Induction

Study Sheet

Induction

Proving a statement for every natural number

Ordinary Induction

Concept
Base case, inductive step
n₀n₀ + 1…base case: the first domino fallsstep: each falling domino knocks over the next
The base case starts the chain; the step makes every link hold.

To prove P(n)P(n) for all n≥n0n \ge n_0: prove P(n0)P(n_0) (base case); then prove that for every k≥n0k \ge n_0, P(k)P(k) implies P(k+1)P(k + 1) (inductive step). Together these knock over every domino.

Induction is the proof method built into the natural numbers: they are exactly the set you reach from 00 by adding 11 repeatedly, so a property that starts true and survives "+1+1" is true everywhere. The inductive step is not "assume what you are proving" — it is "assume it for one specific kk and prove it for the next one". Writing the inductive hypothesis out explicitly, with a specific kk, is what keeps that straight.

Example
Sum of the first nn odd numbers

Claim: 1+3+⋯+(2n−1)=n21 + 3 + \cdots + (2n - 1) = n^2 for all n≥1n \ge 1. Base: n=1n = 1: 1=121 = 1^2. Step: assume 1+3+⋯+(2k−1)=k21 + 3 + \cdots + (2k - 1) = k^2 for some k≥1k \ge 1. Adding the next odd number, 1+⋯+(2k−1)+(2k+1)=k2+2k+1=(k+1)21 + \cdots + (2k - 1) + (2k + 1) = k^2 + 2k + 1 = (k + 1)^2, which is the claim for k+1k + 1. ■\blacksquare

Example
A divisibility

Claim: 7∣8n−17 \mid 8^n - 1 for all n≥0n \ge 0. Base: 80−1=0=7⋅08^0 - 1 = 0 = 7\cdot0. Step: assume 7∣8k−17 \mid 8^k - 1. Then 8k+1−1=8⋅8k−1=8(8k−1)+78^{k+1} - 1 = 8\cdot8^k - 1 = 8(8^k - 1) + 7. The first term is divisible by 77 by the hypothesis and the second is 77, so 77 divides the whole. ■\blacksquare

Example
An inequality that starts late

Claim: 2n>n22^n > n^2 for all n≥5n \ge 5. Base: 25=32>252^5 = 32 > 25. Step: assume 2k>k22^k > k^2 for some k≥5k \ge 5. Then 2k+1=2⋅2k>2k2=k2+k2≥k2+5k>k2+2k+1=(k+1)22^{k+1} = 2\cdot2^k > 2k^2 = k^2 + k^2 \ge k^2 + 5k > k^2 + 2k + 1 = (k + 1)^2. ■\blacksquare

Two uses of k≥5k \ge 5 hide in the chain: k2≥5kk^2 \ge 5k needs k≥5k \ge 5, and 5k>2k+15k > 2k + 1 needs k≥1k \ge 1. The base case is 55 because the claim is false just below it: 24=16=422^4 = 16 = 4^2, 23=8<92^3 = 8 < 9, 22=4=222^2 = 4 = 2^2. An induction must start where the claim starts being true, and the step must be proved only for kk at or beyond that point.

Tip
The shape of the inductive step

Write the k+1k + 1 case; find the kk case inside it (peel off the last term, factor out one power); substitute the inductive hypothesis; finish with algebra. If you cannot find the kk case inside, you may need strong induction.

Side note
Where it goes wrong

Forgetting the base case — "P(k)⇒P(k+1)P(k) \Rightarrow P(k + 1)" is true of "n=n+1n = n + 1" and proves nothing without a start. And proving P(k+1)P(k + 1) from scratch without using P(k)P(k), which usually means induction was not needed.

Try it
Try it: induction

Prove that 2n>n2^n > n for all n≥1n \ge 1.

Show the answer

Base: 21=2>12^1 = 2 > 1. Step: assume 2k>k2^k > k for some k≥1k \ge 1. Then 2k+1=2⋅2k>2k=k+k≥k+12^{k+1} = 2\cdot2^k > 2k = k + k \ge k + 1, using k≥1k \ge 1 in the last step. ■\blacksquare

Strong Induction

Concept
Assume everything below

In strong induction the inductive hypothesis is "P(j)P(j) holds for ALL jj with n0≤j≤kn_0 \le j \le k", and you prove P(k+1)P(k + 1). It is needed when the case k+1k + 1 depends on cases further back than kk — factorizations, Fibonacci-type recurrences, games. It is logically equivalent to ordinary induction.

Why is it allowed to assume so much? By the time the dominoes reach k+1k + 1, every earlier domino has already fallen, not just the one before. Formally, strong induction on PP is ordinary induction on the statement Q(k)Q(k): "P(j)P(j) holds for every jj from n0n_0 to kk". Proving Q(k)⇒Q(k+1)Q(k) \Rightarrow Q(k + 1) is exactly proving P(k+1)P(k + 1) from all the earlier cases.

Example
Every n≥2n \ge 2 is a product of primes

Assume every jj with 2≤j≤k2 \le j \le k is a product of primes. If k+1k + 1 is prime, it is a product of one prime. Otherwise k+1=abk + 1 = ab with 2≤a,b≤k2 \le a, b \le k; each is a product of primes by hypothesis, and so is their product. ■\blacksquare (Ordinary induction fails here: the factors of k+1k + 1 are not kk.)

