Calculus of Variations
Summer Term 2015
Lecture 10
Universität des Saarlandes
2. Juni 2015
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 1 / 28
Purpose of Lesson
Purpose of Lesson:
To introduce the notion of a broken extremal
To discuss the properties of broken extremals
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 2 / 28
Broken extremals
§7. Broken extremals
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 3 / 28
Broken extremals
Until now we mostly stude the extremals curves with at least two
well-defined derivatives.
Obviously this is not always true.
Broken extremals are continuous extremals for which the gradient
has a discontinuty at one or more points.
If a variational problem has a smooth extremal (That therefore
satisfies the Euler-Lagrange equations), this will be better than a
broken one.
But some problems don’t admit smooth extremals.
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 4 / 28
Broken extremals Broken extremals example
Example 10.1
Find y (x) to minimize
Z1
J[y ] = y 2 (1 − y 0 )2 dx
−1
subject to y (−1) = 0 and y (1) = 1.
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 5 / 28
Broken extremals Broken extremals example
Example 10.1 (cont.)
There is no explicit x dependence inside the integral, so we can
find
F − y 0 Fy 0 = c1 = const
y 2 (1 − y 0 )2 + 2y 0 y 2 (1 − y 0 ) = c1
y 2 (1 − y 0 ) 1 + y 0 = c1
h i
y 2 1 − y 02 = c1
If c1 = 0 we get the singular solutions
y =0 or y = ±x + B.
Neither of these satisfies both end-ponts conditions y (−1) = 0
and y (1) = 1, so c1 6= 0 (we think).
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 6 / 28
Broken extremals Broken extremals example
Example 10.1 (cont.)
Given c1 6= 0
h i
y 2 1 − y 02 = c1
y 2 − c1
y 02 =
y2
dy 1
q
=± y 2 − c1
dx y
y
dx = ±p dy
y 2 − c1
q
x = ± y 2 − c1 + c2
(x − c2 )2 = y 2 − c1
The solution is a rectangular hyperbola.
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 7 / 28
Broken extremals Broken extremals example
Example 10.1 (cont.)
Using the end-points conditions we find c1 and c2 from
(x − c2 )2 = y 2 − c1 .
y (−1) = 0 ⇒ (−1 − c2 )2 = −c1
y (1) = 1 ⇒ (1 − c2 )2 = 1 − c1
Addition of these two equations gives
(1 − c2 )2 = 1 + (1 + c2 )2
which has solution c2 = −1/4, and so c1 = −9/16
y 2 = (x + 1/4)2 − 9/16.
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 8 / 28
Broken extremals Broken extremals example
Example 10.1 (cont.)
The end-points are on opposite branches of the hyperbola!
There is NO smooth extremal curve that connects (−1, 0) and
(1, 1).
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 9 / 28
Broken extremals Broken extremal
Sometimes there is no smooth extremal.
We must seek a broken extremal.
Still want a continuous extremal.
What should we do?
Previous smothness results suggest that we should use a smooth
extremal when we can, and so we will try to minimize the number of
corners.
We’ll start by looking for curves with one corner.
But can we apply the Euler-Lagrange equations?
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 10 / 28
Broken extremals Broken extremal
If we have an extremal like this, can we use the Euler-Lagrange
equations?
y
(x0,y0)
(x1,y1)
x*
x
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 11 / 28
Broken extremals Smoothness theorem
Smoothness theorem
Theorem 10.1
If the smooth curve y (x) gives an extremal of a functional J[y ] over the
class of all admissible curves in some ε neighborhood of y , then y (x)
also gives an extremal of a functional J[y ] over the class of all
piecewise smooth curves in the same neighborhood.
Meaning:
We can extend our results to piecewise smooth curves (where a
smooth result exists), not just curves with 2 continuous derivatives.
