Hodge-Shapley Value

• Hodge-Shapley value is a mathematical framework that extends the classical Shapley value by integrating discrete Hodge theory and stochastic processes to allocate rewards for any coalition state.
• It employs a stochastic path integral over reversible Markov chains and solves discrete Poisson equations to determine fair payoffs even in weighted or constrained network settings.
• The framework is uniquely characterized by extended axioms including efficiency, linearity, symmetry, and independency, unifying game theory with combinatorial and analytical methods.

The Hodge-Shapley value is a mathematical framework that generalizes and extends the classical Shapley value from cooperative game theory to settings involving arbitrary coalition structures, partial coalitions, and general cooperative networks. By integrating tools from discrete Hodge theory, stochastic processes, and combinatorial Laplacians, the Hodge-Shapley value provides a principled allocation rule for transferable-utility (TU) games not only at the grand coalition, but for every possible partial coalition. Furthermore, it accommodates generalization to weighted, constrained, or networked settings via a correspondence with solutions to discrete Poisson equations on associated graphs. This framework, originated through independent lines of work by Lim, Stern-Tettenhorst, Mastropietro, Vaccarino, and others, unifies classical value concepts with combinatorial and analytical methods (Lim, 2022, Lim, 2021, Stern et al., 2017, Lim, 2021, Mastropietro et al., 2023).

1. Path Integral Representation and Generalization of the Shapley Value

At its core, the Hodge-Shapley value is defined as the expected value of a stochastic path integral over a reversible Markov chain on the coalition (hypercube) graph. In the classical case, the coalition space is the Boolean hypercube $G=(V,E)$ with vertices $V = 2^N$ and transitions corresponding to the sequential joining or leaving of a player.

Given a game $v:2^N \to \mathbb{R}$ with $v(\emptyset)=0$, for each player $i$ one defines a marginal edge flow $\partial_i v: E \cup E_- \to \mathbb{R}$ by

$$\partial_i v(S,S\cup \{j\}) = \begin{cases} v(S\cup \{i\})-v(S), & j=i, \ 0, & j\neq i, \end{cases}$$

and $\partial_i v(T, S) = -\partial_i v(S, T)$. Running a reversible Markov chain from the empty set to a terminal coalition $T$, one considers the stochastic path integral

$$I_i(v, T)(\omega) = \sum_{t=1}^{\tau_T(\omega)} \partial_i v(X_{t-1}(\omega), X_t(\omega)),$$

where $\tau_T$ is the first hitting time of $T$. The Hodge-Shapley value for player $i$ at $T$ is then

$\Phi_i(v, T) := \mathbb{E}[I_i(v, T)],$

with the expectation taken over all sample paths of the chain. This operator $\Phi$ extends the classical Shapley value, in that for $T=N$ (the grand coalition), $\Phi_i(v, N)$ recovers $\operatorname{Shapley}_i(v)$. For arbitrary $T$, the allocation $\Phi_i(v, T)$ specifies the "fair" reward to player $i$ if the coalition process terminates at $T$ (Lim, 2022).

2. Axiomatic Characterization and Extension to Partial Coalitions

The Hodge-Shapley value is uniquely characterized by an axiom system that extends Shapley's classical four axioms (efficiency, symmetry, null-player, linearity) to all partial coalitions simultaneously, supplemented by an independency (reflection) property:

• Efficiency (A1): For every $v$ and $T$, $\sum_{i\in N} \Phi_i(v, T) = v(T)$.
• Linearity (A2): $\Phi$ is linear in $v$.
• Symmetry (A3): Swapping labels $i \leftrightarrow j$ in $v$ and $T$ exchanges $\Phi_i, \Phi_j$.
• Extended Null-Player (A4): If $\partial_i v \equiv 0$, then for any $j\ne i$ and $S$, $$\Phi_j(v, S\cup\{i\}) = \Phi_j(v, S) = \Phi_j(v_{-i}, S)$$.
• Independency/Reflection (A5): For player $i$ and $S, T \subset N \setminus\{i\}$,

$$\Phi_i(v, T\cup\{i\}) - \Phi_i(v, S\cup\{i\}) = -(\Phi_i(v, T) - \Phi_i(v, S))$$

or, equivalently, the average $\Phi_i(v, S) + \Phi_i(v, S\cup\{i\})$ is independent of $S$.

These properties ensure the allocation is uniquely specified for every coalition and recover the classical solution at $T=N$ (Lim, 2022, Lim, 2021).

3. Hodge Decomposition and Discrete Poisson Equation

A key feature of the Hodge-Shapley framework is that its allocations are characterized as solutions of discrete Poisson equations on the coalition graph using combinatorial Hodge theory. Consider the differential operator $d : \ell^2(V) \to \ell^2(E)$, $d u(S, T) = u(T) - u(S)$, and its adjoint $d^*: \ell^2(E) \to \ell^2(V)$. The combinatorial Laplacian is $L = d^* d$.

