to obstruct the standard Ebert-Upho path from the origin to gdes , this algorithm will nd the path in O (KP ), while the conservative breadth-rst search of the BLD algorithm may still take a long time (in O (N K)) to complete even in this optimistic case. The information provided by the density functions provides this algorithm with two advantages relative to the uninformed BLD algorithm.

160 The density-based algorithm can deduce in O (P ) time that the length of the path to the goal must be at least P , considering only motion constraints and not the eect of obstacles. This algorithm can therefore immediately begin a depthrst search for a path of length P . The BLD algorithm, with no knowledge of the constraints placed on path length by the allowed motion set, will perform a breadth-rst search to avoid missing a (nonexistent) path of length less than P. If exploration from a given node cannot possibly result in a successful path of length P (because of motion constraints), then the density-based algorithm will not explore from that node until after it has checked all nodes that possibly could result in a path of length P . The uninformed BLD algorithm will explore equally from all nodes equally distant from the origin. So I expect that the density-based algorithm will typically be faster than the uninformed BLD algorithm, because it avoids checking fruitless paths that the BLD algorithm will check before nding a solution. On the other hand, the density-based algorithm will never perform signicantly worse than the BLD, because the densitybased algorithm never explores any path longer than the path actually required to reach the goal, but the BLD in general will search every path of length less than or equal to the solution path before nding the solution. In the worst case, where there is no solution, the BLD algorithm will take O (N K) time and the density based algorithm will take O (N K log K) time. The additional factor of log(K) arises from the need to sort paths by solution densityobviously, it is of no real signicance unless K is unusually large.

161

Chapter 7

Conclusions and Future Work

I have presented general closed-form expressions for the force and moment on a deformable Joukowski foil, and the equations of motion for a few examples of deformable swimming bodies. I have built and modelled a three-link planar robot shthe robot sh itself is the rst such platform to be well suited for feedback control experiments, while the model is useful insofar as it assists feedback control and path planning, and gives insight into what physical processes are dominant during any particular sh maneuver. I have outlined how sets of reachable points, and the trajectories necessary to reach them in a minimum time, could be computed for the three-link sh. And I have shown how the density of these sets of reachable points can be convolved to obtain the density of endpoints of longer trajectories, and how these functions can be used as tools in plotting motions for the sh, or for many other mobile robots, with or without the presence of obstacles. There are still many directions in which this work should proceed. Numerical optimization of desirable trajectories for the three-link robot, and for the self-propelled deformable Joukowski foil, is still unnished. It would probably be possible to extend the treatment given to the deformable Joukowski foil and apply it to more general kinds of deformable foil shapes. More ambitiously, it would be interesting to evaluate the potential around multiple foil shapes, so that a sh, for example, could be modelled as a collection of hydrodynamically interacting ns and streamlined body parts. (This contrasts with my current treatment of the the three-link robot in which the three links are assumed to by hydrodynamically separate.) In the realm of path-planning using density functions, the ideas suggested here remain untested on real-world problems. The divide-and-conquer algorithm for charting a path with a logarithmic number of density functions, in particular, requires that the systems density functions, and their products, be reasonably well-behaved and amenable to standard searching techniques. It would be interesting to see whether

162 this holds true of the density functions for a real system of interest (as opposed to an elementary model constructed to illustrate the algorithm).

163

Bibliography
[ACS+ 97] B. Ahlborn, S. Chapman, R. Staord, R.W. Blake, and D.G. Harper. Experimental simulation of the thrust phases of fast-start swimming of sh. Journal of Experimental Biology, 200(17):23012312, September 1997. [AD59] Ira H. Abbott and Albert E. Von Doenho. Theory of Wing Sections. Dover Publications, New York, 1959. [AHB+ 91] Boye Ahlborn, David G. Harper, Robert W. Blake, Dorit Ahlborn, and Michael Cam. Fish without footprints. Journal of Theoretical Biology, 148:521533, 1991. [AK99] Jamie Anderson and Peter A. Kerrebrock. The vorticity control unmanned undersea vehicle (vcuuv) performance results. In Proceedings of the 11th International Symposium on Unmanned Untethered Submersible Technology, 1999. [AWO00] Joseph Ayers, Cricket Wilbur, and Chris Olcott. Lamprey robots. In Proceedings of the 1st International Symposium on Aqua Bio-Mechanisms, pages 327328, 2000. [Ban00] Rudolf Bannasch. Experimental studies on hydrodynamic adaptation in

penguins- mechanisms of turbulence control and their applicability to engineering. In Proceedings of the 1st International Symposium on Aqua Bio-

Mechanisms, pages 1318, 2000. [Bar96] D.S. Barrett. Propulsive Eciency of a Flexible Hull Undersea Vehicle. PhD thesis, Massachusetts Institute of Technology, Cambridge, MA, 1996. [BGT96] David Barrett, Mark Grosenbaugh, and Michael Triantafyllou. The optimal control of a exible hull robotic undersea vehicle propelled by an oscillating foil. In Proceedings of the IEEE AUV Symposium, pages 19, 1996. [BL89] J. Barraquand and J.C. Latombe. On non-holonomic mobile robots and optimal manuevering. Revue dIntelligence Articielle, 3(2), 1989.

164
[CK01] Gregory S. Chirikjian and Alexander B. Kyatkin. Engineering Applications of Noncommutative Harmonic Analysis. CRC Press, Boca Raton, 2001. [CLC98] I-Ming Chen, Hsi-Shang Li, and Arnaud Cathala. Amoebot metamorphic underwater vehicle. In Fifth International Conference on Control, Automation, Robotics, and Vision, Singapore, 1998. [CMOM01] J. Cortes, S. Martinez, J.P. Ostrowski, and K.A. McIsaac. Optimal gaits for dynamic robotic locomotion. International Journal of Robotics Research, 20(9):707728, September 2001. [CMPB00] M. Ceccarelli, A. Marigo, S. Piccinocchi, and A. Bicchi. Planning motions of polyhedral parts by rolling. Algorithmica, 26:560576, 2000. [Gie68] J.P. Giesing. Nonlinear two-dimensional unsteady potential ow with lift. Journal of Aircraft, 5(2):135143, 1968. [GM94] A. Galper and T. Miloh. Generalized kirchho equations for a deformable body moving in a weakly non-uniform ow eld. Proc. Roy. Soc. London A, 446:169193, 1994. [Gra36] J. Gray. Studies in animal locomotion. vi. the propulsive powers of the dolphin. Journal of Experimental Biology, 13:192199, 1936. [Gur92] Leonid Gurvits. Averaging approach to nonholonomic motion planning. In Proc. IEEE Int. Conf. Robotics and Automation, pages 25412546, 1992. [HTT00] Koichi Hirata, Tadanori Takimoto, and Kenkichi Tamura. Study on turning performance of a sh robot. In Proceedings of the 1st International Symposium on Aqua Bio-Mechanisms, pages 287292, 2000. [KC99] A.B. Kyatkin and G.S. Chirikjian. Synthesis of binary manipulators using the fourier transform on the euclidean group. Journal of Mechanical Design, 121:9 14, March 1999. [KM96] S.D. Kelly and R.M. Murray. The geometry and control of dissipative systems. In Proc. IEEE Conf. on Decision and Control, 1996.

165
[KSLO96] L. Kavraki, P. Svestka, J.C. Latombe, and M.H. Overmars. Probabilistic

roadmaps for path planning in high dimensional conguration spaces. IEEE Trans. Robotics and Automation, 12(4):566580, 1996. [KWS00] Naomi Kato, Bugi Wiku Wicaksono, and Yoshihiro Suzuki. Swimming performance of biology-inspired autonomous underwater vehicle with a pair of mechanical pectoral ns. In Proceedings of the 1st International Symposium on Aqua Bio-Mechanisms, pages 311314, 2000. [Lat90] J.-C. Latombe. Robot Motion Planning. Kluwer Academic Publishers, Boston, 1990. [LD91] Harry R. Lewis and Larry Denenberg. Data Structures & Their Algorithms. HarperCollins, New York, 1991. [Lig75] J.L. Lighthill. Mathematical Biouiddynamics. Society for Industrial and Applied Mathematics, Philadelphia, 1975. [LK00] S.M. Lavalle and J.J. Kuner. Rapidly-exploring random trees: Progress

and prospects. In Proc. Workshop on Algorithmic Foundations of Robotics, Hanover, NH, Mar. 2000. [LS93] G. Laerriere and Hector J. Sussmann. A dierential geometric approach to motion planning. In Z. Li and J. F. Canny, editors, Nonholonomic Motion Planning, pages 235270. Kluwer, 1993. [LS95] Frank L. Lewis and Vassilis L. Syrmos. Optimal Control: Second Edition. John Wiley & Sons, New York, 1995. [Mag54] W. Magnus. On the exponential solution of dierential equations for a linear operator. Communications on Pure and Applied Mathematics, VII:649673, 1954. [MDM+ 01] K.A. Morgansen, V. Duindam, R.J. Mason, J.W. Burdick, and R.M. Murray. Nonlinear control methods for planar carangiform robot sh locomotion. In Proceedings of the IEEE International Conference on Robotics and Automation, pages 427434, 2001.

166
[MG93] T. Miloh and A. Galper. Self-propulsion of a manoeuvering deformable body in a perfect uid. Proc. Roy. Soc. London A, 442:273299, 1993. [Mil91] T. Miloh. Hydrodynamical self-propulsion of deformable bodies and oscillating bubbles. In T. Miloh, editor, Mathematical Approaches in Hydrodynamics, pages 2136. SIAM, Philadelphia, 1991. [MR94] J.E. Marsden and T.S. Ratiu. Springer-Verlag, New York, 1994. [MS91] R.M. Murray and S.S. Sastry. Steering nonholonomic systems in chained form. IEEE Control and Decision Conference, pages 11211126, 1991. [MS93a] J.E. Marsden and J. Scheurle. The reduced Euler-Lagrange equations. Fields Institute Comm., 1:139164, 1993. [MS93b] R.M. Murray and S.S. Sastry. Nonholonomic motion planning: Steering using sinusoids. IEEE Transactions on Automatic Control, 38:700716, 1993. [MT68] [MVB02] L.M. Milne-Thomson. Theoretical Hydrodynamics. Macmillan, New York, 1968. K. A. Morgansen, P.A. Vela, and J.W. Burdick. Trajectory stabilization for a planar carangiform robot sh. In Proceedings of the IEEE International Conference on Robotics and Automation, pages 756762, 2002. [NKO00] Motomu Nakashima, Kazuhiro Kaminaga, and Kyosuke Ono. Experimental study of two-joint dolphin robot (propulsive characteristics of 2nd large model). In Proceedings of the 1st International Symposium on Aqua Bio-Mechanisms, pages 311314, 2000. [NS82] Arch W. Naylor and George R. Sell. Linear Operator Theory in Engineering and Science. Springer-Verlag, New York, 1982. [OB98] J.P. Ostrowski and J.W. Burdick. The mechanics and control of undulatory locomotion. Int. J. Robotics Research, 17(7):683701, July 1998. [Ost95] J.P. Ostrowski. The Mechanics and Control of Undulatory Robotic Locomotion. PhD thesis, California Institute of Technology, Pasadena, CA, Sept. 1995. Introduction to Mechanics and Symmetry.

167
[RB98] J.E. Radford and J.W. Burdick. Local motion planning for nonholonomic control systems evolving on principal bundles. In Proc. Mathematical Theory of Networks and Systems, July 1998. [RS90] J. A. Reeds and L. A. Shepp. Optimal paths for a car that goes both forwards and backwards. Pacic Journal of Mathematics, 145(2), 1990. [Saf67] P.G. Saman. The self-propulsion of a deformable body in a perfect uid. J. Fluid Mechanics, 28:285289, 1967. [Saf92] P.G. Saman. Vortex Dynamics. Cambridge University Press, Cambridge, 1992. [Sar75] T. Sarpkaya. An inviscid model of two-dimensional vortex shedding for transient and asymptotically steady separated ow over an inclined plate. Journal of Fluid Mechanics, 68:109128, March 1975. [Sed65] L. I. Sedov. Two-Dimensional Problems in Hydrodynamics and Aerodynamics. Interscience Publishers, New York, 1965. [SLD00] Michael Sfakiotakis, David M. Lane, and Bruce C. Davies. Development of a n actuator for the investigation of undulating n propulsion. In Proceedings of the 1st International Symposium on Aqua Bio-Mechanisms, pages 265270, 2000. [SO95] P. Svestka and M.H. Overmars. Coordinated motion planning for multiple carlike robots using probabilistic roadmaps. In Proc. IEEE Int. Conf. on Robotics and Automation, Nagoya, Japan, Apr. 1995. [ST95] K. Streitlien and M.S. Triantafyllou. Force and moment on a joukowski prole in the presence of point vortices. AIAA Journal, 33(4):603610, Apr. 1995. [Str94] Knut Streitlien. A simulation procedure for vortex ow over an oscillating wing. Technical report, MIT Sea Grant College Program, 1994. [SW89] Alfred Shapere and Frank Wilczek. Geometry of self-propulsion at low Reynolds number. Journal of Fluid Mechanics, 198:557585, January 1989.

168
[TTG93] G. S. Triantafyllou, M. S. Triantafyllou, and M. A. Grosenbaugh. Optimal thrust development in oscillating foils with application to sh propulsion. J Fluid Struct., 7:205224, 1993. [VVP73] V. Voinov, O. Voinov, and A. Petrov. Hydrodynamic interaction between bodies in a perfect incompressible uid and their motion in non-uniform streams. Prikl. Math. mech., 37:680689, 1973. [Wei72] D. Weihs. A hydrodynamical analysis of sh turning manoeuvres. Proc. R. Soc. Lond. B, 182:5972, 1972.

169

Appendix A Forces

Integrals for Unsteady Flow

These integrals arise in determining the force on a deformable Joukowski foil due to the time-varying potential.

(w+w) dz = U t + c

(w1 + w1 ) dz +V

(w2 + w2 ) dz +
k w5 dz

(w3 + w3 ) dz

(w4 + w4 dz

+
k

k 2

+ x

s s (w1 + w1 dz

+ y

s s (w2 + w2 dz

+a

s s (w3 + w3 dz

+U

w1 w1 ( + t t + c 2

+V w4 t

w2 w2 ( + t t +
k

+
k w5

( dz (

w3 w3 + t t

k 2
s s w2 w2 + t t

t +a

+ x

s s w1 w1 + t t

+ y

s s w3 w3 + t t

(A.1)

w1 dz =

2 rc a2 + + c C 2 2 = 2i(a rc )

a2 ( + c )2

d (A.2)

170

w1 dz =

w1 dz
2 a2 rc + + c

=
C

2 2 rc a2 rc + 2 2 (rc + c )2

2 a4 rc ) 2 (rc 2 )2 a4 r 2 = 2i(a2 2 c 2 2 ) (rc )

= 2i(a2 +

(A.3)

w2 dz =

2 a2 rc i + c C 2 2 = 2(a + rc )

a2 ( + c )2

d (A.4)

w2 dz =

w2 dz i
2 a2 rc i + c

2 2 rc a2 rc + 2 2 (rc + c )2

2 a4 rc ) 2 (rc 2 )2 a4 r 2 = 2(a2 + 2 c 2 2 ) (rc )

= 2(a2 +

(A.5)

w3 dz =

2 2 2a2 rc 2a2 c a4 ( c ) 2c rc + + + 2 ( + c ) ( + c ) ( + c )(rc 2 ) a4 c 2 = 2 rc c + a2 c 2 (rc 2 )

i 2

a2 ( + c )2

(A.6)

171

w3 dz =

w3 dz i 2
2 2 2a2 rc 2a2 c a4 ( c ) 2c rc + + + 2 ( + c ) ( + c ) ( + c )(rc 2 ) 2 4 a4 r 2 c a6 c rc a4 rc a4 + 2 c 2 2 2 2 c c (rc 2 )2 (rc ) (rc 2 )3

2 2 rc a2 rc + 2 2 (rc + c )2

= 2 c a2 +

2 a6 c rc a4 c 2 2 (rc 2 ) (rc 2 )3 2 a4 c a6 c rc = 2 c a2 2 2 (rc 2 ) (rc 2 )3

= 2 c a2 +

(A.7)

s s s In order to perform the integrals involving logarithmic terms in w4 and w1 , w2 , w3 ,

we do integration by parts, in each case choosing the branch cut associated with the logarithmic potential to pass through the point zcut = 2a, i.e., the trailing edge of the foil.

+ c + log( ) dz = [log( )z] rc rc = [2izcut ]


C

z
C

d a2 ( + c ) d (A.8)

+ c +

= [2izcut ] 2ic

log(

+ c + c + ) dz = [log( )z] rc rc = [2izcut ] = [2izcut ]


C

d C ( + c ) a2 + c + ( + c ) z

d ( + c ) (A.9)

172

log(

+ c ) dz =

(c ) d C ( + c ) a2 c = ( + c + ) d + c ( + c ) C = [2ic ] z

(A.10)

w4 dz = 2(zcut c ) = 2(2a c ) (A.11)

s w1 dz =

2 2 2 a2 rc a4 c a2 rc + + 2 2 c ( + c ) ( + c )(rc 2 )2 2 2 a2 rc a4 rc a2 d 1 3 2 c c (rc 2 )2 ( + c ) ( + c )2 2a4 r2 c 2a4 r2 (iy ) 2 c2 3 log(( + c )/rc ) dz + 2 c 2 3 log( ) (rc ) (rc ) rc + c c a2 1 2 log( ) dz + 2 2a2 rc 3 2 )3 c (rc 2 2a2 rc a4 ( 2 + r2 ) 2a4 r2 c 2 = 2i rc + a2 + 2 c 2 c2 2 c 2 3 zcut 2 c (rc ) (rc ) 4 2 2 a2 2a r (iy ) 1 2 + 2 + 2 c 2 3 (zcut c ) 2a2 rc 2 (rc ) c (rc 2 )3 2 2a2 rc a4 ( 2 + r2 ) 4a5 r2 c 2 = 2i rc + a2 + 2 c 2 c2 2 c 2 3 2 c (rc ) (rc ) 4 2 1 2 a2 2a rc (iy ) 2 2 (2a c ) 2a rc + 2 + 2 (A.12) 2 (rc 2 )3 c (rc 2 )3

173

s w1

2 2 2 a2 a2 rc a4 c rc + dz = + 2 2 c ( + c ) ( + c )(rc 2 )2 2 2 2 r2 a4 rc a2 rc a2 rc c + 2 d 3 2 c c (rc 2 )2 ( + c ) 2 (rc + c )2 2a4 r2 c 2a4 r2 (iy ) log(( + c )/rc ) dz + 2 c 2 3 log( ) dz 2 c2 3 (rc ) (rc ) rc 1 + c c a2 2 log( 2a2 rc + 2 ) dz 2 )3 3 c (rc a4 r 2 a6 r 2 2 a4 r 4 a6 r 4 a4 = 2i a2 + 2 + 2 c 2 2 2 c c 4 + 2 2 c 2 2 2 c 2 4 c (rc ) (rc 2 ) c (rc ) (rc ) 4 2 2 4 2 2a r (iy ) a c 2a r c + 2 c 2 3 z cut c 2 c 2 3 z cut c + 2 2) (rc ) (rc (rc ) c c a2 1 2 (A.13) + 2 2a4 rc 3 2 2) 2 )3 c (rc (rc

s w1

dz = 2i a +

a4 c
2

2 2a4 rc c 2 (rc 2 )3

2 4 2 4 a4 rc a6 rc a6 rc c a4 rc 2 + 2 + 2 (rc 2 )2 (rc 2 )4 c 2 (r2 2 )2 (rc 2 )4 c

2a c +
3

2 +2a4 rc

c a2 2 (rc 2 )3

a2 c 2 (rc 2 )
2 (rc

c 2)

2 2a4 rc (iy ) (2a c ) 2 (rc 2 )3

(A.14)

174

s w2

2 2 2 a2 a2 rc a4 c rc + dz = (i) 2 2 c ( + c ) ( + c )(rc 2 )2 2 2 a2 a4 rc a2 rc 1 d + 3 2 c c (rc 2 )2 ( + c ) ( + c )2 2a4 r2 c + c 2a4 r2 x +i 2 c 2 3 log( log( ) dz ) dz i 2 c 2 3 (rc ) rc (rc ) rc 2 c a + c 1 2 log( i2a2 rc ) dz + 2 3 2 )3 c (rc 2 2a4 r2 c 2a2 rc a4 ( 2 r2 ) 2 = 2 rc + a2 2 c 2 c2 2 c 2 3 zcut 2 c (rc ) (rc ) 4 2 2a r x 1 2 a2 2 + 2 c 2 3 (zcut c ) + 2a2 rc + 2 2 (rc ) c (rc 2 )3 2 2a2 rc a4 ( 2 r2 ) 4a5 r2 c 2 = 2 rc + a2 2 c 2 c2 2 c 2 3 2 c (rc ) (rc ) 4 2 2 a2 1 2a r x 2 + 2 + 2 c 2 3 (2a c ) + 2a2 rc (A.15) 2 (rc ) c (rc 2 )3

s w2 dz =

2 2 2 a2 rc a2 a4 c rc + 2 2 c ( + c ) ( + c )(rc 2 )2 C 2 2 2 a2 rc a4 rc r2 a2 rc + c + 2 d 3 2 c c (rc 2 )2 ( + c ) 2 (rc + c )2 + c 2a4 r2 x 2a4 r2 c log( log( ) dz ) dz i 2 c 2 3 +i 2 c 2 3 (rc ) rc (rc ) rc 2 + c 1 c a 2 log( + 2 ) dz i2a2 rc 3 c (rc 2 )3 a4 a6 r 2 2 a4 r 4 a6 r 4 a4 r 2 = 2 a2 2 + 2 c 2 2 2 c c 4 2 2 c 2 2 + 2 c 2 4 c (rc ) (rc 2 ) c (rc ) (rc ) 2 4 2 4 2 a c 2a r x 2a r c + 2 c 2 3 z cut c 2 c 2 3 z cut c + 2 2) (rc ) (rc (rc ) 2 c a 1 c 2 +2a4 rc + 2 (A.16) 3 2 c (rc 2 )3 (rc 2 )