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 12 / 28
Broken extremals Smoothness theorem
y
y+εη
y=y(x)
(x1,y1)
(x0,y0)
x
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 13 / 28
Broken extremals Proof Sketch
Proof Sketch
The theorem assumes that there exists a smooth extremal (in this
case a minimum for the purpose of illustration) y . Then for any
other smooth curve ŷ ∈ Bε (y ) we know J[ŷ ] > J[y ].
Assume for the moment that for a piecewise smooth function
ỹ ∈ Bε (y ) we have J[ỹ ] < J[y ]. We can approximate ỹ by a
smooth curve ŷδ ∈ Bε (y ) by rounding off the edges of the
discontinuity.
Given that we can approximate the curve ỹ arbitrarily closely by a
smooth curve ŷδ , for which we already know J[ŷδ ] > J[y ]. We get
a contradiction with J[ỹ ] < J[y ], and so no such alternative
extremal can exist.
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 14 / 28
Broken extremals So What We Do?
So What We Do?
Break the functional into two parts:
Zx ∗ Zx1
J[y ] = J1 [y ] + J2 [y ] = F (x, y1 , y10 )dx + F (x, y2 , y20 )dx
x0 x∗
We require y to have two continuous derivatives everywhere
except at x ∗ , and y1 (x ∗ ) = y2 (x ∗ ).
y
(x0,y0)
(x1,y1)
x*
x
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 15 / 28
Broken extremals So What We Do?
Possible Perturbations:
y
yˆ = y + εη
y = y(x)
€
(x0,y0)
(x1,y1)
x*
xˆ * x
The location of the ”corner” can also be perturbed
c Daria Apushkinskaya (UdS) €
Calculus of variations lecture 10 2. Juni 2015 16 / 28
Broken extremals The First Variation: part 1
The First Variation: part 1
We get the first component of the first variation by considering a
problem with only one fixed end-point, and allowing x ∗ to vary, so
that
Zx̂ ∗
dφ1 (ε) d
0= = F (x, y1 + εη, y10 + εη 0 )dx
dε ε=0 dε ε=0
x0
x ∗ +εX
Z
d
= F (x, y1 + εη, y10 + εη 0 )dx
dε ε=0
x0
Zx ∗
d
= XF (x, y1 , y10 ) + Fy10 η + Fy1 − Fy10 ηdx
x=x ∗ x=x ∗ dx
x0
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 17 / 28
Broken extremals The First Variation: part 1
The First Variation: part 1
The perturbed point (x̂ ∗ , ŷ ∗ ) and perturbed function η must satisfy
certain conditions to be compatible.
Remember that
x̂ ∗ = x ∗ + εX
ŷ ∗ = y ∗ + εY
Notice that
ŷ ∗ = y (x ∗ + εX ) + εη(x ∗ + εX ).
From Taylor’s theorem, for small ε
y (x ∗ + εX ) = y (x ∗ ) + εXy 0 (x ∗ ) + O(ε2 )
= y ∗ + εXy 0 (x ∗ ) + O(ε2 )
εη(x ∗ + εX ) = εη(x ∗ ) + O(ε2 )
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 18 / 28
Broken extremals The First Variation: part 1
The First Variation: part 1
So
y ∗ + εY = y ∗ + εXy 0 (x ∗ ) + εη(x ∗ ) + O(ε2 )
εY = εXy 0 (x ∗ ) + εη(x ∗ ) + O(ε2 )
η(x ∗ ) = Y − Xy 0 (x ∗ ) + O(ε)
Thus, we have
η(x ∗ ) = Y − Xy 0 (x ∗ ) + O(ε) (10.1)
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 19 / 28
Broken extremals The First Variation: part 1
The First Variation: part 1
Substituting the compatibility constraint (10.1) into the our first
variation we get
Zx ∗
h i d
0 = XF + Fy10 η + Fy1 − Fy10 ηdx
x=x ∗ dx
x0
Zx ∗
d
+ Y − Xy10 (x ∗ ) Fy10
= XF + Fy1 − F 0 ηdx
x=x ∗ x=x ∗ dx y1
x0
Zx ∗
h i d
=X F− y10 Fy10 + Y Fy10 + Fy1 − F 0 ηdx
x=x ∗ x=x ∗ dx y1
x0
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 20 / 28
Broken extremals The First Variation: part 1
The First Variation: part 1
So, we get an integral term which results in the E-L equation, plus
the additional constraint
h i
X F − y10 Fy10 ∗
+ Y Fy10 x=x ∗ = 0 (10.2)
x=x
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 21 / 28
Broken extremals The First Variation: part 2
The First Variation: part 2
Note that, for the second component of the First Variation we get a
similar extra term, e.g.
