/ A HIGH THRUST OPTIMIZATION THROUGH HAMILTON - JACOBI THEORY by ALAN ROBERT MITCHELL S.B., Massachusetts Institute of Technology (1965) SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY June, Signature of 1966 Author_Signature redacted L Department of Aeronautics and Astronai;tics, June 1966 Certified by_ Signature redacted Thesis SupQrvisor Accepted by Signature redacted Chairman, Departmental Graduate Committee 038 ii HIGH THRUST OPTIMIZATION THROUGH HAMILTON - JACOBI THEORY by Alan R. Mitchell Submitted to the Department of Aeronautics Astronautics on 20 May 1966 and in partial fulfillment of the requirements for the degree of Master of Science. ABSTRACT This thesis begins a sequence of closed form approximations to minimum time, high thrust space vehicle transfers in an inverse-square, central gravity field. A Hamilton-Jacobi perturbation procedure is used. The familiar zero gravity and constant gravity closed form solutions are derived with the Hamilton-Jacobi formalism, and the solutions are written in a more convenient form, for the ensuing perturbations, than the forms available in the literature. The overall goal of a perturbation solution or of a family of perturbation solutions is to approximate minimum time solutions as they would exist in a gravity field comprised of the first n terms in a Taylor's series, position expansion of gravity about the initial point. The constant gravity solution precisely expresses the effect of the first term in the gravity series. Then a family of perturbations, (which are shown to essentially number two), are required to approximate the effect of the second, linear term in the gravity series. The first member of the linear gravity approximations, -id iii relative to the precise linear gravity solution, produces the well known costate, the position's homogeneous solution, the position's particular solution due to the constant gravity term, and, in some sense, produces an approximation of the position's particular solution due to the optimal linear field thrust program. The linear gravity family of approximations must then improve the position's particular solution due to the thrust, and a procedure for accomplishing this in one step is outlined. A precise example trajectory, within the context of the problem's model, is solved on a computer, and the important results are graphed. Only a brief analysis of the numerical results is made to study the second member of the linear gravity approximations. The second member of the linear gravity approximations and later solutions for a higher order gravity series are left for future research. Thesis Supervisor: Title: James E. Potter Assistant Professor of Aeronautics and Astronautics iv ACKNOWLEDGEMENTS The author is gratefully indebted to William E. Miner of the NASA-Electronics Research Center for inspiring and continually assisting the development of this thesis. Professor James E. Department, To his thesis advisor, Potter of the M.I.T. Aeronautics the author expresses his gratitude for invaluable criticism and support of this thesis. For many fruitful discussions, the author extends a thank you to Theodore N. Edelbaum of Analytical Mechanics Associates and David Schmieder of NASAElectronics Research Center. Marilyn2 Wisowati, who aided the computer programming, and Linda Miles, who typed the manuscript, also earned the author's gratitude. Most essential was the wonderful encouragement of the author's family, to whom the thesis is dedicated. v TABLE OF CONTENTS Page No. Chapter No. 1 Introduction 2 The Formulation 7 The Cost Function and the 2.1 Differential Constraints 2.2 The Optimal Control and the Canonical Equations Canonical Transformations 2.3 and the Hamilton-Jacobi Equation 2.4 The Perturbation Procedure The Gravity Expansion 2.5 3 The Base Solution 3.1 3.2 17 The Results The Generating Function of Zero Gravity 3.3 The Ordinary Integrals 3.4 The Canonical Transformation for Zero Gravity 3.5 The Canonical Transformation for Constant Gravity 4 5 The Numerical Example 4.1 The Chosen Problem 4.2 4.3 4.4 The Differential Equations The Transversality Conditions The Numerical Output The Linear Field Problem 5.1 The Definition of an Accurate Solution 5.2 The Third Perturbation and the Results 5.3 The Derivation of the Third Transformation 5.4 The Ensuing Perturbations as One Perturbation 5.5 A Brief Analysis of 32 44 vi 6 Summary and Conclusion 6.1 6.2 64 A Summary of the Results Concluding Remarks Appendices A The Constant Field Parameters 69 B Schmieder's Sweeping Parameters 77 Figures 1 Example r(t) _ 2 Example e(t, '6 f+ 3 True and Fitted Linear Field>,, 4 9,&A, 83 6 01) 02,C13)c 4 1 15. 85 7 b,, b2, b3 86 _ (t1 80 81 82 5f,84 References b4 87,88 vii NOTATION General it ; one sub bar for a column vector N ; two sub bars for a matrix ; super-T for transpose S or ; (if iL is a defined vector), component label or ; (if m is not defined), time label ; row vector gradient 9;/bv ag/b6 ; matrix of rows m ; total time derivative ; as supermarks, ; enclosing a submark, variable's or L variable's label label Defined Vectors r position from dominant c.g. Lr r , ie. velocity gravity, -Af r the state of the costate and A,,, the R V~i*M of , _! and mass M Atri ihparameter (vector parameter) ; the first, second and third segments of X viii T (vector ; the i" coparameter coparameter) ; the first, spcond and third segments of..X ; thrust -IM~ IT) Defined Scalars t ; m ; mass 0' ; mass flux C ; exhaust velocity 'A ; constant 9 r ;jri U11- Mf time gr2 ~I~1 ;;h canonical generating function H <;> l" total Hamiltonian afterit" perturbation ; segment of H "i orti+i) perturbation solution H (")* ;H H."'+ CHAPTER I Introduction It is desired to approximate the minimum time flight of a high thrust space vehicle in closed form. This requires a perturbation method, (H-J) perturbations are chosen. and Hamilton-Jacobi Historically, in this investigation the perturbation method searched for meaningful problems that it could solve, one of which is the high thrust space vehicle transfer of minimum time. If but a single optimal trajectory and perhaps the field of neighboring extremals were sought, then one would numerically solve sets of ordinary differential equations, like the canonical and the Riccati equations, in a two point boundary value problem. A large field or bundle of optimal trajectories may theoretically be generated through the solution of a first order, partial differential equation, the H-J equation. The numerical approach to the H-J equation, dynamic programming, is virtually impossible for systems of high dimension, viz. the dimensions of present interest.' In any case, dynamic -2- programming is difficult to apply unless the end state that one desires is fixed. Closed form approximations to the H-J solution may result from an H-J perturbation method when the system differential constraints are well behaved, closed form functions, as for the gravity field of space.'*8 H-J perturbation theory might better be called canonical variation of parameters theory. The minor problem that constitutes a perturbation is solved from the canonical or the H-J equations of the minor problem such that one obtains new variables which are canonical constants of the motion in the minor problem and canonical variables in the total problem. A set of variables is canonical in a problem if the set segments in half, (into what are to be called parameters), the parameters and the co- such that there exists a Hamiltonian of the problem with respect to which the variables obey the canonical differential equations: just as the state and co-state are canonical with respect to the original Hamiltonian of the total problem. A transformation of variables is canonical if canonical variables are transformed to canonical variables. Each perturbation solution will take the form of canonical constants of a previous minor problem transforming to canonical constants of the latest minor problem. Each perturbation -3is initiated by an expansion of the latest total Hamiltonian, which may be regarded as the scalar embodiment of the remaining dynamics. Ideally, the minor problems on the average converge to the total problem in the sense that the latest canonical variables tend to become virtual constants in the total problem too. When one is convinced that he has an excellent approximation to the total