Triple Pendulum
Triple Pendulum
Triple Pendulum
1
,
2
,
3
.
l
1
l
2
l
3
3
g
+
Figure 2: Schematic representation of the model.
We rst derive the equations of motion of
the frictionless ideal case. This allows for model
validation by ensuring energy is conserved in
the dynamics. Later we will add in frictional
damping, and see how it changes the dynamics.
Taking down as +y and right as +x, we write
the positions of the centers of mass of the bars
as functions of
i
and the geometric parameters
of the system.
y
1
=
l
1
2
cos
1
(1)
y
2
= l
1
cos
1
+
l
2
2
cos
2
(2)
y
3
= l
1
cos
1
+ l
2
cos
2
+
l
3
2
cos
3
(3)
x
1
=
l
1
2
sin
1
(4)
x
2
= l
1
sin
1
+
l
2
2
sin
2
(5)
x
3
= l
1
sin
1
+ l
2
sin
2
+
l
3
2
sin
3
(6)
These positions are then dierentiated with
respect to time to nd the x and y components
of the velocities as functions of angles and angu-
lar velocities. They will not be shown here for
brevity.
2
The magnitude of the velocity of each bar is
v
i
=
x
i
2
+ y
i
2
(7)
The translational and rotational kinetic energiy
(TKE and RKE) of each bar is:
TKE
i
=
1
2
m
i
v
2
i
(8)
RKE
i
=
1
2
I
i
i
(9)
The gravitational potential energy(GPE) of each
bar is:
GPE
i
= m
i
gy
i
(10)
Using these, the Lagrangian of the system
is:
L = T V =
3
i=1
TKE
i
+RKE
i
GPE
i
(11)
Equation 11 can then be input to Lagranges
Equation:
d
dt
(
dL
q
i
) =
dL
dq
i
(12)
where
i
is used for the generalized coordinate q
i
.
This yields a system of three equations which
contain the angular acceleration terms, omit-
ted for brevity. The solution of this system of
three equations and three unknowns yields the
expressions for the angular velocities. We then
numerically integrate these angular velocities
to produce the path of the pendulum. A char-
acteristic plot of the path is shown in Figure
6. Energy is conserved in the undamped sim-
ulation, and the pendulum appears to behave
as expected. An animation of the system was
also created and used to validate the results.
Videos of the animation are posted on YouTube
for chaotic (http://youtu.be/7lNIAdsInMg) and
periodic (http://youtu.be/THTgIZqDGQ) in-
tial conditions. With this validation of our basic
methods, we add damping to the system.
The bars of the experimental setup are thin,
and the velocities are generally low, so we chose
to model the system damping with a viscous
drag caused by the angular velocity of the joints.
This viscous form of drag can be modeled in
Lagrangian mechanics with the Rayleigh Dissi-
pation Function:
D =
1
2
(
3
i=1
k
i
i
2
) (13)
Lagranges equation is the rewritten as
d
dt
(
dL
q
i
) =
dL
dq
i
dD
d q
i
(14)
This modied form of Lagranges equation
produces a system of three equations which con-
tain the angular velocity terms, as above for the
undamped case. Solving for the angular velocity
terms produces the equations of motion shown
in Appendix A.
4 Results
After creating and validating or model, we col-
lected exprimental data for a triple pendulum
which had been constructed by Ben Smith for
the course. The parameters of this experimental
test setup were measured (Table 1) and input
into our simultaion. The triple pendulum was
tracked by an OptiTrack vision tracking system
and the data was parsed with a Python script
in order to return angles as functions of time
for the three links. This vision tracking was set
up by professor Aaron Hoover. Velocity data
for the three links was created through numer-
ical dierentiation and a low-pass lter using
MATLABs lter function.
Parameter Value Unit
m
1
0.2944 kg
m
2
0.1756 kg
m
3
0.0947 kg
l
1
0.508 m
l
2
0.254 m
l
3
0.127 m
I
1
9.526e-3 kg m2
I
2
1.625e-3 kg m2
I
3
1.848e-4 kg m2
k
1
5e-3 Nms/rad
k
2
0 Nms/rad
k
3
8e-4 Nms/rad
Table 1: Parameters of Experimental Triple Pen-
dulum System
We ran the experiment for a number of vary-
ing initial conditions, some which would create
chaotic and some which would create periodic
3
motion from the start. We used one of the peri-
odic cases (Table 2) to validate our simulation
and tune the damping constants (k
n
). The posi-
tion and velocity of each of the three links was
plotted as a function of time and the damping
constants were tuned until the positional and
velocity plots closely matched the experimental
data. The nal values for the damping constants
(k
n
) are given in Table 1.
