Calculus of Variations & Variational Methods

The calculus of variations is used to obtain extrema of expressions involving unknown functions called functionals. Applications range from simple geometric problems to finite-element methods to optimization theory. References: Ewing, G. M., Calculus of Variations with Applications, Dover, 1985. Hildebrand, F. B., Methods of Applied Mathematics, 2nd Ed., Dover, 1992. Reddy & Rasmussen, Advanced Engineering Analysis, Kreiger 1990. Weinstock, R., Calculus of Variations with Applications to Physics and Engineering, Dover.
292

Calculus of Variations & Variational Methods

Unconstrained Minimization In preparation for an introduction to the calculus of variations, recall maxima, minima (extrema), and inflections of functions from the differential calculus. x0 = critical (stationary) local maximum f ( x) point df f max =0 f max = f ( x0 ) dx = critical value x0

local minimum

f ( x) f max x0

f ( x)

inflection df =0 dx x0
293

df =0 dx

Calculus of Variations & Variational Methods

df dx
= 0, necessary condition for extremum
x = x0

df dx

x = x0

d2 f 2 dx 2 d f =0 2 dx 2 d f dx 2

< 0 local max

x0

= 0 inflection
x0

> 0 local min

x0

294

Calculus of Variations & Variational Methods

Given f ( x, y ) , a necessary condition for a minimum at ( x0 , y0 ) is
df
( x0 , y0 )

f f = dx + dy = dr f y ( x , y ) x 0 0

( x0 , y0 )

=0

Since x and y are independent variables, dx and dy are independent so,

f f = 0 and =0 x y

295

Calculus of Variations & Variational Methods

Constrained Minimization Now minimize f with a constraint, e.g., G ( x, y ) = 0 Lagrange Multiplier Method From (1),
G G dx + dy = dr G = 0 x y

(1)

We introduce the modified function with no constraints

FL ( x, y, ) f ( x, y ) + G ( x, y ) and set F F F dFL = L dx + L dy + L d = 0 x y

(2) (3)
296

Calculus of Variations & Variational Methods

Example: Find the stationary (critical) point of the function
f ( x, y ) = 2 x 2 + y 2 8 x + y + 1

with the constraint 2 x y = 0 Soln:

FL ( x, y, ) = (2 x 2 + y 2 8 x + y + 1) + (2 x y )

FL = 4 x 8 + 2 = 0, x FL = 2 y + 1 = 0, x = 0.5, y FL = 2 x y = 0.

y = 1.0, = 3.0

297

Calculus of Variations & Variational Methods

Calculus of Variations: Functionals and Euler Equations The focus of the calculus of variations is the determination of maxima and minima of expressions that involve unknown functions. Here we look at a few classical problems to introduce some of the concepts.

The Brachistochrone (the one that started it all): Weinstock gives the problem as it was originally stated by John Bernoulli in 1696: Given two points A and B in a vertical plane, to find for the moveable particle M, the path AMB, descending along which by its own gravity, the beginning to be urged from point A, it may in the shortest time reach the point B.

298

Calculus of Variations & Variational Methods

Reddy & Rasmussen state in engineering terms: Design a chute between two points A: (0,0) and B: (xb,yb) in a vertical plane such that a material particle, sliding without friction under its own weight, travels from point A to point B along the chute in the shortest time. xb A

yb

299

Calculus of Variations & Variational Methods

Other Classic Problems Geodesic Problem What is the curve of minimum length that connects two points?

L=

dy 1 + dx dx

(b, yb )

y ( a ) = ya ,

y (b) = yb

( a , ya )
x

300

Calculus of Variations & Variational Methods

Minimum Surface of Revolution What is the curve of minimum length that connects two points?