solution then he may regard the latest variables as canonical constants of the total problem which, with time, through all of the transformations, implicitly define the closed form, general solution of the original canonical equations for the state and co-state. At the expense of retaining canonical form, one gains a well defined perturbation method where the differential equations for each perturbation's set of variables is immediately known in the canonical form, and the next perturbation is always in terms of a scalar function, the Hamiltonian. Also, the canonical form is compatible with the requirements of the digital computer and thus saves considerable effort in numerical investigations. The unfortunate aspects of H-J perturbations include the immense effort involved in retaining canonical form, the ease with which one is lost in minor problems of abstruse physical significance, and the impossibility -4of knowing a priori the form of the solutions beyond a quite early stage. In the present investigation the total problem is modeled as a deterministic flight of a single stage high thrust vehicle in an inverse square central gravitational field. the minimum time One desires flight from a fixed initial state to a manifold in state-time space. The first and second perturbations are for the field free and flat earth problems, and together they are considered the "base" solution. The third perturbation begins a family of perturbations to include the effect of the linear term in a Taylor's series expansion of gravity about the initial point. This family of perturbations will barely be started at the close of this thesis, for more numerical investigation is required before one may ernestly begin the family. However, the third perturbation is singularly important. It immediately produces the well known guidance law of the entire linear field problem, and this will remain invariant: the co-state co-parameters will be related to the and time in an invariant form. Only the parameters will change with each perturbation, always be related to the so-called and they will state and time and invariant functional forms of the invariant co-state. with the third perturbation Therefore, the complexity of including the linear field problem is cut in half. -5- Because an H-J perturbation solution can have sufficient accuracy in a large but limited region and because one requires numerical studies to support the many analytical decisions in forming perturbations, it is necessary to have an example optimal trajectory that one solved on a computer. The example trajectory is chosen to be similar to a Saturn V, second stage transfer to near earth circular orbit. H-J perturbation theory for the calculus of variations has not yet produced a good solution for a meaningful space trajectory problem. H-J perturbation theory has been applied in analogous mechanics' problems (uncontrolled). Delaunay, without the benefit of the computer and after twenty years of work, produced a successful lunar theory through H-J perturbations. S Of course, Delaunay could see the true solution. There exists a vast literature from celestial mechanics on H-J perturbation theory for long term, almost periodic motion and disturbances; eg. the von Zeipel method. This literature is of no use in the present problem. However, the power series expansions in radius and time or mass that are useful here are much simpler to work with than the forms common to celestial mechanics, like Fourier-Bessel series, for the lack of references. and this is some solace -6- Notable in the present research for H-J perturbation solutions of optimal space transfers is work of William E. Miner at the NASA Electronics Center. Mr. Miner is most thrust solutions, the Research intently working on low but his larger, long term interests have made him an invaluable advisor for this high thrust investigation. H-J theory for the calculus of variations may seem to result from purely the first order necessary conditions, but a more rigorous development than found here will show the theory to be intimately dependent upon second order, sufficiency conditions like controllability and the absence ie. of conjugate points, a simply covered region of extremals. Therefore, the following development is prefaced with the assumption that one operates in a "field" in the sense of Bliss. -7- CHAPTER 2 The Formulation 2.1 The Cost Function and the Differential Constraints The minimum time problem is written in a Mayer form, so the cost is simply the final time, at. In this early stage of research one must confine the problem to a single stage, continuous thrust trajectory. The latter condition is insured in formulating a minimum time rather than a minimum fuel problem.9 It is assumed that the change in radius over the initial radius and that the transfer time over some local circular orbit period are small ratios. Position, . , is written in nonrotating, Cartesian coordinates from the dominant center of gravity. Velocity,y = _' , and mass,m , combine with r to form the state, x. _ rM (2.1.1) Gravity, #, _ - r is defined (2.1.2) - -8 Thrust,T, is defined CT)(2.1.3) T= constant max. mass flux constant max. The control, _1 = _1) exhaust velocity (2.1.4) (2.1.5) , is a 3 dimensional unit vector of arbitrary direction. It is desired to take a vehicle initial state,X (t,) = X, t,= 0 , , at a fixed from a fixed initial time, to a manifold in state-time space in minimum time, subject to the differential constraints Ol.= F( 2u) + 24.2 t (2.1.6) -rI The Optimal Control and the Canonical Equations If one introduces the costate, , where costate of C - costate of i (2.2.1) -- costate of m and then defines the Hamiltonian,H , where ) H AT F(X - Ari T+AE49.+ atm : ryT ] - M(2.2.2) -9- then it follows that the optimum control, maximizes (or minimizes, if one prefers) , L:o, H (2.2.3) = One may then write H '>= __F(XU( ,)) Ar+and it is necessary that X N respect to + and \ be canonical with on an optimal trajectory.6 _ 2.3 (2.2.4) C (2.2.5) 9r Canonical Transformations and the Hamilton-Jacobi Equation A set of variables is canonical in a problem if it segments in half, coparameters, into the parameters and such that there exists a Hamiltonian of the problem with respect to which the variables are canonical. Mathematically, parameters, X , there exists a Hamiltonian, , and the coparameters, by the canonical equations: A , of the related -10- H A transformation (2.3.1) of variables XA (2.3.2) (2.3.3) is canonical the if jn variables and the For every canonical are canonical. (I1) variables transformation there exists a generating function of a form like (i 5 5 t)(2.3.4) where det. (2.3.5) X and -A x The function Eq. (2.3.6) (2.3.2) and Eq. -.of (2.3.6) _ S is a generating function because implicitly defines the transformation (2.3.3), and any such function obeying (2.3.5) may so generate a canonical transformation. It then follows that the new Hamiltonian is (H<'S yS + -X (2.3.7) .A1[: -11- Actually, some transformations must be related to a sequence of transformations, S where is a sum of generating functions obeying the group property. The transformation (2.3.2) general solution of Eq. new variables (2.3.1) are constant. H ' 0 (2.3.3) is the if and only if the From the canonical equations, the new variables if and only if and are identically constant . Eq. (2.3.7) is then called the H-J partial differential equation. 0 If Eq. =H (-X) (2.3.8) ,) ) +, if-s--- (2.3.8) Saa S can be solved in closed form then produces the closed form solution to the problem described by Eq. (2.3.1). S'by One may either attempt to find separation of variables or by the characteristic equations. The characteristic equations are canonical equations (2.3.1) plus (2.3.9) If._.A_ appears Eq. (2.3.