Condition Value Unit
1
-0.4603 rad
2
-1.2051 rad
3
-1.5165 rad
theta
n
0 rad/s
Table 2: Initial Conditions for Periodic Experi-
mental Data
The energy of the pendulum was calculated
for both the simulation and experimental results
(Figure 3). The energy is similar for both the
experimental and simulation results, following
the same general decay curve. However, there
is oscillation in the experimental results which
increases and decreases the energy as the pendu-
lum swings. This is incorrect as there is nothing
present in the system which would transfer en-
ergy back into the pendulum once it is lost. We
attribute this error to inaccuracy in the numeri-
cal dierentiation which was used to determine
the pendulums velocity.
0 2 4 6 8 10
2.35
2.3
2.25
2.2
2.15
2.1
2.05
Energy
E
n
e
r
g
y
(
J
)
Time (s)
Simulation
Experiment
Figure 3: Comparison of energy over time for sim-
ulational and experimental data with
damping.
The position and velocity as functions of
time for each of the three pendulum links are
shown in Figures 4 and 5 for the tested periodic
case. As shown, the data matches up with the
experimental results almost perfectly.
0 2 4 6 8 10
0.5
0
0.5
Link 1
0 2 4 6 8 10
1
0.5
0
0.5
A
n
g
u
l
a
r
P
o
s
i
t
i
o
n
(
r
a
d
)Link 2
0 2 4 6 8 10
1
0
1
Time (s)
Link 3
Simulation
Experiment
Figure 4: Comparison of position over time for
simulational and experimental data with
damping.
4
0 2 4 6 8 10
2
0
2
Link 1
0 2 4 6 8 10
5
0
5
A
n
g
u
l
a
r
V
e
l
o
c
i
t
y
(
r
a
d
/
s
)
Link 2
0 2 4 6 8 10
10
5
0
5
10
Time (s)
Link 3
Simulation
Experiment
Figure 5: Comparison of velocity over time for
simulational and experimental data with
damping.
When doing the double pendulum, we found
that for chaotic systems the modeled and exper-
imental results to separate quickly and postu-
lated that this eect was due to damping in the
system which our prior model did not take into
account. This hypothesis was tested by using
our simulation to plot the paths of the triple
pendulum masses for a set of chaotic initial con-
ditions for both the damped and undamped case
(Figure 6). Although there are denite qualita-
tive similarities in the paths of the two masses,
the addition of damping causes their paths to
dier rather signicantly and separate quickly
after the rst period of high kinetic energy.
0.5 0 0.5
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
M1 Path
M2 Path
M3 Path
0.5 0 0.5
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
M1 Path
M2 Path
M3 Path
Figure 6: The paths of the three links for both (a)
undamped and (b) damped cases for a
set of chaotic initial conditions.
5 Diagnosis
The largest problems we encountered with this
project involved our infamiliarity with Mathe-
matica, the symbolic algebra system we used
to solve for the equations of motion. When we
started in this project, we knew just enough to
get us in trouble. Once we had the Lagrangian,
we had diculty in applying Lagranges equa-
tion to it, as it requires taking derivatives of a
function with respect to another function. This
combined with our general unfamiliarity with
Mathematica syntax made it dicult for us to
x problems. To move forward, we had to step
back and go through a few basic Mathematica
tutorials. Once we had a stronger grasp on the
fundamentals, we found it much easier to solve
the problems we were encountering, even though
5
they existed on a higher level of abstraction.
Once we had the equations of motion, we
had diculty in calculating the energies of the
system. We had solved the dynamics both using
lagrangians and vectors, and our notation did
not match between them. Small ambiguities
such as the sign of gravity and the denition
of our axes were not clearly dened when we
began the derivations. This made interpreting
the results of the simulation dicult. When we
went back to the beginning and explicitly wrote
out our system denitions, we were able to more
clearly interpret and use our results.
The motion capture system we used to cap-
ture experimental data of a real triple pendulum
did not tolerate high angular velocities well. It
would lose track of links when they began mov-
ing too quickly. We could not nd an eective
x for this. We made due with what we had
by only testing initial conditions which did not
produce overly large velocities in the system.