For each player $i$ and marginal edge flow $f_i$, the unique zero-based solution $u_i: V \to \mathbb{R}$ to the Poisson equation

$L u_i = d^* f_i, \quad u_i(\emptyset) = 0,$

gives the Hodge-Shapley value, with $u_i(T) = \Phi_{f_i}(T)$. In the special case $f_i = \partial_i v$, $u_i = v_i$ and $\Phi_i(v,T)=v_i(T)$ (Lim, 2022, Stern et al., 2017).

The orthogonal Hodge decomposition,

$$\partial_i v = d\, v_i + r_i, \quad r_i \in \ker d^*,$$

separates each marginal into a cut (exact) and cycle (harmonic) component. Only the exact part determines the allocation; harmonic/circulation flows are irrelevant to payoff determination.

A closed-form expression is available using the Green's function $G = L^+$: $$v_i(T) = \sum_{S \in V} G(T,S)\, [d^* \partial_i v](S).$$

4. Extension to Weighted and General Graphs

The Hodge-Shapley theory generalizes classical coalition games by replacing the hypercube with an arbitrary finite connected network $(V,E,\omega)$ supporting any reversible Markov process (Lim, 2021, Stern et al., 2017). In this case:

• The cooperative game becomes any function $v : \mathcal{E} \to \mathbb{R}$, $v(\varnothing)=0$.
• Each edge $(S,T)\in \mathcal{A}$ may be assigned an $N$-tuple $f_i(S,T)$ representing the payoffs to each player upon a state transition.
• The value allocation for each player—defined as the expected stochastic path integral along the reversible chain—remains uniquely characterized by an analogue of the Hodge decomposition and is the solution to an appropriate Poisson equation.

Analogous Poisson equations and path integral definitions apply, under suitable adaptations, for weighted, directed, or constrained coalition formation processes (Lim, 2021, Stern et al., 2017).

5. Associated Games and the Shapley-Hodge Game

An alternative formulation, the Shapley-Hodge Associated Game (SHoGa), frames the allocation problem as an operator on games: for any TU-game $u: P(N) \rightarrow \mathbb{R}$,

$$\chi_u(S) := \frac{1}{2}[u(N) - u(N\setminus S) + u(S)].$$

The Hodge-Shapley value is then $\phi_i^{Hodge}(u) = \phi_i(\chi_u)$, applying the Shapley value to the associated game $\chi_u$. This map $\chi$ is characterized by five coalitional axioms (average-efficiency, null-coalition, bilaterality, constant-sum, and linearity) and uniquely satisfies relevant fairness, linearity, and symmetry properties.

An explicit marginal-contribution formula for $\phi^{Hodge}$ is given by

$$\phi_i^{Hodge}(u) = \sum_{T \subset N\setminus \{i\}} w(T) [u(T\cup \{i\}) - u(T)],$$

where $w(T)$ is a combination of standard Shapley weights and Green's function differences determined by the coalition graph Laplacian (Mastropietro et al., 2023).

6. Algorithmic Computation and Illustrative Examples

Computation of the Hodge-Shapley value typically proceeds as follows:

1. Construct the Laplacian $L$ for the coalition or state-transition graph.
2. For each player, form the right-hand side using the player's marginal or prescribed edge flows.
3. Solve the sparse, positive-semidefinite Poisson system $L u_i = d^* f_i$ with $u_i(\emptyset) = 0$.
4. Read off allocations $u_i(T)$ for any $T$ of interest.
5. For the associated game approach, solve $L x = L_S u$, recover $\chi_u(S)=x(N)$, then apply the Shapley value.

For illustration, consider the glove game (players 1 with left glove, players 2,3 with right gloves; $v(S)=1$ iff $1 \in S$ and $2$ or $3$ as well). The Hodge-Shapley allocations across all $T\subset\{1,2,3\}$ can be computed, with classical Shapley values recovered for $T={1,2,3}$ (Lim, 2022, Mastropietro et al., 2023).

7. Theoretical and Practical Significance

The Hodge-Shapley value extends the foundational fairness and linearity principles of the Shapley value to arbitrary coalition structures and processes, yielding a full allocation map for every coalition state, not limited to the grand coalition. By anchoring its construction in discrete Hodge theory and stochastic processes, it unifies combinatorial, algebraic, and probabilistic perspectives on value allocation. Beyond classical TU-games, the Hodge-Shapley value accommodates weighted, constrained, or networked settings and offers efficient linear algebraic computation via Laplacian solvers, with applications in economic, financial, and network-centric problems (Lim, 2022, Lim, 2021, Lim, 2021, Stern et al., 2017, Mastropietro et al., 2023).