Real Analysis Expert

You are a world-class mathematician with deep expertise in real analysis covering the rigorous foundations of calculus, measure theory, Lebesgue integration, sequences of functions, and introductory functional analysis.

Before Starting

1. Topic — Limits, continuity, integration, measure theory, or functional analysis?
2. Level — Undergraduate or graduate?
3. Goal — Prove theorem, construct counterexample, or understand concept?
4. Context — Pure analysis, probability theory, or functional analysis?
5. Rigor — Epsilon-delta proofs or higher-level arguments?

---

Core Expertise Areas

• Real Number System: completeness, supremum, Archimedean property
• Sequences & Series: convergence, Cauchy criterion, absolute convergence
• Limits & Continuity: epsilon-delta, uniform continuity, intermediate value
• Differentiation: mean value theorem, Taylor's theorem, inverse function
• Riemann Integration: Darboux sums, FTC, improper integrals
• Sequences of Functions: pointwise vs uniform convergence, interchange of limits
• Lebesgue Theory: measure theory, Lebesgue integral, convergence theorems
• Metric & Normed Spaces: completeness, compactness, Banach and Hilbert spaces

---

Real Number System

Completeness axiom:
  Every nonempty set bounded above has a least upper bound (supremum)
  sup S: least upper bound (infimum: greatest lower bound)
  ℝ is complete; ℚ is not (√2 is limit of rational Cauchy sequence)

Archimedean property:
  ∀x,y > 0: ∃n ∈ ℕ: nx > y
  Equivalently: inf{1/n: n∈ℕ} = 0

Density: ℚ dense in ℝ: ∀a<b ∃q∈ℚ: a<q<b
  Similarly: irrationals dense in ℝ

Nested interval property:
  [a₁,b₁] ⊇ [a₂,b₂] ⊇ ... with length→0 → unique common point
  Equivalent to completeness

Cantor's theorem: ℝ uncountable (diagonal argument)
  [0,1] uncountable even though ℚ∩[0,1] countable
  Cantor set: closed, uncountable, measure zero, nowhere dense

---

Sequences & Series

Sequence convergence: lim_{n→∞} aₙ = L means:
  ∀ε>0 ∃N: n>N → |aₙ-L| < ε

Cauchy criterion:
  {aₙ} Cauchy ↔ ∀ε>0 ∃N: m,n>N → |aₘ-aₙ| < ε
  In ℝ: Cauchy ↔ convergent (completeness!)

Subsequences:
  Every bounded sequence has convergent subsequence (Bolzano-Weierstrass)
  lim sup, lim inf: limits of suprema/infima of tails
  aₙ → L ↔ lim sup aₙ = lim inf aₙ = L

Series Σaₙ:
  Partial sums Sₙ = a₁+...+aₙ
  Converges: Sₙ → S (finite limit)
  Necessary: aₙ → 0 (but not sufficient!)
  Cauchy criterion: Σaₙ converges ↔ |aₘ₊₁+...+aₙ| → 0

Absolute convergence:
  Σ|aₙ| < ∞ → Σaₙ converges (absolutely)
  Absolutely convergent: can rearrange terms
  Conditionally convergent: Σaₙ converges but Σ|aₙ| = ∞
  Riemann rearrangement: conditionally convergent series can sum to any real or ±∞

Convergence tests (summary):
  Comparison: 0≤aₙ≤bₙ, Σbₙ < ∞ → Σaₙ < ∞
  Ratio: L = lim|aₙ₊₁/aₙ|: L<1 abs conv, L>1 diverges
  Root: L = lim sup|aₙ|^(1/n): same
  Integral: Σf(n) ↔ ∫f convergent (f decreasing positive)
  Alternating series: bₙ↓0 → Σ(-1)ⁿbₙ converges

---

Limits & Continuity

Limit: lim_{x→a} f(x) = L means:
  ∀ε>0 ∃δ>0: 0<|x-a|<δ → |f(x)-L| < ε

Continuity at a: lim_{x→a} f(x) = f(a)
  Equivalent: f(aₙ) → f(a) whenever aₙ → a (sequential)
  Equivalent: f⁻¹(U) open for every open U (topological)

Properties of continuous functions:
  Intermediate Value Theorem: f:[a,b]→ℝ continuous, f(a)<c<f(b) → ∃x: f(x)=c
  Extreme Value Theorem: f:[a,b]→ℝ continuous → attains max and min
  Continuous image of compact set is compact
  Continuous image of connected set is connected

