You are on page 1of 19

Acta Mech 208, 97–115 (2009)

DOI 10.1007/s00707-008-0134-3

Slaviša Šalinić

Contribution to the brachistochrone problem with Coulomb


friction

Received: 26 March 2008 / Revised: 12 November 2008 / Published online: 8 January 2009
© Springer-Verlag 2008

Abstract This paper formulates and solves in closed form (expressed by elementary functions) the
brachistochrone problem with Coulomb friction of a particle which moves down a rough curve in a uni-
form gravitational field assuming that the initial velocity of the particle is different from zero. The problem is
solved by the application of variational calculus. Two variants are considered: first, the initial position and the
final position of the particle are given; second, the initial position is given, and the final position lies on a given
vertical straight line. The new approach in treating this problem by variational calculus lies in the fact that the
projection sign of the normal reaction force of the rough curve onto the normal to the curve is introduced as the
additional constraint in the form of an inequality. This inequality is transformed into an equality by introducing
a new state variable. Although this is fundamentally a constrained variational problem, by further introducing
a new functional with an expanded set of unknown functions, it is transformed into an unconstrained problem
where broken extremals appear. Brachistochrone equations in parametric form are obtained for both variants
which are examined, with the slope angle of the tangent to the brachistochrone being taken as the parameter.
These equations contain a certain number of unknown constants which are determined from the corresponding
systems of nonlinear algebraic equations. They are solved by an alternative approach which is based on the
application of differential evolution. The obtained brachistochrones are generally two-segment curves with
the initial line segment representing a free-fall parabola in nonresistant medium. It is shown that regarding the
special values of the parameters the results of the paper coincide with the known results from literature.

1 Introduction

In 1696 Johann Bernoulli formulated the following problem: find a smooth curve down which a particle slides
from rest at a point A to a point B in a vertical plane influenced by its own gravity in the least time. The
formulated problem was solved by the Bernoulli brothers (Johann and Jakob), Newton, Leibniz, Huygens,
and L’ Hospital. It was found that a cycloid is the curve with the minimum descent time. In literature this
problem is well known as the classical brachistochrone problem or Bernoulli’s case of the brachistochrone,
and its solution provided a powerful stimulus to the development of variational calculus. Further research
connected to the brachistochrone problem developed in two directions. On the one hand a classical formulation
was applied and the field of forces in which a particle moves [1,11,14,23–25] was made more complex, and on
the other hand the broadening of the classical brachistochrone problem was applied to systems of particles and
rigid bodies [4,5,7,20,21]. In [15] the problem of optimization of a bobsled traveling on a path was solved as
the problem of a brachistochronic motion of a particle on a surface. The brachistochrone problem of a particle
on a rough surface was investigated in [6] .The formulation of the brachistochrone problem in this paper is the
same as in [14]: a particle moves in a uniform gravitational field in a vertical plane down a rough curve with

S. Šalinić (B)
Faculty of Mechanical Engineering, University of Kragujevac, Dositejeva 19, 36000 Kraljevo, Serbia
E-mail: salinic.s@ptt.rs; salinic.s@mfkv.kg.ac.rs
98 S. Šalinić

Coulomb friction while the initial velocity of the particle is different from zero. In this paper the formulated
problem is solved by the application of variational calculus. Two variants are examined in this paper: in the
first one the initial position and the final position of the particle are given, and in the second variant the initial
position is given, and the final position lies on a given vertical straight line. The paper [14] examined the first
variant by the application of the theory of the singular optimal control. In the references [1,11] both the above-
mentioned variants were examined, but under the condition that the initial velocity of the particle is equal to
zero while the projection sign of the normal reaction force of the rough curve onto the normal to the curve was
supposed, as this projection appears as the absolute value in the expression for the magnitude of the friction
force. There, this assumption was not considered to be an additional constraint. Here, in this paper it will be
shown that joining this assumption as a new constraint with the existing constraints of the problem leads, for
the initial velocity different from zero, to a generalization of the results from [1,11]. A part of the results from
this paper which refers to the first variant coincides with those from [14]. During the process of finding the
solution for the system of nonlinear algebraic equations an alternative approach is applied which comes down
to the transformation of the problem of determining the roots of the system of equations to the problem of the
optimization of a corresponding cost function. This optimization is done by using the differential evolution
[18]. The advantage of this procedure of solving a system of nonlinear equations lies in the fact that it is not
limited by the complexity of nonlinear systems and it does not require initial guesses of the roots, which is not
the case when using the usual numerical methods. Reference [3] determines an optimal shape of the gravity
flow discharge chute for granular material so that the time spent on the transport of the material is minimal.
This being taken into account, the brachistochrone equations of this paper represent an optimal chute shape in
analytical form for initial velocities which in general are different from zero. In this sense the results of this
paper represent improved results from [3] where the optimal chute shape was obtained numerically for the
initial velocity equal to zero and some simplifications connected to the effect of Coulomb friction force.

2 Formulation of the problem within the theory of variational calculus

A particle M of mass m is examined while moving in a vertical plane in a uniform gravitational field upon a
rough curve with Coulomb friction which is treated as a bilateral constraint.This means that the particle must
slide along the curve like a bead on a wire. The case of frictionless unilateral constraint, when the particle
must slide along the curve like a block on an inclined plane, is considered in [19].The shape of this curve
(brachistochrone) is required so that the particle M starting from the position M0 (x0 , y0 ) with the initial speed
v0  = 0 reaches the position O(0, 0) for the minimum descent time T , where y represents the vertical axis
directed upwards, and x is the horizontal axis of a Cartesian coordinate system. Without loss of generality it
is taken that x0 > 0 and y0 > 0.
The differential equation of the motion of the particle M is
ma =mg + N + Fµ , (1)
where N is the normal component of the constraint reaction force, Fµ = −µ |N| v/ |v| is the Coulomb friction
force, µ is the coefficient of friction, g = −g j, g is the acceleration of gravity, v and a are the velocity and
acceleration of the particle, respectively. In Fig. 1b it is noted that the unit vector t of the tangent to the particle
path can be written as t = (− cos ϕ)i + (− sin ϕ)j where ϕ is the slope angle of the tangent, and i and j are the
unit vectors of axes x and y, respectively. According to this dt/dϕ = (sin ϕ)i + (− cos ϕ)j, so it is obvious
that |dt/dϕ| = 1 and t · (dt/dϕ) = 0 is valid. If v represents the projection of v onto t, then it follows that
d dt
a= (vt) = v̇t+v ϕ̇ ,
dt dϕ
where an overdot denotes the derivative with respect to time t. Projecting Eq. (1) on directions t and dt/dϕ
gives
m v̇ = mg sin ϕ − µ |Nn | , mv ϕ̇ = mg cos ϕ + Nn , (2.1, 2)
where Nn = N · (dt/dϕ).
In the initial stage of problem solving, the required brachistochrone curve should be treated as a curve
which in the general case changes its concavity. Since the motion of the particle on a curve is considered as
a motion of a bead on a wire, there are two possibilities for the orientation of vector N on the parts of the
Contribution to the brachistochrone problem with Coulomb friction 99

y (x0,y0) y (x0,y0) y

N Fµ
n
ϕ
M M M,m
dt/dϕ
t v
j v j mg
x x x
O O i O i
(a) (b) (c)
Fig. 1 a General situation, b unit vectors on the particle path, c free body diagram

curve which are concave downwards. On these parts of the curve, the vector N may be directed to the opposite
side from the vector g or to the same side of it. These possibilities of orientation of the vector N represent the
reason to introduce the vector dt/dϕ . Namely, it is obvious that dt/dϕ, in contrast to the unit normal vector n,
does not change the orientation with changing of the concavity of the curve and it is constantly directed to the
same side as g (see Fig. 1b). Taking this into account, if it is assumed that during the course of brachistochrone
motion of the particle the vector N does not change its orientation with respect to g; it is more appropriate to
use the projection Nn than to use the projection of vector N on n because Nn will not change sign during the
changing of concavity of the curve.
Due to the presence of the absolute value bars on Nn in Eq. (2.1) an analysis of the projection sign must be
done. In connection with this, in forthcoming considerations it is assumed that the normal component of the
constraint reaction force is continuously directed opposite to the gravitational force (see Fig. 1c), that is
Nn ≤ 0. (3)
The correctness of this assumption is based on the condition that the results obtained through it for µ = 0
should be reduced to the results which are valid for the classical brachistochrone problem, and for µ  = 0 and
v0 = 0 to the results from [1,11].
Taking into account the assumption (3), eliminating Nn for Nn  = 0 from Eqs. (2) gives
f 1 ≡ v̇ − g sin ϕ + µ(−v ϕ̇ + g cos ϕ) = 0. (4)
Figure 1 shows that the following nonholonomic constraints hold:
f 2 ≡ ẋ + v cos ϕ = 0, f 3 ≡ ẏ + v sin ϕ = 0. (5)
The condition Nn ≤ 0, based on Eq. (2.2), gives another constraint in the problem examined:
v ϕ̇ − g cos ϕ ≤ 0. (6)
Introducing a new unknown function ẇ(t) (see [22]), the inequality (6) is transformed into the equality

f 4 ≡ v ϕ̇ − g cos ϕ + ẇ 2 = 0. (7)
The posed problem can be formulated as the variational calculus problem
T
dt −→ inf (8)
0

with the constraints (4), (5) and (7), while the initial and the final conditions are
t0 = 0 : x(0) = x0 , y(0) = y0 , v(0) = v0 ; t = T : x(T ) = 0, y(T ) = 0. (9)
According to [10], introducing a new functional
T  
4

