The Squeeze Theorem

Trapped between two functions sharing a limit.

One function caught between two others

Some limits cannot be found by substituting or by rewriting. Take f(x) = x² sin(1/x) as x → 0. At x = 0 the expression has no value, because 1/0 is undefined, and sin(1/x) cannot be factored away.

What can be said is that sin of any number lies between −1 and 1. So −1 ≤ sin(1/x) ≤ 1 for every x ≠ 0. Multiply through by x², which is positive for x ≠ 0, so the inequality signs stay as they are: −x² ≤ x² sin(1/x) ≤ x².

So the graph of f lies between the parabola y = x² above and the parabola y = −x² below, touching each wherever sin(1/x) is 1 or −1.

xy

The gold curve is y = x² sin(1/x), between the dashed parabolas y = x² and y = −x². Away from 0 it swings slowly; close to 0 it swings faster and faster, always inside the two parabolas.

The middle has nowhere else to go

The squeeze theorem: if g(x) ≤ f(x) ≤ h(x) for every x near c, except perhaps at c itself, and g(x) and h(x) both tend to the same limit L as x → c, then f(x) tends to L as well.

Here g(x) = −x² and h(x) = x², and both tend to 0 as x → 0. Every value of f lies between them, so f(x) is never further from 0 than x² is. At x = 0.1 the bounds are ±0.01, and f(0.1) = 0.01 × sin 10 = −0.00544. At x = 0.01 the bounds are ±0.0001, and f(0.01) = −0.0000506. Taking x close enough to 0 makes x², and with it every value of f, as close to 0 as you like.

So lim x² sin(1/x) as x → 0 is 0. The theorem says nothing about f at 0 itself, and needs nothing: f has no value there, and the limit is still 0.

δ = 0.6−δ² ≤ f(x) ≤ δ² for |x| ≤ δgap 2δ² = 0.72

δ = 0.6: −x² ≤ f(x) ≤ x², so inside the clamp f is within 0.72 of 0; closing the clamp closes the gap, and a function held between two things that meet must meet them too

Close the clamp to δ = 0.1 and read the gap

x² sin(1/x) in gold between the dashed parabolas y = x² and y = −x². The two upright bars clamp the strip |x| ≤ δ, and inside it every value of f lies between −δ² and δ². Close the clamp to δ = 0.1: the jaws are then 2δ² = 0.02 apart, and they close to nothing as δ does.

Endless swings, one limit

sin(1/x) is 0 whenever 1/x is a multiple of π, that is at x = 1/π, 1/(2π), 1/(3π) and so on: 0.318, 0.159, 0.106, 0.0796, and on toward 0. Between each pair f swings up to the upper parabola or down to the lower one. There are infinitely many of these swings between 0 and any positive x, so no drawing can show all of them.

The swinging never stops, but its size does: the swing at x is at most x² either way. A limit is about how far the values are from L, not about whether they stay still, so the swings do not prevent the limit 0.

xy

The same three curves close to the origin, from x = −0.4 to 0.4. The dashed parabolas pinch together at (0, 0), and the gold curve between them is pinched with them.

Both conditions are needed

The bounds must hold for every x near c. A bound that holds at a few points, or only far from c, says nothing about the limit.

The two bounds must also have the same limit. sin(1/x) itself lies between −1 and 1, but those bounds tend to −1 and 1, two different numbers, so they leave a gap of 2 for sin(1/x) to move in. It does: it is 1 at x = 1/(π/2 + 2kπ) and −1 at x = 1/(3π/2 + 2kπ) for every whole number k, as close to 0 as you like, so sin(1/x) has no limit at 0.

The product law cannot be used for x² sin(1/x) either. It needs both factors to have limits, and sin(1/x) has none. The squeeze avoids that factor’s limit altogether by bounding it.

A squeeze far out

The theorem works for x → ∞ too, with "near c" read as "for all large x". For x > 0, dividing −1 ≤ sin x ≤ 1 by the positive number x gives −1/x ≤ (sin x)/x ≤ 1/x. Both bounds tend to 0 as x → ∞, so (sin x)/x → 0. At x = 100 it is −0.00506, inside the bounds ±0.01.

The usual mistakes

Adding the two limits. If both bounds tend to 5, f tends to 5, not 10: f sits between them, and between 5 and 5 there is only 5.

Multiplying an inequality by a negative number without reversing it. Multiplying by x² is safe because x² is positive; multiplying −1 ≤ sin(1/x) ≤ 1 by a negative x would turn the signs round.

Concluding that there is no limit because the function oscillates. x² sin(1/x) swings forever near 0 and still tends to 0.

Using bounds with different limits, such as −1 and 1 for x² sin(1/x). They are true bounds but they prove nothing; −x² and x² are the bounds that close together.

