Chapter I · Real variables

Theorems on quadratic surds

§14 30 min

Similar and dissimilar surds, why A+B=C+DA + \sqrt B = C + \sqrt D forces A=CA = C and B=DB = D, and why q\sqrt q and −q-\sqrt q are indistinguishable to arithmetic.

Why this matters

When can two expressions with square roots be equal? 8\sqrt8 and 222\sqrt2 are, but can 2\sqrt2 be a rational multiple of 3\sqrt3, or can 1+21 + \sqrt2 equal 3−53 - \sqrt5? §14 answers such questions, and the answer has a remarkable consequence: in any identity built from rationals and q\sqrt q, you may replace q\sqrt q by −q-\sqrt q throughout and the identity stays true. That symmetry is the seed of Galois theory, and in this chapter it explains why the “conjugate” a−ba - \sqrt b is always a root alongside a+ba + \sqrt b.

The idea

Every number made from rationals and one surd q\sqrt q by ++, −-, ×\times can be written as P+QqP + Q\sqrt q with PP, QQ rational: the rule (q)2=q(\sqrt q)^2 = q lets you collapse any power. For example

(1+2)2=3+22,(1+2)3=7+52.(1 + \sqrt2)^2 = 3 + 2\sqrt2, \qquad (1 + \sqrt2)^3 = 7 + 5\sqrt2.

The calculation used only (2)2=2(\sqrt2)^2 = 2, which is equally true of −2-\sqrt2. So the same calculation, run with −2-\sqrt2, gives (1−2)3=7−52(1 - \sqrt2)^3 = 7 - 5\sqrt2. For this to be a theorem rather than a trick, the pair (P,Q)(P, Q) must be determined by the number P+QqP + Q\sqrt q; that is what Hardy proves.

Explore Powers and their conjugates

Try this. With p=1p = 1, q=2q = 2, compare the columns P+QqP + Q\sqrt q and P−QqP - Q\sqrt q with (1+2)m(1 + \sqrt2)^m and (1−2)m(1 - \sqrt2)^m. Look at the last column. Then try p=2p = 2, q=3q = 3, and finally make qq a perfect square such as 44.

(1+2)m=P+Q2(1 + \sqrt{2})^m = P + Q\sqrt{2} and (1−2)m=P−Q2(1 - \sqrt{2})^m = P - Q\sqrt{2}
mPQP+QqP + Q\sqrt qP−QqP - Q\sqrt qP2−qQ2P^2 - qQ^2
010111
1112.414214−0.4142136−1
2325.8284270.17157291
37514.07107−0.07106781−1
4171233.970560.029437251
5412982.01219−0.01219331−1
69970197.99490.0050506341
Each row comes from the one above by (P+Q2)(1+2)=(P+2Q)+(P+Q)2(P + Q\sqrt{2})(1 + \sqrt{2}) = (P + 2Q) + (P + Q)\sqrt{2}, a rule that only uses (2)2=2(\sqrt{2})^2 = 2. That rule is just as true for −2-\sqrt{2}, so the same P and Q give the conjugate powers. Check, for m = 6: (1+2)6≈197.9949(1 + \sqrt{2})^{6} \approx 197.9949 and (1−2)6≈0.005050634(1 - \sqrt{2})^{6} \approx 0.005050634, matching the last row. The last column is (12−2)m=(−1)m(1^2 - 2)^m = (-1)^m, the product of the two conjugates. Since ∣1−2∣<∣1+2∣|1 - \sqrt{2}| < |1 + \sqrt{2}|, the ratio P/QP/Q approaches 2\sqrt{2}: here P/Q=1.41428571P/Q = 1.41428571 against 2=1.41421356\sqrt{2} = 1.41421356.
What to notice

The same integers PP, QQ serve both (p+q)m(p + \sqrt q)^m and (p−q)m(p - \sqrt q)^m. The last column is (p2−q)m(p^2 - q)^m, the product of the two: for p=1p = 1, q=2q = 2 it alternates ±1\pm 1, so P/QP/Q are the approximations 11,32,75,1712,…\tfrac11, \tfrac32, \tfrac75, \tfrac{17}{12}, \dots to 2\sqrt2 of Examples III. When qq is a square, q\sqrt q is rational and the whole argument collapses.

Precise statements

Definition Similar and dissimilar surds Hardy: §14

Two pure quadratic surds are similar if they are rational multiples of the same surd, and dissimilar otherwise. For example 8=22\sqrt8 = 2\sqrt2 and 25/2=522\sqrt{25/2} = \tfrac52\sqrt2 are similar.