J= 1+ λi f i dt, (10)
0 i=1
100 S. Šalinić

where λi are Lagrange multipliers of constraints, and the following transformation of the state variables:
z 1 ≡ x, z 2 ≡ y, z 3 ≡ ϕ, z 4 ≡ v, ż 5 ≡ λ1 , ż 6 ≡ λ2 , ż 7 ≡ λ3 , ż 8 ≡ ẇ, ż 9 ≡ λ4 , (11)
the posed constrained variational problem can be reformulated as the following unconstrained variational
problem:
T

J = 1 + ż 5 [ż 4 − µz 4 ż 3 + g(µ cos z 3 − sin z 3 )] + ż 6 (ż 1 + z 4 cos z 3 ) + ż 7 (ż 2 + z 4 sin z 3 )
0

+ ż 9 (z 4 ż 3 − g cos z 3 + ż 82 ) dt → inf,

t0 = 0 : z 1 (0) = x0 , z 2 (0) = y0 , z 4 (0) = v0 , z i (0) = 0 (i = 5, . . . , 9),


t = T : z 1 (T ) = 0, z 2 (T ) = 0. (12)
The stationarity condition J = 0 yields Euler–Lagrange equations (see [8,9]):
   
d ∂L ∂L d ∂L
= 0, k = {1, 2, 5, 6, 7, 8, 9}, − = 0, j = 3, 4, (13)
dt ∂ ż k ∂z j dt ∂ ż j
as well as boundary conditions:
  T
  T
∂L 
9
∂L

z i = 0, i = 1, . . . , 9, L− ż i t = 0, (14)
∂ ż i 0 ∂ ż i
i=1 0
where (·) represents the noncontemporaneous variation [8,9,17] of the quantity (·), and L represents the
integrand of the functional J .
Based on Eqs. (14), the fact that the quantity z 3 is not given in the initial and final position of the particle
((z 3 )0  = 0, (z 3 )T  = 0) yields the following natural boundary conditions:
   
∂L ∂L
= 0 → (−µvλ1 + vλ4 )0 = 0, = 0 → (−µvλ1 + vλ4 )T = 0, (15)
∂ ż 3 0 ∂ ż 3 T
and the fact that the quantities z i (i = 4, . . . , 9) are not given in the final position of the particle ((z i )T  = 0,
i = 4, . . . , 9) yields another six natural boundary conditions of the form
 
∂L
= 0 → (λ1 )T = 0, (16)
∂ ż 4 T
 
∂L
= 0 → (v̇ − µv ϕ̇ + g (µ cos ϕ − sin ϕ))T = 0, (17)
∂ ż 5 T
   
∂L ∂L
= 0 → (ẋ + v cos ϕ)T = 0, = 0 → ( ẏ + v sin ϕ)T = 0, (18)
∂ ż 6 T ∂ ż 7 T
   
∂L ∂L
= 0 → (ẇλ4 )T = 0, = 0 → (v ϕ̇ − g cos ϕ + ẇ 2 )T = 0, (19)
∂ ż 8 T ∂ ż 9 T
where the notations (·)0 and (·)T indicate that the subscripted quantities are evaluated at t = 0 and t = T ,
respectively. Since the time T is not specified ((t)T  = 0), from Eqs. (14) follows the transversality condition
at the final position of the particle
 

9
∂L
L− ż i =0 (20)
∂ ż i
i=1 T
or, in a developed form, after elimination of the quantities ż 1 , ż 2 , ż 3 and ż 4 in the last relation by using Eqs. (4),
(5) and (7)

1 + λ1 g(µ cos ϕ − sin ϕ) + v(λ2 cos ϕ + λ3 sin ϕ) − λ4 (ẇ 2 + g cos ϕ) T = 0. (21)
Contribution to the brachistochrone problem with Coulomb friction 101

3 Solutions of the Euler–Lagrange equations

Taking into account the conditions (17)–(19), the Euler–Lagrange equation (13) in the original notations of
the unknown functions can be represented on the interval [0, T ] in the form
λ̇2 = 0, λ̇3 = 0,
−gλ1 (µ sin ϕ + cos ϕ) − vλ2 sin ϕ + vλ3 cos ϕ + µv̇λ1 + µv λ̇1 + gλ4 sin ϕ − v̇λ4 − v λ̇4 = 0,
−µϕ̇λ1 + λ2 cos ϕ + λ3 sin ϕ − λ̇1 + ϕ̇λ4 = 0,
v̇ − µv ϕ̇ + g (µ cos ϕ − sin ϕ) = 0,
ẋ + v cos ϕ = 0, ẏ + v sin ϕ = 0,
ẇλ4 ≡ 0, v ϕ̇ − g cos ϕ + ẇ 2 = 0. (22.1–9)
Equation (22.8) implies that on the interval [0, T ] an extremal of functional J can consist of the line
segments on which ẇ ≡ 0, λ4  ≡ 0, and of line segments on which ẇ  ≡ 0, λ4 ≡ 0. According to this,
the functional J belongs to the group of functionals whose extremals may have breaks (the so-called corner
points) in the space R9 . If S is used to represent the total number of corner points of the extremal with the
corresponding time instants t p : ( p = 1, . . . , S), then in these points Weierstrass–Erdmann corner conditions
[8,9] must be satisfied:
   
∂L ∂L
= , i = 1, . . . , 9; p = 1, . . . , S, (23)
∂ ż i t p −0 ∂ ż i t p +0
   
 9
∂L 9
∂L
L− ż i = L− ż i , p = 1, . . . , S. (24)
∂ ż i ∂ ż i
i=1 t p −0 i=1 t p +0

By observing Eqs. (22) it is noted that the conditions (23) ( i = 5, . . . , 9; p = 1, . . . , S) are satisfied iden-
tically. Since the integrand L is not explicitly dependent on time, Eqs. (22) in regard to (20) have the first
integral of the form (see, e.g. [9])

9
∂L
L− ż i = 0 ∀t ∈ [0, T ] (25)
∂ ż i
i=1

so that the conditions (24) are satisfied identically. The quantities x, y, and v are continuous at the corner
points of the extremal of the functional J , which means that
x(t p − 0) = x(t p + 0), y(t p − 0) = y(t p + 0), v(t p − 0) = v(t p + 0). (26.1–3)
In regard to Eq. (26.3), the Weierstrass–Erdmann conditions (23) (i = 3, 4; p = 1, . . . , S) are reduced to the
form
λ1 (t p − 0) = λ1 (t p + 0), λ4 (t p ) = 0, p = 1, . . . , S. (27)

