Upload
lekhanh
View
226
Download
5
Embed Size (px)
Citation preview
Mathematical Induction
Introduction
Mathematical induction, or just induction, is a proof technique. Suppose that for every natural number π, π(π) is a statement. We wish to show that all statements π(π) are true.
In a proof by induction, we show that π(1) is true, and that wheneverπ(π) is true for some π, π π + 1 must also be true. In other words, we show that the conditional π π β π(π + 1) is true for all π.
This creates a logical βchain reactionβ β the truth of π(1) implies the truth of π(2), which implies the truth of π(3), which implies the truth of π 4 , and so on.
On the next slide, we will restate this so-called principle of induction, prove it, and introduce some terminology.
The Principle of InductionTheorem (Principle of Induction): Suppose π(π) is a statement for every natural number π. Further suppose that
1. The first statement, π(1) is true.2. For every π, the conditional π π β π(π + 1) is true.
Then π(π) is true for all π.
Proof: Suppose π(π) is not true for all π. Then there must be at least one π for which it is false. Let π0 be the smallest of all these π. Since we know that π(1) is true, π0 β₯2. Since π0 is the smallest π for which π(π) is false, and π0 β 1 is a natural number, π(π0 β 1) is true. By assumption, the conditional π π0 β 1 β π(π0) is true as well, hence π(π0) is true. That is a contradiction because π(π0) is false.
This proves that π(π) is true for all π.
In an inductive proof, verifying condition 1 is called the base case or basis step. Verifying condition 2 is called the induction step or inductive step. We verify condition 2 by assuming that π π is true (called the inductive hypothesis) for some arbitrary π, and showing that then π π + 1 is true as well.
The Renaming Ritual
Many textbooks, including Rosen, highlight the fact that the inductive hypothesis is made about some arbitrary π value by giving this π a new name such as π. In the inductive hypothesis, rather than assuming π π βfor some arbitrary πβ, they assume π π βfor some arbitrary π = π" and then demonstrate that π π + 1 must be true as well.
This renaming ritual is not necessary, and has the disadvantage that it may cause serious confusion when the statement about π already involves some variable named π. The slide titled βCommon Mistakes (4)β illustrates this.
Example IaLet us prove by induction that for all natural numbers π,
π=1
π
π =π(π + 1)
2
(This equation is the statement π π . )
1. Base case: we verify π 1 . Οπ=11 π =
1β 2
2is true.
2. Inductive step: let us assume π π for some arbitrary π, i.e.
π=1
π
π =π π + 1
2.
We now add the next term of the summation, which in this case happens to be π + 1, to both sides. On the left side, that βupgradesβ the sum to the next higher upper limit. On the right, we hope that the right side of π(π + 1) emerges. (Note that generally, when you write an inductive proof, the statement π·(π + π) will NOT be produced by adding π + π to both sides of π·(π). That algebraic manipulation is SPECIFIC to this example.)
π=1
π+1
π =π(π + 1)
2+ π + 1 =
π
2+ 1 π + 1 =
(π + 1)(π + 2)
2
This statement,
π=1
π+1
π =(π + 1)(π + 2)
2,
is the statement π π + 1 . We just completed the inductive step: we showed that if π π is true, then π π + 1 is true as well. That completes the proof by induction.
π(π + 1) is not generally obtained by adding π + 1 to both sides of π(π).
It bears repeating: the algebraic manipulation we used to obtain π(π + 1) from π(π) in the first example is specific to that example and only works because the quantity under examination grows by π + 1 as we step from the case π to the case π + 1.
Each inductive proof requires its own, individual algebraic steps to conclude that π(π + 1) must be true if π(π) is true.
The next page demonstrates a slightly different algebraic perspective for proving summation formulas. Instead of adding to the summation formula we are assuming is already true, we can split the next summation into what we already know about, plus the next term.
Example Ia, done a little differently Let us prove by induction that for all natural numbers π,
π=1
π
π =π(π + 1)
2
1. Base case: we verify the summation for n=1: 1 =1β2
2. This is true.
2. Inductive step: let us assume that we already know Οπ=1π π =
π π+1
2for some arbitrary π. Then
we split the next sum into the sum we already know, plus the next term:
π=1
π+1
π =
π=1
π
π + π + 1
We now use the inductive hypothesis on the sum we already know:
π=1
π+1
π =π(π + 1)
2+ π + 1 =
π
2+ 1 π + 1 =
(π + 1)(π + 2)
2.
