SPACE TRAJECTORY OPTIMISATION IN HIGH FIDELITY MODELS
Industrial Engineering Faculty Department of Aerospace Science and Technology Master of Science in Space Engineering
Advisor: Prof. Francesco Topputo Graduation Thesis of: Erind Veruari Co-advisor: 837650 Diogene A. Dei Tos, MSc
Academic Year 2015-2016
To my grandparents: even if fate kept us distant, I know you have me close to your heart, and I have you close to mine. Sommario
Il campo della progettazione ed ottimizzazione di traiettorie spaziali procede di pari passo con l’evoluzione del mondo scientifico e tecnologico. Le richieste in questo ambito prevedono trasferimenti che abbiano un alto livello di accuratezza e, al contempo, un basso costo in termini di propellente a bordo. Un esempio esplicativo è rappresentato dal numero crescente di satelliti a bassissima autorità di controllo in orbita (cubesats), il cui studio per missioni interplanetarie si sta intensificando. Tra le varie strategie di progettazione di missione, quelle che sfruttano la dinamica del problema dei tre corpi offrono una serie di soluzioni a basso costo con caratteristiche stimolanti. Tuttavia, il loro utilizzo in modelli reali del sistema solare presenta grandi discrepanze. Il seguente lavoro prende spunto da questa divergenza, muovendosi in due direzioni, una teorica ed una pratica. Quella teorica prevede la riscrittura delle equazioni del moto del problema a tre corpi inserendo le perturbazioni dovute alle azioni gravitazionali degli altri pianeti, così come l’effetto della pressione della radiazione solare. Le equazioni, ottenute a partire dal for- malismo Lagrangiano, vengono poi ruotate in un sistema di riferimento roto-pulsante, nel quale si mantengono le caratteristiche delle orbite progettate in un modello a tre corpi. Esso permette, inoltre, un facile confronto tra le orbite rifinite e quelle di partenza. Dal punto di vista pratico, questo lavoro si occupa di creare alcuni algoritmi che siano in grado di propagare le traiettorie e poi ottimizzarle. Innanzitutto, vengono sviluppati alcuni strumenti numerici che permettano la soluzione del problema ad n-corpi. Per questi viene presentata una strategia di validazione tramite software open-source, i quali sono stati utilizzati nell’ambito della progettazione di missioni già in volo. In secondo luogo, si sfruttano orbite progettate in modelli meno accurati come soluzioni di partenza per la soluzione di un problema non lineare vincolato al contorno, che rappresenta il metodo di ottimizzazione della soluzione. Questi strumenti vengono, infine, applicati al caso del satellite LISA Path Finder (LPF) nell’ambito dell’estensione di durata della missione. Con questo specifico esempio si è voluto testare la capacità dell’algoritmo di ottenere soluzioni convergenti anche in quella regione altamente instabile rappresentata dai punti Lagrangiani del sistema dinamico.
iv Abstract
The design and the optimization of space trajectories goes side by side with the evolution of the scientific and technological world. In this field, there are requirements of transfers with a high-level of accuracy and, at the same time, of low costs in terms of on-board propellant. The high number of ultra-low thrust orbiting satellites (cubesats), whose study for interplanetary mission is growing, gives an instructive example of this situation. Among the different mission design strategies, the ones that exploit restricted three- body problem (R3BP) dynamics provide many low cost solutions with challenging charac- teristics. However, their inclusion in real solar system models results in high discrepancies. This work starts from this incongruence, and takes a twin-track approach, analyzing it from a theoretical and a practical point of view. From a theoretical point of view,taking into consideration the gravitational effects of a set of n-bodies and adding the perturbing effect of the solar radiation pressure. The equations have been computed exploiting the Lagrangian formalism. Afterwards, they have been rotated in a roto-pulsating reference frame (RPF), where the features of the orbits designed in R3BP are preserved. Moreover, RPF allows to easily compare guess and refined solutions. This work aims at creating some algorithms, which can be used for trajectory propa- gation and optimization in practical terms. First of all, numerical tools which solve the n-body problem are established. Then, they are validated using open-source softwares, which have been adopted in the design process of already flown missions. In the sec- ond place, formerly designed orbits in less accurate models are considered as guesses for the solution of the non-linear constrained boundary value problem, which represents the optimization strategy for the trajectory. Finally, these tools are applied to the case of the mission extension for the LISA Path Finder (LPF) satellite. Using this practical example, the converging properties of the algorithm have been tested, specifically in the highly unstable region that is represented by the Lagrangian points of the dynamic system.
v vi Contents
Sommario iv
Abstract v
1 Introduction 1 1.1 Context ...... 1 1.2 Problem definition ...... 2 1.3 State of the art ...... 2 1.4 Motivation ...... 3 1.5 Research question ...... 4 1.6 Structure of the thesis ...... 4
2 Models in inertial frames 7 2.1 Reference frames ...... 7 2.1.1 Inertial reference frames ...... 7 2.1.2 J2000 and EME2000 ...... 8 2.2 n-body problem ...... 10 2.2.1 The general n-body problem ...... 10 2.2.2 The restricted n-body problem and the Lagrangian formulation . 12 2.2.3 Planeto-centred equations of motion ...... 15 2.3 Gravitational perturbation ...... 17 2.4 Non-gravitational perturbations ...... 21 2.4.1 SRP ...... 23 2.5 Equations of motion ...... 24
3 Models in rotating frames 27 3.1 Definition of RPF ...... 27 3.2 Rotation of the equations into RPF ...... 32 3.3 Alternative derivation of the equation of motion ...... 35 3.4 Perturbations ...... 38 3.4.1 Gravitational effects ...... 39 3.4.2 Non-gravitational effects ...... 41 3.5 Logics of the models ...... 42 3.6 Special cases ...... 44 3.6.1 The Restricted 2 Body Problem ...... 44 3.6.2 Restricted 3 Body Problem ...... 46
vii 4 Validation of the models 51 4.1 Integration scheme ...... 51 4.1.1 Runge–Kutta–Fehlberg methods and ODE78 ...... 52 4.2 JPL’s SPICE ...... 54 4.3 GMAT ...... 57 4.4 Validation examples ...... 60 4.4.1 Validation with SPICE ...... 61 4.4.2 Validation with GMAT ...... 72
5 Impulsive trajectory optimization 79 5.1 Introduction to a multiple shooting strategy ...... 79 5.2 The variational equations ...... 82 5.3 Implementation ...... 85 5.3.1 Objective Function ...... 86 5.3.2 Equality Linear Constraints ...... 87 5.3.3 Inequality Linear Constraints ...... 87 5.3.4 Equality Non-Linear Constraints ...... 88 5.3.5 Inequality Non-Linear Constraints ...... 94
6 Application to the LISA PathFinder mission extension 97 6.1 Presentation of the LPF mission ...... 97 6.1.1 Saddle Point and LPF mission extension ...... 100 6.2 Guess Solutions ...... 104 6.3 Optimisation in RPF n-body problem ...... 108 6.4 Analysis of the results ...... 113 6.4.1 Direct solution ...... 113 6.4.2 Indirect solution ...... 114 6.4.3 Divergence and change of class ...... 116
7 Conclusions 119 7.1 Summary of the results ...... 119 7.2 Future work ...... 120
Acknowledgements 123
Bibliography 125
viii List of Figures
2.1 Representation of J2000 and EME2000 reference frame ...... 9 2.2 Geometrical representation of forces on general n-body model ...... 11 2.3 Change of origin: from SSB to Planet ...... 16 2.4 Change of origin: from SSB to Planet ...... 18
3.1 Representation of the roto-pulsating frame ...... 28 3.2 Logics for writing the equations of motion in RPF ...... 44
4.1 GMAT interface ...... 57 4.2 RPF trajectory for asteroids: Ceres - Fortuna - Hermione - Hektor . . . . 63 4.3 RPF trajectory for asteroids: Amor - Apollo - Einstein - Aten ...... 66 4.4 Relative error for 1862 Apollo in long integration period ...... 67 4.5 Relative error for the first 6 asteroids in long integration period . . . . . 68 4.6 Relative error for the second 6 asteroids in long integration period . . . . 69 4.7 RPF trajectory for asteroids: Eureka - Damocles - Chaos - Atira . . . . . 70 4.8 Errors on different primaries for long integration period ...... 71 4.9 Trajectory evolution ...... 75 4.10 Particular of error trend ...... 76 4.11 Trajectory evolution for J002E3 ...... 77
5.1 Multiple shooting strategy ...... 80 5.2 Sparsity pattern of equality linear contraint matrix ...... 87 5.3 Sparsity pattern of equality linear constraint matrix - particular . . . . . 88 5.4 Sparsity pattern of equality non-linear constraint matrix ...... 94
6.1 LPF technolgy, courtesy of ESA ...... 98 6.2 xy motion of LPF in RPF ...... 99 6.3 LPF manifold extension ...... 99 6.4 SP trajectory ...... 100 6.5 Intersection of LPF and SP trajectory ...... 101 6.6 Miss distance depending on departure date ...... 102 6.7 Time evolution for miss distance - Minimum miss distance ...... 103 6.8 Time evolution for miss distance - Maximum miss distance ...... 103 6.9 Synodic trajectory for minimum and maximum miss distance cases . . . 104 6.10 Shooting from 4-body to n-body: 2 manoeuvre, 5 segments ...... 105 6.11 Shooting from 4-body to n-body in different cases ...... 106 6.12 Shooting from 4-body to n-body in different cases ...... 107 6.13 Periodic solution for the phasing problem ...... 108
ix 6.14 Shooting from 4-body to n-body with different number of segments . . . 110 6.15 Shooting from 4-body to n-body: 3 manouevres, 98 segments ...... 111 6.16 Converged direct solution: 6 manoeuvres, 9 segments ...... 114 6.17 Converged indirect solution: 2 manoeuvres, 21 segments ...... 115 6.18 Converged indirect solution: 2 manoeuvres, 19 segments ...... 116 6.19 Converged indirect solution: 2 manoeuvres, 19 segments ...... 117 6.20 Converged indirect solution: 2 manoeuvres, 19 segments ...... 118
