See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/303540681

Tensor Flight Dynamics

Article · January 2012

CITATIONS READS
0 498

1 author:

Peter Zipfel
Modeling and Simulation Technologies
29 PUBLICATIONS 119 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Introduction to Tensor Flight Dynamics View project

Tensors of Rank Two in Tensor Flight Dynamics View project

All content following this page was uploaded by Peter Zipfel on 26 May 2016.

The user has requested enhancement of the downloaded file.

Tensor Flight Dynamics

Peter H. Zipfel*

University of Florida, Research and Engineering Education Facility, Shalimar, FL 32579

Abstract
Tensor flight dynamics models flight dynamics with Cartesian tensors that are invariant under all
coordinate transformations, even time dependent transformations. It elevates Newton’s Second
Law to a law that is not only invariant under inertial coordinate transformations, but under all
transformations, thus bringing Newtonian mechanics under the umbrella of Einstein’s General
Relativity. Its roots go back to the late 1960’s and a paper that was presented at the Second
Atmospheric Flight Mechanics Conference in 1972. The Atmospheric Flight Mechanics Committee
has selected that paper as most influential for flight dynamics of that period. This paper reviews the
history and introduces the kinematic and dynamic concepts. The Special and General Theories of
Relativity are briefly highlighted with two of Einstein’s favorite examples, and their consequences
for classical mechanics are discussed. Recent applications of tensor flight dynamics are
summarized.

1. Introduction
I am honored that the Atmospheric Flight Mechanics Committee selected my 1972 paper1 as most
influential to flight dynamics among those published in the seventies. I still have fond memories of that
Second Atmospheric Flight Mechanics Conference, held in September at the Cabana Hyatt House in Palo
Alto, California. With a fresh Ph.D. diploma in my pocket I presented a key topic from my dissertation2,
and entitled it Perturbation Methods in Atmospheric Flight Mechanics. Of course as author I thought it
was important, but the conference planners relegated it to a back-up slot. Nonetheless, I was able to
present it and was glad that a few stayed to listen. Shortly thereafter I submitted the paper to the AIAA
Journal. The reviewers where rather generous and published it in September of 19733.

My paper focused on linearization, while my dissertation introduced an entirely new framework,

namely, Tensor Flight Dynamics (TFD). When I taught this approach to graduate students at the
University of Florida in 1978, I got a mixed reception. Why do you complicate an already difficult subject
was the complaint from some students. That was in the seventies when Etkin4 ruled flight mechanics and
the term Modeling and Simulation (M&S) was still unknown. What has changed? Students haven‟t
changed, but modeling and simulation of aerospace systems has become more complex. We deal now
with so-called net-centric simulations; i.e., with many interacting objects in the sky, on Earth, and in
space. Flight vehicles navigate by overhead satellites, synchronize their flight paths with other vehicles,
swarm over hostile territory, and attack multiple targets. Each of these objects may carry several
coordinate systems, e.g., inertial, Earth-fixed, body, sensors, etc. They will obliterate the modeling

*
Adjunct Associate Professor, AIAA Associate Fellow, 73 Country Club Rd., Shalimar, FL 32579

1
process quickly if introduced from the start. But coordinate systems have nothing to do with the physics
of the problem. The laws of physics are independent of coordinate systems. A more systematic approach
models the physics first, and then introduces coordinate systems for computation. Tensors model physics,
matrices hold numbers. This has become the motto of TFD: From tensor modeling to matrix coding.

2. Kinematics
Let‟s begin with some simple kinematic concepts. A single point in space, say B, is meaningless.
Where is it located? I have to introduce a second point A. Only then can I relate B to A by the
displacement vector s BA . Its geometric picture is an arrow from A to B with the head at B. The
displacement vector is a relative concept and formulated independent of coordinate systems. An a priori

location in space does not exist. On the other hand, Gibbs‟5 vector mechanics uses the radius vector r
from an origin O of a coordinate system to point B. O is the reference point, assumed to have absolute
meaning in space. The coordinate system itself is still unspecified.

Now we follow point B in time. Both s BA t  and r t  are functions of time. We tacitly assume
that A or O are somehow fixed. Fixed in what? A is a point of frame A which has an infinite number of
points that are mutually fixed; while O is part of the implied coordinate system.

