Theorem 1 (Existence and Uniqueness of Stable Point), Expanded Version
ProvedIncentiveFinNet.theorem_1Fix a network setting : agents, an expected-return matrix whose -th column is , risk aversions , covariance matrices (one per agent), and permitted counterparty sets with . Let , let , and define the matrices
Let be the matrix whose rows are , . Then:
- a stable point exists if and only if lies in the column space of ;
- every stable point satisfies
- if has full rank, there is exactly one stable point .
Here a stable point is a pair with symmetric, skew-symmetric, both zero on prohibited pairs, such that no agent can raise its mean-variance utility by changing to another feasible network at the same prices.
The theorem turns a fixed-point question about selfish agents into linear algebra: the existence of stable contracts and prices is the solvability of one linear system in the prices, and uniqueness follows from the invertibility of its matrix.
Formalization Note The column space of is the range of the linear map ; full rank of the square matrix is ; uniqueness in part 3 is uniqueness of the pair . Every is covered, including and settings with , where is the empty matrix and has full rank.
Differences from the source:
- reading: if k ∉ J_i (and hence i ∉ J_k) → the setting assumes k ∈ J_i ↔ i ∈ J_k; reason: the parenthetical states the symmetry of the permitted sets
- reading: W and P obey the constraints encoded in (Ψ_i) (Definition 2) → k ∉ J_i implies W_ik = 0 and P_ik = 0; reason: the p.9 sentence "if k ∉ J_i (and hence i ∉ J_k), we have W_ik = W_ki = P_ik = P_ki = 0", together with the symmetry of J, W and −P
- reading: g_i(W,P) = max over feasible (W′,P) of g_i(W′,P) (Definition 3) → g_i(W′,P) ≤ g_i(W,P) for every feasible (W′,P); reason: equivalent because (W,P) is itself feasible
- encoding: agents [n], ordered sets J_i and F, vectors in ℝ^|F|, the |F| × |F| matrix Z_F → Fin n, Finset J_i, the subtype F of pairs, functions on F and a matrix indexed by F × F; reason: Lean conventions; the orders of J_i and F change no quantity of the theorem
- encoding: Σ_i ≻ 0 → Matrix.PosDef (symmetric); reason: a covariance matrix is symmetric
- encoding: column space of Z_F; Z_F is full rank; a unique stable point exists → LinearMap.range of Z_F.mulVecLin; rank Z_F = card F; ∃! of the pair (W, P); reason: standard Lean forms of the same notions
import Mathlib import Definitions.Def_IncentiveFinNet_NetworkSetting import Definitions.Def_IncentiveFinNet_Stable import Definitions.Def_IncentiveFinNet_matC import Definitions.Def_IncentiveFinNet_Q import Definitions.Def_IncentiveFinNet_ZF open Matrix
namespace IncentiveFinNet
/-- **Theorem 1 (Existence and Uniqueness of Stable Point)**, Expanded Version, Jalan,
Chakrabarti, Sarkar, *Incentive-Aware Models of Financial Networks*, arXiv:2212.06808v3,
pp.11–12; Expanded Version, A.1, p.33.
With `A_ij = e_iᵀ Q_j M e_j`, `B_{(i,j)} = e_i e_jᵀ Q_i`,
`C_{(i,j)} = (B_{(i,j)} − B_{(j,i)}) − (B_{(i,j)} − B_{(j,i)})ᵀ`, and `Z_F` the `|F| × |F|` matrix
with rows `uvec(C_{(i,j)})_F`, `(i, j) ∈ F`:
1. a stable point `(W, P)` exists iff `uvec(A − Aᵀ)_F` lies in the column space of `Z_F`;
2. every stable point satisfies `Z_F uvec(P)_F = uvec(A − Aᵀ)_F`;
3. if `Z_F` has full rank, a stable point exists and is unique.
**Formalization Note** The column space of `Z_F` is `LinearMap.range Z_F.mulVecLin`; "full rank"
for the square matrix `Z_F` is `rank Z_F = |F|`; uniqueness is of the pair `(W, P)`. -/
theorem theorem_1 {n : ℕ} (S : NetworkSetting n) : ((∃ W P : Matrix (Fin n) (Fin n) ℝ, Stable S W P) ↔ uvecF S (matA S - (matA S)ᵀ) ∈ LinearMap.range (ZF S).mulVecLin) ∧ (∀ W P : Matrix (Fin n) (Fin n) ℝ, Stable S W P → ZF S *ᵥ uvecF S P = uvecF S (matA S - (matA S)ᵀ)) ∧ ((ZF S).rank = Fintype.card (PermittedPairs S) → ∃! WP : Matrix (Fin n) (Fin n) ℝ × Matrix (Fin n) (Fin n) ℝ, Stable S WP.1 WP.2) := by sorry
end IncentiveFinNet
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.