Exploring The Rubber Sheet Spacetime Analogy by Studying Ball Movement in A Bent Trampoline

European Journal of Physics
ACCEPTED MANUSCRIPT
Exploring the rubber sheet spacetime analogy by studying ball

movement in a bent trampoline
To cite this article before publication: Pau Batlle et al 2019 Eur. J. Phys. in press https://doi.org/10.1088/1361-6404/ab1a5c
Manuscript version: Accepted Manuscript

Accepted Manuscript is “the version of the article accepted for publication including all changes made as a result of the peer review process,
and which may also include the addition to the article by IOP Publishing of a header, an article ID, a cover sheet and/or an ‘Accepted
Manuscript’ watermark, but excluding any other editing, typesetting or other changes made by IOP Publishing and/or its licensors”
This Accepted Manuscript is © 2018 European Physical Society.
During the embargo period (the 12 month period from the publication of the Version of Record of this article), the Accepted Manuscript is fully
protected by copyright and cannot be reused or reposted elsewhere.
As the Version of Record of this article is going to be / has been published on a subscription basis, this Accepted Manuscript is available for reuse
under a CC BY-NC-ND 3.0 licence after the 12 month embargo period.
After the embargo period, everyone is permitted to use copy and redistribute this article for non-commercial purposes only, provided that they
adhere to all the terms of the licence https://creativecommons.org/licences/by-nc-nd/3.0
Although reasonable endeavours have been taken to obtain all necessary permissions from third parties to include their copyrighted content
within this article, their full citation and copyright line may not be present in this Accepted Manuscript version. Before using any content from this
article, please refer to the Version of Record on IOPscience once published for full citation and copyright details, as permissions will likely be
required. All third party content is fully copyright protected, unless specifically stated otherwise in the figure caption in the Version of Record.
View the article online for updates and enhancements.
This content was downloaded from IP address 148.212.19.20 on 18/04/2019 at 15:00

Page 1 of 34 AUTHOR SUBMITTED MANUSCRIPT - EJP-104179.R1
1
2
3
4
5
6
7
8
pt
9
10 Exploring the rubber sheet spacetime analogy by
11
12 studying ball movement in a bent trampoline
13
cri
14
15
16
Pau Batlle1 , Adam Teixidó1 , Joan Llobera1 , Isabel Medrano1 ,
17 Luis Carlos Pardo2
18
1
19 Universitat Politècnica de Catalunya, ETSETB, Enginyeria Fı́sica, Barcelona, Spain
20
us
2
21 Grup de Caracterització de Materials, Departament de Fı́sica, EEBE, Universitat
22 Politècnica de Catalunya, Barcelona, Spain
23
24 E-mail: pau.batlle.franch@gmail.com
25
26
27
28
29
an
Abstract. FISIDABO is an educational project aiming to promote critical thinking
among students. In the frame of this project, and motivated by the ubiquitous
qualitative analogy between spacetime warping by massive objects and the deformation
30 of a rubber sheet to explain General Relativity, a group of students of the Engineering
31 Physics degree have undertaken the study of the movement of a small ball on a
32
dM
trampoline bent by the presence of a heavy mass in its centre. As literature states,
33 similarities between how masses move under warped spacetime and under a bent
34
trampoline are only qualitative. Some observed analogy flaws are described which
35
can be useful for teaching general relativity. Triggered by the failure of this analogy,
36
37 a complete study of the movement of particles in a bent rubber sheet is developed.
38 Several models to describe such trajectories and the resulting shape of the trampoline
39 are proposed. Most importantly, a fitting routine to obtain the parameters needed
40 for the description of the real trajectory is presented. The high accuracy between
41 optimized and real trajectories recorded by a video camera implies that the model is
42 able to explain the experiment behaviour.
43
pte
44
45
46
47
48
49 1. Introduction
50
51
ce
52 1.1. The FISIDABO educational project