6 Improvement
The main improvements which could be made
on the model are for more accurate representa-
tions of the damping forces on the system. This
current model takes into eect viscous damping.
A future model could extend this by also includ-
ing aerodynamic drag as the bars swing through
the air.
Furthermore, when testing the system we
found that when the pendulum encountered
rapid, chaotic movements, the bars of the pen-
dulum encountered dramatic out of plane vi-
brations. These vibrations transferred into the
table onto which the system was mounted, shak-
ing the table and removing energy from the
system. Although this is a rather complicated
damping eect, large quantities of energy were
likely lost due to these vibrations and a model of
the experimental setup would not be complete
without including them.
In addition to adding better models of damp-
ing, the model could be made more general by
taking it into the third dimension with a three-
dimensional triple pendulum.
Finally, there are a number of small physi-
cal eects which could be added to the model.
For example, the model could be extended to
include the elongation of the steel pendulum
rods, the change in gravitational constant as
the rods change in elevation or even the gravita-
tional eects of the bars on each other. However,
most of these eects are negligible and would
not signicantly change the results.
7 Reection
Leaning about Lagrangian mechanics from an
operational perspective was interesting. In some
ways it simplies the derivation of the equations
of motion: we had to solve three equations with
Lagrangian mechanics instead of the nine we
would have needed to do it with Newtonian
mechanics. For mechanical systems you are ac-
tually trying to design, the Newtonian perspec-
tive is vastly better because it gives insight into
internal forces which might require structural
changes and which variables are most important
to the system. We learned that while the La-
grangian method can simplify calculating the
dynamics of a system signicantly, it is really
only useful for a small subset of highly theoreti-
cal problems.
Even very small drag forcesmchanged the dy-
namics of the triple pendulum signicantly. This
drives home the idea that a triple pendulum is a
chaotic system. If small changes in drag create
wildly dierent behaviours you need dierent
techniques to learn someting from your model,
because it is unlikely that it will give a highly
accurate prediction of the actual behaviour.
We also learned about using Mathematica to
do large amounts of simple algebra and calculus.
The ability to dene intermediate functions al-
lowed us to work on a higher level of abstraction
than the gory algebraic details. For example,
this allowed us to simple write v
1
instead of the
long expression corresponding to this in terms
of base quantities. This allowed us to see the
key components of what we were doing with-
out having to slog through the algebraic detail.
However, it was absolutely neccesary to have a
strong understanding of what we wanted to do
before using Mathematica. This allowed us to
better debug the code and interpret its output.
6
8 Conclusion
Using Lagrangian energy methods, we success-
fully created a mathematical model featuring a
set of coupled ordinary dierential equations of
motion for the dynamic compound triple pen-
dulum system. These equations of motion were
simulated in MATLAB to create a numerical
model of the system. After tuning damping pa-
rameters, the model t experimental data from
the real compound triple pendulum almost per-
fectly for upwards of ten seconds. The behaviour
of the model shows that the inclusion of damp-
ing forces signicantly alters the dynamics of the
system after the rst few seconds, supporting
our hypothesis. In conclusion, the results of our
work supported our hypothesis and we consider
this project a success.
9 Future Usage
The step between a double and a triple pendu-
lum is a rather minor one. Mostly it involves
more of the same type of mathematics. At most
this is character building if solving by hand, but
because we used a computer algebra system, it
made little dierence. It is probably better to
instead study a double pendulum. That said,
after having done a bob double pendulum, a
compound double pendulum, and a triple pen-
dulum, it is interesting to see that the process
is really exactly the same.
However, the introduction of drag into the
model was quite interesting. Seeing how this
changed the dynamics of the system drove home
the chaotic behaviour of the pendulum. Further-
more, thinking about how to represent a drag
torque in the mathematical model of the system
gave us a greater understanding of what torque
is, although this did not make it into the nal
report. An appropriate problem state for our
project might be:
Consider a planar triple pendulum which
consists of three massed bars joined by pivots.
The pendulum is xed to a stable structure at
the top bar. Each section has mass m
1
, m
2
, and
m
3
with moments of inertia about the center of
mass I
1
I
2
and I
3
.
Write the Lagrangian of the system in
terms of the angular positions and veloci-
ties of the bars.
