Dale's Principle says a neuron is either excitatory or inhibitory — it releases one type of neurotransmitter at all its synapses. An excitatory neuron excites every neuron it connects to. An inhibitory neuron inhibits every neuron it connects to. The principle constrains the sign of the coupling: each row in the connectivity matrix has a uniform sign. This isn't a theorem derived from dynamics — it's an empirical observation promoted to a modeling constraint. Nearly every large-scale neural network simulation enforces it.
Ah-Weng and Rajpal (arXiv 2602.23202, February 2026) build networks of Izhikevich neurons with and without this constraint and find that Dale's Principle hides an entire dynamical regime. When neurons obey the principle — each one purely excitatory or purely inhibitory — the network shows a monotonic transition from asynchronous to synchronous firing as coupling strength increases. Increase the coupling, increase the synchrony. The relationship is smooth and predictable.
When neurons are allowed mixed-sign connectivity — the same neuron exciting some targets and inhibiting others, violating Dale's Principle — a three-regime phase structure appears. At weak coupling, the network is asynchronous. At intermediate coupling, synchronous oscillations emerge. But at strong coupling, instead of locking further into synchrony, the network breaks back into noise-dominated asynchrony. The non-monotonic behavior — synchrony rising and then falling — is invisible to the Dale-constrained model.
The reentrant asynchrony at strong coupling is driven by the mixed-sign connections creating competing feedback loops. An excitatory signal from one neuron simultaneously excites some targets and inhibits others. At moderate coupling, the excitatory pathways dominate and synchrony wins. At strong coupling, the inhibitory pathways become powerful enough to destabilize the synchronized state, creating a frustration that the uniform-sign model can't produce.
The biological evidence for Dale violations is real but contested. Co-release of glutamate and GABA from single neurons has been observed in hippocampus and cortex. Developmental switching between excitatory and inhibitory identity occurs in early circuits. The extent to which mature neural circuits violate Dale's Principle remains an active question, but the modeling assumption — that they don't — is treated as settled in most computational neuroscience.
The paper doesn't claim that real brains violate Dale's Principle wholesale. It claims that the dynamical consequences of the assumption are larger than the assumption appears. A modeling constraint that looks like a simplification — just fixing the sign of each row in a matrix — qualitatively changes the phase structure of the system. The constraint isn't removing noise or irrelevant detail. It's removing a dynamical regime.
This connects to a broader pattern in modeling: constraints inherited from empirical regularities can suppress the dynamics that would reveal whether the regularities hold. If your model enforces Dale's Principle and the network behavior looks monotonic, you can't tell whether the monotonicity is physics or artifact. The constraint and the prediction become circular. The model confirms the principle by forbidding the dynamics that would violate it.