Proving Sequence and Series Convergence
To prove a_n → L:
- Compute
|a_n - L|and simplify algebraically until you get an expression likeC/n^kor similar. - Solve
C/n^k < εfornto find how bignmust be — this is scratch work, not part of the final proof. - Define
Nexplicitly in terms ofε(commonlyN = ⌈C/ε⌉orN = ⌈(C/ε)^(1/k)⌉). - Write the formal proof: "Given
ε > 0, letN = .... Then for alln > N, [algebra] so|a_n - L| < ε. Hencea_n → L."
Example: Prove n/(2n+1) → 1/2.
|n/(2n+1) - 1/2| = |1/(4n+2)| = 1/(4n+2)
Scratch: want 1/(4n+2) < ε ⟺ n > (1/ε - 2)/4, so N = ⌈1/(4ε)⌉ suffices.
Proof: Given ε>0, let N = ⌈1/(4ε)⌉. For n>N, 4n+2 > 4N > 1/ε,
so |n/(2n+1) - 1/2| = 1/(4n+2) < ε. Hence n/(2n+1) → 1/2.
Progress checklist for any convergence problem:
- Identify what type of claim this is: sequence converges to L / sequence diverges / sequence diverges to ∞ / series converges / series diverges
- Do scratch work to find the limit or behavior first (never skip this — you must "know the answer" before writing the proof)
- Choose the matching proof template (below)
- Write the final proof using only forward logical steps (no "we want" language in the final writeup)
- Sanity check: as ε shrinks, does N grow? (reality check for convergence proofs)
Proof templates by claim type
A. a_n → L (direct ε-N proof).
Simplify |a_n - L| to a clean bound, invert the inequality to solve for N, then write the proof forward: assume n > N, derive |a_n - L| < ε.
B. Sequence diverges (bounded oscillation, e.g., (-1)^n).
Proof by contradiction:
- Assume
a_n → Lfor someL. - Pick a concrete small
ε(oftenε = 1/2or based on known gap between terms, e.g.ε = d/3for terms a known distancedapart). - Get
Nfrom the definition; examine two sub-cases ofn > N(e.g., even vs. odd) that force contradictory bounds onL. - Conclude
L < L, a contradiction — hence no limit exists.
C. Sequence diverges because unbounded (e.g., a_n = n, a_n = n² - n).
Either:
- Diverges to infinity: given
M > 0, exhibitNsuch thatn > N ⇒ a_n > M(same mechanics as the ε-N proof but withMinstead ofε). - Diverges (not to infinity, not oscillating but unbounded and non-monotonic): contradiction argument similar to template B, using the fact convergent ⇒ bounded, or forcing
L < L.
D. Using limit laws instead of raw ε-N.
Break the expression into pieces that each converge using algebraic manipulation (factor out dominant terms, e.g., divide num/denom by highest power of n), then invoke: sum rule, scalar multiple rule, product rule, quotient rule (denominator limit ≠ 0). State which rule justifies each step explicitly.
E. Series convergence via direct summation.
- Telescoping: rewrite
a_nasb_n - b_{n+1}. Partial sum collapses:s_n = b_1 - b_{n+1}. Take limit ofb_{n+1}. - Geometric:
s_n = (1 - r^{n+1})/(1-r)via the standards_n - r·s_ntrick. Converges iff|r|<1, giving sum1/(1-r).
F. Series convergence tests (when exact sum is not needed/possible).
- Necessary condition first: check
a_n → 0. If not, series diverges immediately (contrapositive of "convergent ⇒ terms→0"). This never proves convergence, only rules it out. - Comparison test: find
0 ≤ a_n ≤ b_n. IfΣb_nconverges, so doesΣa_n; ifΣa_n = ∞, so doesΣb_n. Construct the comparison series from the dominant behavior ofa_nfor largen. - Ratio test: compute
r_n = a_{n+1}/a_n, findL = lim|r_n|.L<1⇒ absolute convergence;L>1⇒ divergence;L=1⇒ inconclusive, fall back to another test. - Integral test: when
fis positive, decreasing, continuous andf(n)=a_n, convergence ofΣa_nmatches convergence of∫₁^∞ f(x)dx(use for p-series:Σn^pconverges iffp < -1). - Absolute convergence: if
Σ|a_n|converges,Σa_nconverges (stronger, easier to apply than conditional tests).