(i)

175

s w2

dz = 2 a

a4 c
2

2 2a4 rc c 2 (rc 2 )3

2 2 4 4 a4 rc a6 rc c a4 rc a6 rc + 2 2 + 2 (rc 2 )2 (rc 2 )4 c 2 (r2 2 )2 (rc 2 )4 c

2a c +
3

2 +2a4 rc

c a2 2 (rc 2 )3

a2 c 2 (rc 2 )
2 (rc

c 2)

2 2a4 rc x (2a c ) 2 (rc 2 )3

(A.17)

s w3 dz =

2a

2 a3 rc log(( + c )/rc ) dz 2 (rc 2 )2 2 ( + c ) r2 ((r2 2 )2 a2 c ) log( ) dz +2a c c2 2 2 )2 c (rc 2 r2 a2 c a2 r 2 r2 ((r2 2 )2 a2 c ) = 4ai c 2 2 c 2 2 zcut + c c 2 c (rc 2 ) (rc ) c (rc 2 )2 2 2 r2 a2 c 2a3 rc r2 ((r2 2 )2 a2 c ) = 4ai c 2 (A.18) 2 + c c 2 c (rc 2 ) (rc 2 )2 c (rc 2 )2

2 rc a2 c 2 c ( + c )(rc 2 )

a2 ( + c )2

s w3

2 2 2 rc rc a2 c a2 rc 2+ 2 dz = 2a 2 c ( + c )(rc 2 ) (rc + c )2 C a3 r 2 log(( + c )/rc ) dz 2 2 c 2 2 (rc ) 2 r2 ((r2 2 )2 a2 c ) ( + c ) +2a c c2 2 ) dz log( 2 )2 c (rc a2 a4 r 2 c = 4ai 2 c 2 3 c (rc ) 3 2 a rc a2 c 4i 2 z cut c + 2 (rc 2 )2 (rc 2 ) 2 r2 ((r2 2 )2 a2 c ) c +4ia3 c c2 2 2 2) 2 )2 c (rc (rc

(A.19)

176

s w3

2 a4 rc c a2 + 2 dz = 4ai (rc 2 )3 c a2 c a3 r 2 +4i 2 c 2 2 2a c + 2 (rc ) (rc 2 )

4ia

2 2 3 rc ((rc

2 c (rc 2 )2

2 ) 2 a 2 c )

2 (rc

c 2)

(A.20)

Now we evaluate the integrals involving time derivatives of the Kircho potentials.

w1 dz = t

2aa a2 x 2 rc rc a2 i y + + c ( + c )2 ( + c )2

dz

= 4i ( a a rc rc ) d 2 2 a rc = 2i dt

(A.21)

w1 dz = t

2 a a rc 2

rc 2 + c a2 2 x rc 2 + c
2

2 rc 3 rc rc 2 + c a2 i 2 y rc 2 + c
2 2

2 a a 2 c rc 2 + c
2

4 rc rc 2 c rc 2 + c
2

2 rc 3 c

2 2

rc rc 2 + c

dz

3 4 a2 i rc (rc 2 2 ) + a a (rc 5 + rc 3 2 ) + a2 rc 3 c x + i y

rc (rc 2 2 )3

(A.22)

w2 dz = 4 ( a a + rc rc ) t d 2 a + rc 2 = 2 dt

(A.23)

3 4a2 aarc 3 (rc 2 2 ) + rc (rc 2 2 ) + a2 rc 3 c x iy w2 dz = (A.24) t rc (rc 2 2 )3

177

2a4 rc rc c 4a3 ac w3 dz = 2 2rc rc c + 2aac + + r c 2 x 2 2 2 2 )2 t rc (rc a4 i 2 y a4 2 x a4 x a4 i y + irc 2 y (A.25) 2 2 (rc 2 2 )2 rc 2 (rc 2 2 )2 rc 2

2 a2 w3 dz = t rc (rc 2 2 )4 + rc 2 2
4

4 a3 a rc 3 c rc 2 2 2 a a rc c rc 2 2 2 rc c + rc x i y
2

a2 c rc 2 2

4 rc 2 rc 2 rc 2 + rc c x i y (A.26)

+a4 rc 3 2 rc rc c 2 2 x i y + rc 2 x + i y

w4 dz = t

rc

2 rc

rc dz = 0

k w5 dz = i t

k 2 rc

2 2rc rc rc k + 2 k k

2 2rc rc rc k 1 dz + 2 2 rc /k k k rc 2 rc k + rc k 2 a k = 2 + 2 (c + k )2 k

1 k k

(A.27)

s w1 dz = c t

s w1 dz + a c

s w1 dz + rc a

s w1 dz rc

(A.28)

178

s w1 dz = c

4a

i rc 2 c c 6 a rc 2 c y 3 rc 2 2 y 2 i rc 2 c 2 c + + (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4

2 a i rc 4 2 i rc 4 c 2 i rc 4 c 4 a i rc 2 2 (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4


2

(A.29)

s w1 dz = a

8 a4 i rc 2 c 4 a3 i rc 2 2 2 a3 i rc 2 (rc 2 2 )3 (rc 2 2 )3 (rc 2 2 )2 2 a3 i c 2 8 a 4 rc 2 y 4 a 3 rc 2 c y + (rc 2 2 )2 (rc 2 2 )3 (rc 2 2 )3 4 a i

(A.30)

s w1 dz = rc

2 a2 i rc 8 a5 i rc 3 c 2 a2 i rc 9 + 2 c 2 c (rc 2 2 )4 (rc 2 2 )4 4 a4 i rc 3 2 4 a5 i rc c 2 c 8 a2 i rc 7 c + + + c (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4 4 (i rc ) + 2 a4 i rc c 2 c 8 a2 i rc 3 c c 12 a2 i rc 5 c + + (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )4


4 2 2 3

2 a2 i rc c 2 c a4 i r c 3 2 a4 i rc c 2 + + (rc 2 2 )4 (rc 2 2 )3 (rc 2 2 )3 8 a 5 rc 3 y 4 a 4 rc 3 c y a4 i r c 2 + + (rc 2 2 )3 (rc 2 2 )4 (rc 2 2 )4 4 a 5 rc 2 y 2 a 4 rc c 2 c y + (rc 2 2 )4 (rc 2 2 )4

(A.31)

179

s 4a4 w1 2 a rc 6 + 3 a2 i rc 4 c + rc 6 c i rc 6 c dz = 2 2 )5 c (rc i rc 4 c 3 a2 rc 4 c + 2 a2 i rc 4 c 2 a rc 4 2 + 3 a2 i rc 2 c 2 c

+rc 4 c 2 c + 2 i rc 4 c 2 c +2 i rc 2 c 4 c 2 a2 rc 2 c c + a2 i rc 2 c c + 4 a rc 2 c 2 c 2 rc 2 c 3 c i rc 2 c 3 c i c 5 c 6 a rc 4 c y +3 rc 4 c 2 y + 6 a rc 2 c 2 c y 3 rc 2 c 3 c y (A.32)
2 2 2 2 2 2

s w1 4a3 2 2 dz = 2 a2 rc 2 2 c + i rc 2 + 2 2 + c + 8 a r c 2 rc 2 2 2 2 )4 a (rc + rc 2 2 i rc 2 + c 2 rc 2 2 4 r c 2 c c + y

c + y (A.33)

s w1 4a2 a4 i rc 6 i rc 10 + a2 i rc 6 c 2 8 a3 rc 6 c + 5 a4 i rc 4 2 dz = 2 rc rc (rc 2 )5

+4 a2 rc 6 2 + 5 i rc 8 2 3 a2 i rc 4 c 3 c 4 a4 rc 4 c + 2 a4 i rc 4 c
2 2 2 2

2 2

+4 a3 rc 4 c c + 2 a4 i rc 2 c 2 c 2 a2 rc 4 c 2 c 10 i rc 6 c 2 c + 3 a2 i rc 2 c 4 c 2 a4 rc 2 c c + 4 a3 rc 2 c 2 c 2 a2 rc 2 c 3 c
3 3 4 3 3 3

+10 i rc 4 c 3 c a2 i c 5 c 5 i rc 2 c 4 c + i c 5 c 8 a3 rc 6 y + 4 a2 rc 6 c y +4 a3 rc 4 2 y 2 a2 rc 4 c 2 c y + 4 a3 rc 2 c 2 c y 2 a2 rc 2 c 3 c y
2 2

(A.34)

2 s 4a4 rc w2 dz = 2 rc 2 c c + 2 2 c + c 3 i y 2 a rc 2 + 2 2 3 i c y 2 c (rc 2 )4 (A.35)

180

s 32 a4 rc 2 c 16 a3 rc 2 2 w2 dz = 4 a + a (rc 2 2 )3 (rc 2 2 )3 8 a3 c 2 16 a3 i rc 2 (2 a c ) y 8 a3 r c 2 + (A.36) + (rc 2 2 )2 (rc 2 2 )2 (rc 2 2 )3

s 6 a 4 rc 2 2 2 a 4 rc 2 4 a5 c 2 a4 c 2 12 a5 rc 2 c w2 dz = 4rc 1 + + rc (rc 2 2 )4 (rc 2 2 )4 (rc 2 2 )3 (rc 2 2 )3 (rc 2 2 )3 2 a4 2 a4 12 a5 i rc 2 y 6 a4 i rc 2 c y + + + (rc 2 2 )3 (rc 2 2 )2 (rc 2 2 )4 (rc 2 2 )4 4 a5 i y 2 a4 i c y (A.37) + (rc 2 2 )3 (rc 2 2 )3

s 4a4 w2 a2 rc 2 2 3 c + c + 2 i c + rc 2 3 c + (2 + i) c dz = 2 c (rc 2 )5 +2 a i rc 2 rc 2 2 rc 2 + 2 2 + 3 c y

c rc 2 2

(1 + i) rc 4 + c 3 c rc 2 c c + c 2 i c 3 i y (A.38)

s w2 4a3 dz = 2 a2 rc 2 rc 2 c 2 c + c + 2 i c 2 a (rc 2 )4 8 a i rc 2 rc 2 2 c + y

+ rc 2 2

rc 4 + c 3 c r c 2 c c + c 4 i c 4 i y

(A.39)

181

s a2 a4 rc 5 1 w2 12 a2 rc 5 c 2 2 dz = 4a + rc rc rc c 2 (rc 2 2 )5 (rc 2 2 )5 8 a 2 rc 7 c 5 a 4 rc 3 2 2 a 2 rc 9 + 2 5 (rc 2 2 )5 c (rc 2 2 )5 c (rc 2 2 )

a2 rc 3 2 a2 i rc c 2 c 2 (rc 2 2 )4 c (rc 2 2 )2
2

4 a2 i rc 3 2 4 a3 i rc c c + (rc 2 2 )4 (rc 2 2 )4
2

2 a 2 rc c 4 c 2 a4 i rc c c 8 a3 i rc 3 c + + + (rc 2 2 )5 (rc 2 2 )5 (rc 2 2 )4


2

8 a 2 rc 3 c 3 c 4 a 4 i r c 3 c 2 a 4 rc c 2 c + (rc 2 2 )5 (rc 2 2 )5 (rc 2 2 )5


2 3

8 a3 i rc 3 y 2 a 4 rc 3 c + (rc 2 2 )2 (rc 2 + 2 )3 (rc 2 2 )4 4 a2 i rc 3 c y 4 a3 i rc 2 y + (rc 2 2 )4 (rc 2 2 )4 2 a2 i rc c 2 c y (rc 2 2 )4

(A.40)

s 4 a3 i rc 2 3 rc 2 + 4 a c c c w3 dz = 3 c rc 2 c c

(A.41)

s 12 a2 i 2 a rc 2 + c 2 rc 2 c c w3 dz = 2 a rc 2 c c s 16 a3 i rc rc 2 c + a rc 2 + c c w3 dz = 3 rc rc 2 c c

(A.42)

(A.43)

s 4a3 w3 2 4 a rc 2 c rc 2 2 + c rc 2 2 dz = 2 2 )4 c (rc a2 rc 2 3 i rc 2 + 2 2 + 4 i 2

(2 + i) rc 2 i 2 (A.44)

182

s 4a2 w3 6 a rc 4 + 3 rc 4 c i rc 4 c 3 a2 rc 2 c + 6 a2 i rc 2 c dz = 2 2 )3 a (rc

+6 a rc 2 2 3 rc 2 c 2 c + 2 i rc 2 c 2 c i c 3 c

(A.45)

s 8a3 w3 a2 rc 2 c (1 + 2 i) rc 2 + (1 + i) 2 dz = 2 2 )4 rc rc (rc

+c rc 2 2

(1 + i) rc 4 + (1 2 i) rc 2 2 + i c 2 c
2

2 a rc 6 rc 2 c 2 c

(A.46)

183

Appendix B Eect Forces

Integrals for Bernoulli

These integrals arise in the force on a deformable Joukowski foil due to the Bernoulli eect.

C+

dw1 dw1 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
3 3

4 a2 i c (rc 2 2 ) (rc 2 2 ) + a2 rc 4 + c c rc 2 rc 2 + (a c ) c
2

rc 2 (a + c ) c

(B.1)

C+

dw1 dw2 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
3 3

4 a2 c (rc 2 2 ) (rc 2 2 ) + a2 rc 4 + c c rc 2 rc 2 + (a c ) c
2

rc 2 (a + c ) c

(B.2)

C+

dw2 dw1 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
3 3

4 a2 c (rc 2 2 ) (rc 2 2 ) + a2 rc 4 + c c rc 2 rc 2 + (a c ) c
2

rc 2 (a + c ) c

(B.3)

184

C+

dw2 dw2 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
3 3

4 a2 i c (rc 2 2 ) (rc 2 2 ) + a2 rc 4 + c c rc 2 rc 2 + (a c ) c
2

rc 2 (a + c ) c

(B.4)

C+

dw1 dw3 d d

4 a 2 rc 4 2 r c 2 2 + a2 + c 2 c

( + c )2 ( + c )2 a2

2 rc 2 2

d = (rc 2 2 ) + a2 rc 4 c c rc 2 (a + c ) c
2 3 3

rc 2 rc 2 + (a c ) c

(B.5)

C+

dw3 dw1 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
4 3 4

4 a2 c a2 rc 2 c (2 rc 2 3 2 ) (rc 2 2 ) c (rc 2 2 ) + a4 rc 4 c rc 2 c + c c rc 2 rc 2 + (a c ) c
2

rc 2 (a + c ) c

(B.6)

C+

dw2 dw3 d d

4 a2 i rc 4 2 rc 2 2 + a2 + c 2 c rc 2 rc 2 + (a c ) c
2

( + c )2 ( + c )2 a2

2 rc 2

d =
2

(rc 2 2 ) + a2 rc 4 + c c rc 2 (a + c ) c
2

(B.7)

185

C+

dw3 dw2 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
4 3 4

4 a2 i c a2 rc 2 c (2 rc 2 3 2 ) (rc 2 2 ) c (rc 2 2 ) + a4 rc 4 c rc 2 c + c c rc 2 rc 2 + (a c ) c
2

rc 2 (a + c ) c

(B.8)

C+

dw3 dw3 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
2

4 a 2 i rc 4 2 r c 2 2 + a2 + c 2 c
4

3 4 2

a2 rc 2 c (2 rc 2 3 2 ) (rc 2 2 ) c (rc 2 2 ) + a4 rc 4 c rc 2 c + c c rc 2 rc 2 + (a c ) c
2

(rc 2 2 ) rc 2 (a + c ) c

(B.9)

C+

dw1 dw4 d d

( + c )2 ( + c )2 a2

2 rc 2

d = 0

(B.10)

C+

dw4 dw1 d d +

( + c )2 ( + c )2 a2 a3 rc 4

2 rc 2

d = 2 rc
2

2 r c 2 c 2 2 a 2 rc 2 + c 2 (a c ) (a + c ) a3 rc 4
2

(a c ) rc 2 + (a c ) c dw2 dw4 d d

(a + c ) rc 2 (a + c ) c
2 rc 2

(B.11)

C+

( + c )2 ( + c )2 a2

d = 0

(B.12)

186

C+

dw4 dw2 d d

( + c )2 ( + c )2 a2

2 rc 2 4

d =
2 4

2 i rc 2 (rc 2 2 ) + a4 rc 4 c + (rc c ) (rc + c ) c rc 2 rc 2 + (a c ) c 2i a2 (rc 2 2 )


2 2

rc 4 + c 2 c 2 r c 2 c c + c
2

rc 2 (a + c ) c
2

rc 2 rc 2 + (a c ) c ( + c )2 ( + c )2 a2

rc 2 (a + c ) c d = 0

(B.13)

C+

dw3 dw4 d d

2 rc 2

(B.14)

C+

dw4 dw3 d d +c +

( + c )2 ( + c )2 a2 a4 rc 2

2 rc 2

d = 2 i
2

a2 c a2 + c 2 rc 2 a2 c 2 a4 rc 2
2

(a c ) rc 2 + (a c ) c dw4 dw4 d d

(a + c ) rc 2 (a + c ) c
2 rc 2

(B.15)

C+

( + c )2 ( + c )2 a2

d = 0

(B.16)

C+

k dw1 dw5 d d

( + c )2 ( + c )2 a2

2 rc 2

d =

2 a2 k

+ r c 2 c + k
2

a c k k

a + c + k (B.17)

187

C+

k dw5 dw1 d d

( + c )2 ( + c )2 a2
4 2 2

2 rc 2

d =
2

2 rc 2 rc 2 2 a2 rc 2 2
3

(c + k )2 c + k
2 2

a 6 c
2

4 3

rc 2 + c k
2

2 2 2

(c + k )2 rc 4 k + c 2 c k + 2 rc 2 c c + 2 c k c k + c k 2 c 2 + 2 c k + k 2 k r c 8 c + 2 k
2

+a4 c c 2 c

+2 rc 6 k c + k
2

+ 2 c c c + 2 k
2

rc 4 c k 2 k 2 c + k + c 2 3 c + 8 c k 2 k +rc 2 c
2

+ 2 c k c + 4 c k + 2 k
2

2 c 3 c 2 k k + c k 2 c + k
2 2 2

+ 2 c c k c + 2 c k + k k + k

+c 2 c c + 2 c k + k 4 k + k / rc 2 rc 2 + (a c ) c rc 2 (a + c ) c (a c k ) (a + c + k ) c + k
2

(B.18)

C+

k dw2 dw5 d d

( + c )2 ( + c )2 a2

2 rc 2

2 i a2 k + rc 2 c + k a c k k
2

a + c + k (B.19)

188

C+

k dw5 dw2 d d

( + c )2 ( + c )2 a2
2

2 rc 2

d = 2 i
2

rc 2 rc 2 + (a c ) c +a2 rc 2 c c
2

rc 2 (a + c ) c
4 2

(a c k ) (a + c + k ) c + k
2

rc 2 rc 2 c c
3

(c + k )2 c + k
2 2

+ a 6 c

rc 2 + c k
3

(c + k )2 rc 4 k + c 2 c k 2 rc 2 c + 2 c k + c c + c k 2 c 2 + 2 c k + k 2 k
2

+a4 c c 2 c
2 2 3

+ rc 8 c + 2 k
2 2

+rc 4 c c c (3 c + 2 k ) + 2 c (2 c + k )2 k + 2 c 2 + 4 c k + k 2 k +rc 2 c
2

c (c + k )2 + 2 c c + c (c + k )2 k + c 2 4 c + c + 2 c c k + c k 2 k 2 rc 6 k c + k + 2 c c c + 2 k (B.20)

C+

k dw3 dw5 d d 2

( + c )2 ( + c )2 a2

2 rc 2

d =
2

2 i a4 c k rc 2 c rc 2 c c rc 2 c c k

c + k
2

a 2 rc 2 c c

c k + r c 2 c + 2 k

a + c + k

a + c + k (B.21)

189

C+

k dw5 dw3 d d

( + c )2 ( + c )2 a2
2

2 rc 2

d = 2i
2 2

rc 2 rc 2 + (a c ) c a6 c
2 3 2

rc 2 c c
5

rc 2 (a + c ) c
2

(a + c + k ) (a + c + k ) c + k
5

rc 2 c rc 2 c c

(c + k )2 c + k

a 8 c
2

rc 2 + c k

rc 2 c c
2

c c (c + k )2 k + 2 rc 6 c + k
2

rc 2 c c (c + k )2 + 2 c 2 c k + 2 c 2 k +a2 rc 2 c c
2 2 2

2 rc 4 c + k
2

2 c c + k c + k
2 2

(c + k )2 c 3 c k + rc 6 c + 2 k + rc 4 c 2 c 4 c k + k c + k + c 2 c + 2 c k 2 k
4 2 2 2

+rc 2 c 2 c
3 2

a4 rc 2 c c rc 2 c c 2 c 2 + c
2 2 2

c 5 c k + rc 10 c + 2 k (c + k )2 + 2 c 3 c 2 + c
2 2 2

(c + k )2 k k
2

+ c 4 8 c 3 k + 2 c c k + c k 2 + c 2 c 4 k 2 rc 8 2 k c + k + 5 c c c + 2 k r c 4 c c
2

4 k 2 c + k
2

c + 2 k

+2 c k c + k

5 c + 9 k + c 2 7 c + 18 c k + k c + 2 k + 4 c k c + k
2

+rc 6 c 2 k 2 c + k

2 c + 3 k
2

+c 2 9 c + 20 c k + k

(B.22)

C+

k dw4 dw5 d d

( + c )2 ( + c )2 a2

2 rc 2

2 i c + k k a + c + k

a + c + k (B.23)

C+

k dw5 dw4 d d

( + c )2 ( + c )2 a2

2 rc 2

2 i (c + k )2 d = (B.24) k (a + c + k ) (a + c + k )

190

C+

j k dw5 dw5 d d

( + c )2 ( + c )2 a2 a + c + j c + j
2

2 rc 2

d =

a c j

2i (a c k ) (a + c + k ) (rc 2 + j k ) rc 2 + j k + c j 2 j k + c j + k k r c 2 c 2 j k j + k (B.25)

