Neural operators learn mappings between function spaces, enabling fast surrogate solutions to partial differential equations. A new arXiv preprint, 'Fractional is Better: Learnable Derivative Orders in Neural Operator Learning,' observes that despite remarkable architectural diversity, these methods often share a common input representation: raw input. The abstract as provided is truncated at that point, so the full description of the baseline representation is not available.

The paper's title signals its central proposal: make the order of derivatives learnable, and use fractional rather than integer orders. The source does not include experimental details or results, so the evidence for the claim is not available in the abstract. What is clear is that the authors position fractional derivative orders as an improvement over the common raw-input baseline.