x List of Tables
4.1 Butcher’s tableau ...... 53 4.2 Name and properties of chosen asteroids ...... 62 4.3 Validation periods ...... 62 4.4 Name and properties of chosen dwarf planet ...... 62 4.5 Attractors and their gravitational parameters ...... 63 4.6 Absolute and relative maximum positions errors ...... 65 4.7 Absolute and relative maximum velocity errors ...... 65 4.8 Absolute errors for the position in Earth RPF ...... 69 4.9 Postion absolute errors in Jupiter RPF ...... 71 4.10 Mininum max. absolute errors related to the primary ...... 72 4.11 Physical and integration properties ...... 73 4.12 Initial conditions for the integrators ...... 75 4.13 Max relative and absolute errors after 100 days integration ...... 75 4.14 Initial conditions for the integrators ...... 76 4.15 Max relative and absolute errors after 100 days integration ...... 77 4.16 Comparison between the integrators ...... 77
6.1 Percentage of trajectories with miss distance smaller than the tolerance . 102 6.2 Minimum and maximum miss distance ...... 102 6.3 Values for the main parameters ...... 109 6.4 Direct trajectory properties ...... 113 6.5 Comparison between different refined solutions ...... 114 6.6 Indirect trajectory properties ...... 115 6.7 Properties of non-converging solution ...... 117
xi Chapter 1
Introduction
1.1 Context
The evolution of space exploration has highly changed throughout the centuries. It began as a fascinating study of the stars and their motion. It then developed as a philosophical question dethroning the humankind as the centre of the universe. Nowadays, it goes on in the search of new frontiers to reach, and new forms of life to unveil. Science has played the main role in these discoveries. Astronomy led the knowledge of the surrounding universe to enter in the history, whereas the orbital mechanics gave a physical meaning to the motion of celestial bodies. Astrodynamics extended those notions to man-made objects, whose presence has become a natural part even of daily life, through communication satellites and navigation systems. Space trajectory design is one of the branches of the astrodynamics that has mostly evolved in the last twenty years, due to an outstanding interest in space mission and their implication to scientific progress. The design of a trajectory considers the gravitational effects of celestial bodies on orbiting vehicles. The aim is to achieve a transfer whose requirements are met. However, new scientific frontiers need more demanding ones, and this translates into the development of higher fidelity models. The classic approach to the trajectory design has consolidate bases. The formula- tion of the problem dates back to Sir Isaac Newton and his universal gravitational law. The approximated formulation, called Two-Body Problem (2BP), has an analytical so- lution. It can be efficiently used for the trajectory design, where only one main body is considered and the artificial object describes conics around it. In the case of a transfer between different primaries the conics can be patched together. The simplification of the model permits the generation of meaningful solutions. Nevertheless, these solutions must be corrected because of the gravitational effects of other influencing celestial bodies. The perturbation of these attractors are accounted in more complex models, where the solutions are refined. New possibilities emerge as the trajectories are directly designed in more refined mod- els, like the Restricted Three Body Problem (R3BP). It takes advantage of the idea of co-rotating bodies with respect to a common barycentre. Then it considers the spacecraft as a point mass in relative motion. In this way it is easier to understand the mutual in- teraction between celestial bodies and the probe. Mathematically speaking the problem has not an analytical solution, but it is possible to find one integral of motion. In this
1 Introduction model new trajectory families are discovered, bringing to new opportunities for low cost transfers. Also the trajectory design in a 3-body model needs a refinement process. Nowadays, there are new technologies, like cubesats, that greatly sense the effect of the perturba- tions. Even scientific purposes are revealing interesting features that are not defined into a combination of only three bodies. Moreover, given that the requirements on new mission are becoming more stringent and rigorous in terms of fidelity and precision, the current trend moves forward into the improvement of all those tools that allow a deeper comprehension of the trajectory design in multi body dynamics.
1.2 Problem definition
The problem discussed in this work connects to the R3BP model as a starting point for every analysis in the trajectory design branch. It considers the highly non linear context of the model and adds new material of discussion analysing the effects caused by the simultaneous action of many gravitational attractions. In particular, the contribution of all the planets of the solar system s considered, including our Moon. This is an essential step toward a high fidelity model where the celestial bodies will not be seen as perturbations, but as opportunities to enhance the possibilities of a lower cost transfer. As a subsequent step, other perturbations will be considered, such as the solar radi- ation pressure and the inhomogeneity of gravitation field. Even if their influence could be pointless with respect to the gravitational effect of the main bodies in the short term, they are required for mid/long missions, especially for interplanetary transfers or close gravity assist manoeuvres. The meeting point of this analysis with real missions stays in the generation of feasible trajectories where the control is almost zero. The direct transformation of the control capability into fuel costs makes the problem of extreme interest and applicability.
1.3 State of the art
The literature concerning the problem of two body has comprehensive development: a basic introduction could be found in Curtis [10] and a more detailed as well as more dated one in Chobotov [9]. The perturbation analysis is present in Vallado [47] and for useful numerical simulations it is possible to look at Battin [4]. Instead, the R3BP has been extensively examined and expertly presented in Szebehely [42]. However, the problem of orbital mechanics dates back to great names: Newton, La- grange, Euler, Poincaré [34]. And since this section can not cover the vast amount of astrodynamics publications, it will be a brief collection of some authors and missions related to the main topics. This work starts as a continuation of a MSc thesis by Dei Tos [11] at Politecnico di Milano and a following article with Topputo [12]. The necessity of a high fidelity model appears in many different fields of astrodynamics: from the necessity of a study on ballistic capture, as in Hyeraci [24], to the resonant motion of the planets in Topputo [43].
2 1.4 Motivation
The groundwork has been laid from Belbruno [5], whereas Gomez [20] and the Barcelona Group are leading it forward. The missions that express the power of the new dynamics given from a three-body model are those directed to Lagrangian Points, which are mathematical points of equi- librium that do not exist into a model with a unique primary. The Herschel Space Observatory targeted the L2 point in the Sun-Earth system 1. It pointed out, in the infra-red range, cold and/or dusty objects in space. From past missions also Lisa Path Finder 2 deserves a particular remark, since it will be presented in this work. The launch of the James Webb Space Telescope 3 is expected in 2018 and it should take the inheritance of the Hubble Telescope. Even this will be targeted toward the Sun-Earth L2 point. The equilibrium points L4 and L5 are acquiring new interest as they are stability points where information of million ages are hidden. The Lucy mission [32] is headed to the Sun-Jupiter L4 and L5 points, for research and analysis on Trojan Asteroids.
1.4 Motivation
As already seen in the previous section, since R3BP has given life to new fields and missions, one could wonder why it is important to deepen the argument and add new elements. This was true until some years ago but now the technology is reaching levels where the presence of "only" three body cannot be accepted any more. The following examples may explain why. The cubesats, that are mini-satellites with an edge of 10 cm, are gathering attention due to the possibility they give in accessing space with repeatable modules with a very low cost. Given the very small dimensions, these objects cannot combine high performances with high control capability. Therefore, a good knowledge of the a priori motion is needed in order to efficiently use them: the solar radiation pressure or outer planets will induce accelerations that should not be classified as perturbation but should benefit their motion. The research on gravitational waves is producing interesting results with experiments like LIGO 4 and VIRGO 5. However, the space is an ideal place for such experiments. It is, indeed, necessary to find a point where the gravitational effects of all celestial bodies are shut down. Therefore in the perspective of a more realistic mission design, new degrees of complex- ity shall be added to the problem. Nevertheless, this will repay creating higher reliability solutions and reducing costs even more: new choices will derive from greater possibilities. Finally, even the orbit detection and the guidance and navigation control for spacecraft will get benefits: a more reliable knowledge of the designed trajectory means also a more robust transfer, where the communication with the spacecraft can be relaxed.
