A place (a closed point of the base, a prime of a number ring, or a special parameter) in an arithmetic or geometric family where the special fiber fails to be smooth or acquires extra singular components; informally, the family has 'bad' behavior at that place compared with 'good reduction' where fibers are smooth and well behaved.

Principle

Reduction quality is detected by invariants (discriminant, minimal models, monodromy, Kodaira types): vanishing of discriminants or appearance of singularities signals bad reduction. In arithmetic contexts bad reduction often reflects wild ramification or nontrivial monodromy in étale cohomology.

Demonstration

Classical example: an elliptic curve given by a Weierstrass model has bad reduction at primes where the discriminant vanishes; the special fiber can be nodal (multiplicative reduction) or cuspidal (additive reduction). More generally families of curves or abelian varieties develop singular fibers at bad places.

Misapplication

Lumping all non-smooth fibers as equally 'bad' without distinguishing semistable, potentially good, or wild reduction; assuming bad reduction at one model persists under allowable base change (it may become good after extension). Also misusing reduction status to infer arithmetic properties without checking potential changes under extensions.

Consequence

Bad reduction affects arithmetic and geometric invariants: conductors increase, Galois representations gain nontrivial inertia action, and moduli classifications must record reduction types. Practically, it necessitates base changes, minimal model programs, or semistable reduction theorems to control and possibly improve the situation.

Reversal

Good reduction: the special fiber is smooth (or has prescribed mild behaviour like semistability) and arithmetic invariants behave tamely; Galois action on étale cohomology is unramified. Good reduction yields simpler deformation theory and often easier moduli descriptions.

Boundary

The notion depends on context: one must distinguish good, semistable, potentially good, and wild reduction; consider behavior under base change and extension of scalars. 'Bad' is a relative term: what is bad for one problem (wild ramification) may be acceptable in another. The definition typically applies to proper families over discrete valuation rings or arithmetic bases.

Semantic Tension

Tension between coarse usage ('bad' vs 'good') and refined taxonomy (multiplicative, additive, semistable, potentially good). The practical tension is between a binary label useful for heuristics and the nuanced classification needed for precise arithmetic or geometric conclusions.

Synthesis

Bad Reduction is the occurrence that a family acquires singular or otherwise pathological special fibers at certain places, detected by vanishing discriminants, monodromy, or singularity types. It forces refined analysis—distinguishing semistable, additive, and potentially good cases—and often requires base change or minimal model techniques to manage or rectify the degeneration.