This statement
π=1
π+1
π =(π + 1)(π + 2)
2,
is the case n+1 of our summation formula. That completes the proof by induction.
Common Mistakes (1)Students often confuse the statement π·(π) with the algebraic expression about which a statement is being made. They abuse the notation π·(π).
In the previous example, we proved the statement
π π :=
π=1
π
π =π(π + 1)
2
for all π. Notice the parentheses - π π is not the sigma sum, and itβs not the expression π(π+1)
2. It is the
statement that these two quantities are equal.
Therefore, you must not write equations like π π =π(π+1)
2in your inductive proofs, if P(n) has been
agreed upon to be the statement you are proving. If you must have a notation for an expression that is frequently referred to, you are free to define your own notation for it, for example,
ππ =π(π + 1)
2but you must not use π π . You must also not write equations like this:
π π =
π=1
π
π =π(π + 1)
2
Without the critical parentheses, π π is no longer the statement that the middle and right quantity are equal. It is the common value of that quantity. So on top of using wrong notation, you are assuming the conclusion of the proof with this.
Itβs best to avoid the abstraction of βP(n)β in actual inductive proofs!!
To describe the method of induction in theory, it was useful to have a notation β P(n) β to refer to the sequence of statements we are proving. In practice though, using this notation is often worse than useless. Itβs just as easy and better writing style to refer verbally to the statement we are proving, and you avoid any risk of making the mistake described on the previous page, of abusing βP(n)β to refer to both a quantity involved in the statement we are proving and the statement itself.
Observe how the following example proofs all do without the notation βP(n)β and are better for it.
Example IbLet us prove by induction that for all natural numbers π,
π=1
π
π2 =π(π + 1)(2π + 1)
6.
1. Base case: we verify the summation formula for n=1: 1 =1β2β3
6. This is true.
2. Inductive step: let us assume that we have already proved the summation for some arbitrary π, i.e.
π=1
π
π2 =π(π + 1)(2π + 1)
6.
We now add the next term of the summation, which in this case is π + 1 2, to both sides. On the left side, that βupgradesβ the sum to the next higher upper limit. On the right, we expect that the expression
(π+1)(π+2)(2π+3)
6will
emerge. (Observe that the next term of the summation is NOT π + π π‘ππ«π. )
π=1
π+1
π2 =π(π + 1)(2π + 1)
6+ π + 1 2 =
π 2π + 1
6+ π + 1 (π + 1) =
π 2π + 1 + 6 π + 1
6(π + 1)
=2π2 + 7π + 6
6π + 1 =
(π + 2)(2π + 3)
6π + 1 =
(π + 1)(π + 2)(2(π + 1) + 1)
6
This statement,
π=1
π+1
π2 =(π + 1)(π + 2)(2(π + 1) + 1)
6
,
is the summation formula for the case n+1. We just completed the inductive step: we showed that if the formula holds for one n, then it also holds for the next. That completes the proof by induction.
Observe that prudent, early factoring, rather than brute-force distributing, is the algebraic key to an efficient, readable solution here.
Common Mistakes (2)A second, somewhat common mistake is to confuse stating π π + 1 with proving it. Merely stating it does not prove it.
The following variation of the inductive step of Example 1 illustrates that mistake:
βLet us assume that π π is true for some π, i.e.
π=1
π
π =π(π + 1)
2
Then π π + 1 is the statement
π=1
π+1
π =(π + 1)(π + 2)
2
This completes the proof by induction. β
The psychology of this mistake seems to be that merely writing π π + 1 required some algebra work. You had to replace each n by n+1 and simplify, so it may feel like βyou did somethingβ. You did, you just didnβt prove π π + 1 .
Again β the notation P(n) is best avoided completely when you write inductive proofs. It is used here only because it is convenient to describe the mistake in question.
Common Mistakes (3)When π π is a summation formula, people may confuse the π with the running variable in the summation.
