JOVANA
Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

What a Limit Really Means: Epsilon and Delta

The definition that makes calculus rigorous. We turn the vague “gets close to” into a precise challenge–response game between two numbers, [[epsilon-delta|epsilon and delta]]. You name a tolerance; I must hit it. We work a full [[epsilon-n-definition|epsilon–N]] limit and a full [[epsilon-delta-function|epsilon–delta]] limit, line by line.

Why “gets close to” isn't good enough

We say 1/n → 0. Intuitively the terms 1, 1/2, 1/3, ... “get close to” 0. But so do the terms of 1, 1/2, 1/3, ... if we claim they get close to −0.0001 — they hover near it for a while. The phrase “gets close to” cannot tell a true limit from a near-miss, and it never says *how* close or *how soon*. We need a definition with no wiggle room.

The sequence definition, worked

Formally, the epsilon–N definition reads: a_n → L means that for every ε > 0 there exists a cutoff N such that for all n > N we have |a_n − L| < ε. Read it as a two-player game. The challenger names ε; I must produce an N that works. Let us actually win the game for 1/n → 0.

Claim:   lim_{n->inf} 1/n = 0.

Goal (the definition):  for every eps > 0, find N so that
                        n > N  ==>  |1/n - 0| < eps.

Scratch work (find N).  We want  |1/n - 0| = 1/n < eps.
  Since n > 0,  1/n < eps  is the same as  n > 1/eps.
  So any N at least 1/eps will do. Pick N = 1/eps.

Clean proof.
  Let eps > 0 be arbitrary.        <- challenger hands us eps
  Choose N = 1/eps.                <- our response
  Suppose n > N = 1/eps.
  Then  n > 1/eps,  so  1/n < eps.
  And since n > 0,   |1/n - 0| = 1/n < eps.        QED for this eps.

The N depended on eps (smaller eps forces larger N) -- exactly right.
Because eps was ARBITRARY, the promise holds for ALL eps. Limit = 0.

Contrast the FALSE claim  1/n -> -0.0001 :
  take eps = 0.0001. We would need |1/n - (-0.0001)| = 1/n + 0.0001 < 0.0001,
  i.e. 1/n < 0, impossible. The promise FAILS for this one eps -> claim dead.
A complete epsilon–N proof. The order of the quantifiers is the whole game: ε comes first (the challenge), then N is chosen to answer it.

The function definition: delta answers epsilon

For a function, “close” has to be controlled on both axes, so the cutoff N is replaced by a radius δ on the input. The epsilon–delta definition of a function limit: lim_{x→a} f(x) = L means for every ε > 0 there is a δ > 0 such that 0 < |x − a| < δ forces |f(x) − L| < ε. You demand the output be within ε of L; I find a window of width δ around a inside which I can keep that promise.

Claim:   lim_{x->3}  (2x + 1)  =  7.

Goal:  for every eps > 0, find delta > 0 so that
       0 < |x - 3| < delta  ==>  |(2x+1) - 7| < eps.

Scratch work.  Simplify the output gap:
  |(2x+1) - 7| = |2x - 6| = 2|x - 3|.
  We want  2|x - 3| < eps,  i.e.  |x - 3| < eps/2.
  So choosing delta = eps/2 should work.

Clean proof.
  Let eps > 0 be arbitrary.
  Choose delta = eps/2 > 0.
  Suppose 0 < |x - 3| < delta = eps/2.
  Then |(2x+1) - 7| = 2|x - 3| < 2 * (eps/2) = eps.        QED.

The input window (delta) shrinks in proportion to the demanded output
precision (eps). That proportionality IS the slope 2 made rigorous.
An epsilon–delta proof for a function limit. We reverse-engineer δ from ε in scratch work, then present the clean forward argument.

Two habits are now visible and worth keeping for life. First, *scratch work runs backwards* (start from the goal |f(x) − L| < ε and solve for the input window), but the *written proof runs forwards* (declare δ, then derive the conclusion). Second, the answer δ is allowed to depend on ε — that dependence is the quantitative heart of the limit concept. Master this game and every later theorem of analysis becomes a longer version of the same move.