Motivation
The continued fraction of a square root often begins to repeat. For example, the repeated digit two for the square root of two is exact arithmetic, not a numerical pattern inferred from a long decimal expansion. We want a proof that explains both why repetition occurs and why a repeating continued fraction must represent a quadratic irrational.
There are two different mechanisms. A repeated block is a fractional linear transformation with integer coefficients, and its fixed point satisfies a quadratic equation. In the reverse direction, a quadratic irrational has integral surd coordinates. Those coordinates are not initially bounded. We first prove that its conjugate complete quotients enter a certain interval; only then do the coordinates lie in a finite set. Keeping this order avoids the unsupported assertion that an infinite continued fraction must eventually repeat because it is generated by a recurrence.
Use the convergent and complete-quotient formulas. All numbers here are real. The continued fractions are simple: the first digit is an integer and every later digit is a positive integer. Exact floor decisions, rather than floating-point rounding, are understood throughout.
Definitions
Definition 11.2.1
(Complete quotients and reduced quadratic irrationals)
For an irrational , set , , and . Then for . The continued fraction is eventually periodic if its digits repeat from some index onward. It is purely periodic if they repeat from index zero.
A real quadratic irrational is an irrational root of an integer polynomial of degree two. Its other root, or conjugate, is denoted . A quadratic irrational is reduced when and . Conjugation fixes rational numbers and sends the positive square root in its quadratic field to the negative one.
Theorems
Theorem 11.2.2
(Eventual periodicity characterizes real quadratic irrationals)
An irrational real number has an eventually periodic simple continued fraction if and only if it is quadratic. Repetition can be detected exactly by a repeated complete quotient, and for a quadratic irrational it can be detected by a repeated integral surd state.
Theorem 11.2.3
(Reduction gives a finite reversible state system)
Every quadratic irrational has reduced complete quotients from some index onward. A quadratic irrational has a purely periodic simple continued fraction if and only if it is reduced. For a representation
where is a positive nonsquare integer, every reduced state satisfies and . The complete-quotient map permutes the finite set of reduced integral states with this fixed .
Proof sketch or proof idea
A periodic block gives a quadratic fixed point
Proof
(Fixed point of the block matrix)
The operation has matrix and determinant . A period beginning after the initial digit uses only positive digits. Let its product be . Its lower-left entry is positive: this is true for a block of length one, and multiplication by another positive-digit matrix preserves positivity of the required entries. A complete quotient at the beginning of the period therefore satisfies
Equality of the repeating tails follows from uniqueness of an infinite simple continued fraction. The original is a fractional linear expression in with integer coefficients and determinant , obtained by attaching the finite prefix. Such a transformation and its inverse preserve rationality. Thus is irrational, its displayed polynomial has degree two, and is quadratic in the same field. If repetition starts at zero, shift to a later complete quotient first; this avoids making any assumption about the sign of the initial digit.
Put every quadratic irrational into integral surd coordinates
Proof
(The integral surd recurrence)
Suppose is an integer equation with . Its discriminant is positive and nonsquare; otherwise its roots would not be real irrational. The quadratic formula writes as , using either or according to the root. In both cases divides . is allowed to be negative initially.
Rationalizing the next complete quotient gives the exact recurrence
The divisibility hypothesis makes an integer: modulo . The same displayed equation gives the new divisibility condition. Its numerator is nonzero because is nonsquare. Thus the recurrence continues in integral coordinates, but this observation alone does not make the state set finite.
The conjugate must eventually become negative
Proof
(Late conjugates lie in )
Let ; conjugation gives . No is rational. Once at an index , the next lies strictly between and zero because . This interval is then preserved.
Suppose instead that remains positive for all . Positivity of the next term forces , and the following positivity forces . Hence has the same simple continued-fraction digit as at every index. The two numbers and would have identical infinite digits and must be equal, contradicting quadratic conjugacy.
