A necromancer and a paladin combat on a grid. The grid is initially empty. For each turn, the necromancer first chooses a block on the grid, and increases one skeleton in each block within the region centered at the chosen block (which include at most nine blocks); then, the paladin chooses four blocks on the grid, and eliminates one skeleton from each of these blocks (the number of skeletons in each block is a non-negative integer.)
For any turn, we called a block doomed if, at the end of that turn, there are or more skeletons in that block. Find the largest positive integer such that the necromancer has a strategy that guarantees or more blocks on the grid to be simultaneously doomed in some finite turns, regardless of how the paladin acts.
, 2023
Solution
The largest ; in general, for a grid, .

First we prove . Color the grid as in the figure above, and note that in each turn the necromancer can only increase skeletons on at most four white blocks, so the paladin can always keep the skeleton count on the white region at , so the number of doomed blocks the necromancer can possibly guarantee to create is at most the number of gray blocks, namely .
Next we prove . In fact, we may consider the following strengthened version of the game: when it is the paladin's turn, he instead chooses five blocks within the region that the necromancer has just increased skeletons in, and decreases the skeleton count by one in every block on the entire grid except these five blocks. Clearly, if in the strengthened version the necromancer can guarantee , then he can also do so in the original version. Note that, for a region, there are a total of ways for the paladin to choose five of its blocks. Also note that, combining both players' moves, this is equivalent to increasing the skeleton count by in the five blocks chosen by the paladin, keeping the skeleton count unchanged in the remaining four blocks of the region, and decreasing the skeleton count by in every other block on the grid.
Consider the necromancer repeatedly targeting the same region for consecutive turns. By the pigeonhole principle, among all ways of choosing five blocks out of the region, there must be one that the paladin selects at least times. From the discussion in the previous paragraph, this means that within this region, there are at least five blocks whose skeleton count is .
Now, partition the grid into disjoint regions, and number them . Then, starting from the region with the largest number, we perform the following actions in sequence: for the region numbered , let the necromancer repeatedly target that region for consecutive turns. By the discussion above, these turns guarantee that five blocks in this region have their skeleton count increased to at least , and afterward each subsequent turn will decrease this count by . However, after completing the turns for region numbered , we will only perform actions on the regions numbered down to , and the total number of turns for these actions is , so the region numbered will have at least five blocks whose skeleton count is no less than . The above reasoning holds for all , so there are doomed blocks. This completes the proof.