1http://sci.esa.int/herschel/ 2http://sci.esa.int/lisa-pathfinder/ 3https://www.jwst.nasa.gov/ 4https://science.jpl.nasa.gov/projects/LIGO/ 5https://www.ego-gw.it/
3 Introduction
1.5 Research question
The thesis will follow different paths and therefore it will try to answer several ques- tions. The main one is:
can we design space trajectories directly in high fidelity models?
Some sub-questions can help to answer it. The first one is if it possible to validate an n-body model with currently used tools for space mission. It will be searched if it is possible to build an integrator that can be successfully compared with programs such as SPICE of the Jet Propulsion Laboratory at the Caltech, or with open source programs such as GMAT. Once the integrator is validated, the real trajectory design part will be set. At this point many different strategies can be followed: is it reasonable to design the trajectory directly into the n-body model? Or is it better to design it into a less refined model and then optimize it into the high fidelity one? Are these approaches always converging or does this happen only for elementary trajectories?
1.6 Structure of the thesis
A brief description of the space trajectory design problem has already been given in this introductive chapter. Some motivations have been presented on why it is interesting to focus on high fidelity models exploitation. Then it has been formulated a research question that brings into focus the real applications of the model. The rest of the work has been organized as follows.
Chapter 2 deals with all the reference frames that have been considered in the thesis, giving an overview and describing the differences between them. Next, the equations of motion for the n-body problem in an inertial frame are derived: the procedure will exploit the lagrangian formalism for rigid bodies. After an accurate analysis of these equations perturbations are inserted into the system. The gravitational ones are included firstly as a transformation of the lagrangian of the system: mainly are considered those dealing with the inhomogeneity of the gravitational field for celestial bodies. Then the non grav- itational perturbations are added, focusing mainly on the solar radiation pressure.
In Chapter 3 the main frame of this work is presented and analysed: the roto-pulsating frame. After a description of its properties the equations of motion are rotated into this frame. To prove the correctness of this logic, an alternative derivation of the equations of motion is given. Also in this frame the adequately rotated perturbations are added to the equations. A comparison between the different logics is analysed. Eventually, some considerations are taken in order to examine if it is possible to reconstruct special cases (R2BP or R3BP) from the general one (RNBP).
The validation of the models is performed in Chapter 4. Here, the integration scheme is introduced. It follows a description of the two validation sources SPICE and GMAT. Some examples within the two programs are given showing the validity of the high fidelity
4 1.6 Structure of the thesis
model integrator.
Chapter 5 concerns the optimization process of an impulsive trajectory. The multiple shooting strategy is hereby presented. In this approach the trajectory is subdivided in different legs and each leg in separate arcs, with the aim of linking together the arcs, without discontinuities between them. To fulfil this task, the equations of motion are supplied with the variation of a particular matrix, called state transition matrix. The new equations are named variational equations and have important consequences for the evolution of the dynamical system. To carry on an optimization process an objective function is needed and it is linked with the velocity variation, which is connected with fuel consumption through the Tsiolkovsky equation. The smoothness of the trajectory is granted only thanks to some constraints, that are inserted into the dynamical problem through constraint equations. The implementation process is the one described.
Chapter 6 copes with the application of trajectory optimization to a real case: the possible mission extension for Lisa Path Finder. The chapter starts with a description of the LPF mission and its proposal for the mission extension. In fact, the satellite should pass through a particular point called Saddle Point. For this case the initial solutions already exist, but considered in a 4-body model. Starting from those guesses the trajecto- ries are optimized into a full roto-pulsating n-body model and the results are investigated.
The work concludes with Chapter 7 where a brief summary of the results is given. In the meanwhile the limits of the followed strategies are reconsidered and some hints for future works are given.
5 Introduction
6 Chapter 2
Models in inertial frames
This chapter lays the bases for the overall work presented in the thesis. The main topic is the modelling of the n-body problem in inertial frames. Considering the mutual gravitational interaction of n celestial bodies, the celestial me- chanics is that branch of the science that tries to define their motion. The astrodynamics is a branch of the celestial mechanics that tries to define the motion of a man-made object in space influenced by the gravitational fields of celestial bodies. The motion is subject to rigorous equations, that once assembled create a model, called n-body model. The motion of a body cannot be absolutely defined, but has to be specified with respect to other bodies, or a set of particular points, that delimit a frame of reference.
2.1 Reference frames
Before moving forward with the mathematical description of the n-body problem it will be necessary a foreword regarding the concept of reference frame. In fact this is a fundamental notion for this work and will be examined in many aspects. To give a definition for a frame of reference it can be said that it is a set of reference points which allow the description of the motion of an object through a coordinate system. Therefore, the reference frame can be seen as a physical concept related to the state of motion of an observer. According to Salençon [37] it can be also defined as the set of all points in the Euclidean space with the rigid body motion of the observer. For a formal definition of coordinate system the reader is referred to a manual of topology or algebra, such as Pontryagin [35] or Sernesi [39]. Here it will be expressed as the mathematical tool used to represent the physical quantities in the reference frame. This means that it is required to attach a coordinate system to a reference frame in order to have an idea of how and how much the motion of an object is changing. Then, it is possible to chose the coordinate system for sake of convenience: a spherical or cylindrical representation can be preferred to a cartesian one if the motion follows curved trajectories.
2.1.1 Inertial reference frames A reference frame is called inertial when the first principle of dynamics is valid: a free point mass, i.e. where no force acts, or the resultant of the forces is zero, will persist in
7 Models in inertial frames a state of inertia or rectilinear motion. It means that an observer in inertial frame will measure for the point mass a null acceleration for every time instant. Inertial reference frames are extremely useful since the equations of motion do not change between two of them. If two frames are inertial, the only thing changing between them is the velocity. However, by definition, this velocity difference is constant:
VIF 1(t) = VIF 2(t) + VIF 12 (2.1) where VIF 1(t) and VIF 2(t) are, respectively, the absolute velocities of the first and second inertial frame, while VIF 12 is their relative velocity. Taking the derivative of the previous equation it is possible to prove the fact that the acceleration of the two frames are equal:
AIF 1(t) = AIF 2(t) (2.2) where AIF 1(t) and AIF 2(t) represent the accelerations of the inertial frames. The second law of dynamics grants that the forces, F (t), acting on a body of a certain mass, m, depend only on the accelerations, A(t), acting on it:
F (t) = mA(t) (2.3)
Since in the two inertial frames the accelerations are equal, the forces will be equal, hence the equations of motion will be the same. Undoubtedly, this short proof cannot represent an exhaustive description of the properties of an inertial reference frame, which can be found for instance in Landau [29].
2.1.2 J2000 and EME2000 The previous description of the inertial frames is indispensable since in the following work at least two of them are used. The main difficulty for choosing a reference system in space is that it is complex to find an origin that is inertial. Usually, the system of Sun and stars is considered inertial. Therefore, according to what preciously said, all the frames that are in constant rectilinear motion in the Sun-stars frame will be inertial. The complexity is related to the definition of the orientation of the coordinate system that must be attached to it. To solve this problem it is necessary to define some constant direction in space. For this reason the equatorial coordinates are used as explained in Hoehenkerk [23]. The steps to define the celestial frame are related to the definition of the axes.
1. The z-axis is aligned with the polar axis.
2. The x-axis is the result of the intersection between the celestial equator with the ecliptic. The axis points toward the intersection of the sun with the celestial equator on its south to north journey through the sky in the spring and is also known as vernal equinox.
8 2.1 Reference frames
3. The y-axis is the vector orthogonal to both.
However, since it is difficult to measure directly the position of the vernal equinox, it is necessary to rely on star catalogues. In a certain epoch the position will be computed indirectly and after that the reference system can be used. For the J2000 frame the epoch was January 1, 2000, basing on the fifth Fundamental-Katalog, FK5 [18]. Since this frame is earth-based it must be said that it is almost inertial, because it must consider Earth rotation and a series of other factors as the direction variation of the polar axis due to precession and nutation motion: their formulation will be omitted here, but can be found in Seidelmann [38]. What is missing is an origin for the system and usually the Solar System Barycentre (SSB) is chosen. It is also possible to chose the Earth as centre for the reference system, and in this thesis work this frame will take the name EME2000, or Earth Mean Equator in J2000. The two references are just translated one with respect to the other, therefore the position vector of an object can be written as:
R(t) = RE(t) + r(t) (2.4) where R(t) and r(t) represents the position of the object with respect to SSB and Earth, respectively. RE is the position of the Earth with respect to the SSB. However it is possible to say that the passage from the first to the second is a passage from an inertial to a non inertial reference frame.
