Direct Proof

Study Sheet

Direct Proof

From hypothesis to conclusion, one justified step at a time

The Shape of a Direct Proof

Concept
Assume, unpack, compute, conclude

To prove "if PP then QQ" directly: assume PP; rewrite PP using definitions; manipulate until QQ appears in the form of its definition; state QQ. Every "if–then" statement is proved this way unless there is a reason to do otherwise.

The hardest part of a direct proof is usually the middle — the algebra or the counting — and the most common failure is at the ends: forgetting to state what is assumed, or stopping one line before the conclusion has actually been reached in the required form. Ending a proof with "=2(a+b+1)= 2(a + b + 1)" is not the same as ending with "which is even".

Example
Divisibility

Prove: if a∣ba \mid b and b∣cb \mid c then a∣ca \mid c. Proof. Since a∣ba \mid b, b=akb = ak for some integer kk; since b∣cb \mid c, c=bmc = bm for some integer mm. Then c=(ak)m=a(km)c = (ak)m = a(km), and kmkm is an integer, so a∣ca \mid c. ■\blacksquare

Example
An inequality

Prove: for real x,yx, y, x2+y2≥2xyx^2 + y^2 \ge 2xy. Proof. (x−y)2≥0(x - y)^2 \ge 0 because squares are non-negative. Expanding, x2−2xy+y2≥0x^2 - 2xy + y^2 \ge 0, so x2+y2≥2xyx^2 + y^2 \ge 2xy. ■\blacksquare (Written forwards. It was found backwards: start from the goal, subtract, recognize a square.)

Example
Building on a proved fact

Prove: for real a,b≥0a, b \ge 0, a+b2≥ab\dfrac{a + b}{2} \ge \sqrt{ab}. Proof. Because a,b≥0a, b \ge 0, the numbers x=ax = \sqrt a and y=by = \sqrt b are real. Apply the previous inequality to them: x2+y2≥2xyx^2 + y^2 \ge 2xy, that is, a+b≥2aba + b \ge 2\sqrt{ab}. Divide by 22. ■\blacksquare

The hypothesis a,b≥0a, b \ge 0 is used exactly once, to make the square roots real; without it the claim is not even meaningful. Notice also what the proof does not do: it does not redo the algebra of the previous example. A proved result is a tool, and the substitution x=ax = \sqrt a, y=by = \sqrt b is how you pick it up. Equality holds exactly when x=yx = y, that is, when a=ba = b.

Tip
Find it backwards, write it forwards

Scratch work may start from the conclusion and work toward something known. The written proof must run the other way — from known to conclusion — or it proves nothing.

Side note
Where it goes wrong

Starting the write-up with the statement to be proved and manipulating it. "x2+y2≥2xyx^2 + y^2 \ge 2xy, so (x−y)2≥0(x - y)^2 \ge 0, which is true" assumes what it wants to prove; the steps happen to be reversible here, but the proof must SAY so, or be written in the right direction. The same pattern "proves" 1=−11 = -1: square both sides to get 1=11 = 1, which is true.

Try it
Try it: direct

Prove that the sum of any three consecutive integers is divisible by 33.

Show the answer

Let the integers be n,n+1,n+2n, n + 1, n + 2. Their sum is 3n+3=3(n+1)3n + 3 = 3(n + 1), and n+1n + 1 is an integer, so the sum is divisible by 33. ■\blacksquare

Cases

Concept
Proof by cases

When the hypothesis splits naturally (even/odd, positive/negative/zero, n mod 3n \bmod 3), prove the claim in each case. The cases must be exhaustive — every possibility covered — and the conclusion must be the SAME in every case.

Cases work because of a simple rule of logic: if "AA or BB" is true, and AA leads to CC, and BB leads to CC, then CC is true. The art is choosing the split. Split on the feature the conclusion depends on — for squares, the remainder mod 33 or mod 44; for absolute values, the sign — and each case becomes a direct proof with one extra fact to use.

