Researchers have developed exact, training-free neural-network representations for both colorless and colorful Motzkin states. Motzkin states are paradigmatic frustration-free one-dimensional quantum systems known for their exactly solvable combinatorial structures and, crucially, for entanglement scaling that violates the area law. Specifically, colorless Motzkin states exhibit critical logarithmic entanglement divergence (log N) with system size N, while their colorful counterparts host supercritical sublinear (√N) entanglement growth. Such unconventional entanglement behaviors place these states well beyond the expressive capability of standard tensor networks, such as matrix product states (MPS), which are fundamentally constrained by the entanglement area law.

The team systematically constructed these exact representations across four mainstream neural network architectures: recurrent, feedforward, convolutional, and transformer networks. The core design leverages a causal prefix-sum module, implementable via recurrent updates, feedforward mappings, or masked attention layers. This module is combined with position-selective rectified linear gates that enforce the Motzkin height constraints. For the colorful states, a dedicated causal stack module was further introduced to explicitly encode the last-in-first-out color-matching rule.

These results demonstrate that neural architectures can accurately capture highly non-trivial entanglement features inaccessible to conventional tensor networks. The work provides prototypic examples for benchmarking and a constructive design framework for future neural-network quantum state developments targeting strongly entangled quantum systems. This opens new avenues for exploring and modeling complex quantum systems beyond the limitations of current computational tools.