Example
Postage
12131415161718192021filled: the four base cases, checked by handarrow from v − 4 to v: add one 4-cent stamp
Each arrow reaches back four, so four base cases are needed before the arrows take over.

Every n≥12n \ge 12 is a sum of 44s and 55s. Bases: 12=4+4+412 = 4+4+4, 13=4+4+513 = 4+4+5, 14=4+5+514 = 4+5+5, 15=5+5+515 = 5+5+5. Step: for k+1≥16k + 1 \ge 16, (k+1)−4≥12(k + 1) - 4 \ge 12 is a sum of 44s and 55s by hypothesis; add a 44. ■\blacksquare (Four base cases because the step reaches back 44.)

Side note
Where it goes wrong

Too few base cases. If the step uses P(k−3)P(k - 3), the first four values must be checked by hand or the chain has a gap. With only the base 1212 above, the step would produce 16,20,24,…16, 20, 24, \ldots and never reach 1313.

Try it
Try it: strong induction

Define a1=1a_1 = 1, a2=3a_2 = 3 and an=an−1+2an−2a_n = a_{n-1} + 2a_{n-2} for n≥3n \ge 3. Prove that an=2n+1+(−1)n3a_n = \dfrac{2^{n+1} + (-1)^n}{3} for all n≥1n \ge 1.

Show the answer

Bases: n=1n = 1: 4−13=1\dfrac{4 - 1}{3} = 1; n=2n = 2: 8+13=3\dfrac{8 + 1}{3} = 3. Both match. Step: assume the formula for aka_k and ak−1a_{k-1}, with k≥2k \ge 2. Then

ak+1=ak+2ak−1=2k+1+(−1)k+2k+1+2(−1)k−13=2k+2−(−1)k3=2k+2+(−1)k+13,a_{k+1} = a_k + 2a_{k-1} = \frac{2^{k+1} + (-1)^k + 2^{k+1} + 2(-1)^{k-1}}{3} = \frac{2^{k+2} - (-1)^k}{3} = \frac{2^{k+2} + (-1)^{k+1}}{3},

which is the formula for n=k+1n = k + 1. ■\blacksquare Two bases are needed because the step uses the two previous terms. (Checking the bases first is also how you catch a wrong guess: the tempting 2n−(−1)n2^n - (-1)^n gives 33 at n=1n = 1, not 11.)

Where Inductions Go Wrong

Concept
Both halves, every time

An induction has two obligations and each fails in its own way. A false or missing base leaves a valid chain with nothing to start it. A step that fails for one value of kk breaks the chain at that link, and everything beyond is unproved. A circular step uses P(k+1)P(k + 1) to prove P(k+1)P(k + 1).

The step must hold for EVERY k≥n0k \ge n_0, because the chain from P(n0)P(n_0) to P(100)P(100) uses every link from n0n_0 to 9999. A step that "clearly works" for large kk but has a hidden requirement such as k≥2k \ge 2 proves nothing about the small cases, and the small cases are exactly where false claims hide. Test the step on the smallest kk it is supposed to cover.

Example
A valid step, a false theorem

"Claim": n2+n+1n^2 + n + 1 is even for every n≥1n \ge 1. The step is genuinely valid: (k+1)2+(k+1)+1=(k2+k+1)+2(k+1)(k + 1)^2 + (k + 1) + 1 = (k^2 + k + 1) + 2(k + 1), so if k2+k+1k^2 + k + 1 is even, so is the next value. But the base fails: n=1n = 1 gives 33. In fact n2+n+1=n(n+1)+1n^2 + n + 1 = n(n + 1) + 1 is always odd. Flaw: no base case — the chain was never started.

Example
All horses are the same color
n + 1 = 4n + 1 = 2first nlast noverlapno overlap
The overlap the argument needs exists for four horses and is empty for two.

"Claim": in any group of n≥1n \ge 1 horses, all have the same color. Base: one horse matches itself. Step: in a group of k+1k + 1, the first kk share a color and the last kk share a color; the horses in both groups link the two, so all k+1k + 1 match. Flaw: the step needs the two groups to overlap, and for k=1k = 1 (a group of two) the "first one" and "last one" are different horses with nothing in common. The step fails at exactly one link, from 11 to 22, and that is enough to break the whole chain.

Side note
A circular step

A write-up that starts "assume P(k+1)P(k + 1)" and simplifies until it reaches P(k)P(k), which is true, has proved P(k+1)⇒P(k)P(k + 1) \Rightarrow P(k) — the wrong direction. It can be rescued only if every step is reversible and the proof says so; the safe habit is to start from the k+1k + 1 expression and end at the k+1k + 1 claim.

Try it
Try it: all positive integers are equal

Find the flaw. "Let P(n)P(n): if aa and bb are positive integers with max⁡(a,b)=n\max(a, b) = n, then a=ba = b. Base n=1n = 1: then a=b=1a = b = 1. Step: if max⁡(a,b)=k+1\max(a, b) = k + 1, then max⁡(a−1,b−1)=k\max(a - 1, b - 1) = k, so a−1=b−1a - 1 = b - 1 by hypothesis, and a=ba = b."