3.1 General solution on the line segment ẇ ≡ 0, λ4  ≡ 0, i.e. for Nn ≡ 0

From Eqs. (22.1) and (22.2) it follows that


λ2 = const. = Cλ∗2 , λ3 = const. = Cλ∗3 , (28)
and a suitable combination of Eqs. (22.5) and (22.9) with using the transformation d(·)/dt = (d(·)/dϕ)ϕ̇
yields
g cos ϕ dv
ϕ̇ = , = v tan ϕ, (29)
v dϕ
wherefrom integrating Eq. (29.2) leads to
Cv
v= , (30)
cos ϕ
102 S. Šalinić

where Cv is an integration constant. By using the transformation d(·)/dt = (d(·)/dϕ)ϕ̇, Eqs. (22.6) and (22.7)
along with the expressions (29.1) and (30) give the following differential equations:
dx Cv2 dy Cv2 dt Cv
=− , =− tan ϕ, = , (31)
dϕ g(cos ϕ)2 dϕ g(cos ϕ)2 dϕ g(cos ϕ)2
whose general solutions read
Cv2 Cv2
x = C x∗ − tan ϕ, y = C y∗ − (tan ϕ)2 , (32)
g 2g
Cv
t= tan ϕ + Ct∗ , (33)
g
where C x∗ , C y∗ , and Ct∗ are integration constants. Equation (32) represents parametric equations of a free-fall
parabola in a nonresistant medium. By using Eqs. (29.1), (30) and the transformation d(·)/dt = (d(·)/dϕ)ϕ̇,
the equations ϕ̇ · (22.3) and (22.4) yield
d Cv (−λ2 sin ϕ + λ3 cos ϕ)
(λ4 − µλ1 ) = −λ1 + ,
dϕ g(cos ϕ)2
dλ1 Cv (λ2 cos ϕ + λ3 sin ϕ)
= (λ4 − µλ1 ) + . (34)
dϕ g(cos ϕ)2
Solving the obtained system of differential equations for λ1 and λ4 −µλ1 yields the following general solution:
Cv Cv
λ4 − µλ1 = Cλ1 cos ϕ − Cλ4 sin ϕ + λ3 sin ϕ − λ2 sin ϕ tan ϕ,
g g
Cv Cv
λ1 = Cλ1 sin ϕ + Cλ4 cos ϕ + λ2 sin ϕ + λ3 sin ϕ tan ϕ, (35)
g g
where Cλ1 and Cλ4 are integration constants. Combining the equations in (35) gives
Cv Cv
λ4 = (cos ϕ +µ sin ϕ)Cλ1 +(µ cos ϕ − sin ϕ)Cλ4 + λ3 sin ϕ(1 + µ tan ϕ)+ λ2 sin ϕ(µ−tan ϕ). (36)
g g

3.2 General solution on the line segment ẇ  ≡ 0, λ4 ≡ 0, i.e. for Nn  0

The first two equations in (22) yield


λ2 = const. = Cλ2 , λ3 = const. = Cλ3 . (37)

Solving Eq. (22.4) for ϕ̇ and Eq. (22.3) for λ1 v̇ + v λ̇1 and successively substituting the expressions for these
quantities into (22.5) gives

v [λ2 (sin ϕ − µ cos ϕ) − λ3 (cos ϕ + µ sin ϕ)] + λ1 g cos ϕ(1 + µ2 ) = 0. (38)


Further, Eqs. (25) and (38) can be treated as a system of equations in the unknowns λ1 and v whose solution is
1 sec ϕ 1 sec ϕ (tan ϕ + A)
v=− , λ1 = , (39)
B 2µ (tan ϕ + A) − (sec ϕ)2 g (sec ϕ)2 − 2µ (tan ϕ + A)
where
µλ2 + λ3 µλ3 − λ2
A= , B= . (40)
µλ3 − λ2 1 + µ2
Now, as a result of the above, from Eq. (22.5) it follows that
 2
(cos ϕ)2 (tan ϕ − µ)2 + 1 − µ2 − 2µA
ϕ̇ = −g B . (41)
(tan ϕ − µ)2 + 1 − µ2 + 2µA + 4µ tan ϕ
Contribution to the brachistochrone problem with Coulomb friction 103

By using the transformation d(·)/dt = (d(·)/dϕ)ϕ̇, Eqs. (22.6) and (22.7) along with the expressions (39.1)
and (41) give relations between the coordinates x and y and the angle ϕ in differential form:
 
dx (tan ϕ − µ)2 + 1 − µ2 + 2µA + 4µ tan ϕ
=  3 , (42)
dϕ g B 2 (cos ϕ)2 (tan ϕ − µ)2 + 1 − µ2 − 2µA
 
dy (tan ϕ − µ)2 + 1 − µ2 + 2µA + 4µ tan ϕ tan ϕ
=  3 , (43)
dϕ g B 2 (cos ϕ)2 (tan ϕ − µ)2 + 1 − µ2 − 2µA
and after integration

  
1 −1 tan ϕ − µ a2 (tan ϕ − µ) 2µ(a22 + (tan ϕ − µ)a1 )
x= b1 tan + −  2 + C x (44)
2g B 2 a2 (tan ϕ − µ)2 + a22 a22 (tan ϕ − µ)2 + a22

 
1 −1 tan ϕ − µ a2 (tan ϕ − µ) 1
y= b2 tan + −
2g B 2 a2 (tan ϕ − µ) + a2
2 2 (tan ϕ − µ)2 + a22
  
2µ (a22 + µa1 ) (tan ϕ − µ) + a22 (µ − a1 )
−  2 + Cy, (45)
a22 (tan ϕ − µ)2 + a22

where a1 = −(A + µ), a2 = 1 + µ2 + 2µa1 , b1 = (a22 − 3µa1 )/a25 , and b2 = µ(2a22 − 3µa1 )/a25 . The
integration of Eq. (41) yields the dependence of t upon the angle ϕ,
    
1 −1 tan ϕ − µ 2µ a1 (tan ϕ − µ) + a22
t = Ct − b3 tan − 2  , (46)
gB a2 a2 (tan ϕ − µ)2 + a22

where b3 = (a22 − 2µa1 )/a23 .

4 The arrangement of the line segments on the extremal

Let SG1 and SG2 be denotations for line segments on which ẇ ≡ 0, λ4  ≡ 0 and ẇ  ≡ 0, λ4 ≡ 0, respectively,
and let N SG1 and N SG2 be total numbers of line segments SG1 and SG2 on the extremal of the functional J .
In regard to Eqs. (28), (37) and the conditions (23) (i = 1, 2; p = 1, . . . , S) the following holds:
λ2 ≡ const., λ3 ≡ const. ∀t ∈ [0, T ]. (47)
The arrangement of the line segments on the extremal has to satisfy the condition that the total number of
unknown integration constants C x∗ , C y∗ , Cv , Cλ1 , and Cλ4 on the line segments SG1, integration constants C x
and C y on the line segments SG2, the unknown values of angles ϕ p = ϕ(t p ) p = 1, . . . , S, and the unknowns
λ2 and λ3 is equal to the number of available conditions for them to be determined.
If the extremal of the functional J has the beginning and the ending line segment of the type SG1, then
in that case holds N SG1 = S/2 + 1 and N SG2 = S/2. Based on Eqs. (9), (30) and (32) the integration
constants C x∗ , C y∗ , and Cv can be replaced on the beginning line segment by the functional dependencies on
unknown ϕ0 = ϕ(0), and on the ending segment by the functional dependencies on unknowns ϕ f = ϕ(T ) and
v f = v(T ). Now, in regard to (47), there is a total of 9 + S + 5(N SG1 − 2) + 2N SG2 = 4 + 9S/2 unknowns, and
for them to be determined there are available the natural boundary conditions (15) and (16) and the conditions
(26) and (27), which in total amounts to 3 + 5S of conditions. For this variant, the equality 4 + 9S/2 = 3 + 5S
gives that S = 2.
In the case that the beginning line segment is of the type SG1 and the ending of the type SG2, it is valid
that N SG1 = N SG2 = S/2 + 1/2. As in the previous case, the constants C x∗ , C y∗ , and Cv are replaced on the
beginning line segment by the functional dependencies on the unknown ϕ0 , while the constants C x and C y on
the ending line segment based on Eqs. (44) and (45) are replaced by the functional dependencies on ϕ f , λ2 ,
and λ3 . Now, in regard to this and Eq. (47), 5/2 + 9S/2 unknowns should be determined. As in this case the
conditions (15.2) and (16) are equivalent to the determining of the unknowns, at disposal there are 2 + 5S
conditions. For this variant, from the equality 5/2 + 9S/2 = 2 + 5S it follows that S = 1. In a similar manner
104 S. Šalinić

to the above, it can be shown that in the case of the beginning line segment of the type SG2 and the ending
line segment of the type SG1, S = −1 holds, and for the beginning and the ending line segment of the type
SG2, S = −2 holds. Due to the negative values of S these variants do not make a physical sense. It should
be noted that in the last two variants the initial condition v(0) = v0 groups into the available conditions for
determining the unknown constants.
It will further be proved that in general the variant

ẇ(t) ≡ 0, λ4 (t)  ≡ 0 ∀t ∈ [0, t1 ],


ẇ(t)  ≡ 0, λ4 (t) ≡ 0 ∀t ∈ (t1 , t2 ], (48)
ẇ(t) ≡ 0, λ4 (t)  ≡ 0 ∀t ∈ (t2 , T ]

is reduced to the variant

ẇ(t) ≡ 0, λ4 (t)  ≡ 0 ∀t ∈ [0, t1 ],


(49)
ẇ(t)  ≡ 0, λ4 (t) ≡ 0 ∀t ∈ (t1 , T ].