Determine the angular acceleration of the
bars with Lagranges equation.
Write a MATLAB script to simulate the
motion of the pendulum. Validate your
results with a number of initial conditions
and show that energy is conserved.
Now include damping using Rayleighs Dis-
sipation Function or some other method.
What form of damping would be most ac-
curate? Viscous or aerodyanmic, or some-
thing else? What eect does the inclusion
of drag have on the dynamics of the sys-
tem?
References
[1] Widnall, S. Lecture L20 - Energy Methods:
Lagranges Equations. N.p.: 16.07 Dyan-
mics, 2009. PDF.
[2] Bannister, Ross. The Double Pendulum.
The Double Pendulum. University of Read-
ing, June 2001. Web. 09 Dec. 2012.
[3] Von Herrath, Franziska, and Scott Mandell.
The Double Pendulum Problem. 19 May
2000. Lecture.
[4] Marion and Thorton, Classical dynamics
of particles and systems, 4 ed., Saunders.
College Publishing, Fort Worth, TX, 1995.
[5] Jones, Rocky M., and Kuch N. Patel. Ex-
amination of Chaos in Multiple Pendulum
Systems Through Computer Visualization
in Java. Examination of Chaos in Multi-
ple Pendulum Systems Through Computer
Visualization in Java. N.p., n.d. Web. 10
Dec. 2012.
7
Appendix A Equations of Motion
(15)
1
= (2((l
2
3
m
2
3
sin(2
1
2
3
)(4I
2
l
2
2
m
2
) + l
2
2
sin(2
1
2
2
)(m
2
+ 2m
3
)(m
2
m
3
l
2
3
+ 4I
3
(m
2
+
2m
3
)))l
2
1
1
2
+ (l
2
(sin(
1
2
)((m
2
m
3
(m
2
+ 3m
3
)l
2
3
+ 4I
3
(m
2
2
+ 6m
2
m
3
+ 8m
2
3
))l
2
2
+ 4I
2
(m
3
(m
2
+
m
3
)l
2
3
+ 4I
3
(m
2
+ 2m
3
))) + l
2
3
m
2
3
sin(
1
+
2
2
3
)(4I
2
l
2
2
m
2
))
2
2
4k
2
l
2
(cos(
1
2
)(m
3
(m
2
+
m
3
)l
2
3
+4I
3
(m
2
+2m
3
))l
2
3
m
2
3
cos(
1
+
2
2
3
))
2
+l
3
m
3
(sin(
1
3
)(8I
3
m
3
l
2
2
+4I
2
m
3
l
2
3
+16I
2
I
3
)+
l
2
2
sin(
1
2
2
+
3
)(m
2
m
3
l
2
3
+4I
3
(m
2
+2m
3
)))
3
2
4k
3
l
3
m
3
(cos(
1
3
)(2m
3
l
2
2
+4I
2
) l
2
2
cos(
1
2
2
+
3
)(m
2
+ 2m
3
))
3
g(sin(
1
)((m
3
(m
1
m
2
+ 2m
1
m
3
+ 3m
2
m
3
+ m
2
2
)l
2
3
+ 4I
3
(m
2
2
