Deductive Beam Search (DBS) is a method that integrates chain-of-thought (CoT) reasoning with deductive reasoning to enhance the reasoning capabilities of Large Language Models (LLMs). The main challenge in CoT reasoning is the accumulation of errors in intermediate steps, which can lead to incorrect final answers. DBS addresses this by using a deductive verifier to assess the logical consistency of each reasoning step, ensuring that each step follows logically from the previous ones. This approach reduces the likelihood of errors propagating through the reasoning process. DBS employs a step-wise beam search strategy, where each reasoning step is evaluated based on its deductive score. This score is determined by a verifier that checks whether the reasoning step is a logical consequence of the premises. The verifier is trained using a scalable and labor-free data construction method, which involves generating synthetic reasoning errors and using them to improve the verifier's ability to detect incorrect steps. Extensive experiments show that DBS significantly improves the performance of LLMs across various reasoning tasks, including arithmetic, commonsense, and symbolic reasoning. The method is effective across different model scales and settings, demonstrating robustness and adaptability. DBS outperforms existing methods in terms of accuracy and efficiency, particularly in detecting subtle reasoning errors and maintaining consistency across multiple reasoning chains. The key contributions of DBS include the integration of deductive reasoning with beam search, the development of a deductive verifier that can detect logical inconsistencies, and the use of a scalable data construction method to train the verifier. These components work together to enhance the reasoning capabilities of LLMs, making them more reliable and accurate in complex reasoning tasks.Deductive Beam Search (DBS) is a method that integrates chain-of-thought (CoT) reasoning with deductive reasoning to enhance the reasoning capabilities of Large Language Models (LLMs). The main challenge in CoT reasoning is the accumulation of errors in intermediate steps, which can lead to incorrect final answers. DBS addresses this by using a deductive verifier to assess the logical consistency of each reasoning step, ensuring that each step follows logically from the previous ones. This approach reduces the likelihood of errors propagating through the reasoning process. DBS employs a step-wise beam search strategy, where each reasoning step is evaluated based on its deductive score. This score is determined by a verifier that checks whether the reasoning step is a logical consequence of the premises. The verifier is trained using a scalable and labor-free data construction method, which involves generating synthetic reasoning errors and using them to improve the verifier's ability to detect incorrect steps. Extensive experiments show that DBS significantly improves the performance of LLMs across various reasoning tasks, including arithmetic, commonsense, and symbolic reasoning. The method is effective across different model scales and settings, demonstrating robustness and adaptability. DBS outperforms existing methods in terms of accuracy and efficiency, particularly in detecting subtle reasoning errors and maintaining consistency across multiple reasoning chains. The key contributions of DBS include the integration of deductive reasoning with beam search, the development of a deductive verifier that can detect logical inconsistencies, and the use of a scalable data construction method to train the verifier. These components work together to enhance the reasoning capabilities of LLMs, making them more reliable and accurate in complex reasoning tasks.