Take a negative rational number -p/q and raise it to successive powers. The integer parts oscillate in sign. The fractional parts — the remainders after subtracting the nearest integer — form a sequence that clusters in specific regions of the interval [0,1].
How small can those clustering intervals be? This paper proves sharp bounds: the fractional parts of (-p/q)^n cannot all lie in an interval shorter than a critical threshold determined by p and q. Below that threshold, some power must escape the interval. The bound is tight — there exist starting values that achieve it.
Note (2026-11-09): I checked the abstract of the Lu and Zheng paper. It proves a lower bound: for any nonzero real xi and coprime integers p>q>=1 with xi irrational or q>1, the sequence xi(-p/q)^n is not contained in any interval of length less than (1+q/p-q^2/p^2)/p. The abstract does not say the bound is attained, which this essay states; the essay's dynamics framing and its gcd explanation are not in the abstract. The essay also contrasts rational with irrational, but in the abstract the irrational case is the starting value xi, for which the same bound holds; the essay's sequence is the case xi = 1.
The proof connects to dynamics on the real line. Powers of a rational number generate an orbit under multiplication, and the fractional parts are this orbit viewed modulo 1. The clustering question becomes: how concentrated can a multiplicative orbit be on the unit interval? The answer depends on the arithmetic properties of the base — specifically, the greatest common divisor structure of p and q.
What makes the negative case different from the positive is the sign alternation. Positive powers of p/q always grow in the same direction. Negative powers alternate, and the fractional parts inherit this oscillation. The oscillation doesn't help with clustering — it constrains it differently. The orbit visits both sides of 1/2 regularly, which limits how tightly it can concentrate in any subinterval.
A question about arithmetic becomes a question about dynamics. The number's rationality constrains its orbit's geometry. Irrationals can cluster arbitrarily tightly; rationals cannot.