Theorem Coprime non-squares give dissimilar surds Hardy: §14

If MM and NN are integers with no common factor, neither of which is a perfect square, then M\sqrt M and N\sqrt N are dissimilar.

Theorem Comparing A+BA + \sqrt B and C+DC + \sqrt D Hardy: Theorem, §14

If AA, BB, CC, DD are rational and A+B=C+DA + \sqrt B = C + \sqrt D, then either A=CA = C and B=DB = D, or BB and DD are both squares of rational numbers.

Corollary Conjugates Hardy: Corollary, §14

If A+B=C+DA + \sqrt B = C + \sqrt D, then A−B=C−DA - \sqrt B = C - \sqrt D, unless B\sqrt B and D\sqrt D are both rational.

Proofs

Proof Coprime non-squares give dissimilar surds
Strategy

If M\sqrt M were a rational multiple of N\sqrt N, then MN\sqrt{MN} would be rational, hence an integer, so MNMN would be a perfect square. With no common factor, the prime factors of that square must split cleanly between MM and NN, making each a square.

MN\sqrt{MN} would be rational

Suppose M=(a/b)N\sqrt M = (a/b)\sqrt N with aa, bb integers. Multiplying by N\sqrt N: MN=(a/b)N\sqrt{MN} = (a/b) N, a rational number.

So MNMN is a perfect square

An integer that is the square of a rational is the square of an integer (§3). So MN=P2MN = P^2 for an integer PP.

Split the prime powers

Write P=aαbβcγ⋯P = a^{\alpha} b^{\beta} c^{\gamma} \cdots with distinct primes a,b,c,…a, b, c, \dots. Then a2αa^{2\alpha} divides MNMN. Since MM and NN share no factor, the prime aa divides only one of them, and that one is divisible by all of a2αa^{2\alpha}. The same holds for each prime of PP.

Both are squares, a contradiction

So MM is the product of some of the factors a2α,b2β,…a^{2\alpha}, b^{2\beta}, \dots and NN of the rest: both are perfect squares, contrary to hypothesis.

Proof If A+B=C+DA + \sqrt B = C + \sqrt D then A=CA = C, B=DB = D, or both roots are rational
Strategy

Assume B≠DB \neq D. Then B−D\sqrt B - \sqrt D is a non-zero rational, and dividing the rational B−DB - D by it shows that B+D\sqrt B + \sqrt D is rational too. Sum and difference rational means each root is rational.

If B=DB = D, then A=CA = C

Subtract B=D\sqrt B = \sqrt D from both sides. So assume B≠DB \neq D from now on.

B−D\sqrt B - \sqrt D is a non-zero rational

It equals C−AC - A, which is rational. It is not 00, since B=D\sqrt B = \sqrt D would give B=DB = D on squaring.

B+D\sqrt B + \sqrt D is rational

B−D=(B−D)(B+D)B - D = (\sqrt B - \sqrt D)(\sqrt B + \sqrt D), so B+D=(B−D)/(B−D)\sqrt B + \sqrt D = (B - D)/(\sqrt B - \sqrt D), a quotient of rationals.

Both roots are rational

B=12{(B+D)+(B−D)}\sqrt B = \tfrac12\{(\sqrt B + \sqrt D) + (\sqrt B - \sqrt D)\} and D=12{(B+D)−(B−D)}\sqrt D = \tfrac12\{(\sqrt B + \sqrt D) - (\sqrt B - \sqrt D)\} are rational, so BB and DD are squares of rationals.

The corollary is immediate: unless both roots are rational, A=CA = C and B=DB = D, and then of course A−B=C−DA - \sqrt B = C - \sqrt D. Read differently, the theorem says that PP and QQ in P+QqP + Q\sqrt q are uniquely determined (when q\sqrt q is irrational): if P+Qq=P′+Q′qP + Q\sqrt q = P' + Q'\sqrt q with Q≠Q′Q \neq Q', then q=(P−P′)/(Q′−Q)\sqrt q = (P - P')/(Q' - Q) would be rational. That uniqueness is what lets us “switch the sign of q\sqrt q” in the widget.

Worked examples and exercises

Exercise ●○○ Hardy: Examples VIII, 1

Prove from first principles that 2\sqrt2 and 3\sqrt3 are not similar surds.