For completeness, the uniqueness used here does not rely on a decimal expansion. A finite positive-digit prefix maps tails into an interval whose endpoints are the prefix convergent and the value with tail one. The determinant identity shows that its length is at most , where is the last convergent denominator. These denominators grow without bound, since their recurrence adds a positive multiple of the preceding denominator. Two numbers sharing every prefix therefore have distance zero. The argument applies equally to a conjugate proposed to have all those positive tails.
Proof X-ray
Where finiteness enters the periodicity argument
Integral coordinates give a discrete set, but not yet a bounded set. Negativity of the conjugate makes the state reduced. The inequalities for a reduced state then bound both integer coordinates. A repeated state finally gives identical future states because the next digit is the exact floor of its represented number. Each step supplies a condition absent from the preceding one; discreteness alone cannot establish repetition.
Reduced coordinates are bounded, hence eventually repeat
Proof
(Bounded reduced coordinates)
For a reduced state and , the difference is positive, so . The inequalities and give and . Also . Thus only finitely many integer pairs are possible, even before imposing divisibility or the remaining inequality .
All sufficiently late states are reduced by the preceding conjugate argument. Two of them therefore have the same pair. The recurrence, including its floor decision, is deterministic, so their subsequent states and digits repeat forever. This proves quadratic implies eventual periodicity, without any estimate of how soon reduction begins.
Why reduced starting points have no nonperiodic prefix
Proof
(A permutation of reduced states)
For fixed consider all integral states satisfying the stated divisibility condition and the two reduced inequalities. We have shown that this finite set maps into itself. It also has a unique reduced predecessor.
Indeed, if is reduced with conjugate , a preceding complete quotient must be , with conjugate . Requiring determines
The endpoints are not integers because is irrational. This also gives and . The predecessor belongs to the same integral-state set: put and . Then is a positive integer, direct rationalization gives , and divides . Its reduced inequalities were built into the choice of a. Uniqueness follows from the same conjugate interval. Thus the map is a permutation of a finite set, so every reduced starting state is periodic from its first step.
Conversely, a purely periodic irrational has positive initial digit: otherwise its repeated initial digit would violate positivity at a later index. It is therefore greater than one. Its complete quotient and conjugate sequences are periodic, and the latter eventually lie in . Periodicity brings that interval condition back to the initial conjugate. This proves the pure-periodicity criterion.
An exact certificate, and what the finite bound does not promise
For a reduced state is positive, so testing whether exceeds an integer is the same as comparing the positive square root of with . If that integer is negative, the comparison is immediate. If it is nonnegative, compare with its square. Equality cannot occur because is nonsquare. This gives exact floor decisions using integer arithmetic; a sufficiently accurate decimal approximation is not part of the certificate.
A claimed period can therefore be checked by listing its integer pairs, verifying the recurrence and each floor inequality, and exhibiting the return to its first pair. For fixed the representation of a state is unique: equality of two expressions forces equality of the rational and irrational coefficients, hence equality of both pairs. If every intermediate pair is distinct, the displayed state cycle has minimal length. A shorter digit period would give the same complete quotient by infinite-expansion uniqueness, so would force an earlier state return as well.
Put . The bounded rectangle has at most integer pairs. This bounds the length of every reduced cycle, although many pairs fail divisibility or the reduced inequalities. It does not bound how many steps an arbitrary initial quotient needs before reduction. It also does not give a polynomial-time algorithm in the binary length of : a bound proportional to can be exponential in that input length. The proof establishes exact termination and verifiable periodicity, while keeping those claims separate from sharper computational estimates.
Worked examples
Worked example 11.2.4
(The square root of two has a prefix, but one plus it is reduced)
Start with and . The next state is , representing . Its digit is two, and the recurrence returns the same state. Thus . Its conjugate is below , so the initial square root is not reduced and its fraction is not purely periodic. By contrast has conjugate in , and equals . The distinction concerns the starting complete quotient, not whether a period exists somewhere.
