How many checkers can you place on a 8-by-8 checkerboard without any checker having three (or more) neighbors?
"Neighbor" refers to a checker that is adjacent horizontally or vertically only.
This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try
refreshing the page, (b) enabling javascript if it is disabled on your browser and,
finally, (c)
loading the
non-javascript version of this page
. We're sorry about the hassle.
Use the pigeonhole principle: Divide the grid in four 4 × 4 squares: At least one square must have 12 blocks if there is 45 or more black squares.
We will prove first that in each subsquare just 1 2 squares fit: Suppose 1 3 fit. Let c , e , b denonte the number of the squares in the corner, edge and "body" of the grid. Also, denote v i , j denote the number of "uses" of the cell i , j and n the total number of "uses", Here, we let a particular cell "use" each neighbor one time and use itself two times. So we can see that each cell has at most 4 two uses, whichever cell we are talking about. So each corner contributes to 4 in the total count, each edge 5 , and each body 6 , and in our 4 × 4 mini square we have 4 c + 5 e + 6 b = n ≤ 4 × 1 6 = 6 4 (four uses available for each one of the 1 6 squares), and we want to maximize c + e + b . Greedily using the cheapest first until exhaustion, we try c = 4 , e = 8 , b = 1 . But in our diagram we see that is impossible. We can easily see that we can't exchange more than two of the border(corner or edge) squares with the body ones(that would cause that the total sum exceeds 6 4 ). The only way of exchaning two borders without breaking our limit is removing two edges and adding two bodies, say: c = 4 , e = 6 , b = 3 . But in the diagram, that is impossible. So we may exchange at most one square in the border with the one in the body. Back in our diagram, we see that that never works, since removing any single square in the border never lets us put two in the body.
So at most 1 2 squares fit in one subsquare. But, can 1 2 fit? In wich ways? First if we consider using just the borders we have a perfect example. A little juggling lefts us with the possibilities:
By simmetry and convinience, we can fix the subsquare with the missing top-right corner in the bottom left(this is the least we can influence the rest of the squares, it would be have the same effect if we chose the one with the top-right and bottom-left corners missing). We see that the possible squares(colored blue) for the subsquare in the right would be:
Now, the minimal sum of neighbors using 1 1 squares (with the lonely corner as contributing with three) is 3 + 4 c + 5 e + 6 b = 3 + 4 × 2 + 5 × 6 + 6 × 2 = 5 3 > 5 2
Since we had 1 3 × 4 uses at most in total (four for each square, 13 possible squares) this is impossible. So the 4 × 4 subsquare in the bottom right can have at most 1 0 checkers and analogously the top-left subsquare must also have at most 1 0 checkers. So in order to have at least 4 5 total squares, the top-right subsquare must have at least 1 3 checkers. Contradiction.
ohh! great effort
Ohh thank you. :-)
What's the difference?
Kevin Bourrillion originally had a ASCII-based picture that didn't really look all that good, so I made a better picture. He then took the above picture and edited it into his post.
It's a lot better than the original "picture" I had, so I edited it into my post. There is no difference. :-)
Exactly my method. :D
I agree the answer is 44, but there is at least one other pattern that also works.
In my above comment I proved that only 4 × 4 subsquares of 10 checkers can go next to a 1 2 -subsquare. So let's build all possible checkerboards of 4 4 checkers. So If we had a single 1 2 -subsquare, we would be forced to have one of this structures:
Since each 1 2 -subsquare invalidates three places in the subsquare next to it, the 1 0 subsquares would need to have this patterns(which are the same up to rotation):
Clearly that is impossible. So we can't have any 1 2 -subsquare. Therefore, since all our four subsquares would have 1 1 or less each one and they need to add up to the 4 4 , they must all be 1 1 -subsquares.
We will now build all the optimal patterns. With a quick-and-diry script to generate all the possible 1 1 -subsquares(there are 1 3 6 of them) and then join all the valid combinations of four of them(using the pertinent branch cuts), we have the following optimal ways(equivalent up to rotation): I would be interested to see that way you propose, because that would imply there is a bug in my code :)
Why hold us in suspense? :-)
I didn't want to spoil it. If you want a hint, just move everything diagonally. Ruins the symmetry, but still adds up to 44.
I made a mistake. When moving everything diagonally, there are two possibilities, but one only adds up to 43 and the other contains two cells that violate the rule. I don't know if there is another solution for 44, but if there is, it will be a very different pattern, and Luis has made a good argument for 44 being the max.
Nice explanation :)
Problem Loading...
Note Loading...
Set Loading...
Diagonal "zigzags" are the most efficient way to pack them in, since one "zigzag" and the next can safely touch on their corners.
(picture courtesy of Daniel)
Imgur
I lack a proof that more is impossible (a colleague found and verified it using a computer program). Can you come up with one?