Tengyu Ma Xiaoming Sun Huacheng Yu Institute for Institute for Institute for Interdisciplinary Advanced Study, Interdisciplinary Information Sciences Tsinghua University Information Sciences Tsinghua University Tsinghua University 3 cooperative players each is assigned a hat of color red or blue each can only see others’ hat guess own color or pass players win if: at least one correct and no wrong guess goal : to maximize winning probability pass strategy1: only a pre-specified player guesses randomly winning prob. = pass 1 2 strategy2: if other two have same color, guess the opposite, otherwise pass. winning prob. = 3 4 is optimal 3 4 𝒏 cooperative players: ◦ coordinate a strategy initially assigned a blue or red hat ◦ uniformly and independently guess a color or pass winning condition: ◦ at least 𝒌 correct guesses and no wrong guess goal: to maximize winning prob. 𝑘 = 1 case is well studied by [?], [?].. Observation 1: randomized strategy does not help Observation 2: related to the minimum 𝑘-dominating set of 0,1 𝑛 Definition: A 𝑘-dominating set for a graph 𝐺 = 𝑉, 𝐸 is a subset 𝐷 of 𝑉, such that every vertex not in 𝐷 has at least 𝑘 neighbors in 𝐷 pass win! pass pass pass lose pass win! pass pass lose pass 0,1,0,1,1,1 ∈ 0,1 winning point 6 1,1,0,1,1,1 ∈ 0,1 losing point 6 pass win! pass pass lose pass winning point has at least 𝑘 losing points as neighbors all losing points → 𝐷 ⊂ 0,1 𝑛 ◦ 𝐷 is 𝑘-dominating set of 0,1 ◦ winning prob. = 1 − 𝐷 2𝑛 𝑛 reduction can be done vice versa by counting argument: ◦ 𝐷 ≥ 𝑘 2𝑛 𝑛+𝑘 ◦ ⇒ winning prob. ≤ 𝑛 𝑛+𝑘 Theorem: ◦ There exists a 𝑘-dominating set 𝐷 of size 𝑘 2𝑛 , 𝑛+𝑘 𝑘 2𝑛 𝑛+𝑘 as long as is an integer, for large enough 𝑛 (𝑛 ≥ Ω(𝑘2𝑘 ) ). ◦ It follows that there exists a strategy of the hat guessing games with winning 𝑛 prob. 