Indeed, based on v f = v(T ) and the boundary conditions (15.2) and (16), from Eqs. (30), (35) and (36) it is
obtained that

v f λ3 sin ϕ f v f λ2 sin ϕ f
Cv = v f cos ϕ f , Cλ1 = − , Cλ4 = − . (50)
g g

Now, based on Eq. (26.3) ( p = 2), the conditions (27) ( p = 2) are brought into the homogeneous system of
equations in the unknowns λ2 and λ3 . For the nontrivial solution of this system of equations it is necessary
and sufficient that the determinant of its coefficient matrix is equal to zero,

v 2f (sin(ϕ2 − ϕ f ))2
= 0, (51)
g 2 (cos ϕ2 )2

wherefrom it follows that ϕ2 = ϕ f must hold due to the fact that the case v f = 0 does not make physical sense.
Finally, the required brachistochrone in general represents a two-segment curve along which the arrangement
of the line segments is defined by Eq. (49). The remaining constants Ct and Ct∗ as well as the descent time
T are easily determined from Eqs. (33) and (46) by means of the conditions t (ϕ0 ) = 0, t (ϕ f ) = T , and
t (ϕ1 − 0) = t (ϕ1 + 0).

5 Particular solutions

In further considerations the subscripts − and + indicate that the subsequent expressions are valid on the line
segments SG1 and SG2, respectively. Based on the natural boundary conditions (16) and v+ (ϕ f ) = v f ,
Eq. (39) yields

cos ϕ f
A = − tan ϕ f , B= . (52)
vf

Further, including these relations into Eqs. (39) and (40) yields

1 + µ tan ϕ f µ − tan ϕ f
λ2 = − , λ3 = , (53)
v f sec ϕ f v f sec ϕ f

1 sec ϕ tan ϕ − tan ϕ f v f sec ϕ f sec ϕ
λ1+ = , v+ = . (54)
g (sec ϕ) − 2µ tan ϕ − tan ϕ f
2
(sec ϕ) − 2µ tan ϕ − tan ϕ f
2
Contribution to the brachistochrone problem with Coulomb friction 105

Taking into account the previous results, for x+ (ϕ f ) = 0, y+ (ϕ f ) = 0, and t+ (ϕ f ) = T from Eqs. (44)–(46)
the following particular solutions are obtained:

   
v 2f −1 tan ϕ − µ a2 (tan ϕ − µ) −1 a1 a1 a2
x+ = b1 tan + − tan − 2
2g(cos ϕ f )2 a2 (tan ϕ − µ)2 + a22 a2 a1 + a22

2µ(a22 + (tan ϕ − µ)a1 ) 2µ
−  2 + 2 2 , (55)
1 + a2
2
a2 (tan ϕ − µ) + a2
2 2 2 a 2 a

   
v 2f −1 tan ϕ − µ a2 (tan ϕ − µ) −1 a1 a1 a2
y+ = b2 tan + − tan − 2
2g(cos ϕ f )2 a2 (tan ϕ − µ)2 + a22 a2 a1 + a22
  
1 2µ (a22 +µa1 ) (tan ϕ −µ)+a22 (µ−a1 ) 1 2µ2
− −  2 + 2 2+ 2 2 , (56)
(tan ϕ −µ)2 + a22 a22 (tan ϕ −µ)2 +a22 a1 +a2 a2 a1 + a22
        
vf −1 tan ϕ −µ −1 a1 2µ a1 (tan ϕ −µ)+a22 2µ
t+ = T − b3 tan −tan − 2  + 2 , (57)
g cos ϕ f a2 a2 a2 (tan ϕ − µ)2 +a22 a2

To give Eqs. (55)–(57) a physical sense, the constant a2 has to be a real nonzero number, and this request gives
the following condition:

1 − µ2 + 2µ tan ϕ f > 0. (58)

For the interval [0, t1 ] and x− (ϕ0 ) = x0 , y− (ϕ0 ) = y0 , v− (ϕ0 ) = v0 , λ1− (ϕ0 ) = (λ1 )0 , and t− (ϕ0 ) = 0 ,
where (λ1 )0 is unknown, Eqs. (30), (32), (33), (35) and (36) give the following particular solutions:
v0 cos ϕ0
v− = , (59)
cos ϕ
v 2 (cos ϕ0 )2
x− = x0 − 0 (tan ϕ − tan ϕ0 ), (60)
g
v 2 (cos ϕ0 )2  
y− = y0 − 0 (tan ϕ)2 − (tan ϕ0 )2 , (61)
2g
v0 cos ϕ0
t− = (tan ϕ − tan ϕ0 ), (62)
g
v0  
λ1− = (sin ϕ0 − cos ϕ0 tan ϕ) cos(ϕ − ϕ f ) − µ sin(ϕ − ϕ f ) + (λ1 )0 cos(ϕ − ϕ0 ), (63)
vf g
v0 (1 + µ2 )
λ4− = sec ϕ sin (ϕ − ϕ0 ) sin ϕ − ϕ f + (λ1 )0 [µ cos(ϕ − ϕ0 ) − sin(ϕ − ϕ0 )] . (64)
vf g

From the nature of the posed problem it follows that 0 ≤ v < ∞, and keeping in mind the condition (58)
the denominator in the expression (54.2) is a positive one for all values of ϕ. Based on these facts from the
expressions (54.2) and (59) it is obtained that ϕ must be limited by

− π/2 ≤ ϕ ≤ π/2. (65)

This condition has a clear geometrical confirmation in Fig. 1b. Taking into account the condition (65), when
calculating the values of the function tan−1 (·) in the expressions (55), (56) and in the further considerations,
its principal branch should be used.
The unknowns ϕ0 , ϕ f , ϕ1 , v f and (λ1 )0 must satisfy the conditions (26.1–3) and (27) which now may be
written in the following form:

x− (ϕ1 ) = x+ (ϕ1 ), y− (ϕ1 ) = y+ (ϕ1 ), v− (ϕ1 ) = v+ (ϕ1 ),


λ1− (ϕ1 ) = λ1+ (ϕ1 ), λ4− (ϕ1 ) = 0. (66)
106 S. Šalinić

Table 1 Numerical values of the parameters of brachistochrone curves (v0 = 4 m/s)

µ ϕ0 ϕ1 ϕf v f (m/s) T (s)/T f (s)


0.1 0.58531127245408 0.84117455638771 0.22401024995530 6.73721842520438 0.660383/0.638551
0.3 0.28235958342084 0.78027892637736 0.30540274812492 5.77055051965695 0.704483/0.638551
0.5 0.10575791620352 0.77885729850727 0.39319970497597 5.15099759679747 0.738284/0.638551
0.7 −0.01373921733802 0.79564939443358 0.47323884833173 4.73388728568082 0.765107/0.638551
0.9 −0.09998594135303 0.81725104456084 0.54112903164358 4.43219537358280 0.786945/0.638551

Table 2 Numerical values of the parameters of brachistochrone curves (µ = 0.4)

v0 (m/s) ϕ0 ϕ1 ϕf v f (m/s) T (s)/T f (s)


2 0.24509068455450 0.91899705086443 −0.01004439564158 3.22263322540435 1.044554/0.638551
4 0.18453836454896 0.77585076195059 0.34970876377582 5.42815823131868 0.722413/0.638551
6 0.24973096435396 0.72875067194272 0.59261088139213 7.81923211356930 0.531104/0.638551
8 0.34913093489764 0.73378407987148 0.72957850623775 10.12665194266618 0.417899/0.638551

