Graph-Hodge Decomposition for Accounting

Author: Nirvan Chitnis Date: 2025-11-08 Status: Mathematical Foundation (Appendix)

Abstract

We present the graph-Hodge decomposition of accounting flows, interpreting journal entries as 1-cochains on the account-entity graph. This establishes that internal flows (balanced postings) lie in the divergence-free subspace, while boundary/owner/measurement terms constitute sources. This view connects double-entry bookkeeping to differential geometry and cohomology theory.

1. Motivation: Why Hodge Theory?

Traditional view: Double-entry accounting as a ledger of debits and credits.

Graph-theoretic view: Accounts as nodes, postings as directed edges (incidence matrix representation).

Hodge view: Postings as 1-cochains on a graph, admitting orthogonal decomposition into:

$$\text{All flows} = \underbrace{\text{Gradient (sources)}}_{\text{Boundary}} \oplus \underbrace{\text{Divergence-free (internal)}}_{\text{Balanced}} \oplus \underbrace{\text{Harmonic (cycles)}}_{\text{Zero in acyclic graph}}$$

Practical implication: This decomposition formalizes the distinction between: - Internal flows (Kirchhoff’s law: $\mathbf{1}^\top P = 0$) - Source/sink terms (equity bridge: NI, OCI, owner flows, measurement)

2. Graph Setup

2.1 Account-Entity Graph

Define a directed graph $G = (V, E)$ where:

Example: Revenue recognition:

Cash|ParentCo ←── 100 ──── Revenue|ParentCo

This is a 1-cochain assigning value $+100$ to the edge $(Revenue \to Cash)$.

2.2 Incidence Matrix

The incidence matrix $P \in \mathbb{R}^{|V| \times |E|}$ encodes postings:

$$P_{ij} = \begin{cases} +1 & \text{if edge } j \text{ ends at vertex } i \\ -1 & \text{if edge } j \text{ starts at vertex } i \\ 0 & \text{otherwise} \end{cases}$$

Kirchhoff’s Law (Theorem 1): Balanced postings satisfy:

$$\mathbf{1}^\top P_j = 0 \quad \forall j$$

(Sum of column $j$ is zero: debits = credits.)

3. Hodge Decomposition

3.1 Three Subspaces

For a graph $G$ with Laplacian $L = PP^\top$, the space of all flows $\mathbb{R}^{|V|}$ decomposes as:

$$\mathbb{R}^{|V|} = \text{Im}(L) \oplus \ker(L) \oplus \text{Harmonic}$$

Where:

  1. Gradient (curl-free): $\text{Im}(L) = \{\nabla \phi : \phi \in \mathbb{R}^{|V|}\}$ — Flows derivable from a potential (e.g., equity changes from source terms)
  2. Divergence-free: $\ker(L) = \{\mathbf{1}\}$ — Constant flows (total mass conservation)
  3. Harmonic: $\{x : Lx = 0, \mathbf{1}^\top x = 0\}$ — Cycles (zero for acyclic graphs like accounting trees)

Key insight: Balanced postings live in $\ker(L^\top) = \text{divergence-free subspace}$.

3.2 Application to Equity Bridge

The equity bridge formula:

$$\Delta E = \text{Internal} + \text{Source}$$

corresponds to:

$$\Delta E = \underbrace{P \cdot \text{postings}}_{\text{Internal (div-free)}} + \underbrace{\nabla \phi}_{\text{Source (gradient)}}$$

Where: - $P \cdot \text{postings}$ = internal flows (NI + OCI, after NCI attribution) - $\nabla \phi$ = boundary sources (owner flows, measurement adjustments)

Hodge decomposition: $\Delta E$ splits uniquely into these orthogonal components.

4. Discrete Exterior Calculus View

4.1 Cochain Complex

Accounting flows form a cochain complex:

$$0 \to C^0(G) \xrightarrow{d_0} C^1(G) \xrightarrow{d_1} C^2(G) \to 0$$

Where: - $C^0(G) = \mathbb{R}^{|V|}$: 0-cochains (account balances, scalar potential $\phi$) - $C^1(G) = \mathbb{R}^{|E|}$: 1-cochains (journal entry postings, edge flows) - $C^2(G) = \mathbb{R}^{|\text{faces}|}$: 2-cochains (consolidation boundaries, unused in practice)

Coboundary operators: - $d_0 = P^\top$: gradient (converts potential to flow) - $d_1$: curl (detects cycles; zero for acyclic graphs)

Balanced postings condition: $\mathbf{1}^\top P = 0 \iff P \in \ker(d_1)$ (1-cochains in the kernel of curl are “closed”).

4.2 Hodge Star and Inner Product

Define an inner product on 1-cochains:

$$\langle P_i, P_j \rangle = P_i^\top W P_j$$

where $W$ is a diagonal weight matrix (e.g., monetary amounts).