S = 2 y ds
a

y
2

(b, yb )

du = 2 u 1 + dx a dx
b

( a , ya )

y = u ( x) x

y ( a ) = ya ,

y (b) = yb

301

Calculus of Variations & Variational Methods

The Euler Equation Each of these problems presents the problem of finding a continuously differentiable function u(x) that minimizes the integral of the form
I (u ) = F ( x, u , u) dx
a b

(1)

And that satisfies the end conditions u (a ) = ua , u (b) = ub We now suppose that u(x) is the minimizing function, then choose any continuously differentiable function (x) and create a oneparameter trial function y = u ( x) = u ( x) + ( x)
302

Calculus of Variations & Variational Methods

that vanishes at the end points u = ua and u = ub. Then for any constant , u ( x) = u ( x) + ( x) with (a ) = (b) = 0 satisfies the end conditions. u (a ) = ua , u (b) = ub u(x) u(x) u ( x) = u ( x) + ( x)

( x)
a b x
303

Calculus of Variations & Variational Methods

Now we substitute this function into the integral to be minimized
I ( ) = F ( x, u , u ) dx = F ( x, u + , u + ) dx
a a b b

where I is now a function of once u(x) and (x) are assigned. Since we have assumed that u(x) is the minimizing function then I() is minimized when = 0. So, like the differential calculus, we now determine the stationary function (analogous to stationary points), but in this case we know in advance that = 0. So, dI ( ) =0 d b F du dI ( ) F du dx = + a d u d u d b F F = + dx a u u 304

Calculus of Variations & Variational Methods

Now when 0
b F dI (0) F = + dx = 0. a d u u

Now we integrate the second term by parts,

F d F F dI (0) dx + = 0. = a u a d u dx u
b b

Since this must hold for all choices of (x), then

F d F = 0. u dx u
This is the Euler equation or Euler-Lagrange equation.

(2)

305

Calculus of Variations & Variational Methods

So, if u(x) minimizes I(u), it must satisfy the Euler equation. Solutions of the Euler equation are called extremals of the problem and an extremal that satisfies the end conditions is called a stationary function. First Integrals Equation (2) is written with partial derivatives that treat x, u, and u as independent variables. The second term can be expanded to give F F du F du + + x u u u dx u u dx So, Eq. (2) is equivalent to

306

Calculus of Variations & Variational Methods

d 2u du + ( Fux Fu ) = 0. Fuu 2 + Fuu dx dx
(3)

This is a second-order ODE, unless 2 F / u2 0, so in general there are two arbitrary constants to satisfy the end conditions. One can show that (3) is also equivalent to
1d F du F F = 0. u dx u dx x

So, if F does not involve x explicitly, F F = C if 0. u x This is a first integral of Eulers equation. F u
307

Calculus of Variations & Variational Methods

Also, if F does not involve u explicitly, another first integral is F = C if u Examples F 0. u

308

Calculus of Variations & Variational Methods

Variational Notation The notation of the calculus of variations shows many similarities to the differential calculus. Beginning with the integrand of the functional I(u)

F = F ( x, u , u )
We substituted for

y = u ( x)
with the trial function

y = u ( x) = u ( x) + ( x) = u ( x) + u

where u = ( x) is defined as the first variation of u(x).

309

Calculus of Variations & Variational Methods

For a fixed x, F = F ( x, u + , u + ) F ( x, u , u). Expanding via the Taylor series, F F F = F ( x, u , u ) + u + u + H.O.T. F ( x, u , u ) u u F F = + + H.O.T. u u Thus, the first variation of F is F F F = u + u u u
310

Calculus of Variations & Variational Methods

This is analogous to the total differential, F F F dF = dx + du + du x u u Graphically,

u ( x) = u ( x) + ( x)

( x)
du

u = u u
du = the first-order approximation of the change along u(x) corresponding to a change in x of dx

dx

x fixed

u = the first-order approximation to the change

u to u at a fixed x
311

Calculus of Variations & Variational Methods

Variational laws are analogous to differentiation,

( F1 F2 ) = F1 F2 + F2 F1 ,
F1 F2 F1 F1 F2 = . 2 F2 F2

The variation and derivative are commutative operators, i.e.,

d du ( u ) = dx dx This will be a very important result for us later.