We earlier proved that
π=1
π
π =π(π + 1)
2
is true for all n. Notice that both sides are functions of n alone: n is the ONLY free variable in this formula. You may think that you are seeing another free variable, k, on the left, but this hallucination is easily dispelled by remembering the meaning of the sigma notation: it means simply 1 + 2 + 3 +β―+ π. Our formula is
1 + 2 + 3 +β―+ π =π π + 1
2.
There is no βπβ there. If you replace π by π + 1, you simply break the sigma sum and turn it into something unrelated.
People who are unclear about the distinction between π and π may think that they need to increment π when they form Οπ=1
π+1 π , either instead of incrementing π, or in addition to it:
π=1
π
(π + 1 ) or
π=1
π+1
(π + 1) .
Neither is equal to Οπ=1π+1 π.
From a programming perspective, you can think of the k inside the sigma notation as a local variable inside a function that is invisible from the outside.
function sum(int n)
int sum = 0;
for (int k=1;k<=n;k++)
sum +=k;
Common Mistakes (4)If you practice the βrenaming ritualβ of assuming in the inductive step that π π is true βfor some π = π β rather than just βfor some πβ, and if your π π involves a sigma sum that uses π as a running variable, youβre in danger of writing wholly nonsensical expressions like this:
π=1
π
π
If you donβt see whatβs wrong with that, consider the following equivalent code:
function sum(int k)int sum = 0;
for (int k=1;k<=k;k++)
sum +=k;
The best way to avoid this is to not rename the induction variable πand to assume that π π is true for some π in the inductive step.
Common Mistakes (5)
Whatβs wrong with the following inductive step?
βInductive Step. Let us assume π(π) is true for all n..β
In the inductive step, we prove that the conditional π π β π(π + 1) is true for all π, i.e. we prove
βπ π π β π π + 1 .
We do this by assuming that π(π) is true for some (arbitrary) π, which sets up universal generalization. Then we show that π π + 1 must be true as well. By universal generalization, since our π was arbitrary, that shows
βπ π π β π π + 1 .
In the proof fragment above, it is assumed that π(π) is true for all π. That is assuming the conclusion. If we already know that π(π) is true for all π, we donβt need a proof.
Common Mistakes (6)
Some students who practice the renaming ritual confuse the act of giving the specific π for which we assume π(π) is already proved in the inductive step a new name with the act of assuming that π(π) has already been proved.
Instead of writing βInductive Step. Let us assume π(π) for some π =πβ,
they write βInductive Step. Assume π = πβ.
This is meaningless since in that βassumptionβ, π and π are just two undefined variables. Even if π was defined, giving π a new name is not the same as assuming that π(π) is true.
The general theme of this misunderstanding is to not distinguish properly between numbers and the statements they refer to. Donβt write βwe proved n=1β. Write βwe proved the case n=1β. Donβt write βn=1 is trueβ when you mean βthe first statement is trueβ.
Example IIaTheorem: For any positive integer π, 4π β 1 is divisible by 3.
Proof by induction: Base Case: for π = 1, 4π β 1 = 4 β 1 = 3, which is divisible by 3.
Inductive Step: suppose we have already proved that 4π β 1 is divisible by 3 for some positive integer π. Our goal is to show that 4π+1 β 1 = 4 β 4π β 1 must then be divisible by 3 as well. There are different ways to proceed from here.
Variation 1: We can split four times 4π into three times that quantity, plus one times that quantity:
4 β 4π β 1 = 3 β 4π + (4π β 1).
According to the inductive hypothesis, 4π β 1 is a multiple of 3. Since 3 β 4π is a multiple of 3 as well, so is the sum of the two terms, which is 4π+1 β 1.
Variation 2: According to the inductive hypothesis, 4π β 1 is a multiple of 3, i.e. 4π β 1 =3π for some integer π, or 4π = 3π + 1. Substituting that into 4π+1 β 1, we get
4π+1 β 1 = 4 β 3π + 1 β 1 = 12π + 3 = 3(4π + 1).
Thus 4π+1 β 1 is a multiple of 3.
Example IIbTheorem: 21 divides 4π+1 + 52πβ1 for all natural numbers π.Proof by induction: we first verify the statement for π = 1: 21 divides 42 + 51 = 21.
Now suppose that 21 divides 4π+1 + 52πβ1 for some natural number π. This means 4π+1 + 52πβ1 = 21π for some integer π. Let us consider
4π+2 + 52(π+1)β1 = 4π+2 + 52π+1
By using laws of exponentiation, we can rewrite this as 4 β 4π+1 + 25 β 52πβ1.