Next we ask what is the change of the vector during a small dt? In both cases we use the ordinary

d d   dr
time derivative s BA and r . Vector mechanics writes v  and is finished. In the first case,
dt dt dt
which I call tensor mechanics, we use a more explicit nomenclature. Realizing that it is the velocity of
d
point B relative to frame A, we label it v B 
A
s BA . Then we express both s BA and v BA in a coordinate
dt
A
 ds 
system [] , and write [v ]  sBA    BA  . The coordinate system [] A is associated with
A A Ad A
B
dt  dt 
frame A, such that any two points of frame A have constant components. Notice the difference. Vector
mechanics implicitly assumes a coordinate system, while tensor mechanics distinguishes carefully
between frames and coordinate systems, making it clear in the nomenclature. A superscript next to the
variable name indicates a frame, while outside the bracket the superscript designates the coordinate
system. Points are always relegated to subscripts. The nomenclature I adopted is based on the hypothesis
that points and frames suffice to model any problem in flight dynamics. If true, this nomenclature is
self-defining. In forty years I found no exception.

dr
Vector mechanics makes another implicit assumption. The time derivative operates on the
dt

components of r that are related to the implied coordinate system, say A. What if we want to calculate the

time rate-of-change of r with respect to (wrt) another coordinate system, say C ? If C changes orientation

in time, expressed by the angular velocity vector  , then
 
dr dr  
(1)  r
dt C dt A

2
The two Cartesian coordinate systems and the radius vector are depicted in Figure 1.

B
3A
r ●
2C
3C

2A

1C
O

1A


Figure 1 Radius vector r related to coordinate systems A and C

In tensor dynamics, which formulates expressions independent of coordinate systems, a more

general time operator is used. It is called the rotational time derivative, based on Wundheiler‟s paper of
the 1930‟6, and applied to TFD in Appendix D of my textbook7. The velocity of point B relative to frame
A, v BA , is obtained from the displacement vector, s BA , by the rotational time derivative with respect to
(wrt) the frame A, D A  ,

(2) v BA  D As BA

Since no coordinate system is implied, this equation is valid in all coordinate systems, and is therefore an
invariant tensor concept. Tensor dynamics recognizes the fact that velocity is a concept of physics and
therefore independent of coordinate systems. Thus it relates the velocity to a physical frame and not to a
coordinate system as vector mechanics does. In vector mechanics coordinate systems and frames are
synonymous. In tensor dynamics they are entirely different things. The distinction being that frames
model such things as Earth, aircraft, antenna, etc., while coordinate systems establish a one-to-one
algebraic correspondence with the Euclidian three-space; i.e., physics vs. computation.

To relate the velocity of point B to two different frames, say A and C, we pick from frame C a
point C that coincides with point A of the displacement vector and abbreviate

(3) s  s BA  s BC

The two frames and the displacement vector are shown in Figure 2.

3
●B
S

●
C
C=A A

Figure 2 Displacement vector s related to frames A and C

Corresponding to Eq.(1), but in tensor formulation, the velocities are related by

(4) DC s  D As  Ω AC s

where Ω AC is the skew symmetric tensor of the angular velocity vector ω AC of frame A wrt frame C.
Eq.(4) is called the Euler transformation and applies to any vector s.

Equations (2), (3), and (4) are tensor relationships independent of coordinate systems and
therefore hold in any coordinate system. By a circuitous route Eq. (4) leads to the rotational time
derivative, though the strict mathematical proof is provided in Appendix D of my textbook. Express
Eq.(4) in coordinate system [] A associated with frame A

(5) [ D C s] A  [ D A s] A  [Ω AC ] A [s] A

A
 ds 
The first term on the right-hand side is the ordinary time derivative   , because the coordinate
 dt 
system [] A is associated with frame A. To deal with the second term, we realize that Ω AC has a matrix
form. We introduce the coordinate system []C associated with frame C and the transformation matrix
[T ] AC of coordinates A wrt C

AC
 dT 
(6) [Ω AC
]  [T ]
A AC
 
 dt 

with the over-bar indicating the transposed of the matrix. Substituting Eq.(6) into Eq.(5) and with the
ordinary time derivative term, we arrive at the rotational time derivative expression

AC
 ds   dT 
A

(7) [ D s]     [T ] AC   [ s] A
C A

 dt   dt 

which holds for any frame C with its associated coordinate system C, and any other coordinate system A.
Because A is arbitrary, Eq.(7) is valid for all coordinate systems and therefore the rotational time

4
derivative is a tensor concept. When operating on a tensor of rank one (vector) it again produces a tensor
of rank one. On the other hand, the ordinary time derivative does not have this quality. – Enough of
kinematics; let‟s turn to dynamics.