𝑛+𝑘 theorem is not true for small 𝑛 ◦ example: 𝑛, 𝑘 = 5,3 , (11,5) 0,1 𝑛 ∖𝐷 𝐷 each has 𝑘 neighbors in 𝐷 each has 𝑛 neighbors in 0,1 𝑛 ∖ 𝐷 𝑉1 each has 𝑑1 neighbors in 𝑉1 𝑉2 each has 𝑑2 neighbors in 𝑉2 𝑘-DS of 0,1 n → 𝑘, 𝑛 − 𝑘 -RP of 0,1 possible (𝑑1 , 𝑑2 )-RP of 0,1 𝑛 : possible 𝑘-DS corresponds to the case 𝑛 ◦ the parameters are of the following form 𝑑1 = 𝑘, 𝑑2 = 2𝑠 − 𝑘, 𝑛 = 2𝑠 − 𝑟 𝑑1 = 𝑘, 𝑑2 = 2𝑠 − 𝑘, 𝑛 = 2𝑠 − 𝑘 easy case 𝑑1 = 1, 𝑑2 = 2𝑠 − 1, 𝑛 = 2𝑠 − 1 hard case 𝑑1 = 𝑘, 𝑑2 = 2𝑠 − 𝑘, 𝑛 = 2𝑠 − 𝑟, 𝑟 = 2, … , 𝑘 from the cases 𝑑1 , 𝑑2 , 𝑛 = (1,2𝑠 − 1,2𝑠 − 1) to 𝑘, 2𝑠 − 𝑘, 2𝑠 − 1 -- nontrivial, [?] from 𝑘, 2𝑠 − 𝑘, 2𝑠 − 𝑟 to 𝑘, 2𝑠+1 − 1,2𝑠+1 − 𝑟 + 1 solve the case 3,5,6 from 3,1,3 given 3,1 -RP of 0,1 3 , (𝑉1 , 𝑉2 ) : 111 011 𝑉1 101 001 𝑉2 110 010 000 100 now construct (3,5)-partition for 0,1 for each 𝑝 = (𝑝1 , 𝑝2 , 𝑝3 ) ∈ 𝑉2 𝐸𝑝 : sys. of equations over 0,1 6 𝑥1 + 𝑥2 = 𝑝1 𝑥3 + 𝑥4 = 𝑝2 𝑥5 + 𝑥6 = 𝑝3 6 𝑠𝑜𝑙 𝐸𝑝 , the collection of solutions of 𝐸𝑝 ◦ 𝐸𝑝 is an independent set 0,1 6 = 𝑠𝑜𝑙 𝐸000 ∪ 𝑠𝑜𝑙 𝐸001 ∪ … ∪ 𝑠𝑜𝑙(𝐸111 ) 𝑠𝑜𝑙(𝐸011 ) 𝑠𝑜𝑙(𝐸001 ) 𝑠𝑜𝑙(𝐸000 ) 𝑠𝑜𝑙(𝐸101 ) 𝑉1 𝑉2 𝑠𝑜𝑙(𝐸010 ) 𝑠𝑜𝑙(𝐸000 ) 𝑠𝑜𝑙(𝐸110 ) 𝑠𝑜𝑙(𝐸100 ) find a perfect matching in 𝑉2 cut each black set by an additional eqn. for 𝑠𝑜𝑙 𝐸011 and 𝑠𝑜𝑙 𝐸010 use eqn.: 𝑥1 + 𝑥3 + 𝑥5 + 𝑥6 = 0 𝑠𝑜𝑙(𝐸011 ) 𝑠𝑜𝑙(𝐸001 ) 𝑠𝑜𝑙(𝐸000 ) 𝑠𝑜𝑙(𝐸101 ) 6 = 2 * the index of the different bit 𝑉1 𝑉2 𝑠𝑜𝑙(𝐸010 ) 𝑠𝑜𝑙(𝐸000 ) 𝑠𝑜𝑙(𝐸110 ) 𝑠𝑜𝑙(𝐸100 ) find a perfect matching in 𝑉2 cut each black set by an additional eqn. for 𝑠𝑜𝑙 𝐸001 and 𝑠𝑜𝑙 𝐸101 use eqn.: 𝑥1 + 𝑥3 + 𝑥5 + 𝑥2 = 0 𝑠𝑜𝑙(𝐸011 ) 𝑠𝑜𝑙(𝐸001 ) 𝑠𝑜𝑙(𝐸000 ) 𝑠𝑜𝑙(𝐸101 ) 2 = 2 * the index of the different bit 𝑉1 𝑉2 𝑠𝑜𝑙(𝐸010 ) 𝑠𝑜𝑙(𝐸000 ) 𝑠𝑜𝑙(𝐸110 ) 𝑠𝑜𝑙(𝐸100 ) all the grey points → 𝑉2′ ⊂ 0,1 6 , 𝑉1′ = 𝑉2′ 𝐶 . ◦ 𝑉1′ , 𝑉2′ is a 3,5 -RP of 0,1 6 this idea is extendable to general cases 𝑠𝑜𝑙(𝐸011 ) 𝑠𝑜𝑙(𝐸001 ) 𝑠𝑜𝑙(𝐸000 ) 𝑠𝑜𝑙(𝐸101 ) 𝑉1 𝑉2 𝑠𝑜𝑙(𝐸010 ) 𝑠𝑜𝑙(𝐸000 ) 𝑠𝑜𝑙(𝐸110 ) 𝑠𝑜𝑙(𝐸100 ) Main contribution: ◦ foy any odd 𝑘, and 0 < 𝑐 ≤ 𝑘, when 𝑠 ≥ 𝑐 + log 𝑘 − 1, there exists a 𝑘, 2𝑠 − 𝑘 -regular 𝑠 −𝑐 2 partition of 0,1 . ◦ particularly, it follows that for large enough 𝑛, there exists 𝑘-dominating set of size 𝑘 2𝑛 , 𝑛+𝑘 as long as 𝑘 2𝑛 𝑛+𝑘 is integer.