+ 6m
2
m
3
+
m
1
m
2
+ 4m
2
3
+ 4m
1
m
3
))l
2
2
+ 4I
2
(m
3
(m
1
+ 2m
2
+ m
3
)l
2
3
+ 4I
3
(m
1
+ 2m
2
+ 2m
3
))) + l
2
3
m
2
3
(sin(
1
2
3
)(4I
2
l
2
2
m
2
) 2l
2
2
cos(2
2
2
3
) sin(
1
)(m
1
+ m
2
)) + l
2
2
sin(
1
2
2
)(m
2
+ 2m
3
)(m
2
m
3
l
2
3
+
4I
3
(m
2
+2m
3
))))l
1
+2k
1
(4I
2
(m
3
l
2
3
+4I
3
) +l
2
2
(m
3
(m
2
+2m
3
)l
2
3
+4I
3
(m
2
+4m
3
)) 2l
2
2
l
2
3
m
2
3
cos(2
2
2
3
))
1
))/(64I
1
I
2
I
3
+8I
3
l
2
1
l
2
2
m
2
2
+8I
1
l
2
2
l
2
3
m
2
3
+8I
2
l
2
1
l
2
3
m
2
3
+32I
3
l
2
1
l
2
2
m
2
3
+16I
2
I
3
l
2
1
m
1
+16I
1
I
3
l
2
2
m
2
+
64I
2
I
3
l
2
1
m
2
+16I
1
I
2
l
2
3
m
3
+64I
1
I
3
l
2
2
m
3
+64I
2
I
3
l
2
1
m
3
+4I
3
l
2
1
l
2
2
m
1
m
2
+4I
2
l
2
1
l
2
3
m
1
m
3
+16I
3
l
2
1
l
2
2
m
1
m
3
+
4I
1
l
2
2
l
2
3
m
2
m
3
+16I
2
l
2
1
l
2
3
m
2
m
3
+48I
3
l
2
1
l
2
2
m
2
m
3
8I
1
l
2
2
l
2
3
m
2
3
cos(2
2
2
3
) 2l
2
1
l
2
2
cos(2
1
2
2
)(m
2
+
2m
3
)(m
2
m
3
l
2
3
+4I
3
(m
2
+2m
3
))2l
2
1
l
2
3
m
2
3
cos(2
1
2
3
)(m
2
l
2
2
+4I
2
)+2l
2
1
l
2
2
l
2
3
m
1
m
2
3
+6l
2
1
l
2
2
l
2
3
m
2
m
2
3
+
2l
2
1
l
2
2
l
2
3
m
2
2
m
3
+ l
2
1
l
2
2
l
2
3
m
1
m
2
m
3
2l
2
1
l
2
2
l
2
3
m
1
m
2
3
cos(2
2
2
3
) 4l
2
1
l
2
2
l
2
3
m
2
m
2
3
cos(2
2
2
3
))
(16)
2
= (2((l
2
1
sin(2
1
2
2
)(m
2
+ 2m
3
)(m
2
m
3
l
2
3
+ 4I
3
(m
2
+ 2m
3
)) l
2
3
m
2
3
sin(2
2
2
3
)((m
1
+
2m
2
)l
2
1
+4I
1
))l
2
2
2
2
+l
1
(sin(
1
2
)((m
3
(m
1
(m
2
+m
3
) +2m
2
(2m
2
+3m
3
))l
2
3
+4I
3
(m
2
+2m
3
)(m
1
+
4m
2
+ 4m
3
))l
2
1
+ 4I
1
(m
3
(m
2
+ m
3
)l
2
3
+ 4I
3
(m
2
+ 2m
3
))) l
2
3
m
2
3
sin(
1
+
2
2
3
)((m
1
+ 2m
2
)l
2
1
+
4I
1
))l
2
1
2
+ 4k
1
l
1
(cos(
1
2
)(m
3
(m
2
+ m
3
)l
2
3
+ 4I
3
(m
2
+ 2m
3
)) l
2
3
m
2
3
cos(
1
+
2
2
3
))l
2
1
+
(l
3
m
3
(sin(
2
3
)((m
3
(m
1
+3m
2
)l
2
3
+4I
3
(m
1
+3m
2
+2m
3
))l
2
1
+4I
1
(m
3
l
2
3
+4I
3
))l
2
1
sin(2
1
3
)(m
2
m
3
l
2
3
+4I
3
(m
2
+2m
3
)))
3
2
+4k
3
l
3
m
3
(cos(
2
3
)((m
1
+3m
2
+2m
3
)l
2
1
+4I
1
) l
2
1
cos(2
1
3
)(m
2
+2m
3
))
3
+g(sin(
2
)((m
2
m
3
(2m
2
+3m
3
)l
2
3