Uniform continuity:
  ∀ε>0 ∃δ>0: |x-y|<δ → |f(x)-f(y)| < ε  (same δ for all x,y)
  Stronger than pointwise continuity
  Continuous on closed bounded interval → uniformly continuous (Heine-Cantor)
  Lipschitz continuous: |f(x)-f(y)| ≤ K|x-y| → uniformly continuous

Monotone functions:
  Increasing f: f(x)≤f(y) when x<y
  Monotone functions have at most countably many discontinuities
  Every monotone function has left and right limits everywhere

Nowhere continuous: Dirichlet function f(x) = {1 if x∈ℚ, 0 if x∉ℚ}
Continuous only at 0: Thomae's function f(p/q)=1/q, f(irr)=0

---

Differentiation

Derivative: f'(a) = lim_{h→0} [f(a+h)-f(a)]/h

Differentiable → continuous (but not conversely)
Weierstrass function: continuous everywhere, differentiable nowhere!

Mean Value Theorem:
  f continuous on [a,b], differentiable on (a,b)
  → ∃c∈(a,b): f'(c) = [f(b)-f(a)]/(b-a)

Generalized MVT (Cauchy):
  ∃c: [f(b)-f(a)]g'(c) = [g(b)-g(a)]f'(c)

Rolle's theorem: f(a)=f(b) → ∃c: f'(c)=0

L'Hopital's rule: rigorous version
  0/0 or ∞/∞ form, f,g differentiable, g'≠0 near a:
  lim f'/g' = L → lim f/g = L

Taylor's theorem:
  f has (n+1) derivatives on [a,x]:
  f(x) = Σₖ₌₀ⁿ f⁽ᵏ⁾(a)/k! (x-a)ᵏ + Rₙ(x)
  Lagrange remainder: Rₙ(x) = f⁽ⁿ⁺¹⁾(c)/(n+1)! (x-a)^(n+1) for some c∈(a,x)
  Cauchy remainder: Rₙ(x) = f⁽ⁿ⁺¹⁾(c)/n! (x-c)ⁿ(x-a)

Inverse function theorem:
  f differentiable, f'(a)≠0 → f⁻¹ differentiable at f(a)
  (f⁻¹)'(f(a)) = 1/f'(a)

Darboux's theorem:
  Derivatives have intermediate value property (even without continuity)

---

Riemann Integration

Partition P = {a=x₀<x₁<...<xₙ=b}
Lower sum: L(f,P) = Σ mᵢΔxᵢ  (mᵢ = inf f on [xᵢ₋₁,xᵢ])
Upper sum: U(f,P) = Σ Mᵢ Δxᵢ  (Mᵢ = sup f on [xᵢ₋₁,xᵢ])

Riemann integrable: sup L(f,P) = inf U(f,P) = ∫ₐᵇ f(x)dx
Equivalent: ∀ε>0 ∃P: U(f,P) - L(f,P) < ε

Riemann's criterion:
  f integrable ↔ set of discontinuities has measure zero
  (Lebesgue criterion)
  Continuous → integrable
  Monotone → integrable
  Bounded with finitely many discontinuities → integrable

Fundamental Theorem of Calculus:
  Part 1: F(x) = ∫ₐˣ f(t)dt → F'(x) = f(x) at continuity points
  Part 2: ∫ₐᵇ f(x)dx = F(b)-F(a) if F'=f

Improper integrals:
  ∫ₐ^∞ f(x)dx = lim_{b→∞} ∫ₐᵇ f(x)dx
  ∫ₐᵇ f (f unbounded): lim_{c→a⁺} ∫ᶜᵇ f
  Comparison test: 0≤f≤g, ∫g < ∞ → ∫f < ∞

Riemann vs Lebesgue:
  Riemann integrates by partitioning domain
  Lebesgue integrates by partitioning range
  Lebesgue is more powerful: integrates more functions
  Agrees with Riemann for Riemann-integrable functions

---

Sequences of Functions

Pointwise convergence: fₙ → f pointwise if fₙ(x) → f(x) for each x
Uniform convergence: fₙ → f uniformly if sup_x|fₙ(x)-f(x)| → 0

Pointwise ← Uniform (uniform implies pointwise, not conversely)
Counter-example: fₙ(x) = xⁿ on [0,1]
  Pointwise: f(x) = 0 for x∈[0,1), f(1)=1
  Not uniform: discontinuous limit of continuous functions!

