Why is Chinese chess (xiangqi) harder for computers than international chess?

Question

Why is Chinese chess (xiangqi) harder for computers than international chess?

When one looks at Chinese chess in comparison to international chess, one cannot help but notice how extremely limited several of the pieces are:

1. Kings are confined to 9 positions overall (the palace) and have 4 possible moves at most (only 1 step orthogonally).
2. The two advisors are confined to 5 shared positions overall (diagonals of the palace), where except for the central palace position they are confined to 1 move.
3. The two elephants are confined to 7 shared positions (two touching diamonds side by side, like an 8 on its side), due to having a move that naturally restricts them to a quarter of the board (only every other square the same color as its starting square (on a checkered board)), but then also being restricted to not move across the central dividing river which cuts it down to an expected 1/8. In fact the 7 positions out of 90 which the elephants can reach is an even worse ratio than 1/8 which is due to the specific board dimensions penalizing it even further.

Is there something about the two horses, two chariots (rooks), two cannons, and five chinese pawns that makes this game harder for computers than international chess? Or is it mostly due to the larger board size, 90 vs 64? Or is there some other consideration that I have overlooked?

Answer 1

As far as I can see there is no fundamental difference between chess and xiangqi that would make xiangqi more difficult for a computer.

The state space complexity of chess is somewhat higher, whereas the game tree complexity of xiangqi is higher. Also the branching factor of xiangqi is 38 compared to 35 for chess, not much of a difference.

I suspect that while chess has been in the focus of computer research for many decades, there just weren't the same resources devoted to optimising xiangqi engines.

Possibly the slower game (because of the weaker pieces) requires the calculation of longer variations (see game tree complexity), increasing the hardware requirements, but I think that should only be a minor difference compared to demonstrably more complex games like shogi or go.

It is a bit difficult to find out precisely how strong xiangqi engines are right now, but any of the two factors would explain if they lagged one or two decades behind.

Answer 2

Xiangqi has more theoretical complexity mainly due to larger board. So if you need to do exhaustive search, it has larger search space.

Xiangqi is also more positional. For attacking each side has two rooks, 4 knight-strength pieces, and 5 pawns. So for a crude comparison with western chess, Q+3P are replaced with 4 purely defensive pieces. We know positional chess is difficult for computer.

Answer 3

Xiangqi is very tactical, not positional, doesn't have chains or pawns for a long time. The difficulty to computers is the many positions available to the pieces, and the board is 10 x 10.

I think makruk or international chess is more more positional than xiangqi and shogi.

Answer 4

Xiangqi is less "connected" than western chess. For instance, the five pawns cover only every other file; they're not side by side. There is no "queen" that unifies both lateral and diagonal movement. And the defensive pieces are either limited to the palace (advisors) or to patrolling the surround countryside (elephants); they cannot be used to help attack the opposing king. Lastly, there is no pawn promotion, which means that it is hard to win if too many pieces have been exchanged.

Because of the above "disconnects," it is harder to construct viable winning combinations, or even winning positions than in western chess. In the latter, for instance, with queening, the win of a pawn in an otherwise level position means winning the game. That's not necessarily so in Chinese chess.