2
$\begingroup$

In Algorithmic Game Theory by Noam Nisan, a (general) mechanism for $n$ player is defined as

$$ M=(\{T_i\},\{X_i\},A,\{v_i\},a,\{p_i\}),\tag1 $$

where $T_1,\dots,T_n$ are the players' type spaces (private information), $X_1,\dots,X_n$ are the players' action spaces, $A$ is the set of alternatives, each $v_i:T_i\times A\to\mathbf R$ is a valuation for player $i$, $a:\times X_i\to A$ is the outcome function, and each $p_i:\times X_i\to\mathbf R$ is the amount that player $i$ pays. (pg. 224 here).

On the other hand, a direct revelation mechanism (pg. 218) is defined as

$$ (f,p_1,\dots,p_n), $$

where $f:V_1\times\dots\times V_n\to A$ is a social choice function from a set $V_i$ of individual preferences $v_i:A\to\mathbf R$ to an alternative, and each $p_i:V_1\times\dots\times V_n\to\mathbf R$ is the amount that player $i$ pays.

Question: How can I recover a direct revelation mechanism from the definition of a general mechanism? More specifically, what should the parameters in $(1)$ be set to, in order to obtain a direct revelation mechanism?

Thank you.

$\endgroup$

1 Answer 1

2
$\begingroup$

A direct revelation mechanism is one in which a player's type space is also their action space ($X_i=T_i$ for all $i$) and the outcome function is the same as the social choice function ($a(t)=f(t)$ for all $t\in T_1\times\cdots\times T_n$).

$\endgroup$
3
  • $\begingroup$ I should have been more clear - in the definition of a direct revelation mechanism, $f$ maps from $V_1\times\dots\times V_n$ to $A$, and each $v_i\in V_i$ is a function $A\to\mathbf R$. Whereas in the definition of a general mechanism, one of the parameters is a set of $n$ (predefined) preferences $v_i:T_i\times A\to\mathbf R$. How should we reconcile these two definitions? I've fixed my question to account for this. $\endgroup$
    – andrew
    Nov 8, 2021 at 4:10
  • 1
    $\begingroup$ Each $t_i$ induces a preference $\succeq$ over $A$, so $T_i$ and $V_i$ can be identified by one another. $\endgroup$ Nov 8, 2021 at 6:36
  • $\begingroup$ @andrew: I think the difference is probably due to sloppy notations. Another possible reconciliation is that $v_i:A\to \mathbf R$ is defined with the assumption of "truth telling" (as if types can be observed), whereas $v_i:T_i\times A\to \mathbf R$ explicitly models the incompleteness of information. This interpretation is supported by the open paragraph of Sec 9.4, which says: "The mechanisms considered so far extract information from the different players by motivating them to "tell the truth."" $\endgroup$
    – Herr K.
    Nov 8, 2021 at 15:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.