h i
0
−X F − y2 Fy2
0
∗
− Y Fy20 x=x ∗ = 0. (10.3)
x=x
The sign is reversed because it corresponds to the x0 term (as
opposed to the x1 term for δJ1 ).
The combined First Variation (minus the terms that result from the
Euler-Lagrange equation which must be zero) is
h i h i
X F − y10 Fy10 ∗
+Y F 0
y1 x=x ∗ −X F − y 0
F
2 y20
∗
−Y Fy20 x=x ∗ = 0.
x=x x=x
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 22 / 28
Broken extremals Conditions
Conditions
We rearrange to give
nh i h io
0 = X F (x, y1 , y10 ) − y10 Fy10 − F (x, y2 , y20 ) − y20 Fy20
x=x ∗
n o
+ Y Fy10 − Fy20 ∗
.
x=x
Note that the point of discontinuity may vary freely, so we may
independently vary X and Y or set one or both to zero. Hence, we
can separate the condition to get two conditions
h i
F (x, y1 , y10 ) − y10 Fy10 − F (x, y2 , y20 ) + y20 Fy20 ∗
=0
n ox=x
Fy10 − Fy20 ∗
=0
x=x
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 23 / 28
Broken extremals Weierstrass-Erdman
Weierstrass-Erdman
We can write the conditions as
h i h i
F (x, y1 , y10 ) − y10 Fy10 ∗
= F (x, y ,
2 2y 0
) − y 0
F
2 y20
x=x x=x ∗
Fy10 = Fy20
x=x ∗ x=x ∗
Called the Weierstrass-Erdmann Corner Conditions.
Rather than separating y into y1 and y2 we may write the corner
conditions in terms of limits from the left and right, e.g.
F − y 0 Fy 0 x=x ∗− = F − y 0 Fy 0 x=x ∗+
Fy 0 = Fy 0
x=x ∗− x=x ∗+
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 24 / 28
Broken extremals Solution
Solution
So the broken extremal solution must satisfy
The Euler-Lagrange equations
The Weierstrass-Erdmann Corner Conditions
F − y 0 Fy 0 x=x ∗− = F − y 0 Fy 0 x=x ∗+
Fy 0 = Fy 0
x=x ∗− x=x ∗+
must hold at any ”corner”.
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 25 / 28
Broken extremals Broken extremals example
Example 10.1 (cont. II)
Find y (x) to minimize
Z1
J[y ] = y 2 (1 − y 0 )2 dx
−1
subject to y (−1) = 0 and y (1) = 1.
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 26 / 28
Broken extremals Broken extremals example
Example 10.1 (cont. II)
In the example considered
F − y 0 Fy 0 = y 2 1 − y 02
Fy 0 = −2y 2 1 − y 0
Remember that y = 0 and y = x + A are valid solutions to the
Euler-Lagrange equations, and that for both of these solutions
Fy 0 = F − y 0 Fy 0 = 0,
so we can put a ”corner” where needed.
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 27 / 28
Broken extremals Broken extremals example
Example 10.1 (cont. II)
The solution must also satisfy the end-point conditions, so
y (−1) = 0 and y (1) = 1, and therefore, as valid solution has
x ∗ = 0 and
y1 = 0 for x ∈ [−1, x ∗ ]
y2 = x for x ∈ [x ∗ , 1]
c Daria Apushkinskaya (UdS) Calculus of variations lecture 10 2. Juni 2015 28 / 28