(c + k )2 j j + k k + rc 2 j + k

+a2 rc 4 2 c + j + k

C+

k k dw5 dw5 d d

(c + k )2 c + k

( + c )2 ( + c )2 a2
2

2 rc 2

2 d = 2 i rc k k

k + k

+ a 2 rc 2 2 c + k + k + c 2 k k + c k + k a c k a + c + k rc 2 k
2

(a c k ) (a + c + k ) rc 2 + k 2

(B.26)

C+

s dw1 dw1 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
3

4 a2 i a2 rc 4 c + c + c (rc 2 + 2 ) rc 2 (rc 2 2 )3

(B.27)

191

C+

s dw1 dw1 d d

( + c )2 ( + c )2 a2

2 rc 2

d = 4 a2 i
2

rc 2 rc 2 + (a c ) c c rc 2 + 2 a6 rc 2 c a4 rc 2 2
2 2 7

+ a 2 rc 2 2
6

(rc 2 2 )3 rc 2 (a + c ) c
3 4 2 2

2 rc 2 c 2 c c + r c 4 c + c i y c + i y

rc 4 (c i y ) + rc 2 c (c + i y ) + c 2 c
2 3

c c

+ r c 6 c + c i y + r c 4 c c c + 4 c + 2 i c c y +rc 2 c c c 2 c + i y (B.28)

C+

s dw2 dw1 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
3

4 a2 a2 rc 4 c + c + c (rc 2 2 ) rc 2 (rc 2 2 )3

(B.29)

C+

s dw1 dw2 d d

( + c )2 ( + c )2 a2
2

2 rc 2

d =
2

rc 2 rc 2 + (a c ) c +a6 rc 2 c
2

4 a2
2

(rc 2 2 )3 rc 2 (a + c ) c
3 3 4

c rc 2 + 2
2

c rc 4 + r c 2 c + c c +a2 rc 2 2
4 2 6

+ i rc 4 + (rc c ) (rc + c ) c

2 rc 2 c 2 c c + rc 4 c + c + i y
2 3

+a4 rc 2 2

c c rc 6 c + c i y + rc 4 c c 2 2 i c + c y +rc 2 c c + c 2 c + i y (B.30)

192

C+

s dw3 dw1 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
4 2

4 a2 a4 rc 4 c c + c c c (rc 2 2 ) + a2 (rc 2 2 ) rc 2 (rc 2 2 )4

rc 4 + 2 r c 2 c c c

(B.31)

C+

s dw1 dw3 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
2

rc 2 rc 2 + (a c ) c rc 2 2
3 9

4 a2

a 2 rc 2 2 +a8 rc 2 c c
4 6

(rc 2 2 )4 rc 2 (a + c ) c
3 4

rc 6 2 r c 2 c c + c 2 c + r c 4 c c i y
2 3 2

rc 2 (c i y ) + c c c + i y + r c 2 c c
7 2 3

+a4 rc 2 2

rc 8 c 2 c r c 4 c 2 c 3 + 3 c 2 c 3 c c + c + 2 i c 2 + c
2

+rc 6 2 c 2 c + c c + i y +a6 c rc 2 2
6 2 2

2 c + c 2 c + i y

2 rc 2 c c + c 2 c + rc 8 4 c + c 4 i y + rc 6 c c 5 c + c + 8 i y c 3 c c + c
2

+rc 4 c

c + i y 2 c 2 c + 2 i y

(B.32)

C+

s dw4 dw1 d d

( + c )2 ( + c )2 a2
s dw1 dw4 d d

2 rc 2

d =

2 (a rc ) (a + rc ) rc 2

(B.33)

C+

( + c )2 ( + c )2 a2

2 rc 2

d = 0

(B.34)

193

C+

k s dw5 dw1 d d

( + c )2 ( + c )2 a2
3

2 rc 2 2

d = a 2 rc 2 c c
2 3

rc 2 rc 2 c c

k (c + k )2 c + k
3

k (c + k )2 rc 2 + k

a4 rc 8 k + c 3 c k k + rc 6 c k 3 c + c + 2 k +rc 4 c c
2

rc 2 c 2 c k k 2 c c + c k + 3 c k c 2 k (2 c + k ) + 2 c 2 c c c k + c k 2 k 2 k
2

+ 2 c k (c + 2 k ) k rc 2 rc 2 c c
3

+ 2 i (c + k )2 c + k 2

y
2

k (a + c + k ) (a + c + k ) c + k

(B.35)

C+

s k dw1 dw5 d d

( + c )2 ( + c )2 a2 rc 2 c c
3 3

2 rc 2 2

d = 2

a + c + k
2

a + c + k
3

rc 2 rc 2 c c

c + k

a 2 rc 2 c c

rc 2 + k
2

2 2

+a4 k rc 4 k c 3 c k + rc 2 c 2 c c c k + c k 2 k

+ 2 i c + k

y (B.36)

C+

s dw1 dw2 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
3

4 a2 a2 rc 4 c c + c (rc 2 2 ) rc 2 (rc 2 2 )3

(B.37)

194

C+

s dw2 dw1 d d

( + c )2 ( + c )2 a2
2 4

2 rc 2

d =
2

4 a2 2 )3
3

rc 2 rc 2 + (a c ) c +a2 rc 2 2 +a6 rc 2 c
2 2

(rc 2

rc 2 (a + c ) c
4 2

c rc 2 2

2 rc 2 c + 2 c c + rc 4 c + c i y c i y + rc 4 (c + i y )
2 3 2 6

rc 2 c (c i y ) + c 2 c c c

+a4 rc 2 2

+ rc 6 c + c + i y c + c 2 c + i y (B.38)

rc 4 c c 2 + 2 i c c y + rc 2 c

C+

s dw2 dw2 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
3

4 a2 i a2 rc 4 c c + c (rc 2 + 2 ) rc 2 (rc 2 2 )3

(B.39)

C+

s dw2 dw2 d d

( + c )2 ( + c )2 a2
2

2 rc 2

d =
2

4 a2 i

rc 2 rc 2 + (a c ) c +a6 rc 2 c
2

c 2 c

(rc 2 2 )3 rc 2 (a + c ) c
3 4

c rc 2 2
2

c i y + rc 4 (c + i y ) + rc 2 c (c + i y )
4 6

+a2 rc 2 2 +a4 rc 2 2
2

2 rc 2 c + 2 c c + rc 4 c + c + i y + rc 6 c + c + i y + rc 4 c c c 4 c
2 3

c c

2 i c + c y + rc 2 c

c + c 2 c + i y

(B.40)

195

C+

s dw3 dw2 d d

( + c )2 ( + c )2 a2

2 rc 2

d =
4 2

4 a2 i a4 rc 4 c c c + c c (rc 2 2 ) + a2 (rc 2 2 ) rc 2 (rc 2 2 )4

rc 4 2 r c 2 c + c c

(B.41)

C+

s dw2 dw3 d d

( + c )2 ( + c )2 a2
2

2 rc 2

d =
2

rc 2 rc 2 + (a c ) c +a2 rc 2 2 +a4 rc 2 2
3 2 5

4 a2 i (rc 2

2 )4

+a8 rc 2 c c

rc 2 (a + c ) c
3 4

rc 2 2

c c c i y + rc 2 (c + i y )
2 3 2

rc 6 + 2 r c 2 c c c 2 c r c 4 c c + i y
6

rc 8 c 2 c rc 4 c 2 c 3 + 3 c 2 c + 3 c c + c 2 i c 2 + c + r c 2 c c
7 2

+rc 6 2 c 2 + c + c c i y a6 c rc 2 2
6 2

2 c + c 2 c i y
2

2 rc 2 c c c 2 c r c 6 c c 5 c + c 8 i y c 3 c c + 2 c 2 c 2 i y + c
2

+rc 8 4 c c 4 i y + rc 4 c

c + i y (B.42)

C+

s dw4 dw2 d d

( + c )2 ( + c )2 a2
s dw2 dw4 d d

2 rc 2

d =

2 i (a2 + rc 2 ) rc 2

(B.43)

C+

( + c )2 ( + c )2 a2

2 rc 2

d = 0

(B.44)

196

C+

k s dw5 dw2 d d

( + c )2 ( + c )2 a2
3

2 rc 2

d = 2i
2 2 2

rc 2 rc 2 c c rc 2 rc 2 c c
3 3 2

k (a + c + k ) (a + c + k ) c + k + a 2 rc 2 c c
3

k (c + k )2 c + k

k (c + k )2 rc 2 + k

+a4 rc 8 k + c 3 c k k + rc 6 c k 3 c + c + 2 k + rc 2 c 2 c k k 2 c c + c k 3 c k +rc 4 c c (2 c + k ) c + 2 k + 2 c 2 c c + c + c k + 2 k 2 k + 2 c + k (c + 2 k ) k
2 2 2

+ 2 i (c + k )2 c + k

(B.45)

C+

s k dw2 dw5 d d

( + c )2 ( + c )2 a2 rc 2 c c
3 3

2 rc 2 2

d = 2i

a + c + k
2

a + c + k
3

rc 2 rc 2 c c

c + k

+ a 2 rc 2 c c

rc 2 + k
2

2 2

+a4 k rc 4 k + c 3 c k + rc 2 c 2 c c + c + c k + 2 k

2 i c + k

y (B.46)

C+

s dw1 dw3 d d

( + c )2 ( + c )2 a2
2

2 rc 2 4ai

d =
2 2

rc 2 rc 2 + (a c ) c rc 2 rc 2 2

(rc 2
2

2 )2

rc 2 (a + c ) c
4 2

2 a 6 rc 4 c
3 3

+ a 2 rc 2 2 rc 2 2

rc 4 + 5 r c 2 c 2 c c
2

+a4 c

5 rc 4 2 rc 2 c + 2 c c

(B.47)

197

C+

s dw3 dw1 d d

( + c )2 ( + c )2 a2
2

2 rc 2

d =
2

rc 2 rc 2 + (a c ) c +a2 rc 2 2
2

4 a3 i (rc 2

2 )2

rc 2 (a + c ) c
4

a4 rc 2 c
3

rc 4 r c 2 c + c 2 c

rc 2 2
5

rc 4 + 3 r c 2 c 2 c c

rc 6 2 c c + 2 r c 4 c 2 c + c + r c 2 c 2 c

+ c

(B.48)

C+

s dw2 dw3 d d

( + c )2 ( + c )2 a2
2 2

2 rc 2 4a

d =
2 2

rc

rc + (a c ) c +rc 2 rc 2 2

(rc
2

2 )2

rc (a + c ) c
4 2

2 a 6 rc 4 c
3 3

+ a 2 rc 2 2

rc 4 5 r c 2 c + 2 c c
2

a4 c

rc 2 2

5 rc 4 2 r c 2 c + 2 c c

(B.49)

C+

s dw3 dw2 d d

( + c )2 ( + c )2 a2 4 a3
2

2 rc 2

d =
2

rc 2 rc 2 + (a c ) c +a2 rc 2 2
2

(rc 2

2 )2

rc 2 (a + c ) c
4

a4 rc 2 c
3

rc 4 + rc 2 + c 2 c

+ rc 2 2

rc 4 3 r c 2 c + 2 c c
5

rc 6 + 2 r c 4 2 c c c 2 c c + r c 2 c 2 c

+ c

(B.50)

198

C+

s dw3 dw3 d d

rc

rc + (a c ) c rc 2 c rc 2 2
4

2 rc ( + c )2 ( + c )2 a2 2 4 a 2 7

d =
2

(rc
3

2 )3

rc (a + c ) c
5 2

2 a 8 rc 4 c c
2 3 4

+ a 2 c rc 2 2 a 6 c

rc 4 + 4 r c 2 2 2 c 2 c
2

+a4 rc 2 rc 2 2

rc 2 c 7 c

rc 2 2

5 rc 4 c 4 r c 2 c + 2 c c

(B.51)

C+

s dw3 dw3 d d

( + c )2 ( + c )2 a2
2 2

2 rc 2

d =
2

rc

rc + (a c ) c

4 a3 (rc

2 )3
5 3

rc (a + c ) c

a6 rc 2 c c
2 4

rc 2 + 2

+c rc 2 + 2 +a2 rc 2 c rc 2 2 +a4 c rc 2 2

4 rc 4 5 r c 2 2 + 2 c 2 c
2 2

5 rc 4 6 r c 2 2 + c 2 c + 2 c
5

4 rc 8 2 r c 6 2 2 r c 4 c 2 c 3 r c 2 c c + 2 c 2 c

(B.52)

C+

s dw4 dw3 d d

4 a3 c 2 rc 6 5 rc 4 2 + 4 rc 2 c 2 c + (a c ) c (a + c ) c rc 2 rc 2 + (a c ) c
s dw3 dw4 d d 2

( + c )2 ( + c )2 a2

2 rc 2

d =
2 3

rc 2 (a + c ) c
2 rc 2

(B.53)

C+

( + c )2 ( + c )2 a2

d = 0

(B.54)

199

C+

k s dw5 dw3 d d

( + c )2 ( + c )2 a2
2

2 rc 2 2

d = 4 a
2 2

rc 2 rc 2 + (a c ) c rc 2 rc 2 c c

rc 2 c c
2

rc 2 (a + c ) c

k (a + c + k ) (a + c + k ) c + k
4

k (c + k )2 c + k a2 rc 2 c c
4

c c k (c + k )2 k

2 2

+rc 6 k c + k + rc 2 c k 2 c (c + k )2 + c k (2 c + k ) k 3 c 2 + 4 c k + 2 k 2 k +rc 4 c c + k 2 k k + c c + k
2

a 6 c

3 rc 2 c 2 c k k + c 3 c k k
2

+rc 6 k c + 2 k + rc 4 c 2 c + k +a4 c
2

k 2 c + k

c k c + 2 c k 2 k
2

rc 2 c c

c c k 2 c 2 + 2 c k + k 2 k + r c 6 k 4 c + 5 k c (c + k )2 3 c 2 k
2

+rc 2 c k c + 2 k +rc 4 2 c 2 c + k
2

3 k 2 c + k

+ c k 5 c 4 c k + 4 k

(B.55)

C+

s k dw3 dw5 d d

( + c )2 ( + c )2 a2
2

2 rc 2

d =
2

4 a c rc 3 r c c c

+ r c 2 a2 c 2 + rc 2 c c rc 2 c c
2

k + a 2 c rc 2 c c k a 2 rc 2 k a + c + k

a + c + k

(B.56)

C+

s s dw1 dw1 d d

rc 2 rc 2 c c

4 a2 i

( + c )2 ( + c )2 a2
5

2 rc 2 5

d = + a 2 rc 2 c c
2

c rc 2 c c

2 rc 2 c + c c
2

+rc 4 2 c + c + i y

+ a 4 rc 2 rc 4 2 c + c i y c 2 c
2

c + i y (B.57)

+2 rc 2 c c 2 + c c + c + i c y

200

C+

s s dw1 dw2 d d

( + c )2 ( + c )2 a2
5

2 rc 2

d =
2

rc 2 rc 2 c c
3 4

4 a2

c rc 2 + c c

+ a 2 rc 2 c c
2

2 rc 2 c c c + rc 4 (2 c + i y )

+a4 rc 2 rc 4 c + i y + c 2 c

c + i y + 2 r c 2 c c 2 c c + i y (B.58)

C+

s s dw2 dw1 d d

( + c )2 ( + c )2 a2
5 2

2 rc 2

d =
2

4 a2 rc 2 rc 2 c c
3 4 5

c rc 2 c c +a4 rc 2 c 2 c

+ a 2 rc 2 c c

2 rc 2 c + c c + rc 4 (2 c i y )
2

c + i y + rc 4 c + i y 2 rc 2 c c 2 c + i c y (B.59)

C+

s s dw2 dw2 d d

( + c )2 ( + c )2 a2
5 2

2 rc 2 2

d =

4 a2 i rc 2 rc 2 c c
3 4 5

c rc 2 c c +a4 rc 2 c 2 c

+ a 2 rc 2 c c

2 rc 2 c + c c + rc 4 2 c + 2 c + i y
2

c + i y + rc 4 2 c + c + i y + 2 rc 2 c c 2 c c + c i c y (B.60)

201

C+

s s dw1 dw3 d d

( + c )2 ( + c )2 a2 rc 2 rc 2 + (a c ) c

2 rc d = 2 4 a i 2

rc 2 rc 2 c c +a4 rc 2 c c +2 a8 rc 2 c +a6 c rc 2 c c
2 4 4

rc 2 c c
5

a 2 rc 2 c c
4

rc 2 (a + c ) c
2

2 3 2

rc 4 + 6 r c 2 c 2 c c

rc 6 + 9 r c 2 c 4 c c + r c 4 3 c 2 + 2 c c + 5 c rc 4 c 2 c c + i y + r c 2 c 2 c + c + i y
6

2 c c + rc 6 4 c 5 c + 4 i y rc 4 c 9 c 2 + 10 c c + 2 c + 6 i c y +2 rc 2 c
2

2 c + c 2 c + i y

(B.61)

C+

s s dw3 dw1 d d

( + c )2 ( + c )2 a2
2

2 rc 2 2

d =
3

rc 2 rc 2 c c

4 a3 i

4 2

rc 2 c c

2 rc 4 + 3 r c 2 c 2 c c

+ a 2 rc 6 r c 2 c 2 c + 4 r c 4 c c + c (B.62)

C+

s s dw2 dw3 d d

( + c )2 ( + c )2 a2 rc 2 rc 2 + (a c ) c

2 rc 2 2

d = 4a
4 2 3 2

rc 2 rc 2 c c +a4 rc 2 c c +2 a8 rc 2 c +a6 c rc 2 c c
2 4 4

rc 2 c c
5

+ a 2 rc 2 c c
4

rc 2 (a + c ) c
2

rc 4 6 r c 2 c + 2 c c

rc 6 + 9 rc 2 c 4 c c + rc 4 3 c 2 + 2 c c 5 c rc 4 rc 2 c 2 c 2 c + i y + c 2 c c + i y
6

2 c c + rc 6 4 c 5 c 4 i y + rc 4 c 9 c 2 10 c c + 2 c + 6 i c y +2 rc 2 c
2

2 c + c 2 c i y

(B.63)

202

C+

s s dw3 dw2 d d

( + c )2 ( + c )2 a2

2 rc 2 2

d =

4 a3 rc 2 rc 2 c c
2 4 2 3

a2 rc 6 4 r c 4 c c c r c 2 c 2 c

+ rc 2 c c

2 rc 4 3 r c 2 c + 2 c c

(B.64)

C+

s s dw3 dw3 d d

( + c )2 ( + c )2 a2 rc 2 rc 2 + (a c ) c a2 rc 2 c c

2 rc 2

d = 8 a2 i
3 2

rc 2 c rc 2 c c
4 2

rc 2 c c

+ a 6 rc 2 c c

rc 2 (a + c ) c 5 rc 2 + c c
3 4

3 rc 4 c + 6 r c 2 c 2 c c
5

+a4 c rc 2 c c

2 rc 6 + 12 rc 4 c c 2 c c + rc 2 c 2 c

+ 3 c

(B.65)

203

Appendix C

Coecients in

Equation (3.156)
B1 = 4 2 a 2 rc 2 2 r c 2 rc 2 2 2 )4 4 a5 rc 2 rc 2 2 c i c i y i y
4

2 (rc

+ a 6 rc 2 rc 2 + c 2 c + c + 2 i c

B2

2 a4 rc 2 2 rc 4 c 3 c + rc 2 c c + i c + 2 y = 4 a 5 rc 2 i c c rc 2 2 2 a 2 i rc 2 2 2 2 )4 (rc 2 a4 rc 2 2 rc 2 2 + i rc 4 rc 2 c 2 + c 3 c 2a3 a2 (2 + i) rc 2 c + 2 a (1 + i) rc 2 rc 2 2 = 2 2 )3 (rc c (3 + i) rc 2 2 2 rc 2 2 a6 rc 2 2 c + i rc 2 + 2 c c c


2

i r c 2 rc 2 2

B3

204

C1 = C2 = C3 =

2 a2 c (rc 2 2 ) (rc 2 2 ) + a2 rc 4 + c 3 c 4 a2 i c (rc 2 2 ) rc 6 3 rc 4 2 + 3 rc 2 c 2 c + c 3 c a2 c 2 a2 c (rc 2 2 ) (rc 2 2 ) + a2 rc 4 + c 3 c rc 2 rc 2 + c a c rc 2 2 2


2

rc 2 rc 2 + c a c

rc 2 c a + c
2

rc 2 rc 2 + c a c
3

rc 2 c a + c
2

C 4 = 2 a2 i

rc 2 2
2

rc 2 c a + c
4

+ a 4 c rc 2 c 3 + r c 4 c c
2

2 c 4 c + a2 rc 8 4 rc 2 c 4 c + rc 4 c 2 c 7 c + c rc 6 c 3 c + 2 c / rc 2 rc 2 + c a c C 5 = 2 a2 rc 2 2 2
2

rc 2 c a + c
4 2

rc 2 2

+a2 rc 8 + rc 6 c 3 c 2 c + 4 rc 2 c 4 c + rc 4 c 2 c 7 c + c +a4 c rc 2 c 3 + 2 c 4 c + rc 4 c + c / rc 2 rc 2 + c a c C 6 = 2 a2 c rc 2 2 +a2 c rc 2 2


3 3 2

rc 2 c a + c
2

+ a 6 rc 2 c 5 + r c 4 c 2 c + c 6 c
2

2 rc 4 4 r c 2 2 + c 2 c
2

+a4 rc 8 c + c 6 c + rc 6 c 3 2 c c / rc 2 rc 2 + c a c rc 2 2

+ rc 4 2 c 4 c + c 2 c
2

rc 2 c a + c

205

D 1 = 2 a 2 2 c rc 2 2 +a4 rc 2 2
2

a 2 rc 2 2
2

4 rc 2 c 3 4 c 4 c + r c 4 3 c + 2 c i y

+a6 rc 2 c 2 rc 4 c i y + c 2 c (c i y ) + rc 2 c 2 2 c 2 c i y 2 c 6 c + rc 6 c + c i y
2 2