Applying the notation to the functional

I (u ) = F ( x, u , u )dx,
a b

312

Calculus of Variations & Variational Methods

Then F d F I = u + u dx = 0. a u u dx Integrating the second term by parts,
b

b d F F d F u dx = u u dx. u dx u a a dx u b

Then,
F d F F I = u dx + u = 0 a u dx u u a
b b

u = 0 at end points

F d F I = u dx = 0 a u dx u
b

313

Calculus of Variations & Variational Methods

So, F d F = 0. u dx u Thus, we have recovered the Euler-Lagrange equation using the variational notation.
End Conditions We now look at the end conditions (more generally, boundary conditions). Start with

I = F ( x, u , u )dx
a

Applying Liebnizs rule,

314

Calculus of Variations & Variational Methods

I = F dx + [ F ]b b [ F ]a a
a b

(fixed end points)

F F = u u dx. a u u
b

Integrating by parts,
F d F F I = u + u dx a u dx u u a
b b

F d F F F = u dx + ( u )b ( u ) a a u dx u u b u a
b

315

Calculus of Variations & Variational Methods

This then gives us two possible end point conditions: u = 0 at x = a, b essential boundary conditions (u specified at x = a, b)
F = 0 at x = a, b natural boundary conditions u (u not specified at x = a, b)

Example: Calculate the first variation of I ( y ) = y 1 + ( y) 2 dx

a

I = y 1 + ( y) 2 dx
a

Applying the variational chain rule: b b yy 2 2 I = 1 + ( y ) y + y dx + y 1 + ( y ) x 2 a a + 1 ( ) y

= 0 (fixed endpts)

316

Calculus of Variations & Variational Methods

Higher Dimensions Now we extend the methods just presented to multiple variables, in particular we fist look at the function u(x,y,z) and a region in space R enclosed by the surface S. In this case, the functional I(u) has the form R

I (u ) = F ( x, y, z , u , u x , u y , u z ) dx dy dz
R

d
r

or using vector notation I (u ) = F (r, u , u ) d .

R

317

Calculus of Variations & Variational Methods

With the variational notation, the first variation of I(u) is F F F F I = u + ux + uy + u z d R u u x u y u z Note that the variation is evaluated at a fixed point (x,y,z), thus x, y, z = 0.
F F F F + + I = u + u d u x u y u z R u y z x

Note the operator in the parentheses looks a lot like a dot product. If we set F F F x + y + z a= e e e u x u y u z
318

Calculus of Variations & Variational Methods

Then F I = u + a ( u ) d . R u For integration by parts, recall ( a) = a + a a = ( a) a So, F I = a u d + (a u ) d . R u R
319

Calculus of Variations & Variational Methods

Now use the divergence theorem on the last integral F a) u dS . I = a u d + (n R u S Again, the necessary condition for minimizing I(u) is that I = 0. Thus, the integrand of the volume integral must be zero, i.e., F a = 0 u or F F F F u x u x y u y z u z = 0.
320

Calculus of Variations & Variational Methods

This is the 3D Euler-Lagrange equation. The surface integral must also vanish, this gives three possible boundary conditions i.) u specified on S u = 0 on S (essential) a = ii.) n iii.) F F F nx + ny + nz = 0 on S (natural) u x u y u z

u = 0 on part of S

(mixed) a = on remainder of S n

321

Calculus of Variations & Variational Methods

Note the steps used to obtain the Euler equation and the associated boundary conditions: 1. Take the first variation I(u) 2. Use integration by parts to factor u in the integrand 3. Use the divergence theorem to create a surface integral containing the boundary conditions

322

Calculus of Variations & Variational Methods

Example (from Reddy & Rasmussen): Two-dimensional conduction and convective heat transfer y Convective heat transfer T P
, Specified temperature T

CT Ch R Q Cq

Specified heat transfer x

323

Calculus of Variations & Variational Methods

T = temperature T = ambient temperature Q ( x, y ) = internal heat generation k = k1 ii + k2 jj = thermal conductivity = conductive heat transfer q h = convective heat transfer coefficient

324

Calculus of Variations & Variational Methods

Minimize: I (T ) =
R 1 2

ds + h ( 1 { 2 T T )T ds T k T + QT }dx dy qT
Cq Ch

First variation:

I (T ) =
R

1 2

1 ( ) ( T T Q T dx dy q T ds h k + + { } 2 T T ) T ds
Cq Ch

Note that in this problem, we already have explicit boundary integrals. As before the interior integral (over R) will also generate a boundary integral after it is integrated by parts.

