MCS3: Continuous Mathematics Flashcards
Lemma
If g: ℝ → ℝ is a stricly increasing function, then
Optimization Tricks
arg min f(x) = arg min g(f(x)) for x ∈ F
Lemma
If g: ℝ → ℝ is a 1-1 function, then
Optimization Tricks
arg min f(x) = g(arg min f(g(x))) for x ∈ F
Knowledge
Midpoint Rule
2 points
- Approximate f: [a, b] → ℝ by 0ᵗʰ-order Taylor polynomial at m
- M₁[f, a, b] = (b - a)f(m)
Knowledge
Bounds for err(M₁)[f, a, b]
2 points
- Lower bound: -((b - a)³/24)D̅₂
- Upper bound: -((b - a)³/24)D̲₂
Knowledge
Trapezium Rule
2 points
- Approximate f: [a, b] → ℝ by its linear interpolation
- T₁[f, a, b] = ((b - a)/2)(f(a) + f(b))
Knowledge
Bounds for err(T₁)[f, a, b]
2 points
- Lower bound: ((b - a)³/12)D̲₂
- Upper bound: ((b - a)³/12)D̅₂
Knowledge
Simpson’s Rule
2 points
- Approximate f: [a, b] → ℝ by quadratic interpolation
- S₂[f, a, b] = ((b - a)/6)(f(a) + 4f(m) + f(b))
Knowledge
Bounds for err(S₂)[f, a, b]
2 points
- Lower bound: ((b - a)⁵/2880)D̲₄
- Upper bound: ((b - a)⁵/2880)D̅₄
Formula
Composite Midpoint Rule
Mₙ[f, a, b] = ((b - a)/n)(f((x₀ + x₁)/2) + … + f((xₙ₋₁ + xₙ)/2))
Knowledge
Composite Midpoint Rule Error Bounds
2 points
- O(n⁻²)
- Depends on -f’’
Formula
Composite Simpson’s Rule
Sₙ[f, a, b] = ((b - a)/(3n))(f(x₀) + 4f(x₁) + 2f(x₂) + 4f(x₃) + … + 4f(xₙ₋₁) + f(xₙ))
Knowledge
Composite Simpson’s Rule Error Bounds
2 points
- O(n⁻⁴)
- Depends on f’’’’
Knowledge
Monte Carlo Integration
4 points
- MCₙ[f, R] = A(R)(1/n)∑f(X̲ᵢ)
- A(R) is the area or volume of R
- X̲ᵢ are indpenedently uniformly random on R
- Pdf is p(x̲) = 1/A if x̲ ∈ R otherwise p(x̲) = 0
Lemma
MCₙ[f, r] is unbiased
E[MCₙ[f, r]] = ∫ᵣ f(x̲) dx̲
Lemma
Var[MCₙ[f, R]] = …
V / n for a constant V
Lemma
MCₙ[f, R] is consistent
P[|err(MCₙ)[f, R]| ≥ ε] → 0 as n → ∞ for any ε > 0
Lemma
For large n err(MCₙ)[f, R] approaches
N(0, V/n)
Knowledge
IEEE 754 Single Precision Standard
4 points
- 23 bits for m
- 8 bits for e
- 1 bit for sign
- ε = 2⁻²⁴ ≈ 6⋅10⁻⁸