Higher-dimensional Golden Section, Generalized Fibonacci Sequence, and Multiple Online Leasing

Abstract

We discover a connection between the classical golden section and the online leasing problem: when the number of skis increases from one to two, the optimal competitive algorithm is to rent two pairs until the golden section point$(\sqrt{5}-1)s/2$ within [0,s], then buy one pair, and finally buy the second at time $s$, achieving the optimal competitive ratio of $(5+\sqrt{5})/4$. Building on this, we explore higher-dimensional golden sections and multiple online leasing. For the golden section, we define geometric-type and harmonic-type higher-dimensional golden sections, $τ^{(m)}$ and $T^{(m)}$, based on two equivalent propositions dividing the unit segment into $m$ segments. We establish equations for the geometric-type ratio $λ$ and harmonic-type basis $ω$, and investigate their algebraic characteristics. Extending the $k$-generalized Fibonacci sequence to a bi-infinite $m$-generalized Fibonacci sequence, we establish their relationship. We introduce novel concepts including generating sequences and the Fibonacci matrix, investigating its submatrices. Results such as a determinant formula involving the Fibonacci matrix demonstrate its effectiveness in studying generalized Fibonacci sequences. For online leasing, we consider multiple online leasing of $m$ pairs of skis. We design a control and balancing strategy and prove it is optimal via competitive analysis. We show the optimal strategy is precisely the $m$-dimensional harmonic-type golden section method. Since explicit solutions may not exist for dimensions higher than four, we propose a continuous relaxation problem. We explore using its relaxed solution to approximate the optimal solution or the largest real root. We propose the mantissa weighting method to solve the equation, achieving high accuracy through error analysis.