325

Calculus of Variations & Variational Methods

Integration by parts: First, use the vector identity T k T = a,

( a ) = a + a, a = ( a ) a. So, T k T = ( T k T ) T ( k T ). Now,

I = ( T k T ) T ( k T ) + Q T dx dy
R

T ds + h (T T q
Cq Ch

) T ds.
326

Calculus of Variations & Variational Methods

Now concentrate on the area integral. In 3D, the divergence theorem gives, dS ( T k T ) d = ( T k T ) n
R S

So, in 2D, ( k T ) dS , ( T k T ) dS = T n
R S

I = T ds + h (T T ) T ds T ( k T ) + Q T dx dy q
R

Then,

CT + Cq +Ch

T n ( k T ) dS ,

Cq

Ch

327

Calculus of Variations & Variational Methods

I = ( k T ) + Q T dx dy
R

( k T ) T ds n
CT

+ ( k T ) q ( k T ) h(T T ) n T ds + n T ds.
Cq Ch

Now that T is isolated we can write the Euler equation with accompanying boundary conditions ( k T ) + Q = 0 Euler Equation

essential boundary condition: on CT ) set n ( k T ) T ds = 0 by setting T = 0 (T = T

CT

328

Calculus of Variations & Variational Methods

Multiple Dependent Variables So far we looked at the case where the functional is a function of one dependent variable, i.e., I(u). (Note, u is a dependent variable since it depends on x.) Now we look at the case I(u,v) where the functional is a function of two dependent variables. The following development is for the 2D case, leaving the 3D case as an obvious extension. Begin with the functional to be minimized

I (u, v ) = F ( x, y , u, v, u x , v x , u y , v y ) dx dy
R
329

Calculus of Variations & Variational Methods

Then for the first variation F F F F F F F = u + v + ux + vx + uy + vy . u v u x v x u y v y essential: u, v specified on C u = v = 0 on C. natural: u, v not specified on C.
ds
R C
330

Calculus of Variations & Variational Methods

As before, the necessary condition for a minimum is I = 0 for min I(u,v) Substituting the first variation F into the integral,
I =
R

u +

F F F F F ux + uy + v + vx + v y dxdy = 0. u x u y v x v y v

Now that weve separated the integral, it should be clear how to proceed based on our earlier developments. We next rearrange the integral to integrate by parts,
I =
R

F u

u +

F F F F F ( u ) + ( u ) + v + ( v ) + ( v ) dxdy = 0. u x x u y y v x x v y y v

331

Calculus of Variations & Variational Methods

Now, for integration by parts lets focus on one of the four terms in the previous expression,
F F F ( u ) = u u u x x x u x x u x

Recall the gradient theorem,

dS . d = n
R S

Then,
ds dx dy = n
R C

dx dy = n ds x x R C dx dy = n y ds y R C
332

Calculus of Variations & Variational Methods

So,
F F F ( u ) dx dy = u u dx dy u x x x u x x u x R R F F = u dx dy n x u ds u x x u x C R

Just as in the previous cases, the integration by parts contributes one portion to the interior of the region and one portion to the boundary. The other three similar terms are expanded in the same fashion to give,
F F F F F F + I = u v dx dy u x u x y u y R v x v x y v y F F F F nx + ny u + nx + n y v ds = 0 + u x u y v y C v x
333

Calculus of Variations & Variational Methods

So, we have
F F F = 0 u x u x y u y Euler Equations F F F 0 = v x v x y v y

u = v = 0 on C (essential)
F F nx + ny = 0 u x u y on C (natural) F F nx + ny = 0 v x v y

or mixed boundary conditions.

334