+8I
3
(m
2
2
+3m
2
m
3
+2m
2
3
))l
2
1
+4I
1
(m
3
(m
2
+
m
3
)l
2
3
+ 4I
3
(m
2
+ 2m
3
))) l
2
1
sin(2
1
2
)(m
3
(m
1
(m
2
+ m
3
) + m
2
(2m
2
+ 3m
3
))l
2
3
+ 4I
3
(m
2
+
2m
3
)(m
1
+ 2m
2
+ 2m
3
)) + l
2
3
m
2
3
(sin(
2
2
3
)(m
2
l
2
1
+ 4I
1
) + l
2
1
sin(2
1
+
2
2
3
)(m
1
+ m
2
))))l
2
2k
2
(4I
1
(m
3
l
2
3
+ 4I
3
) + l
2
1
(m
3
(m
1
+ 4m
2
+ 2m
3
)l
2
3
+ 4I
3
(m
1
+ 4m
2
+ 4m
3
)) 2l
2
1
l
2
3
m
2
3
cos(2
1
2
3
))
2
))/(64I
1
I
2
I
3
+8I
3
l
2
1
l
2
2
m
2
2
+8I
1
l
2
2
l
2
3
m
2
3
+8I
2
l
2
1
l
2
3
m
2
3
+32I
3
l
2
1
l
2
2
m
2
3
+16I
2
I
3
l
2
1
m
1
+16I
1
I
3
l
2
2
m
2
+
64I
2
I
3
l
2
1
m
2
+16I
1
I
2
l
2
3
m
3
+64I
1
I
3
l
2
2
m
3
+64I
2
I
3
l
2
1
m
3
+4I
3
l
2
1
l
2
2
m
1
m
2
+4I
2
l
2
1
l
2
3
m
1
m
3
+16I
3
l
2
1
l
2
2
m
1
m
3
+
4I
1
l
2
2
l
2
3
m
2
m
3
+16I
2
l
2
1
l
2
3
m
2
m
3
+48I
3
l
2
1
l
2
2
m
2
m
3
8I
1
l
2
2
l
2
3
m
2
3
cos(2
2
2
3
) 2l
2
1
l
2
2
cos(2
1
2
2
)(m
2
+
2m
3
)(m
2
m
3
l
2
3
+4I
3
(m
2
+2m
3
))2l
2
1
l
2
3
m
2
3
cos(2
1
2
3
)(m
2
l
2
2
+4I
2
)+2l
2
1
l
2
2
l
2
3
m
1
m
2
3
+6l
2
1
l
2
2
l
2
3
m
2
m
2
3
+
2l
2
1
l
2
2
l
2
3
m
2
2
m
3
+ l
2
1
l
2
2
l
2
3
m
1
m
2
m
3
2l
2
1
l
2
2
l
2
3
m
1
m
2
3
cos(2
2
2
3
) 4l
2
1
l
2
2
l
2
3
m
2
m
2
3
cos(2
2
2
3
))
8
(17)
3
= (2(32I
1
I
2
k
3
3
l
2
l
3
m
3
2
2
(sin(
2
3
)((l
2
2
(m
1
m
2
+ 4m
1
m
3
+ 6m
2
m
3
+ m
2
2
) + 4I
2
(m
1
+
3m
2
+ 2m
3
))l
2
1
+ 4I
1
(4I
2
+ l
2
2
(m
2
+ 4m
3
))) + l
2
1
sin(2
1
2
3
)(m
2
+ 2m
3
)(4I
2
m
2
l
2
2
))
l
1
l
3
m
3
1
2
(sin(
1
3
)(8I
1
(m
3
l
2
2
+2I
2
) +2l
2
1
((m
1
m
3
m
2
2
)l
2
2
+2I
2
(m
1
+4m
2
+4m
3
))) l
2
2
sin(
1
2
2
+
3
)(m
2
+ 2m
3
)((m
1
+ 2m
2
)l
2
1
+ 4I
1
)) + 4k
3
l
2
1
l
2
2
m
2
2
3
+ 16k
3
l
2
1
l
2
2
m
2
3
3
+ 8I
2
k
3
l
2
1
m
1
3
+
8I
1
k
3
l
2
2
m
2
3
+ 32I
2
k
3
l
2
1
m
2
3
+ 32I
1
k
3
l
2
2
m
3
3
+ 32I
2
k
3
l
2
1
m
3
3
4k
1
l
1
l
3
m
3
1
(cos(
1
3
)(2m
3
l
2
2
+
4I
2
) l
2
2
cos(
1
2
2
+
3
)(m
2
+ 2m
3
)) 16I
1
I
2
gl
3
m
3
sin(
3
) 4I
2
l
2
1
l
2
3
m
2
3
3
2
sin(2
1
2
3
)