Figure 2.1: Representation of J2000 and EME2000 reference frame
9 Models in inertial frames
Taking the derivative of the previous equation:
V (t) = VE(t) + v(t) (2.5) where V (t) and v(t) are the velocities of the object with respect to SSB and Earth, respectively. VE is the velocity of Earth with respect to SSB. This means that the two reference frames are not moving of mutual constant rectilinear motion. It will be converted also in a new acceleration term that will modify the equation of motion with an apparent term:
A(t) = AE(t) + a(t) (2.6) where A(t) and a(t) are the accelerations of the spacecraft with respect to SSB and Earth, respectively, while AE is the velocity of Earth with respect to SSB. In practice, the new term is due to the fact that the Earth is rotating around the solar system barycentre. So this is what we meant for a quasi-inertial frame: in fact the two frames J2000 and EME2000 are inertial with respect to distant stars (the rotation of EME is insignificant due to the very high distance), but they are not strictly inertial one with respect to the other.
2.2 n-body problem
The general n-body problem is the starting case of the orbital mechanics. It studies the motion of n gravitational sources, due to their mutual interaction. Since the primordial formulation of the dynamics, this problem has been one of the most studied: Newton built its dynamics together with the universal gravitational law. Even if the general relativity has changed the concepts of motion and mechanics, the classical formulation of the problem still represents the more direct way to give a physical meaning to a problem that can appear as a strictly mathematical problem. To get a strong connection between the physics that lays behind the n-body problem and the mathematics that will be needed to find some solution, the following section will put forward its classical expression. After that the Lagrangian formalism will be used to define the restrict n-body problem.
2.2.1 The general n-body problem The most complete description for the motion of n bodies subjected to the mutual gravitational interaction is the n-body problem. Each body of mass mi, i = 1, .., n, de- pends on the masses mj of the other n-1 bodies. The Newton’s law for general gravitation expresses the force acting on the i-th body, Fi, as:
j=n j=n X X Rj − Ri Fi = Fji = Gmjmi i = 1, .., n (2.7) kR − R k3 j=1 j=1 j i j6=i j6=i
10 2.2 n-body problem
Figure 2.2: Geometrical representation of forces on general n-body model
In fact, equation (2.7) represents a set of 3n equations, where the Ri is the position T vector of the i-th body, whose cartesian coordinates can be written as Ri = [Xi Yi Zi] .G is the gravitational constant and Fji is the force acting on the i-th body due to the gravitational effect fo the j-th body. The latter is directly proportional to the product of the masses of the two bodies and inversely proportional to the square of their distance, that is the expression of the universal gravitational law. From the already cited second law of dynamics it is possible to relate the acceleration ai acting on a body with the forces that are generating the motion. This is due to the fact that the inertial mass and the gravitational mass are equal, so it can be written as:
¨ Fi = miai = miRi (2.8) ¨ where Ri is the differential expression of the same acceleration. In the assumption that the forces acting on the bodies are only the gravitational ones, it is possible to combine equations (2.7) and (2.8) writing:
j=n ¨ X Rj − Ri miRi = Gmjmi i = 1, .., n (2.9) kR − R k3 j=1 j i j6=i
Equation (2.9) represents again a set of 3n differential second order equations (that can be converted always in a set of 6n first order equations) where the acceleration of the i-th body is related only to the relative position of that body with respect to the
11 Models in inertial frames other attractive sources. The problem can be further simplified noting that if masses are constant and nonnull the equations set (2.9) can be written as:
j=n ¨ X Rj − Ri Ri = µj ∀i = 1, .., n (2.10) kR − R k3 j=1 j i j6=i
In equation (2.10), µj represents the gravitational parameter of a body, which is by definition the product of its mass with the gravitational constant. Since the problem is highly non-linear, solutions are hard to find. However, from Picard-Lindelöf [30] theorem, it is possible to prove that solutions exist and are unique. It is possible to see this problem as an initial value problem, or a Cauchy problem, that can be rewritten in the state space form:
x˙ = f(x) (2.11)
As remarked previously any second order differential equation can be formulated as a system of first order equations, and in the considered case it is possible to set xi = ˙ T [Ri, Ri] Given a set of 6n initial condition at a time t0, the solution is granted to be existent and unique if the function f and its gradient ∂f/∂x are continuous in a manifold containing t0 and x0. In the n-body problem the only discontinuity can be given from the denominator of equation (2.10). Since the only way for this denominator to be 0 should be Rj = Ri, and this cannot happen except in the case of collisions, the right hand side is continuous. The same reasoning can be done for the gradient. Regarding the discontinuity point, some strategies can be adopted to analytically remove it, by regularizing the equations of motion. In this way it is possible to have a continuous right hand side in the whole domain, according to the Levi–Civita transformation [40].
2.2.2 The restricted n-body problem and the Lagrangian formu- lation Solving the general n-body problem is a very interesting matter of study, however it is more related to astronomy or orbital mechanics. Astrodynamics is involved in finding the motion of an artificial object under the gravitational effect of the other celestial bodies. The mass of this object is much smaller than those of the planets, therefore, the force that it exerts on them is negligible. So the problem in converted from a general formulation into a restricted one. The motion of the planets can be seen as independent from the motion of the satellite, then their equations can be decoupled from the rest of the system. If those are independently solved then the problem transforms into finding a solution for the motion of the satellite. To give an expression for the restricted n-body problem, in this work it has been fol- lowed an approach that considers a Lagrangian formalism for rigid bodies, which has been mathematically postulated by Lagrange in 1811. All planets are assumed as point sources.
12 2.2 n-body problem
Then this way of writing the equations of motion gives the possibility of modifying them easily if non inertial reference frames are considered, or even if new perturbations are inserted. On the other hand writing the equations of motion for non inertial frames, ex- ploiting D’Alembert principle for dynamic equilibrium or cardinal equations can exhibit huge complexities, especially on the arrangement of apparent forces.
To express the Lagrangian formalism it is necessary to define a new function, called Lagrangian:
L = L (q, q˙, t) (2.12)
where L is the Lagrangian function, q and q˙ are the new coordinates and their deriva- tives, while t is the time variable. The Lagrangian of a system is a function that can describe its motion, because ac- cording to the principle of the least action (or Hamilton’s principle) a physical system in motion between two points chooses always the path that minimizes the integral of the Lagrangian for the whole trajectory [21]. It expresses a motion where it is possible to define an action functional S(q) in a time interval [t1,t2] that has a minimum:
Z t2 S(q) = L (q(t), q˙(t), t) dt (2.13) t1 From this definition it is also possible to write the equations of motion, since it must be true that:
d ∂L ∂L − = 0 (2.14) dt ∂q˙ ∂q The opportunity given from this formalism is that a suitable choice for the coordinates q can bring to a simpler way of mathematical formulation for the problem.
In this procedure the Lagrangian free coordinates q coincide with the physical ones for position R. It is then possible to express the Lagrangian as the difference between the kinetic,T , and the potential energy,V , that in the more general way can be written as:
L (R(t), R˙ (t), t) = T (R(t), R˙ (t), t) − V (R(t), R˙ (t), t) (2.15)
The next step is to define the components of the Lagrangian for the system, starting from the kinetic energy. This reads:
1 T = m R˙ · R˙ (2.16) 2 As always the kinetic energy depends directly on the mass and on the square of the velocity of the object. In this work the choice has been to keep, when possible, a vectorial
13 Models in inertial frames
notation for all quantities, in order to have a better understanding, even when those are derived or manipulated.
The definition of the potential energy includes all the gravitational effects on the body:
X µj V = −m (2.17) kR − Rjk j∈S
The used notation is as similar as possible with the one used for the general n-body problem: the µj represents the gravitational parameter of the j-th attractor, while Rj is its position vector. Therefore the quantity expressed in equation (2.17) is the gravitational potential seen as the summation of all the influences on the object. The expressions j ∈ S stays for a general model where all the desired bodies are considered, so S is a set of attractors. Redefining the Lagrangian the result is:
1 X µj L (R, R˙ ) = mR˙ · R˙ + m (2.18) 2 kR − Rjk j∈S
In order to get the equations of motion for the dynamic system, stating that the action is minimized along a trajectory, the minimum equation can be expressed as:
d ∂L ∂L − = 0 (2.19) dt ∂R˙ ∂R At this point the power of the Lagrangian formalism appears and shows how a physical problem has been translated into a mathematical expression. Everything from this point on can be automated. The first step will be to compute the different derivatives and then everything can be combined together.
d ∂L = mR¨ dt ∂R˙ (2.20) ∂L X R − Rj = −m µj 3 ∂R kR − Rjk j∈S
The complete set of equations describing the motion of the restricted problem is:
¨ X R − Rj R + µj 3 = 0 (2.21) kR − Rjk j∈S
Some little notes should be added as a comment for equation (2.21). In appearance it is very similar to what expressed in eq. (2.10), however the meaning is different. In fact equation (2.10) is a set of 3n second order differential equations, so what is written is just one of the equations of the system. In eq. (2.21) the second order differential equations
14 2.2 n-body problem
are only 3, because there is no need to solve for the motion of the planets. This fact has turned the problem to nonautonomous, because there is a hidden cariable in that equation, or:
Rj = Rj(t) (2.22)
The motion of the planets is now directly dependent upon time, which enters in this way into the equation of motion. Another important note is the frame in which these equations are written. Every step of this procedure is true if an inertial frame is chosen, in this case the J2000, and the results will be expressed as positions with respect to solar system barycentre (SSB).
