SigmaField
Mathematical Foundations for Quant Finance
0 of 256 lessons

Proof Gym PG4 ε–δ limits Lesson 4 of 6

ProofLesson PG4.4 in the Proof Gym, stage PG4: ε–δ limits. About 40 minutes.

Supremum, infimum and completeness

Simulate

Check the claim with numbers

Take the set of the numbers 1−1n1 - \frac{1}{n} for whole numbers n≥1n \ge 1: 0, 12\frac{1}{2}, 23\frac{2}{3}, 34\frac{3}{4}, and so on. Every element is below 1. Does the set have a largest element, and what is the smallest number above all of it?

⌘/Ctrl + Enter runs the cell

Among the first 100,000 elements, the largest is 99999100000\frac{99999}{100000}, still below 1, and the smallest is 0. Yet every number tried just below 1 is passed: an element exceeds 1−0.0011 - 0.001 from n=1001n = 1001 on. So 1 looks like the smallest number above the set, though no element reaches it.

Some sets have no such number, since nothing lies above all of 1, 2, 3 and so on. And the smallest number above a set need not be a fraction. Bisection (lesson 1.3.2) closes in on the top of the set of xx with x2<2x^2 < 2:

⌘/Ctrl + Enter runs the cell

After 50 halvings, lo and hi agree with 2=1.4142135623730951\sqrt{2} = 1.4142135623730951 to about 15 digits. But 2\sqrt{2} is not a fraction (stage PG2 proves it), so with fractions alone this set would have no smallest number above it. The real numbers fill such gaps, and this lesson says how.

Conjecture

Every element of the set

A={1−1n:n≥1}A = \left\{ 1 - \frac{1}{n} : n \ge 1 \right\}

is below 1, and every number below 1 is passed by some element of AA. So 1 is the smallest number above AA, and AA has no largest element.

One hundred thousand elements stay below 1, and three numbers just below 1 are passed. A proof must handle every element and every number below 1.

Upper bounds and the supremum

The braces build a set: AA is the set of all numbers 1−1n1 - \frac{1}{n} for whole numbers n≥1n \ge 1. An upper bound of a set AA is a number uu with a≤ua \le u for every element aa. If AA has one, it is bounded above, and then it has many: anything above an upper bound is one too.

The supremum of AA, written sup⁡A\sup A, is its least upper bound: an upper bound that is at most every other upper bound. Lower bounds and the infimum inf⁡A\inf A, the greatest lower bound, are the same ideas turned upside down.

Rule: Supremum

s=sup⁡A means:a≤s for every a∈A, ands≤u for every upper bound u\begin{gathered} s = \sup A \text{ means:} \\ a \le s \text{ for every } a \in A, \text{ and} \\ s \le u \text{ for every upper bound } u \end{gathered}

The symbol ∈\in reads "is in". A supremum need not belong to AA. When it does, it is the maximum of AA, its largest element.

For the set of the conjecture, 2, 1.5 and 1 are upper bounds. But 0.9 is not, because the element 1−1111 - \frac{1}{11} is above it: the simulation's first element above 1−0.11 - 0.1.

The ε-test

The second half of the definition is easier to check in another form. An upper bound ss is the supremum exactly when every number below ss is passed by some element.

Rule: The ε-test for a supremum

an upper bound s is sup⁡A whenfor every ε>0 some a∈Ahas a>s−ε\begin{gathered} \text{an upper bound } s \text{ is } \sup A \text{ when} \\ \text{for every } \varepsilon > 0 \text{ some } a \in A \\ \text{has } a > s - \varepsilon \end{gathered}

The test works both ways: a supremum always passes it, and an upper bound that passes it is the supremum.

Here is why. Suppose the upper bound ss passes the test, and u<su < s. Take ε=s−u>0\varepsilon = s - u > 0: some element is above s−ε=us - \varepsilon = u, so uu is not an upper bound, and ss is the least.

Conversely, let s=sup⁡As = \sup A and ε>0\varepsilon > 0. Then s−εs - \varepsilon is below the least upper bound, so it is not an upper bound, and some element is above it.