53
54
55 Physics is mostly taught at a merely theoretical level. In order to exemplify theoretical
56 concepts, students perform some experiments in the lab in a very controlled way. The
Ac
57
results of their practical work are usually known beforehand and no place for creativity
58
59 or critical thought is left. Labs are thus just a way to “prove” that physics explained in
60
AUTHOR SUBMITTED MANUSCRIPT - EJP-104179.R1 Page 2 of 34
1
2
3 2
4
5 front of a blackboard work, at the cost of hiding how real science is built. The result
6 of the experiments performed by researchers is not known beforehand and they are
7
affected by several errors and uncontrolled circumstances that make their interpretation
8
pt
9 not straightforward. If an experiment provides an unexpected result it is not easy
10 to know whether it is due to a failure of the physical model aiming to describe the
11 experiment, or due to some external variable not taken into account that is masking the
12
13 final result.
cri
14
15 Emotions are a key factor to enhance students performance in complex subjects such
16 as physics [1, 2], and these emotions are hard to find in controlled lab experiments. A
17 place where the entanglement of emotions and the laws of physics takes place naturally
18
19 are amusement parks. The goal of the FISIDABO project [3] is to involve students in
20 simple research projects designed by themselves and performed in Tibidabo, Barcelona’s
us
21 amusement park. Additionally, the opportunity to be involved in a more complex project
22
23 is given to some students with high capabilities in physics and mathematics. This part
24 of the FISIDABO project has been given the name of HIPATIA.
25
26 The goal of FISIDABO HIPATIA is to provide the students with a frame as similar
27
28
29
an
as possible to real science. The project is led by researchers in different areas such
as astrophysics or chemical physics that have first-hand information on how science is
30 working in the real world. The project has four parts:
31
32 • First, the students must write a proposal to perform an experiment at one of the
dM
33 rides of the Tibidabo amusement park. The best project is then chosen based on
34
35 its originality and feasibility.
36
37 • In a second step, the students design the experiment, given all the constraints
38 imposed by the amusement park rides.
39
40 • The students are given a limited time to perform the experiment at the amusement
41 park.
42
43
pte
• Finally, the students analyze the data, and depending on the quality of their work,
44
45 they write a scientific article as the one presented.
46
47 The final goal is, in any case, to take the students out of their comfort zone and make
48 them face the complexity of real experiments, promoting their critical thinking.
49
50 In previous experiments, students have investigated the dynamics of crash cars [4] or
51
ce
the movement of a pendulum in a rotating plane [5]. The project presented in this
52
53 article was undertaken by a group of students of Physics Engineering from “Universitat
54 Politècnica de Catalunya” lead by a researcher in the field of chemical physics.
55
56
Ac
57
58
59
60
1
2
3 3
4
5 1.2. Field of research overview and role of this work
6
7 One of the most novel concepts Einstein introduced in his theory of GR is the connection
8
pt
9 between gravitational forces and spacetime geometry. In this theory, gravity is not
10 thought anymore as a force between objects due to its masses. Instead, gravity appears
11 as the result of the mass of an object warping the spacetime around it. Objects follow
12
13 geodesics when moving in spacetime only under the effects of gravity, the warping
cri
14 alters the geodesics, in turn changing the trajectories. Since spacetime warping can
15 be a difficult concept to understand and visualize, an analogy is often used in physics
16
17 education: spacetime can be thought as a thin rubber sheet. The way masses warp the
18 rubber sheet and affect other smaller masses can help students to develop an intuition
19 about how gravity works under GR.
20
us
21 Because of how illustrative the ball-surface system is, it has been used not only in
22 the aforementioned analogy but in other ones (see [6] for an example regarding binary
23 star systems), and has also been studied in some previous works. In [6], the authors
24
25 analyze the shape of a spandex in the large radius regime. In [7], the trajectories of
26 rolling marbles over some specific surface shapes are studied in depth. Finally, in [8],
27
28
29
an
the authors specifically compare the obtained orbits in a warped spandex to the ones
predicted both by Newtonian Mechanics and General Relativity, concluding that they
30 differ in both cases from the spandex orbits. The same article states, however, that
31 the analogy between the two remains conceptually useful for GR beginners, conclusion
32
dM
33 which was also achieved in [9], which reported gains in students who were presented
34 with this analogy. It is therefore important to make clear which conclusions can be
35 derived from the analogy and which can not. In this article, concrete examples of flaws
36
37 of this analogy are provided, which arise because of the dissimilitude between GR and
38 classical mechanics and are supported by the data obtained from an experiment done
39 in the Tibidabo amusement park. These examples could be used by teachers to help
40
41 students understand that the analogy just acts as a qualitative example.
42
43 Regarding the general study of the ball-surface physical system, this article provides, on
pte
44 the one hand, a physical model of the shape of the surface generated in the experimental
45 conditions, and on the other hand, a more general model for the movement of a ball
46
47 over a surface, which does not require the surface to be radially symmetric such as in
48 previous works.
49
50
51
ce
2. Experimental methods
52
53
54 The two main goals of the experimental part of this article were, on the one hand,
55
56 warping the trampoline and measuring its shape, and on the other hand, recording the
Ac
57 movement of a small ball orbit around another, bigger ball, placed in the centre.
58
59
60
1
2
3 4
4
5 2.1. Warping the trampoline and recording its shape
6
7 An illustration of the experimental design including the central ball along with the
8
pt
9 dimensions of the trampoline can be found in Figure 1.
10
11
12
13
cri
14
15
16
17
18
19
20
us
21
22
23
24
25
26
27
28
29
an
30
31 Figure 1: This figure shows the dimensions (in m) of the trampoline and adjacent
32 platforms used in the experiment together with the position of the central ball
dM
33
34
35
36 The ball placed in the centre of the trampoline was a bowling ball with a mass of 7.26
37 kg and a radius of 10.6 cm, playing the role of a central mass or a star for planets to
38
orbit around. Since the mass from the ball itself was not enough to warp the trampoline
39
40 as much as needed, lifting weights with a combined mass of 60 kg were attached to the
41 trampoline in the centre of the ball to increase the overall deformation of the trampoline,
42
as seen in Figure 2. This figure is part of a video which records a lateral view of the
43
pte
44 trampoline that was later used for reconstruction of the shape of the trampoline, as
45 explained in the section raw data analysis.
46
47
48
49
2.2. Recording the movement of the ball
50
51
ce
The small ball, playing the role of a planet orbiting a star, was a petanque ball with
52
a radius of 3.52 cm and a mass of 0.718 kg. Different throws of the small ball on the
53
54 surface of the trampoline were performed. The ball was ejected with an angle and a
55 velocity such that it ‘orbited’ around the central mass for more than 5 seconds, which
56
was useful to record the movement allowing for the posterior analysis of the trajectory
Ac
57
58 of the ball.
59
60
1
2
3 5
4
5
6
7
8
pt
9
10
11
12
13
cri
14
15
16
17
18
19
20
us
21
22 Figure 2: Lateral view of the warped elastic bed due to the total mass of 67.3 kg. It
23
24
was assured that the lifting weights did not touch the floor
25
26
27
28
29
an
The experiments were recorded with a video camera located over the central mass posi-
tion and an approximate height of 1 meter. A video take was recorded for each different
30
ball throw. The camera used had a wide angle lens and no deviation between the images
31 (frames) taken by the camera and the real position of the ball was assumed. The videos
32 were recorded at 180 fps (frames per second). An example of a frame of the videos
dM
33
34
captured by the camera can be seen in Figure 3.
35
36
37
38
39
40
41
42
43
pte
44
45
46
47
48
49
50
51
ce
52
53
Figure 3: Example of a frame obtained from one of the ball throws. Both balls used in
54
55 the experiment can be seen
56
Ac
57
58
59
60
1
2
3 6
4
5 3. Raw Data Analysis
6
7
8 Using the raw data obtained during the experiment, both the filmed 2D trajectories
pt
9 of the small ball and an approximation for the shape of the elastic bed were found.
10 One should note that both things are necessary to fully reconstruct the experiment
11
12 and for later physical analysis: the trajectory of the rolling ball is filmed from above,
13 only giving information of its position (x, y) on the horizontal plane with respect to the
cri
14 central ball. The shape of the trampoline completes the position of the petanque ball
15
16 in 3D by providing the altitude at which the ball must be at every point (x, y).
17
18
19 3.1. Obtaining the trajectories
20
us
21
22 The position data was extracted from the 180 fps videos by analyzing them frame by
23 frame using the program Tracker Video Analysis and Modeling Tool. [10] The program
24 was used to manually track the trajectory of the centre of the small ball with respect to
25
26 the middle of the central mass, which acts as our coordinate origin (since its movement
27
28
29
an
is almost nonexistent, and therefore can be ignored). After exporting the position data,
other magnitudes such as velocity, acceleration and angular magnitudes such as distance
30 from the centre and swept angle of the ball were calculated for each take at each frame.
31 All the data was smoothed using a Savitzky-Golay filter for convenience in later use.
32 The trajectories observed are shown in Figure 4.
dM
33
34
35
36
37
38
39
40
41
42
43
pte
44
45
46
47
48
49
50
51
ce
52
53
54
55
56 Figure 4: Smoothed trajectories of the petanque ball obtained after video analysis
Ac
57
58
59
60
1
2
3 7
4
5 3.2. Measuring the shape of the trampoline
6
7 To recover the height of the trampoline, an image tool that allows to measure the
8
pt
9 positions of specific points is used (see Figure 5). Previously, the image perspective was
10 corrected using an image editor, in order to make the trampoline longer axis parallel
11 to the photo’s left to right direction. The coordinate system used along this article has
12
13 its origin in the centre of the bowling ball, x-axis and y-axis are parallel to the longer
cri
14 and the shorter axis of the trampoline respectively and the z-axis is the vertical one.
15 Therefore, in the image of Figure 5, the x-axis goes from left to right and the z-axis
16
17 goes from bottom to top.
18
19
20
us
21
22
23
24
25
26
27
28
29
an
30
31
32
dM
33
34
35 Figure 5: Perspective corrected trampoline. Red dots are the collected position points
36
37
38 Finally, heights and x-axis positions in metres can be recovered from the image
39 measurements using a reference object with a known length, such as the diameter of the
40 bowling ball. The coordinates of the measured surface points on the x-axis and z-axis
41
42 are referred as (x, z).
43
pte
44
In order to have a simple expression for the height of the trampoline, a function z = h(x)
45 is fitted to the surface experimentally recovered points. The h function found to both
46 have a simple expression and minimize the mean squared error with respect to the
47
48
measured heights has the form:
49
50 h(x) = −ke−cx with k, c > 0 (1)
51
ce
52 Where by minimizing the distance between all measured z coordinates and their
53
54
corresponding h(x), the parameters are estimated to be k = 0.44 m and c = 1.8 m−1 .
55
56
This function only provides the height along the x-axis. A first step to reconstruct the
Ac
57 full trampoline’s shape would be to consider that it has radial

p symmetry. Then the
58 expression for the complete height f would be f (x, y) = h( x2 + y 2 ). However, since
59
60
1
2
3 8
4
5 the trampoline is not circular, elliptic symmetry is more sensible. The final expression
6 for the height f becomes:
7
8 f (x, y) = h(r(x, y)) = −ke−cr(x,y)
pt
9 p (2)
10 where r(x, y) = x2 + (σy)2
11
12 With σ being a free parameter referred to as the “stretching parameter”. σ only stretches
13
in the y-axis in order to recover that the height is h(x) when y = 0. This fitted
cri
14
15 exponential function serves as a first, simple approximation for the trampoline shape.
16 Another method to approximate the shape which does not require the former symmetry
17
assumption nor the stretching parameter will be developed in the modelling section.
18
19
20
us
21 4. Exploring the spacetime rubber sheet analogy
22
23
24 As discussed previously, the spacetime rubber sheet analogy can help to develop an
25 intuition about spacetime curvature. However, a student who has been taught about
26
27
28
29
an
this analogy might assume that this analogy can be used to learn more about the basics
of GR. In that case, this student would incur in various mistakes, some of which are
exemplified in this section. Despite the theoretical backgrounds behind the movement of
30
31 the ball in the trampoline and planet movement in GR being totally different, concrete
32 examples of failure of this analogy could help in its explanation in order to reduce
dM
33 possible confusions in students. The examples provided are, on the one hand, a shape
34
35 comparison between the observed trajectories and the geodesics of the trampoline, and
36 on the other hand, a comparison between the apsidal precession of the orbits observed
37 and the predicted by GR for orbits of planets in the solar system.
38
39
40
41 4.1. Trajectory shape comparison
42
43
pte
Since trajectories in spacetime are known to be geodesics, and considering the

