Can a child node also be a parent?

Yes. Any nonleaf node below the root can be both a child of one node and a parent of others.

Must every child represent a different world state?

No. Search-tree nodes reached through different paths may represent the same state.

Can a child have several parents?

In an ordinary rooted tree, no. In a general directed graph using parent-child terminology, yes, depending on the representation.

AI Glossary / Graphs and symbolic AI / Intermediate

Child node

Definition

A child node is a node directly below another node, its parent, in a rooted tree. In a search tree, a child is commonly generated by applying one action to the state represented by its parent. The relationship is immediate: a grandchild is a descendant, but it is not a direct child of the original node.

What it means

Tree terminology describes structure rather than importance or age. The root has no parent, while other nodes in an ordinary rooted tree have one parent. A node can have no children, one child, or several children depending on the tree. A node with no children is called a leaf.
AI uses these relationships in search trees, decision trees, parse trees, and other representations. In a directed graphical model, child can also mean a node at the destination of a directed edge, and such a node may have several parents. The representation must therefore be identified before assuming tree-specific properties.

How it works

In a search algorithm, expanding a node creates child nodes for permitted successor actions. Each child can store the resulting state, the action, accumulated path cost, depth, and a link to its parent. These records allow a solution path to be reconstructed after a goal is reached.
A child node is not always a previously unseen state. Two different action sequences can reach the same state and therefore create different search-tree nodes representing it. Graph-search methods often detect such repetitions. In a learned decision tree, by contrast, child nodes represent the results of a split test rather than successive physical actions.

Three practical examples

A route search creates children corresponding to the locations reachable by one allowed move from the current state.
A classification tree sends a case to one child or the other according to whether a feature satisfies the node's split condition.
A syntactic parse tree places phrase components beneath a parent phrase node, expressing their immediate structural relationship.

Think of it this way

Think of a folder containing immediate subfolders. Those subfolders are its children; folders inside them are deeper descendants. The analogy is limited because a search-tree child is usually a generated decision possibility, not a permanent storage container, and general directed graphs can permit relationships that ordinary folder trees do not.

Worked example

Consider a search tree rooted at R. R has children A and B. A has children C and D, while B has child E. The immediate children of R are exactly A and B: C, D, and E are descendants at depth 2, assuming the root has depth 0.
Suppose the edge costs are R-A = 3, R-B = 2, A-C = 4, A-D = 1, and B-E = 6. D's parent is A, its depth is 2, and its accumulated path cost from R is 3 + 1 = 4. These are different quantities: being two levels deep does not imply cost 2. If D and E happen to represent the same underlying world state, they remain different nodes in this search tree because they were reached along different paths, with costs 4 and 8 respectively. A graph-search algorithm may use that information to avoid unnecessary repeated exploration.

Illustrative case study

At fictional Cedar Search, a team displays an AI planner's explored tree. Users initially assume that two nodes with the same state label indicate an export error. The team adds parent links and path costs, showing that the state was reached through two different sequences.
The display also distinguishes immediate children from all descendants so users can interpret expansion counts correctly. This fictional case illustrates a representation issue that affects explanations of search behavior. It does not claim a performance improvement or prescribe one duplicate-handling policy for every search algorithm.

Common misconceptions and mistakes

Child and descendant are not interchangeable. Child means one structural step away, while descendant can be many steps away. Likewise, sibling nodes share a parent; they need not share the same state or outcome.
Depth is not accumulated cost unless every edge has the same unit cost. A leaf in a partially explored search tree may simply be unexpanded, so it is not automatically a terminal state or goal. Finally, a child relationship in a Bayesian network does not by itself establish a causal claim; directed edges have the semantics assigned by that model.

Business use

For businesses, this terminology helps make AI decision diagrams and planning traces understandable. A useful interface identifies what nodes represent, what edges mean, and whether the diagram is a tree or a more general graph. In troubleshooting, parent links can show which earlier decision produced a candidate. In governance or review, avoid describing an unexpanded child as a rejected option unless the algorithm actually evaluated and rejected it.

Check your understanding

R connects to A, and A connects to C in a rooted tree. Is C a child of R?

Answer: No. C is a child of A and a descendant of R, two edges below it.

Related terms

Parent node
Root
Search tree
leaf node
descendant
sibling
successor function

Learn next

Draw the six-node worked tree and label each node's parent, depth, and path cost. Then merge nodes that represent the same state into a graph. Comparing the two diagrams clarifies which relationships belong to the search history and which belong to the underlying problem.

Sources

Sources: NIST Dictionary of Algorithms and Data Structures, child: https://xlinux.nist.gov/dads/HTML/child.html . Russell and Norvig, original AIMA search-tree materials: https://aima.eecs.berkeley.edu/slides-pdf/chapter03-6pp.pdf .

Previous: Chi-squared distribution Next: Chinese wall