Principle:Diagram of thought Diagram of thought Problem Formulation
| Knowledge Sources | |
|---|---|
| Domains | Reasoning, Prompt_Engineering |
| Last Updated | 2026-02-14 04:30 GMT |
Overview
Problem formulation is the principle of explicitly declaring the root objective of a reasoning process as a distinguished, uniquely identified node so that all subsequent reasoning steps can trace their provenance back to a single, well-defined origin.
Description
In directed-acyclic-graph (DAG) based reasoning, every graph must have a clearly designated starting point from which all other nodes derive. Problem formulation addresses the challenge of anchoring a reasoning session: without a formal root node, downstream propositions, critiques, and summaries lack a shared reference point, making the graph ambiguous and unverifiable.
When a user presents a task, problem formulation transforms that free-form description into the root node of the reasoning DAG. This root node carries a distinguished role that separates it from all other node types (propositions, critiques, summaries). By doing so, it establishes the following invariants:
- Single entry point: The DAG has exactly one node with in-degree zero, ensuring every reasoning path can be traced back to the original problem statement.
- Role separation: The problem node is typed differently from reasoning nodes, preventing confusion between the task description and the steps taken to solve it.
- Dependency foundation: All subsequent nodes must, directly or transitively, reference the problem node through typed edges, guaranteeing that no reasoning step exists in isolation from the stated objective.
Within structured reasoning frameworks, problem formulation acts as the contract between the user and the reasoning system: it defines what must be solved before any reasoning about how to solve it begins.
Usage
Apply this principle at the very start of any reasoning session that employs DAG-based decomposition. The design trigger is the receipt of a new task or problem statement from a user. Before any proposition, critique, or synthesis step is generated, the task description must be wrapped in the designated problem-node structure. This ensures the reasoning graph is well-rooted from its first token of output. Omitting this step would produce a DAG with no identifiable origin, undermining traceability and auditability.
Theoretical Basis
The formal basis for problem formulation rests on the graph-theoretic requirement that a DAG used for structured reasoning must possess a unique source node (a node with in-degree zero from which all other nodes are reachable). In the Diagram of Thought framework, this source node is the problem node, assigned id=1 by convention.
From the perspective of category theory, the problem node functions as an initial object in the category whose objects are DAG nodes and whose morphisms are the typed edges (use, critique, refine). An initial object I in a category C is one for which there exists exactly one morphism from I to every other object in C. In the DoT DAG, this means:
- There is a (possibly indirect) directed path from the problem node to every other node in the graph.
- This path is unique up to the acyclicity constraint enforced by the rule that edge sources must have strictly smaller IDs than edge destinations (
src < dst).
The DoT framework further grounds this in topos theory. The reasoning DAG is interpreted as a diagram in a slice topos, where validated propositions are subobjects and the final summary is a colimit -- a universal construction that glues together all validated evidence. The problem node serves as the base object of this slice: without it, the slice topos has no anchor, and the colimit construction is undefined. In other words, the formal guarantees of principled synthesis (invariance to isomorphic rearrangement of the graph, robustness of the final answer) all depend on the existence of a well-defined root from which the diagram emanates.
The acyclicity constraint (@edge sources must have IDs strictly less than destinations) further ensures that no reasoning step can circularly depend on the problem node's descendants and then feed back into the problem itself, preserving the DAG structure and making deterministic extraction of the reasoning path possible.