2.2.3 Planeto-centred equations of motion
The equations of motion previously written define the evolution of a system referred to the solar system barycentre. However, this is not always the case of interest. Often it is more interesting to consider the dynamics of a satellite with respect to a planet, therefore, it would be convenient to have a system that can directly be integrated in that frame. In this case the results will be relative to the planet and it would be easier to have a direct view on how the quantities of the problem depend on the gravitational effect of that source with respect to others. Also, this procedure can yield to more accurate results since numerical cancellations may occur when (2.21) is integrated in the proximity of a celestial body in the set S. A further evolution of this strategy can be found in Amato [1]: the regularized equations of motion are integrated in a planeto-centred frame, that changes the primary if a close encounter happens in order to increase numerical efficiency and accuracy. It is possible to prove that applying a coordinate transformation, the result does not change. This can be done before or after writing the equations of motion. In this work it has been chosen to apply the transformation after the equations of motion have been written. Recalling equation (2.21) it is possible to see how the dynamical system depends on position and acceleration. To write the equation of motion then, it is necessary to apply a coordinate transformation acting on position, that will be directly transformed into a condition on acceleration:
R = RP + r (2.23)
In eq. (2.23) R and RP represent the position vectors respectively of satellite and planet, while r is the relative positions, that will be the new coordinate, as shown in fig. 2.3. Taking the derivatives of eq. (2.23) will bring to new relations:
˙ ˙ R = RP + r˙ (2.24) ¨ ¨ R = RP + r¨ (2.25)
15 Models in inertial frames
Figure 2.3: Change of origin: from SSB to Planet
Plugging eq. (2.25) and (2.23) into (2.21) it is possible to get that:
X r + RP − Rj ¨ ¨ r + RP + µj 3 = 0 (2.26) kr + RP − Rjk j∈S
Even if the reference system is changing, there is no modification for the set of planets S where the gravitational forces are considered. Physically all the planets belonging to S that had an effect over the spacecraft when the equations were written in SSB, will still attract it, even if the centre of the reference is changing. Now, eq. (2.23) refers not only to the spacecraft, but to every body in the system, therefore the last equation, can be rewritten in a more suitable way:
X r − rj ¨ ¨ r + RP + µj 3 = 0 (2.27) kr − rjk j∈S
where rj = Rj − RP are the relative positions of the planets with respect to the one taken as origin. In fact this is quite evident, since a change of reference cannot change relative positions or distances. However, it is possible to see another term, that represents the acceleration of the planet with respect to SSB. To express this term, let’s just recall eq. (2.10):
¨ X RP − Rk X rk RP = − µk 3 = µk 3 (2.28) kRP − Rkk krkk k∈S2 k∈S2
16 2.3 Gravitational perturbation
where S2 is a subset f S, that comprehends all the planets but the one chosen as origin for the frame. The last equation has been expressed both in SSB and in a planeto-centred version. Plugging it into eq. (2.27) the complete equations of motion can be written:
X rk X r − rj ¨ r + µk 3 + µj 3 = 0 (2.29) krkk kr − rjk k∈S2 j∈S It is worth to note that in the set S also the planet with respect to which the equations are written is included. However, for this body is true that rp = 0, hence, the complete equations can also be expressed as:
r X r − rk rk ¨ r + µP 3 + µk 3 + 3 = 0 (2.30) krk kr − rkk krkk k∈S2 From eq. (2.30) it is clear that if the summation term were neglected, what’s remaining is nothing else than the classical differential formulation of the two body problem, whose solution can be given in an analytical fashion.
2.3 Gravitational perturbation
In the n-body problem presented before, the attractors have been considered as point masses. But in real world this is not true; actually all bodies have a distributed mass. It is possible to prove that if this mass were distributed with a spherical symmetry the potential would be exactly the same. Since this is not always possible there is the need to define a new potential that accounts for the oblateness of the celestial bodies. These procedure has many references in literature: here it will be followed the strategy given in Vallado [47].
To define the new potential it is possible to see it as the summation of the potential due to infinite point masses:
X mq U = −G (2.31) ρ q q
Each point has a mass of value mq and it is at a distance ρq from the point where the potential is evaluated. Passing from a discrete formulation to a continuous, it is possible to transform the summation into an integral over the surface Q:
Z 1 U = −G dm (2.32) Q ρq
Calling r the distance of a point P from the centre of mass, rq the distance of the infinitesimal mass, the vectorial relation is true (fig. 2.4):
r = rq + ρq (2.33)
17 Models in inertial frames
Figure 2.4: Change of origin: from SSB to Planet
that scalarly can be written as:
2 2 2 ρq = r + rq − 2rrq cos Λ (2.34) The rules of cosines have been used to express the modulus of the vectors, combined with the knowledge of the angle between r and rq. Λ is called also ground angle and is expressed as:
r · rq cos Λ = (2.35) rrq If the ratio α is defined as:
rq α = (2.36) r it is possible to write: √ 2 ρq = r 1 − 2α cos Λ + α (2.37) Plugging the last equation into eq. (2.32) it is possible to obtain the new formulation for the potential:
G Z 1 U = − √ m 2 d (2.38) r Q 1 − 2α cos Λ + α
18 2.3 Gravitational perturbation
Since it will always be α < 1 and cos Λ < 1 it is possible to expand the denominator using the binomial theorem:
∞ 1 X l √ = α Pl[cos Λ] 2 (2.39) 1 − 2α cos Λ + α l=0
This equivalence is due to the fact that integrating the quantity over the square root is much more complex than expressing it as a summation of polynomials. The expression in eq. (2.39) represents the classical definition of Legendre polynomials, and according to Rodrigues’ formula [36] it has the following expression:
γ = cos Λ l 1 X (−1)j(2l − 2j)! (2.40) P [γ] = γl−2j l 2l j!(l − j)!(l − 2j)! j=0
This argument has been deeply studied in literature, therefore for the next equations there will not be mathematical proofs. In fact, it will be shown only those expression needed for the acceleration model. For further information the reader can refer to Vallado [47]. The ground angle Λ is a complex angle to be measured. Moreover, from this notation there is the necessity to decouple the effect of the central bodies, from those coming from the position of the point where the potential is computed. To do this, it is possible to write an equivalence between ground angle and latitude and longitude of the point mass (φq and λq respectively) and of the potential point (φP and λP respectively). From spherical trigonometry:
cos Λ = sin Φq sin ΦP + cos Φq cos Φp cos(λq − λP ) (2.41)
Note that this equivalence permits a new formulation of the potential through the so called associated Legendre polynomials, and the harmonics coefficients:
∞ l l µ X X RB U = − Pln[sin Φp] Cln cos(nλP ) + Sln sin(nλP ) (2.42) r r l=0 n=0
In eq. (2.42) RB is the radius of the body, Pln the associated Legendre polynomials, Cln and Sln the harmonics coefficients. It is possible to see how the meaning of that expression is that the gravitational potential can be seen as a superposition of many dif- ferent potentials, which come from the harmonics coefficients which represent the different combinations:
- Zonal harmonics: are the harmonics due to the n = 0 terms. They represents symmetrical fields about the polar axis, as simple bands of latitude, where the function is simply increasing or decreasing.
19 Models in inertial frames
- Sectorial harmonics: are the harmonics due to the n = l terms. They represent bands of longitudes where the function increases or decreases. - Tesseral harmonics: are all the harmonics where n 6= l 6= 0. Those are a combi- nation of the previous two shapes and are represented by tiles. It is possible to prove that the first term of the summation is zero ad the 0th order term is represented by the µ/r ratio. In this work it has been chosen to approximate eq. (2.42) by considering only the first term. It is the only relevant term because the trajectory of the spacecraft is far from the planet.
