Opening the Black Box: A Survey on the Mechanisms of Multi-Step Reasoning in Large Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Pan, Liangming, Liang, Jason, Ye, Jiaran, Yang, Minglai, Lu, Xinyuan, Zhu, Fengbin
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914268428894208
author Pan, Liangming
Liang, Jason
Ye, Jiaran
Yang, Minglai
Lu, Xinyuan
Zhu, Fengbin
author_facet Pan, Liangming
Liang, Jason
Ye, Jiaran
Yang, Minglai
Lu, Xinyuan
Zhu, Fengbin
contents Large Language Models (LLMs) have demonstrated remarkable abilities to solve problems requiring multiple reasoning steps, yet the internal mechanisms enabling such capabilities remain elusive. Unlike existing surveys that primarily focus on engineering methods to enhance performance, this survey provides a comprehensive overview of the mechanisms underlying LLM multi-step reasoning. We organize the survey around a conceptual framework comprising seven interconnected research questions, from how LLMs execute implicit multi-hop reasoning within hidden activations to how verbalized explicit reasoning remodels the internal computation. Finally, we highlight five research directions for future mechanistic studies.
format Preprint
id arxiv_https___arxiv_org_abs_2601_14270
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Opening the Black Box: A Survey on the Mechanisms of Multi-Step Reasoning in Large Language Models
Pan, Liangming
Liang, Jason
Ye, Jiaran
Yang, Minglai
Lu, Xinyuan
Zhu, Fengbin
Computation and Language
Artificial Intelligence
Large Language Models (LLMs) have demonstrated remarkable abilities to solve problems requiring multiple reasoning steps, yet the internal mechanisms enabling such capabilities remain elusive. Unlike existing surveys that primarily focus on engineering methods to enhance performance, this survey provides a comprehensive overview of the mechanisms underlying LLM multi-step reasoning. We organize the survey around a conceptual framework comprising seven interconnected research questions, from how LLMs execute implicit multi-hop reasoning within hidden activations to how verbalized explicit reasoning remodels the internal computation. Finally, we highlight five research directions for future mechanistic studies.
title Opening the Black Box: A Survey on the Mechanisms of Multi-Step Reasoning in Large Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2601.14270