- Triangle inequality, both forms:
|a+b| ≤ |a|+|b|and|a+b| ≥ |a|-|b|. Used constantly to split and bound. - Add-and-subtract trick: insert
±(same term)to split a hard expression into two manageable pieces, e.g.s_nt_n - st = s_n(t_n-t) + t(s_n-s). - Convergent ⇒ bounded: use this to get a uniform bound
Mon a sequence when proving product/ratio limit laws. - Uniqueness of limits: proof by contradiction assuming two limits
K ≠ L, pickingε = |K-L|/3. - Monotone bounded sequence theorem: bounded + increasing (or decreasing) ⇒ convergent, via completeness axiom (sup/inf). Use when an explicit limit is hard to guess but monotonicity and boundedness are easy to show.
- Completeness axiom / sup-inf: invoke when you need existence of a limit without knowing its value explicitly (e.g. constructing √2, proving Archimedean property).
Example 1 (ε-N, polynomial ratio):
Input: Prove (4n²-3)/(5n²-2n) → 4/5.
Output: Factor: = (4 - 3/n²)/(5 - 2/n). By limit laws, numerator → 4, denominator → 5 (denominator nonzero eventually), so quotient → 4/5 by the quotient rule.
Example 2 (divergence by contradiction, oscillating):
Input: Prove {(-1)^n} diverges.
Output: Assume (-1)^n → L. Let ε=1/2. Find N; take even k>N giving 1/2<L<3/2, odd ℓ>N giving -3/2<L<-1/2. Combined: L<L, contradiction. Diverges.
Example 3 (telescoping series):
Input: Evaluate Σ 1/(n(n+1)) for n≥1.
Output: 1/(n(n+1)) = 1/n - 1/(n+1). Partial sum s_n = 1 - 1/(n+1) → 1. Series converges to 1.
Example 4 (ratio test):
Input: Determine convergence of Σ n/3^n.
Output: r_n = a_{n+1}/a_n = (n+1)/(3n) → 1/3 < 1. By the ratio test, converges absolutely.
Example 5 (p-series via integral test):
Input: Determine convergence of Σ 1/n^3.
Output: f(x)=x^{-3} positive, decreasing, continuous. ∫₁^∞ x^{-3}dx converges (p=-3<-1). Hence series converges.
- Always separate scratch work (solving for N, guessing the limit) from the final proof (forward deduction only). State this distinction when writing up solutions.
- When choosing
N, prefer ceiling expressions⌈...⌉to guarantee an integer satisfying strict inequalities. - Do a "reality check": smaller
εshould require largerN. If not, recheck algebra. - For divergence-by-oscillation proofs, pick
εsmall enough to separate the suspected cluster points (often half their distance or a third of it, to leave strict inequality room). - For series, always check the necessary condition (
a_n → 0) before attempting harder convergence tests — it's the fastest way to rule out convergence. - When bounding for the comparison test, construct the comparison sequence from the dominant term's asymptotic behavior (e.g.,
1/(n-√n)compare to1/n-type harmonic bound). - Explicitly name which limit law (sum/product/quotient/scalar) justifies each algebraic step when using limit laws instead of raw ε-N arguments.
- Don't present scratch work as if it were the proof — "we want 1/n < ε" is not a valid proof step; inequalities should flow forward from the choice of N.
- Don't forget to check the denominator's limit is nonzero before applying the quotient rule.
- Don't conclude convergence just because
a_n → 0; this is necessary, not sufficient (classic trap: harmonic series). - Don't apply the ratio test when
lim|r_n| = 1— it is inconclusive, not evidence of divergence or convergence. - Don't mix up "diverges" with "diverges to infinity" — the latter is a stronger, more specific claim requiring the
M-Ndefinition, not just "no limit exists." - When proving a sequence is unbounded/oscillating diverges, don't forget both sub-cases (e.g. even/odd, or whatever partition is relevant) — a contradiction needs both bounds on
Lto collide.