Conjugate radical powers: even sums, integer parts, and recurrence relations

Original publication: Updated:

Why adding conjugate powers removes the radical

Let a,ba,b be integers with b≥0b\ge0, and let nn be a positive integer. Then

(a+b)n+(a−b)n is an even integer.\boxed{(a+\sqrt b)^n+(a-\sqrt b)^n\text{ is an even integer}.}

For example, with a=2,b=3a=2,b=3, the first three sums are 4,14,524,14,52. This cancellation can be proved in two equally rigorous ways.

Proof 1: integer coefficients by induction

We prove simultaneously that integers pn,qnp_n,q_n exist with

(a+b)n=pn+qnb,(a−b)n=pn−qnb.(a+\sqrt b)^n=p_n+q_n\sqrt b,\qquad(a-\sqrt b)^n=p_n-q_n\sqrt b.

For n=1n=1, choose p1=a,q1=1p_1=a,q_1=1. If the claim holds at n=kn=k, multiplying by the corresponding conjugate base gives

pk+1=apk+bqk,qk+1=pk+aqk.p_{k+1}=ap_k+bq_k,\qquad q_{k+1}=p_k+aq_k.

Both remain integers, and the two expansions retain opposite radical coefficients. Induction proves the claim. Adding the expansions yields 2pn2p_n, an even integer.

Proof 2: the binomial theorem

Expand both powers with the binomial theorem. Terms containing an odd power of b\sqrt b have opposite signs and cancel. Terms with an even exponent 2j2j occur twice, leaving

(a+b)n+(a−b)n=2∑j=0⌊n/2⌋(n2j)an−2jbj.(a+\sqrt b)^n+(a-\sqrt b)^n=2\sum_{j=0}^{\lfloor n/2\rfloor}\binom n{2j}a^{n-2j}b^j.

The sum is an integer, so the result is even. The indexed sum specifies every term explicitly; there is no need to regard this proof as less rigorous than induction.

A small conjugate determines an integer part

Suppose also 0<a−b<10<a-\sqrt b<1. Then

0<(a−b)n<1,0<(a-\sqrt b)^n<1,

so, writing the even sum as 2pn2p_n,

2pn−1<(a+b)n<2pn.2p_n-1<(a+\sqrt b)^n<2p_n.

Consequently

⌊(a+b)n⌋=2pn−1,\boxed{\lfloor(a+\sqrt b)^n\rfloor=2p_n-1,}

which is odd. The strict inequalities are what identify the floor; merely knowing the conjugate sum is integral is not enough.

Problem 1: an odd integer part for every positive exponent

Problem. Prove that the integer part of (7+38)n(7+\sqrt{38})^n is odd for every positive integer nn.

Because 6<38<76<\sqrt{38}<7, we have 0<7−38<10<7-\sqrt{38}<1. The conjugate sum

(7+38)n+(7−38)n=2pn(7+\sqrt{38})^n+(7-\sqrt{38})^n=2p_n

is an even integer by either proof above. Therefore

2pn−1<(7+38)n<2pn,2p_n-1<(7+\sqrt{38})^n<2p_n,

and its floor is the odd integer 2pn−12p_n-1.

Problem 2: a negative conjugate and a last digit

Let α>β\alpha>\beta be the roots of x2−4x−1=0x^2-4x-1=0, and define sn=αn+βns_n=\alpha^n+\beta^n.

  1. Find s1,s2,s3s_1,s_2,s_3, and a recurrence expressing sns_n through the previous two terms.
  2. Find the greatest integer not exceeding β3\beta^3.
  3. Find the units digit of the greatest integer not exceeding α2003\alpha^{2003}.

Part 1: initial values and recurrence

The roots are α=2+5\alpha=2+\sqrt5, β=2−5\beta=2-\sqrt5. Thus

s1=4,s2=18,s3=76.s_1=4,\qquad s_2=18,\qquad s_3=76.