It can be easily shown that Eqs. (66.4) and (66.5), in regard to (66.3), represent a homogeneous system
of equations in the unknowns 1/v f and (λ1 )0 . For the nontrivial solution of this system of equations it is
necessary and sufficient that the determinant of its coefficient matrix is equal to zero,
 
v0 cos ϕ f [cos 2 (ϕ0 − ϕ1 ) − 2µ cos ϕ0 sin (ϕ0 − 2ϕ1 ) − 1] − 2µ (cos ϕ1 )2 sin ϕ f
= 0. (67)
2g(cos ϕ1 )2
According to the condition t− (ϕ1 ) = t+ (ϕ1 ), the descent time T is given by
        
vf −1 tan ϕ1 − µ −1 a1 2µ a1 (tan ϕ1 − µ) + a22 2µ
T = b3 tan − tan − 2  + 2
g cos ϕ f a2 a2 a2 (tanϕ1 − µ)2 + a22 a2
v0 cos ϕ0
+ (tan ϕ1 − tan ϕ0 ). (68)
g
In the case of v0  = 0, Eqs. (66.1–3) and (67) are used to determine the unknowns ϕ0 , ϕ f , ϕ1 and v f , and
then the obtained parameter values are substituted into Eqs. (66.4) or (66.5) to determine the unknown (λ1 )0 .
For x0 = π m, y0 = 2 m, v0 = 4 m/s, g = 9.81 m/s2 and various values of the friction coefficient µ, as well
as for x0 = π m, y0 = 2 m, µ = 0.4, g = 9.81 m/s2 and various values of the initial speed v0 , by numerically
solving the equation system (66.1–3) and (67), the values of the parameters ϕ0 , ϕ f , ϕ1 and v f are obtained.
These values are represented in Tables 1 and 2. Based on these data and Eqs. (55), (56), (60) and (61) in
Fig. 2 the brachistochrone curves are shown. The points (x(ϕ1 ), y(ϕ1 )) are marked by dotted lines in Fig. 2. In
Fig. 2a it can be seen that a decreasing value of the friction coefficient µ is also accompanied by decreasing a
portion of the parabolic line segment on the brachistochrone, which is an anticipated result taking into account
that for µ = 0 the brachistochrone represents a cycloid. By analyzing the data in Table 2 it is noted that by
increasing the values of the initial speed v0 , the value of the angle ϕ1 tends to the value of the angle ϕ f , that is,
the brachistochrone tends to become a parabola. In Sect. 7 the critical value of v0 for which ϕ1 = ϕ f holds will
be determined. Also, the values of the angle ϕ0 in Table 1 show that for a certain value of friction coefficient
µ it can be obtained that ϕ0 = 0. Indeed, by putting ϕ0 = 0 and taking the friction coefficient µ as unknown
instead of ϕ0 in Eqs. (66.1–3) and (67), it is found for x0 = π m, y0 = 2 m, v0 = 4 m/s, g = 9.81 m/s2 that
µ = 0.67318121365721. In order to obtain a certain √ notion about the descent time T , in Tables 1 and 2 a
comparison of time T with the free-fall time T f = 2y0 /g of the particle from the position M0 (x0 , y0 ) to the
x-axis is shown. It is interesting to observe that in Tables 1 and 2 the time T is very close to T f and for certain
values of µ and v0 even less than T f !

6 Reachability conditions

Based on the expression (52.2) and the condition (65) it can be concluded that the quantity ϕ̇+ , given by
expression (41), has the opposite sign of the term
(tan ϕ − µ)2 + 1 − µ2 + 2µA + 4µ tan ϕ ⇔ (tan ϕ)2 + 2µ tan ϕ + 1 − 2µ tan ϕ f . (69)
Contribution to the brachistochrone problem with Coulomb friction 107

y [m]
(π m, 2 m)
2 v0= 4 m/s

µ=0.1
1.5 µ=0.3

µ=0.5

1 µ=0.7
µ=0.9

0.5

x [m]
0
0.5 1 1.5 2 2.5 3
(a)
y [m]
(π m, 2 m)
2 µ=0.4

v0= 6 m/s
1.5 v0= 8 m/s

1
v0= 2 m/s
v0= 4 m/s

0.5

x [m]
0
0.5 1 1.5 2 2.5 3
(b)
Fig. 2 Coulomb friction brachistochrones for: a various values of the friction coefficient µ, b various values of the initial speed
v0

The expression (69) represents a quadratic function of tan ϕ. Taking into account the condition (58), the ana-
lysis of the sign of the expression (69) yields that for (µ2 − 1)/(2µ) < tan ϕ f < (1 − µ2 )/(2µ) we have
ϕ̇+ < 0 for all ϕ ∈ [−π/2, π/2]. Further, for tan ϕ f ≥ (1 − µ2 )/(2µ) it stands that
        
π −1 −1 π
ϕ̇+ < 0 ∀ϕ ∈ − , tan −µ − µ − 1 + 2µ tan ϕ f
2 ∪ tan −µ + µ − 1 + 2µ tan ϕ f ,
2 ,
2 2
      
ϕ̇+ > 0 ∀ϕ ∈ tan−1 −µ − µ2 − 1 + 2µ tan ϕ f , tan−1 −µ + µ2 − 1 + 2µ tan ϕ f .

Based on the conducted analysis of the sign of the quantity ϕ̇+ , it follows that there is physical sense only in
the case when
ϕ1 > ϕ f . (70)
In regard to the above, it holds that ϕ̇+ < 0.
Similar to [1], multiplying Eq. (4) by vdt and integrating over the interval (t1 , T ] yields

v 2f ϕ f
v2 y1
− 1 + gx1 (µ − ) = µ v 2 dϕ, (71)
2 2 x1
ϕ1

where x(ϕ1 ) = x1 , y(ϕ1 ) = y1 and v(ϕ1 ) = v1 . Since it was previously shown that ϕ decreases on the interval
(t1 , T ], the right-hand side of Eq. (71) is negative so the quantities ϕ0 , ϕ1 , ϕ f and v f must have such values
so that the left-hand side of Eq. (71) is negative, that is, according to Eqs. (66.1)–(66.3) it is necessary for the
following condition to be satisfied:
108 S. Šalinić

v 2f − v02 + 2g(µx0 − y0 ) + 2µv02 cos ϕ0 sec ϕ1 sin(ϕ0 − ϕ1 ) < 0. (72)


Furthermore, based on Eqs. (29.1) and (65) the angle ϕ increases along the line segment SG1. Hence, it holds
that
ϕ1 > ϕ0 . (73)
The relations (58), (70), (72) and (73) represent conditions which must be satisfied by the parameters ϕ0 ,
ϕ1 , ϕ f and v f in order to have a brachistochrone, i.e. so that the final position of the particle would be reachable.
In the general case the system of equations from which the parameters ϕ0 , ϕ1 , ϕ f and v f are determined may
have more than one solution. In regard to this, based on the conditions (58), (70), (72) and (73), the solution
which makes physical sense, that is the solution which corresponds to the required brachistochrone stands out.

7 Special cases

For ϕ1 = ϕ0 , the whole of the brachistochrone represents the line segment SG2. Hence, Eqs. (55), (56) repre-
sent the parametric equations of the brachistochrone. From Eq. (66.5) it follows that (λ1 )0 = 0. Further, Eq.
(66.4) shows that ϕ0 = π/2 since ϕ0 = ϕ f is not possible because it was shown in Sect. 6 that ϕ decreases
along the line segment SG2. Finally, for ϕ0 = π/2 from Eq. (54.2) v0 = 0 is obtained. The case of the
brachistochrone for ϕ0 = π/2 and v0 = 0 is examined in detail in [1,11,23]. For this case it follows from the
condition (72) that the initial point (x0 , y0 ) must not find its position below the straight line y = µx.
For ϕ1 = ϕ f the whole of the brachistochrone represents a free-fall parabola in a nonresistant medium.
For this case it follows from (67) that ϕ f = ϕ0 + tan−1 µ. Now, Eqs. (66.1) and (66.2) yield the following
relations:
µv02 µv02 [µ + 2 tan ϕ0 − µ(tan ϕ0 )2 ]
x0 = , y0 = , (74)
g(1 − µ tan ϕ0 ) 2g(1 − µ tan ϕ0 )2
which lead to the quadratic equation
µx0 (tan ϕ0 )2 − 2(µy0 + x0 ) tan ϕ0 + 2y0 − µx0 = 0

whose roots are (tan ϕ0 )1,2 = y0 /x0 + 1/µ ± y02 /x02 + 1 + 1/µ2 . Substituting these roots in Eq. (74.1)

it is obtained that v02 = −gy0 ∓ y02 g 2 + (1 + 1/µ2 )x02 g 2 , wherefrom, due to the assumption of y0 > 0,
the following expressions for the value of the initial slope angle and the critical value of the initial speed are
obtained:
     

y 1 y 2
1  y02 1 y0
ϕ0 = tan −1 0
+ − 0
+ 1 + 2 , (v0 )cr = gx0  +1+ 2 − (75)
x0 µ x02 µ x02 µ x0

when the whole of the brachistochrone has the shape of a free-fall parabola in a nonresistant medium. The
expressions in Eq. (75) coincide with the appropriate expressions in [14] where they are given in dimensionless
form. Substituting ϕ1 = ϕ f = ϕ0 + tan−1 µ into Eq. (68), the first term in Eq. (68) vanishes, and taking into
account (75) the expression for the descent time T becomes