It would be nice here if we could just factor out a common factor in order to substitute the inductive hypothesis, but that is not possible. So we split the second term to make at least a partial factorization possible:
4 β 4π+1 + 25 β 52πβ1 = 4 β 4π+1 + 4 β 52πβ1 + 21 β 52πβ1
Therefore,
4π+2 + 52(π+1)β1 = 4 4π+1 + 52πβ1 + 21 β 52πβ1= 21 4π + 52πβ1 .
We have demonstrated that 4π+2 + 52(π+1)β1 is divisible by 21. This completes the proof.
Example IIITheorem: if π is a finite set with π = π, then π« π = 2π, for π = 0,1,2,3, β¦
Proof by induction:1. Base Case (n=0): if π is a set with π = 0, then π« π = 1. This statement is true, because π = 0 implies π = β , hence π« π = β , so π« π = 1.
2. Inductive Step: suppose we already know for some arbitrary π that if a set has π elements, its power set must have 2π elements. We must then verify that if π is a set with π = π + 1, then π« π = 2π+1. To verify this conditional, we will assume its premise and show that the conclusion must be true.
Let π be a set with π = π + 1. Let us single out an arbitrary element of π and call it π₯. Define π = π β {π₯}. Then π = π. Now let us count how many subsets π has. The number of subsets of π equals π΄ + π΅, where
π΄ = the number of subsets of π that donβt contain π₯,π΅ = the number of subsets that do contain π₯.
A subset of π that doesnβt contain π₯ is a subset of π , hence π΄ = π« π = 2π by the inductive hypothesis.
A subset of π that contains π₯ is the union of a subset of π with {π₯}, hence π΅ =π« π = 2π. Therefore, π« π = π΄ + π΅ = 2π + 2π = 2π+1. That completes our
proof.
Example IVTheorem: π2 < 2π for all integers π β₯ 5.
Proof by induction: we first verify the statement for π = 5: 52 < 25 because 25 < 32. This was the base case.
Inductive step: Now suppose that π2 < 2π for some natural number π β₯ 5. We will show (π + 1)2< 2π+1.
Since(π + 1)2= π2 + 2π + 1, we only need to show that 2π + 1 < 2π for π β₯ 5.Then, π2 + 2π + 1 < 2π + 2π = 2π+1 by the inductive hypothesis and thus (π + 1)2< 2π+1.
It remains to show that 2π + 1 < 2π for all π β₯ 5. We do this once again by induction.
Base case: 2π + 1 < 2π is true for π = 5 because 11 < 32.
Inductive step: suppose 2π + 1 < 2π for some π β₯ 5. Our goal is to show 2(π + 1) +1 < 2π+1. Let us add two to the inequality of the inductive hypothesis: 2π + 3 <2π + 2. Since 2 < 2π for π β₯ 2, and certainly for π β₯ 5, 2π + 3 < 2π + 2 < 2π +2π = 2π+1. Therefore, 2(π + 1) + 1 < 2π+1.
An Example of an Incorrect Proof by Induction
False Theorem: for all nonnegative integers π, π = 2π.This statement is of course absurd. π = 2π is true for only one number: π = 0. Now let us consider the following (incorrect) proof by induction. Define π π for all nonnegative integers to be the statement π = 2π.
Base case: for π = 0, the statement π = 2π is correct.
Inductive Step: suppose π = 2π for some arbitrary nonnegative integer π has already been shown. Multiply both sides of this equation by
π+1
πto get π + 1 = 2 π + 1 . That completes the proof.
It is clear that this proof is incorrect, because any alleged proof of a false statement must be incorrect. Let us try to pinpoint the mistake.
The base case was correctly verified. 0 = 2 β 0 is a true statement.
The argument of the inductive step is correct for all π except the first. When π = 0, then the argument asks us to multiply by the undefined quantity
1
0.
Indeed, any argument that claims to prove π π β π(π + 1) for all π must at least be invalid for π = 0, because π 0 is a true statement and π(1) is a false statement, so the conditional π 0 β π 1 is false. (The following conditionals, π 1 β π 2 , etc, are actually true, and our proof above showed that. )