3. Dynamics
Flight dynamics is entirely in the realm of classical mechanics, specifically Newtonian
mechanics. Today‟s textbooks introduce Newton‟s Second Law as invariant relative to inertial coordinate
systems, though Newton never used this terminology. His law states that the time-rate-of change of linear
 
momentum dp dt equals the sum of the externally applied forces f and takes place in that direction

dp 
(8)  f
dt

Implying vector mechanics‟ phantom coordinate systems, Eq. (8) has the same form, i.e., is invariant for
all coordinate systems that are moving uniformly (linearly and at constant speed) relative to each other.
Notice the double duty of the coordinate systems. They are used as reference frames and as computational
tools. It was Galileo who had provided Newton with the Principle of Relativity: the laws of mechanics are
the same in uniformly moving (non-accelerated) coordinate systems. This principle together with
Galileo‟s principle of addition of velocities and Newton‟s three laws constitute the pillars that support
classical mechanics.

Enter Einstein with his Special Theory of Relativity8. He adopted from Maxwell‟s
electrodynamics the constancy of speed of light in vacuum and elevated that experimental result to a
universal principle. He started a revolution. With Einstein‟s beloved railway carriages I will illustrate the
consequences. Figure 3 shows the two cases. The left picture demonstrates Galileo‟s principle of
superposition of velocities. A shot is fired at the center of the carriage. The pressure waves propagate at
the speed of sound a in both directions v  a . Suppose the bow waves could be seen by the conductor
standing on the embankment observing the carriage travelling past him with speed w . He would notice
that the forward pressure wave propagates faster forward wrt the embankment, v  a  w , than
backward, v  a  w . Next to the conductor stands Einstein. He is more interested in the propagation of
the muzzle flash (right side of Figure 3). Inside the carriage the light waves propagate at the speed of light
c in both directions v  c , outside on the embankment Einstein notices that they also propagate in both
directions at the same v  c . He is overjoyed that his theory of Special Relativity is validated: The speed
of light is independent of the motion of the light source or the observer, with the caveat that the train
moves with constant (uniform) speed. These are the two pillars of Einstein‟s Special Relativity: (1) Speed
of light is constant and (2) all laws of physics are invariant in non-accelerated (inertial) coordinate
systems.

5
Sound Light
v=a v=a v=c v=c
w w

v=a-w v=a+w v=c v=c

Figure 3 Galileo‟s superposition of velocities versus Einstein‟s constancy of speed of light

So Einstein maintained Galileo‟s Principle of Relativity but discarded Galileo‟s Superposition of

Velocities. Absolute time is replaced by every observer carrying his own clock, with the consequences of
time dilation and length contraction, and the famous E  mc .
2

But Einstein was still brooding. What is so special about inertial coordinate systems? Shouldn‟t
the laws of nature be independent of the motion of the observer even if the observer is accelerated? After
all, physical phenomena occur with or without an observer. Discarding Galileo‟s Relativity he formulated
his Principle of General Relativity in 19169 and, in the same year, published a popular version translated
and reprinted by Dover10. He reconfirmed it in the 15th Edition, 1952, in Appendix 5: “Natural laws must
be covariant with respect to arbitrary continuous transformations of the coordinates”. Here „covariant‟
is synonymous with invariant. This postulate led him to the Principle of Equivalence of inertia and
gravitational mass with the consequence that the Euclidean space metric had to be replaced by the more
general Riemannien metric, determined by the distribution of mass in space. Let‟s use another one of
Einstein‟s favorite illustrations, the merry-go-round11. Climb in one of its chairs and feel the centrifugal
force pushing you outward; at least Newton would use this terminology. Now close your eyes and
disregard the wind in your face. Doesn‟t it feel like gravity is pulling on you? In both instances the forces
are proportional to the mass of your body. We use different names, but they are entirely equivalent
according to Einstein.