4I
1
l
2
2
l
2
3
m
2
3
3
2
sin(2
2
2
3
) +8I
2
gl
2
1
l
3
m
2
3
sin(2
1
3
) +8I
1
gl
2
2
l
3
m
2
3
sin(2
2
3
) +2k
3
l
2
1
l
2
2
m
1
m
2
3
+
8k
3
l
2
1
l
2
2
m
1
m
3
3
+ 24k
3
l
2
1
l
2
2
m
2
m
3
3
4k
2
l
2
l
3
m
3
2
(cos(
2
3
)((m
1
+ 3m
2
+ 2m
3
)l
2
1
+ 4I
1
)
l
2
1
cos(2
1
2
3
)(m
2
+ 2m
3
)) 8I
1
gl
2
2
l
3
m
2
3
sin(
3
) 8I
2
gl
2
1
l
3
m
2
3
sin(
3
) 4k
3
l
2
1
l
2
2
m
2
2
3
cos(2
1
2
2
) 16k
3
l
2
1
l
2
2
m
2
3
3
cos(2
1
2
2
) 8I
2
gl
2
1
l
3
m
2
m
3
sin(
3
) 16k
3
l
2
1
l
2
2
m
2
m
3
3
cos(2
1
2
2
)
l
2
1
l
2
2
l
2
3
m
1
m
2
3
3
2
sin(2
2
2
3
) + l
2
1
l
2
2
l
2
3
m
2
m
2
3
3
2
sin(2
1
2
3
) 2l
2
1
l
2
2
l
2
3
m
2
m
2
3
3
2
sin(2
2
2
3
) + 2gl
2
1
l
2
2
l
3
m
1
m
2
3
sin(2
1
3
) gl
2
1
l
2
2
l
3
m
2
2
m
3
sin(2
1
3
) + 2gl
2
1
l
2
2
l
3
m
2
m
2
3
sin(2
2
3
) + gl
2
1
l
2
2
l
3
m
2
2
m
3
sin(2
2
3
) + 4I
2
gl
2
1
l
3
m
1
m
3
sin(2
1
3
) + 8I
2
gl
2
1
l
3
m
2
m
3
sin(2
1
3
) +
4I
1
gl
2
2
l
3
m
2
m
3
sin(2
2
3
) 2gl
2
1
l
2
2
l
3
m
1
m
2
3
sin(2
1
2
2
+
3
) 2gl
2
1
l
2
2
l
3
m
2
m
2
3
sin(2
1
2
2
+
3
) gl
2
1
l
2
2
l
3
m
2
2
m
3
sin(2
1
2
2
+
3
) + gl
2
1
l
2
2
l
3
m
2
2
m
3
sin(
3
) gl
2
1
l
2
2
l
3
m
1
m
2
m
3
sin(2
1
2
2
+
3
)))/(64I
1
I
2
I
3
+ 8I
3
l
2
1
l
2
2
m
2
2
+ 8I
1
l
2
2
l
2
3
m
2
3
+ 8I
2
l
2
1
l
2
3
m
2
3
+ 32I
3
l
2
1
l
2
2
m
2
3
+ 16I
2
I
3
l
2
1
m
1
+ 16I
1
I
3
l
2
2
m
2
+
64I
2
I
3
l
2
1
m
2
+16I
1
I
2
l
2
3
m
3
+64I
1
I
3
l
2
2
m
3
+64I
2
I
3
l
2
1
m
3
+4I
3
l
2
1
l
2
2
m
1
m
2
+4I
2
l
2
1
l
2
3
m
1
m
3
+16I
3
l
2
1
l
2
2
m
1
m
3
+
4I
1
l
2
2
l
2
3
m
2
m
3
+16I
2
l
2
1
l
2
3
m
2
m
3
+48I
3
l
2
1
l
2
2
m
2
m
3
8I
1
l
2
2
l
2
3
m
2
3
cos(2
2
2
3
) 2l
2
1
l
2
2
cos(2
1
2
2
)(m
2
+
2m
3
)(m
2
m
3
l
2
3
+4I
3
(m
2
+2m
3
)) 2l
2
1
l
2
3
m
2
3
cos(2
1
2
3
)(4I
2
m
2
l
2
2
) +2l
2
1
l
2
2
l
2
3
m
1
m
2
3
+6l
2
1
l
2
2
l
2
3
m
2
m
2
3
+
2l
2
1
l
2
2
l
2
3
m
2
2
m
3
+ l
2
1
l
2
2
l
2
3
m
1
m
2
m
3
2l
2
1
l
2
2
l
2
3
m
1
m
2
3
cos(2
2
2
3
) 4l
2
1
l
2
2
l
2
3
m
2
m
2
3
cos(2
2
2
3
))
9