Hint
If 3=(a/b)2\sqrt3 = (a/b)\sqrt2 then 3b2=2a23b^2 = 2a^2. Look at the power of 22 (or of 33) on each side.

Solution

Suppose 3=(a/b)2\sqrt3 = (a/b)\sqrt2 with positive integers a,ba, b. Squaring, 3b2=2a23b^2 = 2a^2. In the prime factorisation, the power of 33 dividing 3b23b^2 is odd (one more than twice the power in bb), while the power of 33 dividing 2a22a^2 is even (twice the power in aa). Unique factorisation makes this impossible. (Alternatively: 6=23=(a/b)⋅2\sqrt6 = \sqrt2\sqrt3 = (a/b) \cdot 2 would be rational, but 66 is not a square.)

Exercise ●●○ Hardy: Examples VIII, 3

If aa and bb are rational, show that a+b\sqrt a + \sqrt b cannot be rational unless a\sqrt a and b\sqrt b are both rational; and the same for a−b\sqrt a - \sqrt b, unless a=ba = b.

Hint
Use the identity a−b=(a−b)(a+b)a - b = (\sqrt a - \sqrt b)(\sqrt a + \sqrt b), as in the proof of the theorem.

Solution

Suppose s=a+bs = \sqrt a + \sqrt b is rational. If s=0s = 0 then a=b=0\sqrt a = \sqrt b = 0 (both are ≥0\ge 0), which are rational. Otherwise a−b=(a−b)/s\sqrt a - \sqrt b = (a - b)/s is rational too, and so are a=12(s+(a−b)/s)\sqrt a = \tfrac12(s + (a - b)/s) and b\sqrt b. Now suppose d=a−bd = \sqrt a - \sqrt b is rational and a≠ba \neq b. Then d≠0d \neq 0 and a+b=(a−b)/d\sqrt a + \sqrt b = (a - b)/d is rational, and we conclude as before. (If a=ba = b, the difference is 00 whatever aa is.)

Exercise Conjugate roots ●●○ Hardy: Examples VIII, 6–7

Show that every polynomial in x=p+qx = p + \sqrt q with rational coefficients can be written P+QqP + Q\sqrt q with PP, QQ rational. Deduce that if a+ba + \sqrt b (bb not a square) is a root of a polynomial equation with rational coefficients, so is a−ba - \sqrt b.

Hint
Induction on the power: (P+Qq)(p+q)=(pP+qQ)+(P+pQ)q(P + Q\sqrt q)(p + \sqrt q) = (pP + qQ) + (P + pQ)\sqrt q.
Hint
For the second part, compute f(a+b)=P+Qbf(a + \sqrt b) = P + Q\sqrt b and f(a−b)f(a - \sqrt b) by the same steps.

Solution

By the recurrence in the hint, each power xmx^m is Pm+QmqP_m + Q_m\sqrt q with rational PmP_m, QmQ_m, and a rational linear combination of such expressions has the same form.

Let ff have rational coefficients. The computation of f(a+b)=P+Qbf(a + \sqrt b) = P + Q\sqrt b uses only rational arithmetic and the rule (b)2=b(\sqrt b)^2 = b; running the identical computation with −b-\sqrt b (which also squares to bb) gives f(a−b)=P−Qbf(a - \sqrt b) = P - Q\sqrt b. If f(a+b)=0f(a + \sqrt b) = 0, then P+Qb=0+0bP + Q\sqrt b = 0 + 0\sqrt b, so by the theorem (uniqueness of PP, QQ, as b\sqrt b is irrational) P=Q=0P = Q = 0. Hence f(a−b)=0f(a - \sqrt b) = 0 too.

Exercise Rationalising a denominator ●○○ Hardy: Examples VIII, 8

Write 1/(p+q)1/(p + \sqrt q) in the form P+QqP + Q\sqrt q, and compute 1/(3+2)1/(3 + \sqrt2).

Hint
Multiply top and bottom by the conjugate p−qp - \sqrt q.

Solution
1p+q=p−q(p+q)(p−q)=pp2−q−1p2−qq,\frac{1}{p + \sqrt q} = \frac{p - \sqrt q}{(p + \sqrt q)(p - \sqrt q)} = \frac{p}{p^2 - q} - \frac{1}{p^2 - q}\sqrt q,

valid when p2≠qp^2 \neq q (always the case if q\sqrt q is irrational and pp rational). For example 1/(3+2)=(3−2)/7=37−1721/(3 + \sqrt2) = (3 - \sqrt2)/7 = \tfrac37 - \tfrac17\sqrt2. Combined with the previous exercise, every quotient of polynomials in x=p+qx = p + \sqrt q can be written P+QqP + Q\sqrt q (Examples VIII, 9).