44
45 trampoline to play the complete role of spacetime in the analogy analized, a sensible
46 first approach the student could have in mind is analizing the trampoline geodesics.
47 Geodesics are the curves that minimize the distance between two points, and it can
48
49
be proven mathematically that geodesics in surfaces correspond to trajectories followed
50 by particles only under the action of forces which are perpendicular to the surface
51
ce
[11]. Therefore, the geodesics of the surface were calculated by simulating the path of a
52
53
particle only under the normal force of the surface, which is perpendicular to the surface
54 itself. This can be achieved using the Algorithm 2 described in the trajectory simulation
55 section, setting all forces but the normal force to zero. As it can be observed in Figure
56
6, due to the shape of the trampoline, its geodesics are always “open”, this is, no
Ac
57
58 orbits can be observed. A bounded orbit would not satisfy the property of forces being
59 perpendicular to the surface at all points. This however does not match the observed
60
1
2
3 9
4
5 trajectories, shown in Figure 4. Instead of following trampoline geodesics, orbits are
6 formed thanks to Earth’s gravity acting on the petanque ball and trapping it in the
7
trampoline deformation. Also, the orbits gradually reduce in size due to friction. Since
8
pt
9 the trampoline geodesics do not correspond to the ball trajectories over the surface, the
10 analogy cannot be applied in this manner. Assuming that particles should follow the
11 trampoline geodesics in the conditions of this experiment is an incorrect extrapolation
12
13 of the rubber sheet analogy.
cri
14
15
16
17
18
19
20
us
21
22
23
24
25
26
27
28
29
an
30
31
32
dM
33
34
35 Figure 6: Various trampoline geodesics
36
37
38
4.2. Apsidal precession comparison
39
40
41 As noted in the previous subsection, trajectories around the central ball do not follow
42
the trampoline geodesics but orbit the centre instead. It is precisely because of this
43
pte
44 fact that the analogy can be used in the first place: the visual resemblance between the
45 ball movement and planet orbits can help students to visualise the effects of spacetime
46
warping.
47
48
To study this resemblance, a comparison between the observed orbits and predicted
49
50 GR orbits such as the ones observed in the solar system is made. In particular,
51
ce
classical orbital parameters such as the period of revolution, semi-major axis, orbit
52 eccentricity and apsidal or perihelion precession are calculated, to later discuss whether
53
54 these parameters obey the relation predicted by GR, given by
55
56
L2
Ac
57 σperi ∝ (3)
58 T 2 (1 − e2 )
59
60
1
2
3 10
4
5 where we observe the dependence of the perihelion precession σperi on the parameters L
6 (semi-major axis), T (period of revolution) and e (eccentricity of the orbit). Historically,
7
the correct prediction of the apsidal precession of Mercury given by the expression above
8
pt
9 is considered a major proof of validity of the GR theory, since previous gravitational
10 theories failed to completely explain why and how Mercury’s perihelion precessed over
11 time.
12
13 The study of the distance between centres r (orbital radius) as a function of time t can
cri
14
15 be used to obtain the orbital data. The r(t) plots of takes 1 to 4 can be found in Figure
16 7, and Figure 8 represents some of the parameters we want to obtain.
17
18
19
20
us
21
22
23
24
25
26
27
28
29
an
30
31
32
dM
33
34
35
36 Figure 7: Graphical representation of r as a function of t, of takes 1 to 4
37
38
39
40
41
42
43
pte
44
45
46
47
48
49
50
51
ce
52
53
54
55 Figure 8: Some parameters of the ellipse
56
Ac
57
58 In order to obtain the parameters from the data, it is considered for each orbit, that
59
60
each interval between two consecutive maximum values of r(t) correspond to separate
1
2
3 11
4
5 “ellipses”. As in planetary orbits, we consider the bowling ball to be placed on a focus
6 of the ellipse. Also, since friction causes orbital radius to decay over time, orbital
7
parameters will slowly change while the ball is following the ellipse.
8
pt
9 Let tmin be the value in this interval where the radius is minimum and tmax1 and tmax2
10
11 be the two extremes of the interval (where r(t) attains a local maximum). Two more
12 time values for each ellipse are needed for the calculation: the two instants, tmid1 and
13 tmid2 , at which the ball is crossing the minor axis. An estimation of tmid1 and tmid2 is
cri
14
15 obtained by considering the crossing to happen when 1/4 or 3/4 of the length is travelled
16 from the aphelion, analogously to a real ellipse. The orbital parameters are calculated
17 as explained on Table 1.
18
19
20 Orbital parameter Calculation Method
us
21 Period of revolution (T ) tmax2 − tmax1
22
23
r at Perihelion (rperihelion ) r(tmin )
24 r at Aphelion (raphelion ) (r(tmax1 ) + r(tmax2 ))/2
25 r(t) at minor axis (rmid ) (r(tmid1 ) + r(tmid2 ))/2
26
27
28
29
Semi-major axis (L)
Semi-minor axis (l)
an
(r
paphelion + rperihelion )/2
2
qrmid − (L − rperihelion )
2
2
Eccentricity 1 − Ll 2
30
31 Apsidal precession |Angle diff between maxima| - 2π
32
dM
33 Table 1: Calculation methods used for different classical orbital parameters
34
35
36 The results of the ten analyzed orbit fragments or ellipses can be found in Table 2. Only
37 their mean and standard deviation are presented because their variation is relatively
38
39 small.
40
41 Orbital parameter Mean value observed Standard deviation
42 Semi-major axis (L) 0.24 m 0.04 m
43
pte
44 Semi-minor axis (l) 0.23 m 0.04 m

45 Period of revolution (T ) 1.4 s 0.2 s
46 Eccentricity (e) 0.29 0.07
47
48 Apsidal precession (σperi ) -2.1 rad 0.5 rad
49
50 Table 2: Results of the calculated orbital parameters
51
ce
52
53
54
The first interesting result is that all the apsidal precession values that are found are
55 negative. Oppositely, GR predicts them to be positive for planets. In other words,
56 unlike what happens in planet orbits maxima in radial distance repeat before a 2π
Ac
57
58
radian revolution is completed.
59
60
More concisely, the GR relation 3 predicts that if Figure 9 plot were to be made for
1
2
3 12
4
5
6
7
8
pt
9
10
11
12
13
cri
14
15
16
17
18
19
20
us
21 Figure 9: Relationship between apsidal precession and orbital parameters
22
23
24
25
planets orbiting a star, a linear relation with positive slope should be observed, which
26 is clearly not the case of the obtained experimental data.
27
28
29
an
In conclusion, the analogy is flawed when it is taken into the context of analyzing the
orbit, as the GR prediction about the relation between the studied orbital parameters
30
has been found to not apply to the ball orbits in the trampoline. This conclusion
31
32 coincides both with the theoretical differences between GR and classical mechanics
dM
33 explaining ball movement and with the results found in the literature. In [7], the
34
authors prove that no Newtonian-like orbits (excluding circular orbits) will appear on
35
36 any cylindrically-symmetric surface, and in [8], C.A. Middleton and M. Langston extend
37 this result, showing that the same inconsistency remains for orbits obtained by a full
38
GR treatment.
39
40
The results of this section indicate that taking the analogy one step further, by not
41
42 only using it to get familiarized with spacetime curvature but also using it to predict
43
pte
properties about the orbits on the rubber sheet, does not work. This is especially
44
important in teaching, where the usual explanation of this analogy could be paired with
45
46 one of the exposed examples to make sure students are aware of the limitations of the
47 analogy, and make them reflect on the differences with reality.
48
49 Apart from the aforementioned geometric intuition, one can not hope to learn much
50 more about GR by observing ball trajectories on a trampolines.
51
ce
52 As an additional comment, the rubber sheet being spacetime is not the most accurate
53
54 way to present the analogy. Instead, the sheet being a slice of just space deformed by
55 masses is a good visualization of the space geometry around a black hole, as is pointed
56 out by K. Thorne in his published lecture notes [12].
Ac
57
58
59
60
1
2
3 13
4
5 5. Physical Models
6
7
8 Having concluded in the previous section that the rubber sheet spacetime analogy does
pt
9 not match the movement of the orbiting ball in the experimental conditions, a physical
10 model to explain the observed trajectories is lacking. In this section, a model of the
11
12 trampoline shape and two models of the ball movement under the influence of the bent
13 trampoline are presented. These three models are based solely on classical mechanics.
cri
14
15 In both physical models of the ball movement presented, specifying the surface where
16 the ball moves is required. With the purpose of providing it to the model, the trampoline
17
18 surface parametrization can be obtained both as the result of the model presented in
19 this section or the approximation by a radial exponential developed previously.
20
us
21 The classical problem of the movement of rolling balls on a surface of revolution (i.e.
22 cyllindrically symmetric surface) was already studied as an analogue to Newtonian
23
24 gravity in 1968 by K. Mirenberg [13]. Later, using an elastic spandex sheet as the
25 surface, its resulting shape after deformation by a mass is studied in [6] and its theoretical
26
27
28
29
an
treatment further improved in [14] and [8], [6] also studies analogues to classical gravity
such as Kepler’s laws, finally a study of trajectories on a surface of shape r−1 , r−2 ,
r−3 is done in [7]. In the present work, a model for ball movement on a general surface
30 given by an arbitrary smooth function is presented, therefore dropping the cyllindrically
31
32 symmetric hypothesis from previous works. Also, in the rolling sphere model, the
dM
33 full effects of torque and moment of inertia are included into the formulation. The
34 trampoline shape used for the model came from the numerical simulation of another
35
36 model, also explained in this section, which approximates such shape by solving a system
37 of ordinary differential equations.
38
39
40 5.1. Trampoline shape model
41
42
43 The trampoline in this experiment can be modeled as an ideally elastic surface, where
pte
44 a known mass has been left on its middle and its rectangular boundary (see Figure 1)
45
46 has been fixed in place.
47 To be able to simulate approximately the mentioned scenario, the surface is discretized
48 into a grid of points. Newton’s Laws are applied to the grid points, which have a
49
50 really small mass (coming from the trampoline’s elastic cords) together with a great
51
ce
mass distributed in a circle in the centre, representing the bowling ball and the weights.
52 These points are united with its neighbours by springs of elastic modulus K which are
53
54 under a certain initial tension T0 when the trampoline is flat, not deformed by the central
55 mass. Both T0 and K are not known a priori, and in the following section a method
56 to adjust them computationally will be explained. The equations developed from the
Ac
57
58 stated assumptions and their complete expressions can be found in the appendix. As an
59 abridgement, the forces that act on a point with mass ∆m are tensions T and gravity,
60
1
2
3 14
4
5 which add up to a total force F~ ,
6
7 F~ = T~y− + T~y+ + T~x− + T~x+ − ∆mgẑ (4)
8
pt
9 (tensions depend on the four neighbours’ positons and g = 9.81ms−2 is the gravitational
10
11 acceleration) then the equation of motion for a point in position ~r is
12
13 F~
~r¨ = − µ~r˙
cri
14 (5)
∆m
15
16 where the dot indicates a time derivative, following Newton’s notation. A friction term
17
µ is added to the usual Newton’s Law, because in the real experiment friction occurs in
18
19 and between elastic cords, which dampens oscillations of the surface of the trampoline.
20
us
21 The goal is to determine the shape of the weight-loaded trampoline at equilibrium,
22 when it remains static. Two forces give shape to the trampoline: gravitational forces
23
and elastic forces. At equilibrium these forces compensate each other. As explained
24
25 in the following section, when implemented computationally, the system is simulated
26 until these equilibrium conditions are reached, using numerical methods to solve the
27
28
29
differential equations of the model.
an
30
31 5.2. Point particle model
32
dM
33
The first physical ball movement model consists of the following assumptions:
34
35
(i) The petanque ball is a point particle of mass m, without any moment of inertia or
36
37 radius.
38
39 (ii) The point particle can only move over the surface of the trampoline. This happens
40 because the ball is affected by gravity, which forces it to always touch the elastic
41
trampoline, as seen in the filming of the experiment.
42
43
pte
(iii) The trampoline is not deformed by the passage of the petanque ball, for the sake
44
45 of simplicity it is considered to be a rigid surface.
46
47 (iv) A friction proportional to the normal force is exerted against the particle movement.
48
49
50 Model equations The equations of motion for the punctual particle are immediately
51
ce
obtained from the force F~ exerted on it,

