Existing game theory applied to SCC, to my knowledge, only applies to the complete game tree for SCC. Game theory applied to SCC says what I put in
lichess.org/forum/team-jomegas-tabia/theoretical-chess#4.
I see no application to SCC with the complete game tree to the subjects listed in the links in #5
lichess.org/forum/team-jomegas-tabia/theoretical-chess#5That leaves us in my mind with the question, "Can existing game theory help in analyzing SCC when all one has is a partial game tree (per definition in
en.wikipedia.org/wiki/Game_tree).
At the moment, I don't see how game theory can help there. Game theory assume you know what the payoffs are. Either the payoffs for the terminal nodes in the extensive form representation, or in the matrix in the normal form representation. But we don't know that. At best we have a probability we might assign a payoff.
I see that this might be covered in game theory as part of its talk on games with incomplete information.
en.wikipedia.org/wiki/Extensive-form_game#Incomplete_informationLook at that link carefully. They introduce "Nature" to assign the probabilities as a third player and then they say "In the case of private information, every player knows what has been played by nature." Then the example still has the final payoffs (listed on the left and right of the picture).
Indeed, in the formal definition you can see that the payoffs are known; they are specified by the payoff profile function.
en.wikipedia.org/wiki/Extensive-form_game#Formal_definitionThe probabilities are then "nature's" choices that determine the actual payoffs applicable per the player's choices. Look back at the simple example they have in the game depicted by the picture under the title section "Incomplete Information" - the payoffs are listed on the left and right of that picture and effectively decided by nature's choice. In that example, nature "played" first. Nature chooses the ability of the job applicant. In the general mathematical model the probabilities are specified and known by the players.
Let's try to apply this to SCC with a partial game tree. To make this incredibly simple, assume we have the limiting case of the empty partial game tree, which means we are only looking at the initial position of chess. "Nature" in the sense of that web page, is choosing the payoffs for the game, but what can nature be choosing? Chess is deterministic.
The only thing I can think of is to pretend that nature chooses the probabilities of the payoffs for the 20 possible White moves. The payoffs can still be taken as the usual (1,0), (0,1) or (1/2,1/2) but now the picture splits into 20 pieces depending on White's first move (in analogy to that picture on that page splitting in 2 depending on player 1's choice of U or D and nature's choice of its first move).
White knows the probabilities and the payoffs. So how does White proceed? I think it is simple. For each of the 20 moves there are probabilities p1, p2, p3 for the above payoffs. White computes a probability-payoff for a move as p1 * 1 + p2 * 0 + p3 * 0.5, orders the moves by the largest of these to the smallest and then picks any move with a largest probability-payoff.
For example if we take the Lichess master database as nature then for 1.e4 we have
p1=0.33, p2 = 0.25, p3=0.42 and the probability-payoff is 0.54. Continuing with the other first choices for White and assuming ones not listed have a probability of 0,0,0 we can choose a move with the largest probability-payoff.
A human or chess program could assign similar probabilities determined in any way they see fit. Typically, this is what a human means when they say they've evaluated their chance of a win/loss/draw in a position. Computers typically give their evaluation of a position not as a list of 3 probabilities for win/loss/draw, but as a real number indicating their assessment of the position on some point-count scale - such as what Stockfish does.