Hodge star operator $\star: C^1 \to C^1$ (duality) gives the orthogonal complement:

$$C^1 = \text{Im}(d_0) \oplus \ker(d_1)$$

Physical interpretation: - $\text{Im}(d_0)$ = gradient flows (source-driven, e.g., shareholder injections) - $\ker(d_1)$ = divergence-free flows (internal circulation, e.g., revenue → retained earnings)

5. Laplacian and Spectral Properties

5.1 Graph Laplacian

The graph Laplacian is:

$$L = PP^\top \in \mathbb{R}^{|V| \times |V|}$$

Properties: 1. Symmetric positive semidefinite: $x^\top L x \geq 0$ 2. Kernel is constant functions: $\ker(L) = \text{span}\{\mathbf{1}\}$ 3. Eigenvalues: $0 = \lambda_1 < \lambda_2 \leq \ldots \leq \lambda_n$

Accounting interpretation: - $\lambda_1 = 0$: Conservation of total equity (constant mode) - $\lambda_2 > 0$: Algebraic connectivity (how “integrated” the accounting system is)

5.2 Spectral Decomposition

Any flow vector $x$ decomposes as:

$$x = \sum_{i=1}^n \alpha_i v_i$$

where $v_i$ are eigenvectors of $L$. The conservation-preserving projection is:

$$\Pi_{\text{const}} x = \alpha_1 v_1 = \frac{\mathbf{1} \mathbf{1}^\top}{n} x$$

(Projects onto total equity; removes internal fluctuations.)

6. Connection to Physics: Kirchhoff’s Laws

6.1 Electrical Circuit Analogy

Accounting Electrical Circuits
Accounts Nodes
Journal entries Currents
Balanced postings Kirchhoff’s Current Law
Equity potential Voltage
Laplacian $L$ Admittance matrix

Kirchhoff’s Current Law (KCL): At every node, $\sum I_{\text{in}} = \sum I_{\text{out}}$.

Accounting analog: At every account, $\sum \text{debits} = \sum \text{credits}$ (balanced postings).

6.2 Discrete Conservation Laws

The discrete divergence is:

$$\text{div}(P_j) = \mathbf{1}^\top P_j$$

Theorem (Discrete Divergence-Free Flows): A posting $P_j$ is balanced iff:

$$\text{div}(P_j) = 0$$

This is the discrete analog of $\nabla \cdot \mathbf{F} = 0$ in fluid mechanics (incompressibility).

7. Practical Implications

7.1 Validation via Hodge Decomposition

Algorithm:

  1. Compute Laplacian $L = PP^\top$
  2. Decompose equity changes: $\Delta E = \Delta E_{\text{grad}} + \Delta E_{\text{div-free}}$
    • $\Delta E_{\text{grad}} = L^+ \nabla \phi$ (source terms)
    • $\Delta E_{\text{div-free}} = (I - L L^+) \Delta E$ (internal flows)
  3. Verify: $\|\Delta E_{\text{div-free}}\|$ should match NI + OCI (after NCI attribution)

Where $L^+$ is the Moore-Penrose pseudoinverse.

7.2 Diagnostics

Harmonic component check: For acyclic accounting graphs (tree structure), harmonic component should be zero:

$$\text{Harmonic}(\Delta E) = \Delta E - \text{Grad}(\Delta E) - \text{Const}(\Delta E) \approx 0$$

If non-zero: Indicates cycles (e.g., circular ownership, intercompany eliminations not fully resolved).

8. Example: Simple Equity Bridge

Setup: Two accounts: Retained Earnings (RE), Cash.

Journal entry: $+100$ Cash (debit), $+100$ Revenue (credit, flows to RE).

Incidence matrix:

$$P = \begin{bmatrix} +1 \\ -1 \end{bmatrix} \quad \text{(Cash debit, Revenue credit)}$$

Laplacian:

$$L = PP^\top = \begin{bmatrix} 1 & -1 \\ -1 & 1 \end{bmatrix}$$

Kernel: $\ker(L) = \text{span}\{\begin{bmatrix} 1 \\ 1 \end{bmatrix}\}$ (constant total equity).

Gradient flow: Source term (shareholder injection) would add to both accounts equally (increasing total equity).

Divergence-free flow: Revenue → RE (internal redistribution, no equity change).

9. Future Directions

  1. Persistent homology: Track evolution of accounting graph topology over time (e.g., M&A changing graph structure).
  2. Spectral clustering: Group accounts by Laplacian eigenvectors (automatic classification into operating/financing/investing).
  3. Hodge-theoretic anomaly detection: Unusual flows as outliers in gradient/divergence decomposition.

10. Conclusion

Main Result: Accounting flows admit a Hodge decomposition into:

$$\text{Flows} = \text{Gradient (sources)} \oplus \text{Divergence-free (internal)} \oplus \text{Harmonic (cycles)}$$

Practical Impact: - ✅ Formal justification for Kirchhoff’s law (balanced postings = divergence-free) - ✅ Orthogonal decomposition of equity changes (source vs. internal) - ✅ Spectral methods for classification and anomaly detection

Connection to Literature: This extends Ellerman (2014) and Liang (2023) by placing accounting in the framework of discrete exterior calculus and spectral graph theory.

References

  1. Lim, L.-H. (2020). Hodge Laplacians on Graphs. arXiv:1507.05379 — Discrete Hodge theory foundations
  2. Desbrun, M., et al. (2005). Discrete Exterior Calculus. arXiv:math/0508341 — Cochain complexes and operators
  3. Ellerman, D. (2014). Accounting and Category Theory. arXiv:1412.4229 — Double-entry as adjoint functors
  4. Liang, P. (2023). Kirchhoff’s Current Law for Accounting. — Graph-theoretic incidence matrices
  5. Chung, F. R. K. (1997). Spectral Graph Theory. AMS. — Laplacian eigenvalues and algebraic connectivity

Generated with Claude Code (claude-sonnet-4-5) Co-Authored-By: Claude