AI summary by TwinMind
Lecture 6 Part 2: Calculus of Variations and Gradients of Functionals
MIT OpenCourseWare
42m 32s
Introduction to Calculus of Variations
- Topic also called "calculus of variations" - classic name
- Same idea as derivatives but applied to functions as vectors
- Formula: F(U + dU) - F(U) = F'(U)dU
- Key difference: U is in vector space of functions, F(U) is scalar (real number)
Functions as Vectors
- Functions can be treated as vectors in vector space
- Can add two functions, multiply by scalars
- Linear algebra concepts apply: linear operators, orthogonal functions, dot products, Gram-Schmidt, eigenfunctions
- Need to be careful with functions - can oscillate infinitely, blow up
- For rigorous math, must restrict to well-behaved function spaces
- Implicitly assume functions where integrals are defined
First Example: Simple Functional
- U(X) function from [0,1] to real numbers
- F(U) = ∫₀¹ sin(U(X))dX (takes function, gives number)
- Called a "functional" - maps function to scalar
- To find gradient: linearize change in F for small change dU
- dU is also a function (small perturbation)
- DF = ∫₀¹ cos(U(X))dU(X)dX
- This is linear operator acting on dU
Dot Products of Functions
- Standard Euclidean dot product for functions: ∫₀¹ U(X)V(X)dX
- Norm of function: ||U|| = √(∫₀¹ U(X)²dX)
- Once have dot product, can define gradient
- Gradient found by inspection: ∇F = cos(U(X))
- Sometimes called Fréchet derivative (general linear operator concept)
Arc Length Example (More Complex)
- F(U) = ∫₀¹ √(1 + U'(X)²)dX (arc length of curve U(X))
- Geometric meaning: length of curve from x=0 to x=1
- Linearization involves Taylor expansion of √(1 + (U' + dU')²)
- First order: U'dU'/√(1 + U'²)
- Need integration by parts to eliminate dU' terms
- Boundary term: U'/√(1 + U'²) × dU|₀¹
- Integral term: -d/dx(U'/√(1 + U'²)) × dU
Boundary Conditions and Minimization
- Fix endpoints: U(0) = U(1) = 0
- Find U that minimizes arc length F(U)
- With fixed endpoints, dU = 0 at boundaries
- Boundary terms disappear
- Gradient: ∇F = -d/dx(U'/√(1 + U'²))
- At minimum: ∇F = 0
- Working out derivative using quotient rule
- Result: U'' = 0, so U' = constant (straight line)
- Confirms shortest distance between two points is straight line
General Euler-Lagrange Equations
- General functional: F(U) = ∫ᴬᴮ f(X, U, U')dX
- f depends on X, U(X), and U'(X)
- DF has terms: ∂f/∂U × dU and ∂f/∂U' × dU'
- Integration by parts on second term
- Boundary term: ∂f/∂U' × dU|ᴬᴮ = 0 (fixed endpoints)
- Gradient: ∇F = ∂f/∂U - d/dx(∂f/∂U')
- Setting equal to zero gives Euler-Lagrange equation
- Second-order differential equation in U
Applications of Euler-Lagrange Equations
- Optimal control theory
- Advanced mechanics (courses 8.06, 8.08)
- Least action principle - alternative to Newton's laws
- Allows writing equations in weird coordinate systems
- Brachistochrone problem (18th century)
- Mass sliding down frictionless ramp
- Find shape for fastest descent time
- Solution is cycloid (curve traced by point on rolling wheel)
- Principle of least action in mechanics
Mathematical Framework Summary
- Functions treated as vectors in infinite-dimensional space
- Linear operators, dot products generalize to function spaces
- Derivative = linearized change in output
- Integration by parts converts dU' terms to dU terms
- Gradient found via dot product representation
- Extrema found by setting gradient to zero
- Results in differential equations (Euler-Lagrange)