Show the answer

The hypothesis P(k)P(k) is only about POSITIVE integers, and a−1a - 1 or b−1b - 1 may be 00. For a=1a = 1, b=2b = 2: max⁡(a,b)=2\max(a, b) = 2, but a−1=0a - 1 = 0 is not a positive integer, so P(1)P(1) says nothing about the pair (0,1)(0, 1). The step fails already at the first link, from k=1k = 1 to k=2k = 2.

Induction Beyond Formulas

Concept
Induct on size

Induction proves statements about sets, boards, polygons and games — anything with a size nn. The step: take an arbitrary object of size k+1k + 1, find inside it one or more objects of size at most kk, apply the hypothesis to them, and rebuild.

The word "arbitrary" carries the weight. The step must start from ANY object of size k+1k + 1 and break it down, not start from an object of size kk and build up, because building up only reaches the objects your construction happens to produce. Breaking down guarantees that every object of size k+1k + 1 is covered.

Example
Subsets of a set

Claim: a set with nn elements has 2n2^n subsets. Base: the empty set has exactly one subset, itself, and 20=12^0 = 1. Step: take any set with k+1k + 1 elements and pick one element xx. Subsets not containing xx are the subsets of the other kk elements: 2k2^k by hypothesis. Subsets containing xx are those same 2k2^k sets with xx added. Total 2k+2k=2k+12^k + 2^k = 2^{k+1}. ■\blacksquare

Example
Tiling with L-trominoes
removedone trominoat the center
Split into quadrants; one central tromino gives each of the other three quadrants its own missing square.

Claim: for every n≥1n \ge 1, a 2n×2n2^n \times 2^n board with any one square removed can be tiled by L-shaped pieces of three squares. Base n=1n = 1: a 2×22 \times 2 board minus one square IS an L. Step: cut a 2k+1×2k+12^{k+1} \times 2^{k+1} board with one square missing into four 2k×2k2^k \times 2^k quadrants. The missing square is in one quadrant. Place one L at the centre so that it covers the corner square nearest the centre in each of the other three quadrants. Now every quadrant is a 2k×2k2^k \times 2^k board with one square missing, and the hypothesis tiles each. ■\blacksquare

The single central tromino is the whole idea: it converts three complete quadrants into three copies of the smaller problem. A quick consistency check is that the count works out — 4n−14^n - 1 squares must be a multiple of 33, and it is, since 4≡14 \equiv 1 gives 4n−1≡0(mod3)4^n - 1 \equiv 0 \pmod 3.

Side note
Where it goes wrong

Building up. "Take a tiled 2k2^k board and attach three more quadrants" only proves the result for the boards that construction produces — here, boards whose missing square sits where you chose. Start from an arbitrary board of the larger size, with the missing square anywhere, and break it down.

Try it
Try it: counting diagonals

Prove by induction that a convex polygon with n≥3n \ge 3 vertices has n(n−3)2\dfrac{n(n - 3)}{2} diagonals.

Show the answer

Base n=3n = 3: a triangle has 0=3⋅020 = \tfrac{3\cdot0}{2} diagonals. Step: take any convex polygon with k+1k + 1 vertices and remove one vertex VV; what is left is a convex kk-gon with k(k−3)2\tfrac{k(k - 3)}{2} diagonals. The diagonals of the big polygon are: those, plus the k−2k - 2 diagonals from VV to its non-neighbours, plus the segment joining VV's two neighbours (a side of the kk-gon, a diagonal of the big one). Total

k(k−3)2+(k−2)+1=k2−k−22=(k+1)(k−2)2,\frac{k(k - 3)}{2} + (k - 2) + 1 = \frac{k^2 - k - 2}{2} = \frac{(k + 1)(k - 2)}{2},

the formula for k+1k + 1. ■\blacksquare

Why It Works

The well-ordering principle says every non-empty set of positive integers (or of integers ≥n0\ge n_0) has a smallest element. Suppose the base case and the step have both been proved, but P(n)P(n) fails for some n≥n0n \ge n_0. The failures form a non-empty set; let mm be its smallest element. Then m≠n0m \ne n_0, because the base case holds. So m−1≥n0m - 1 \ge n_0, and P(m−1)P(m - 1) is true because m−1m - 1 is smaller than the smallest failure. The step, applied with k=m−1k = m - 1, gives P(m)P(m) — contradicting the choice of mm. So there are no failures. ■\blacksquare

Given PP, define Q(k)Q(k): "P(j)P(j) holds for every jj with n0≤j≤kn_0 \le j \le k". The strong-induction step, "all of P(n0),…,P(k)P(n_0), \ldots, P(k) imply P(k+1)P(k + 1)", is exactly "Q(k)⇒Q(k+1)Q(k) \Rightarrow Q(k + 1)". So ordinary induction on QQ proves every Q(k)Q(k), and in particular every P(k)P(k). The two methods prove the same theorems; strong induction just lets you write the hypothesis you actually use.