Worked example 11.2.5
(An exact five-step period for the square root of thirteen)
Begin with . The successive integral states and their digits are
For example, the first update uses ; the last uses and . Each floor follows from , with the indicated integer numerators and denominators. All five states before the return are distinct. Hence , and this complete-quotient cycle has length five. Merely observing five repeated decimal blocks would not supply these exact state equalities.
Worked example 11.2.6
(Recovering an irrational from its periodic block)
For , the block matrix is . Its fixed-point equation is , so the positive root is . Its value lies between one and two, and its conjugate lies between and zero. The reduced criterion confirms the chosen root has a pure period; the negative root cannot be a positive complete quotient. The next quotient is , whose digit is two, and the following quotient is again. A quadratic equation alone must still be paired with the interval information selecting the intended root.
Common mistakes
Counterexample mode
A deterministic recurrence need not be periodic
The recurrence on integers is deterministic and never repeats. Integral surd recurrences also range over an infinite set until the reduced inequalities are proved. The finite-state step needs both integrality and boundedness. A rounded numerical state is different again: an apparent numerical return does not prove equality of two quadratic irrationals.
Common mistake
(A quadratic irrational need not be purely periodic)
Quadratic irrationals need not start with a pure period, as shows: its first digit does not repeat, because its conjugate lies below , so is not reduced. Only a later complete quotient, here , begins the period.
Common mistake
(Rational inputs are not periodic examples)
Rational numbers terminate under this algorithm; their alternative finite last-digit representation does not make them irrational periodic examples.
Common mistake
(Conjugation does not preserve inequalities)
The proof uses conjugates to bound states, not the assertion that conjugation preserves an inequality. Conjugation can change signs and order: , while its conjugate is negative. Each conjugate recurrence and interval was therefore checked separately.
Quick check
Quick check
(For , must every integer pair with and occur in the square-root cycle?)
Compare the number of pairs in the rectangle with the length of the cycle computed above.
Solution · Answer
No. Divisibility, reduced inequalities and membership in that particular orbit impose further restrictions. The rectangle contains the pairs with and , but the cycle of uses only five of them. For instance lies in the rectangle, yet . The rectangle is a finite upper bound for candidate states, not a classification of a single period.
Exercises
Quick check
(Starting from , compute the complete-quotient states for until the first return. Identify which starting complete quotient is reduced.)
Use and with .
Solution · Answer
Since , the next state is , digit one. The next is , digit two, which returns to . Thus . Its first complete quotient has conjugate in , so is reduced. The initial conjugate is not in that interval.
Quick check
(Find the positive irrational represented by using its matrix equation. Verify both reduced inequalities and its two exact digits.)
Multiply and solve the fixed-point equation.
Solution · Answer
The matrix product is , giving . The positive root is , with conjugate in . Its floor is two. The next quotient has floor one, and its next quotient returns to . This checks the exact period and root choice.
Quick check
(Let . Find the quadratic equation for . Explain why is eventually periodic but not purely periodic.)
Substitute into .
Solution · Answer
The equation becomes after reciprocal substitution. The positive root is , between zero and one. Its continued fraction begins with zero and then has every digit one. It cannot be purely periodic, since repeating its initial zero would put a zero among later digits. It is not reduced because it is less than one.
Quick check
(Explain why attaching any finite admissible simple-continued-fraction prefix to a positive quadratic tail gives an irrational in the same quadratic field. State why the fractional linear denominator cannot vanish for such a tail.)
Write the prefix as a fractional linear map whose matrix has determinant .
Solution · Answer
A prefix gives with integer coefficients and determinant , so its value lies in the tail's quadratic field. If that value were rational , the equation would either make rational or force both coefficients to vanish; the latter contradicts the nonzero determinant. The denominator cannot vanish at an irrational , since that would be a nontrivial rational linear equation; is also excluded by the determinant. Thus the attached value is an irrational element of a degree-two field and is quadratic.
Related notes
The Pell-equation unit uses periodicity-related recurrences in a Diophantine setting. The present theorem establishes exact eventual and pure periodicity; it does not bound minimal period lengths sharply, prove the full Hurwitz approximation theorem, or classify all quadratic norms.