52
53
54 F~ = −mgẑ + F~f + F~ntramp (6)
55
56 Both friction F~f and normal F~ntramp forces are generated by the trampoline,
Ac
57
58
F~f = −µFntramp v̂, (7)
59
60
1
2
3 15
4
5 " #!
6 m ẋ
h i h i
7 F~ntramp = g + ẋ ẏ Hf (~rp ) ~
−∇f 1 (8)
~ )2 + 1
(∇f ẏ
8
pt
9
10 Where Fntramp stands for modulus of F~ntramp , v̂ for unitary velocity vector, ∇f
~ and Hf
11 for gradient and Hessian of the surface parametrization f (x, y). The positions of the
12
13 particle in the x-axis & y-axis are x & y respectively. The derivation of this equations
cri
14 is found on the appendix.
15
16
17
Model parameters The calculation of the forces and the model trajectory prediction
18
19 depend on the parameters listed on Table 3. Note that the mass of the particle m does
20 not appear in the table because its value does not affect the accelerations calculated by
us
21
the model.
22
23
24 Parameter Description Known value
25 µ Friction against velocity No
26
27
28
29
f (x, y)
[x, y](0)
an
Parametrized trampoline shape
Initial position of the particle
Estimated
Yes
[ẋ, ẏ](0) Initial velocity of the particle Yes
30
31 Table 3: Point particle model parameters
32
dM
33
34
35
36 5.3. Rolling sphere model
37
38
The second ball movement model is an improvement of the previous one while still
39
40 relying on classical Newtonian mechanics. Assumption (i), the petanque ball being a
41 point particle, is dropped, and instead it is considered to be a rigid uniform sphere of
42
radius R. Superseding it a new assumption is made:
43
pte
44
(i’) The sphere rolls on the surface of the elastic trampoline without sliding.
45
46
Assumption (ii) has to be more detailed due to the ball now occupying a volume:
47
48
(ii’) There is always a point P of the surface of the sphere that is in contact with the
49
50 trampoline, which is not the same point as the centre of mass O of the sphere. This
51
ce
is illustrated in Figure 10.

52
53 The assumption (iii) stays the same, the elastic trampoline is not deformed due to the
54 petanque ball’s weight. The assumption (iv) is extended by adding a friction torque
55
56 which diminishes the sphere’s spinning.
Ac
57
58
59
60
1
2
3 16
4
5
6
7
8
pt
9
10
11
12
13
cri
14
15
16
17
18 Figure 10: 2D slice of the sphere rolling over the trampoline. The thick black line
19 represents the trampoline surface, and the left top axis marks the coordinate origin
20
us
21
22 h i
23 Model equations Let ~rO and ~rP = xP yP zP be the positions of O and P
24 respectively. If the isurface of the trampoline is parametrized as f (x, y), then ~n(~rP )
25 h
26 = −∇f ~ (xP , yP ) 1 is a vector normal to it at point P, the corresponding unitary
an
27 ~ · n̂, the normal component of the sphere’s angular
normal vector is n̂. Then, let ωs be ω
28
29
velocity, named spin.
30
31
The defined variables ~rO , ~rP and ωs are unknown, however their first or second
32 derivatives can be expressed as:
dM
33 
34 ¨ ˙ ˙ 5 ~ ~
~rO = −(~rO · n̂)n̂ + 7 Fs + Ff − g(ẑ − (ẑ · n̂)n̂)

35

36 ~r˙P = (Id + RA)−1 ~r˙O (9)
37

Rω˙s = −~r˙O · (n̂ × n̂)
˙ − sign(ωs )ηF tramp


38 n
39
40 Where:
41
42 2
43
F~s = Rωs n̂ × n̂˙ (10)
pte
5
44 ~
45
Ff = −Crr Fntramp v̂ (11)
46 g
Fntramp = m − ~r˙O · n̂˙ (12)
47 n
48 n̂˙ = A~r˙P (13)
49  
50 0
Id − n̂ · n̂T  Hf

51
ce
A= 0 (14)

52 n

53 0 0 0
54
55
56
Here η is the spin friction coefficient, Crr is the rolling resistance coefficient and Hf
Ac
57 is evaluated at ~rp . This defines a system of ordinary differential equations which is