Example
n2+nn^2 + n is always even

Case 1: n=2kn = 2k. Then n2+n=4k2+2k=2(2k2+k)n^2 + n = 4k^2 + 2k = 2(2k^2 + k), even. Case 2: n=2k+1n = 2k + 1. Then n2+n=n(n+1)=(2k+1)(2k+2)=2(2k+1)(k+1)n^2 + n = n(n + 1) = (2k + 1)(2k + 2) = 2(2k + 1)(k + 1), even. Both cases give even; every integer is in one of them. ■\blacksquare

Example
Squares modulo 4

Claim: every perfect square leaves remainder 00 or 11 on division by 44. Case n=2kn = 2k: n2=4k2n^2 = 4k^2, remainder 00. Case n=2k+1n = 2k + 1: n2=4(k2+k)+1n^2 = 4(k^2 + k) + 1, remainder 11. ■\blacksquare Consequence: a sum of two squares leaves remainder 00, 11 or 22, never 33. So 40034003, which leaves remainder 33, is not a sum of two squares — proved without trying a single pair.

Example
Squares modulo 3

Claim: if 3∣a2+b23 \mid a^2 + b^2, then 3∣a3 \mid a and 3∣b3 \mid b. First, cases on aa mod 33: if a=3ka = 3k then a2=9k2a^2 = 9k^2 leaves remainder 00; if a=3k±1a = 3k \pm 1 then a2=9k2±6k+1a^2 = 9k^2 \pm 6k + 1 leaves remainder 11. So each of a2,b2a^2, b^2 leaves remainder 00 or 11, and their sum leaves 00, 11 or 22. It leaves 00 only when both leave 00, which happens only when 3∣a3 \mid a and 3∣b3 \mid b. ■\blacksquare

Side note
Where it goes wrong

Cases that miss something ("nn positive or negative" forgets zero), or proving a different conclusion in each case and calling it done. A split into "n=3kn = 3k and n=3k+1n = 3k + 1" is also incomplete: it forgets n=3k+2n = 3k + 2.

Try it
Try it: cases

Prove that for every integer nn, n3−nn^3 - n is divisible by 33.

Show the answer

Write n3−n=(n−1)n(n+1)n^3 - n = (n - 1)n(n + 1). Among three consecutive integers one is a multiple of 33 (cases n≡0,1,2(mod3)n \equiv 0, 1, 2 \pmod 3: the multiple is nn, n−1n - 1, n+1n + 1 respectively). So the product is divisible by 33. ■\blacksquare

Divisibility Arguments

Concept
Linear combinations

If d∣ad \mid a and d∣bd \mid b, then d∣ax+byd \mid ax + by for all integers xx and yy. Proof: a=dka = dk and b=dmb = dm, so ax+by=d(kx+my)ax + by = d(kx + my). ■\blacksquare

This one line is behind most divisibility proofs. To show that two numbers share no common factor bigger than 11, combine them — multiply each by a well-chosen integer and subtract — until what is left is 11. Any common divisor divides the combination, so it divides 11. The only creative step is choosing the multipliers, and they are usually the coefficients of nn swapped.

Example
A combination

Prove: if 7∣a7 \mid a and 7∣b7 \mid b, then 7∣3a−5b7 \mid 3a - 5b. Proof. Write a=7ka = 7k and b=7mb = 7m. Then 3a−5b=21k−35m=7(3k−5m)3a - 5b = 21k - 35m = 7(3k - 5m), a multiple of 77. ■\blacksquare

Example
Consecutive integers share no factor

Prove: gcd⁡(n,n+1)=1\gcd(n, n + 1) = 1 for every positive integer nn. Proof. Let dd be a positive common divisor. By the combination rule dd divides (n+1)−n=1(n + 1) - n = 1, so d=1d = 1. ■\blacksquare