Exercise Denesting a square root ●●○ Hardy: Examples VIII, 10

If pp, qq and p2−qp^2 - q are positive, show that p+q=x+y\sqrt{p + \sqrt q} = \sqrt x + \sqrt y with x=12{p+p2−q}x = \tfrac12\{p + \sqrt{p^2 - q}\} and y=12{p−p2−q}y = \tfrac12\{p - \sqrt{p^2 - q}\}. Use it to simplify 3+8\sqrt{3 + \sqrt8}.

Hint
Square x+y\sqrt x + \sqrt y and compute x+yx + y and xyxy.

Solution

We have x+y=px + y = p and xy=14{p2−(p2−q)}=14qxy = \tfrac14\{p^2 - (p^2 - q)\} = \tfrac14 q, with x>0x > 0 and y≥0y \ge 0 (since p>p2−qp > \sqrt{p^2 - q}). Then

(x+y)2=x+y+2xy=p+2⋅12q=p+q,(\sqrt x + \sqrt y)^2 = x + y + 2\sqrt{xy} = p + 2 \cdot \tfrac12\sqrt q = p + \sqrt q,

and since both x+y\sqrt x + \sqrt y and p+q\sqrt{p + \sqrt q} are positive, they are equal. For p=3p = 3, q=8q = 8: p2−q=1p^2 - q = 1, so x=2x = 2, y=1y = 1, and 3+8=2+1\sqrt{3 + \sqrt8} = \sqrt2 + 1. The formula gives something simpler than the original only when p2−qp^2 - q is a rational square, which is Examples VIII, 11.

Check yourself

Check yourself True or false?

12\sqrt{12} and 27\sqrt{27} are similar surds.

Why

True: 12=23\sqrt{12} = 2\sqrt3 and 27=33\sqrt{27} = 3\sqrt3, both rational multiples of 3\sqrt3. The theorem on dissimilar surds needs MM, NN with no common factor and neither a square; 1212 and 2727 share the factor 33.

Check yourself Choose one

Given (2+3)3=26+153(2 + \sqrt3)^3 = 26 + 15\sqrt3, what is (2−3)3(2 - \sqrt3)^3?

Why

Replace 3\sqrt3 by −3-\sqrt3 throughout: the computation used only (3)2=3(\sqrt3)^2 = 3. Check: 26−153≈26−25.981=0.01926 - 15\sqrt3 \approx 26 - 25.981 = 0.019, and (2−3)3≈0.2683≈0.019(2 - \sqrt3)^3 \approx 0.268^3 \approx 0.019.

Check yourself Enter a number

With (2+3)3=26+153(2 + \sqrt3)^3 = 26 + 15\sqrt3, compute P2−3Q2P^2 - 3Q^2 for P=26P = 26, Q=15Q = 15.

Why

676−3⋅225=676−675=1676 - 3 \cdot 225 = 676 - 675 = 1. In general P2−qQ2=(P+Qq)(P−Qq)=(2+3)3(2−3)3=(4−3)3=1P^2 - qQ^2 = (P + Q\sqrt q)(P - Q\sqrt q) = (2 + \sqrt3)^3 (2 - \sqrt3)^3 = (4 - 3)^3 = 1.

Check yourself Choose one

Why does the theorem allow the exception ”BB and DD are both squares of rationals”?

Why

When the square roots are rational, A+BA + \sqrt B is just a rational number, and a rational number can be split as “rational plus square root of a square” in many ways: 1+4=2+1=31 + \sqrt4 = 2 + \sqrt1 = 3.

Hardy’s language

Summary

Summary
  • Surds are similar when they are rational multiples of one surd; M\sqrt M, N\sqrt N with coprime non-square MM, NN are dissimilar.
  • If A+B=C+DA + \sqrt B = C + \sqrt D with A,B,C,DA, B, C, D rational, then A=CA = C and B=DB = D, unless both square roots are rational.
  • Equivalently, PP and QQ in P+QqP + Q\sqrt q are unique, so q\sqrt q may be replaced by −q-\sqrt q in any rational identity.
  • Hence conjugate surds a±ba \pm \sqrt b are roots of the same rational polynomials, and denominators p+qp + \sqrt q can be rationalised.