%) tell one that If one can solve Eq. form when H linearly in S * then Eq. (2.3.9) with is a constant of the motion. (2.3.1) or Eq. (2.3.8) in closed is the total Hamiltonian of a problem then one can solve the total problem in closed form. In that -12- case it makes no sense to work backwards characteristic equations to form from the S611, for that amounts to solving the total problem in two interdependent ways. However, one generally cannot solve the total problem in closed form, and so a closed form approximation through a pefturbation method is solution. Such is the only way to obtain a literal the case Because H-J perturbations work backwards 5 "i (j i) 2.4 , where " in the present problem. are chosen, it makes sense to from the characteristic equations S to form relates to the solution for the perturbation. The Perturbation Procedure If one can find a significant part of that would lead to a closed form total solution if it were the total problem's Hamiltonian then it may be considered for the (i + o)stperturbation. H <=H =H. 0 Assume that Eq. H + + (2.4.2) ' S is +(2.4.1) (2.4.2) solvable in closed form with or without the aid of the characteristic equations. H0 (' is considered to be significant may result from physical, intuitive reasoning or from numerical That H#I ;+1_(2.4.3) -13- studies that compare the canonical equations to H"' . The derivatives of the variables due H, o'i+)need not be greater than those due toH"' , and for H. but the former rates should at least be of the same order of magnitude as the latter. of many perturbations And on the average, the effect should be to make the latest variables more nearly constants of the total motion. One may regard a perturbative H-J equation as a way of getting a canonical generating function that transforms the constants of the ih perturbative problem into the constants of the (i+1)" where the latter problem is true problem. perturbative problem, ideally closer to the The canonical constants of a perturbative problem are canonical variables of the total problem. From Eq. (2.3.7), the new total Hamiltonian is H I It is intuitively clear from Eq. (2.4.3) that the new total Hamiltonian is the embodiment of the remaining dynamics. It is convenient to write each Hamiltonian without the explicit appearance of For example, if (i4')(2.4.4) time. -14- then the canonical equations or Eq. , that problem. Therefore, time is separable in (2.4.5) or the parameters number Applying the generating Eq. ; 41 (minor) S +5"(1 We assume the coparameters . tell one is a constant of its perturbative A'=t Ii (2.4.2) (2.3.6) S (2.4.6) As (2.4.7) S t='64(2.4.8) Assume that H (I in total was written without time. The above equations tell one that time will always appear as (:,,-t) in the transformation, the new total Hamiltonian, H "'. and thus in If one wants the new total Hamiltonian to be a constant of the total problem then one may use a time transformation to starred variables: S+ jn-. +(2.4.9) -15- As a result, the starred and unstarred variables are the same except for =* X- * -- (2.3.7) From Eqs. (2.4.10) (2.4.3)... and H ~'(2.4.11) 4-H It happens that in the present problem the time transformation may always take the above 5 so we will assume that after every an Therefore, . H(j41)will form, there follows mean and the stak-is only retained on In summary, one has a total, time invarient Hamiltonian of the in variables, H® . portion, H, A significant ,that one can solve in the H-J equation is used to define the generating function of the(i+1)S problem, . The generating rules are applied and one hopes to explicitly define the old variables in terms of the ( j*O5s problem. by set of variables-constants of their minor Time will appear as . (x,, - , which is replaced The new total Hamiltonian is therefore constant, and is found by Eq. (2.4.11). -162.5 The Gravity Expansion The trajectories with which we deal are characterized by a small change in initial It therefore makes physical sense to r(t,). H< )with initiate the expansion of expansion of r r campared to the a Taylor's series about its initial value. r (2.5.1) +,23/(2.5.2) + 2-2 7,1N(+0 ) (2.5.3) where 7(2.5.4) _ N L -3(2.5.5) Placing Eq. (2.5.2) forA into the total Hamiltonian of the state and costate, H"o'of Eq. will allow the perturbative process to begin. H0 (2.2.4), will represent the field free case, and H. will represent the constant field case, (flat earth) . The constant field case is the base solution. 3 Then H.,will be an important first member of a family of minor problems in . designed to incorporate the effect of the linear term -17- CHAPTER 3 The Base Solution 3.1 The Results The closed form integrals of the canonical equations for the problems of zero gravity and constant 6,10 gravity are well known in one form or another.' In a variation of parameters study, Ross has written the differential equations of the total problem for the constant g integration "constants", producing better equations for numerical integration'. more: way. We desire much improving closed form approximations all the The initial conditions for the state and costate in these simple problems could be used as a second set of canonical variables, but a much simpler form for the first canonical transformation results from the following somewhat arduous development. To the author's knowledge, no one has produced a simpler constant 9 transformation than, for example, Ross's, canonical but a simpler form is desirable for a perturbation theory. It will be shown that the following transformation is the general solution of the constant -18- problem. Segment the parameters and coparameters: 2 V (3.1.1) A4* (R I (3.1.2) = = (3.1.3) a (3.1.4) -LL(R) 2 =~3I + M(RA (3.1.5) The state is then given by a r=fA- (3.1.6) + tr <V -M *g, cf / -M if "')JI -7C - R (3.1.7) 3j & )m where X=T (3.1. 8) (3.1.9) In~ ll -19- r' (.&LS + -&(F0-ALLVI (3.1 .10) + (3.1.11) tAI-Ati problem are the same, except for..A(m, I 2A - The coparameters of the zero + r0 , The parameters of the zero $ to iM 9 AR I3 .1.12) problem are simply related Y> M *, am 3.2 (3.1.13) The Generating Function for Zero Gravity Using Eq. (2.2.4) the zero- Hamiltonian is defined. (3.2.1) Therefore, , - (3.2.2) , H'"= K;, +Iot' rn' the first H-J equation is 0 = H,"_X')+S(,it) (3.2.3) -20- J. If the first coparameter is , there is a zeroeth to relate the notation to chapter two, coparameter, this is clearly the costate, Similarly, _. For brevity, is the zeroeth parameter,X. and the state,X the parameters or parameter vector is termed simply "parameter", and so forth for the term "coparameter". One defines of Eq. (3.2.3) JL from the separability constants or from the characteristic equations' constants of integration. ThenX. is defined by the generating equation (2.3.6). When first developing a perturbative generating function, like S , whether by separation of variables or from the characteristic equations, one usually will not define the coparameters until the complete integrals are written and the most convenient definitions are evident. The following development will, however, employ post facto rationale in defining -A- From equation separates in t and r. (3.2.3) it is evident that S Instead of proceeding from this factwe will first write the canonical equations for H , () and and there will result a _ reduced problem involving A(m, . From the canonical equations Ho H'= 0 (3.2.3) -21- t d it y-_rH0 O1 =- AUr) The coparameter A .is " =QO'- HO = - 01 j , (3.2.3) (3.2.5) segmented as 1\_(M) } A and Eqs. (3.2.4) (3.2.6) (3.2.4) and (3.2.5) may be integrated to yield A (, HA A (r) = - A At) '( R) .1 ii consian (3.2.7) COfnStdnt (3.2.8) (3.2.9) i(o If one writes the time-mass relation -t = 6oIm,-m) (3.2.9) may be recast as =1 A(RI (V) ) then Eq. (3.2.10) (3.2.11) -22- The coparameterj\. thus relates simply to 1r) for , 'S A and , Eqs. . Using the generating equations (3.2.7), to yield most of (3.2.8), and (3.2.11) S b ()= b can be used H t (3.2.12) 1 -//0 =(M) (3.2.13) - A -r -A- (V) 4.-rn -(tr) S ~ = {..A~Mt + .. Alai VLA )+ 4- r (i) RI V (3.2.15 into Eq. I (3.2.15) (3.2.3) produces ) Inserting Eq. (3.2.14) A(R) ?) S = . a reduced H-J equation for O4JL(M) +CmIJ However, f rom Eq. (3.2.15) bm - (r -- S I 0 ' I d A (3.2.16) and therefore Eq. S mAtR)j (3.2.16) (3.2.17) c(mS may be rewritten (3.2.18) ~~(M) -C rT~j +Mj\ -23- The integration of Eq. (3.2.18) S=~~ {#J.~r ~ '(I.v In The terms Ie and 6 I -A-W)l is straight forward. mL AtI A (3.2.19) are defined + ITl- ~ -AV~n~~(l fl[JL(i+ The entire generating function I~ (3.2.20) I--tV)I Th 2. 