58 integrated to obtain the sphere trajectory. Again, the full derivation of these equations
59
60
can be found in the appendix.
1
2
3 17
4
5 Model parameters The forces and trajectories predicted by the spherical model depend
6 on the parameters listed in Table 4. As in the point particle case, the sphere mass m does
7
not appear in the table because its value does not affect the calculated accelerations.
8
pt
9 Also, the initial position of the contact point ~rP (0) does not appear because it is found
10 as the solution of the following non-linear equation (which is solved numerically when
11 implementing the model computationally):
12
13
cri
14 ~rO = ~rP + Rn̂(~rP ) (15)
15
16
17 Parameter Description Known value
18
19 Crr Resistance against rolling coefficient No
20 η Friction against spin coefficient No
us
21
ωs (0) Initial value of the ball’s spin No
22
23 f (x, y) Trampoline’s surface shape Estimated
24 ~rO (0) Initial position of the ball’s centre Yes
25
26
~r˙O (0) Initial velocity of the ball’s centre
an Yes
27 R Ball’s radius Yes
28
29 Table 4: Rolling sphere model parameters
30
31
32
dM
33 6. Parameter fitting and Model Validation
34
35
36 6.1. Algorithmic implementation of the physical models
37
38
39 The previous section proposed physical models for the shape of the trampoline and ball
40 movement, which included parameters with unknown values. In order to both validate
41 these models and to find an approximation for the unknown values, computational
42
43
simulations were performed. In this section the implementation of each of the physical
pte
44 models used to validate them is explained. These simulations were combined together in
45 order to simulate the full experiment, this is, simulating first the shape of the trampoline
46
47
after adding weight to its centre and using the result to simulate the trajectory of a ball
48 on the obtained surface. A diagram of the structure of the combined simulations can
49 be found in Figure 11. This diagram refers to the simulations done using the simulated
50
51
trampoline via the trampoline model exposed in the previous subsection 5.1, but it
ce
52 should be noted that the exponential fitting derived before (in subsection 3.2) can also
53 be used instead.
54
55
56
Ac
57
58
59
60
1
2
3 18
4
5
6
7
8
pt
9
10
11
12
13
cri
14
15
16
17
18
19
20
us
21
22
23
24 Figure 11: Program flow diagram
25
26
27
28
29
an
6.1.1. Simulating the trampoline shape Given some values for K and T0 , Algorithm 1
was used to obtain a 3-D reconstruction of the trampoline:
30
31
Algorithm 1 Trampoline Reconstruction Algorithm
32
dM
33 1: positions ← Initial positions of a 2-D Grid
34 2: velocities ← Zero for each grid point in each axis
35
3: while not in equilibrium do
36
37 4: accelerations ← Forces(positions, velocities)
38 5: velocities ← Integrate(accelerations).
39
40
6: positions ← Integrate(velocities).
41 7: F inalP ositions ← positions
42
8: ReconstructedShape ← Interpolate(F inalP ositions)
43
pte
44
45
46 where Forces refers to applying equation 5 from the trampoline shape model and
47
48
Newton’s Law to obtain the accelerations at step t given the positions and velocities
49 at the same step t, Integrate refers to applying a numerical integration method to
50 solve the differential equation to obtain the position and velocities at t + ∆t given the
51
ce
52
accelerations at t and Interpolate refers to applying an interpolation method in order to
53 be able to know the shape of the trampoline at all points given the final 3-D positions
54 of all the points in the grid. The presence of a friction term in the equations implies
55
that equilibrium will be eventually reached. In this case the specific integration and
56
Ac
57 interpolation numerical methods used were fourth order Runge-Kutta and third order
58 splines, respectively.
59
60
1
2
3 19
4
5 Note that in order to calculate the forces, the values of K and T0 need to be specified,
6 and therefore, the final shape of the trampoline depend on their values. The objective
7
is to obtain the ones that generate the most similar trampoline in the x-axis to the
8
pt
9 measurements of the real Tibidabo weight-loaded trampoline. After finding the best-
10 fit values for these two parameters, the predicted trampoline can be used as a good
11 estimation of the real one at every point in the horizontal plane, not only in the x-axis.
12
13 The optimal values for the parameters are found to be K = 112.31N and T0 = 3.28N ,
cri
14 which generate the trampoline in Figure 12. From this point, K and T0 will always be
15 considered to take these obtained optimal values.
16
17
18
19
20
us
21
22
23
24
25
26
27
28
29
an
30
31
32 (a) Trampoline (red) and real points (blue) (b) Visualization of the optimization result
dM
33
34 Figure 12: Two different views of the simulated trampoline which is most similar to the
35 real one in the x-axis cross-section
36
37
38
39 6.1.2. Simulating the trajectories For both the point particle model and the rolling
40 sphere model, ball trajectories can be simulated using the Algorithm 2. Unlike the
41
42
previous Trampoline Reconstruction algorithm, in order to obtain the full trajectory
43 information the output has to be the positions, velocities and accelerations of the ball
pte
44
45
46
47
48
49
50
51
ce
52
53
54
55
56
Ac
57
58
59
60
1
2
3 20
4
5 at each step, and not only at the last step.
6
7
Algorithm 2 Trajectory Simulation Algorithm
8
pt
9 1: positions ← Empty container of size StepsNumber
10 2: velocities ← Empty container of size StepsNumber
11
3: accelerations ← Empty container of size StepsNumber
12
13 4: positions[0] ← Initial position of the ball in each axis
cri
14 5: velocities[0] ← Initial velocity of the ball in each axis
15
6: i ← 0
16
17 7: while i ≤ StepsN umber do
18 8: accelerations[i] ← Forces(positions[i], velocities[i])
19
9: velocities[i+1] ← Integrate(accelerations[i])
20
us
21 10: positions[i+1] ← Integrate(velocities[i])
22 11: i←i+1
23
24
25
26 here, Forces refers to calculating and adding all forces acting on the ball (gravity, friction
27
28
29
an
forces, spin forces in the rolling sphere model and the normal force of the surface, which
depend on the shape of the surface), and obtain the acceleration applying Newton’s
30 Law. Both the shape of the trampoline and the model used affect the final trajectory,
31 since the model determines how to calculate the value of the forces acting on the ball,
32
dM
33 and in particular uses the shape of the trampoline to adjust the normal force. As in
34 the previous algorithm, Integrate refers to an Ordinary Differential Equation numerical
35 solver, and the fourth order Runge-Kutta method was used as well.
36
37 Using this algorithm, we can find the best values for the unknown model parameters
38
39
by simulating and optimizing how similar the simulated trajectories are to the observed
40 ones, and then conclude which of the models is able to better describe the observed
41 trajectories.
42
43
pte
44
45 6.2. Model comparison and optimization of unknown parameters
46
47 6.2.1. Fixed trampoline optimization The first optimization over the unknown
48
49
parameters of the models exposed in tables 3 and 4 aimed to do two comparisons:
50
51
(i) Comparison between the exponential trampoline (Equation 2) with k and c fitted
ce
52 to the real measures and the trampoline obtained using the trampoline shape model
53 and Algorithm 1
54
55 (ii) Comparison between the point particle model and rolling sphere model
56
Ac
57 The four possible combinations were simulated, optimizing for all the unknown
58
59
parameters, including σ from Equation 2. For each of the four takes shown previously
60
1
2
3 21
4
5 in Figure 4, the loss function minimized is the mean of the square euclidean distances
6 between observed trajectories xi and simulated trajectories x̃i at each timestep i:
7
8 1 X
N
pt
9 L= ||xi − x̃i ||2
10 N i=0
11
12
13
here, L is thought as a function of the values of the unknown parameters and its value
cri
14 is minimized using a search method combined with Nelder-Mead algorithm.
15
16 Both the trampoline obtained with the trampoline shape model and the exponentially
17 fitted trampoline achieved similar performance in all the cases. The point particle
18
model failed to determine parameters that could reproduce similar trajectories to the
19
20 ones observed, but the rolling sphere model was able to take advantage of the initial
us
21 spin parameter (which does not exist in the point particle model) to obtain a better fit
22
to the four takes. As expected, this more complex model was able to better explain the
23
24 observed trajectories. However, the spin value found in the optimal case was larger than
25 the expected real one, probably to compensate for mistakes of the model with respect
26
27
28
29
an
to the real scenario. An example of the performance of the rolling sphere model in the
best fit case can be found in Figure 13. The Root Mean Square Error (RMSE, square
root of the fitness function used) averaged over the takes achieved in this case is 0.09 m.
30
31
32
dM
33
34
35
36
37
38
39
40
41
42
43
pte
44
45
46
47
48
49
50
51
ce
52 Figure 13: Example of a predicted trajectory of the rolling sphere model over the
53 trampoline obtained using Algorithm 1, compared to the real observed trajectory
54
55
56
Ac
57 6.2.2. Free trampoline optimization Another optimization was performed, over the
58 same fitness function, this time only with the exponential trampoline (not using the
59
60
shape model trampoline) and letting c and k as parameters to be optimized, both in the
1
2
3 22
4
5 case of the point model and the rolling sphere model. The results of both models can be
6 seen in Figures 14 and 15. The RMSE achieved for each point is 0.05 m in the case of
7
the rolling sphere model and 0.07 m in the case of the point model, a slight difference.
8
pt
9 However, the c and k parameters found while optimizing the point particle model are
10 0.25 m−1 and 0.83 m, far from the real ones, while on the other hand, the parameters
11 found by using the rolling sphere model are reasonably close to the observed ones at 2.4
12
13 m−1 and 0.43 m, recalling that the observed ones were 1.8 m−1 and 0.44 m.
cri
14
15 Both facts lead to the conclusion that the rolling sphere model greatly outperforms the