A lathe and a tapering pin

The application below has exactly this function as the error in a machined surface, x² sin(1/x) mm at a distance x mm from the tip, and uses the same bound to say how far out the surface can be near the tip.

Worked example: The Ripple a Lathe Leaves Near the Tip of a Tapering Pin: An Error Trapped Between Two Curves

Question A lathe cuts a long tapering pin. At a distance x mm from the tip the surface is out by e(x) = x2 sin1x mm, a ripple whose spacing tightens toward the tip; at the tip itself the model has no value. (a) Use the squeeze theorem to find limx → 0 e(x). (b) Within 0.3 mm of the tip, give a bound on how far out the surface can be, using |sin| ≤ 1.

  1. 1.The sine of any number lies between −1 and 1, so −1 ≤ sin1x ≤ 1 for every x except 0. The awkward part of the model is inside those bounds, however fast it swings.

    −0.2−0.0900.090.200.10.20.30.4distance from the tip, x mmerror (mm)sin of anything is between −1 and 1
    −0.2−0.0900.090.200.10.20.30.4distance from the tip, x mmerror (mm)sin of anything is between −1 and 1
    However fast sin1x swings, it stays between −1 and 1.
  2. 2.Multiply the inequality through by x2. That is a positive number for x ≠ 0, so the inequality signs stay as they are: −x2 ≤ e(x) ≤ x2.

    −0.2−0.0900.090.200.10.20.30.4distance from the tip, x mmerror (mm)x2−x2multiply by x2, which is positive−x2at most e(x), e(x) at most x2
    −0.2−0.0900.090.200.10.20.30.4distance from the tip, x mmerror (mm)x2−x2multiply by x2, which is positive−x2at most e(x), e(x) at most x2
    Multiplying by x2, which is positive, keeps the inequality: −x2 ≤ e(x) ≤ x2.
  3. 3.Both bounding curves have the same limit at the tip: limx → 0 (−x2) = 0 and limx → 0 x2 = 0. The error is trapped between two curves that arrive at the same place.

    −0.2−0.0900.090.200.10.20.30.4distance from the tip, x mmerror (mm)x2−x2x2→ 0 and −x2→ 0 at the tip
    −0.2−0.0900.090.200.10.20.30.4distance from the tip, x mmerror (mm)x2−x2x2→ 0 and −x2→ 0 at the tip
    Both bounding curves tend to 0 at the tip, and the model itself has no value there.
  4. 4.(a) By the squeeze theorem, limx → 0 e(x) = 0. The ripple dies away at the tip: the model has no value at x = 0, but every value near it is near 0, so the surface is true there.

    −0.2−0.0900.090.200.10.20.30.4distance from the tip, x mmerror (mm)x2−x2both bounds → 0(a) squeezed, so the error → 0no value at the tip, but the limit is 0
    −0.2−0.0900.090.200.10.20.30.4distance from the tip, x mmerror (mm)x2−x2both bounds → 0(a) squeezed, so the error → 0no value at the tip, but the limit is 0
    (a) By the squeeze theorem limx → 0 e(x) = 0: the ripple dies away at the tip.
  5. 5.(b) The same bound answers this. For 0 < x ≤ 0.3, the error is at most x2, and x2 is at most 0.32 = 0.09. So the surface is out by at most 0.09 mm. Check: at x = 0.3 the model gives e(0.3) = 0.09 sin 3.33 = −0.017 mm, which is well inside the bound.

    −0.2−0.0900.090.200.10.20.30.4distance from the tip, x mmerror (mm)x2−x2at most 0.09 mmwithin 0.3 mm: error at most 0.3 × 0.3(b) at most 0.09 mm
    −0.2−0.0900.090.200.10.20.30.4distance from the tip, x mmerror (mm)x2−x2at most 0.09 mmwithin 0.3 mm: error at most 0.3 × 0.3(b) at most 0.09 mm
    (b) For 0 < x ≤ 0.3 the error is at most x2 ≤ 0.32 = 0.09 mm.

Answer: (a) limx → 0 e(x) = 0 mm, so the ripple dies away at the tip; (b) at most 0.09 mm

Common mistakes

  • Trying to find the limit by substituting x = 0. Then sin1x has nothing to be the sine of, since 10 is undefined, and the product law for limits needs both factors to have a limit. The squeeze theorem gets round the factor altogether by bounding it.
  • Saying that the limit does not exist because sin1x swings faster and faster near the tip. The swinging never stops, but its size is cut down by the x2 in front of it, and it is the size, not the swinging, that a limit is about.

More limits problems, worked step by step →

Practice The Squeeze Theorem in the app