Non-additive two-option ski rental

Amir Levi, Boaz Patt-Shamir

Research output: Contribution to journalArticlepeer-review


We consider the following generalization of the classical problem of ski rental. There is a game that ends at an unknown time, and the algorithm needs to decide how to pay for the time until the game ends. In our generalization, there are two "payment plans" called "options," such that each option i (for i=1, 2) consists of two kinds of costs: bi is the (one time) cost to start using Option i, and ai is the (ongoing) usage cost per unit of time for Option i. We assume w.l.o.g. that a1>a2 and b1<b2. Additionally, we assume the existence of a transition cost c, which is incurred if we switch from Option 1 to Option 2. (In the classical version, b1=0, a2=0 and c=b2.)We give deterministic and randomized algorithms for this general setting and analyze their competitive ratio. We also prove that the competitive ratios of our algorithms are the best possible by presenting matching lower bounds for both the deterministic and the randomized cases.

Original languageEnglish
Pages (from-to)42-52
Number of pages11
JournalTheoretical Computer Science
StatePublished - 13 Jun 2015


  • Buy or rent
  • Competitive analysis
  • Duration prediction
  • Randomized algorithms
  • Ski rental


Dive into the research topics of 'Non-additive two-option ski rental'. Together they form a unique fingerprint.

Cite this