Example
A fraction that never simplifies

Prove: for every positive integer nn, the fraction 5n+33n+2\dfrac{5n + 3}{3n + 2} is in lowest terms. Proof. Let dd divide both 5n+35n + 3 and 3n+23n + 2. Then dd divides 3(5n+3)−5(3n+2)=15n+9−15n−10=−13(5n + 3) - 5(3n + 2) = 15n + 9 - 15n - 10 = -1. So d=1d = 1, and the numerator and denominator have no common factor. ■\blacksquare

The multipliers 33 and 55 are the coefficients of nn, swapped, so that the nn terms cancel. That is the whole trick; what is left is a constant, and if the constant is ±1\pm1 the fraction never simplifies. If it had come out as ±2\pm 2, the conclusion would only be that any common factor divides 22, and you would check parity separately.

Side note
Where it goes wrong

Running the rule backwards. "d∣a+bd \mid a + b, so d∣ad \mid a and d∣bd \mid b" is false: 3∣1+23 \mid 1 + 2, yet 3∤13 \nmid 1. Divisibility of a combination says nothing about the pieces unless you also know it for all but one of them.

Try it
Try it: never simplifies

Prove that 4n+33n+2\dfrac{4n + 3}{3n + 2} is in lowest terms for every positive integer nn.

Show the answer

If dd divides 4n+34n + 3 and 3n+23n + 2, it divides 3(4n+3)−4(3n+2)=12n+9−12n−8=13(4n + 3) - 4(3n + 2) = 12n + 9 - 12n - 8 = 1. So d=1d = 1. ■\blacksquare

Putting It Together

Claim: the sum of two rational numbers is rational.

Broken 1. "Let a=p/qa = p/q and b=p/qb = p/q. Then a+b=2p/qa + b = 2p/q, which is rational." Flaw: the same letters for both numbers, so this proves only that a+aa + a is rational.

Broken 2. "Let a=p/qa = p/q and b=r/sb = r/s. Then a+b=p+rq+sa + b = \dfrac{p + r}{q + s}, which is rational." Flaw: the algebra is wrong — 12+12=1\tfrac12 + \tfrac12 = 1, not 24\tfrac24. A one-line number check exposes it.

Repair. "Let a=p/qa = p/q and b=r/sb = r/s with p,q,r,sp, q, r, s integers and q,s≠0q, s \ne 0. Then a+b=ps+qrqsa + b = \dfrac{ps + qr}{qs}. The numerator and denominator are integers, and qs≠0qs \ne 0 because neither factor is 00. So a+ba + b is rational. ■\blacksquare" The hypothesis q,s≠0q, s \ne 0 is used, visibly, at the one place it is needed.

Example
A model direct proof

Claim: the difference of the squares of two consecutive odd numbers is divisible by 88. Proof. Let the numbers be 2k+12k + 1 and 2k+32k + 3, with kk an integer. Then (2k+3)2−(2k+1)2=(4k2+12k+9)−(4k2+4k+1)=8k+8=8(k+1)(2k + 3)^2 - (2k + 1)^2 = (4k^2 + 12k + 9) - (4k^2 + 4k + 1) = 8k + 8 = 8(k + 1), which is divisible by 88. ■\blacksquare (Check: 72−52=24=8⋅37^2 - 5^2 = 24 = 8\cdot3.)

Try it
Try it: the whole proof

Prove that for any integers aa and bb, if a−ba - b is even, then a2−b2a^2 - b^2 is divisible by 44.

Show the answer

Suppose a−b=2ka - b = 2k with kk an integer. Then a+b=(a−b)+2b=2k+2b=2(k+b)a + b = (a - b) + 2b = 2k + 2b = 2(k + b) is even too. So a2−b2=(a−b)(a+b)=2k⋅2(k+b)=4k(k+b)a^2 - b^2 = (a - b)(a + b) = 2k\cdot2(k + b) = 4k(k + b), divisible by 44. ■\blacksquare