# Swapping the $i$th largest card between $2$ hands of cards

Introduction to "game"

There are $2$ players and $2n$ cards, labelled $1, 1, 2, 2, 3, 3, 4, 4, \dots, n, n$. ($2$ of each card from $1$ to $n$)

Firstly, the $2n$ players are each given $n$ cards randomly. More specifically, a random ordering of the $2n$ cards is achieved and the first player gets the first $n$ cards, the second player gets the other $n$ cards. We can assume each player can see all the cards.

On every turn, each player will sort his hand of cards. Label the cards in the hands $a_1, a_2, a_3,\dots, a_n, b_1, b_2, b_3,\dots, b_n$.

The players find a pair of cards $a_i\neq b_i$. Then they exchange the cards $a_i$ and $b_i$. After which, the turn ends.

Clearly if $a_i=b_i$ for all $i$, then $a_i=b_i=i$, and the "game" terminates.

Example

Let $n=4$. We call the $2$ players $A, B$. At the start of the game, they receive:

$A: 1, 1, 2, 2$

$B: 3, 3, 4, 4$

Turn 1: They decide to swap the first card.

$A: \color{red}{3}, 1, 2, 2$

$B: \color{red}{1}, 3, 4, 4$

Before the second turn starts, they sort their cards.

$A: 1, 2, 2, 3$

$B: 1, 3, 4, 4$

Turn 2: This time, they decide to swap the third card. They cannot swap the first card as the numbers on the first card are the same.

$A: 1, 2, \color{red}{4}, 3$

$B: 1, 3, \color{red}{2}, 4$

They sort their cards.

$A: 1, 2, 3, 4$

$B: 1, 2, 3, 4$

The "game" ends as the players both have $1, 2, 3, 4$. This game ends in $2$ turns.

Will it terminate? (The answer is yes, and an idea of a bound based on the monovariant will be added as a self-answer.)

In how many moves (at most) does it take for a $2n$ card game to terminate? What arrangements will cause it to take such a long time to terminate?

Assuming that the players pick the next card to exchange randomly (among all the possible moves, choose a valid move, each valid move with equal probability), what is the expected number of moves for the game to terminate?

As suggested by hardmath:

If we constrain the players to take the lowest card that works or the highest card that works, the game will definitely end in at most $n-1$ turns. Both cases are symmetrical, so we work on highest.

Base case: if $n=1$, no moves have to be done. Otherwise, $n>1$.

We now proceed on the inductive step. Suppose the highest cards of the players are different. If they are the same, we delete them, reducing to the $n-1$ case, which can be done in at most $n-2$ moves. Otherwise, one player has both the highest card. When the move is done, a copy of the highest card is given to the other player. Now, both players have the highest card, and it can be deleted. In total, there are $1+(n-2)=n-1$ moves for this.

From this, we can obtain that on average, the game ends in $O(n^2)$ turns, as it takes (on average) $O(n)$ turns for the highest card to be swapped.

Code (for reference):

I ran a Monte Carlo simulation of random games. A simulation of $1000$ games of $1000$ cards gave an average number of swaps of $1310.703$.

C++ code for reference:

#include <cstdio>
#include <algorithm>
#include <cmath>
using namespace std;
int N;
int K;
int maxCounter = 0;
int sumCounter = 0;
int counter;
int arr[999999];
/** Results:
(X cards, 10^6 rounds)
1 -> 0 0
2 -> 1 0.334546 (guess: 1/3)
3 -> 2 0.733746
4 -> 3 1.180559
5 -> 5 1.667006 (guess: 5/3)
6 -> 6 2.179321
7 -> 8 2.721661
8 -> 9 3.283566
9 -> 10 3.867299
10 -> 12 4.471352
Other things I tried: (cards, rounds)
100 1000 -> 143 84.411
200 1000 -> 307 197.332
300 1000 -> 503 319.691
500 1000 -> 960 580.373
1000 1000 -> 1910 1310.703
**/
int main(void) {
srand(23);
printf("Number of cards: ");
scanf("%d", &N);
for (int i=0;i<N;i++) {
arr[i] = arr[i+N] = i;
}
printf("Number of rounds: ");
scanf("%d", &K);
for (int j=0;j<K;j++) {
random_shuffle(arr, arr+2*N);
counter = 0;
while (counter>=0) {
sort(arr, arr+N);
sort(arr+N, arr+2*N);
int pass = 1;
for (int i=0;i<N;i++) {
if (arr[i] != arr[i+N]) {
pass = 0;
break;
}
}
if (pass) break;
counter++;
int theMove = rand()%N;
while (arr[theMove] == arr[theMove+N]) theMove = rand()%N;
int temp = arr[theMove];
arr[theMove] = arr[theMove+N];
arr[theMove+N] = temp;
}
//printf("%d ", counter);
maxCounter = max(maxCounter, counter);
sumCounter += counter;
if (j%(K/100) == 0) printf("Done: %d\n", j);
}
printf("Maximum: %d\nAverage: %lf\n", maxCounter, sumCounter/(double)K);
return 0;
}


Results of the code:

The average number of moves appear to grow faster than $O(n)$.