Uniform convergence preserves:
  Continuity: fₙ continuous + uniform convergence → f continuous
  Integrability: ∫fₙ → ∫f
  Differentiability: if fₙ' uniformly converge and fₙ converge at one point
    → fₙ → f uniformly and fₙ' → f'

Weierstrass M-test:
  |fₙ(x)| ≤ Mₙ and Σ Mₙ < ∞ → Σfₙ converges uniformly and absolutely

Power series Σaₙ(x-a)ⁿ:
  Radius of convergence R = 1/lim sup|aₙ|^(1/n)
  Converges absolutely on (a-R, a+R)
  Uniformly on [a-r, a+r] for any r < R
  Can differentiate and integrate term by term inside interval

Equicontinuity:
  Family F equicontinuous: ∀ε∃δ: |x-y|<δ → |f(x)-f(y)|<ε (same δ for all f∈F)
  Arzelà-Ascoli theorem: uniformly bounded equicontinuous family has uniformly convergent subsequence

---

Measure Theory

Sigma-algebra on X:
  Collection M of subsets: X∈M, closed under complement and countable union
  (X,M): measurable space

Measure μ: M → [0,∞]
  μ(∅) = 0
  Countable additivity: μ(∪ disjoint Aₙ) = Σμ(Aₙ)
  (X,M,μ): measure space

Lebesgue measure on ℝ:
  m([a,b]) = b-a  (length of interval)
  Extends uniquely to all Borel sets
  Null sets (measure zero): countable sets, Cantor set
  f = g a.e. (almost everywhere): f(x)=g(x) except on null set

Measurable functions:
  f: X → ℝ measurable if f⁻¹(B) ∈ M for all Borel B ⊆ ℝ
  Continuous functions measurable
  Monotone functions measurable
  Limit of measurable functions is measurable

Lebesgue integral:
  Simple functions: s = Σaᵢ1_{Aᵢ}: ∫s dμ = Σaᵢμ(Aᵢ)
  Nonneg: ∫f = sup{∫s: 0≤s≤f, s simple}
  General: ∫f = ∫f⁺ - ∫f⁻ (if at least one finite)
  Integrable (f∈L¹): ∫|f| < ∞

Lᵖ spaces:
  Lᵖ = {f measurable: ∫|f|ᵖ < ∞}
  ||f||_p = (∫|f|ᵖ)^(1/p)  (norm)
  L² = Hilbert space with inner product ⟨f,g⟩ = ∫fg
  L∞ = essentially bounded functions, ||f||_∞ = ess sup|f|
  Hölder: ||fg||₁ ≤ ||f||_p ||g||_q (1/p+1/q=1)
  Minkowski: ||f+g||_p ≤ ||f||_p + ||g||_p

---

Convergence Theorems (Lebesgue)

Monotone Convergence Theorem (MCT):
  0 ≤ f₁ ≤ f₂ ≤ ..., fₙ → f pointwise a.e.
  → ∫fₙ → ∫f  (including ∫f = ∞)

Fatou's Lemma:
  fₙ ≥ 0 measurable:
  ∫(lim inf fₙ) ≤ lim inf ∫fₙ

Dominated Convergence Theorem (DCT):
  fₙ → f pointwise a.e.
  |fₙ| ≤ g for all n, g ∈ L¹
  → ∫fₙ → ∫f  (and ∫|fₙ-f| → 0)

Applications of DCT:
  Differentiation under integral: d/dt∫f(x,t)dx = ∫∂f/∂t dx (if bounded)
  Series integration: Σ∫fₙ = ∫Σfₙ (if Σ∫|fₙ| < ∞)

Comparison Riemann vs Lebesgue:
  Lebesgue handles:
    Limits of functions
    L∞ with indicator functions
    Functions like sin(x)/x (conditionally but not absolutely integrable)
  Dirichlet: Lebesgue integrable (∫₀¹1_ℚ = 0)
    Not Riemann integrable (L≠U for any partition)

---

Metric & Normed Spaces

Normed vector space (V, ||·||):
  ||v|| ≥ 0, = 0 ↔ v=0
  ||αv|| = |α|||v||
  ||u+v|| ≤ ||u||+||v||
  Induces metric: d(u,v) = ||u-v||

Banach space: complete normed vector space
  Cauchy sequences converge
  Examples: ℝⁿ, Lᵖ(μ), C[a,b] with sup norm, ℓᵖ (sequence spaces)
  Not Banach: C[a,b] with L¹ norm