In Figure 4 the merry-go-round is stripped down to the essentials. A is the inertial frame and C is
the frame of the merry-go-round. Both frames share the common point of rotation C =A. B is the rider
who is a point of frame C.

CA
s ●B
C ●
C= A

Figure 4 Mass point B on the merry-go-round C , rotating at ω

CA
wrt base A

The centrifugal force is obtained from Newton‟s law. Assuming no external forces means that the linear

momentum p is conserved in the A coordinate system

6

dp
(9) 0
dt A

Taking the time derivative wrt the rotating coordinate system C introduces an extra term, namely, the
 
angular velocity of coordinate system C relative to A,  CA , vectorially multiplied with p

dp  
(10)   CA  p  0
dt C
 
Introducing p  mv , the centrifugal force that you experience on the merry-go-round is

dv  
(11) m  m CA  v
dt C

For coordinate systems A and C , Newton‟s law takes on different forms in Eqs.(9) and (10).

Now there is a conflict with Einstein‟s Principle of General Relativity! Paraphrased it states that
all laws of physics have the same form in all coordinates system. Is Newton‟s law an exception?
Certainly not. Classical mechanics is a subset of Einstein‟s General Theory. How can the conflict be
resolved?

First, we have to distinguish carefully between reference frames and coordinate systems. Frames
are models of physical entities and coordinate systems relate ordered algebraic numbers to the Euclidean
space metric. Second, we use the tensor calculus of Ricci and Levi-Civita12. Einstein applied it to great
advantage. For us the simplest Cartesian tensors will suffice. Finally, with the new rotational time
derivative we can formulate Newton‟s law invariant under all Cartesian coordinate transformations, even
those that have time dependent elements (rotationally accelerated).

The invariant formulation of Newton‟s Second Law is

(12) DIp  f

Where p is the linear momentum tensor of rank one, operated on by the rotational time derivative D I
wrt to inertial frame I and f is the sum of all external forces Given two arbitrary Cartesian coordinate
systems [] A and []C we can express Eq.(12) in both systems

(13) [ D I p]C  [ f ]C and [ D I p] A  [ f ] A

Because the forms are the same (no extra terms), Newton‟s law is an invariant tensor form and abides by
Einstein‟s Principle of General Relativity, though it is an utterly classical law of physics. Notice that the
I
rotational derivative operates on the inertial frame I, and not an inertial coordinate system [] . In case
the coordinate system [] A is an inertial system, the rotational time derivative reduces to the ordinary
time derivative

7
A
 dp 
(14) [ D I p] A     [ f ] A
 dt 

and we have the traditional expression of Newton‟s law in matrix notation, comparable to Eq.(8).

Solving the merry-go-round problem again in tensor form, we use Newton‟s Second Law, which
expresses the fact that the rate-of-change of linear momentum of point B wrt to the inertial frame A
vanishes, if no external forces are present

(15) D Ap BA  0

We shift from the inertial frame A to the rotating frame C with the generalized Euler transformation

(16) D Ap BA  DC p BA  ΩCAp BA  0

and obtain

(17) DC p BA  ΩCAp BA

Replacing the linear momentum by mass x velocity p BA  mv BA and introducing a coordinate system []C
associated with frame C , yields the matrix expression

(18) 
m DC vBA 
C
 m[Ω CA ]C [vBA ]C


where D v C
B
A C  dvBA 
  is the acceleration of point B relative to frame A expressed in coordinates
 dt 
[]C associated with frame C. This produces a relationship corresponding to Eq.(11)

C
 dvBA 
(19) m   m[Ω CA ]C [vBA ]C
 dt 

Our human specimen, not knowing that he is riding on a merry-go-round, could interpret the force as a
gravitational pull. With [vBA ]C  [Ω CA ]C [sBC ]C , the numerical value he experiences is calculated from

C
 dv A 
(20) m B   m[Ω CA ]C [Ω CA ]C [ sBC ]C
 dt 

A bystander, located in the inertial frame, however, would tell him that he is subject to a centrifugal force,
because of his rotating reference frame. In the first case m is interpreted as gravitational mass, while in the
second case as inertial mass; hence Einstein‟s Equivalence Principle.

