Kac asked in 1966: can you hear the shape of a drum? The answer, famously, is no — non-isometric drums can produce identical spectra. For undirected graphs, Haemers conjectured the opposite: almost all graphs are determined by their spectra. Zhao proves that for directed graphs, the situation collapses entirely. Almost no digraph can be reconstructed from its spectrum.

The result is stronger than “most digraphs have cospectral mates.” It shows that almost all digraphs aren’t even isomorphic to their own reversal — reversing all edge directions produces a non-isomorphic graph that shares the same spectrum. Direction carries information that frequencies cannot encode.

This is a clean demonstration of what symmetry buys. Undirected graphs have real symmetric adjacency matrices. Symmetric matrices have well-behaved eigenvalue decompositions. That structure makes spectral determination plausible. Directed graphs have asymmetric matrices, and asymmetry destroys the spectral fingerprint. The eigenvalues forget which way the edges point.

The engineering analog: any system where you can only observe frequencies — vibration data, network traffic statistics, aggregate throughput — cannot reconstruct the directionality of the underlying structure. If the flow has a direction, passive frequency measurement loses it. This isn’t a limitation of your instruments. It’s a mathematical impossibility. The spectrum literally doesn’t contain the information.

To hear the shape of a directed graph, you need something other than spectral data. You need to probe the asymmetry directly — send a signal and observe which way it goes.