+rc 4 c 4 c 2 + 6 2 + c + 2 i c y 2 i c y + rc 2 c 2 2 c 3 c c + i c y / rc 2 rc 2 + c a c D2 = 2 a2 2 i c rc 2 2 + rc 4 r c 2 c 2 c 2 c rc 2 2
3

rc 2 c a + c
4

+ a 2 rc 2 2

i rc 4 c + 2 i 2 rc 2 c 3 + rc 4 c + 2 c 4 c r c 4 c + c 3 c
2 2 2

rc 4 y + a6 rc 2 c 2 i rc 2 c 3 + i rc 2 c 2 c 2 c
2

y + a 4 rc 2 2
2 2

2 i r c 4 c 3 + i rc 6 c c

2 c 6 c + rc 4 c 2 c 2 2 2 c / rc 2 rc 2 + c a c D 3 = 2 a 2 r c 2 rc 2 2 +a4 rc 2 2 2 a2 2 rc rc
2 6 2

+ r c 2 2 c 5 c 3 c y rc 2 c a + c
4 2

+rc 2 rc 4 + 2 rc 2 c c c + c 2 c rc 2 2
2 3

a 2 rc 2 2

3 rc 4 + 10 rc 2 c 2 4 c 3 c
2

+a6 rc 2 c 2 rc 4 4 rc 2 c 2 + c 2 c / rc 2 rc 2 + c a c D4 =
2

rc 6 4 c 5 c + r c 4 c 9 c + 4 c + r c 2 4 c 4 c 2 c rc 2 2
2

rc 2 c a + c

D5 = 2 a2 2 i c rc 2 2 +a4 rc 2 2
2

+ a 2 rc 2 2
2

i 4 rc 2 c 3 + 4 c 4 c + rc 4 c + 2 c rc 4 + r c 2 c 2 c 2 c
2

rc 4 y + a6 rc 2 c 2 i rc 4 c + c 3 c + 2 rc 2 c 2 c + c +rc 2 2 c 5 c 3 c
2

i 2 c 6 c + rc 6 c + c + rc 4 c 4 c 2 2 2 + c r c 2 rc 4 + c 2 c 2 r c 2 c c + c
2 2

/ rc 2 rc 2 + c a c

rc 2 2

rc 2 c a + c

206

D6 = 2 a2 2 c rc 2 2 +rc 4 3 c 2 c i y

+ a 2 rc 2 2

4 rc 2 c 3 + 4 c 4 c
2

+ a6 rc 2 c 2 c 2 c (c i y )
2

+rc 2 c 2 2 c 2 c i y + rc 4 c i y +rc 2 c 2 2 c 3 c c i c y / rc 2 rc 2 + c a c D7 = 2 a i 2 rc 2 rc 2 2 +a4 rc 2 2


2 2 2 2

+ a 4 rc 2 2

2 c 6 c

+rc 6 c c i y rc 4 c 4 c 2 6 2 + c 2 i c y 2 i c y rc 2 2
2 3

rc 2 c a + c
4

+ a 2 rc 2 2

3 rc 4 10 rc 2 c 2 + 4 c 3 c
2

+a6 rc 2 c 2 rc 4 + 4 rc 2 c 2 + c 2 c

rc 6 4 c 5 c + rc 4 c 9 c + 4 c + rc 2 4 c 4 c 2 c
2

/ rc 2 rc 2 + c a c D8 = D9 = 2 i (a2 + rc 2 ) rc
4

rc 2 2

rc 2 c a + c

i r c 4 rc 2 2 +a4 rc 2 2
2

+ a 2 i rc 2 2
3

2 r c 4 c 2 + rc 2 4 2
2

rc 2 2

i r c 4 rc 4 c 4 2 r c 2 2 + c 2 c
2

+ c rc 6 5 c 2 c

3 rc 2 c 3 c + 4 c 4 c + 2 rc 4 2 3 c + c +a10 rc 2 c 4 i rc 4 + 2 c 2 c + 4 rc 2 c c + c +a8 c rc 2 2

+ 2 r c 4 c rc 2 2 y + 2 c rc 2 + 2 y
2 3 2

i rc 4 c 3 rc 2 2 + 8 rc 8 c 11 rc 2 c 6 c + 4 c 7 c
2

+2 rc 6 c c 2 2 2 7 c +a6 rc 2 2
3

+ r c 4 c 2 7 c 3 4 c 2 c + 4 c c + 6 c
2

+2 rc 4 4 rc 4 + c 4 + 8 rc 2 2 4 c 2 c +rc 6 6 c 2 + 5 2 + 6 c +rc 2 17 c 5 c + 2 c 3 c / rc 2 rc 2 + c a c
3 2 2

y
2 2 3 2

i 2 rc 4 c 2 rc 2 2 + 3 rc 8 4 c 6 c
3

rc 4 c 13 c 3 10 c 2 c + 10 c c + 6 c
4

+ 2 r c 2 rc 4 c + c 2 c 2 r c 2 c c 2 + c rc 2 2 rc 2 c a + c
2

207

D10 =

r c 4 rc 2 2 a4 rc 2 2 +a6 rc 2 2
4

+ a 2 rc 2 2
2

2 rc 4 c 2 + rc 2 4 2
2

rc 2 2

+a10 rc 2 c 4 rc 4 + 2 c c (c i y ) + 2 rc 2 c 2 c 2 c i y r c 4 rc 4 c 4 2 r c 2 2 + c 2 c
2

c rc 2 + 2
3

rc 2 c 2 c + 4 c 3 c + rc 4 5 c 2 c 2 i y 2 r c 4 c 2 rc 2 2 + 3 r c 8 4 c 6 c
2 2 2 2 2

+rc 2 c 2 c 17 c 3 + 2 c c + 2 i c y + rc 6 6 c 2 + 5 2 2 c 3 c i y +rc 4 c 13 c 3 10 c c + 2 c a8 c rc 2 2 3 c 2 i y + c 2 10 c 4 i y
2 2 3 2

rc 4 c 3 rc 2 2 + 11 rc 2 c 6 c 4 c 7 c

+8 rc 8 c i y rc 4 c 2 7 c 3 + 4 c c 6 c 8 i c y + c 2 4 c + 2 i y +2 rc 6 c c 2 + 2 2 + c 7 c 8 i y / rc 2 rc 2 + c a c +2 a2 c rc 2 2 +a4 rc 2 rc 2 2
5 3 2 2

rc 2 2
7

rc 2 c a + c
2

D11 = 2 a i ( + ) rc 2 c rc 2 2

+ a 8 r c 2 c 4 c 5 rc 2 + 2

r c 4 rc 4 4 r c 2 2 + 2 c 2 c + a 6 c rc 2 2
2

c 3 4 rc 2 + 3 2 + 8 rc 4 c 9 rc 2 c c 2 + c
2

+c 2 c 11 c 2 + c

2 r c 2 c 4 rc 2 2
2

+ 4 rc 8 11 rc 6 2 10 rc 2 c 5 c + 4 c 6 c + rc 4 5 c 4 + 7 c 2 c / rc 2 rc 2 + c a c D12 = 2 i rc (rc 2 2 )4 rc 2 2
3

rc 2 c a + c
4

a6 r c 4 c + r c 2 c rc 2 2
2

+ a 2 c rc 2 2

+a4 c rc 2 2

rc 2 2 r c 2 + 2

208 E 1 = 2 a 2 c rc 2 2
5

+ 2 a 3 rc 4 rc 2 2

(1 + i) rc 2 + 3 i c y 4 rc 2 c 4 c + c 5 c
2 2

+c 2 (1 + i) c + 3 i y

a 2 rc 2 2

+rc 6 (5 + i) c + 4 c i y + rc 4 c 3 c 2 + c 5 c + 2 i c + 3 i y c c 2 i y / rc 2 rc 2 2
5 5

+ a4 rc 2 rc 4 4 c + (4 + i) c + i y c 2 c (c i y )
2

+rc 2 c 5 c 2 + 2 c + c (3 + 2 i) c 2 i y E 2 = (2 2 i) a2 c rc 2 c c +rc 2 (1 i) c 4 c + 3 c c
3 4

+ a 2 rc 2 c c

c 2 c

+ rc 6 2 i c + (3 i) c y

+ rc 4 (1 + i) c 3

+2 i c 2 c 2 c + c c 2 i c + y c 2 c
2

+ a4 rc 2 rc 4 (2 4 i) c + c y

c + y + 2 rc 2 c i c 2 + (1 2 i) c c + c c + y
5 5

/ rc 2 rc 2 c c

E 3 = 2 a2 c rc 2 2

2 a 3 rc 4 rc 2 2

(1 + i) rc 2 + 3 i c y
2

+c 2 (1 + i) c + 3 i y

+ a4 rc 2 c 2 c (c i y )
2 2

+rc 4 4 c + (4 i) c i y + rc 2 c 5 c 2 + 2 c + c (3 2 i) c + 2 i y +a2 rc 2 2 / rc 2 rc 2 2
5

4 rc 2 c 4 c c 5 c + rc 6 (5 + i) c 4 c i y

+rc 4 c 3 c 2 + c 5 c + 2 i c + 3 i y + c c + 3 i y i y

209

E 4 = 2 i a3 a2 rc 2 c c + (5 + 4 i) c
2 3

3 rc 8 8 c 2 c + rc 6 7 c 2 (1 2 i) c c
3 5

+ rc 2 9 c 3 c + 22 c c

+ r c 4 c 7 c 3

+ (7 + 2 i) c 2 c 14 c + c c (7 + 4 i) c + 16 i y +a6 rc 2 c
4

2 rc 4 + rc 2 8 c 2 + (2 + 4 i) c + 2 i c c y + 4 i a 5 rc 4 c
2 4

+c 2 c 3 c + 2 i y

rc 2 c c ((1 + 2 i) c
3

+ (2 + i) c + 4 y 8 i a3 rc 4 c +4 y ) rc 2 c c +a4 c rc 2 c c +rc 4 c
2 5

rc 2 c c
5 4

((1 + 2 i) c

+ (2 + i) c + 4 y + 4 i a rc 4 rc 2 c c

(1 + 2 i) c + (2 + i) c
3

3 rc 6 4 c 2 c + rc 2 4 c 3 c + 11 c c
2 7 3 3

+rc 4 (4 + 2 i) c 2 + (3 + 4 i) c c 7 c + 8 i c y 4 c 2 c + r c 2 c c 11 c + c 2 8 c + 2 i y
2 3

+rc 8 4 c 3 c 4 i y + rc 6 c 13 c 2 + (5 4 i) c c (6 + 8 i) c + 10 i c y 17 c 3 (10 2 i) c 2 c + 4 i c c + 7 c 8 i c c + c y
2

/ rc 2 rc 2 + (a c ) c

rc 2 c c

rc 2 (a + c ) c
4 2

E 5 = 2 a 4 i a7 rc 4 (1 + 2 i) c (2 + i) c c 8 i a5 rc 4 (1 + 2 i) c (2 + i) c c (2 + i) c rc 2 c c
3 5

rc 2 c c
3

rc 2 c c
8 2

+ 4 i a3 rc 4 ((1 + 2 i) c
5

+ 2 r c 2 rc 2 c c
6

+ a 2 rc 2 c c

9 rc 6 4 c 2 c
3 2

rc 4 (4 + 2 i) c 2 + (3 4 i) c c + 11 c +a4 rc 2 c c +a8 rc 2 c
4

+ rc 2 4 c 3 c + 15 c c

7 rc 8 + 8 c 2 c + rc 6 7 c 2 + (7 + 2 i) c c (23 + 4 i) c
2 3 3

+rc 4 c 7 c 3 (7 2 i) c 2 c + (11 4 i) c c + 16 c +i c y )) + a6 c rc 2 c c +c 2 (34 2 i) c + 8 i y


2

rc 2 7 c 3 c + 24 c c
2 2 3

8 rc 4 + c 2 c 3 c + 2 i y 2 rc 2 4 c 2 (8 + i) c c + (1 + 2 i) c 4 c 2 c rc 4 c
6 7 2

17 c 3 + (8 + 4 i) c c + 7 c

+ rc 8 4 c 17 c 4 i y + rc 6 c 13 c 2 (17 + 4 i) c c
3 4

+ (14 + 8 i) c + 10 i c y + rc 2 11 c c + 2 i c 3 c y / rc 2 rc 2 + (a c ) c
2

rc 2 c c

rc 2 (a + c ) c

210

E 6 = 2 a2 a6 rc 2 c 4 c (16 + 3 i) rc 2 2 2 + 6 a5 (1 + i) rc 4 c 4 rc 2 2 12 a3 (1 + i) rc 4 c 2 rc 2 2 3 rc 2 c (3 + i) rc 2 2 2
2 3

+ 6 a (1 + i) rc 4 rc 2 2
5

rc 2 2

+ a 2 rc 2 2
2

3 (4 + i) rc 6 c
2 2

24 rc 2 c 4 c + 4 c 5 c + rc 4 c (26 + 6 i) c 2 3 (4 + i) c a4 c rc 2 2 +rc 4 c 2 (13 + 3 i) c 2 2 (19 + 3 i) c / rc 2 rc 2 + c a c E 7 = 2 i a2 rc 2 c c a4 rc 2 c c


2 5 2

4 rc 8 + 2 (16 + 3 i) rc 6 2 + 4 c 6 c + 2 rc 2 c 3 c 7 c 2 + c rc 2 2
2 3

rc 2 c a + c
5

+ r c 2 rc 2 c c
2

+a6 rc 2 rc 4 + 2 c c + i c c + rc 2 5 c c + (2 + 4 i) c

rc 6 c 4 c + rc 2 c 2 c (1 + 2 i) c + 4 i y 2 rc 2 + c c c i c + 2 y
5

+rc 4 c c + (4 + 4 i) c + 8 i y 4 a5 rc 2 rc 2 c c / rc rc 2 c c
5

E 8 = 2 a2 rc 2 c c 4 a5 rc 2 c + i c +a4 rc 2 c c / rc rc 2 c c
5

r c 2 rc 2 c c
2

rc 2 c c
2

2 rc 2 + c c
2

rc 6 + (1 + 2 i) rc 2 c 2 c + c 4 c + rc 4 c c + (4 + 4 i) c
2

+a6 rc 2 rc 4 + 2 c c i c c + rc 2 5 c c (2 + 4 i) c E 9 = 4 a3 (1 i) rc 6 (2 a + i c ) + i rc 4 (2 + i) a2 c 2 c + (1 + i) rc 2 c i a2 + 2 a c + (1 + 2 i) c 2 c i c 4 c / rc rc 2 c c
4 2 3

211

F1

2 r c 2 c 2 2 a 2 rc 2 + c 2 = 2 2rc (a c ) (a + c ) + a3 rc 4 (a c ) rc 2 + (a c ) c
2

a3 rc 4 (a + c ) rc 2 (a + c ) c
2

F2

2 a2 c 2 i 2+ 2 = 2 rc a2 + c 2 +a3 rc 2

1 a2 c a2 + c 2 rc 2 a2 c 2 a4 rc 2
2

(a c ) rc 2 + (a c ) c + c

1 (a + c ) rc 2 (a + c ) c
2

F3 = i +

a4 rc 2 (a + c ) rc 2 (a + c ) c
2

F4 F5

(a c ) rc 2 + (a c ) c a2 = 1 + 2 rc a2 = i 1 + 2 rc

F6 =

2 a3 c 2 rc 6 5 rc 4 c c + 4 rc 2 c 2 c + (a c ) c (a + c ) c rc 2 rc 2 + (a c ) c
2

rc 2 (a + c ) c

212

G k = a2 (a c ) c 2 (a + c ) c k 2 a2 (c + k )2 k + rc 12 (c + k )2 c + k 1 rc 10 4 c 3 c + 8 c 2 c k + a2 + c c + 4 c +rc 8 2 c c (c + k )2 3 c c + 2 k 2 +a2 2 c
3

k 2 + 2 c k 3 + k 4
2 2

c + k

c + k 4 c
2

+ a 4 c k 2 c + 2 k c 2 c + 2 c c k k + c k 2
2 2

c 2 c + 2 c c k k + c k 2
2 2 2 2

2 c 2 c + 4 c c k c 2 + 4 c c 2 c +rc 4 c
2 2 2 2

k 2 2 c k 3 k 4 k c + k
2

a6 c k 2 + c 3 c (c + k )2 c c + 4 k 2 2 c c k
2 2

+a4 c

k + c k 2 + c 2 c + 3 k 2
2

+2 c c 2 c + 2 c c k + 2 c + c
2 2 4

k 2 + 4 c k 3 + k 4 k
2

+ c 2 c 2 k 2 + k 2 c + k 2 + 2 c k c 2 c k + 2 k 2 2 a2 c 2 c k 4 c + k
4 3 2 2 2

k
2 2

+ c 3 c + 2 c k + c 3 k 2 k
2

+2 c 2 k c + 2 c k + c 3 k 2 k + c + 4 c k 3 k 2 k +a4 k 2
2 2

+ c k 2 c

c + 4 k + 2 c

c + 4 k k c + k
2 2

2 rc 6 c c 2 c (c + k )2 2 c c + 3 k 2 + 2 c c c + 2 k
2 4 3

c k

c + k

a 2 c k 4 c + k
2 2 2 2

+2 c 3 c + c k k k +2 c +2 c k 2 c
3 3 2

c + c + k k 2 c 2 k 2 c + c k k k + c k k + 2 k 2 k
2 2 2 2

k 4 k

c + k + 2 c

c + k k + c + c k k 2 k
2 3 2

rc 2 c k 2 c 4 c (c + k )2 c + k +a2 c 2 4 c 3 k 2 c k 2 c + k 4 c k c + 2 c k + c k k
2 2 3 2 2 2

+ a6 c 2 c k + c + k

c 2 c + 2 c k + c 8 k k
2

+ a 4 2 c 3 c 2 k k + c k 2 c + k + 2 c c k c + 2 c k + k k + k
2

c 2 c c + 2 c k + k 4 k + k / rc 2 rc 2 + (a c ) c

rc 2 (a + c ) c

(a c k ) k 2 (a + c + k ) c + k

213

G k = i a2 (a c ) c 2 (a + c ) c k 2 a2 (c + k )2 k + rc 12 (c + k )2 c + k 2 rc 10 4 c 3 c + 8 c 2 c k a2 + c c 4 c +rc 8 2 c c (c + k )2 3 c c 2 k 2 a2 2 c
3

k 2 2 c k 3 k 4
2 2

c + k

c + k + 4 c
2

+ a 4 c k 2 c + 2 k k

c 2 c + c k 2 + 2 c k c + k
2 2

c 2 c + c k 2 + 2 c k c + k k 2 2 c k 3 k 4 k c + k
2 2 2

+ 2 c 2 c + 4 c c k c 2 4 c c 2 c +rc 4 c
2 2 2 2 2 2

a6 c k 2 + c 3 c (c + k )2 c c 4 k 2 c k 2 + 2 c k c + k
2 2

+a4 c

+ c 2 c + 3 k 2 k 2 + 4 c k 3 + k 4 k
2

+2 c c 2 c + 2 c c k + 2 c + c
2 2

+ c 2 c 2 k 2 + k 2 c + k 2 + 2 c k c + 2 c k + 2 k 2 2 a2 c 2 c k 4 c + k
4 3 2 2 2 2 2

k
2 2

+ c 3 c + 2 c k + c 3 k 2 k
2

+2 c 2 k c + 2 c k + c 3 k 2 k + c 4 c k 3 k 2 k
3 2

+ c k 2 c
3

c 4 k + 2 c

c 4 k k
2

+ rc 2 c k 2 c 4 c (c + k )2 c + k a 2 c 2 4 c 3 k + 2 c k 2 c + k
2 3 2 2 2 2 2

+a6 c 2 c k + c + k

+4 c k c + 2 c k + c + k k

+ c 2 c + 2 c k + c + 8 k k

+a4 2 c 3 c 2 k k + c k 2 c + k c 2 c c + 2 c k + k 4 k + k +a4 k 2
2 2 2

+ 2 c c k c + 2 c k + k k + k c + k
2 2

2 rc 6 c c 2 c (c + k )2 2 c c 3 k 2 + a 2 c k 4 c + k
4 3

c + k

c + k
2

+ 2 c c c + 2 k

2 c 3 c + c k k k +c
4 2 2

c + c + k k + 2 c k 2 c + c 2 k k k + c k k + k 2 k
2 2 2 2 2 3

k 2 k

2 c 2 k 2 c + c k k 2 k 2 k + 2 c k k + k + c / rc 2 rc 2 + (a c ) c
2

k + 4 k
2

rc 2 (a + c ) c

(a c k ) k 2 (a + c + k ) c + k

214

G k = i rc 2 rc 2 c c 3
4 2

(c + k )2 rc 2 c + c k 2
2 2

c + k

2 2

+a8 c k 2 rc 4 c + c 2 c k + rc 2 c c + 4 c k + k +a6 c rc 4 c + k
2

rc 2 c c 2 k c + k 2
2

c c k 2 (c + k )2 k + 2 rc 6 k 2 c + k c + k + c c + 6 c k 2 + c k + 2 k 2 k
2 2 2 3 2

rc 2 c k 2 2 c c k c 2 c + 2 k + c k 2 + c + k +a4 rc 2 c c c k 2 c + k
2 2 2 2 4 2 3 3

c 5 c k 2 k + rc 10 k 2 c + 2 k + rc 2 c k 2 2 c c k c + k + 4 c 3 k c + k
2

c + 2 k + c 2 c + k
2

c + 2 c k 2 + c k + 4 k 2 k
2

+c 4 c + 4 c k 2 k

+ rc 8 2 k 2 c + k 2
2 2 3

c + k

+ c 2 c
3

c 2 k 2
2

+4 c c 2 k 2 k + 2 c + k 2 k 9 k 2 k k 2 c + k
4 2 2

+ r c 4 c c

2 c k c + k
2

c 5 c k 2 + c k

5 c + 4 c k 2 + 5 c k + 8 k 2 k
2 2

+c 2 c 4 c k 2 + 2 c c 6 k 2 k + c + 2 k 2 k +2 rc 6 c k 2 c + k +2 c k c + k
4 3 3

c + c k 2 + c k + 2 k 2 k
2

2 c + 2 c k 2 2 c k + 3 k 2 k
2 2

c 2 2 c + 4 c k 7 c k 2 k + k 2 k + c a2 rc 2 c c
2 2 3 2 2 2 2 2

3 k 2 + 2 k
2

2 2 2 2 2 2

c 3 c k 2 (c + k )2 k + rc 8 (c + 2 k ) c + k + 2 c k 2 c + k
4 3

rc 2 c k 2 4 c c k c + k 2 c 3 k c + 2 c k 2 k
3 2 2 4 2

c 4 c + 2 c k 2 k
2 3

+ c 2 c + 2 c k 2 c k 2 k + 2 k 2 k + c
2 3 2 2 2

k 2 + k

+rc 6 c 2 2 c 4 c k + 2 k 2 k + c k 2 2 k +c k 2 2 k
4 3

+ 2 c k 2 c 4 c k + 2 k 2 k
2

+ k 2 c + 2 c k + 2 k 2 k + c k 2 + k
2

rc 4 c c 2 c + 2 c k + 4 c k 2 k k 2 k + c +2 c k c + 2 c k + 4 c k 2 k k 2 k + c +k 2 4 c + 8 c k + 4 c k 2 k k 2 k + 2 c / rc 2 rc 2 + (a c ) c
2 4 3 2 2 2 3 2 2