8
This concludes our Gedanken experiments. Notice the approach of tensor dynamics: Model the
physical phenomena in tensors first, followed by introducing coordinate systems for computation. This
approach “from tensor modeling to matrix programming” has borne many fruits in aerospace applications.

4. Flight Dynamics
The foundation of flight dynamics is classical mechanics. What I have discussed so far is a
general, coordinate independent formulation of Newtonian mechanics. From Newton‟s Second Law and
Euler‟s law of attitude dynamics we get the six degrees of freedom equations of motion of aerospace
vehicles. These are the core of any flight simulation and the starting point for linearized perturbation
equations.

After two decades of teaching fight dynamics at the University of Florida my students prevailed
and I published the textbook “Modeling and Simulation of Aerospace Vehicle Dynamics”7 with the help
of the AIAA. It consists of two parts: Modeling of Flight Dynamics and Simulation of Aerospace
Vehicles. In the first part you can find a comprehensive course of analytical tensor mechanics as applied
to flight dynamics and you can use the second part to start your three, five, or six degrees of freedom
flight simulation. May the book serve as a comprehensive summary of tensor flight dynamics and its
applications. Here I will just mention the latest progress.

Recently, I have applied the Tensorial Approach to Modeling Robotic Dynamics, as published in
the proceedings of the ICNPAA Conference of Mathematical Problems in Engineering, Aerospace and
Sciences 2008, University of Genoa, Italy13. The classical method uses vector mechanics to formulate the
equations of motion of multi-linked manipulators. Each link carries several coordinate systems. From the
outset physics and computation are intertwined and cause much confusion to the novice. To demonstrate
the tensor method, I derived the Newton-Euler equations of robotic manipulators in their invariant tensor
form. For validation with current textbooks, I also expressed them in coordinate form. Complete
agreement was reached. While present methods mix physical and computational aspects right from the
outset, this article shows how the two-step derivation clarifies these two distinctly different phases, and
how tensor modeling contributes to a better understanding of robotic dynamics.

My 1972 paper1 found its way into my textbook as Chapter 7, Perturbation Equations. I was able
to derive the general perturbation equations of flight mechanics even for unsteady reference flight.
Without going into the details they are listed below in tensor from, valid in any Cartesian coordinate
system.

(21)  
m D Bp v IB  m R BpBr Ω BrI R BpBr v IB  m  Ω BI R BpBr v IBr   f a   ft  E  R BpBr f gr

Bp D  ω
I Bp Bp  ω
 R BpBr Ω BrI R BpBr I Bp   Ω BI R BpBr I Br
Br ω   m a   mt
Bp BI BI BrI
(22)

The prefix  on the left side indicates the state variables, linear and angular velocities; on the right side,
the force and moment perturbations.

Creagh used these equations as starting points for his recently published article in the Journal of
Guidance, Control and Dynamics14. In an earlier paper15 he presented his research on reference frames of
spinning vehicles, making the clear distinction between frames and coordinate systems.

9
Tensor flight dynamics is particularly useful for modeling modern-day aerospace vehicles. Over
four decades I have witnessed the increased sophistication of computer-based synthesis and analysis. In
the early days I used FORTRAN, but now with the object oriented C++ language we can create and
instantiate many different flight objects. So here we have a symbiosis of tensor flight dynamics as
modeling tool, and C++ as matrix cruncher.

The table summarizes the simulations that I created since programming in C++. They span a wide
range of aerospace vehicles from subsonic to hypersonic cruise missiles, from air-to-ground to air-to-air
missiles, and three-stage boosters. All are based on the common frame work CADAC (Computer Aided
Design of Aerospace Concepts) 16

