Calculus · Limit of a sequence

Limit of a sequence

A sequence converges to L when, for every ε > 0, all its terms after some N lie within ε of L. Shrink the band and watch N grow; drag N lower to see which term breaks the rule.

The limit of a sequence in this visualization

A sequence a1,a2,a3,…a_1, a_2, a_3, \ldots converges to L when every band around L, however narrow, eventually holds all of its terms:

The ε–N definition
∀ε>0  ∃N:n>N  ⟹  ∣an−L∣<ε\forall \varepsilon > 0\ \ \exists N:\quad n > N \implies |a_n - L| < \varepsilon

On the plot ε is the half-width of the red band and N is the yellow line. Every term to the right of the line has to be strictly inside the band. A term on the band’s edge is not inside: for 1/n1/n and ε = 0.1, the term a10=0.1a_{10} = 0.1 is exactly 0.1 away, so N = 10, not 9.

When the distance to L only decreases, the smallest N is the last term still outside the band. For 1/n1/n, 1/n<ε1/n < \varepsilon exactly when n>1/εn > 1/\varepsilon, so N=⌊1/ε⌋N = \lfloor 1/\varepsilon \rfloor. When it goes up and down, a bound comes first: ∣sin⁡n/n∣≤1/n|\sin n / n| \le 1/n, so every term after 1/ε1/\varepsilon is inside, and the terms before it are checked one by one. For ε = 0.1 the bound gives 10 and the check finds 8. Any larger N works too; the page shows the smallest.

The definition only looks past N, so the first terms can do anything. n2/2nn^2/2^n climbs to a3=1.125a_3 = 1.125 before it falls, and still converges to 0. Changing finitely many terms never changes a limit.

As ε shrinks, the smallest N never decreases, but how fast it grows depends on the sequence. For 1/n1/n it is multiplied by 10 each time ε is divided by 10: 10, 100, 1000, 10000. For 0.8n0.8^n it grows by about 10, because 0.8n<ε0.8^n < \varepsilon once n>ln⁡ε/ln⁡0.8n > \ln\varepsilon / \ln 0.8, and dividing ε by 10 adds ln⁡10/ln⁡1.25≈10.3\ln 10 / \ln 1.25 \approx 10.3. On the log–log plot under the terms, 1/n1/n is a straight line of slope −1 and 0.8n0.8^n bends down ever more steeply. Both converge; the definition does not care how fast.

A sequence has at most one limit. If L and L′ are a distance d apart, bands of half-width d/2 around them do not overlap, and no tail lies in both. The terms of 1/n1/n get ever closer to −0.2, yet each stays more than 0.2 away, so −0.2 is not the limit. Being bounded is not enough either: (−1)n(-1)^n alternates between 1 and −1, and a band around L holds both only when ε>1+∣L∣\varepsilon > 1 + |L|. For ε ≤ 1 no L works, so (−1)n(-1)^n has no limit.

Things to notice

  • A limit need not be one of the terms: no term of 1/n1/n is 0.
  • “Tends to infinity” is a different definition: for every M there is an N with an>Ma_n > M for n > N. Such a sequence diverges and has no finite limit, even though textbooks write lim⁡an=∞\lim a_n = \infty.

Controls

  • Drag the yellow handle on the bottom scale to try a smaller N: the first term after it that lies outside the band is crossed and its value shown.
  • Drag the blue handle on the left scale to try another candidate limit. A press or a drag on the distance plot sets ε.

Related