Analogical reasoning involves mapping structural relations from a well-understood source domain onto a novel target domain. Gentner's structure-mapping theory specifies that productive analogies preserve relational structure rather than surface features — 'the atom is like the solar system' is compelling because both share the pattern of a central body with orbiting satellites. Analogical transfer in problem solving requires noticing structural similarity between a previously solved problem and a new one, a step that often fails without explicit prompting even when the analogy is apt.
Present the Gick and Holyoak radiation problem: without a hint, few subjects use an analogous prior military problem; with a hint, transfer improves dramatically. This shows that structural mapping requires active access to the source, not mere prior exposure.
From your study of problem-solving strategies, you know that effective solvers represent problems in terms of their deep structure—what is known, what is unknown, what operations are applicable. From schema theory, you know that schemas abstract recurring relational patterns away from surface details. Analogical reasoning is what happens when those abstractions cross domain boundaries: you notice that the relational structure of a well-understood source domain maps onto an unfamiliar target domain, and you exploit that mapping to generate insight, make predictions, or find solutions.
Gentner's structure-mapping theory gives a precise account of what makes an analogy productive. A surface analogy notes attribute similarities between individual objects: the sun is yellow, gold is yellow. A structural analogy preserves relational patterns: "the atom is like the solar system" is compelling because in both cases a large central body exerts an attractive force causing smaller bodies to orbit it. The relational pattern—*central body, attractive force, orbital path*—transfers intact; the objects themselves (electrons vs. planets, electrostatics vs. gravity) are otherwise radically different. Structure-mapping predicts which analogies will be judged apt, will facilitate learning, and will generate correct inferences about the target domain. Analogies that rest only on surface similarity without structural correspondence tend to mislead.
The classic experimental demonstration is the Gick and Holyoak radiation problem. Subjects are presented Duncker's problem: a doctor must destroy a tumor using radiation, but any dose powerful enough to destroy the tumor will also kill healthy tissue en route. The solution—use multiple low-intensity beams converging on the tumor from different angles, each individually safe—is difficult to discover spontaneously. Subjects who had earlier read an analogous military story (a general captures a fortress by splitting his army into small groups approaching from multiple directions) solved the radiation problem at much higher rates—but only when the experimenter explicitly told them the two stories were related. Without the retrieval hint, subjects failed to notice the structural correspondence even though they had just read the analogous story. This is the core empirical finding: analogical transfer requires noticing structural similarity, not merely having encountered the source analog. Prior exposure is necessary but not sufficient.
The practical implication is that expertise partly consists in rebuilding problem categorization around deep structure rather than surface features. Novices index problems by surface features—a physics problem with an inclined plane looks like other inclined-plane problems; a word problem about trains looks like other train problems. Experts index by the underlying structure—what forces are present, what quantities are conserved, what type of constraint is operative. This is why expert physicists sort mechanics problems by principle (conservation of energy, Newton's second law) while novices sort by surface appearance (ramp problems, pulley problems). Instruction that explicitly teaches students to identify structural roles—rather than pattern-matching on surface features—builds the analogical access that enables spontaneous transfer to new problems.