Since α+β=4\alpha+\beta=4 and αβ=−1\alpha\beta=-1,

sn=4sn−1+sn−2(n≥3).\boxed{s_n=4s_{n-1}+s_{n-2}\qquad(n\ge3).}

The recurrence and integer initial values also give an independent proof that every sns_n is an integer.

Part 2: watch the sign of the small conjugate

Here −1<β<0-1<\beta<0, so −1<β3<0-1<\beta^3<0. The required greatest integer is −1\boxed{-1}.

When the conjugate is negative, its power changes sign with parity. The floor argument used for a positive small conjugate must be adjusted accordingly.

Part 3: a recurrence modulo ten

Since 20032003 is odd, −1<β2003<0-1<\beta^{2003}<0. Therefore

s2003<α2003<s2003+1,s_{2003}<\alpha^{2003}<s_{2003}+1,

and ⌊α2003⌋=s2003\lfloor\alpha^{2003}\rfloor=s_{2003}.

Modulo ten, the initial values and recurrence produce

4,8,6,2,4,8,6,2,….4,8,6,2,4,8,6,2,\ldots.

To justify repetition rather than simply guess it, apply the recurrence to a block (4,8,6,2)(4,8,6,2):

4⋅2+6≡4,4⋅4+2≡8,4⋅8+4≡6,4⋅6+8≡2(mod10).4\cdot2+6\equiv4,\quad4\cdot4+2\equiv8,\quad4\cdot8+4\equiv6,\quad4\cdot6+8\equiv2\pmod{10}.

Thus each four-term block generates the next. Since 2003≡3(mod4)2003\equiv3\pmod4, the required units digit is 6\boxed{6}.

Problem 3: a product relation and a greatest common divisor

The original article also presents a problem from the University of Tokyo's 2017 science-track paper. Put p=2+5p=2+\sqrt5 and

an=pn+(−1p)n,n=1,2,….a_n=p^n+\left(-\frac1p\right)^n,\qquad n=1,2,\ldots.
  1. Find a1,a2a_1,a_2.
  2. For n≥2n\ge2, express a1ana_1a_n using an+1,an−1a_{n+1},a_{n-1}.
  3. Prove that every ana_n is a positive integer.
  4. Find gcd⁡(an,an+1)\gcd(a_n,a_{n+1}).

Because (2+5)(2−5)=−1(2+\sqrt5)(2-\sqrt5)=-1, we have −1/p=2−5-1/p=2-\sqrt5. This is the same conjugate pair as in Problem 2, so a1=4a_1=4 and a2=18a_2=18.

Let q=−1/pq=-1/p. Since pq=−1pq=-1,

a1an=(p+q)(pn+qn)=pn+1+qn+1+pq(pn−1+qn−1)=an+1−an−1.\begin{aligned} a_1a_n&=(p+q)(p^n+q^n)\\ &=p^{n+1}+q^{n+1}+pq(p^{n-1}+q^{n-1})\\ &=\boxed{a_{n+1}-a_{n-1}.} \end{aligned}

Hence an+1=4an+an−1a_{n+1}=4a_n+a_{n-1}. The positive integer initial values imply, by induction, that all later terms are positive integers.

Finally, the Euclidean algorithm and the recurrence give

gcd⁡(an+1,an)=gcd⁡(an−1,an).\gcd(a_{n+1},a_n)=\gcd(a_{n-1},a_n).

Repeatedly reducing the index yields

gcd⁡(an,an+1)=gcd⁡(a1,a2)=gcd⁡(4,18)=2.\boxed{\gcd(a_n,a_{n+1})=\gcd(a_1,a_2)=\gcd(4,18)=2.}

This shows how the same conjugate-power structure supports floor arguments, congruences, integer recurrences, and divisibility proofs. For related techniques, see rational and irrational parts and Diophantine equations.

Related articles

About the author

Koki Saito

Physician and graduate of the University of Tokyo Faculty of Medicine. Drawing on his entrance-exam experience, he explains how to choose mathematical methods and construct complete solutions.

Full profile →

Have a question or spotted an error? Contact us.