2 k 2 + k 2 k 2 + k
2

k 2 + 2 k

2 2

rc 2 c c

rc 2 (a + c ) c

k 2 (c + k )2 a2

c + k

215

Gk = 4

rc 2 rc 2 c c +a2 rc 2 c c

3 3

(rc k ) (rc + k ) (c + k )2 c + k k 2 (c + k )2 k c + c + k + k
2 2

2 2 3 2 2

+ r c 4 c + k

rc 2 k 2 c c + k k
2 3

+ a 4 k rc 8 k c 3 c k k
2 2 2

rc 6 k 3 c c + 2 c + 4 c k + k +rc 4 c c 4 c + c +k 2 c + 4 c
3 2 2

+ r c 2 c 2 c k 3 c k + c c + 4 c k + 2 k

5 k 8 k + 4 c k k k + 2 k k
2 3

k k + c 8 k 5 k k + 4 k k

4 i (c + k )2 c + k
2

/ rc 2 rc 2 c c

k 2 (a + c + k ) (a + c + k ) c + k

G k = i rc 2 rc 2 c c 5

(c + k )2 rc 2 + k 2
3 2

c + k

2 2 3 2 2

+a2 rc 2 c c rc 2 k 2 c c + k k
2 2

k 2 (c + k )2 k rc 4 c + k
2 2

c + c + k + k

+ a 4 k rc 8 k + c 3 c k k

+rc 6 k 3 c c + 2 c + 4 c k + k +rc 4 c c

+ r c 2 c 2 c k c c + 4 c c k + 2 c 3 c k
2 2

4 c c + 5 c k + 2 c k + 4 k 2 + 4 c 2 c c + c + c k + 2 k 2 k + 4 c c + 2 c k + 5 c k + 4 k 2 k / rc 2 rc 2 c c
3

k 2 (a + c + k ) (a + c + k ) c + k

216

Gk = 2 a 6

rc 4 c rc 2 + (a c ) c
2 2

rc 2 c c
2

rc 2 (a + c ) c

2 2

+rc 4 rc 2 + (a c ) c +a2 c rc 2 + (a c ) c +rc 4 2 a2 9 c 2 a2 c rc 2 c c


2 2

rc 2 (a + c ) c rc 2 c c
4 2

2 a 2 c 2 + rc 2 c c
3

k
2

3 rc 2 c c

rc 2 (a + c ) c
5

k 2

a2 rc 12 rc 8 5 a2 + 12 c 2 c + 2 rc 6 c 5 a2 + 14 c 2 c
2

a2 + 3 c 2 c 4 rc 2 c 3 a2 3 c 2 c + 2 (a c ) c 4 (a + c ) c 2 rc 6 5 rc 4 c c + 4 rc 2 c 2 c + (a c ) c (a + c ) c a2 c
3 3

k 3

k 4 rc 2 + (a + c ) c
3

/ (a c k ) k (a + c + k ) + + rc 2 c c c + k
3

rc 2 + (a c ) c
2

rc 2 + c c c + k
2

rc 8 3 rc 6 c c + rc 4 5 a2 + 3 c 2 c rc 2 c 7 a2 + c 2 c
4

+2 a2 (a c ) (a + c ) c

/ rc 2 rc 2 + (a c ) c

rc 2 c c

rc 2 (a + c ) c

Gk = 7 Gk 8 Gk 9 Gk 10

4 rc k 2 a2 = (c + k )2 2 rc 2 = 2 k 2 i (c + k )2 = k (a + c + k ) (a + c + k ) c + j
2

Gk = i 11 / Gk 12 =

(c + k )2 j j + k k rc 2 j + k k + r c 2 c 2 j k j + k a + c + j (a c k ) (a + c + k ) rc 2 + j k rc 2 + j k

+a2 rc 4 2 c + j + k c j 2 j k + c j + k 2 i (c + k ) (c + k )3 k a2 rc 2 + c k a c j

(a + c + k )2 (a + c + k )2 rc 2 + k k

217

H1 = 2(2a c ) Hk = 2

2 rc 2 + k 2 a k + + c + k

a2 (+c )2

218

Appendix D

Coecients in

Equation (3.160)
J1 = Re i a 6 r c 2 c + 2 c + 2 a 2 c r c 2 2 2 (rc 2 )3
3 3

r c 2 c r c 2 2

(D.1)

+a4 2 c 2 c rc 4 c + 3 c + rc 2 2 c + 5 c = J2 =
2 y (2 a2 rc ) + 2 2 (rc

Re

a4 rc 4 c 3 c = J3 = x 2 a2 + rc 2 Re
2 2a2 c

a 6 r c 2 c 2 c + r c 2 c r c 2 2 2 (rc 2 )3 a6 r c 2 2 a4 2 2 (rc 2 ) (rc 2 )3

a6 r c 2 a4 2 2) (rc 2 )3

(D.2)
3

2 a2 c rc 2 + 2
3

(D.3)

+ r c 2 c c 5 c c + 2 c 2 c

(D.4)
2

2 rc 2

2 2 2a4 (rc 2 2 ) 2a6 c a8 r c 2 + + 2 + 2 (rc 2 ) (rc 2 )2 (rc2 2 )4

(D.5)

K1 = Re K1a log(

2 r2 2 rc + ac 2 ) + K1b log( c 2 ) 2 rc rc K1d a K1e +K1c log( ) + 2 2 2 )(r 2 2 )5 (rc 2 )5 c (rc + ac c

(D.6)

K1a =

2 i a6 rc 4 rc 4 c a2 c 3 rc 4 c + a2 c 2 c 2 rc 2 c 2 c + 2 rc 2 c c + c 3 c c 2 c 4 a6 rc 4 rc 4 + a2 c 2 + 2 rc 2 2 c 2 c c (rc 2
2 5 2)

c 2 (rc 2 2 )

(D.7)

219

K1b =

c 2 c 4 i a 8 r c 6 + a 8 r c 4 c 2 c (rc 2 + 2 )
5

2 i a 4 r c 4 + a 6 c 2
3

r c 2 c 3 c 2 i a 6 r c 6 + a 6 r c 4 c 2 + a 6 r c 2 c 4 + a 6 c 6 4 i a 8 rc 4 + 4 3 c (rc 2 + 2 ) c 3 (rc 2 + 2 ) + 2 i 2 a 6 r c 6 a 6 r c 2 c 4 a 6 c 6 rc 2 c (rc 2 +


3 2 2)

2 i a 4 r c 6 + a 6 c 4
2

2 i a6 r c 2 c

2 i a 6 c rc (rc 2 + 2 )
2

(D.8)

K1c =

2 i a6 rc 4 rc 2 + a c 2

r c 2 c + r c 2 c + a 2 + c 2 c a c c c c (rc 2 2 )
2 5

4 a 6 rc 4 rc 2 + a c 2
2

c (rc 2 2 )

rc 2 + a c + 2 y
5

(D.9)

K1d = i a rc 2 c 2 c

rc 2 2
6 6

+ r c 2 c 2 c r c 4 c c
2

rc 2 2

+ a 3 2 rc 2 2
3

r c 4 c c

+a2 c rc 2 2 a6 rc 2 2
3

+ a 9 r c 2 c c
2

r c 4 + c c
3

r c 6 c + c c r c 2 c +a8 rc 2 c c
2

c 3 c 2 c + c c + 2 c
3 3 2

+ rc 4 2 (c + 2 i (i + ) x )

rc 2 2
6

r c 4 + c c + 2 r c 2 c x 7 c + c + 2 i c 2 i c 2 y + 2 y
2 3

+a4 rc 2 c rc 2 2

2 rc 6 6 rc 4 2 + c 2 c (2iy ) + rc 2 c c
2

a7 c rc 2 2

r c 6 c + c c r c 2 c
2 2

c 3 c 2 c + c c + 2 c
3

+rc 4 c c 2 2 c x + 2 ( + ) c (ix ) +a5 rc 2 c rc 2 2 2 rc 6 6 rc 4 2 + c 2 c (2iy ) / c c


3

+rc 2 2 c 9 c + 2 i y + c c + 2 i c 2 i c 2 y + 2 y

(D.10)

220

K1e = i a2

r c 4 c 4 c

rc 2 2

+ a 6 r c 2 c 2 c
2

r c 4 + 2 c 2 c 3 c + c i y rc 2 2 2
2

+rc 2 c 2 c + c + 2 i y a4 rc 2
2 2 2

2 a 5 r c 2 c 2 c
5

rc 2 2 (c i y )

+2 a3 c 3 c rc 2 2 2

rc 3 r c 2
3 3

(c i y )
2

rc 6 c + c 3 c + rc 4 c c + 2 (1 i + i ) c 2 (c i y ) y
2

rc 2 c 2 c 4 c 3 + 2 c 2 c + c 2 i c 2 + c +2 a2 2 rc 3 rc 2
2

r c 6 3 r c 4 2 + c 3 c

c c i y / c 3 c
2

+rc 2 c 2 c 3 c + ( + ) (i c + y )

(D.11)

K2 = Re K2a log(

2 r2 2 rc + ac 2 ) + K2b log( c 2 ) 2 rc rc K2d K2e a 2 +K2c log( ) + 2 2 (rc 2 )5 c (rc + ac 2 )(rc 2 )5

(D.12)

K2a =

2 a 6 r c 4 r c 4 c + a 2 c 3 r c 4 c + a 2 c 2 c + 2 r c 2 c 2 c + 2 r c 2 c c c 3 c c 2 c c 2 (rc 2 2 )
5

(D.13)

K2b =

2 a 4 r c 4 + a 6 c 2 c 2 c
3

2 a 4 r c 6 + a 6 c 4
2

4 a8 rc 4 c (rc 2 + 2 )
4

r c 2 c 3 c 2 a 6 r c 6 a 6 r c 4 c 2 + a 6 r c 2 c 4 a 6 c 6 c 3 (rc 2 + 2 )
2 3

2 2 a 6 4 a 8 r c 6 a 8 r c 4 c + 5 r c 2 c c (rc 2 + 2 )

2 2 a 6 r c 6 a 6 r c 2 c 4 + a 6 c 6 rc 2 c 3 (rc 2 + 2 )

rc 2

2 a 6 c (rc 2 + 2 )

(D.14)

K2c =

2 a6 rc 4 rc 2 + a c 2

r c 2 c r c 2 c + a 2 + c 2 c + a c + c c c (rc 2 2 )
2 5

(D.15)

221

K2d = a rc 2 c 2 c
2

rc 2 2
6 3

r c 2 c 2 c
3

rc 2 2

+ a 9 r c 2 c c
5

rc 4 + c c
3

a2 c rc 2 2 a8 rc 2 c c rc 2 2
2 2

r c 4 + c c

+ a3 2 rc 2 + 2

r c 4 + c c
3

r c 4 c c 2 r c 2 c x + a 4 r c 2 c r c 2 2
3

2 rc 6 + 6 rc 4 2 + c 2 c

c + c

+rc 2 c c

7 c + c + 2 i c 2 i c 2 y + 2 y rc 2 2
2 2

+a5 rc 2 c

2 rc 6 + 6 rc 4 2 + c 2 c c 3 + c 2 c + c c 2 c
6 2 2

c + c

+rc 2 2 c 5 c + c + 2 i c 2 i c + 2 i c c + c y +a6 rc 2 2
3

r c 6 c c c r c 2 c
2

+ rc 4 2 (c + 2 (i + ) (ix ))
2 3 3

+a7 c rc 2 2

r c 6 c c c r c 2 c

c 3 + c 2 c + c c 2 c

+rc 4 c c 2 + 2 c x + 2 ( ) c (ix )

/ c c

(D.16)

K2e = a2

r c 4 c 4 c

rc 2 2

+ a 6 r c 2 c 2 c
2

rc 4 + 2 c 2 c 3 c + c + i y
2

+rc 2 c 2 c + c 2 i y +2 a2 2 rc 3 rc 2
2 2

+ 2 a 5 r c 2 c 2 c

rc 2 2 2

rc 2 2 (c i y )

2 a3 c 3 c rc 2 2 2 r c 6 3 r c 4 2 c 3 c
2 5 3 2

rc 3 r c 2

(c i y )
2

c + c i y + rc 2 c 2 c 3 c + ( ) (i c + y ) c + 2 i y / c 3 c
2

a4 rc 2 2

r c 6 c + c 3 c + r c 2 c 2 c 4 c 3 2 c 2 x c +rc 4 c c + 2 (i + ) c 2 (i c + y )

(D.17)

K3 = Re K3a log(

2 rc 2 K3b K3c )+ 2 + 2 2 2 )(r 2 2 )4 rc (rc 2 )4 (rc + ac c

(D.18)

K3a =

2 i 2 a3 rc 4 + 3 a5 c 2 c 2 c
2

6 i a 5 r c 4 + a 5 c 4 12 i a5 c 8 i a 7 rc 4 12 i a5 c 2 + + 2 4 2 2 rc (rc 2 + 2 ) r c 2 c (rc 2 + 2 ) c (rc 2 + 2 ) (D.19)

222

K3b = 2 i a3 a5 rc 2 c 2 c c rc 2 2 +a3 c rc 2 2
2 2 4

rc 6 + i (2 i + ) rc 4 2 2 rc 2 c 2 c + c 3 c
3 4 3 2

i ( + ) rc 4 c + rc 2 rc 2 + c 2 c + 2 rc 2 c c c a c
2

+a4 rc 6 c c rc 2 c 3 c
3

rc 2 2
4

r c 6 2 r c 2 c 2 c + c 3 c
2

+rc 4 c c + i (2 i + ) c (1 i + i ) rc 4 c c c + rc 2 c c 2 + 2 c

+a2 rc 2 2

/ c c

(D.20)

K3c = 2 i a3 a4 rc 2 c c a 2 rc 2 2 2

r c 2 5 2 + a 3 r c 2 c
2

rc 2 2 2
4

rc 2 2
2 3 3 2

rc 3 r c 2
2

+ c rc 2 2

rc 6 + i (i + ) rc 4 2 2 rc 2 c 2 c + c 3 c / c c

+a2 rc 2 2

(1 i + i ) rc 4 c + c c rc 2 3 c 2 c + 2 c

(D.21)

L1 = Re i a8 rc 4 c

+ r c 4 rc 2 2
4 4

+ 2 a 6 c
2

rc 2 2
3

2 r c 4 c c
2

2 a2 rc 2 2 +a4 rc 3 rc 2
2

rc 6 + 3 rc 2 c 2 c c 3 c + rc 4 c 3 c + c c 2 + 4 2 + c
2 2 2

r c 6 2 c 2 c 2 r c 4 c c + c + r c 2 c / rc 2 rc 2 + (a c ) c rc 2 2
2

rc 2 (a + c ) c

(D.22)

L2 = Re

4 a2 c rc 2 2

r c 6 c 3 r c 4 a 2 + c 2 c + 3 r c 2 c a 2 + c 2 c + a 4 c 4 c rc 2 rc 2 + (a c ) c
2

rc 2 (a + c ) c

(D.23)

L3 = Re i a8 rc 4 c

+ r c 4 rc 2 2
4 4

+ 2 a 6 c
2

rc 2 2
3

2 r c 4 + c c

+2 a2 rc 2 2 +a4 rc 3 rc 2
2

r c 6 + 3 r c 2 c 2 c c 3 c r c 4 c 3 c + c
2

rc 6 + 2 c 2 c + 2 rc 4 c (2iy ) + rc 2 c / rc 2 rc 2 + (a c ) c
2

c 2 4 2 + c

2 2

rc 2 2

rc 2 (a + c ) c

(D.24)

223

L4 = Re a10 rc 4 c c c + rc 4 c c +2 c c rc 2 c
4 2

rc 2 + 2

2 a 8 c

rc 2 2
3

rc 4 2 c c
2 3 2

c + 2 c
2

+ 2 a 2 rc 2 2 a 6 rc 2 2
2 2 3

r c 6 2 c c + r c 2 c 2 6 c c c 2 c 4 c
5 5

rc 4 c 6 c 2 3 2 + c
3

r c 6 c 7 c 4 c 2 c + 2 r c 2 c
3 2 2

2 c 2 2 2 + c
3

+rc 4 c c 2 + 11 2 + 6 c 2 rc 2 c 2 c

+ a 4 rc 2 2 + r c 4 c
2

2 r c 8 c 5 c 4 c 4 c
3 3

c 2 12 2 2 c

6 c 3 46 c 2 c 9 c c + c
2

+rc 6 6 c 2 c + 36 c c + 4 c / rc 2 rc 2 + (a c ) c

rc 2 2

rc 2 (a + c ) c

(D.25)

L5 = Re i a10 rc 4 c
4 3

c + c + r c 4 c + c a 6 rc 2 2
2 2 3

rc 2 2
5

2 a 8 c
3

rc 2 2

r c 2 c 2 c c
2 2

+2 c c + rc 4 2 c + c 2 rc 2 c

4 c 2 c + rc 6 c + 7 c rc 4 c c 2 + 11 2 6 c
5

2 c 2 + 2 2 + c
3

+ 2 a 2 rc 2 2
2

2 c 4 c + rc 6 2 c + c + rc 2 c 2 c
3

6 c + c

rc 4 c 6 c 2 + 3 2 + c 2 rc 2 c 2 c
2

a 4 rc 2 2

4 c 4 c + 2 r c 8 c + 5 c
2

c 2 + 12 2 2 c

+ 2 rc 6 c 3 c 2 18 2 + 2 c
2 3 2

rc 4 c

6 c 3 46 c 2 c + 9 c c + c rc 2 2
3

/ rc 2 rc 2 + (a c ) c

rc 2 (a + c ) c

(D.26)

L6 = Re i a12 rc 4 c c +rc 2 c c 2 3 c +rc 2 c 2 c


2 2 2

+ r c 4 2 rc 2 2
6

8 3

+ a10 c

rc 2 2
2

2 2

2 r c 4 c + 2 c c

4 2 3

+ a 2 rc 2 2

2 c 5 c 4 rc 4 2 2 c 2 + c
3 4

+ r c 6 3 c 2 + c
5

7 c 2 + c
4 2

+ a 8 c r c 2 2
4

7 rc 6 c + 2 rc 2 c 3 c 4 c 2 c + rc 4 9 c 2 c + 2 c
3

+a4 rc 2 2
2

2 rc 2 2
2 4

2 rc 8 12 rc 6 2 + 4 c 4 c + 2 rc 2 c c
4

8 c 2 + c

2 2 2

+rc 4 22 c 2 c c +c

+ a 6 rc 2 rc 2 2

r c 4 c 2 + 5 c
2

2 r c 2 2 c 2 + 6 c
4

c 4 + 8 c 2 c + c

/ rc 2 rc 2 + (a c ) c

rc 2 2

rc 2 (a + c ) c

(D.27)

224

M1 = Re 2 i rc 4 rc 2 2 +a6 rc 2 2
3

2 a 2 rc 2 2
4 2

rc 6 + 3 rc 2 c 2 c c 3 c + rc 4 c 3 c + c
2

4 r c 6 + 2 c 2 c + r c 2 2 c 2 6 c

r c 4 c 2 + 6 2 4 c
4

i rc 2 c rc 4 + c 2 c + 2 rc 2 c (2iy ) y + a10 rc 2 c +rc 2 c 2 + 2 2 + 3 c + i c y


2 2

r c 4 c 2 c x
2

+ a 4 rc 2 2
7 4

r c 8 2 c 3 c + r c 2 c 2 c
3 4 2

c 2 + 5 c

+rc 6 c 2 2 + c + i c y + rc 4 2 c 3 c 4 c c + c i c 2 c y a8 c rc 2 2 2 c 2 c + r c 2 c c
2

c 2 5 c

+ r c 8 4 c + 3 c 4 i y
2

+rc 6 c c 2 + 3 2 + 6 c + 8 i c y rc 4 c
2 3 2

3 c 3 + 7 c 2 c

+7 c c 3 c + i c 4 c + c y / rc 2 rc 2 + (a c ) c rc 2 2
4

rc 2 (a + c ) c

(D.28)

M2 = Re 2 a2 2 2 rc 2 2 +a2 rc 2 2 +a4 rc 2 2
3 5 6

+ a 8 r c 2 c
4

c 2 c c i y + r c 2 c 2 3 c + i c y
2

2 rc 6 2 c 2 c rc 2 2 c 2 + c
2 3 2

+ r c 4 c 2 + 2 + 3 c + i c y + r c 6 c c + 4 c i c y 2 c 2 c
7 2

r c 8 2 c 2 c + r c 2 c c
4 2

8 c + c 2 3 c i y

+rc 4 c c 5 c 2 + 2 2 + 6 c +rc 2 c c c 2 + 5 c
2

+ 2 i c c c y
2 3

a 6 c r c 2 2

+ r c 8 4 c c 4 i y r c 6 c c 2 + 2 + 3 c 8 i c y
2 2

+rc 4 c

3 c 3 + c 2 c + 4 c c 3 c i c 4 c + c y / rc 2 rc 2 + (a c ) c rc 2 2
4

rc 2 (a + c ) c

(D.29)