   2
 x 1 + y0 + 1 − y02 + 1 + 1
 0 µ x02 µ2
 x0
T =    . (76)
 y02 y0
g x2
+ 1 + 1
µ2
− x0
0

Hence, in the case of v0 = (1 − )(v0 )cr , where 0 <   1, the time T can be approximately calculated by
using Eq. (76).
The previous presentation clearly states that in the case of v0 > (v0 )cr along the extremal of the functional
J it is valid that ẇ ≡ 0 for all t ∈ [0, T ], that is, the parametric equations of the brachistochrone are deter-
mined by (60) and (61). In other words, in this case λ4 ≡ 0 for all t ∈ [0, T ] cannot be applied, since based
Contribution to the brachistochrone problem with Coulomb friction 109

on Eq. (54.1) and the natural boundary conditions (15.1) and (16) it would be valid that A = − tan ϕ0 and
A = − tan ϕ f , which is equivalent to the relation ϕ0 = ϕ f . The elimination of the parameter tan ϕ from Eqs.
(60) and (61) by means of Eq. (62) yields

gt 2
x− = x0 − v0 t cos ϕ0 , y− = y0 − v0 t sin ϕ0 − . (77)
2
Substituting t = T , x− (T ) = 0 and y− (T ) = 0 into (77) yields a system of equations in the unknowns ϕ0
and T whose set of solutions is

−v02 x0 ± v04 x02 − g 2 x04 + 2gv02 x02 y0 x0 
(tan ϕ0 )1,2 = , (T )1,2 = 1 + (tan ϕ0 )21,2 . (78)
gx02 v0

Here, the subscripts 1 and 2 in Eqs. (78) correspond to the signs − and + , respectively, in front of the square
root in the expression (78.1). The obtained set of solutions shows that there are two parabolas as the candidates
for the brachistochrone. It is taken that the brachistochrone is the parabola with the least time T . It is noted that
T1 > T2 so that for v0 > (v0 )cr the parametric equations of the brachistochrone are determined by Eqs. (60)
and (61), where
⎛  ⎞
−v02 x0 + v04 x02 − g 2 x04 + 2gv02 x02 y0
ϕ0 = tan−1 ⎝ ⎠, (79)
gx02

 ⎛  ⎞2


x0  −v 2 x + v 4 x 2 − g 2 x 4 + 2gv 2 x 2 y
0 0 0 0 0 0 0 0
T = 1 + ⎝ ⎠ . (80)
v0 2
gx0

Substituting the expression (75.2) into (80) yields the expression (76), which means that for the case of
v0 = (1 + )(v0 )cr , where 0 <   1, the descent time T can be approximately calculated by using the
expression (76).
In the case of µ = 0 and v0 = 0, from Eq. (66.3) it follows that ϕ1 = π/2 so that, as in the case of µ  = 0
and v0√= 0, ϕ1 = ϕ0 = π/2. In the examined case from Eqs. (66.1) and (66.2) it follows that the final speed
v f = 2gy0 , which is in accordance with the law of conservation of mechanical energy, as well as the relation
x0 = y0 [(π/2 − ϕ f )/(cos ϕ f )2 − tan ϕ f ] based on which the value of the angle ϕ f is determined. In regard
to the previously determined parameters, Eqs. (55) and (56) get the following form:

v 2f v 2f
x(ϕ) = [2(ϕ − ϕ f ) + sin 2ϕ − sin 2ϕ f ], y(ϕ) = (cos 2ϕ f − cos 2ϕ), (81)
4g(cos ϕ f )2 4g(cos ϕ f )2
and they represent parametric equations of a cycloid, while the descent time of the particle from the position
M0 to the position O is

2y0 1  π
T = − ϕf . (82)
g cos ϕ f 2

When µ = 0 and v0  = 0, Eq. (67) results in cos 2(ϕ0 − ϕ1 ) = 1, that is, ϕ1 = ϕ0 . In this case this means
that Eqs. (81) are valid, while the unknowns v f , ϕ0 and ϕ f are determined from Eqs. (66.1)–(66.3), which are
reduced to the form
v 2f
(cos 2ϕ f − cos 2ϕ0 ) = y0 ,
4g(cos ϕ f )2
v 2f
[2(ϕ0 − ϕ f ) + sin 2ϕ0 − sin 2ϕ f ] = x0 ,
4g(cos ϕ f )2
v f cos ϕ0 − v0 cos ϕ f = 0. (83)
110 S. Šalinić

For the values v0 = 4 m/s, g = 9.81 m/s2 , x0 = π m and y0 = 2 m, numerically solving the equation
system (83) yields ϕ0 = 1.01490886543894, ϕ f = 0.19775875427255
 and v f = 7.43236167042481 m/s.
The obtained value for v f coincides with the value of v f = v02 + 2gy0 , which is derived from the law of
conservation of mechanical energy.

8 The modification of the brachistochrone problem

Further discussion presents a modification of the brachistochrone problem from Sect. 2 which may be stated as
follows: determine an equation of the rough guidance so that the particle starting from the position M0 (x0 , y0 )
with initial speed v0  = 0 reaches the y-axis for the minimum descent time T . Since now y(T ) = y f is not
given, there holds (z 2 )T  = 0, so that one more natural boundary condition of the form
 
∂L
= 0 ⇐⇒ (λ3 )T = 0 (84)
∂ ż 2 T

should be added to the natural boundary conditions (15)–(19).


An analogous approach as in Sect. 4 may be used to show that in this case the brachistochrone is a two-
segment curve with the segment arrangement defined by (49). Considerations from Sect. 3 are left unchanged,
and particular solutions from Sect. 5 are simplified. Based on Eq. (47) and the natural boundary condition (84),

λ3 ≡ 0 ∀t ∈ [0, T ] (85)

holds. Based on the last relation, from Eq. (40) it follows that

λ2
A = −µ, B=− , (86)
1 + µ2

and then, based on Eqs. (47) and (52),



1 + µ2
λ2 ≡ − ∀t ∈ [0, T ], ϕ f = tan−1 µ. (87)
vf

Taking into account the previous results, Eqs. (54)–(57) are reduced to the form

1 sec ϕ (tan ϕ − µ) v f 1 + µ2 sec ϕ
λ1+ = , v+ = , (88)
g (sec ϕ)2 − 2µ (tan ϕ − µ) (sec ϕ)2 − 2µ (tan ϕ − µ)

  
1 + µ2 v 2f ⎨ 1 −1 tan ϕ − µ 1 + µ 2 (tan ϕ − µ)
x+ =  3 tan  +
2g ⎩ 1 + µ2 (tan ϕ − µ)2 + 1 + µ2
1+µ 2

2µ 2µ
− 2 + 2 , (89)
(tan ϕ − µ)2 + 1 + µ2 1 + µ2

  
1 + µ2 v 2f ⎨ 2µ −1 tan ϕ − µ 1 + µ2 (tan ϕ − µ)
y+ = yf +  3 tan  +
2g ⎩ 1 + µ2 (tan ϕ − µ)2 + 1 + µ2
1 + µ2

(tan ϕ − µ)2 2µ tan ϕ 2µ2
+  − 2 + 2 , (90)
1 + µ2 (tan ϕ − µ)2 + 1 + µ2 (tan ϕ − µ)2 + 1 + µ2 1 + µ2

 
1 + µ2 v f 1 tan ϕ − µ 2µ (tan ϕ − µ)2
t+ = T−  tan−1  +  . (91)
g 1 + µ2 1 + µ2 (tan ϕ − µ)2 + 1 + µ2 (1 + µ2 )
Contribution to the brachistochrone problem with Coulomb friction 111

Table 3 Numerical values of the parameters of brachistochrone curves in the modified brachistochrone problem (v0 = 4 m/s)

µ ϕ0 ϕ1 y f (m) v f (m/s) T (s)/T f (s)


0.1 0.50389329351951 0.77252567345527 0.36858252086529 6.18630232397962 0.656797/0.576717
0.3 0.27211071350183 0.77232235733837 0.04307437457487 5.69981384006614 0.704428/0.631637
0.5 0.15993406120446 0.81881600321527 −0.22547700029464 5.55929706359072 0.736630/0.673584
0.7 0.09509292479804 0.87268020562648 −0.45057257369388 5.64953563557170 0.757846/0.706829
0.9 0.05538079545021 0.92333491937119 −0.63474985900535 5.90284079514762 0.771090/0.732909