Table CADAC++ Simulations

TYPE VEHICLE OBJECTS DoF EARTH FEATURES
Cruise Missile Missile; Target; Satellite 5 Spherical Remote Targeting
Fighter Aircraft Aircraft 6 Flat Generic F16
Air-to-Ground Missile Missile; Aircraft; Target 6 Flat Weather Deck, MC
Air-to-Air Missile Missile; Target Aircraft 6 Flat MC
Plane + Transfer Vehicle +
National Aerospace Generic X30,
Interceptor; Tracking Station; 6 WGS84
Plane Weather Deck, MC
Satellite
Generic Defense Defensive Missile; Aircraft;
6 Flat MC
Missile Offensive Missile
Insertion Guidance, Weather
Three Stage Booster Rocket with Three Stages 6 WGS84
Deck, MC
Long Range Strike
Missile; Target 5 Spherical Hypersonic, FLAMES®
Missile
Two Pulse Rocket, Integral
Dual Role Missile Missile; Target; Recce Aircraft 6 Flat
Rocket Ramjet, MC
Booster + Waverider + Munition;
Global Strike 6 WGS84 Wave Rider, MC
Satellite; Target
Defensive Missile; Aircraft;
Self Defense Missile 6 Flat Real time, MC, FLAMES®
Offensive Missile
Small Smart Bomb Bomb; Satellite; Target 6 WGS84 Weather Deck, MC
Hypersonic Cruise
Missile; Satellite; Target 6 WGS84 Scramjet, Weather Deck, MC
Missile

Most of these simulations model more than one vehicle. There are missiles, aircraft, targets, and satellites
that interact during flight. Depending on the application, the Earth may be modeled as flat, spherical or
elliptical (WGS84). Special features include wind and turbulence, Monte Carlo stochastic runs (MC), and
compatibility with FLAMES®, a mission level analysis tool.

5. Conclusions
After forty years, tensor modeling of flight dynamics has become an accepted approach. The
AIAA Flight Mechanics Committee played an important role by allowing me to present and publish my
1972 paper on Perturbation Methods in Atmospheric Flight Mechanics. Some Universities have

10
incorporated tensor flight dynamics into their graduate curriculum. But professors share my experience
that students are reluctant to forsake vector mechanics for the daunting tensor formulation; though I can
foresee a time when Newton‟s law will be taught as a truly invariant law of physics, beginning in High
School.

1 nd
Zipfel, Peter H., “Perturbation Methods in Atmospheric Flight Mechanics”, 2 Atmospheric Flight Mechanics
Conference, 11-13 September 1972, Palo Alto, CA.
2
Zipfel, Peter, H., “On Flight Dynamics of Magnus Rotors”, Department of the Army, Tech Report 117, November
1970, DTIC AD0716345.
3
Zipfel, Peter H., “Perturbation Methods in Atmospheric Flight Mechanics”, AIAA Journal, Vol. 11, No. 9,
September 1973, pp. 11247-1251.
4
Etkin, Bernard, Dynamics of Atmospheric Flight, John Wiley & sons, Inc, 1972.
5
Gibbs, Willard, J. The Scientific Papers of J Willard Gibbs, Vol II Dynamics, Dover 1961.
6
Wundheiler, Alexander, “Kovariante Ableitung und die Cesaroschen Unbeweglichkeitsbedingungen”,
Mathematische Zeitschrift Vol 36, 1932, pp. 104-109.
7 nd
Zipfel, Peter, H., Modeling and Simulation of Aerospace Vehicle Dynamics, 2 Edition, AIAA, 2007.
8
Einstein, Albert, “Zur Elektrodynamik Bewegter Körper”, Annalen der Physik Vol. 17, 1905, pp. 891-921.
9
Einstein, Albert, “Die Grundlage der Allgemeinen Relativitätstheorie”, Annalen der Physik Vol. 49, 1916, pp. 284-
338.
10
Einstein, Albert, Relativity, Dover 1961.
11
Ibid., p. 88.
12
Ricci, G., and Levi-Civita, T, “Methodes de Calcul Differentiel Absolu et Leurs Applications”, Mathematische
Annalen, Vol. 54, 1901.
13
Sivasundaram, S (Editor) Mathematical Problems in Engineering, Aerospace, and Sciences, Volume 5, Cambridge
Scientific Publishers, 2010.
14
Creagh, Michael, “Attitude Guidance for spinning Vehicles with Independent Pitch and Yaw Control”, Journal of
Guidance, Control, and Dynamics, Vol. 33 No. 3, 2010, pp. 915-922.
15
Creagh, Michael, “Comparison of Reference Frames in the Linearization of Flight Dynamics for Spinning
Vehicles”, AIAA Atmospheric Flight Mechanics Conference, Chicago, Illinois, Aug. 10-13, 2009.
16
Zipfel, Peter H., “CADAC: Multi-Use Architecture for Constructive Aerospace Simulations”, The Journal of
Defense Modeling and Simulation, April 2011.

11

View publication stats