M3 = Re 2 i a 2 rc 2 c rc 2 2 +a8 rc 2 c
4 2

+ a 2 rc 2 2

4 c 2 c + r c 4 3 c + 2 c 2 r c 2 2 c + 5 c 4 r c 8 + 2 r c 2 c c 5 c c + 4 c 2 c a 4 rc 2 rc 2 2 c 2 4 2 + 10 c rc 2 2
3 2 2 3 4 6 2

c 2 c + r c 2 3 c + 5 c

a 6 c r c 2 2
2 2

+2 rc 6 c 3 c + 5 c + rc 4 c

10 c 2 13 2 + 6 c + c c
2

r c 4 c c

+rc 2 c 2 c 2 + 7 2 12 c

/ rc 2 rc 2 + (a c ) c

rc 2 (a + c ) c

(D.30)

225

M4 = Re

2 i a2 c rc 2 2

+ c rc 3 r c 2

rc (rc 2 2 )

+ a 4 c c

+ r c 2 c + 2 c

(D.31)

M5 = Re 2 a2 2 2 rc 2 2 +rc 4 c
2 3

a 6 c r c 2 2
4

2 c 2 c r c 2 c c

c 2 + 5 c
2

3 c 3 + 3 c + 2 4 c + i y c 2 c + 4 i y +rc 8 4 c + c 4 i y + a 8 r c 2 c
4 2

+ rc 6 c c 2 3 c + c c + 8 i y
2 2

c 2 c x + r c 2 c 2 3 c + i c y r c 4 c 2 2 + 3 c + i c y rc 6 c 4 c c + i c y
2 3

+a2 rc 2 2 +a4 rc 2 2
3

2 rc 6 + 2 c 2 c + rc 2 2 c 2 + c
6 2 2

rc 8 + 2 c 2 c rc 2 c c

8 c + c 2 3 c + i y

+rc 4 c c 5 c 2 2 2 + 6 c

+ 2 i c c + c y
2

/ rc 2 rc 2 + (a c ) c

rc 2 2

rc 2 (a + c ) c

(D.32)

M6 = Re 2 i rc 4 rc 2 2 +a6 rc 2 2
3

2 a 2 rc 2 2
4

r c 6 + 3 r c 2 c 2 c c 3 c r c 4 c 3 c + c
2

4 rc 6 + 2 c 2 c + rc 2 2 c 2 6 c
2

+ rc 4 c 2 + 6 2 + 4 c
4

+i rc 2 c rc 4 + c 2 c 2 rc 2 c c + c rc 2 c 2 2 2 + 3 c + i c y +rc 4 c
2 3 2 2

y + a10 rc 2 c
7

r c 4 c 2 c c i y
4

+ a 8 c r c 2 2

2 c 2 c + rc 2 c 3 c
5

+ 5 c c

3 c 3 3 c + 2 7 c + i y + c 2 7 c 4 i y + a 4 rc 2 2
2 4 2 2

+ r c 8 4 c 3 c 4 i y
2

+rc 6 c c 2 3 2 + 6 c + 8 i c y

rc 8 2 c 3 c + rc 2 c 2 c
3

c 2 + 5 c

+rc 6 c 2 + 2 + c + i c y rc 4 c 2 c 3 + 4 c c + c + i c 2 y / rc 2 rc 2 + (a c ) c rc 2 2
4

rc 2 (a + c ) c

(D.33)

226

M7 = Re 2 a 2 rc 2 c rc 2 2 a2 rc 2 2 a6 c rc 2 2
5

+ a 8 r c 2 c

r c 2 3 c 5 c c 2 c
3 6 4

r c 4 3 c 2 c 2 r c 2 c c 5 c c 4 c 2 c
2 2

4 r c 8 + 2 r c 6 3 c 5 c c + 4 c 2 c 2 r c 2 c c +rc 4 c 10 c 2 + 13 2 + 6 c
2 2

c + 5 c

a4 rc 2 rc 2 2

r c 4 c + c + c c

c 2 + 4 2 + 10 c
2

rc 2 c 2 c 2 + 7 2 + 12 c
3

/ rc 2 rc 2 + (a c ) c

rc 2 2

rc 2 (a + c ) c

(D.34)

M8 = Re

2 a2 c rc 2 2

+ c rc 3 r c 2

+ a 4 r c 2 c 2 c + c c
2

rc (rc 2 2 )

(D.35)

M9 = Re 2 rc 4 2 c c
3

rc 2 2
2 3

+ 2 a 2 rc 2 2 + a12 rc 2 c
2 6 4

rc 6 3 c 2 c + rc 2 c 2 11 c 2 c c r c 4 c 3 c 4 c 2 c x
2 2

4 c 4 c 2 rc 4 c 5 c 2 4 2 + c rc 2 2 c + 7 c 4 i y rc 2 c +a4 rc 2 rc 2 2
5 2

+ 2 a10 c

rc 2 2
4

4 c c + r c 4 c c 2 6 c + 4 c c i y
2 2 3

2 c 3 + 9 c + 2 c 2 x + rc 6 9 c + 3 c + 8 i y 5 c + 2 c + r c 2 c
2

3 r c 6 c 3 c + 4 c 2 c

3 c 3 20 c c + 2 c

c 2 53 c + 4 i y
3

+ 2 rc 4 c 3 c 2 + 21 2 + 4 c + 2 i c y
2 2

+a8 rc 3 rc 2 rc 2 c c
2

4 c 2 c
2

2 c 3 + 5 c + c 2 x

33 c 3 21 c c + c

43 c 4 i y + c 2 35 c + 36 i y
3 2

+rc 4 c

27 c 3 45 c c + 27 c + c 2 39 c + 76 i y
6

+rc 8 15 c 19 c + 16 i y + rc 6 c 13 c 2 + 11 2 + 24 c 60 i c y +2 a6 rc 2 2 +2 rc 2 c c
2 3 4 4

8 c 3 c + rc 8 3 c + 2 c
2 2 3 4 2

c 4 + 12 c c + c + c 3 x + rc 6 c 2 c 2 + 8 2 7 c + 2 i c y y
2 2

+rc 4 c 4 c 4 7 c 3 c + 5 c 2 c 20 c c 2 c 4 i c c 2 + c / 2 rc 2 rc 2 + (a c ) c rc 2 2
5

rc 2 (a + c ) c

(D.36)

227

M10 = Re i 2 rc 4 2 c + c +rc 2 c 2 c
2

rc 2 2
2

2 a 2 rc 2 2
2 6

4 c 4 c + rc 6 3 c + 2 c
4 2

11 c + 2 c 2 rc 4 c 5 c 2 + 4 2 + c + 2 a10 c
2

+ a12 rc 2 c

r c 4 c + 3 c + 4 c 2 c
3

c i y

+rc 2 2 c + 7 c + 4 i y
4

rc 2 2

4 c c + rc 2 c

2 c 3 + 9 c + 2 c 2 c i y
2 3

+rc 4 c c 2 6 c 4 c x + rc 6 9 c 3 c + 8 i y +a4 rc 2 rc 2 2
5

4 c 2 c

5 c + 2 c + 3 rc 6 c + 3 c + rc 2 c
4 3

3 c 3 20 c c 2 c
2

+c 2 53 c 4 i y +2 i c y )) + a8 rc 3 rc 2 +rc 2 c c +rc 4 c
2 3 2 3 2 2

+ 2 rc 4 c 3 c 2 21 2 + 4 c 4 c 2 c
2

2 c 3 5 c c 2 c i y

33 c 3 + 21 c c + c

43 c + 4 i y + c 2 35 c 36 i y + rc 8 15 c + 19 c + 16 i y
3 4 2

27 c 3 45 c c 27 c + c 2 39 c + 76 i y
6 2

+rc 6 c 13 c 2 11 2 + 24 c 60 i c y +2 a6 rc 2 2
4

8 c 3 c + r c 8 3 c + 2 c 2 r c 2 c c
2 2 3 4

c 4 + 12 c c c + c 3 c i y
2

rc 6 c 2 c 2 + 8 2 + 7 c 2 i c y + rc 4 c 4 c 4 + 7 c 3 c +5 c 2 c + 20 c c 2 c 4 i c c 2 + c / 2 rc 2 rc 2 + (a c ) c
2

y rc 2 (a + c ) c
2

rc 2 2

(D.37)

M11 = Re 2 a 2 rc 2 c 2 rc 2 2 +a2 rc 2 2
2 6

+ a10 rc 2 c c

3 rc 2 + 2
2

2 rc 2 2
3 5

r c 4 2 r c 2 2 4 c 2 c
4 2 2

+ a 6 rc 2 2
5 2 2

4 rc 10 + 7 rc 8 2
6 7 3

31 rc 6 c 2 c + 25 rc 4 c 3 c 5 rc 2 2 rc 4 + c 4 c + 11 rc 4 c c + 2 rc 2 c 2 c 4 c 3 c +a4 rc 2 2
4

4 c 3 c + r c 2 c 2 c
2

c 2 32 c

+ rc 6 c 2 14 c
5 2

+ rc 4 2 c 3 c + 46 c c
4 2

+a8 c

rc 2 2

4 rc 6 6 rc 4 2 4 c c + rc 2 3 c 2 c + 10 c rc 2 2
4

/ rc 2 rc 2 + (a c ) c

rc 2 (a + c ) c

(D.38)

228

M12 = Re 2 a8 rc 4 2 rc 2 2 rc 2 2 a4 rc 2 2
3

5 2

+ a2 c 2 rc 2 + 2 + a 6 rc 2 2
2

5 3 2

2 r c 4 3 r c 2 2 + 2 c 2 c

c c + r c 2 c 2 2 c / rc rc 2 2
5

(D.39)

N 1 = Re N 1a log(

2 rc + ac 2 r2 2 ) + N 1b log( c 2 ) 2 rc rc N 1e N 1d a + 2 +N 1c log( ) + 2 2 (rc 2 )6 c (rc + ac 2 )(rc 2 )6

(D.40)

N 1a =

6 a 6 r c 4 r c 4 c + a 2 c 3 r c 4 c + a 2 c 2 c + 2 r c 2 c 2 c + 2 r c 2 c c c 3 c c 2 c c 2 r c 2 c c
6

(D.41)

N 1b =

6 i a4 rc 4 + a6 c 2
4

4 i 3 a 8 r c 6 + 2 a 8 r c 4 c 2 c 2 rc 2 + c c
5

c 2 c r c 2 c r c 4 c 4 c 12 i a6 rc 2 c + 2 a6 c 3 6 i a 8 r c 8 + 2 a 8 r c 6 c 2 + a 8 r c 4 c 4 + 6 r c 6 c c 2 rc 2 + c c + 6 i a 8 r c 4 a 6 r c 6 2 a 6 r c 4 c 2 a 6 r c 2 c 4 c 2 rc 2 + c c
2 4

12 i a6 c
3

2 i 3 a4 rc 8 + a6 rc 2 c 4 + 9 a6 c 6
2

4 i a 6 r c 4 2 a 6 c 4 c 2 rc 2 + c c (D.42)
3

2 i a 6 3 r c 8 + 5 r c 2 c 6 + 3 c 8 rc 4 c 4 rc 2 + c c

12 i a6 rc 2 c 2 + 2 c 4 rc 6 rc 2 + c c

N 1c =

6 a6 rc 4 rc 4 c rc 4 c + 2 rc 2 c 2 c + a2 c c + 2 rc 2 c c c 3 c + a2 c c 2 c c
2

r c 2 c c

(D.43)

229

N 1d = i a2 rc 2 3 a rc 2 c c rc 2 c c a7 c c +a6 c c
2 3 4

+ 3 r c 2 c r c 2 c c
4

7 2

rc 6 + 4 rc 4 c c + c 2 c + 2 rc 2 c c
5 2

c + 2 c
2

r c 2 c c
3

r c 6 c 2 c 2 r c 4 c 2 c + c 2 r c 2 c c
3

c + 4 c + 3 i y

+a2 c rc 2 c c

6 rc 8 22 rc 6 c c c 3 c + rc 4 c c
2

30 c + (1 2 i + 2 i ) c
3 2

2 rc 2 c 2 c a5 c rc 2 c c +rc 2 c
2 2

10 c + (2 + 2 i 2 i ) c + 3 ( + ) y i ( + ) c + c + c 2 c c c
2 3

3 r c 6 c + 2 r c 4 c
3

c 2 + c

6 c 3 + 2 c + 3 c c c + 2 i y + 2 ( ) c 2 2 i c + 3 y
3 2 2

a4 rc 2 c c +rc 2 c +a3 c
2

3 rc 6 c + 2 (1 i + i ) rc 4 c c + c 2 c c c
2 3 5 2

c 2 + c

6 c 3 c c + 2 c + 2 (i + ) c 2 2 i c + 3 y 6 rc 8 22 rc 6 c c c 3 c + rc 4 c c 28 c + (1 2 i + 2 i ) c / c c
4

r c 2 c c
2

+2 rc 2 c 2 c

3 c 4 c + i y + c 2 i (i + ) c + 3 ( ) y

(D.44)

N 1e = i rc 4 c 4 c

r c 2 c c
2

+ a 2 c

r c 2 c c r c 2 c c r c 2 c c
4

3 r c 6 2 r c 2 c 4 + 2 c 5 c r c 2 + 2 c c 3 i c y r c 2 + 2 c c 3 i c y
2 7

+2 a7 rc 4 c 3 c

r c 2 2 c c
2

2 a5 rc 4 c 4 c rc 2 2 c c a6 rc 2 c c
2

2 i (i + ) rc 8 c 4 + 3 rc 10 c + 2 c 5 c
2

+rc 2 c 4 c +rc 4 c 3 c

c 4 7 c 2 c 2 c c + i y
2 2

11 c 3 5 c c + 2 c
2

c 3 i y + c 2 6 c 4 i y
2 3 3

+rc 6 c 2 c 4 i c 3 4 i c 3 + c 2 c + 2 c c 6 c + 6 (i + ) c 2 y +a8 rc 2 c 2 c r c 8 + r c 6 c c 4 c + 2 c 4 c
2 2 2

c + i y
2

2 rc 2 c 3 c c 2 + 8 c c + 6 c + i c y rc 4 c 2 9 c 2 + 11 c + 2 c 8 c 5 i y +2 i c y 4 y 2 +2 c 5 c
3

+ a 4 c c r c 3 r c c c c 2 + c
2

6 rc 10 22 rc 8 c c + rc 6 c 2 c (3 + 2 i 2 i ) c + 30 c 4 c 2 + 3 c c + 2 c + 2 i c 3 c y /rc 2 c 4 c
2 2

r c 2 c 4 c

+2 rc 4 c 3 c c 2 + i c (3 i + 2 2 ) c + y + c 7 c + 3 y 3 y

(D.45)

230

N 2 = Re N 2b log(

2 rc 2 a N 2d N 2e ) + N 2c log( ) + 2 + 2 2 2 )(r 2 2 )6 rc (rc 2 )6 c (rc + ac c

(D.46)

N 2b =

12 a4 rc 4 + a6 c 2 c 2 c
4

48 a6 c 3 r c 6 c

r c 2 c 8 12 a rc 8 a8 rc 4 c 4 c
2

24 a6 c
3

12 a4 rc 8 + 3 a6 c 6 r c 4 c 4 c c
2 2

24 a8 rc 6 rc 2 + c c
2 5

12 a8 rc 4 c 2 rc 2 + c c

rc 2 + c c
4

rc 4 c 4 rc 2 + c c

12 a6 rc 8 c 8

rc 6

48 a6 c 4 rc 2 + c c

(D.47)

N 2d = 2 a2 rc 6 a4 rc 4 + 4 a7 c 80 a3 rc 4 c 2 a6 c 2 + 182 a2 rc 4 c 2 + 14 a5 c 3 c 12 16 6 12 3 16 7 4 6 2 a rc + a rc c a rc c 3 a rc c 2 + 7 a2 rc 16 c 2
3

+60 a rc 4 c 3 24 a4 c 4 105 rc 4 c 4

6 a6 rc 14 a2 rc 18
4

c c 2 6 a5 rc 14 + a8 rc 10 c 40 a4 rc 14 c + 6 a7 rc 10 c 2 18 a3 rc 14 c 2 3 a6 rc 10 c 3 + 63 a2 rc 14 c 3 + c c 2 a9 rc 8 34 a5 rc 12 5 a8 rc 8 c + 114 a4 rc 12 c + 3 a7 rc 8 c 2 + 2 a rc
2 2

2 45 a rc
4

12

c 17 a rc 8 c 3 105 a2 rc 12 c 3
6

a rc

+ 5 a rc c + 2 a7 c 2 + 102 a3 rc 4 c 2 2 a6 c 3
4 4 2

172 a2 rc 4 c 3 10 a5 c 4 45 a rc 4 c 4 + 12 a4 c 5 + 63 rc 4 c 5 c 2 a2 rc 2 c 4 a5 rc 4 9 a4 rc 4 c 70 a3 rc 4 c 2 + 90 a2 rc 4 c 3 + 2 a5 c 4 + 18 a rc 4 c 4 2 a4 c 5 21 rc 4 c 5 c +2 a2 rc 4 c 2 5 a5 7 a4 c 20 a3 c 2 + 20 a2 c 3 + 3 a c 4 3 c 5 c 4 a6 rc 2 (a c ) c 3 c
3 4

(D.48)

231

N 2e = (1 + i) rc 4 c 4 c rc 2 c c (1 i) a4 rc 2 c rc 2 c c
3

+ (1 + i) a2 c rc 2 c c
2

3 i rc 6 + rc 2 c 4 (1 i) c 5 c
4

6 rc 8 16 rc 6 c c + (1 i) c 6 c 2 c 4 c + rc 4 c 2 c i c + 14 c (3 3 i) rc 10 + 2 c 5 c + (2 2 i) rc 6 c 4 3 c 2 c
2 3 5 2

+rc 2 c 4 c (1 + i) c c (1 + i) y +a6 c rc 2 c c
4 2

(1 + i) rc 4 c 3 c 3 3 c c + 2 i c + c 2 8 i c (1 i) y +i rc 2 c 8 + (2 + 7 i) c 4 c + 2 i c 6 c c + y 2 c 4 c
3

+ a8 rc 2 c 2 c (1 + i) rc 8 rc 6 c c (4 4 i) c
2

c + y + rc 4 c 2 9 i c 2 (4 + 4 i) c c + (11 8 i) c

+2 c y 4 c y + 4 i y 2 + rc 2 c 3 c 2 i c 2 + 6 c (2 i) c + y c (2 + 3 i) c + 2 y / r c 2 c 4 c (D.49)

N 3 = Re N 3a log(

2 rc + ac 2 r2 2 ) + N 3b log( c 2 ) 2 rc rc N 3d N 3e a + 2 +N 3c log( ) + 2 2 )(r 2 2 )6 (rc 2 )6 c (rc + ac c

(D.50)

N 3a =

6 a 6 r c 4 r c 4 c + a 2 c 3 r c 4 c + a 2 c 2 c + 2 r c 2 c 2 c + 2 r c 2 c c c 3 c c 2 c c 2 r c 2 c c
6

(D.51)

c 2 c 12 i a6 rc 2 c 2 a6 c 3 r c 6 c

N 3b =

6 i a 4 r c 4 + a 6 c 2
4

6 i a 8 r c 4 + a 6 r c 6 2 a 6 r c 4 c 2 + a 6 r c 2 c 4 + 2 i 3 a 6 r c 8 5 a 6 r c 2 c 6 + 3 a 6 c 8 rc 4 c 4 rc 2 + c c
2

r c 2 c r c 4 c 4 c 2 4 i 3 a 8 r c 6 2 a 8 r c 4 c 2 6 i a 8 r c 8 2 a 8 r c 6 c + a 8 r c 4 c 4 6 5 c 2 rc 2 + c c c 2 rc 2 + c c
4

12 i a6 c
3

2 i 3 a 4 r c 8 + a 6 r c 2 c 4 9 a 6 c 6
2

c 2 rc 2 + c c

4 i a 6 r c 4 2 a 6 c 4 c 2 rc 2 + c c
3

12 i a6 rc 2 c 2 2 a6 c 4 rc 6 rc 2 + c c

(D.52)

232

N 3c =

6 a6 rc 4 rc 4 c rc 4 c + 2 rc 2 c 2 c + a2 c c + 2 rc 2 c c c 3 c + a2 c c 2 c c
2

r c 2 c c

(D.53)

N 3d = i a2 rc 2 3 a rc 2 c c rc 2 c c +a7 c c +a6 c c
2 3 4

+ 3 r c 2 c r c 2 c c

7 2 4

r c 6 4 r c 4 c c 2 r c 2 c c 2 c c + c 2 c
2 5 2

r c 2 c c
3

rc 6 + c 2 c + 2 rc 4 c 2 c + c 2 rc 2 c c 6 rc 8 22 rc 6 c c + c 3 c + rc 4 c c
3 2 2

c 4 c 3 i y

+a2 c rc 2 c c

30 c + c + 2 i c 2 i c
2

+2 rc 2 c 2 c

10 c + (2 + 2 i 2 i ) c + 3 ( + ) y 3 r c 6 c + 2 i r c 4 c c + i c c
3 2 3

a5 c rc 2 c c +c c
3

2 c 3 + c 2 c + 2 c c + c

r c 2 c

6 c 3 + 2 c + c c 5 c + 6 i y

+2 ( ) c 2 2 i c + 3 y a4 rc 2 c c +c c
3 3

3 rc 6 c + 2 i (i + ) rc 4 c c
2 2 3 5 2

2 c 3 + c 2 c + 2 c c + c
2

r c 2 c

6 c 3 + c c + 2 c + 2 (i + ) c 2 2 i c + 3 y 32 c + c + 2 i c 2 i c / c c
4

+a3 c

r c 2 c c
2

6 rc 8 22 rc 6 c c + c 3 c + rc 4 c c

+2 rc 2 c 2 c

c 8 c + 3 i y + c (2 + 2 i 2 i ) c + 3 ( + ) y

(D.54)