2 ! µ RB U = − 1 − J2 P20[sin Φp] (2.43) r r where it has been called C20 = −J2. Expressing the associated Legendre polynomials of second order, the equation (2.43) can be rewritten as:
2 2 µ µJ2R 3r U = − 1 − B z − 1 (2.44) r 2r3 r2
where rz is the third component of the relative distance. The expression of eq. (2.44) can be found in literature (Vallado [?]) and gives some indications on how to compute the potential for non homogeneous bodies. Moreover, this expression has to be considered inside the set of equations of motion, therefore distances must be written with respect to the frame in which the dynamics is defined. Remembering that the attractors belong to a set, the gravitational potential can be written as:
2 2 2 ! X µj X µjJ2jRBj X 3µjJ2jRBj rk V = U = − + 3 − 5 (2.45) kR − Rjk 2kR − Rjk 2 kR − Rjk j∈S j∈S j∈S
It is worth noting that rz is a scalar values. However, to maintain a vectorial notation, it can be rewritten as a product between vectors:
T rz = Iz (R − Rj) = [0 0 rk] (2.46) where:
0 0 0 Iz = 0 0 0 0 0 1
In this way rz is a vector that has only zeros, but for the third component. The scalar quantity can therefore be expressed as:
2 T T rk = rk rk = (R − Rj) Iz (R − Rj) (2.47)
20 2.4 Non-gravitational perturbations
since
T Iz Iz = Iz
From the Lagrangian equations it is needed to compute the gradient of the potential. To ease the derivation of the high number of terms, it can be split into three parts: dV V = Ω1 + Ω2 + Ω3 =⇒ = ∇V = ∇Ω1 + ∇Ω2 + ∇Ω3 dR
X µj ∇Ω1 = 3 (R − Rj) (2.48a) kR − Rjk j∈S
2 X 3µjJ2jRBj ∇Ω2 = 5 (R − Rj) (2.48b) kR − Rjk j∈S
2 T X µjJ2jRplj 5(R − Rj) Iz(R − Rj) ∇Ω3 = 5 2Iz(R − Rj) − 2 (R − Rj) (2.48c) kR − Rjk kR − Rjk j∈S
Thus the equations of motion can be simply written as:
R¨ + ∇V = 0 (2.49)
Comparing the last equations with those derived from the simplified potential model of eq. (2.21), it can be seen how the terms acting are always those due to a gradient of a potential. Moreover, it is possible to notice the power of a notation with Legendre polynomials. This permits not only a more complete physical representation, but also an immediate comprehension of how simplified models are only specific cases.
2.4 Non-gravitational perturbations
Every time non conservative elements are inserted in a problem, it becomes hard to use potentials and energies, therefore the concept of work will come at hand for this prob- lem. In the case of non conservative forces, according to Lagrange [28], the Lagrangian formalism can be written as in eq. (2.50):
d ∂L ∂L − = Q (2.50) dt ∂q˙T ∂qT where Q are the generalized non conservative forces. According to this formulation, it is necessary to add a term due to the generalized forces acting on a system to the previously presented equations of Euler-Lagrange, on the right hand side. These represent the variation of the generalized virtual coordinates
21 Models in inertial frames
δq that generate the same virtual work on the system. To better explain this concept, the virtual work δW as the work done by a generic force Fi over a virtual displacement δRi shall be introduced:
N X ∂Ri δW = F T δq (2.51) i ∂q i=1 where δq is the virtual displacement of the free coordinates and it has been computed using the derivation rules for a change of coordinates. Now, by definition it should be that the virtual work done by these force and the generalized forces must be the same, therefore holds:
δW = QT δq (2.52) Thus, only if the two right hand sides are equal the virtual work can be equal:
N X ∂Ri QT δq = F T δq (2.53) i ∂q i=1 The last equation, transformed into a vectorial relation, shows the equivalence between the generalized forces and the real ones:
N T X ∂Ri Q = Fi (2.54) ∂q i=1 Spending only a few words on the last expression, it is possible to say that the gener- alized forces can be seen as conversion due to a change of coordinates for the real forces. In the case when the two coordinates are the same, as in the case under consideration, the last equation simplifies up to see that the partial derivative is converted into the identity matrix and the generalized forces coincide with the summation of the real forces.
∂Ri = I (2.55a) ∂q
N X Q = Fi (2.55b) i=1 where I is the identity matrix. Finally eq. (2.50)can be written in a more suitable way as: N d ∂L ∂L X − = Fi (2.56) dt ∂q˙T ∂qT i=1 Eq. (2.56) concludes the first part of this section and shows the mathematical proce- dure to follow in order to insert the non conservative forces into the equations of motion. The next step will be to give an example of this kind of forces that enter into the n-body model.
22 2.4 Non-gravitational perturbations
2.4.1 SRP The most known and studied is the so called Solar Radiation Pressure perturbation. This is a perturbation due to the solar flux interacting with the spacecraft: in fact it is possible to think of it as a change in momentum by the photons coming from the Sun and the satellite in its motion. For this reason this interaction can be seen as a push of the spacecraft by the solar radiation, as if a pressure is exerted on it. A very simple proof can be obtained considering the power passing from the photons to the spacecraft. It is known that the power is conserved (unless dispersion as heat) and the two expression for the power are:
( Pi = ΨASC (2.57) Po = F c
This means that the input power Pi into the system is due to a solar flux of photons Ψ that encounters an area of spacecraft ASC which can be pushed. The output power Po come as a consequence of the momentum exchanged by the photons with the spacecraft: they apply a force F moving at light speed c. In any case, if the two powers are equal, the following expression can be written:
Ψ F = ASC (2.58) c The force due to the radiation pressure can be seen as the one created by a pressure pSR acting on a surface:
F = pSRASC (2.59)
Thus it is possible to relate the solar radiation pressure to the flux of photons incoming from the Sun and the speed of light:
Ψ pSR = (2.60) c The solar flux is a quantity that depends only on the distance from the Sun and it deceases with the square of that distance:
2 d0 Ψ = Ψ0 (2.61) d
where Ψ is the solar flux at the distance d, while Ψ0 at d0. So if the solar flux is known at a given distance d0, it is known in every other point of the space. In fact it is already known that a distance of 1 AU the solar flux is equal to a quantity of 1368 W/m2. This quantity has a variation depending on solar cycles and solar activities, which can be found explained in a detailed way in Hargreaves [22], but for this work it will be assumed constant.
23 Models in inertial frames
So, what has been computed is just the modulus of the force. Every force being a vector, it is also needed a direction for a full description of the physical quantity. In this case it should be considered the shape of the satellite, and in particular its surfaces. For a good approximation it can be considered the spacecraft as spherical, and the force as pushing it away from the Sun:
2 Ψ0ASC d0 R − Rs F = (2.62) c d kR − Rsk
Eventually, a correction to the last formula should be made. In fact the area that faces the solar radiation pressure should be corrected by a reflectivity coefficient. This means that, depending on the material of the satellite it can react in different ways. The coefficient cr can have a value from 0 to 2.A 0 value means that the object is translucent to the incoming radiation, and there is no interaction with it. A value equal to 1 means that all the incoming radiation is absorbed (as in the case of a black body) and the total force is completely transmitted to the system. If the reflectivity coefficient were 2 it would mean that all the radiation is reflected and twice the force is transmitted to the satellite (i.e. flat mirror perpendicular to light source). However, one last note could be added: usually, in astrodynamics is not possible to think of objects that do not absorb light. Therefore the coefficient of reflectivity is only the part bigger than 1 that should be added to the absorbed component of solar pressure. In this case cr would be a number between 0 (black body) and 1 (flat mirror).
2 Ψ0d0 R − Rs F = (1 + cr)ASC 3 (2.63) c kR − Rsk
2.5 Equations of motion
Finally, in this section, the complete equations of motion will be summarized in the cases of free and perturbed dynamics. They will be written in the J2000 frame, taking as origin the Solar System Barycentre.
Unperturbed equations of motion:
R¨ + ∇V = 0 X µj (2.64) ∇V = 3 (R − Rj) kR − Rjk j∈S
24 2.5 Equations of motion
Perturbed equations of motion: J2
R¨ + ∇V = 0 ∇V = ∇Ω + ∇Ω + ∇Ω 1 2 3 µ X j ∇Ω1 = 3 (R − Rj) kR − Rjk j∈S 3µ J R2 (2.65) X j 2j Bj ∇Ω2 = 5 (R − Rj) kR − Rjk j∈S 2 T µjJ2jR 5(R − R ) I (R − R ) X plj j z j ∇Ω3 = 5 2Iz(R − Rj) − 2 (R − Rj) kR − Rjk kR − Rjk j∈S
Perturbed equations of motion: J2 + SRP The most complete set of equations of motion will be written, considering both the effects of the inhomogeneous gravitational field and the SRP. It is importante to note that in this term will appear also the mass of the spacecraft mSC , to get the acceleration aSRP that the force described by eq. (2.63) applies to the system.
¨ R + ∇V = aSRP ∇V = ∇Ω1 + ∇Ω2 + ∇Ω3 X µj ∇Ω = (R − R ) 1 3 j kR − Rjk j∈S 2 X 3µjJ2jRBj ∇Ω = (R − R ) 2 5 j (2.66) kR − Rjk j∈S 2 T X µjJ2jRplj 5(R − Rj) Iz(R − Rj) ∇Ω = 2I (R − R ) − (R − R ) 3 5 z j 2 j kR − Rjk kR − Rjk j∈S A Ψ d2 R − R SC 0 0 s aSRP = (1 + cr) 3 mSC c kR − Rsk
25 Models in inertial frames
26 Chapter 3
Models in rotating frames
Inertial frames represent a comfortable reference to write equations of motion. No apparent forces appear and according to Einstein’s relativity they are the frames where the equations of motion appear in their simplest form [14]. This means that the compre- hension of the mathematical terms is straightforward and immediate. However, these frames aren’t always the most suitable choice. Indeed, in many cases it is very convenient to have a frame that rotates. Simply thinking of a pendulum, it is much easier to represent the tension in a frame that follows the rotation of the mass, or even a satellite rotating around the Earth, which rotates around the Solar System Barycentre.