21) -~(3. S'' is thus ~icr +J +cr1A,)M 3.3 V) 'r ~'(LY 4- eC lJ~+ m~ The Ordinary Integrals In forming S only the canonical equations for the costate were used in conjunction with the H-J equation. Also, the Eqs. (3.2.7), (3.2.8), and (3.2.11) represent one-half of the desired zero field canonical transformation. The rest of the transformation results from the generating equation for However, applied to it proves useful to anticipate the _ trans- formation by integrating the canonical equations for the -24- state, forming a more common description of optimal, zero field flight in terms of the costate and the initial values of the state and costate. = = dt I SCm (3.3.5) (3.2.10) and (3.2.11) one may reform From Eqs. Eq. -r- A (1) (3.3.5) to a -"- - C -_ _ ~ (3.3.6) -cm .L It,, I kil Eqs. (3.2.20) and subscript (3.3.7) indefinite integral of ( (3.3.7), is called Using and W , where the (3.2.21) describe t ' in Eq. . WI&1i ) (M) = refers tot= t,. as expressed in Eq. _ _. S of Eq. s'"={(t in Eq. (3.1.8) (3.2.22) as + m) +A(', r 4 0- C ts M One may easily check that the original H-J Eq. (3.3.8) - V(m,) M()m) Lt(m) + _V(M)and the definition for-JX' one may rewrite The S of Eq. (V) V () (3. 9) (3.3.9) satisfies (3.2.3), so this final verification -25- is left for the interested reader. Finally, the equation for r is written, (3.3.10) d-t Ma rn The integral of Eq. (3.3.11) (3.3.11) is most easily evaluated by parts.t r -dm 3.3.11) -M fn +i t m=r',-o'r m F -- E2 + x) 3(3.3.12) (3.3.13) The next section will show that the canonical transformation for X (3.3.7) and (3.3.13). results in equations like Eqs. The difference is that will tA suggestion of William E. Miner's saving considerable effort. -26- comprise all of the lower limits of integration. The reason for first presenting the above results is that the prior knowledge enables one to spot unessential , terms of the transformation, viz. functions of which may be systematically eliminated. 3.4 The Canonical Transformation for Zero Gravity With Eq. (3.2.22) and 6- 5 (3.4.1) one may define the complete zero - canonical trans- formation. The cumbersome mechanics of applying Eq. is in Appendix A; (3.4.1) the results of Eq. and some additional alterations of Sect. 3.1. (3.4.1) were given in There are two details that one must describe to explain the results in Sect. 3.1, and this explanation will now be given in a qualitative form. Applying Eq. equation , S (as defined in (3.2.22) results in a transformation which, when compared to Eqs. r (3.4.1) to (3.3.7) and (3.3.13) for If seems to have many unnecessary, horrendous and terms. One is soon convinced that the unnecessary terms are functions of only_ . If one can find a scalar function of.A whose gradient with respect toA produces the -27- unwanted terms of the original transformation then one should subtract of I the scalar function from the first form From Eq. . (3.4.1), the new form of represent the zero - S will still canonical transformation, but without producing the unwanted terms. Secondly, there is a more general reason to S further alter the form of : the differential equations for homogeneous, scaling. Acriand Notice that Aig.are and that, because the optimal steering law only depends on the direction of Air')and not its one has that the trajectory initial conditions of r)and by a nonzero constant, versality condition, r(tlis uneffected length, if the are all multiplied _i k .f Assuming that the trans- (which is discussed in chapter 4), does not impose a natural scaling condition, one concludes that the costate is scalable in one's boundary value problem such that the state's invariant. that From Eqs. -Ais also scalable. effected by scaling' to render and (3.2.11) one wants it is seen Nevertheless,X may be One has the freedom to alterS invariant under scaling. desirable because tIf (3.2.8) trajectory remains is associated with k to be negative, This may be X , which is then a minus . sign should be written in front of Eq. (2.2.3), which amounts to minimizing H"" with respect to t-r -28- invariant under scaling. Also, keeping a variant of scaling gives one numerical problems as one attempts to use the freedom to one's benefit, and insodoing one may make the perturbations much too dependent on the example trajectory from which numbers are obtained. The transformations involving found in Sect. 3.1 are chosen to be invariant under scaling. In closing, it is mentioned that if one wishes to change to doubly starred variables such that -A (3.4.2) h and (3.4.3) then from the generating Eq. (2.3.6) one requires the generating function (3.4.4) where (3.4.5) The above result is useful in following Appendix A. 3.5 The Canonical Transformation for Constant Gravity From Eq. (2.4.11) and Eqs. (3.2.1) and (3.2.2) one may write the total Hamiltonian of the zero-field canonical variables H-A i- --_\ (3.5.1) -29- A H into its proper variables one would replace through Eq. qr) in its (3.2.11) and write proper form through Eqs. (2.1.2) and (3.1.7) . To put The perturbation for the constant field problem is obvious, where we employ the gravity expansion (2.5.3). =Ho1 HO (3.5.2) Hj./(21 ( Aui1VA4 L/A(MI (3.5.3) (R) The second H-J equation is therefore +M*67521) 'b -I- 'r (3.5.4) a 5 =S21z (3.5.5) The canonical equations of and -. S tell tAi are still constant. M , (A- HO- A (3.5.4) cons-t ant (3.5.6) COS-tCnt (3.5.7) const ant (3.5.8) / A(V) or Eq. / one that H0a J(MtA-KJ.()V ~ctj * where,. ) (3.5.9) -30- Using Eq. (3.5.9) in Eq. (3.5.4) one reduces ( 2) =jL Eq. (3.5.10) M + ) the H-J equation to (3.5.10) integrates to give IkeM W2 2 r# 2I (3.5.11) (3.5.9) and applying into Eq. Inserting the generating equations one obtains: 0 (3.5.12) Y-A I r M 0 + R)'*g1 (3.5.13) , Of course, just as-J a and Jk are segmented similarly, it is convenient to segmentX and E M similarly. a X In writing X {(3.5.14) as XL, and X,, , being the last component, we have assumed that the time transformation has occurred as described in Sect. 2.4. -31- The above results are easily obtained from the field free answer in a generally similar form without proceeding through so much formalism, but the formalism is necessary to retain canonical form. Hopefully, the formalism will return great benefits as we continue, no for longer are closed form solutions or approximations well covered in the literature. ferred to Sect. Again, the reader is re- 3.1 for the full transformation--an elementary change from the field free answer. -32- CHAPTER 4 The Numerical Example 4.1 The Chosen Problem It is desired to investigate the or zero g result of Sect. 3.1, constant and later transformations too, on the basis of an example optimum trajectory, solved on a computer. No longer is physical reasoning adequate to choose the next perturbations, possible exception of with the of Chapter 5, and therefore one must expand the latest Hamiltonians on the basis of numerical evidence. The chosen problem is similar to a Saturn V, second stage going into near-earth, circular orbit.12 The resulting trajectory will burn for over nine minutes, will subtend a central angle greater than twenty degrees, but it experiences a change in radius ratio of only 1:64. Although the problem is planar, the evidence is useful for near-planar problems. The following information describes the chosen problem in MKS units: mllmq T -33- C =4,,/0 sa PH YSICAL CONSTANTS + S=4.5v/0 33 c'=/ m, 7.5 10 k 6.44c/O6} r= 0m (4.1.1) B.5,v/0a ={2,6/O} S lra=6.5V/O'm |Lia|= (g r '(t.)a'* Tg = 0 rf FINAL 7.8 5 v/O Y"- BOUNDARY CONiDITIONS i 4.2 The Differential Equations The differential equation for the costate is actually found from the canonical equations for and yL and , (3.2.11). in conjunction with definitions i, or (3.2.7) This is coupled with the differential equation for the state, _ __ such that giving _/..