233

N 3e = i rc 4 c 4 c

r c 2 c c
2

+ a 2 c

r c 2 c c
2 2

3 r c 6 2 r c 2 c 4 + 2 c 5 c

+2 a5 rc 4 c 4 c rc 2 2 c c 2 a7 rc 4 c 3 c
2

r c 2 c c

r c 2 + 2 c c 3 i c y r c 2 + 2 c c 3 i c y
2 7

r c 4 3 r c 2 c c + 2 c 2 c
2 4

a6 rc 2 c c
2

2 (1 i + i ) rc 8 c 4 + 3 rc 10 c + 2 c 5 c c 4 7 c 2 c 2 c c i y
2 2 2

+rc 2 c 4 c +rc 4 c 3 c

11 c 3 + 5 c c + 2 c
2 2

c + 3 i y + c 2 6 c + 4 i y
3

+rc 6 c 2 c 4 i c 3 + 4 i c 3 + c 2 c 2 c c 6 c 6 i c 2 y 6 c 2 y + 6 c 2 y +a8 rc 2 c 2 c r c 8 r c 6 c c + 4 c + 2 c 4 c
3

c i y
2

2 rc 2 c 3 c c 2 8 c c + 6 c + i c y + rc 4 c 2 9 c 2 + 16 c c 11 c + 2 i c y + 2 i c y 4 y 2 +a4 c c rc 3 rc c c +2 c 5 c
3 2 2

6 rc 10 22 rc 8 c c + rc 6 c 2 c (3 2 i + 2 i ) c + 30 c r c 2 c 4 c
2

c 2 + c

4 c 2 + c 3 c + 2 i y + 2 c c + 3 i y / r c 2 c 4 c
2

+2 rc 4 c 3 c c 2 + i c (3 i 2 + 2 ) c + y +c 7 c + 3 ( + ) y

(D.55)

N 4 = Re N 4a log(

2 rc + ac 2 r2 2 ) + N 4b log( c 2 ) 2 rc rc a N 4d +N 4c log( ) + 2 2 )2 (r 2 a 2 )2 (r 2 2 )5 c (rc + ac c c c N 4f N 4e + 2 + 2 2 (rc 2 )5 (rc + ac 2 )(rc 2 )5 2

(D.56)

N 4a =

4 i a5 rc 4 rc 4 2 rc 2 c c + c 2 a2 + c c 2 rc 2 + c c
5

c c 4 i y

(D.57)

234

N 4b =

8 i a 3 r c 4 + a 5 c 2 c 2 c
3

8 i a 3 r c 6 + a 5 c 4
2

24 i a7 rc 4 c rc 2 + c c +
4

16 i a5 r c 2 c +

r c 2 c 3 c 24 i a7 rc 6 + a7 rc 4 c 2 c rc 2 + c c
5

8 i a 5 r c 6 + 2 a 5 r c 4 c 2 + 2 a 5 r c 2 c 4 + a 5 c 6 c 3 rc 2 + c c
2 3

8 i 2 a 5 r c 6 2 a 5 r c 2 c 4 a 5 c 6 rc 2 c 3 rc 2 + c c

rc 2

16 i a5 c rc 2 + c c

(D.58)

N 4c =

4 i a5 rc 4 rc 4 + 2 rc 2 c c + a2 c 2 c c
2

r c 2 c c

c + c + 4 i y

(D.59)

N 4d =

2ia 2 rc

r c 2 c r c 2 c c
7

a2 rc 2 c c

2 c 2 c + r c 4 c + 2 c r c 2 c c c + 6 c
5 3

+a4 rc 2 c c
2 4 2

2 c 2 c + r c 6 c + c r c 2 c c c 3 + 2 c c + 2 c
2 3 2 3

+rc 4 c 3 c 2 c + c c c +a10 rc 2 c 2 c 2 c
4 6

c + i y + 3 r c 4 2 c + c 2 i y + 3 r c 2 c c c + 2 c + 2 i y 2 c c + rc 2 c
2 2

+a6 rc 2 c c

2 c 3 c c + 8 c + 2 c 2 c + i y
2 6 2 5

+rc 6 2 c c 2 i y rc 4 c c 2 + 4 c c 6 c + 2 i c y a8 c
2

r c 2 c c

2 c c + rc 2 c 3 c

+ 5 c
2

+2 rc 6 8 c + 3 c 8 i y + rc 4 c 6 c 2 + 11 c c + 3 c + 12 i c y

(D.60)

235

N 4e = 2 i a a c c rc 2 c c +a7 rc 2 c c a4 rc 2 c c
3 3

r c 4 c c
6

+ c r c 2 c c
2 2

r c 4 c c
2

2 r c 4 + 3 r c 2 c c + c c

c + 2 c
3

r c 6 c + 2 c c + r c 2 c

2 c 3 + c 2 c 2 c c 4 c
2

+rc 4 c c (1 + 2 i 2 i ) c + 4 i (i + ) c + i y +a6 rc 2 c c +rc 2 c 5 c + 4 c + 4 i y +rc 4 c c


2 2

r c 2 c c
3

2 r c 4 + c c

c + 2 c
5 3

+ a 2 c r c 2 c c
2 6

4 rc 8 12 rc 6 c c c 3 c + rc 2 c 2 c
2 2 3

c + 4 c

(15 + 2 i 2 i ) c + c + 4 i c 4 i c 4 y + 4 y r c 6 c + 2 c c + r c 2 c 2 c 3 + c 2 c 2 c c 4 c
5 3

a5 c rc 2 c c
2

rc 4 c (3 2 i + 2 i ) c 2 + 4 c c + i y + 2 c c 2 i c + 2 i c + 2 y 2 y +a3 c r c 2 c c
2

4 rc 8 12 rc 6 c c c 3 c + rc 2 c 2 c

c + 4 c

+rc 4 c c 2 c 2 + c (19 + 2 i 2 i ) c + 4 i y +c c + 4 i c 4 i c 4 y + 4 y / c c

(D.61)

N 4f = 2 i a rc 6 c

r c 2 c c

a 4 r c 2 c c
3 2

r c 8 c 2 c 4 c + r c 2 c 3 c

c 3 + c 2 c + 6 c

+rc 6 (4 4 i + 4 i ) c 4 + 3 c c + 2 (1 i + i ) c 3 c 2 i y +rc 4 c 2 c 8 c 3 4 c +a6 rc 2 c 2 c


2 3

c i y + c 2 13 c + 4 i y
2 2

2 rc 6 c 3 c + 2 rc 4 c c + 4 c + i y + rc 2 c 2 c 15 c + 19 c 4 i y +2 a3 rc 4 c 3 c rc 2 2 c c 2 a5 rc 4 c 2 c
2 2

r c 2 c c

2 c + c 2 i y 2 c + c 2 i y
5 3

r c 4 3 r c 2 c c + 2 c 2 c

+a2 c c rc 2 c c +rc 4 c 3 c
2

4 rc 10 12 rc 8 c c + 2 c 5 c rc 2 c 4 c

c + 8 c

7 c + 10 c 4 i y + 2 rc 6 c 2 c (1 2 i + 2 i ) c + (3 i + i ) c +2 ( + ) y ))) /rc 2 c 3 c
2

(D.62)

236

N 5 = Re N 5a log(

2 rc + ac 2 r2 2 ) + N 5b log( c 2 ) 2 rc rc N 5d a +N 5c log( ) + 2 2 2 c (rc + ac 2 )2 (rc ac 2 )2 (rc 2 )5 N 5e N 5f + 2 + 2 2 )(r 2 2 )5 (rc 2 )5 (rc + ac c 2

(D.63)

N 5a =

4 a 5 r c 4 c + c

rc 4 2 rc 2 c c + c 2 a2 + c c 2 rc 2 + c c
5

(D.64)

N 5b =
7

8 a 3 r c 4 + a 5 c 2
6

24 a rc

c rc 2 + c c

c 2 c a r c 4 c 2
7 5

8 2 a 5 r c 6 2 a 5 r c 2 c 4 + a 5 c 6 rc 2 c 3 rc 2 + c c
2

r c 2 c 3 c 8 a 5 r c 6 2 a 5 r c 4 c 2 + 2 a 5 r c 2 c 4 a 5 c 6 24 a7 rc 4 4 3 c 3 rc 2 + c c rc 2 + c c rc 2 16 a5 c rc 2 + c c
2

8 a 3 r c 6 + a 5 c 4
2

16 a5 r c 2 c

(D.65)

N 5c =

4 a5 rc 4 c + c c

rc 4 2 rc 2 c c + a2 + c 2 c rc 2 + c c
5

(D.66)

N 5d = 2 a rc 2 c rc 2 c c a4 rc 2 c c
5

a 2 r c 2 c c
5 2 3

r c 4 c 2 c r c 2 c c 6 c c 2 c 2 c
2 3 4

r c 6 c c + 2 c 2 c + r c 2 c c c 3 + 2 c c 2 c + a10 rc 2 c r c 4 6 c 3 c + 2 i y
2 3 2

rc 4 c 3 + c 2 c + c c + c +2 c 2 c +a6 rc 2 c c
4 6

c i y + 3 r c 2 c c c 2 c + 2 i y
2

2 c c r c 2 c
2 2

2 c 3 + c c + 8 c + 2 c 2 c i y
2 6 2 3

+rc 6 2 c + c 2 i y + rc 4 c c 2 + 4 c c + 6 c 2 i c y a8 c r c 2 c c 2 c c r c 2 c c 3 + 5 c
2

+2 rc 6 8 c 3 c 8 i y

+ rc 4 c 6 c 2 11 c c + 3 c + 12 i c y

/rc 2

(D.67)

237

N 5e = 2 a a7 rc 2 c c
2

2 r c 4 + 3 r c 2 c c + c c 2 c c
5

2 6 3

+a c c rc 2 c c +a6 rc 2 c c r c 2 c c
3 2

r c 4 + c c

+ c r c 2 c c
2

r c 4 + c c

2 r c 4 + c c 2 c c + r c 2 c 5 c 4 c + i y 4 rc 8 12 rc 6 c c + rc 2 c 2 c 4 c c + c 3 c
6 2 2 3 5

+a2 c rc 2 c c +rc 4 c c
3

(15 + 2 i 2 i ) c + (1 4 i + 4 i ) c + 4 ( ) y r c 6 c 2 c c r c 2 c
2 6

a4 rc 2 c c

2 c 3 + c 2 c + 2 c c 4 c
2 2

+rc 4 c c (1 + 2 i 2 i ) c + 4 (i + ) i c + y a5 c rc 2 c c
2

r c 6 c 2 c c r c 2 c
2

2 c 3 + c 2 c + 2 c c 4 c
3 5

+rc 4 c (3 + 2 i 2 i ) c 2 + 4 c c + i y + c (2 4 i + 4 i ) c + 4 ( ) y +a3 c r c 2 c c 4 rc 8 12 rc 6 c c + rc 2 c 2 c 4 c c + c 3 c +rc 4 c c 2 c 2 + c (11 + 2 i 2 i ) c 4 i y +c (1 4 i + 4 i ) c + 4 ( ) y /c c

(D.68)

N 5f = 2 a rc 6 c

r c 2 c c

+ a 6 r c 2 c 2 c

2 r c 6 c 3 c
2

2 rc 4 c c 4 c + 3 i y + rc 2 c 2 c 15 c + 19 c + 4 i y 2 a3 rc 4 c 3 c rc 2 2 c c +2 a5 rc 4 c 2 c +a2 c c rc 2 c c +rc 4 c 3 c
2 2 2 2

r c 2 c c

r c 4 3 r c 2 c c + 2 c 2 c

2 c c 2 i y 2 c c 2 i y
3 5

4 rc 10 12 rc 8 c c + rc 2 c 4 c 8 c c + 2 c 5 c
2 6 2 3

7 c + 10 c + 4 i y + 2 rc 6 c 2 c c + 2 i c 2 i c + 3 c i c + i c + 2 y 2 y a4 rc 2 c c r c 8 c 2 c 4 c + r c 2 c 3 c
2 3

c 3 + c 2 c + 6 c

+rc 4 c 2 c 8 c 3 4 c

c + i y c 2 13 c + 4 i y / r c 2 c 3 c
2

+rc 6 4 i (i + ) c 4 + 3 c c + 2 (i + ) c 3 i c + 2 y

(D.69)

238

N 6 = Re N 6b log(

2 rc 2 N 6d )+ 2 2 2 )2 (r 2 a 2 )2 (r 2 2 )4 rc (rc + ac c c c N 6e N 6f + 2 + 2 2 )(r 2 2 )4 (rc 2 )4 (rc + ac c

(D.70)

N 6b =

2 i 2 a2 rc 4 + 7 a4 c 2 c 2 c
2

28 i a4 c 24 i a6 rc 4 + r c 2 c rc 2 + c c 14 i a4 rc 4 + a4 c 4 c 2 rc 2 + c c
2

28 i a4 c 2 rc 2 rc 2 + c c

(D.71)

N 6d = 4 i a2 rc 2 rc 2 c c
2 2

+ a 8 r c 2 c
4

3 r c 4 + 6 r c 2 c c c 2 c
2 4 5 2

+a4 rc 2 rc 2 c c +a6 c r c 2 c c

r c 4 2 r c 2 c c c 2 c + 5 c
6 3

8 rc 6 12 rc 4 c c + 2 c c + rc 2 c +a2 rc 2 c c

c 2 4 c
2

2 c c + rc 2 c 2 + 4 c

/rc 2

(D.72)

N 6e = 2 i a2 3 a5 rc 2 c 2 c + 3 a4 rc 2 c c c rc 2 c c +a3 c rc 2 c c a c
2 2 2 3 3

r c 4 c 2 c

2 2 3 4

rc 6 + (2 3 i + 3 i ) rc 4 c c 8 rc 2 c 2 c + 3 c 3 c
3 2 3

3 i rc 4 c + 3 i rc 4 c + 3 rc 4 c + rc 2 c 2 c + 6 rc 2 c 3 c c rc 6 8 rc 2 c 2 c + 3 c 3 c rc 4 c 3 c + (2 + 3 i 3 i ) c 3 (1 i + i ) rc 4 c 3 c c + rc 2 c c 2 + 6 c
4 2 2

r c 2 c c

+a2 rc 2 c c

/c c

(D.73)

N 6f = 2 i a2 3 a4 rc 2 c c
2

r c 2 5 c c 3 a c c r c 2 2 c c
2

r c 3 r c c c
2 3

+3 a3 rc 2 c +c rc 2 c c +a2 rc 2 c c
2

r c 4 3 r c 2 c c + 2 c 2 c
4

rc 6 + i (i + 3 3 ) rc 4 c c 4 rc 2 c 2 c + c 3 c
3

3 (1 i + i ) rc 4 c + c c rc 2 9 c 2 c + 4 c

/c c

(D.74)

239

N 7 = Re N 7a log(

2 rc + ac 2 r2 2 a N 7d ) + N 7b log( c 2 ) + N 7c log( ) + 2 2 rc rc (rc 2 )5 c

(D.75)

N 7a =

8 a 6 r c 3 2 r c 2 + c c

rc 4 2 rc 2 c c + c 2 a2 + c c 2 r c 2 c c
6

(i c + y )

(D.76)

N 7b =

8 i a 4 r c 6 + 2 a 6 c 4 r c 3 c 3 c
5 2

8 i 2 a 6 r c 2 + 9 a 6 c 2 r c 5 c c rc 2 + c c
2 4

24 i a8 rc 7 + a8 rc 5 c 2 c rc 2 + c c
6

8 i 4 a 8 r c 5 + a 8 r c 3 c 2 c rc 2 + c c +

8 i a 8 r c 3 3 a 6 r c 3 c 2 3 a 6 r c c 4

8 i 2 a 6 r c 2 c + 5 a 6 c 3 rc rc 2 + c c
3

8 i a 6 r c 6 + 2 a 6 r c 2 c 4 + 7 a 6 c 6 rc 3 c 3 rc 2 + c c

8 i 2 a 6 r c 2 c + 9 a 6 c 3 rc 5 rc 2 + c c

(D.77)

N 7c =

8 a 6 r c 3 2 r c 2 + c c

rc 4 + 2 rc 2 c c + a2 c 2 c c
2

(i c + y )

r c 2 c c

(D.78)

N 7d = 4 i a2

r c 4 c 4 c

r c 2 c c
2

+ a 6 r c 2 c 2 c

r c 6 + 2 c 3 c

2 c + c i y

+rc 2 c 2 c 11 c + 5 c 3 i y + rc 4 c 5 c + 2 c + 5 i y +2 a3 c 3 c rc 3 rc c c 2 a5 rc 2 c 2 c
2 2

2 r c 4 3 r c 2 c c 2 c 2 c
2 3

(c i y ) (c i y )

2 r c 6 5 r c 4 c c + r c 2 c 2 c + 2 c 3 c 2 r c 8 6 r c 6 c c + 2 c 4 c

a2 c c rc 3 rc c c +rc 2 c 3 c +a4 rc 2 c c
2 2 2

c + c + i y

+rc 4 c 2 c i (i + 4 4 ) c + 10 c + (i + 4 4 ) y i (3 i + 2 2 ) c 8 c + (3 i + 2 2 ) y
6 2

r c 8 c c 4 c + 2 r c 2 c 3 c
2 2

2 c 3 + c c + c 2 c i y + c

c i y / r c c 3 c
2

+4 (i + ) rc 6 c 3 (i c + y ) +2 rc 4 c 2 c (3 + i i ) c 3 + 2 c c c c + 2 i y + c 2 2 c + (i + ) y

(D.79)

240

N8

Re N 8a log(

2 N 8d rc + ac 2 r2 2 a ) + N 8b log( c 2 ) + N 8c log( ) + 2 2 rc rc (rc 2 )6 c

(D.80)

N 8a =

8 a 6 r c 3 2 r c 2 + c c

rc 4 2 rc 2 c c + c 2 a2 + c c 2 r c 2 c c
6

(c i y )

(D.81)

N 8b = +

8 a 4 r c 6 + 2 a 6 c 4 r c 3 c 3 c
5 2

8 2 a 6 r c 2 9 a 6 c 2 r c 5 c

24 a8 rc 7 a8 rc 5 c 2 c rc 2 + c c +
6

8 4 a 8 r c 5 a 8 r c 3 c 2 c rc 2 + c c

8 a 8 r c 3 3 a 6 r c 3 c 2 + 3 a 6 r c c 4 c rc 2 + c c
2 4

8 2 a 6 r c 2 c 5 a 6 c 3 rc rc 2 + c c
3

8 a 6 r c 6 + 2 a 6 r c 2 c 4 7 a 6 c 6 rc 3 c 3 rc 2 + c c

8 2 a 6 r c 2 c 9 a 6 c 3 rc 5 rc 2 + c c

(D.82)

N 8c =

8 a6 rc 3 2 rc 2 + c c

rc 4 2 rc 2 c c + a2 + c 2 c c
2

r c 2 c c

(c i y )

(D.83)

N 8d = 4 a2

r c 4 c 4 c

r c 2 c c
2

+ a 6 r c 2 c 2 c

r c 6 + 2 c 3 c

2 c + c + i y

+rc 2 c 2 c 11 c + 5 c + 3 i y + rc 4 c 5 c + 2 c 5 i y +2 a3 c 3 c rc 3 rc c c +2 a5 rc 2 c 2 c
2 2

2 rc 4 + 3 rc 2 c c + 2 c 2 c
2 3

(c i y ) (c i y )

2 r c 6 5 r c 4 c c + r c 2 c 2 c + 2 c 3 c 2 r c 8 6 r c 6 c c + 2 c 4 c

a2 c c rc 3 rc c c +rc 2 c 3 c
2

c + c i y

(3 2 i + 2 i ) c 8 c + (3 i 2 + 2 ) y
2 6 2

+rc 4 c 2 c (1 4 i + 4 i ) c + 10 c + (i 4 + 4 ) y +a4 rc 2 c c
2

r c 8 c c 4 c + 2 r c 2 c 3 c
2

2 c 3 c c + c 2 c + i y
2 2

+c

c + i y

+ 4 (1 i + i ) rc 6 c 3 (c i y ) c 2 i y / r c c 3 c
2

+2 rc 4 c 2 c i (3 i + ) c 3 2 c c c

+c 2 2 c + (i + ) y

(D.84)

241

N9

Re N 9a log(

2 N 9d rc + ac 2 r2 2 a ) + N 9b log( c 2 ) + N 9c log( ) + 2 2 rc rc (rc 2 )6 c

(D.85)

N 9a =

16 i a7 rc 5 rc 2 + c c
5

8 i a 7 rc 3 rc 2 + c c
4

16 i a5 rc 5 c 2 rc 2 + c c
3

8 i a 5 rc 3 c 2 rc 2 + c c
2

(D.86)

N 9b =

8 i a 3 r c 4 + 2 a 5 c 2 r c c 2 c
2

16 i a7 rc 3 rc 2 + c c

64 i a5 c 32 i a7 rc 5 + r c 3 c rc 2 + c c
4

32 i a5 rc c 2 rc 2 + c c
3

8 i a 5 r c 4 6 a 5 c 4 rc c 2 rc 2 + c c
2

64 i a5 c 2 rc 3 rc 2 + c c
2

(D.87)

N 9c =

8 i a 5 r c 3 r c 2 + c c c
2

rc 4 2 rc 2 c c + a2 + c 2 c rc 2 + c c
5

(D.88)

N 9d = 4 i a3 3 a4 rc 2 c c
2

r c 4 3 r c 2 c c 2 c 2 c
2 2 3

2 2

+2 a c c rc 3 rc c c +2 a3 rc 2 c

rc 4 + rc 2 c c + 2 c 2 c c r c 2 c c
3 2 2

r c 6 2 r c 4 c c r c 2 c 2 c + 2 c 3 c
2

2 rc 8 + (3 2 i + 2 i ) rc 6 c c
4 2 2 5

+ (1 2 i + 2 i ) rc 4 c 2 c + 6 rc 2 c 3 c 2 c 4 c +2 a2 c rc 2 c c

(1 i + i ) rc 6 + i (2 i + ) rc 4 c c + c c 3 rc 2 c

c 2 + c
2

/ r c c c

(D.89)