3.1 Definition of RPF
The first historical steps into formulating rotating frames for astrodynamics were taken into the restricted three-body problem. In this problem two main bodies are taken into account, whose rotation is relative to their common barycentre. If a set of coordinates is chosen such that the x-axis is aligned, at every time, t, with the conjunction of the two bodies, and the z-axis with the relative angular momentum, the rotating frame will follow the motion. Moreover, in this frame, the two bodies will be always fixed in the chosen points of the x-axis. The frame will be then properly adimensionalized and will be ready to be used. Exploiting this basic idea, already followed from Langrange and Euler, in this work it has been developed further, according to the works of the Barcelona group [19]. It has been called roto-pulsating frame because:
- roto: it is a frame that rotates with two main bodies called primaries. The origin of the frame will coincide with the barycentre of the two primaries.
- pulsating: the distance between the primaries is not fixed, but variable in time, according to the ephemeris. It will be adimensionalized accordingly, such that in the new frame their relative distance will be always the unit length.
Following these definitions it is possible to write one equation that converts position
27 Models in rotating frames
Figure 3.1: Representation of the roto-pulsating frame and times, according to the procedure followed in Dei Tos [11]:
R(t) = b(t) + k(t)C(t)ρ(τ) (3.1a)
τ = ω(t − t0) (3.1b)
The description of the variables in eq. (3.1), depicted in fig. 3.1 follows: - R(t) represents the position vector of the satellite in the inertial frame of reference J2000. It depends on the dimensional time, t - b(t) represents the position vector of the barycentre of the primaries in the inertial frame of reference J2000. Its presence in the equation (3.1a) shows that the frame has been translated from the SSB to the barycentre of the primaries. It depends on dimensional time, because it is a function of the position of the primaries. In fact, from the definition of barycentre it is possible to say:
mpRp + msRs b(t) = (3.2) mp + ms where m and R represent masses and positions, while the subscripts P and S stay for primary body or secondary body. Usually is taken as primary body the one with higher mass. It is assumed that the position of the primary and the secondary are known: this information comes from the ephemeris. Their masses are assumed to be constants.
28 3.1 Definition of RPF
- k(t) is the adimensionalization factor for the distances. It is the present distance of the primaries, that in the new frame is equal to one. This ca be proved to be true if k(t) is chosen as the modulus of the relative distance between the primaries, bmrP(t).
k(t) = kRp(t) − Rs(t)k = krP(t)k (3.3) The position in the inertial frame f the primaries can be expressed as:
( Rp(t) = b(t) + k(t)C(t)ρp(τ)
Rs(t) = b(t) + k(t)C(t)ρs(τ)
Taking the difference of one with the respect to the other it is possible to see that:
Rp(t) − Rs(t) = k(t)C(t) ρp(τ) − ρp(τ)
Comparing the norms of the two sides of the equation results that:
kR (t) − R (t)k ρ (τ) − ρ (τ) = p s = 1 p p k(t) where the term related to the rotation of the frame doesn’t affect its modulus, as it will be proved later. - C(t) introduces the rotation into the frame. It is an orthogonal matrix or direct cosine matrix whose task is to rotate the J2000 frame into the RPF frame. The C(t) matrix can be computed as formed by three unit vectors, which are aligned with the desired directions:
r (t) P e1(t) = k(t) C(t) = e1(t) e2(t) e3(t) =⇒ e2(t) = e3(t) ∧ e1(t) (3.4) rP(t) ∧ vP(t) e3(t) = hP(t)
drP(t) where vP(t) = . dt Given this orthonormal definition of the direct cosine matrix, it will have a very useful property, i.e. the transpose of the matrix coincides with its inverse:
CT = C−1 (3.5) In the previous expressions it has been introduced also the relative velocity of the 1 primaries vP and the magnitude of their relative angular momentum hP defined as : 1From now on ∧ represents the cross product
29 Models in rotating frames
hP(t) = krP(t) ∧ vP(t)k (3.6)
- ρ(τ) is the non dimensional position vector in the RPF frame. In order to have it fully non dimensional it should also depend on a non dimensional time, such as τ.
- τ is the non dimensional time introduced into the model to make the new variable fully adimensional. The adimensional time is obtained from a dimensional time considering the relative angular velocity of the planets around the barycentre.
- ω is the relative angular velocity of the planets with respect to their common barycentre. It is used in the RPF as a non-dimensionalizing factor. In this work this quantity has been chosen as constant and has been computed according to Dei Tos [11] as the ratio between 2π and the orbital relative revolution period between the primaries, T :
2π ω = (3.7) T
The advantage of a similar choice is that the adimensional time is scaled accordingly to the revolution period, such that at for the value t = T corresponds, τ = T . The computational of the orbital period is something amply described in literature, for instance in Bradley [6], where it is related with the mass of the primaries and the mean value of the semi-major axis a˜ for a sufficiently high interval of time to remove secular perturbations, as:
s a˜3 T = 2π (3.8) G(mp + ms)
- t0 is the starting dimensional time, needed to express the relations in the new adimensional time.
At this point, since all the introduced quantities have been presented, the RPF can be described as a frame that introduces a first translation from SSB to the system and then it starts rotating as the planets do. To keep the distance between the primaries fixed it can be imagined as a pulsating motion for the frame: it tightens and stretches following the positions of the main bodies. To continue and rotate the equations into RPF there is the need to take some deriva- tives. As already mentioned, in non-inertial frames there are some apparent forces. To understand their meaning, eq. (3.1) is derived in time.
R˙ = b˙ + k˙ Cρ + kCρ˙ + kCρ0τ˙ (3.9a) τ˙ = ω (3.9b)
30 3.1 Definition of RPF
Taking the first derivative of the position vector, the derivatives of the translational motion, b˙, will appear (it can be seen as velocity of the RPF with respect to the inertial frame) and that of the rotational motion. In particular this will have three terms linked to k˙ , the rotation of the frame, C˙ , and the relative velocity, ρ0. Focusing on the relative velocity on the RPF, it has been said that it depends on the adimensional time and no longer on the dimensional one. In this way it is possible to explain the presence of the τ˙. Following the chain rule:
dρ dρ dτ = = ρ0τ˙ (3.10) dt dτ dt Note that all time dependencies have been dropped to ease the notation.
R¨ = b¨ + k¨C + 2k˙ C˙ + kC¨ ρ + 2 k˙ C + kC˙ ρ0τ˙ + kCρ00τ˙ 2 (3.11a) τ¨ = 0 (3.11b) The second derivative represents the inertial acceleration of the spacecraft and its conversion. It is possible to see how many components arise. Given that all these parameters will be used, it is interesting to have an overview of also their derivation procedure.
The adimensionalization factor k:
rP · vP k˙ = (3.12a) k ˙ k(vP · vP + rP · aP) − krP · vP k¨ = (3.12b) k2
The first derivative of the rotation matrix:
kv − k˙ r e˙ = P P 1 k2 e˙2 = e˙3 ∧ e1 + e3 ∧ e˙1 (3.13) h (r ∧ a ) − h˙ (r ∧ v ) e˙ = P P P P P P 3 2 hP The second derivative of the rotation matrix:
(2k˙ 2 − kk¨)r − 2kk˙ v + k2a e¨ = P P P 1 k3 e¨2 = e¨3 ∧ e1 + 2(e˙3 ∧ e˙1) + e3 ∧ e¨1 (3.14) (2h˙ 2 − h h¨ )r ∧ v − 2h h˙ r ∧ a + h2 (r ∧ j + v ∧ a ) e¨ = P P P P P P P P P P P P P P 3 3 hP
31 Models in rotating frames
The angular momentum hP:
˙ (rP ∧ vP) · (rP ∧ aP) hP = (3.15a) hP 2 ˙ ¨ hP (rP ∧ aP) + (rP ∧ vP) · (vP ∧ aP + rP ∧ jP) − h(rP ∧ vP) · (rP ∧ aP) hP = 2 hP (3.15b) All the relative computed values can be expressed as the difference of the absolute value of the secondary and the primary:
r = R − R P s p vP = Vs − Vp (3.16a) aP = As − Ap jP = Js − Jp From the ephemeris only the value of position and velocity is known. The other terms represent the second and third time derivatives of the position and they have to be computed according to new formulas:
dvP daP aP = jP = dt dt where aP is the relative acceleration, while jP the relative jerk. The absolute values for these can be computed according to the universal gravitational law:
X R − Rj A = − µj 3 (3.17) kR − Rjk j∈S
The absolute acceleration for the primary body AP will be computed according to eq. (3.17), choosing as position RP. The same for the secondary body. To obtain a formulation for the jerk, it si necessary to derive the acceleration in eq. (3.17) with respect to time.