16 point particle model in explaining the experiment. The obtained results could be further
17 improved by a model taking into account the possible sliding of the ball over the surface.
18
19
20
us
21
22
23
24
25
26
27
28
29
an
30
31
32
dM
33
34
35
36
37
38
39
40
41
42
43
pte
44 Figure 14: Point particle model best fit trajectories after optimization (in orange),
45 compared to the observed ones (in black)
46
47
48
49
50
51
ce
52
53
54
55
56
Ac
57
58
59
60
1
2
3 23
4
5
6
7
8
pt
9
10
11
12
13
cri
14
15
16
17
18
19
20
us
21
22
23
24
25
26
27
28
29
an
Figure 15: Rolling sphere model best fit trajectories after optimization (in blue),
compared to the observed ones (in black)
30
31
32
dM
33 7. Conclusions
34
35
36 The two main questions this article addressed were examining the rubber-sheet analogy
37
and predicting the orbit of balls rolling on a warped trampoline, by using in both cases
38
39 data collected from the experiment performed in Tibidabo amusement park.
40
41 Firstly, no way of extending the analogy accurately beyond the rubber sheet being a
42 visualization of the spacetime warping caused by masses has been found.
43
pte
44 It can not be considered as a complete spacetime where bodies follow geodesics, neither
45
are the orbits a ball follows on a warped trampoline compatible with Newtonian and
46
47 GR planetary orbits. One cannot expect to predict ball movement over the trampoline
48 using solely GR. This should be kept in mind while teaching the analogy, which has
49
been shown to be effective in GR teaching but could become misleading if one does not
50
51 clarify where it fails.
ce
52
53 Regarding the second question, two different physical models were proposed and
54 compared to describe the phenomenon. Computational simulations of both the
55
trampoline and the ball movement concluded that the proposed rolling sphere model can
56
Ac
57 accurately predict the experimental results, outperforming the other proposed model,
58 the point particle model.
59
60
1
2
3 24
4
5 Acknowledgments
6
7
8 We thank the referees for all their valuable input. We would also like to thank the
pt
9 amusement park Tibidabo for offering their facilities and believing in the FISIDABO
10 project to promote youth talent in physics. We additionally want to thank Artur Paz
11
12 for his support on the logistics of the experiment. We want to extend our thanks to
13 professor Francisco Marquès and PhD student Yago Herrera for their advice in the
cri
14 physical models, and to Sandra Wells for her proofreading of the article. Finally we
15
16 want to acknowledge the continuous efforts given by Enginyeria Fı́sica from Universitat
17 Politècnica de Catalunya.
18
19
20
us
21
References
22
23 [1] Ponnambalam M 2018 Emotional Component in Teaching and Learning The Physics Teacher 56
24 92-95
25 [2] Gupta A, Elby A and Danielak B A 2018 Exploring the entanglement of personal epistemologies
26 and emotions in students thinking Phys. Rev. Phys. Educ. Res. 14 010129
27
28
29
[3] FISIDABO webpage, fisidabo.upc.edu
an
[4] Buenda J J, Lopez H, Sanchis G and Pardo L C 2017 Modelling human behaviour in a bumper car
ride using molecular dynamics tools: a student project European Journal of Physics 38 035802
30 [5] Gurri P, Amat D, Espar J, Puig J, Gerard Jimenez G, Sendra L and Pardo L C 2017 Pendulum
31 dynamics in an amusement park European Journal of Physics 38 035005
32
dM
[6] White G D and Walker M 2002 The shape of “the Spandex” and orbits upon its surface American
33 Journal of Physics 70 48-52
34
[7] English L Q and Mareno A 2012 Trajectories of rolling marbles on various funnels American
35
Journal of Physics 80 996-1000
36
37 [8] Middleton C A and Langston M 2014 Circular orbits on a warped spandex fabric American Journal
38 of Physics 82 287-294
39 [9] Tiffany R W 2014 Gravity & Einstein: Assessing the Rubber Sheet Analogy in Undergraduate
40 Conceptual Physics Boise State University Theses and Dissertations 862.
41 [10] Tracker by Douglas Brown, found as open software in physlets.org/tracker/
42 [11] Pokorny P 2012 Geodesics Revisited Chaotic Modeling and Simulation (CMSIM) Jan 2012 281298
43
pte
[12] Thorne K 2003 Warping spacetime, published in The Future of Theoretical Physics
44 and Cosmology: Celebrating Stephen Hawkings 60th Birthday. Available on
45 www.ws5.com/spacetime/SpaceTime%20Warp.pdf
46 [13] Mirenberg K J 1968 Introduction to Gravity-Well Models of Celestial Objects
47
[14] Lemmons D S and Lipscombe T C 2002 Comment on “The shape of ‘the Spandex’ and the orbits
48
upon its surface”, by Gary D. White and Michael Walker American Journal of Physics 70
49
1056-1060
50
51 [15] Carroll S M 1997 Lecture Notes on General Relativity. Available on arXiv:gr-qc/9712019.
ce
52 [16] Kenneth Hong C M 2006 General Relativity: Warping of Space and Time. Article for GEK1508
53 Course; Einstein’s Quantum Weirdness.
54
55
56
Ac
57
58
59
60
1
2
3 25
4
5 8. Appendix
6
7
8 The mathematical derivations of the equations used to implement in a computer program
pt
9 the physical models are found in this appendix.
10
11
12
8.1. Trampoline shape simulation
13
cri
14
15 The Tibidabo trampoline is a mesh of elastic cords, which can be approximately
16 considered as a continuous elastic surface. The Physical Models section presented
17
18 a common approach to simulate such surfaces: discretizing them as a grid of point
19 masses joined by springs. In the end, the simulation is a mesh of connected springs,
20 which resembles the original trampoline even more than the intended continuous surface.
us
21
22 Crossings of the original elastic cords can be associated to the points of the grid and a
23 segment of cord between crossings to a spring. This makes the transformation between
24 the simulation and the real trampoline straightforward.
25
26
27
28
29
an
Definitions The points in the grid are indexed by (i, j), their positions are ~rij .
Following the mentioned association, the index range will be:
30
31
32 0 ≤ i ≤ Crossingsx−axis + 1 0 ≤ j ≤ Crossingsy−axis + 1
dM
33
34
35 The extreme values of both indices correspond to boundary points that will be fixed in
36 their place, at height 0.
37
38
39 Axis Number of crossings Size (m) ρcord (kg · m−1 )
40 x 214 3.80
41 0.01
y 150 1.60
42
43
pte
44
Table 5: Tibidabo trampoline characteristics. Except for the size, they are estimations
45
46 From the data in Table 5 the cord length between crossings in each axis is calculated,
47
48 ax and ay , when the trampoline is flat (no weight loaded):
49
50 3.80 3.80
ax = = 0.01767m, ay = = 0.00744m (16)
51
ce
215 151
52
53 The simulation springs acquire this length when the spring mesh is flat and tensed,
54 under the initial tension T0 , however the length of a non-deformed spring on the x-axis
55 lx or y-axis ly is unknown. To find them, consider the tension T any spring following
56
Hooke’s Law generates when deformed by an amount ∆l:
Ac
57
58
59
∆l ax − lx
T =K =⇒ T0 = K (17)
60 l lx
1
2
3 26
4
5
6
K K
=⇒ lx = ax , ly = ay . (18)
7 K + T0 K + T0
8 Where K is the Young’s modulus of the springs, which should coincide with the Young’s
pt
9
modulus of the cords. The chosen parameter is K instead of the more common k = Kl
10
11 because it is an intrinsic property of the cord, independent of the segment under study,
12 whatever its length.
13
cri
14 The last definition is that the total mass of the point (i, j) is referred as ∆mij :
15
16 M
∆mij = (lx + ly )ρcord + IC (i, j) (19)
17 nC
18
19 The first summand is meant to represent the mass of the elastic cords: four segments
20 of cord come out of every crossing, as shown in Figure 16, and half of each segment
us
21 contributes to the mass of the point. The second summand adds more mass to the
22
23 points inside a circle C in the centre of the mesh of springs, to represent the bowling
24 ball and the lifting weights. IC is a function that indicates with a 1 or a 0 whether the
25 point is in the circle C. nC is the total number of grid points inside C. The total mass
26
27
28
29
loaded on the trampoline is M = 67.3Kg.
an
30 Equations of motion To clarify the considered situation, the following figure shows a
31 close view of one of the mobile grid points:
32
dM
33
34
35
36
37
38
39
40
41
42
43
pte
44
45
46
47
48
49
50
51
ce
52
53 Figure 16: Point mass and its spring-linked neighbors
54
55
56 In the last figure, it can be seen that any tension vector T~ is parallel to its spring.
Ac
57
58 Adding up all the forces actuating on the point (i, j):
59
60
F~(i,j) = T~y−(i,j) + T~y+(i,j) + T~x−(i,j) + T~x+(i,j) − ∆mij gẑ (20)
1
2
3 27
4
5 The expression for one of the tension forces, T~x− (the one generated by the neighbour
6 (i − 1, j) ) is developed:
7
8 ~ry−(i,j) ≡ ~ri−1,j − ~ri,j
pt
9
10 |~ry (i,j) | − ly
11 Ty−(i,j) = K
ly (21)
12
13 ~ry (i,j)
T~y−(i,j) = Ty−(i,j)
cri
14 |~ry (i,j) |
15
16
17 The calculation of the three other tension forces is analogous, only the corresponding
18 indices need to be modified.
19
20 Finally, the equation of motion for each point (i, j) is:
us
21
22 F~(i,j)
23 ~rïj = − µ~r˙ij (22)
24 ∆mij
25
26 In which the added term, depending on the constant µ, is the dissipative term that
27
28
29
in the desired equilibrium shape.
an
continuously reduces the mechanical energy of the system, forcing it to eventually halt
30
The mesh of springs and point masses will be initialized in a flat disposition, and when
31
32 after a certain number of simulation steps the motions and accelerations come sufficiently
dM
33 close to zero the point masses disposition will be the estimation of the loaded Tibidabo’s
34
trampoline shape.
35
36
37
38 8.2. Point particle model
39
40
Total force F~ exerted on the point particle:
41
42
43 F~ = −mgẑ + F~f + F~ntramp (23)
pte
44
45 The first term, −mgẑ, is the gravitational force, the second, F~f , the dynamical friction
46
force, and the last, F~ntramp , the normal force coming from the elastic trampoline. The
47
48 particle cannot cross the trampoline surface because of the normal force F~ntramp , which
49 instead makes the particle stay above the surface.
50
51
ce
Friction force is considered opposite to the particle velocity ~v and proportional to the
52
normal force modulus (assumption (iv) of the point particle model):
53
54
55 F~f = −µFntramp v̂ (24)
56
Ac
57 Where v̂ indicates the unitary vector of ~v and µ is an unknown constant.