(defines course, Yfln() is known, X(t.) and a unique trajectory. and it Of will be shown that A(mi) is not important for minimum time problems. -34- (3.5.1) From Eq. (4.2.1) where = (4.2.2) - (= 2 one uses Eq. , equation for cj (3.1.7) Writing the canonical : __ <~I> I LPr - r (4.2.3) . and For +M*/L, 'JLv Note that it makes no difference in Eq. _1 (4.2.4) (4.2.4) one if deals with the second set of perturbative variables first set, I From Eq. 3c3 LAAIT MATRIX (3.1.7) : (4.2.5) (4.2.3) : From Eq. (-- A o as shown. / - rather than the f (,R) rK]r Er,r (4.2.6) (iE7) 31~ (4.2.7) Similarly, d IL-r - 4H <'>-2 _ 1 (4.2.8) From Eq. (3.1.7) br -SM v (4.2.9) 0) . one notes from Eqs. (4.2.7) (3.2.7) ()4 )' (1) and and and (4.2.10) (3.2.11) or A to (4.2.2) that (4.2.11) . To relate Eqs. (4.2.12) tri To close the loop one merely writes the equations for it and r (4.2.13) and dre dt - (4.2.14) - (4.2.1 (R V) d -35- The following flow diagram illustrates the computational process, where the numbers appearing within the boxes refer to equations of this section. -36(4.Z-7) i1 (0 Diagram 4.1 4.3 The Transversality Conditions Given and excluding A1l0 If unique trajectory. *a one requires JA(O)to 0) and the four parameters Ai,(0), is open, define a then one effectively has five degrees of freedom in the integral of X X R.), (f.)) (4.3.1) Three degrees of freedom are necessary for the final boundary conditions in (4.1.1). The apparent surplus of two degrees of freedom is consumed by the transversality and the so-called scalability conditions. The analysis that produces the optimal control and the canonical equation, as given in Sect. 2.2, furnishes another necessary condition, condition., d r (t,) and 6 There is dft of the problem. al , a condition relating if also the transversality todf 2 , dma, they are allowed by the statement If the statement of the problem inter- relates some of the given differentials, then the -37- problem's requirements and the transversality condition must be satisfied simultaneously. Imagine a minimum time problem that, academic reasons, allows Or 0 sQ<M just as 1(T,) and zero. also contains H must maximize O' (C Cr'(t) , C' on an optimal trajectory. ,,, - >0 0 CM- 0' and , (4.3.2) U Then the control to vary between C'm for (4.3.3) The transversality condition is then \,,(td 20 H dt, ((+ (4.3.4) dt2 The differentials to d L, drf and , dt, coasting arcs, ( > dm, (4.3.4) H Am) d M. are not constrained and, because we have allowed and then argue from Eq. and A " are independent. One can that = -l(4.3.5) O(4.3.6) Ur =0 (4.3.7) -38- Eq. (4.3.7) must be satisfied with the requirements of the boundary value problem: Tdr =0 (4.3.8) J~~a=0 (4.3.9) r =0 Tqd~ -4-(JTd First, consider the "switching" condition (4.3.3) in the light of Eqs. then has that (4. 3.10) (2.2.5), and (4.3.6). ,\rr'O for I '' f . The switching , except possibly function is thus positive for t<tz for a finite number of points; Next, consider Eq. ergo no switching. (4.3.5). (4.3.11) 4C Because, as already mentioned, scalable, One _ and are _ (ie. can be multipled by a nonzero constant X without changing f(t)), one has that condition (4.3.11) can always be met and therefore is trivial and need not be obeyed.t Scalability is the first result needed to reconcile the surplus in degrees of freedom. One can, for example, arbitrarily define I t required if ---(T)1)1 = See Sect. h <0. ) (4.3.12 3.4 for the alteration that is rr -39- Finally, Eqs. (4.3.7), (4.3.8), and (4.3.9) (4.3.10) are nontrivially satisfied only if OT T' OTT rQT det. (4.3.13) LEq. is perpendicular to Because (4.3. 13) reduces to Fortunately, the \_ left side of Eq. )x 0 (4.3.14) is a (4.3.14) constant of the motion and can be incorporated at t, . -4+ That this is true can be discovered by formulating the problem in polar coordinates, and, after realizing that the costate to the central angle is identically zero, transforming the costates of the polar formulation to those of the Cartesian formulation. ~~ (4.3.15) .T. X k4- Schmieder's sweeping parameters were found to be a convenient pair of initial variables with conditions (4.3.12) and (4.3.15), for determining the correct initial conditions of the boundary value problem (4.1.1)./ Computer t2 was defined by the attainment of -40- _LT(ta), the desired and and the4r(ta) twenty meters r2 _l)conditions ~ta Without a machine were met by sweeping the parameters. iteration scheme, r was easily satisfied to within and the angle between and r,. was only off by a few minutes of arc. _e- Schmieder's results are given in Appendix B in a form commensurate with the present problem. Then, in a separate part of the machine program, and many important functions ofJL were calculated from the algebraic results of Sect. The Numerical Output Fig. in Many of the and will now be presented. results have been graphed, 4.4 3.1. r(t)and 1 illustrates transfers requiring a is characteristic of rLta The hump _r(tI. huge change in velocity but relatively little change in radius. Fig. 2 illustrates (t from direction one, and i3 direction two. Transfers , C.C.W. angle of C .W.angle ofT from to circular orbit that experience a hump in r(+) apparently have -> (1 r - is most perpendicular to r t , and, r of course, 0 . r- 0 This so that as is because is almost adequate to keep it so, as in a circular orbit, while up"Mm -41- the basic job for the thrust is to raise the magnitude of in the last few seconds to the required value; ie, '''-0 r and . Thus, as t-' and approaches 900 T athe angle between t-*O. If there were a large radius to be gained then the end condition would occur at the t 0. to r at Fig. t 1T T/= 0. () first stationary point of , 2 shows that T , where is almost perpendicular and in a perfect run one would expect Such facts, however, are only of passing interest because one hopes to approximate a large bundle of trajectories. Fig. 3 is of great interest here, for it shows that the true by the ,C _ i (OI)and /I) trajectory is well approximated predicted in the linear field model, where A(Olwere matched. the more striking if one The close fit is all fits the model curves to the true curves to obtain minimum integral-square difference. The fit of Fig. 3 has a maximum error at in terms of pUi) , t which, represents four degrees of arc. A minimum integral-square error fit will have a maximum error of less than one degree of arc. Therefore, an H-J solution that merely approximates the linear field model might prove to be satisfactory for a wide class of trajectories; if used with sampleddata procedures, the ,." -'/mI -42- overall accuracy of such an approximation would improve. Indeed, even the flat earth solutions of Sect. 3.1, similar forms, or have been actually applied in sampled 'I A~f data operations.' The reader is referred to Chapter 5 for a derivation of the linear field A. Because of the form of Eqs. p, (3.1.7) it was decided to graph 9a in Fig. 4, and and (3.1.6) and andR.-mV + a),= A 2 in Fig. (R - )=A. 5.t In Chapter 5 it will be shown that these terms of are uniquely important and interesting in their significance for the linear field problem, and, used in defining in fact, are to be for the third perturbative solution. Fig. 6 illustrates the following terms: /f/j....( (4.4.1) d,4 I -. X = 00 Fig. 7 illustrates: 05 = b, 2 6,= ~Ofci , 6?= R,i tHere, , az (4.4.2) is that found from Eq. (3.1.7) minus -43- Figs. perturbation. 4 and 5 are useful for the third Figs. 6 and 7 are useful for the the linear term of 9 . ensuing perturbations to incorporate the effect of -44- CHAPTER 5 The Linear Field Problem 5.1 The Definition of an Accurate Solution The present family of perturbations are designed to incorporate the effect of the linear term of the gravity expansion. If higher order terms of are never to be included, then an accurate closed form approximation of the linear field minimum time solution may be defined as one whose difference from the linear field solution is in some sense of the same order of magnitude as the difference between the linear field solution and the inverse-square field minimum time result. For example, the trajectory of the linear field model using the same initial ( and A , Chapter 4 may be compared with a computer solution for which in turn may be compared with one's literal solution using the same initial X and A . The quantities of comparison might be the components of r at every ti If one planned, for example, to later incorporate the quadratic terms of 4 in the Hamiltonian, then one's tz -45- closed form approximation of the linear field solution must be more accurate: perhaps as accurate as the quadratic field's minimum time solution appears when compared with the inverse-square solution. Now, with a rough goal in mind, the linear field perturbations may be started. 