X V − Vj (R − Rj) · (V − Vj) J = − µj 3 − 5 (R − Rj) (3.18) kR − Rjk kR − Rjk j∈S The absolute jerk for the primaries can then be computed applying a substitution to the general position R and velocity V their specific values.
3.2 Rotation of the equations into RPF
The main quantities that have an important role into the description of the roto pulsating frame have been introduced. The next step will be to exploit these quantities and build the equations of motion in this frame. As already said, in this frame there are two important benefits:
32 3.2 Rotation of the equations into RPF
- The equations of motion result directly adimensionalized due to the conversion from one reference to the other. This limits computational errors, since quantities which can be of different magnitudes are opportunely scaled.
- The solutions designed in the R3BP ca be easily interpreted and refined in this frame, that represents the natural extension of R3BP dynamics.
- It is easy to check if the refined solution retained the features of the R3BP one.
The equations of motion can be rotated following different procedures, which should bring to the same results. In this section a first procedure will apply the transformation directly on the differential equations, which can be the simplest way. In a second mo- ment the equations of motion will be derived for a system where the transformation is applied before the statement of the Lagrangian. Following the structure of the previous Chapter the free dynamics is expressed and then the equations are modified due to the perturbation terms.
Plugging (3.1a) and (3.11a) into (2.21) comes out:
2 00 ˙ ˙ 0 ¨ ˙ ˙ ¨ ¨ X C(ρ − ρj) τ˙ kCρ + 2τ ˙kC + 2τk ˙ C ρ + kC + 2kC + kC ρ+b+ µj 2 3 = 0 (3.19) k kρ − ρjk j∈S
where the adimensional position of the attractors ρj are computed inverting (3.1a):
CT (R − b) ρj = (3.20) k
Is worth noting the development of the relative position terms. It can be seen how they are now independent from b:
R − Rj = kC(ρ − ρj)
Regarding the distance, this should not depend on the reference system, since they are presented as relative quantities. It can easily proved that:
n n n kR − Rjk = k kρ − ρjk
It can be seen that the rotation matrix does not appear and this is a direct consequence of the orthonormality property. In fact:
33 Models in rotating frames
n n q kR − Rjk = (kC(ρ − ρj)) · (kC(ρ − ρj)) q n T = (kC(ρ − ρj)) (kC(ρ − ρj)) q n 2 T T = k (ρ − ρj) C C(ρ − ρj) q n 2 T −1 = k (ρ − ρj) C C(ρ − ρj) q n T = k (ρ − ρj) (ρ − ρj)
n n = k kρ − ρjk
In order to allow implementation, eq. (3.19) should be written in the canonical form:
! ! 1 2k˙ 1 k¨ k˙ ρ00 + I + 2CT C˙ ρ0 + I + 2 CT C˙ + CT C¨ ρ τ˙ k τ˙ 2 k k ! 1 1 T ¨ 1 X ρ − ρj + 2 C b + 3 µj 3 = 0 (3.21) τ˙ k k kρ − ρjk j∈S where I is the identity matrix. It is possible to note that all the products of the direct cosine matrix with its transpose have disappeared, as expected. Eq. (3.21) differs from eq. (2.21) not only on the number of terms in the expression, which come out after the rotation, but mostly on the domain of the integration time. The independent variable ρ is in fact dependent on τ. Its derivatives are then taken with respect to this variable. However, it is possible to see in equation (3.21) many terms which depend on time t and have also derivatives taken with respect to it. The dissimilarity is mandatory for the equation of motion of this system and is originated from two different requirements:
- On one side the independent variable will be the one which is integrated. The numerical procedures introduce numerical errors and those are even bigger when considering very small or very big numbers. Since the characteristic distances of the system are in the order of the astronomical units, the magnitude is in the order of 1010 − 1011 m. This translates into very high characteristic times: to have meaningful trajectories for such a system, hundreds of days are required, therefore in the order of 106 −107 s. Adimensionalizing this number will reduce the numerical error introduced in the integration procedure. - on the other side, there are some terms that depend on the real time, or rather on the epoch. As already said, to have a high fidelity model, it is needed to get the
34 3.3 Alternative derivation of the equation of motion
position of the celestial bodies from their ephemeris, which are dependent on the considered epoch time. All those variable then cannot depend on the integration time. Indeed, they have to be computed on-line, but their dependence should be left to the dimensional time, which makes them quantities with a precise physical meaning. This makes the problem non-autonomous.
3.3 Alternative derivation of the equation of motion
An useful exercise to better comprehend the terms of the equations of motion is to try to obtain them in a different way. In the previous section these were derived in an inertial frame, then they were rotated into RPF at the level of the differential equations. Taking into consideration operating the transformation before the Lagrangian definition, all the quantities shall be expressed with respect to the same independent variable, the non dimensional variable. In order to make the exercise mathematically complete:
R(τ) = b(τ) + k(τ)C(τ)ρ(τ) (3.22)
Since the Lagrangian formulation has many time dependent variables, the chain rule for a generic quantity follows:
∼ (t) =∼ (τ)
d ∼ d ∼ dτ ∼˙ = = =∼0 τ˙ dt dτ dt
d∼˙ d(∼0 τ˙) d ∼0 dτ˙ dτ d ∼0 dτ˙ ∼¨ = = =τ ˙ + ∼0 =τ ˙ + ∼0 =τ ˙ 2 ∼00 + ∼0 τ¨ dt dt dt dt dt dτ dt
According to its definition, the Lagrangian is:
L = T − V where the relations depending on τ hold:
R0 = b0 + k0Cρ + kC0ρ + kCρ0 (3.23)
mpRp + msRs b = (3.24) mp + ms
35 Models in rotating frames
In the new frame, the kinetic energy can be written, exploiting eq. (3.23), as follows:
1 T = R˙ · R˙ 2 τ˙ 2 = R0 · R0 2 τ˙ 2 = b0 + k0Cρ + kC0ρ + kCρ0)T (b0 + k0Cρ + kC0ρ + kCρ0 2 (3.25) τ˙ 2 = (b0T b0 + k02ρT ρ + k2ρT C0T C0ρ + k2ρ0T ρ0 2 + 2k0b0T Cρ + 2kb0T C0ρ + 2kb0T Cρ0 + 2kk0ρT CT C0ρ + 2kk0ρT ρ0 + 2k2ρT C0T Cρ0) Furthermore, the potential energy can also be defined in the new frame: X µj V = − kR − Rjk j∈S X µj (3.26) = − kkC(ρ − ρj)k j∈S It is possible to see that both T and V have been defined according to their expression in the inertial frame, and after that the rotated quantities have been introduced. This is due to the fact that the Lagrangian formulation requires certain quantities that are related with physical properties of the system. At this point it is possible to express the Lagrangian as dependent on the two energies, which have been completely rotated into the new frame:
L (ρ, ρ0, τ) = T (ρ, ρ0, τ) − V (ρ, τ) (3.27) The equations of motion derive from the Lagrange formalism, which can be written as:
d ∂L ∂L − = Q = 0 dt ∂ρ˙ ∂ρ Once again it must be noted that the previous equation is considered for inertial frames, where the independent variable is time t, depending on ρ and no more on R . Since in this case the variable does not depend on t, but on τ, those equations can be rotated following the chain rule:
dτ d ∂L dt ∂L − = dt dτ ∂ρ0 dτ ∂ρ d ∂L 1 ∂L τ˙ − = (3.28) dτ ∂ρ0 τ˙ ∂ρ d ∂L τ¨ ∂L ∂L − − = 0 dτ ∂ρ0 τ˙ 2 ∂ρ0 ∂ρ
36 3.3 Alternative derivation of the equation of motion
New terms have appeared on the formulation of the problem and it will be necessary to compute them all:
∂L =τ ˙ 2 k2ρ0 + kCT b0 + kk0ρ + k2CT C0ρ ∂ρ0
d ∂L = 2¨τ k2ρ0 + kCT b0 + kk0ρ + k2CT C0ρ dt ∂ρ0 +τ ˙ 2(2kk0ρ0 + k2ρ00 + k0CT b0 + kC0T b0 + kCT b00 + k02ρ + kk00ρ + kk0ρ0 + 2kk0CT C0ρ + k2C0T C0ρ + k2CT C00ρ + k2CT C0ρ0)
τ¨ ∂L =τ ¨ k2ρ0 + kCT b0 + kk0ρ + k2CT C0ρ τ˙ 2 ∂ρ0
∂L 0 = k02ρ + k2C0T C0ρ + k0CT b0 + kCT b0 + kk0ρ0 + k2C0T Cρ0 − ∇V ∂ρ
Plugging all the terms into (3.28) it is then possible to arrive at the equations of motion, which are completely dependent on the adimensional time τ: