← Research

Spectral Forensics

One equation, three domains — the Fiedler vector that finds fraud also finds disease mutations
JIM’S OVERSIMPLIFICATION

Two datasets that share hidden structure have high coupling. A bridge frequency connecting them reveals the relationship. Same math as finding a common friend between two strangers.

K IN THIS DOMAIN

K here is spectral coupling between datasets. Two signals that share structure have high K. A bridge frequency connecting them reveals the relationship.

There is one equation that finds fraud in financial networks, finds disease-causing mutations in proteins, and places 40 million transistors on a chip in 4.5 seconds. Same equation. Same code. Different inputs.

The equation takes any network — any set of things connected to other things — and finds the weakest link. The point where the network would split in two if you cut one connection. The bottleneck. The fault line.

In a financial network, that fault line is where the laundering happens. Enron, Madoff, Wirecard, FTX, Danske Bank — five fraud patterns tested, five detected. The fraudulent structure always sits at the bottleneck because that is where the money has to pass through the fewest hands. The equation finds it by looking at the coupling structure, not the transactions.

In a protein, the same fault line identifies which mutations will cause disease. Change an amino acid at the structural bottleneck and the entire protein’s fold propagates the damage. Change one at the surface and nothing happens. The equation predicts pathogenicity with 82% accuracy using only the shape of the contact network. No training data. No machine learning. Just the coupling structure.

In a computer chip, the same fault line identifies where to partition 40 million gates into groups that minimize wire crossings. Place strongly coupled gates near each other. Weakly coupled gates far apart. 4.5 seconds for 40 million placements.

This is not a claim that proteins are financial networks. The physics differ. What is identical is the graph structure. A bottleneck in a money flow and a bottleneck in a protein backbone and a bottleneck in a chip layout are all the same mathematical object: a sign change in the second-smallest eigenvector of the Laplacian. One number. Same code path in all three domains.

The equation is from 1973 (Fiedler). We did not discover it. We ran it on three things nobody thought to run it on at the same time.

THE EQUATION

L = D - A

L  = graph Laplacian
D  = degree matrix (diagonal: how connected each node is)
A  = adjacency matrix (who connects to whom)

Second-smallest eigenvector of L = Fiedler vector
It splits ANY graph into two communities. Finds bottlenecks. Detects anomalies.
The nodes don't matter. The coupling does.

THREE DOMAINS, SAME CODE

1. Financial transactions (Trace tool)
  Nodes = accounts. Edges = money flows.
  Fiedler vector finds: circular flows, wash trading, layered structures.
  Result: 5/5 fraud patterns detected (Enron, Madoff, Wirecard, FTX, Danske).

2. Protein contact maps (Mutation Scanner)
  Nodes = amino acids. Edges = physical contacts (<8Å).
  Fiedler vector finds: structural bottlenecks, damage propagation paths.
  Result: 0.82 AUC as single structural feature for pathogenicity.

3. Circuit netlists (Chip Fast)
  Nodes = gates. Edges = wires.
  Fiedler vector finds: critical paths, placement bottlenecks, missing connections.
  Result: 40M gates placed in 4.5 seconds using spectral partitioning.

THE CROSS-PREDICTION

Could you detect financial fraud by treating transactions like amino acid contacts? Could you find protein mutations by treating contact maps like transaction flows?

The answer is yes — because the spectral gap doesn't care what the nodes are. A bottleneck in a protein is structurally identical to a bottleneck in a money laundering network. Both show up as sign changes in the same eigenvector.

Already demonstrated: The Networks page shows the same Laplacian math on Zachary's Karate Club (social), chip placement (engineering), protein contacts (biology), and fraud graphs (finance). Fiedler value = 0.4685 on Zachary. Same code path in all four.

HONEST LIMITS

What this is: A well-known property of spectral graph theory (Fiedler 1973, Cheeger inequality).
We are not discovering spectral partitioning. We are showing it works across our three
specific domains with the same code and the same K/R/E/T framework.

What it isn't: A claim that proteins ARE financial networks. The analogy is
structural (both are graphs with coupling), not mechanistic (the physics differ).
GUMPResearch · Support · [email protected] · terms