242

P 1 = Re ( c ) + +

4 a 4 r c 4 c a 2 c 2 rc 2 rc 2 + (a c ) c rc 2 (a + c ) c
2 2

2 3 2

10 a4 rc 2 c c rc 2 + (a c ) c 2 a 6 c c
4

rc 2 (a + c ) c rc 2 (a + c ) c

8 a 4 c 2 c
2

rc 2 + (a c ) c

2 a 4 c 3 c
2

rc 2 (a + c ) c
4

rc 2 rc 2 + (a c ) c

rc 2 rc 2 + (a c ) c

rc 2 (a + c ) c

(D.90)

P 2 = Re i c + +

i a 2 c 4 i a 4 r c 4 c 2 2 rc rc 2 (a + c ) c rc 2 + (a c ) c 10 i a4 rc 2 c c
2 2

2 3 2

rc 2 + (a c ) c

rc 2 (a + c ) c
4

8 i a 4 c 2 c
2

rc 2 + (a c ) c

2 i a 6 c c rc 2 rc 2 + (a c ) c
2

2 i a 4 c 3 c
2

rc 2 (a + c ) c
4

rc 2 (a + c ) c

rc 2 rc 2 + (a c ) c

rc 2 (a + c ) c

(D.91)

P 3 = Re

i a6 2 i a 4 c 2 i a 2 c 4 2 2 2 2 + i 2 r c 2 a2 c 2 rc a c 2 rc a c 2 + 2 i a 5 rc 2 (a c ) rc 2 + (a c ) c
2

2 i a 5 rc 2 i a4 + 2 + 2 rc (a + c ) rc 2 (a + c ) c

(D.92)

P 4 = Re a2 c rc 2 2 +a4 c 2 c
3

r c 2 c r c 2 2

3 3

+ rc 2 2 c + 3 c + rc 4 5 c 2 c + 4 i y

/ rc 2 rc 2 2

(D.93)

P 5 = Re i a2 c rc 2 2

+ r c 2 c r c 2 2

3 3 3

+a4 rc 4 5 c 2 c rc 2 c c 3 c c c 2 c

/ rc 2 rc 2 2

(D.94)

243

P 6 = Re

2 a5 2 a 3 c 2 2 a4 rc 2 2 2 + r c 2 a2 c 2 rc a c 2 (a c ) rc 2 + (a c ) c 4 a3 rc 2 (rc 2
2 2)

2 a4 rc 2 2 a3 + 2 2 rc (a + c ) rc 2 (a + c ) c = 2 rc 2 a4 rc (rc 2 2 )
2

(D.95)

P 7 = Re 2 rc

2 a4 rc (rc 2 2 )
2

(D.96)

Q1 = Re

2 (rc

+ ac

2 2 )2 (rc

ac

2 )(a

Q1a c k )(a + c + k )(c + k )2 Q1b + (a c k )(a + c + k )(c + k )

(D.97)

244

Q1a = a8 c k
4 3

r c 2 + c k r c 2 r c 2 c c
3 2 2 2

(c + k )

c + k

2 2

rc 2 + c k a4 rc 10 k + c 5 c k
3

+rc 8 2 c + 4 c k 2 c
2

c + k k k 4 c (2 c + k ) k (c + 2 k ) k c 2 c + k c k 2 c + k
2

+rc 2 c c k 5 c 3 c k + c k 2 c + k 2 c k c + k c + c k 2 k k +c 2 c + k
3

c + c k 8 k k
2

+ r c 4 c
2

2 c 3 k c + 8 c k + 2 k
2 2

3 c + 7 c k 2 c k 2 k k 18 k k c + 5 c k + 4 c k 4 k k c + c k k k
3 2 2 3 2 3 2 2

+2 c k c + k

+2 rc 6 c 2 k k c + k +c 2 k 2 c + 9 c k + 2 k
2 2 2

c c + k
2

2 c + 3 c k 6 k k + c k 2 k + k
2 2

a6 c rc 4 c + k
3 2 2

c c (c + k ) k + 4 rc 6 k c + k + r c 2 c k c 2 k 3 c + 4 k
2 2

c + c k + 2 c (3 c + k ) k + 2 (c + k ) k
2

+c c c + 2 c k + k 2 k + k +a2 rc 2 c c
2 2 3 2 2 3

+ c k 2 c + 4 c k + k k + 2 k
2

c 3 c (c + k ) k + rc 8 c + k
2 2 2

+ rc 6 c 2 k + k k 2 c 4 c k + k 2 k k + r c 4 c 3 k
2 2

c 2 c + 5 c k + 4 c k + k +2 c k 2 c + k
2 4 4 3 2

2 k + k
3 2 2 3

2 c + k
2 3

+ c k 4 c + 12 c k + 12 c k 2 c k 2 k k + k k
2

+c 2 3 c + 8 c k + 7 c k + 2 k k + 2 c k +rc 2 c c k c 3 c 2 k k + 2 c k 2 c + k
3 2 2 3

2 k + k
2 2

+ c 2 c + 2 c k 4 k k + c k 2 k + k
2

+c k 2 c + 4 c k 2 k k + c k k + 2 k

/ r c 2 k

(D.98)

Q1b =

a2 + (c + k )

k 2

a2 k 2 rc 2 (c + k )

(D.99)

Q2 = Re

Q2a 2 2 (rc + ac 2 )2 (rc ac 2 )(a c k )(a + c + k )(c + k )2 Q2b + (a c k )(a + c + k )(c + k )

(D.100)

245

Q2a = i a8 c k +a6 c
2

r c 2 + c k + r c 2 r c 2 c c
2 2 3 2

(c + k )

c + k

r c 2 + c k c + c k
3 2

c c (c + k ) k
2

4 r c 6 k c + k r c 4 c + k
2

2 c (3 c + k ) k + 2 (c k ) k +c c c + 2 c k + k 2 k + k +a4 rc 10 k
3 2 4 3 4 2

+ r c 2 c k c 2 k 3 c + 4 k

+ c k 2 c + 4 c k + k k + 2 k
3 2 3

c 5 c k + rc 8 2 c + 4 c k + 4 c (2 c + k ) k + (c + 2 k ) k +2 c k c + k + k
2 2 2

rc 2 c c k 5 c 3 c k c k 2 c + k +c 2 c + k 2 rc 6 c 2 k k c + k +c c + k +rc 4 c
2 2 2

+ 2 c k c + k

c + c k + 2 k k

c + c k + 8 k k
2 2

c + c k + k k + c 2 k 2 c + 9 c k + 2 k
3 2 2

2 c + 3 c k + 6 k k + c k 2 k + k
2

c k 2 c + k
3

+ 2 c 3 k c + 8 c k + 2 k
2 2 2

+2 c k c + k +c 2 c + k
3

c + 5 c k + 4 c k + 4 k k
2 2

3 c + 7 c k + 18 k k + 2 c k k + 2 k
2

a2 rc 2 c c +rc 2 c c k 2 c k 2 c + k
2 3

c 3 c (c + k ) k
2 2

+ r c 8 c + k
3 2

2 2

+ c 3 k c + 2 k + c 2 c + 2 c k + 4 k k + c k 2 k + k
3 2 2 2

+c k 2 c + 4 c k + 2 k k + c k k + 2 k rc 6 c 2 k + c 2 c + 5 c k + 4 c k + k +k k 2 c + 4 c k + k k + 2 k
4 3 2 2 2 3 2

2 k + k
2 2 2

+ r c 4 c 3 2 c k k + 2 c k 2 c + k
2

+c 2 3 c + 8 c k + 7 c k 2 k k + 2 c k +c k 4 c + 12 c k + 12 c k k k + 2 c k
4 3 2 2 3 2

2 k + k / r c 2 k (D.101)

k + 2 k

Q2b =

i a2 + (c + k )

a2 k 2 + rc 2 (c + k ) k 2

(D.102)

Q3 = Re

2 (rc

+ ac

2 2 )2 (rc

ac

Q3a 2 )(a +

c + k )(a + c + k )(c + k )2 Q3b + (a + c + k )(a + c + k )(c + k )

(D.103)

246

Q3a = i a10 c k +a8 c


3 2

r c 2 + c k
3

+ r c 2 c r c 2 c c
2 2 2 2

(c + k )

2 3

c + k
2

r c 2 + c k
2 3

c c k (2 c + k ) k + rc 6 c + 2 c k + 2 k +rc 2 c k c k 2 k 2 c k c + c k + k +a2 rc 2 c c
3 2 2 3 2

+ r c 4 c c 3 c k + c k + 2 k
2 2

c 2 c + 6 c k + 3 k
2

c 4 c (c + k ) k + rc 8 c c c + k + rc 6 2 c c
3 2

c + c + k c + c + k
2

+2 c c + k

c + c + k k c k
2 3

c c + k
2 2 3

+c 3 c 3 + 5 c c 6 c 2 k + 2 c k 3 c k 2 k + (c + k ) 4 c + 3 c (c + k ) k +c (c + 2 k ) k 4 c c k c + k
5 2 3 4 3

+ rc 4 2 c k 2 c + k
2 3 3 2 2 3 3

+ c 4 c 6 c k + k
2 2

+ 2 c 3 k c 6 c k + k k 2 7 k
2 2 2 2 2

+c 2 3 c 8 c k 6 c k 2 k 2 c k + k 2 k + c +rc 2 c c k 2 c c k c + k +2 c 3 k c + 3 c k 2 k
2 2 2 2

2 c k 2 c + k
4 3 3

+ c 4 c + 3 c k 2 k
3

2 2

c 2 c + 2 c k 3 c k 2 k + 2 k 2 k + c c 2 c
2

k 2 + k

+a6 c rc 2 c c +rc 8 2 c + 8 c k + 7 k +2 c
2 2

2 c 2 + 2 c k + k 2 k
3 2 2

+ r c 4 c c

3 c 2 c + 2 c k + k 2
2 3 2 2 3

9 c 2 c + 8 c k + k 2 k + c 23 c 2 c + 36 c k + 6 k 2 k + c 2 + 20 c k + 4 k 2 k rc 6 8 k k c + k
2 2 2 2 2

+ c 4 c + 22 c k + 23 c k + k
2 3

+rc 2 c k 2 c k c + k 2 c 2 k c + 7 c k + 4 k +a4 rc 2 c c
4 4

c 3 c + 4 c k 3 k
2

2 2 2 2

+ c c + 2 c k 3 c k 2 k 2 k 2 k + c
3 2 2 2

k 2 + k

c 4 c k (2 c + k ) k rc 12 c + 2 k + rc 10 c 4 c + 7 c k 3 k
3

+rc 8 2 c
3 2

4 c 2 c + 2 c k + k 2 2 c
2 2 2

7 c 2 2 c
2

2 3

+5 c k + 2 k 2 k + c 9 c 2 + 2 c 2 c k + k 2 k + c 2 + 6 c k + 4 k 2 +rc 6 c 4 c c 2 c 2 c + 2 c k + k 2 + 2 c
2 2 2 2 4 2 2 2 2

7 c 3 5 c c + 9 c 2 k 2 c k + 3 c k 2 k
2 3 2 2 2 2 2 3

c 15 c 3 + 8 c c + 6 c 2 k + 8 c k + 9 c k 2 k 5 c 3 + 2 c c + 24 c 2 k + 4 c k + 15 c k 2 k +rc 2 c c k 2 c c k c + k 2 c 3 k c + 5 c k + 8 k +rc 4 c
2 2 2 3 2 2 3

+ c k 2 c + k
3

c 4 c + 6 c k + 3 k
2 2 2 3

c 2 c + 2 c k + 7 c k 2 k + 9 k 2 k + c
2 2 3 2 3

2 k 2 + k

c k 2 c + k

+ 2 c c k c + k

c + 4 k + c 4 3 c 4 c k + 15 c k + 7 k
3 3 3 2

+2 c 3 k 2 c 3 c k + 9 c k + 16 k +c 2 3 c + 10 c k + 15 c k 2 k + 4 c k + 19 k 2 k + c
5 4 2 2

2 k 2 + 11 k

/ r c 2 r c 2 c c k

(D.104)

247

Q3b =

i a2 + c 2

a 4 c 6 r c 4 c a 2 r c 2 c + 6 r c 2 c 2 c + a 2 c c
3

rc 2 + c c i a 4 r c 2 c + 2 a 2 r c 2 c + r c 2 c 5 k 2

2 i a 4 r c 2 + 3 a 2 r c 2 c 2 + 2 r c 2 c 4 k
2

2 i a 2 r c 4 + a 4 c 2 2 r c 4 c 2 + 2 r c 2 c 3 c + a 2 c 2 c rc 2 + c c

k
2

i a 4 c r c 4 c a 2 r c 2 c + r c 2 c 2 c + a 2 c c rc 2 + c c

k 2

(D.105)

Q4 = Re

rc 2 2 (a c k ) (a + c + k ) + a 2 rc 2 2
2 2

rc 3 r c 2
3 2

(c + k ) c + k
2

c k 2 k

+rc 2 k 2 + (c + k ) k c 2 c + k k + k
2

c k 2 (c + k ) k
2

+ rc 2 k 2 k c c (c + k ) + a4 k 2 k c c (c + k ) k
2 2 2 2 2

+ rc 4 k 2 (c + k ) + k c + k + r c 2 c 2 c
2

rc 4 c + 2 c k + k c + k + k +2 c k + k / rc 2 rc 2 2
3

+ c k k c + 2 k + c k c + k k
2

+ 4 i a4 rc 4 k (c + k )

a2 + (c + k )

k c + k

y (D.106)

k 2 (c + k ) (a + c + k ) (a + c + k ) k c + k

Q5 = Re

Q5a 2 (rc 2 )3 (a + c + k )(a + c + k )(c + k )2 +

Q5b 2 (rc 2 )3 (a + c + k )(a + c + k )(c + k )

(D.107)

248

Q5a = i rc 2 rc 2 c c

k (c + k )
3

c + k

2 2

r c 2 + c k i y k + c c + i y + k

+a6 k rc 6 k k + c 2 c k k + rc 4 2 c

+c 4 c c + i y + c + c + 4 i y k k + 2 c + i y k + c 2 c + 2 i y + k k
2

+rc 2 c c k k c 2 c + k c c + 3 k +a4 rc 10 k c 2 c k 2 (2 c + k ) k rc 8 k 3 c c c k + k +rc 6 k 3 c 2 c + c 3 c 2 + 2 c c + 2 c k k c 2 7 c c + 2 c k 4 c k + k 2 +c


2 2 2 3 3 3

k + (c + 2 k ) k

+ r c 4 c 3 c k k

2 i y k 3 + 2 c k 2 (3 i y + k ) + 2 c 2 k c + 3 i y + k + c 3 2 c + 2 i y + 3 k +c 4 c 3 c + i y + 3 c 2 (c + 4 i y ) k + 2 c 5 c c + 6 i y k 2

+ 7 c c + 4 i y k 3 k

+ 2 c i y k 3 + c 3 2 c + 2 i y + k + c k 2 4 c + 6 i y + 3 k k
3

+c 2 k 3 c + 6 i y + 4 k

+ r c 2 c c k k c 3 c + k
2

+ 2 c 2 c

k + k

+c k 2 k c + 3 k c k k 2 c + 2 c k 6 c k + k k +a2 rc 2 c c
3

k c (c + k ) k rc 4 (c + k ) + c + k
2 2

2 2

+rc 2 k c 3 + c 2 2 k + k + c c k 2 + 2 c + k k + k +k 2 c + 4 c k + k k + 2 k

/ r c 2 k k

(D.108)

Q5b = i a2 + (c + k ) +a2 rc 2 c c
3

r c 2 r c 2 c c

(c + k )

rc 2 + k 2 + a4 k rc 4 k + c 3 c k + rc 2 2 c 2 c i y +c c + c 4 i y k + 2 (c i y ) k 2 /k 2 (D.109)

Q6 = Re

2 (rc

2 2 )2 (rc

+ ac

2 2 )2 (rc

Q6a ac 2 )(a + c + k )(a + c + k )(c + k )2 Q6b + 2 (D.110) 2 )2 (a + + )(a + + )( + ) (rc c k c k c k

249

Q6a = 2 a rc 2 rc 2 c c

6 4

k (c + k )
2

c + k
2

r c 2 + c k

a8 c k c 2 c k k rc 2 c c k k c + 3 k +rc 4 c c + k +a6 c
2 2 4 2 3 2

k c + c k k
2 2

c 2 c k (c + k ) k rc 8 c + k + r c 6 c c c + k
3

c k 2 c (c 2 k ) k
2

2 (c + k ) k
2

2 c k + c (4 c + 7 k ) k 2 (2 c + 5 k ) k
2

+rc 2 c c k k 2 c k c k
2

c + 2 k k 2 k c + 3 k
2 3 2 2

+c 2 c 4 c k 7 k
2 4 2 2

+ r c 4 c
2

c 3 k c + k

+ c k 2 k c + 3 k
3

+k 3 k c + 3 c k + 3 k a2 rc 2 c c

c 2 k c + 6 c k 7 c k 13 k
2 3

c 2 c k (c + k ) k + rc 8 k c + k rc 6 c k 2 c k
2 3 3 2 2 3 3

c + k
2

+rc 2 c c k k c

c 2 + 2 c k + 2 k 2 + c k (2 c + 3 k ) k (c + k ) (3 c + k ) k + k 3 2 c + c k 3 c k k
3 2 2 3 2 3 2

+rc 4 c 3 k c + k +c k 2 4 c + 5 c k + k +a4 rc 2 c c
3 2 4 3

+ c 2 k 3 c + 4 c k + 3 c k + 3 k

c 4 c k k + rc 8 2 c k + c (c + k ) k c (2 c + 3 k ) k (c + k ) k
2

+rc 2 c c k k c 2 c k +rc 4 c
2

2 c k c 4 k
2

c + k + k 2 c + k c + k
3

c + 6 k

c 3 k c + k
2

+ k 3 c 4 k
2 2

c + 2 k
2 3 3

+c 2 k c 2 k +rc 6 c 2 c 2 k c + k
2

3 c + 6 c k + k

+ 2 c k 2 c 8 c k 7 k
3 2

+ 4 k 2 k c + k

+ c k 4 c + 4 c k + 17 c k + 5 k

/rc 2 k k (D.111)

Q6b =

2 a c r c 3 r c c c

r c 2 a 2 c 2 + r c 2 c c k 2

k a2 + (c + k )
2

2a a2 c rc 2 + c c k 2 + a2 rc 2 k 3 k 2

(D.112)

250

Q7 = Re

2 rc a4 + rc 4 2 rc 2 2 + c 2 c (rc 2 2 )
2

4 a 2 r c 4 c

rc (rc 2 2 ) c + k 2 2 + 2 a 4 r c 2 c + 2 a 2 r c 2 c 2 c a 4 c c a 2 c 3 c (D.113) 2 rc (rc 2 2 ) c + k


2 2

4 a 2 r c 4 2 a 2 r c 2 2 + a 4 c + a 2 c 2 c
2

Q8

Re

2 rc 2 a2 rc 2 2

(rc 2 2 ) k 2 (c + k )

+ a 4 c k + c r c 2 2
2

(c + k )

(D.114)

Q9 = Re

2 a2 rc 6 + c a2 + c 2 c k + rc 4 c 2 c + k + rc 2 c c 2 c 2 k a2 c + 2 k (rc 2 2 ) k c + k
2 2

(D.115)

Q10 = 0

(D.116)

Q11 = Re i a4 rc 4 + j c j + c + j k + c + j
2

k r c 2 c j + k j j k j + k + c j + k 2 j + k rc 2 + j k (D.117)
2

(c + k )

r c 4 j c j + c + j k k + r c 2 c j + k k

+a2 2 c + j + k

rc 4 2 c + j + k

+rc 2 2 j k + c j + k / a c j

a + c + j (a c k ) (a + c + k ) rc 2 + j k

Q12 = Re

i 4 a2 (c + k )

rc 2 k k + a4 rc 2 + (2 c + k ) k (c + k ) (a + c + k ) (a + c + k )
2 2

rc 2 + (2 c + k ) k

rc 2 + k k

(D.118)

251

R1 = Re
3

2 2 a + i r c c i r c c + 2 a c c

2 a2 rc 2 2 a4 rc 2 + 2 c 2 (rc 2 2 )
2 2

log(
2

rc 2 + a c 2 )+ rc 2
2 2 2

2 a2 rc 2 c
2 2 2

2 a4 rc 2 (rc
2 2 2)

log(

a ) c c

2 i i a rc a rc + a 2 rc 2 c
2 2

2 a r c 2 + a 2 c

2 a + c c

2 a r c + a c c (rc 2 + 2 )

(D.119)

R2 = Re a6 c c k c k + c 2 c rc 2 2 a c 2 rc 2 + 2
3 2 2

2 2

2 k + rc 2 (2 c + k )

c + k c + k

r c 2 2 4 c 3 k + 2 c c

c + c k + rc 4 2 c + k
2

+2 a3 rc 2 2 +rc 2 c 2 c c

2 k c c + c

+ c 2 + 2 2 + c

k 2 k + c k k c k 2 c + k

rc 4 (c k ) + c c + k
2

+a5 c k rc 2 2 c + c + k 2 c c c k + c c + k +a2 rc 2 2 c 2 c k c 2 c k + 4 2 c + k + 2 c
3 2

c + k

+rc 4 2 c 2 k + c 2 c k +rc 2 c 2 4 c + 4 c
3 2 2

c + k + c k c + k

k k c k k + c k 3 c + 4 k k 2 k c k k + c k c + k
2

+a4 rc 4 2 c + c

+c c k 2 c 2 c k + 4 2 c + k + c rc 2 c 2 2 c 2 k
2

c + k

c + k + c k c + k + c 2 4 c k 2 k k rc 2 + 2 k c + k (D.120)

/ 2 (a c ) 2 rc 2 + (a c ) c