• As formulated the two players "decide" which of their not yet identical ranked cards to exchange. Have you considered the version of the "game" in which the exchange is always made on the highest (or lowest) possible ranked cards? – hardmath Dec 28 '15 at 12:07
• Note that, however the choice of an exchange is decided, once the highest ranked cards are equal (resp. lowest ranked cards are equal), this will remain so throughout the game. Thus the case is reduced to essentially $n-1$ cards after that point. – hardmath Dec 28 '15 at 12:13
• @hardmath I essentially had the same thing after a little thought. I added it in nevertheless, it might help with the problem. I have found an average-case bound with it. – Element118 Dec 28 '15 at 12:18
• $\sum_i ia_i+ib_i$ won't decrease. Will it always increase? – Empy2 Dec 28 '15 at 12:25
• @Michael, edited. – Element118 Dec 28 '15 at 12:53

Here is a plan with $O(n\log n)$ moves.
Start with $n=2^{m}-1$. Let one hand be $1,1,2,2,...,2^{m-1}$, so the other hand begins with the other $2^{m-1}$.
In $(n-1)/2=2^{m-1}-1$ moves, swap the cards in the middle row every time.
You will end with both cards labelled $2^{m-1}$ at the middle, with the higher cards forming a similar structure above $2^{m-1}$ and the lower cards forming a similar structure below $2^{m-1}$. So one hand has the lowest quartile of cards, and the other hand has the highest quartile of cards. For example, you would have $1,1,2,4,5,5,6$ in one hand and $2,3,3,4,6,7,7$ in the other.
Then repeat the process, for the higher cards and for the lower cards. The total number of moves is $M(m)=2M(m-1)+2^{m-1}-1$.
2 moves are possible for $n=3$, so $M(m)=(m-\frac32)2^{m-1}+1$ for $n=2^m-1$.

This is not a complete answer to the question yet.

The monovariant $\sum i(a_i+b_i)$ was suggested by Michael in the comments.

When the swap is performed, the monovariant remains the same.

The monovariant will increase when any one of the $2$ lists become unsorted after the swap. We try to prove this is the case.

We break into $2$ cases:

Case 1: $|a_i-b_i|>1$

We consider where the cards labelled $a_i+1$ are placed. They either exist as cards $a_j$ where $j>i$ or $b_k$ where $k<i$. In both subcases in case 1, there will be a rearrangement since the hands are not ordered after the swap.

Case 2: $|a_i-b_i|=1$

Without loss of generality, let $a_i-b_i=1$. Then we consider the cards labelled $a_i$ and $b_i$. For convenience, label the cards $x$, $x+1$.

$\begin{matrix} A:&\dots&x+1&\dots\\ B:&\dots&x&\dots \end{matrix}$

Now we consider the positions of the other $2$ cards labelled $x$, $x+1$.

Considering all the cards labelled $1, 2, 3,\dots, x-1$, there are an even number of such cards. All these cards must appear before the cards $x, x+1$ in both lists.

For the unplaced $x, x+1$, if only one appears before the $i$th position, then there are an odd number of positions for an even number of cards, which is impossible.

Hence either both appear before the $i$th position or after the $i$th position.

There are 2 cases (blue: cards to swap, red: after swap):

$\begin{matrix} A:&\dots&\color{blue}{x+1}&x+1&\dots\\ B:&\dots&\color{blue}{x}&x&\dots \end{matrix}\rightarrow\begin{matrix} A:&\dots&\color{red}{x}&x+1&\dots\\ B:&\dots&\color{red}{x+1}&x&\dots \end{matrix}$

$\begin{matrix} A:&\dots&x+1&\color{blue}{x+1}&\dots\\ B:&\dots&x&\color{blue}{x}&\dots \end{matrix}\rightarrow\begin{matrix} A:&\dots&x+1&\color{red}{x}&\dots\\ B:&\dots&x&\color{red}{x+1}&\dots \end{matrix}$

Exhausting both cases proves the monovariant.

Hence, it has been proven that this terminates.

The ending value of the monovariant is $\sum i(2i) = \frac{n(n+1)(2n+1)}{3}$.

The starting value of the monovariant is at least $\sum 2i(n+1-i)$ (the 2 hands are $n, n-1, n-2, \dots, 3, 2, 1$, which is minimal by Rearrangement Inequality).

• Since the two hands are sorted, the minimum might come from one hand being 1,1,2,2,3,3,... – Empy2 Dec 28 '15 at 14:14