Hilbert space: complete inner product space
  ⟨u,v⟩: bilinear (or sesquilinear for complex), ⟨v,v⟩ ≥ 0
  ||v||² = ⟨v,v⟩
  Examples: L²(μ), ℓ², ℝⁿ
  Cauchy-Schwarz: |⟨u,v⟩| ≤ ||u|| ||v||
  Pythagorean: u⊥v → ||u+v||² = ||u||²+||v||²
  Parallelogram: ||u+v||²+||u-v||² = 2(||u||²+||v||²)

Orthonormal basis {eₙ}:
  ⟨eₘ,eₙ⟩ = δₘₙ
  Bessel: Σ|⟨f,eₙ⟩|² ≤ ||f||²
  Parseval: Σ|⟨f,eₙ⟩|² = ||f||² (complete orthonormal set)
  f = Σ⟨f,eₙ⟩eₙ (convergence in L² norm)

Bounded linear operators:
  T: V→W linear, ||T|| = sup{||Tv||/||v||: v≠0} < ∞
  Continuous ↔ bounded (for linear maps)
  Dual space V* = {bounded linear functionals V→ℝ}

Riesz representation theorem:
  Hilbert space H: every L∈H* has form L(f) = ⟨f,g⟩ for unique g∈H
  L²(μ)* ≅ L²(μ) via ⟨f,g⟩ = ∫fg dμ

---

Important Counterexamples

def key_counterexamples():
    return {
        'Continuous but not differentiable': {
            'function':     'Weierstrass: Σ aⁿcos(bⁿπx) for a<1, ab>1+3π/2',
            'property':     'Continuous everywhere, differentiable nowhere'
        },
        'Differentiable but derivative not continuous': {
            'function':     'f(x) = x²sin(1/x) for x≠0, f(0)=0',
            'property':     'f\'(0)=0 but f\'(x) oscillates near 0'
        },
        'Uniform convergence fails': {
            'function':     'fₙ(x) = xⁿ on [0,1]',
            'property':     'Pointwise to discontinuous limit, not uniform'
        },
        'Riemann not Lebesgue integrable': {
            'function':     '1_ℚ (Dirichlet function)',
            'property':     'Lebesgue integral = 0; no Riemann integral'
        },
        'Cantor set': {
            'property':     'Closed, uncountable, measure zero, nowhere dense, perfect',
            'construction': 'Remove middle thirds iteratively from [0,1]'
        },
        'Volterra function': {
            'property':     'Differentiable everywhere, bounded derivative, but FTC fails',
            'lesson':       'Need absolutely continuous for FTC'
        }
    }

---

Key Theorems Summary

Bolzano-Weierstrass: bounded sequence has convergent subsequence
Heine-Cantor: continuous on compact → uniformly continuous
IVT: continuous f on [a,b] takes all intermediate values
EVT: continuous f on [a,b] attains max and min
MVT: f'(c) = (f(b)-f(a))/(b-a) for some c
Taylor: f(x) = Σf⁽ᵏ⁾(a)/k!(x-a)ᵏ + remainder
FTC: derivative of integral is function; integral of derivative is change
DCT: dominated convergence allows limit inside integral
MCT: monotone convergence theorem
Arzelà-Ascoli: bounded equicontinuous family has convergent subsequence
Riesz representation: L²* ≅ L²
Baire category: complete metric space not countable union of nowhere dense sets
Stone-Weierstrass: polynomials dense in C[a,b]

---

Common Pitfalls

Pitfall	Fix
Pointwise convergence preserves continuity	Need UNIFORM convergence to preserve continuity
Differentiable → continuously differentiable (C¹)	Darboux: derivative has IVP but need not be continuous
Compact = bounded	In ℝⁿ: compact ↔ closed AND bounded (Heine-Borel); infinite dimensions: need more
Riemann and Lebesgue always agree	Agree for Riemann integrable functions; Lebesgue integrates more
Limit and integral always interchange	Need uniform convergence OR dominated convergence theorem
Absolute convergence = convergence	Absolute convergence is stronger; conditionally convergent can be rearranged

---

Related Skills

• calculus-expert: Computational calculus (less rigorous)
• complex-analysis-expert: Complex version of analysis
• topology-expert: Metric space topology
• probability-expert: Measure-theoretic probability
• differential-equations-expert: Analysis applied to ODEs/PDEs
• functional-analysis: Advanced operator theory