Voltammetric Translation with Transformers: Converting Cyclic Voltammograms to Differential Pulse Voltammograms.

Here we report a voltammogram translation framework that converts cyclic voltammograms into differential pulse voltammograms, enabling protocol interoperability without requiring paired experimental corpora. We formulate CV → DPV conversion as a sequence-to-sequence regression problem and implement a query-based transformer that encodes the full CV cycle with explicit scan-direction context and decodes DPV on a fixed target potential grid via cross-attention. To overcome the scarcity of paired measurements, we generate a large synthetic paired CV-DPV data set using physics-informed simulations spanning six representative mechanisms (E, EC, ECatal, EE, ECE, and ECEC) and broad parameter variations. Across deep-learning and classical regression baselines, the Transformer achieves the lowest prediction error and markedly outperforms slope-based heuristics. Cross-attention analysis indicates a two-mode strategy: baseline anchoring away from faradaic features and protocol-consistent local retrieval in peak-forming regions, including an offset comparable to the DPV pulse amplitude. Importantly, the experimental validation directly probes mechanism coverage learned from simulation: the translator accurately reproduces DPV for