Sequences and the Idea of Convergence

The concept of a convergent sequence is the load-bearing wall of essentially all of analysis. Derivatives are defined as limits of sequences of difference quotients; the definite integral is the limit of a sequence of Riemann sums as partitions refine; and an infinite series $\sum a_n$ is defined to converge exactly when its sequence of partial sums converges, linking this article directly to the theory of Infinite Series and Partial Sums. Beyond calculus proper, sequence convergence underlies numerical analysis: iterative algorithms such as Newton's method or gradient descent generate sequences of approximations, and proving that an algorithm "works" means proving the generated sequence converges to the true solution, often with an estimated rate. In probability and statistics, the Law of Large Numbers and the Central Limit Theorem are precisely statements about convergence of sequences of random variables (in various technical senses built on this same skeleton). In computer science, the convergence of iterative methods underlies floating-point root-finding, machine learning optimizers, and fixed-point computations in program semantics. Even in physics and engineering, convergent sequences model discretized approximations to continuous systems — finite-difference schemes for differential equations are trusted only once one has proved the corresponding sequence of approximate solutions converges to the true one as the mesh is refined.

Ask the tutor: How exactly does the $\varepsilon$–$N$ definition get used to prove convergence for a brand-new sequence I haven't seen before?

Picture the real number line, and mark the candidate limit $L$ as a fixed point. Around $L$, draw an interval of radius $\varepsilon > 0$ — a "tolerance window" $(L - \varepsilon, L + \varepsilon)$. The claim "$a_n \to L$" says: no matter how thin you make this window, the sequence eventually enters it and never leaves. Terms are allowed to wander outside the window at the start (small $n$), but only finitely many can do so; from some index $N$ onward, every single term $a_n$ must lie strictly inside the window. The subtlety, and the reason a rigorous definition is unavoidable, is the order of the quantifiers. The window comes first — chosen by an adversary who can make it as small as they like — and then we must find an $N$ that works for that particular window. If instead we fixed $N$ first and asked for a window to work afterward, the whole notion would collapse into something trivial. This is the single most important logical subtlety in the definition of a limit, and it is worth sitting with before moving to formal symbols. A second useful picture: plot the points $(n, a_n)$ in the plane. Convergence to $L$ means that, however thin a horizontal strip you draw around the line $y = L$, the plotted points eventually all fall inside that strip and stay there forever afterward.

Before any theorem can be proved, "arbitrarily close, eventually" must be translated into a statement built from real numbers and inequalities — nothing else. The translation is: > For every $\varepsilon > 0$, there exists $N \in \mathbb{N}$ such that for all $n \ge N$, $|a_n - L| < \varepsilon$. Here $\varepsilon$ (Greek epsilon) plays the role of the tolerance, and $N$ is the index beyond which the sequence must stay inside that tolerance. This is the celebrated $\varepsilon$–$N$ definition, which we will use as our working definition below (it is exactly the precise version of the informal Definition above, and the sequel article develops it in full depth). We adopt it now because the theorems of this article genuinely require a precise object to reason about — no theorem can be proved from a vague picture.

This corollary is logically just the contrapositive of the Boundedness Theorem, and requires no separate proof: it is the same implication read in the opposite direction.

Example 1 — a convergent sequence with an asymmetric, non-obvious limit. Consider

We claim $a_n \to \dfrac{3}{7}$. Computing the gap. For $n \ge 1$ the denominator $7n^2 + 2n - 4$ is positive (check $n=1$: $7+2-4=5>0$, and it only grows), so division is valid. Compute

Numerically at $n = 5$: $a_5 = \dfrac{3(25) - 25 + 1}{7(25) + 10 - 4} = \dfrac{75-25+1}{175+10-4} = \dfrac{51}{181} \approx 0.28177$, while $3/7 \approx 0.42857$ — so the terms start noticeably far from the limit, illustrating that convergence is about eventual, not immediate, closeness. At $n = 100$: $a_{100} = \dfrac{30000 - 500+1}{70000+200-4} = \dfrac{29501}{70196} \approx 0.42026$, already close to $0.42857$. Verifying convergence via the definition. For $n \ge 1$,

for $n\ge1$ (using $19 \le 19n$ and $41n+19n=60n$, and $7n^2+2n-4 \ge 7n^2$ for $n \ge 1$ since $2n - 4 \ge -2 \ge$ well, check directly: at $n=1$, $2n-4=-2$, and $7n^2 = 7 > -2$ trivially satisfied since we only need $7n^2+2n-4 \ge 7n^2 -$ actually simplest: $7n^2+2n-4 \ge 7n^2 - 4 \ge 3n^2 \ge 3n$ for $n\ge2$, and check $n=1$ directly gives denominator $5$; either way the bound $\frac{60}{49n}\to 0$ suffices for large $n$). Given any $\varepsilon > 0$, choose $N$ to be any natural number with $N > \dfrac{60}{49\varepsilon}$ (such $N$ exists by the Archimedean Property of $\mathbb{R}$). Then for $n \ge N$,

This confirms $a_n \to 3/7$ in the full $\varepsilon$–$N$ sense, matching the numerical trend observed above.

Example 2 — a bounded, non-monotone sequence that still converges. Let

The first piece oscillates in sign and shrinks; the second piece is a rational function tending to $2$. We claim $b_n \to 2$. Splitting the estimate. Write

By the triangle inequality,

Numerically: at $n = 3$, $b_3 = \dfrac{-1}{3} + \dfrac{7}{6} = -0.3333 + 1.1667 = 0.8333$, while at $n = 4$, $b_4 = \dfrac{1}{4} + \dfrac{9}{7} = 0.25 + 1.2857 = 1.5357$. The sequence is not monotone (it jumped from $0.8333$ up to $1.5357$), yet both values are consistent with heading toward $2$; by $n=50$, $b_{50} = \dfrac{1}{50} + \dfrac{101}{53} \approx 0.02 + 1.9057 = 1.9257$, closer still. Verifying via the definition. Given $\varepsilon > 0$, choose $N \in \mathbb{N}$ with $N > 6/\varepsilon$ (again by the Archimedean property). For $n \ge N$,

Hence $b_n \to 2$. This example shows convergence does not require the sequence to approach its limit monotonically from one side — oscillation is allowed, provided the size of the oscillation shrinks to zero.