58
59 The Normal force is calculated as follows:
60
1
2
3 28
4
5 If the particle can only move on the surface (assumption (ii)), then the velocity vector
6 will have to be tangent to the surface, which is expressed as:
7
8 ~v · ~n = 0 ∀~n ∈ (TP S)⊥ (25)
pt
9
10
P is the position of the particle, then ~n will be any vector normal to the surface (denoted
11
12 by S) at the point P .
13
cri
14 A normal vector at every point is found using the implicit form of the surface:
15
16 F (x, y, z) = z − f (x, y) = 0
17
h i h i (26)
~ = −fx −fy 1 = −∇f
~n = ∇F ~ 1
18
19
20 Where fx indicates the partial derivative of f with respect to the x-axis. By always
us
21 choosing this normal vector, it is possible to take the derivative of (25) with respect to
22
23 time:
24 ~a · ~n + ~v · ~n˙ = 0 (27)
25
26 From F~ = m~a and since F~f k ~v and F~ntramp k ~n (k meaning parallel):
27
28
29 ~a · ~n = −gẑ +
an
F~f
+
F~ntramp
!
· ~n
m m
30 (28)
31 F tramp
32 = −g + n n
m
dM
33
34 By denoting Hf the Hessian of f :
35 h i h i
36 ~v · ~n˙ = ẋ ẏ ż · −∇f ~ 1 ˙
37
38
" " #" # #
h i fxx fxy ẋ
39 = ẋ ẏ ż · − ,0
40 fyx fyy ẏ (29)
41 " #
42
h i ẋ
= − ẋ ẏ Hf
43 ẏ
pte
44
45 Substituting in (27):
46
47
" #
Fntramp h i ẋ
48 −g + n − ẋ ẏ Hf =0 (30)
49 m ẏ
50 " #!
51 m ẋ
ce
h i
Fntramp = g + ẋ ẏ Hf (31)
52 n ẏ
53
54 " #!
55 m h i ẋ h i
F~ntramp = Fntramp n̂ = g + ẋ ẏ Hf ~
−∇f 1 (32)
56 ~
(∇f )2 + 1 ẏ
Ac
57
58 The resulting normal force F~ntramp depends on the shape of the surface where the particle
59 ~ and Hf .
60
moves, specifically on ∇f
1
2
3 29
4
5 When the parametrization of the surface, f (x, y), is known, gradient and the Hessian
6 of the parametrized surface can be calculated. If the firstly proposed approximation
7
of the trampoline surface is used: f (x, y) = h(r(x, y)), h(r) = kecr , r(x, y) =
8 p
pt
9 x2 + (σy)2 :
10 " # " # " #
11 f x rx
x
12
~
∇f = 0
= h (r) 0
= h (r) σr2 y (33)
fy ry
13 r
cri
14
15
16 h i
17 Hf = (∇f ~ )x (∇f~ )y
18 " # " #
19 rx
h i rxx r xy
20 = h00 (r) rx ry + h0 (r)
ry ryx ryy
us
21
22
 2 
x σ 2 xy (34)
" #
1 x2 σ 2 xy
23 00 r 2
r 0 r
− r3 r3
= h (r)  2 2 2  + h (r) 2
σ2 4 2
24 σ xy σ y σ xy
− σr3y
r2 r r3 r
25
26
" #
0 0
an
h (r) ~ ∇r
~ >+ h (r) 1 0
27 = h00 (r) − ∇r
28 r r 0 σ2
29
30
31 h0 (r) = ckecr h00 (r) = c2 kecr (35)
32
dM
33
34 Equation (23) showed every force that actuates on the point particle. Afterwards,
35 explicit expressions to calculate each force have been found: (24), (32). By applying
36 Newton’s Second Law the particle’s acceleration ~a is obtained, then, its trajectory is
37
38 predicted, given its initial conditions of position ~x(0) and velocity ~v (0) and using an
39 integrator for ~a = ~x¨.
40
41
42
8.3. Rolling sphere model
43
pte
44
45 The first step to build the equations of the rolling sphere model is to define ~rO as the
46 position of the sphere’s centre of mass and ~rP as the position of the contact point between
47
48 the surface and the sphere. The final goal is to obtain the mass centre O acceleration,
49 ~rÖ . To accomplish it, all the constraints and relations of the physical model must be
50 expressed mathematically.
51
ce
52 If n̂ is the unitary vector of ~n, which always has a positive z component and therefore
53
54 points upwards, then the following relation holds:
55
56 ~ ,
~rP = ~rO + OP ~ = −Rn̂
OP (36)
Ac
57
58 Where R is the radius of the sphere rolling over the surface. The relation can be easily
59
60
checked looking at Figure 17.
1
2
3 30
4
5
6
7
8
pt
9
10
11
12
13
cri
14
15
16
17
18
19
20 Figure 17: Diagram of the sphere rolling over the surface and its position vectors ~rO
us
21
and ~rP
22
23
24
25 Now, the model assumption (ii) can be rewritten as: the derivatives of the two position
26 vectors, i.e. the centre O velocity and the contact point P velocity, are parallel to the
27
28
29
surface:
an
30
31 ~r˙O · ~n = 0 , ~r˙P · ~n = 0 ∀~n ⊥ Surf ace (37)
32
dM
33 The assumption (i), rolling without sliding, implies that the velocity of the sphere ~vS(P )
34
35 at the point of contact P is zero (~vS(P ) does not have to be confused with the velocity
36 of the contact point ~r˙P , which is nonzero when the sphere moves). Then, rigid body
37 kinematics tell us:
38 ~ = ~vS(P ) = 0
39 ~ × OP
~vO + ω (38)
40
41
~ is the angular velocity vector and ~vO ≡ ~r˙O .
Where ω
42
43
The next equation to consider is Newton’s Second Law, for simplicity the forces are split
pte
44 between the one exerted by the elastic trampoline on P and the force exerted by the
45 gravity:
46
47
m~rÖ = F~ tramp − mgẑ (39)
48 The following equation is the so called Newton’s Second Law for rigid bodies, which
49
50 takes a simple form due to our body being a sphere:
51
ce
52 ~˙
~τ = I ω
53 (40)
~ × F~ tramp = ~τ
OP
54
55
56 Only the force exerted by the trampoline F~ tramp contributes to the torque ~τ , because
Ac
57 the gravity force acts on the centre of mass. I is the momentum of inertia, which on a
58
59 solid sphere takes the value I = 25 mR2 .
60
1
2
3 31
4
5 All relations have been considered, only algebraic manipulation is left to determine all
6 ~ , ~rÖ ). First, the variables are split into its normal and
the unknown variables (F~ tramp , ω
7
tangent components to the surface:
8
pt
9 F~ tramp = F~ntramp + F~ttramp
10
11 ω
~ =ω ~s + ω~t (41)
12
13 ~rÖ = ~an + ~at
cri
14
15
n denotes normal, s denotes spin and t denotes tangential component. To isolate the
16 ~ t and F~ttramp from the equations (38) and (40), the triple product identity
values of ω
17 will come in useful:
18
~a × (~b × ~c) = ~b(~a · ~c) − ~c(~a · ~b) (42)
19
20
us
21
22 ~ = −~r˙O
23 ~ × OP
ω
24 ~ × (~ω × OP
OP ~ ) = −OP ~ × ~r˙O
25
26 ω ~ · OP
~ (OP ~ ) − OP
~ (OP ~ ·ω ~ ) = −OP~ × ~r˙O (43)
27
28
29
ω
an
~ R2 − R2 (~ω · n̂)n̂ = Rn̂ × ~r˙O
R~ωt = n̂ × ~r˙O
30
31
32
dM
33 OP ~˙
~ × F~ tramp = I ω
34
35
~ × (OP
OP ~ × F~ tramp ) = OP
~ × Iω~˙
36 (44)
−R2 F~ttramp = I(OP~ ×ω ~˙ )
37
38 2 ~ ) = 2 mR(n̂ × ω
F~ttramp = m(ω ~˙ × OP ~˙ )
39 5 5
40
41 When the cross product is taken at both sides, a component of the vectors is lost, the
42 ~ . Therefore a scalar equation for each vectorial equation (38) and (40)
one parallel to OP
43
pte
was forgotten when the vectorial product was done. They can be recovered by scalarly
44 ~ = −Rn̂.
45
multiplying both sides by a vector parallel to OP
46
47
~r˙O · n̂ = ω ~ · n̂ = 0
~ × OP
(45)
48 Iω ~˙ · n̂ = OP
~ × F~ tramp · n̂ = 0
49
50 The first equation was already known, it is a reformulation of the condition of staying
51
ce
in contact with the trampoline (37), but the second one is new. They tell that both ~r˙O
52
53 and ω~˙ do not have a normal component, they are tangent or parallel to the surface.
54
55 To know the value of F~ttramp explicitly, the angular acceleration ω ~˙ is needed. The tangent
56 component ω~ t is already known, and its derivative is:
Ac
57 !
58 d n̂ × ~
r˙O n̂˙ × ~r˙O n̂ × ~rÖ
59 ~˙ t =
ω = + (46)
dt R R R
60
1
2
3 32
4
5 The normal component cannot be explicitly known, but its derivative can still be
6 calculated:
7
8 ~˙ · n̂ + ω
ω˙s = (~ω · n̂)˙ = ω ~ · n̂˙ = ω~ · n̂˙
(47)
pt
9 ~˙ s = (ωs n̂)˙ = ω˙s n̂ + ωs n̂˙ = (~ω · n̂)n̂
ω ˙ + ωs n̂˙
10
11 It can be found that ω ~ · n̂˙ = ω ˙ which requires less information to be known. This
~ t · n̂,
12
13 identity happens thanks to the fact that n̂ · n̂˙ = 0 because the unitary vector n̂ has a
cri
14 constant modulus. Now, it is possible to compute the value of F~ttramp explicitly:
15
2
16 F~ttramp = mR(n̂ × ω ~˙ )
17 5
18 n̂ × ω~˙ = n̂ × (ω ~˙ s + ω
~˙ t )
19 !
20 ˙ × ~r˙O n̂ × ~rÖ
n̂
= n̂ × (~ω · n̂)n̂˙ + ωs n̂˙ + (48)
us
21
+
R R
22 !
23 ˙ × ~r˙O n̂ × ~rÖ
n̂
24 = n̂ × ωs n̂˙ + +
25
R R
26
27
28
29
the centre’s acceleration ~rÖ :
an
To simplify the previous expression it will be useful to use the tangent ~at component of
n̂ × (n̂ × ~rÖ ) = −~at