5.2 The Third Perturbation and the Result From Eqs. (2.4.11), (3.5.1) and (2.5.3) one has that the total Hamiltonian of the constant field variables may be written H One puts Eq. i> -A,,,(Mo QIf/( (92) (5.2.1) in terms of its proper variables with (3.1.5) for incidentally, Eq. _jr) , Eq. (3.1.7) for A (2.5.5) for . , and, The total Hamiltonian for the linear field problem of the constant field : variables is called <2> = -A It is J 2 - a ) / W,N > that is of present concern, to approximate the dynamics expressed by region of accuracy, as outlined in Sect. otherwise stated, (5.2.2) and it is desired :4 -(>to 5.1. some Unless the "problem" refers in this chapter __"WNW -iloo, -46- to the linear field problem embodied by From the problem's canonical equations, .... is clear that the rates of change of R of Eq. appear in the term 2.t - _ bAH e jt-.AIR= dt Using Eqs. -L (5.2.3) (3.1.5), and V as they (5.2.2). = 9%,) N FV<a>Zt (V) and d 2 a with respect to time are caused by it at H=R) (5.2.3) (5.2.4) and (5.2.4): =L2 \ j49 (5.2.5) It is convenient to define r 0 (5.2.6) so that 0 0 (5.2.7) * ' N =0* .0O0 1 The three scalar differential equations of Eq. are thus uncoupled, I, b ,cde,) (5.2.5) hence introducing the constants one has that -47- 2/t} + + b sinb a sin-vt + d coshb4'2)t 4 sinVJt {eCosvt Ccos Vt = From Eq. (5.2.3) (5.2.8) r>\ and (5.2.8): Cisinb4vt+bcosh4 yt) -M'*(cos2Vt*+b sinI429t) a_ (.~ -,s'rnvt+fcos vt) (5.2.9) -M*le cosyt -4r sinyt) from Eq. Therefore, (3.1.5) one has that r( sinh4At +bcosh4- 2't) sin Yt - d cos Z/ -fcos _ Clearly, if ~ - 'S {$'} i/f(5.2.10) were approximated by 2 + , ~c (5.2.11) then one would still obtain the linear fields' the optimal guidance law, form for although the literal equation -48- for the state would be different. Fig. 4 shows that for the chosen problem of Chapter 4 the difference A) - is of similar shape to 9, with generally of greater absolute value than is not particularly close to the difference ( ?a -Aa is, ?, . to first derivatives,nz.i, of Eq. (5.2.11) ; so A, approximation, ?a a AR'S ; so are somewhat approximate to ,'s All things considered, derivatives. ?, Fig. 5 shows that constant, albeit large compared to nth a magnitude n+h the approximation seems to be a reasonable beginning. From Eq. (5.2.1) or Eq. (5.2.2) and Eq. (5.2.11) one has that the third perturbative Hamiltonian is defined 2M ' 4 0-M 1 (5.2.12) It will be shown that the canonical transformation to the constants of the third pertubative problem, N, embodied by may be defined as follows. .3 as Segment 2 and as JL with the superscript and "3". were segmented, One has for J\.: 33 A = A,,cosjt (R/ Li~ o(5. 2.13) +Y-A-1voisn/ A.(.) cos/t - T s A hit but -49- +( ircoshri vt + v d'M*sinhrFyt).A.vI.] 4- ( ca coSvt - crVM ts in E(:Isinyt - 6M'cos * 3 'tJA. t - r'VM*s invt) 4-( dcs 3 (5.2.14) 3 -A,1 COsh ljt -4 -- AI ISinbri 'vt .3 ie -L cos it A - 3 Cos vt -CV 2Nr VJ- L 3 in Yt se) 3 VrA-A(V)s (5.2.15) in ;'t N(Nd7' 2 (5.2.16) +- a) One then has = v form: ashI)t cr'c 0 0 0 C'l 0 -VTAsinh-Rt A 0 j L{ + 0 0 R sinvt 0 J 0 0 -dI 3 a'-'Cos 0 + 9) - 6 ~sinj (5.2.17) i -W. IIIs MINII ..au--.--. -50- 2 c ++M*cos;hfri't) -vrsinh-r?/t 0 ti v('ir~7/t+MCOs2 0 I 0 coshzvt+4r avm*s inhbt) 0 (cos/t-vM*sinvt) cos,/i.M~sihVt) 0 L 0 0 +0 (5.2.18) (5.2.19) M=x n 5.3 The Derivation of the Third Transformation The H-J equation of the third perturbation is (0 o0 With the results of time in (5.3.1) ,),,LH.(9 37A. t (5.2.8), (5.2.9) and the separation one has almost solved Eq. S (5.3.1). First, define (3) HO 3 =-/AMl,, (5.3.2) cornstdn+ 3 Next, use to obtain -. (a,b,c,d,e,f) IR) = c J\-t. 3 and.-A-Lv (5.3.3) -51- (5.3.4) One then has the results (5.2.13) and (5.2.14). The 2 generating equations for -. A. _ (3)= t+[ V Of course, one substitutes . 3 and integrated to define most of as ( always, are then applied, (3 3 5.3.5) (M-1V4)for t in Eqs. (5.2.13) and (5.2.14) in order to get the forms indicated in Eq. (5.3.5). The entire process already taken in forming H-J equation S for back into Eq. is quite similar to that . S One may write by using the results obtained (5.3.1). PR- oM*V+ Z- -go) and Eq. (5.3.6) a- (5.2.15) (5..3.5) .. for (5.2.12) for - (5.3.6) use is made of Eq. ) In writing Eq. Eq. a reduced . From 2T ~M~it, - 2 6_A (5.3.7) -52- However, the operator - ergo Eqs. to rewrite Eq. is, (5.2.3) and (5.3.7), equivalent (5.2.4) allow one to (5.3.7) as A N(-(M) (5.3.8) b 5 'S=(S Using Eq. in Eq. (5.3.8) in (5.3.6) one has ' (5.3.9) 3(M1 Using Eq. for riti one may rewrite (5.3.9) s!~ -"-(MI Integrating Eq. (5.3.10) for may then use Eq. 4- S (5.3.5) to form S'"= /AM~ M' +V2M (5.3.10) is elementary, and one ' Eq. (5.2.6) + [LRJR 2 ES2 3 3 3 Sihh~rZJt)M* Z-'os-Vt-AII 5 inhrJ2'it+ -A4tRj C(5..11 414~~ftj The square brackets around ... /(~and -. J1L~ indicate that 3 they are to be written as functions of -A-,k and -53- Applying the Eq. generating equations to (5.3.11) , ----s S' i'= (5.3.12) 2 and solving Eq. (5.2.18) and (5.3.12) for (5.2.19)t results in (5.2.17), When these answers are combined with the constant field transformation of Sect. 3.1 one obtains an interesting result. 3 .3 V/cosh "dt +F V R, S inA-P cos COSI - P'2 Sin I/t + (5.3. 13) 33 R cosht + j-V, sinY+Tt+ Rcost j-gvsiVt = where (/ CO~,.,~J.~ 3 2 (5.3.14) )f (5.3.15) t This step is straight forward, and is left for the interested reader. -54- and -' = -4-Co'jj(-~ - From Eqs. and due to _T ~R) (3.1.6) and are the RI _ .A.ftV ,0(5. 3.16) (3.1.7) one has that particular solutions , in the constant problem, but with _A_ 3 put in terms of.., blem), (a constant of the linear field pro- and time through the linear field guidance law (5.2.13), (5.2.14) and (5.2.15). It will be shown in the next section that all the remaining perturbations of the linear field problem leave -A- might expect, are invariant forms. so that Furthermore, _V and constant, as one because the guidance law does not depend explicitly on position and because gravity is linear, one's differential equation for 4 L takes the linear form in the linear field: + + =' (5.3.18) a 4- 4 Ttf(5.3.17) + Eq. (5.3.18) means that the solution of Eq. a superposition of the homogeneous particular solutions due to 9, , (5.3.17) is solution and the two and T . Now, referring -7--7 AT -55- again to Eqs. (5.3.18) and that the additions to ) (5.3.19), one discovers are in and that the additions to precisejy i+h transformation, '..4 for . In If the assumption ,of the linear field family: 3 SV delay, 4h)) one then expects the following form for a 5.4 -+4! 9')but only expressed approximately for conclusion, an ( are in facth).t tMexpressed One therefore has and fact 4- (5.3.19) tal i (5.3.19) is unclear one should not for the next section will discuss it further. The Ensuing Perturbations as One Perturbation The results of the third perturbation are now analysed, and a general form for all the ensuing perturbations of the linear problem is developed. is shown that Eqs. It (5.3.19) apply, and that the net gration with respect to M* or Using Eq. the results of Sect. (5.2.2) t . result of all the perturbations is essentially one inte- for 5.3 one has that , rule (2.4.11) and -.. Mom ON- mr-I -56- <3>) -V 'J(4JM =-A __NN (5.4.1) It is assumed that the ensuing perturbations in time orM on expansions of the part of for the(*+3) rd series. are based Call 4 . H perturbational Hamiltonian. -4 -4h (\ We have assumed that A n(5.4.2) , (ie. -- '2 -(R) v ), -L-(V) R does not depend on never changes form. Because 3 3 3 A JLRand one has that or V are still constant regardless of the definition of -AMw- H[ 4 [ =t S'H+ (' Clearly, the integration Am(N5.4.3) A'" (5.4.4) constants .4 4 -- 6- (Rand .wA (V) 3 of the H-J equation 3 --7LA V) (5.4.4) may be defined as 3 .A...c) and From the generating rule for A 3 -r 3 4) (5.4.5) and the definitions4 (5.4.6) (R I(R\ 3 4 -57- one has that t4) 3 A4 A(+) S 4 v 4~ (5.4.7) (5.4.4) may be reduced to 4 i MA~ ~(4) LiM ) +1S Therefore, the H-J equation S (4)(4 -(5.4.8) A4 J-cMI/I Mz~\1I~A VJLijVo0V (5.4.9) 'I'd M*i The square brackets in Eq. (5.4.9) mean that the quantity is to be written as a function of A form of (5.4.9) depends on (5.2.7) for So, finding S 4 M and . The second NV amounts to integrating Eq. (5.4.9) (4) for over S , 's and because one has complete control, ideally, expansion then one assumes that a significant can be defined such that Eq. (5.4.9) is integrable. Applying the generating equations for to one has R) 4 (4) (5.4.1 4 'r V A (41(5.4.11) ~~ S ( )) ) that -58- Definition (5.4.10) and (5.4.6) and results (5.4.11) thus verify for f=,4 the assumptions of (5.3.19) for the general form of the perturbation. It is clear that the total Hamiltonian of the linear problem for the 4th variables is, from Eqs. HSo, (5.4.1) and (5.4.2), -- yi' J-(M) jA 4 (5.4.12) the pattern may be repeated for all the perturbations, for again and again R and V will be absent from H One may write the general problem as: 3 MR A - I (5.4.13) (RI ~A(9); 3 -A tv) A i i-1 R=P M M +-( ) '/J ]iVT )-S) /tw " S (5.4.14) (V ^M ~d M*(5.4.15) (5.4.16) (5.4.17) However, thinking of many perturbations i.4 is not the clearest point of view. for LJ and , One has the results (5.3.13) and (5.3.14), and the only -59- 3 3 i change at perturbation (5.3.19). prescribed in 4 3 = A R -4- ( R is in V , as (5.4.16) : From Eq. S b 6A(RI and (IT 4 R +( S (5.4.18) = R -( .3 .'