4

In the question on nimbers, the original poster asks for the meaning of Nimber multiplication in the context of impartial games.


Edit: As noted by Mark Fischler in the comments below, the following is wrong

My gut instinct is $*a \times *b$ means that if $*a$ is a game equivalent to $a$ stones, and $*b$ is a game equivalent to $b$ stones, then if you replace every stone in the $*a$ game with a copy of the $*b$ game, you get a game with the Nimber $*a \times *b$, but I haven't been able prove it.

yberman
  • 1,974
  • Note: the original question did not provide a meaningful definition to the construct as described on wikipedia. – yberman Feb 25 '16 at 15:48
  • 1
    If you look at the nimber multiplication chart in wikepedia, it does not seem to agree with your conjectured meaning. For example, $5 \times 4 = 2$ which is not the case if you have filve piles of 4 stones and a move consists of making moves in any subset of the piles. – Mark Fischler Feb 25 '16 at 16:02
  • Thank you @MarkFischler, I will note this as wrong in the question. – yberman Feb 25 '16 at 16:16
  • 2
    What moves are available in your game $a \times b$? If $a=5, b=4$, it would seem I have twenty stones on the table grouped into five piles of four. I should be able to remove any set of complete piles (move in $a$) or any set of stone from one pile (move in $b$). Note your multiplication seems to be non-commutative. – Ross Millikan Feb 25 '16 at 16:19
  • @RossMillikan If you have 2k+1 piles of m stones, doesn't that have the same nimber as 1 pile of m stones? I have to review my definition of nimbers. – yberman Feb 25 '16 at 16:22
  • @Yosef Yes, those would be equivalent since the $2k$ piles would cancel out. – Théophile Feb 25 '16 at 16:28
  • That is why I asked what moves were available. You said to replace each stone in $a$ with a pile of $b$. If we just play that as a game of Nim, yes the matching piles cancel. For $a=5$ I would call that $5 \times b$, not $5 \times b$ and you can prove that for odd $a$, $a \times b=b$ this way. In $a \times b$ it seems I should still be able to move in $a$, which should allow me to take as many whole piles (but not partial ones) as I want. – Ross Millikan Feb 25 '16 at 16:36
  • You are correct @RossMillikan. Four piles of 5 will equal $0$, and five piles of 4 would equal $4$. I was just mulling over it. – yberman Feb 25 '16 at 19:00

3 Answers3

9

I don't think there's an intuitive way to understand nimber multiplication. Also note that it is distinct from repeated addition, so it isn't as simple as replacing stones with copies of piles. Multiplication is defined recursively as $$ab = \text{mex}\left(\{a'b+ab'+a'b': a'<a, b'<b\}\right)$$ where $\text{mex}$ is the minimum excluded element.

Although it's hard to see what this means in terms of a concrete game, it's possible to understand why it's defined this way. The point is that we want to create an algebraic system without zero divisors, so $(a-a')(b-b') \neq 0$ whenever $a\neq a'$ and $b\neq b'$. In other words, $$ab - ab' - a'b + a'b' \neq 0,$$ so $$ab \neq a'b + ab' - a'b'.$$ Since subtraction and addition are the same operation, we have $$ab \neq a'b + ab' + a'b'.$$ The definition above takes the first nimber that meets this criterion.

Théophile
  • 24,749
3

Winning Ways Volume 3, Chapter 14, considers coin-turning games. One such game is Turning Corners, where a move is to turn over the four corners of a rectangle with horizontal and vertical sides, but the North-East most coin needs to be turned from heads to tails. A sample move shows that it defines nimber multiplication:

a           X           H
.  
.  
.  
a'          X           X
.  
.  
.  
0  .  .  .  b' .  .  .  b

A move will turn the H to tails and Xs over, so from the position $(a,b)$, a move moves to the sum $(a',b)+(a,b')+(a',b')$, so the value of the position $(a,b)$ is the nimber product $ab$ inductively.

E. Z. L.
  • 135
  • So the nimber of a game state is the nim-sum of all the products where there are heads. Do we know whether a player given a nonzero nimber state can always reduce it to zero? – Oscar Lanzi Jul 21 '21 at 21:09
  • what if I choose the a' and b' to be 0, then my adversary wouldn't be able to make a move such that I win the game. is that correct? it just seems too trivial and makes no sense to me. – Liu Weibo Mar 20 '23 at 19:56
  • If you do that, you result in (a',0)+(0,b') – E. Z. L. Mar 21 '23 at 23:10
  • I meant (a,b)+(0,b), which is certainly not an autmoatic win – E. Z. L. Mar 21 '23 at 23:11
2

The "game" in this video may be analyzed in terms of a limited form of nimber multiplication, specifically involving the nimbers $1,2,3$.

The rules of the game are as follows:

  • Start with disks (or hexagons in the video) of three different colors. Fill a base row of a triangular array with some permutation of $n$ disks.

  • This generates $n-1$ adjacent pairs. On top of each adjacent pair add a disk of the same color if the two disks below are identical, add a disk of the third color otherwise. Repeat this process for the next layer up until you reach one final disk at the top.

For certain values of the layer count $n$ the color of the top disk will depend only on the two other corners of the triangle; all the other disks in the base layer and all the intermediate layers cancel themselves out. The special layer counts are

$n=2,4,10,...; n\in\{3^k+1,k\in\mathbb{N}\}.$

We may analyze this in terms of nimber multiplication by rendering the colors as the nimbers $1,2,3$. With this rendering the placement rule is simply that the nimber product of any two horizontally adjacent disks with the one above is $1$, to wit

$1\otimes1\otimes1=2\otimes2\otimes2=3\otimes3\otimes3=1\otimes2\otimes3=1.$

Thus disks with colors $a$ and $b$ will be topped by a disk whose color corresponds to $a^{-1}b^{-1}$, where the product is understood to be the nimber product. For an $n$-layer array, the top disk will have a combined product of all the base disks raised to some powers; for instance with six layers the top disk carries the product

$a^{-1}b^{-5}c^{-10}d^{-10}e^{-5}f^{-1}$

where the disks are in order from $a$ to $f$ across the base row. The exponent will in general be $(-1)^{n-1}$ times the entry in Pascal's Triangle corresponding to the disk's position. The exponents always sum to $(-2)^{n-1}\equiv1\bmod 3$.

Since each nimber cubed is $1$, the product may be simplified by reducing each exponent to a residue $\in\{-1,0,1\}\bmod 3$. For the six-layer case above this gives

$a^{-1}bc^{-1}d^{-1}ef^{-1}$

whereas a product depending only on the corners would have been $a^{-1}f^{-1}$. Thus six layers is not a "special number". But for four layers we do properly get $a^{-1}d^{-1}$ as the intervening elements in Pascal's Triangle are multiples of $3$ ($1~\color{blue}{3~3}~1$). We see, given well-known properties of Pascal's Triangle, the "special numbers" correspond to $n-1$ being a power of $3$ as indicated in the video.

Oscar Lanzi
  • 48,208