Table 4 Numerical values of the parameters of brachistochrone curves in the modified brachistochrone problem (µ = 0.4)

v0 (m/s) ϕ0 ϕ1 y f (m) v f (m/s) T (s)/T f (s)


2 0.47110662220799 1.07004595651011 −1.05269009368873 5.28994154521823 0.993055/0.788900
4 0.20777161393425 0.79339962396564 −0.09695728766761 5.59502470188276 0.722122/0.653846
6 0.05582652535646 0.56310014048209 0.73877466739483 6.74964525879815 0.519436/0.507080
8 0.00311624714332 0.41592403130681 1.24560767454141 8.62913334381805 0.392666/0.392174

On the interval [0, t1 ] Eqs. (59)–(62) stay unchanged, and Eqs. (63) and (64) are reduced to

v0 1 + µ 2
λ1− = (λ1 )0 cos (ϕ − ϕ0 ) − sin(ϕ − ϕ0 ), (92)
gv f

v0 1 + µ 2
λ4− = (λ1 )0 [µ cos(ϕ − ϕ0 ) − sin(ϕ − ϕ0 )] + (tan ϕ − µ) sin (ϕ − ϕ0 ) . (93)
gv f
Now, Eq. (67) is reduced to
 
v0 cos 2 (ϕ0 − ϕ1 ) − 2µ cos ϕ0 sin (ϕ0 − 2ϕ1 ) − 2µ2 (cos ϕ1 )2 − 1
 = 0. (94)
2g(cos ϕ1 )2 1 + µ2
For x0 = π m, y0 = 2 m, v0 = 4 m/s, g = 9.81 m/s2 and various values of the friction coefficient µ,
as well as for x0 = π m, y0 = 2 m, µ = 0.4, g = 9.81 m/s2 and various values of the initial speed v0 , by
numerically solving the equation system (66.1–3) and (94) the values of the parameters ϕ0 , ϕ1 , y f and v f are
obtained. These values are represented in Tables 3 and 4. Based on this data and Eqs. (60), (61), (89) and (90)
the brachistochrone curves are represented in Fig. 3. The points (x(ϕ1 ), y(ϕ1 )) are marked by dotted lines in
Fig. 3. Similar to Sect. 5, the time T is given by

 
1 + µ2 v f 1 −1 tan ϕ1 − µ 2µ (tan ϕ1 − µ)2
T =  tan  + 
g 1 + µ2 1 + µ2 (tan ϕ1 − µ)2 + 1 + µ2 1 + µ2
v0 cos ϕ0
+ (tan ϕ1 − tan ϕ0 ). (95)
g

As in Sect. 5, in Tables 3 and 4 a comparison of time T with the free-fall time T f = 2(y0 − y f )/g of
the particle from the position M0 (x0 , y0 ) to the horizontal straight line y = y f is shown. Based on the values
in Tables 3 and 4 it can be observed that the values of T and T f are very close.

8.1 Reachability conditions

The analysis of the reachability of the final position in this variant of the brachistochrone problem is further
conducted in the same way as in Sect. 6. Having in mind that now ϕ f = tan−1 µ, the term defined in (69)
is reduced√to the form (tan ϕ)2 + 2µ tan ϕ + 1 − 2µ2 . The analysis
√ of the sign of this term gives that for
0 ≤ µ < 3/3, ϕ̇+ < 0 for all ϕ ∈ [−π/2, π/2]. Further, for 3/3 ≤ µ ≤ 1 we have
$ π      π%
ϕ̇+ < 0 ∀ϕ ∈ − , tan−1 −µ − 3µ2 − 1 ∪ tan−1 −µ + 3µ2 − 1 , ,
2 2
    
ϕ̇+ > 0 ∀ϕ ∈ tan−1 −µ − 3µ2 − 1 , tan−1 −µ + 3µ2 − 1 .
112 S. Šalinić

y [m]
v0= 4 m/s µ=0.7 (π m, 2 m)
2 µ=0.9
µ=0.1
1.5 µ=0.3
µ=0.5

0.5

x [m]
0
0.5 1 1.5 2 2.5 3
-0.5 (a)

y [m] µ=0.4
v0= 6 m/s (π m, 2 m)
2 v0= 8 m/s

1.5 v0= 2 m/s

v0= 4 m/s
1

0.5
x [m]
0
0.5 1 1.5 2 2.5 3
-0.5
(b)
-1

Fig. 3 Coulomb friction brachistochrones for modified brachistochrone problem for: a various values of the friction coefficient µ,
b various values of the initial speed v0

Based on the conducted analysis of the sign of the quantity ϕ̇+ and the fact that ϕ f = tan−1 µ, it follows that
there is physical sense only in the case when ϕ1 > tan−1 µ. In regard to above, it holds that ϕ̇+ < 0. Hence,
the analogous procedure as in Sect. 6 leads to the conditions of reachability defined by the relations (70), (72)
and (73).

8.2 Special cases

For ϕ1 = ϕ0 the whole of the brachistochrone represents the line segment SG2 and therefore, the parametric
equations of the brachistochrone are determined by Eqs. (89) and (90). In this case, the analogous procedure
as in Sect. 7 gives ϕ0 = π/2 and v0 = 0. This case is examined in [1,11].
For ϕ1 = ϕ f from Eq. (94) it is obtained that v0 (µ − tan ϕ0 ) sin 2ϕ0 = 0, which gives, after the case
ϕ0 = ϕ f = tan−1 µ has been ruled out, that ϕ0 = 0 must be valid. According to this, the brachistochrone
in this case represents a free-fall parabola with a horizontal initial velocity in a nonresistant medium. Now,
Eqs. (66.1) and (66.2) yield the critical value of the initial speed and the value y f of the y-coordinate of the
final position of the particle:

gx0 µx0
(v0 )cr = , y f = y0 − . (96)
µ 2
Substituting ϕ1 = ϕ f = tan−1 µ and ϕ0 = 0 into Eq. (95), the first term in Eq. (95) vanishes and the expression
for the descent time T becomes

µx0
T = . (97)
g
Similar to Sect. 7 it is shown that the brachistochrone for v0 > (v0 )cr represents a free-fall parabola in a
nonresistant medium and that Eqs. (77) are valid. Substituting t− = T ∗ and x− (T ∗ ) = 0 into (77.1) yields the
descent time T ∗ = x0 /(v0 cos ϕ0 ). For given values of the parameters x0 and v0 , the time T ∗ has the minimal
value for ϕ0 = 0, which is
Contribution to the brachistochrone problem with Coulomb friction 113

x0
T = . (98)
v0
Based on y(T ) = y f and ϕ0 = 0, from Eq. (77.2) it follows that

gx02
y f = y0 − , (99)
2v02
and, finally, the angle ϕ f is given by
   
−1 ẏ− (T ) −1 gx0
ϕ f = tan = tan . (100)
ẋ− (T ) v02

Based on the previous results, similar to Sect. 7 it can be easily shown that for the case of v0 = (1 ± )(v0 )cr ,
where 0 <   1, the descent time T is approximately given by the expression (97).
For µ = 0 and v0 = 0 it is ϕ f = 0, while from Eq. (66.3) it is obtained that ϕ1 = π/2 so that ϕ0 = ϕ1 .
Further, Eqs. (66.1), (66.2), and (91) for this case give the following values of the quantities v f , y f and T :
 
4gx0 2x0 π x0
vf = , y f = y0 − , T = , (101)
π π g
and then, in regard to Eqs. (89) and (90), the parametric equations of the brachistochrone read
x0 2x0 x0
x(ϕ) = (2ϕ + sin 2ϕ) , y(ϕ) = y0 − + (1 − cos 2ϕ) . (102)
π π π
Finally, for µ = 0 and v0  = 0 stands that ϕ f = 0, and from Eq. (94) it follows that cos 2(ϕ0 − ϕ1 ) = 1, that
is, ϕ1 = ϕ0 . In regard to this, Eqs. (66.1)–(66.3) are reduced to the following equation system:
v 2f v 2f
(2ϕ0 + sin 2ϕ0 ) = x0 , yf + [1 − cos 2ϕ0 ] = y0 , v f cos ϕ0 = v0 (103)
4g 4g
in the unknowns v f , ϕ0 , and y f . From Eq. (91) follows the descent time T = v f ϕ0 /g, and similar to the
previous case, the parametric equations of the brachistochrone read
v 2f v 2f
x(ϕ) = (2ϕ + sin 2ϕ) , y(ϕ) = y f + (1 − cos 2ϕ) . (104)
4g 4g