30
31 n̂ × (n̂˙ × ~r˙O ) = n̂(
˙ ~r˙O · n̂) − ~r˙O (n̂ · n̂)
˙ =0
32
dM

33 ~ tramp 2 ˙ ~at (49)
=⇒ Ft = mR ωs n̂ × n̂ −
34 5 R
35 2 2 2
36 = mRωs n̂ × n̂˙ − m~at =: F~s − m~at
37 5 5 5
38 Substituting in the 2nd Newton’s Law (39) restricted to the tangent space:
39
40 m~at = F~ttramp − (mgẑ)t
41 2
42 = F~s − m~at − mg(ẑ)t (50)
43
5
pte
44
5 h~ i
= Fs − mg(ẑ)t , (ẑ)t = ẑ − (ẑ · n̂)n̂
45 7
46
Thus, the net effect from the tangent force produced by the trampoline F~ttramp , the
47
48 force that appears in order to change the rotation of the ball when necessary, is that
−1
49 all the tangent forces get their effect reduced by 5/7 = mR I
2 + 1 plus an extra force
50 ~ I ˙
Fs = R ωs n̂ × n̂ which is proportional to the value of the spin ωs and to how the surface
51
ce
52
˙
is changing, n̂ × n̂.
53
54 The normal component of the acceleration is calculated, equally to the point particle
55 model, taking the derivative from the condition of moving inside the surface ~r˙O · n̂ = 0:
56
Ac
57
58 ~rÖ · n̂ + ~r˙O · n̂˙ = 0
59 (51)
60 an = −~r˙O · n̂˙
1
2
3 33
4
˙
Now, the acceleration of the centre O is expressed as a function of ωs , n̂ and n̂:
5
6
7
˙ + 2 Rωs n̂ × n̂˙ − 5 g(ẑ − (ẑ · n̂)n̂)
~rÖ = −(~r˙O · n̂)n̂ (52)
7 7
8
pt
9 The value of ωs can be numerically integrated from its derivative, which only depends
10 on ω ˙
~ t and n̂:
11
12 n̂ × ~r˙O ˙
~ t · n̂˙ =
ω̇s = ω · n̂
13 R (53)
cri
14 ˙
Rω̇s = −~rO · (n̂ × n̂) ˙
15
16 However, to continue a problem is found: n̂ and n̂˙ depend on ~rP and ~r˙P , but only ~rO
17
and ~r˙O can be obtained from ~rÖ integration. The derivative of equation (36) allows
18
19 obtaining ~r˙P from ~r˙O as the solution of a matrix linear system, (as n̂˙ depends linearly
20 of ~r˙P ):
us
21
~r˙P + Rn̂(~
˙ rP , ~r˙P ) = ~r˙O (54)
22
23 Then, ~rP can be obtained from ~r˙P through integration and finally the problem is solved
24
since all the required values to calculate ~rÖ are known. Before concluding, to solve the
25
26 system of linear equations it was needed to explicitly know them:
27
28
29
n̂˙ =
d
dt n
an

~n ~n˙
= − 2 ṅ
n n
~n
30 ˙~n · ~n
31 0 = n̂˙ · ~n = − ṅ =⇒ ṅn = ~n˙ · ~n (55)
32 n
dM
˙ ˙

33 ~
n T

~
n ~
n Id − n̂ · n̂
34 n̂˙ = − ~n˙ · ~n 3 = t
= ~n˙
n n n n
35
36 The derivative of the normal vector, ~n˙ , can be recovered from calculations in the point
37
particle model part of the appendix:
38 " # 
39
   
ẋP 0 ẋP
40 ˙~n =  Hf   Hf ˙ ˙
ẏP  =  0 ~rP Given ~rP =  ẏP  (56)
 
41 
42 0 0 0 0 żP
43
pte
44 The linear system becomes explicitly:

45
  
0
46 T

Id − n̂ · n̂  Hf
~r˙O = ~r˙P + Rn̂˙ = Id + R 0 ~r˙P (57)
 
47
n

48 0 0 0
49
50
51
Although it has not been needed yet, the value of the normal force exerted by the
ce
52 trampoline F~ntramp can be calculated and it has exactly the same value as in the point
53 particle model:
54
55 m~rÖ · n̂ = (F~ tramp − mgẑ) · n̂
56
man = FN − mgẑ · n̂
Ac
57 (58)
58 gẑ · ~n g
59
~
Fntramp
=m + an n̂ = m − ~r˙O · n̂˙ n̂
n n
60
1
2
3 34
4
5 Extra Considerations On the developed model, the friction force between the
6 trampoline and the ball has been ignored. If the ball rolled ideally, the friction with the
7
trampoline surface would be zero, as they are stationary one to the other at the point
8
pt
9 of contact P. However, the rolling is not ideal and the contact between the surface and
10 the sphere is not a point, the contact area is broader because the trampoline deforms to
11 accommodate the ball. This broader contact allows a friction force or rolling resistance
12
13 to appear, this force does not generate torque and as a simple approximation is opposed
cri
14 to velocity and proportional to the normal force modulus F~f = −Crr Fntramp v̂, as in the
15 point particle model.
16
17 Another friction force that was not considered during the development of the model and
18
19 is added a posteriori is the friction force on the spin ωs , which makes a spinning ball
20 stop. It is also caused due the contact zone between the sphere and the bed not being
us
21 pointlike. This friction can be implemented adding a term to the derivative ω˙s , that
22
23 will be proportional to Fntramp and oppose to the spin:
24
25 Rω˙s = −~r˙O · (n̂ × n̂)
˙ − sign(ωs )ηF tramp
n (59)
26
27
28
29
an
This spin friction is not considered to generate a net force on the centre of mass of
the ball and only creates a torque on the ball in the n̂ direction. Therefore, it does not
30 interfere with the torque generated by the tangential force of the trampoline OP ~ × F~ttramp
31 which is orthogonal to n̂. It is clarifying to note that the value of the tangential torque
32
~˙ )t or equivalently by: the
dM
~ × F~ttramp is determined completely by (ω
~τt = ~τ − (~τ · n̂)n̂ = OP
33
34 imposition that the sphere rolls without sliding and the value of ωs · n̂. ˙ Then, adding
35 this spin friction ad hoc does not change the already deduced value of F~ttramp .
36
37 Moreover, air could also create friction against the movement of the ball. That friction
38
39 would not depend on the normal force and instead be proportional or quadratic with
40 respect to velocity. This type of friction is not considered in this model.
41
42 To conclude, the ordinary differential equation (ODE) system that should be solved is
43
pte
the one expressed in the Physical Models section, in equation (9) and its subsequent
44
45 equations. The parameters on which the ODE depends are listed on Table 4 of the same
46 section.
47
48
49
50
51
ce
52
53
54
55
56
Ac
57
58
59
60

Exploring The Rubber Sheet Spacetime Analogy by Studying Ball Movement in A Bent Trampoline

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Exploring The Rubber Sheet Spacetime Analogy by Studying Ball Movement in A Bent Trampoline

Загружено:

Авторское право:

Доступные форматы

European Journal of Physics

Exploring the rubber sheet spacetime analogy by studying ball

Manuscript version: Accepted Manuscript

This Accepted Manuscript is © 2018 European Physical Society.

View the article online for updates and enhancements.

This content was downloaded from IP address 148.212.19.20 on 18/04/2019 at 15:00

52 1.1. The FISIDABO educational project

57 full trampoline’s shape would be to consider that it has radial

Since trajectories in spacetime are known to be geodesics, and considering the

44 Semi-minor axis (l) 0.23 m 0.04 m

obtained from the force F~ exerted on it,

is illustrated in Figure 10.

57 is evaluated at ~rp . This defines a system of ordinary differential equations which is

57 Where v̂ indicates the unitary vector of ~v and µ is an unknown constant.

n̂ × (n̂ × ~r¨O ) = −~at

44 The linear system becomes explicitly:

Вам также может понравиться