. +(4 R S ") n T (5.4.19) - If Z5 Similarly 3- Using Eq. v--rn (5.4.15) and Eq. (5.4.20) (5.4.2), where we forget due to the gradients of Eqs. the terms and C(; ) -r (5.4.19) (5.4.20), one obtains Z~~= 2J T V I~ E 3~Ym -o (5.4.21) Therefore, rather than thinking of many P 'S and many one can ask for an entire, truncated expansion of ~4% 4where +4" (5.4.22) (5.4.23) -60- such that Cin Eq. (5.4.21) is said to result in a closed form. 3 In summary, (5.4.24) V(5.4.25) expanding to obtain a truncated series such that is obtainable in closed form via Eq. (5.4.21) produces the linear field approximate solution. The solution is had explicitly by putting into Eqs. resultsin Eqs. (5.4.24) (5.3.13) One no form or partial a single, and (5.4.25), and then these (5.3.14). longer cares explicitly for canonical differential equations. clear goal: This is not and define There remains and integrate Eq. something that may be done uniquely nor, any case, simply. Nevertheless, perturbational process (5.4.21). in one has simplified the to a well defined point that requires one more, huge effort for the completion of the problem. Of course, if the term later point then formalism, _( is to be included at a one would later revert to the H-J and one may then expect a far worse experience than he has had before: the uncoupled linearities typifying the transformations of the linear problem would be but fond memories. -61- A Brief Analysis of 5.5 Before closing the chapter, a brief analysis is made on the basis of Figures 4 through 7. of From Sects. 3.1 and 5.2: 2 I-I(5.5.1) -L4 For the purpose of expanding through Eqs. A , it is written as (3.1.8), (3.1.10) and A (3.1.11) for -V7rand 3 and for.A..(,,.A.) and (5.2.15) a * A, and Eqs. (5.2.13), (5.2.14) Notice that Aj Figure 4 shows that determining (5.5.2) is quite significant in and it is not easily approximated by a low order polynomial in time. is significant in determining approximated by a constant or, Figure 5 shows that P , but it may be better, a low order polynomial in time. Figures 6 and 7 are basically concerned with ,A A andCy. 'I S = c+, + a += (4(5.5.3) ,+ + (5.5.4) ----- ~---- U ~-.- -~ -~- - - -62- d; 6 illustrates that none of the Fig. t insignificant in the determination of The terms 0, and as terms are and -6 might be approximated by constants by taking their first terms in their Taylor's series about t"0. Also: Q4 Fig. = -+#- + 62(5.5.5) o + b +b 4 7 shows that b3z -b4, approximates (5.5.6) so that the constant that Cy may be called unity. first half of the flight b,% For about the -b2 , but overall C4 is not justifiably approximated by unity. In summary, expanded in powers of t and f+nhi27/# ,IA-/\4 similarly for appear in Eq. if /h IY 0 are, and and for example, * 0 n/t _ , and as they (5.5.1), then one might have a quickly may lead to a closed convergent scheme, such that form for The expansions, to reiterate, the basis of of 2 i V ._./I ,i(R) and , _Al as functions written and time, as if take place on .... and were truely constant in the inverse squard field problem. - I -~ -63- The actual definition of requires much more thought and experimentation, but the problem reduction of Sect. 5.4 at last brings one to a well defined perttrbational problem. -64- CHAPTER 6 Summary and Conclusions 6.1 A Summary of the Results A summary of best given in terms of the results of this thesis is approximating minimum time transfers in a linear gravity field. The approximating perturbations may be classed into four steps: constant gravity; "homogeneous" "approximate" linear gravity. gravity zero gravity; linear gravity; and By "homogeneous" linear is meant a gravity field that varies with time during the flight as if there were no thrust and merely a linear gravity series in postion. after the first two, linear gravity has been solved. This third problem, By "approximate" is meant a gravity field that varies with time during the flight in the manner of one's best and final approximation of the linear field problem. This fourth problem has been shown to consist of an expansion of a scalar function of time such that one defines an integrable part with an ignorable remainder. Note that although the third problem's approximation of the gravity time series does not reflect -65- the effect of thrust, the resulting solution for the state does approximate the effect of the thrust and It is the resulting solution for the costate is precise. interesting that the half of the third problem's solution for the position that remains invariant upon the fourth perturbation is that part of the Hamiltonian whose effect The remains to be included in the fourth perturbation. vector function in question, , is the third problem's approximation of the position's particular solution due to the optimal linear field thrust program. 6.2 Concluding Remarks In performing the fourth perturbational solution, it will be useful to see if any parts of can be included without a series expansion. Also, a second example trajectory should be solved on a computer in order to better segregate numerical properties that are typical of a rather large class of high thrust transfers. One might, for example, require that a slight plane change and a relatively large change in radius occur. The troubles that would be encountered in incorporating the higher order gravity terms are not clear in the author's mind, but certainly they would be -66- significantly greater than those experienced so far. Transfers that require many terms in the gravity series for a model, ie medium or low thrust transfers, might better be handled with a zero thrust, inverse-square field base soltuion rather than a full thrust, field or constant field base solution. zero The former base solution has been accomplished by MinerP Actually, a Taylor's series expansion of gravity will not work for most low thrust transfers, mechanics' and in this case the celestial literature on the von Zeipel method and so forth will prove quite useful. Finally, if one achieves an adequate literal solution then there remains an implicit function difficulty in doing guidance. That is, one desires an explicit guidance law like A (6.2.1) but one actually has, for example, =\R)-V~) 1 \/.A A(6.2.3) ___ Lt _ _6-.2.4) (6.2.2) -67- S1 1(6.2.5) For illustration the form of E is chosen to relect stationary differential constraints and stationary boundary conditions, hence (_ and 0 . The "constants" are implicitly defined at t )Y() L_(t) _-E= , by M.t) (given a defined final boundary value- minimization problem), and thus >(V)is implicitly defined by () and %_(t). One would probably determine the constants numerically at discrete moments, a method commonly called sampled data feedback guidance. Of course, one requires some feedback control because one's solution is imperfect and one's knowledge of the state and the control is uncertain. At any rate, if the numerical difficulties in obtaining the constants are too great then, from a practical standpoint, this spells the ultimate doom of one's closed form solution. This possibility must be objectively tested along with questions of accuracy that ignore the guidance problem. For example, expanding ~_ is necessary for the fourth perturbation, and it may again be required to apply the tIn between sample points, Eq. (6.2.4) represents an explicit guidance law, albeit open loop. - -M64 - _M _ -68- solutions in a guidance scheme. It must be remembered that this thesis merely investigates one of many approaches to obtaining a literal approximation of the minimum time, high thrust transfer, and the state-of- the-art of doing guidance is so young that no one method, existant or anticipated, has a clear superiority. -69- The Constant Field Parameter APPENDIX A It is desired to generate by Eqs. (3.4.1) and (3.3.9). as described , The superscripts "i " are First, we generate V dropped for convenience. T V- c+T - 