9 Application of the differential evolution in numerical computations

The systems of nonlinear equations from which the unknowns ϕ0 , ϕ f , ϕ1 and v f are determined, are presented
with a high nonlinearity. When determining the roots of these equation systems by using standard numerical
methods (see e.g. [2]), difficulties are encountered due to the sensitivity of the convergence of the solution
upon the accuracy degree of the initial guess for the roots of the nonlinear equation systems, the possibility of
multiple roots etc. This is why an alternative approach was applied in this paper to determine the roots of the
nonlinear equation systems, which is based on optimization of an appropriate cost function (see [12,13,16]).
This approach is further exposed in the case of solving the system of Eqs. (66.1)–(66.3) and (67) for the
unknowns ϕ0 , ϕ f , ϕ1 and v f .This system can be presented in the following form:
F1 (ϕ0 , ϕ f , ϕ1 , v f ) = 0,
. . . . . (105)
F4 (ϕ0 , ϕ f , ϕ1 , v f ) = 0.
The problem of finding the solution to the equation system (105) can be transformed into the optimization
problem where the cost function is given as

4
F= |Fi | (106)
i=1
114 S. Šalinić

with the restrictions



π π
− ≤ ϕ0 , ϕ f , ϕ1 ≤ , 0 ≤ v f ≤ v02 + 2gy0 . (107)
2 2
It is obvious that the values (ϕ0∗ , ϕ ∗f , ϕ1∗ , v ∗f ), for which the cost function F has a global minimum, represent
the solution to the equation system (105). The differential evolution [18], which represents the optimization
method which is simple, fast in convergence and has many other advantages in comparison to other optimiza-
tion methods (for more details see [18]), was used in this paper to determine previously mentioned parameter
values. The other methods of optimization of the cost function are given in [12,13,16] .There is more than one
variant of the differential evolution, and this paper used DE/local-to-best/1/exp (see [18]). All the computa-
tions were conducted for the following values of the control variables of the differential evolution: crossover
probability constant C R = 0.95, scaling factor M F = 0.5, maximum number of iterations (generations) is
200, population size N P = 10D where D is a number of variables of the cost function. Numerical values in
Tables 1, 2, 3 and 4 are shown with 14 decimals which assures the accuracy F < 10−12 .

10 Conclusions

This paper has examined two variants of the brachistochrone problem with Coulomb friction, the variant when
the coordinates of the initial and the final position has been given and the variant when the initial position has
been given with the requirement that the final position is on a vertical straight line. It has been shown for both
variants that depending on the value of the initial speed v0 of the particle, the following cases are possible:
• v0 = 0 — brachistochrone is a one-segment curve,
• 0 < v0 < (v0 )cr — brachistochrone is a two-segment curve with the initial line segment representing a
free-fall parabola in a nonresistant medium,
• v0 ≥ (v0 )cr — brachistochrone is a one-segment curve which represents a free-fall parabola in a nonresis-
tant medium.
In the second variant for the given µ, by changing the initial speed of the particle in the interval 0 < v0 ≤
(v0 )cr brachistochrones are found which intersect the vertical straight line in the final position with the same
angle tan−1 µ, and for v0 > (v0 )cr this angle is tan−1 (gx0 /v02 ).
The approach described in this paper to solve the brachistochrone problem with Coulomb friction gives an
opportunity to analyze the effect of different variants of initial and final conditions on the form of the analytical
solution of the brachistochrone problem by using variational calculus. This is specifically important in practical
applications such as in optimizing the shape of the gravity flow discharge chute for granular materials [3].

Acknowledgments The author would like to extend his gratitude to Professor Vukman Čović and Professor Miroslav M. Vesković
for their helpful comments and discussions during the preparation of this work.

References

1. Ashby, N., Brittin, W.E., Love, W.F., Wyss, W.: Brachistochrone with Coulomb friction. Am. J. Phys. 43(10), 902–906 (1975)
2. Chapra, S.C., Canale, R.P.: Numerical Methods for Engineers: with Software and Programming Applications,
4th edn. McGraw Hill, New York (2002)
3. Charlton, W., Chiarella, C., Roberts, A.W.: Gravity flow of granular materials in chutes: optimizing flow properties. J. Agr.
Eng. Res. 20, 39–45 (1975)
4. Čović, V., Vesković, M.: Extension of the Bernoulli’s case of brachistochronic motion to the multibody system having the
form of a kinematic chain with external constraints. Eur. J. Mech. A/Solids 21, 347–354 (2002)
5. Čović, V., Lukačević, M.: Extension of the Bernoulli’s case of a brachistochronic motion to the multibody system in the
form of a closed kinematic chain. Facta Univ. Ser. Mech. Autom. Control Robot. 2(9), 973–982 (1999)
6. Čović, V., Vesković, M.: Brachistochrone on a surface with Coulomb friction. Int. J. Nonlinear Mech. 43(5), 437–450 (2008)
7. Djukić, Dj.: On the brachistochronic motion of a dynamic system. Acta Mech. 32, 181–186 (1979)
8. Elsgolc, L.E.: Calculus of Variations. Pergamon Press, Oxford (1963)
9. Gelfand, I.M., Fomin, S.V.: Calculus of Variations. Prentice Hall, Englewood Cliffs (1964)
10. Gregory, J., Lin, C.: An unconstrained calculus of variations formulation for generalized optimal control problems and for
the constrained problem of Bolza. J. Math. Anal. Appl. 187, 826–841 (1994)
11. Hayen, J.C.: Brachistochrone with Coulomb friction. Int. J. Nonlinear Mech. 40, 1057–1075 (2005)
12. Hirsch, M.J., Meneses, C.N., Pardalos, P.M., Resende, M.G.C.: Global optimization by continuous grasp. Opt. Lett. 1(2),
201–212 (2007)
Contribution to the brachistochrone problem with Coulomb friction 115

13. Karr, C.L., Weck, B., Freeman, l.M.: Solutions to systems of nonlinear equations via a genetic algorithm. Eng. Appl. Artif.
Intell. 11, 369–375 (1998)
14. Lipp, S.C.: Brachistochrone with Coulomb friction. SIAM J. Control Optim. 35(2), 562–584 (1997)
15. Maisser, P.: Brachystochronen als zeitkürzeste Fahrspuren von Bobschlitten. ZAMM-Z. Angew. Math. Mech. 78(5), 311–319
(in German) (1998)
16. Martinez, J.M.: Algorithms for solving nonlinear systems of equations. In: Spedicato, E. (ed.) Continuous Optimization: The
State of the Art, pp. 81–108. Kluwer, Dordrecht (1994)
17. Papastavridis, J.G.: On a Lagrangean action based kinetic instability theorem of Kelvin and Tait. Int. J. Eng. Sci. 24(1), 1–17
(1986)
18. Price, K.V., Storn, R.M., Lampinen, J.A.: Differential Evolution: a Practical Approach to Global Optimization,
2nd edn. Springer, Berlin (2005)
19. Stork, D.G., Yang, J.: The general unrestrained brachistochrone. Am. J. Phys. 56(1), 22–26 (1988)
20. Zeković, D.: The brachistochrone motion of mechanical system with nonholonomic, nonlinear and nonstationary con-
straints. PMM-J. Appl. Math. Mech. 54(6), 765–768 (1990)
21. Zeković, D., Čović, V.: On the brachistochronic motion of mechanical systems with linear nonholonomic nonhomogeneous
constraints. Mech. Res. Commun. 20(1), 25–35 (1993)
22. Valentine, F.A.: The problem of Lagrange with differential inequalities as added side conditions. Contributions to the calculus
of variations, 1933–1937, pp. 407–448. University of Chicago Press, Chicago (1937)
23. Vander Heijden, A.M.A., Diepstraten, J.D.: On the brachystochrone with dry friction. Int. J. Nonlinear Mech. 10, 97–112
(1975)
24. Vratanar, B., Saje, M.: On the analytical solution of the brachistochrone problem in a nonconservative field. Int. J. Nonlinear
Mech. 33(3), 489–505 (1998)
25. Wensrich, C.M.: Evolutionary solutions to the brachistochrone problem with Coulomb friction. Mech. Res. Commun.
31, 151–159 (2004)

You might also like