VCG Payment Identity
Theorem. Under the VCG mechanism with valuation profile $v$ and allocation $a^* = \mathrm{efficientAllocation}\ v$, every agent $i$'s quasi-linear utility under truthful reporting equals $$ u_i \;=\; W(v, a^*) \;-\; \max_{a \in A} W_{-i}(v, a). $$
Equivalently: agent $i$'s utility is their marginal contribution to social welfare.
In Lean: VCGMechanism_quasiLinearUtility_eq_socialWelfare_sub_maxWelfareWithout.
Why this is the soul of VCG
The two headline VCG properties fall out of this identity almost immediately:
-
Truthfulness ([[mechanism_design.vcg_truthfulness_and_ir]]): $u_i$ depends on $i$'s reported valuation only through $a^*$ (the $W_{-i}$ term is report-independent, see [[mechanism_design.vcg.welfare_without]]
_update_selflemmas). Hence maximising $u_i$ over $i$'s report is equivalent to maximising $v_i(a^*)$ given the other agents' reports — which truthful reporting achieves by definition of $a^*$. -
Individual rationality (under non-negative valuations): $u_i \ge 0$ iff $W(v, a^*) \ge \max_a W_{-i}(v, a)$, i.e. agent $i$'s presence weakly increases the achievable social welfare. This is automatic when $v_i \ge 0$ pointwise — adding agent $i$'s valuation can only raise the max.
Proof
sketch
Substitute the Clarke-pivot payment
$p_i = \max_a W_{-i}(v, a) - W_{-i}(v, a^*)$ into the quasi-linear
utility $u_i = v_i(a^*) - p_i$:
$$
u_i \;=\; v_i(a^*) + W_{-i}(v, a^*) - \max_a W_{-i}(v, a)
\;=\; W(v, a^*) - \max_a W_{-i}(v, a),
$$
using socialWelfare_eq_value_add_welfareWithout.
References
- [AGT Chapter 9, §9.3.4] Nisan, Roughgarden, Tardos, and Vazirani, Algorithmic Game Theory. VCG mechanism and the payment identity.
- Vickrey, W. (1961); Clarke, E. H. (1971); Groves, T. (1973). Original VCG papers.