Completeness

Rule: Completeness

every non-empty setof real numberswith an upper boundhas a supremum\begin{gathered} \text{every non-empty set} \\ \text{of real numbers} \\ \text{with an upper bound} \\ \text{has a supremum} \end{gathered}

We take this as a basic property of the real numbers, as lesson PG3.1 took well-ordering for the whole numbers. The fractions alone lack it, as the set of xx with x2<2x^2 < 2 suggests.

Completeness says a supremum exists without saying what it is. Lesson PG4.5 uses exactly that, to prove that a sequence converges before its limit is known.

Format 2

Fill the gap

The proof of the conjecture, with its key steps left out. Type each missing piece in nn, or as a number, then check.

1 is an upper bound. For every n≥1n \ge 1, 1−(1−1n)=1 - \left(1 - \frac{1}{n}\right) = , which is positive. So every element of AA is below 1.

Nothing smaller is an upper bound. Let ε>0\varepsilon > 0. The Archimedean property, proved at the end of this lesson, gives a whole number n>1εn > \frac{1}{\varepsilon}, so 1n<ε\frac{1}{n} < \varepsilon and 1−1n>1−ε1 - \frac{1}{n} > 1 - \varepsilon. By the ε-test, sup⁡A=1\sup A = 1.

For ε=0.001\varepsilon = 0.001, the first such whole number is n=n = , as the simulation found.

No largest element. Each element is passed by the next one, which is bigger by (1−1n+1)−(1−1n)=\left(1 - \frac{1}{n + 1}\right) - \left(1 - \frac{1}{n}\right) = . That is positive, so no element is the largest, and the supremum 1 is not in AA.

The infimum. The elements increase, so the first one, 1−11=1 - \frac{1}{1} = , is the smallest. So inf⁡A=0\inf A = 0, and here the infimum is in AA: it is the minimum.

How to type answers
  • Powers with ^, brackets round a power that is a sum: r^k, r^(k+1).
  • Fractions with / and brackets: (1-r)/(1+r). Multiply with * or side by side: 2k+1.
  • Press Enter in any box to check them all.

Format 3

Find the error

A proof that every non-empty set with an upper bound has a largest element. The claim is false, so some line must be wrong. Select it.

Simulate

A claim of your own to prove

Lesson PG4.1 chose each NN with the Archimedean property: for every real number xx there is a whole number n>xn > x. For any particular xx, Python finds one with math.floor, which rounds down to a whole number.

⌘/Ctrl + Enter runs the cell

Even for xx just above 101510^{15}, the whole number 1000000000000001 is larger. In the form lesson PG4.1 used: for ε=10−6\varepsilon = 10^{-6}, n=1000001n = 1000001 has 1n<ε\frac{1}{n} < \varepsilon.

Conjecture

For every real number xx there is a whole number nn with n>xn > x.

Python finds a whole number above each xx it is given. The claim is that one exists for every real number, and its proof needs completeness.

Format 4

Write it

Prove the Archimedean property from completeness. Argue by contradiction: suppose that some real number xx has n≤xn \le x for every whole number nn.

Saved in this browser as you type.

Write your own first: a proof you have tried to write teaches far more than one you only read.

The form lesson PG4.1 used follows at once. For ε>0\varepsilon > 0, take a whole number n>1εn > \frac{1}{\varepsilon}; then 1n<ε\frac{1}{n} < \varepsilon.

Summary card

  • An upper bound uu of AA has a≤ua \le u for every a∈Aa \in A. The supremum sup⁡A\sup A is the least upper bound; the infimum inf⁡A\inf A is the greatest lower bound.
  • The ε-test: an upper bound ss is sup⁡A\sup A exactly when, for every ε>0\varepsilon > 0, some element is above s−εs - \varepsilon.
  • Completeness: every non-empty set of real numbers with an upper bound has a supremum. It need not be in the set: sup⁡{1−1n}=1\sup\left\{1 - \frac{1}{n}\right\} = 1.
  • The Archimedean property: for every real xx some whole number exceeds xx. Its proof needs completeness.

Free further reading

Optional. This lesson covers what you need; these add more examples and practice. All are free to read.