Improper Integrals and Their Convergence

What is an Improper Integral

The standard Riemann integral requires that a function is defined and bounded on a closed finite interval. But what if the domain of integration is infinite or the function is unbounded?

Improper integrals extend the theory to such cases.

Type I (infinite limit): ∫ₐ^∞ f(x) dx = lim(b→∞) ∫ₐᵇ f(x) dx.

Type II (unbounded function): ∫ₐᵇ f(x) dx (f → ∞ as x→a+) = lim(ε→0+) ∫ₐ₊ₑᵇ f(x) dx.

If the limit exists and is finite—the integral converges. Otherwise—it diverges.

Convergence Criteria

Comparison test: If 0 ≤ f(x) ≤ g(x) and ∫g converges, then ∫f converges.

Benchmarks: ∫₁^∞ dx/xᵖ converges for p > 1, diverges for p ≤ 1.
∫₀¹ dx/xᵖ converges for p < 1, diverges for p ≥ 1.

Dirichlet’s test: ∫ₐ^∞ f(x)g(x) dx converges if F(x) = ∫ₐˣ f(t)dt is bounded, and g is monotonic and tends to 0.

For example, ∫₁^∞ sin(x)/x dx converges (conditionally), but ∫₁^∞ |sin(x)|/x dx diverges.

The Most Important Improper Integrals

Gamma function: Γ(s) = ∫₀^∞ xˢ⁻¹e⁻ˣ dx for s > 0.

Properties: Γ(1) = 1, Γ(n) = (n-1)! for natural n, Γ(1/2) = √π.

The gamma function generalizes the factorial to real numbers and appears in probability theory, statistics, and physics.

Dirichlet integral: ∫₀^∞ sin(x)/x dx = π/2.

Gaussian integral: ∫₋∞^∞ e^(-x²) dx = √π — the basis of the normal distribution.

Absolute and Conditional Convergence

∫ f converges absolutely if ∫ |f| converges. Absolute convergence implies convergence.

If ∫ f converges but ∫ |f| diverges — conditional convergence. This is analogous to conditionally convergent series.

A conditionally convergent integral is "fragile": rearranging terms can change its value.

Stirling’s Formula

How does n! behave for large n? n! ≈ √(2πn) · (n/e)ⁿ (Stirling’s formula). This follows from the asymptotics of the Gaussian integral. The formula is used in combinatorics, information theory, and statistical physics.

Beta Function and Its Relation to the Gamma Function

Beta function: B(p, q) = ∫₀¹ t^(p-1)(1−t)^(q-1) dt for p, q > 0.

Key relation: B(p, q) = Γ(p)Γ(q)/Γ(p+q). This allows calculation of complicated trigonometric integrals.

Example: ∫₀^(π/2) sin^n x dx = (√π/2) · Γ((n+1)/2) / Γ((n+2)/2).
For n = 2: ∫₀^(π/2) sin²x dx = π/4. ✓

Laplace Transform as an Improper Integral

Laplace transform $\hat f(s) = \int_0^{\infty} f(t) e^{-st} ,dt$ is an improper integral over an infinite interval with parameter s. For s > σ₀ (where σ₀ is the abscissa of convergence), the integral converges absolutely.

Laplace images of basic functions: L{1} = 1/s, L{e^{at}} = 1/(s−a), L{sin ωt} = ω/(s² + ω²). Differentiation becomes multiplication: L{f'(t)} = s·L{f(t)} − f(0). This property makes the Laplace transform a primary tool in ODE theory and control theory: a differential equation becomes an algebraic one.

Cauchy Principal Value and Conditional Convergence

A conditionally convergent integral is "fragile": its value may depend on how the limit is taken.
Cauchy principal value: V.P. ∫₋∞^∞ f(x) dx = lim(R→∞) ∫₋R^R f(x) dx.
For odd functions V.P. ∫₋∞^∞ x dx = 0, though the integral diverges in the usual sense.
The concept of the principal value is critical in the theory of singular integrals and in hydrodynamics equations.

Dirichlet’s Criterion for Improper Integrals

Dirichlet’s criterion: If f(x) monotonically tends toward zero as x → ∞ and ∫ₐˣ g(t) dt is bounded, then ∫ₐ^∞ f(x)g(x) dx converges. This generalizes Leibniz’s test for series to integrals.
Example: ∫₁^∞ sin(x)/x dx converges (f = 1/x ↘ 0, primitive g = sin is bounded), but ∫₁^∞ |sin(x)|/x dx diverges (conditional but not absolute convergence).

Comparison of Convergence Criteria for Improper Integrals

The arsenal of convergence criteria: comparison test (if 0 ≤ f ≤ g and ∫g converges, then ∫f converges), limit comparison test (if f/g → L ∈ (0, ∞), then ∫f and ∫g either both converge or both diverge), Dirichlet’s criterion, and Abel’s criterion for conditionally convergent integrals. In practice, the standard benchmarks are integrals like ∫₁^∞ dx/x^α (converges for α > 1, diverges for α ≤ 1) and ∫₀^∞ e^(−αx) dx (converges for α > 0). Most problems are reduced to comparison with these two types.

Question for thought: Why use Stirling’s formula to compute n! in informatics instead of direct multiplication? For which n is the error of Stirling’s formula less than 1%?

Gamma Function as Analytic Continuation of the Factorial

The gamma function Γ(n) = ∫₀^∞ t^(n−1) e^(−t) dt satisfies the relation Γ(n+1) = n·Γ(n) and Γ(1) = 1, so Γ(n+1) = n! for integers n ≥ 0. But Γ is also defined for fractional arguments: Γ(1/2) = √π (hence ∫₋∞^∞ e^(−x²) dx = √π — the Gaussian integral). Stirling’s formula n! ≈ √(2πn)(n/e)ⁿ follows from the asymptotics of Γ via the saddle point method.
Applications: in combinatorics (estimating the number of permutations), in quantum mechanics (Bose-Einstein and Fermi-Dirac statistics), in information theory (calculating entropy of large systems).