6LAW (RI~- +I T VV) M~ m-A ._ Cofstari vector from Eqs. One expects that the excess constant, knows that ie a function of V and (3.3.8) o - (A-2) (3.3.7) (A-i) terms of Eq. and -AM (A-i) are a -A(V), while one is a constant. fC& a If one can find . )c -'-A (MJ' ) (A-3) and a scalar function S whose gradient with respect to JLY)produces A can augment S with (-S) AI(f, A-v then one and rid the transformation of L . However, -Y) M.M -70- From Eq. V(M) V) -AL V , (3.2.20) From Eqs. I 2f (3.3.7) )f JA and -AR (A-4) (3.2.21) -b 4 ( A -x) -- _-t Tvt ALVI kA-6) (T4 3 If one uses the above results in the left side of Eq. (A-3) and writes the result in terms of the orthogonal vectors -.M and (A( 1 then one obtains LA<adVf 4l~d'rV1 A,,)'(CR4Y AN trivially. efficients of -A(R) and proof and not a proof using A A To prove that the co- 1-tmt A(vi (which, because one desires 4 Tvlr (A-7) )v AR) Of course, the question mark refers to whether really contains y J ', LAfY /tjcontain , m trivially, means an algebraic ), one can probably -71- find simplifying relations among t , I However, the author used "brute force" and 0 algebra to prove that A6 =fci+ - V (A-8) 4+ v LI~m LA-1 - ( ZA (A-9) Before attempting to find to generate R 6 it is desirable , which also suffers from unnecessary terms. = = - S - R + m +(g i -Again, -C \1 {'L may replace r in ~ '(M) S ~ (A-10) because only the gradients of this function are important. From Eq. -Ct (3.3.13) 1JLU) (A-ll) m -72- or- + (r 3 0 As with V, lm =Consfdni (v)tai - ve.ctor one attempts to write the R generating function relation to correspond with the integrals", (A-12). "ordinary However, in the case of b R one must _ (Y) to make the com- first perform the gradient of parison. (A-12) T__ ) -A k.=C - -A (A-13) 1-_1R one may write -(R m+J AriiL) m - With the above results, in the form (A-14) the relation (A-10) may be put (A-12). C IAT -- -A V), _:A - (A - 1 5 ) Also, ----- -I1 -73- Therefore, one expects that I -4- ' (vm To prove that for A AI (iA) Jivd4 IAVIIIJp~JstiV i(A-16) -I 0 ) ) one proceeds just as one did The result is . -Ac x AA =L I-At P) P ItJill-VA +j A-(R) -A The question of augmenting S by V (A-17) Fe) , where augmenting S by (- O . If one can find 0 and & IT . , arises along with the question of such that they are equal then this is the ideal situation. The author could not quite obtainI9 = &, but the following and A . development explains what was accomplished to rid Note that I.At~)IJ_ --T -'/' A -74- Also A 3 T + T = (A-19) = (~i71 However J -j- n -A -IAjJ 'Y UMAv + Lkrill.A(v) (A-20) y-i.A(V)1(I A (t)R 1) ( (A-21) .' = &fc Y Am-lI Ih ( AI)) Define an augmented S (A-22) by S =S-G As a result, (A-23) the term transformation. 3 b- 0= __ disappears from the But what of _A ? Note that CrcIfC[ 4- IJ) II J II' ] j (A-24) -75- So, 9 is ' plus a remainder that in the V leaves the second term of Eq. From Eqs. 3 (A-l), transformation (A-24) 's final form. (A-3) and (A-24) one has that gives the following V transformation: V ={-N r -0iYn) .A And S gives the followingR + transformation: roJ-L T. ILA,.I I.,IJ From Eq. (3.3.7) term of Eq. h t for V' (A-26) one has that the additional (A-25) may be combined with L would become /P , +I[4vj4aI .ALoltlecnIJ If one imagines scaling all of the terms then one has from Eqs. *LI (A-25) and that many terms of the transformation, and (A-26) like V and R, are not invarient under scaling, although _r are invariant. where and If one could form slightly different arguements of the logs, where the arguements are in- variant under scaling, then V and R would be invariant too. Namely, ' and 3 of Sect. 3.1 are so normalized. It is stated without further motivation that if one desires to change to doubly starred variables such that -76- R -AV) R (A-27) -AR (A-2 8) V) =R +(w ,xe,(A-2 OLroA 4-LLC -AIN L401fAcRu _ ) 9) (A-30) J-kALVJI then one requires the generating function S = () R + -t V (A-31) Using So"on the transformation of S of Sect. 3. 1. 3S gives the results was defined after some experimentation. -72- Schmieder's Sweeping Parameters APPENDIX B -MAay b(ts) s= DIAGRAM F-\ It will be shown that o( and Q( ME ~y be used as the two degrees of freedom necessary to solve the problem of Chapter 4. Define = J (B-1) and notice from Diag. B-1 that fixed att,. , =d (B-2) (B-3) _3 = fd n- As already mentioned in Chapter 4, at and Eqs. (B_4) and (B-4) A V0 one must define to define a trajectory. From (B-I): sI COs = s(t +N (1- = co5( , + o) (B-5) (B-6) --- -78- Hence, CE, condition defines _(Jtff,j. Using the transversality (4.3.13) one can show that or The result - =0 (t (, >-(nC To define (B-7) rt, ,rlti) (B-9) depends on Therefore, writing Eq. (B-8) (B-9) r (t) =0, i\r.)(one needs 2 'irh+L also determines 0 Sri . J(til Sd 0( b( (t,). From Eq. O9fSC) (B-10) at txr (B-4) (B-10) t, one has that (B-11) But (B-12) and from Newton's Laws = u( 0-S In C( -- 3COW (B-13) - -I -79- Using the above three equations one has ((B-14) or \ (- (B-15) 0.E.D4 -80- (/0om) y(/3 rn 6.5/ 8.0 Ii 7. S- '/ . r(t.. 6.YSO 7.0 , rEt 2 a) 6.47 6.46 6.0 78/ (t.2)A / Lr / I 0 / 6.41 3,~0 A I I 6.43 rGt 6, -.10 I kmz.62 miles A I 4/ I a4- a- 6.4! I4.. 3.0 J(i )42.74 ' 6.401 2.5 f, .z SJ5 I - 60 /20 /80 300 240 -RO .360 FIG. / o EXAMPLE r(t) M.K.S. UNITS -1 180 540 600 -L (sec.) 7-e (de.) 3, (def) i 24 60 22 I 'I I so / it / I 0 A it /190 /a 40 NA I 30 / /6 / N I / S /4 I / XL / S /0 ,J\. *1 /0 S I 0 I, I I V Co a 6 11 * I -&0 oll J- 9 4- tLz 3w55 'p. 0 0 60 /20 /80 2-f0 300 -360 420 FIG. 2 E XA MPLE A2T 6(t)~ r 2 e %.. 480 5-0 600 (sec .) COMM trou it Val .r) (t=0) tr ArI (t,) 1trae A f- I II- .7 -Attr u e S .5 ~r~traAe 0 60 /20 180 2-0 JOO 360 -20 FIG. 3 TRUE A AND FITTED LINEAR FIELD A, 480 50 600 ($c. 83- (/o 6 ) J,; A, i .3 .2 .1 ..-. J --- ~- -w 60 -.1 120 /80 2A0 300 ,36&<'2O 480 510 .- ~Z... A M=r -4 FIG. 4 I&A, - .3 \ I 600 (s ec.) A& j ( /0 y 4.0 3.5 - 3.0 A |1 A I.- I1 .100" 'A Aa 7000 .000Aa .7 .A0 1.5, A-o /.0 A-01 0.5 I0 t2 ---- 55'5-9 60 120 /80 2-40 300 J60 420 A~ FIG. 5 y &A 2 -80 540 600 1 sec. (norm dliZE-D) /1 C, /6 /.4 / 2*1 / N N // N./.-*/ -.-- 018 0.6 ~-. --- ------ . .- 0.-1 0. 2 t z 0.0 60 /20 /80 240 300 360 420 -0.2 - 0.4 -0.6 FIG.6 (a, ,a), a3 , a4 ,a,) 490 540 655 600 W- c 5 e C.) A (normaliZED) 3.0 7 / b, '.5- / 6N N 0.0 N - N --........... N. I NI *1 za' 0.0 60 /20 /80 240 300 360 -0.5 - 120 -180 Jr0 I1 /.0 --- -..- -2.0 - --.-...,..........B 4 Io' .000 -3.0 FIG. 7 ) (b,,b b3 b 3 600 (sec) -87- References 1) Bellman, Richard E. and Dreyfus, Stuart E., Applied Dynamic Programming, Princeton University Press, 1962, Chapters V, VI, XII. 2) Bliss, Gilbert Ames, Lectures on the Calculus of Variations, University of Chicago Press, Phoenix Science Series, 3) 1961, Chapters III, VII, VIII,IX. Bliss, Gilbert Ames, The Problem of Mayer With Variable End Points, Transactions of the American Mathematical Society, XIX, 1918, pp 305-314. 4) Boyce, M. G., An Application of Calculus of Variations to the Optimization of Multistage Trajectories, Progress Report No. 3 on Studies in the Fields of Space Flight and Guidance Theory, NASA MTP-AERO-63-12, 1963. 5) Brouwer, Dirk and Clemence, Gerald M., Methods of Celestial Mechanics, Academic Press, New York, 1961, Chapters XI, XII, XVII. 6) Bryson, Arthur E. and Ho, Yu-Chi, Optimal Programming, Estimation and Control, Harvard University, (unpublished 1966) 7) Goldstein, Herbert, Wesley Press, Inc., 8) Miner, William E., Low Thrust Optimization through Hamilton-Jacobi Theory, NASA-Electronics Research Center, (unpublished). 9) Rosenberg, R. M., An Optimization Problem in Dynamics, 4th U. S. National Congress of Applied Mechanics, Vol. 1, 10) Classical Mechanics, Addison1950, Chapters 2,7,8,9. 1962. Ross, S., Optimal Ascent into Orbit-A New Look at an Old Problem, NASA, Washington, D.C., (XV International Astronautical Congress, Warsaw, 1964). - - .1910111 1. 1 111 .... .. .------ 1 -88- NASA, 11) Schmieder, David, Sweeping Parameters, Huntsville, (internal note). 12) Polovitch, R. S. and Morgan, W. B., Path Adaptive Guidance for Saturn V Three Dimensional Ascent to Orbit, NASA, Huntsville, MTP-Aero-63-70, 1963. 13) Baker, Clyde D., Saturn Guidance Concepts, NASA research Achievements Review Series #15-16, NASA TMX-53373, 1965. 14) Perkins, F. M., Explicit Tangent-Steering Guidance Equations for Multi Stage Boosters, Aerospace Corporation, #TDR-469(5540-lo) -3, 1965.