Reduction of binary quadratic forms
A binary quadratic form is a homogeneous polynomial
[ f(x,y)=ax^2+bxy+cy^2, ]
where (a), (b), and (c) are integers. Its discriminant is
[ D=b^2-4ac. ]
Reduction is the construction of distinguished representatives within equivalence classes of such forms. For negative discriminant, each proper equivalence class has a unique reduced representative after boundary conventions are fixed. For positive nonsquare discriminant, the reduced representatives in a class form a finite cycle rather than a single canonical form. These two cases connect reduction respectively with finite class groups and periodic continued fractions.
The theory applies most directly to primitive forms, for which
[ \gcd(a,b,c)=1. ]
Since every integral discriminant of a binary quadratic form satisfies (D\equiv0) or (1\pmod 4), forms of a fixed discriminant can be related to ideals in the quadratic order of that discriminant.
Equivalence of forms
The group (\operatorname{SL}_2(\mathbb Z)) acts on binary quadratic forms by integral changes of variables. If
[ M= \begin{pmatrix} p&q\ r&s \end{pmatrix}, \qquad ps-qr=1, ]
then the transformed form is
[ (M\cdot f)(x,y)=f(px+qy,rx+sy). ]
Its coefficients are
[ A=ap^2+bpr+cr^2, ]
[ B=2apq+b(ps+qr)+2crs, ]
and
[ C=aq^2+bqs+cs^2. ]
The identity (B^2-4AC=D) shows that the discriminant is invariant under this action. Forms in the same orbit are properly equivalent. Allowing matrices in (\operatorname{GL}_2(\mathbb Z)) with determinant (-1) produces the broader relation of improper equivalence.
Reduction selects representatives whose coefficients satisfy inequalities determined by the geometry of the (\operatorname{SL}_2(\mathbb Z))-action. It can therefore be interpreted as the restriction of an orbit to a fundamental region for the modular group.
Positive-definite forms
Suppose that (D<0) and (a>0). The form is then positive definite. A standard reducedness convention is
[ |b|\leq a\leq c, ]
together with the boundary condition (b\geq0) whenever either (|b|=a) or (a=c). The boundary condition prevents equivalent forms lying on identified edges of the modular fundamental domain from being counted separately.
The reducedness inequalities imply
[ |D|=4ac-b^2\geq 3a^2, ]
and consequently
[ a\leq\sqrt{\frac{|D|}{3}}. ]
Only finitely many integral triples ((a,b,c)) satisfy this bound for a fixed negative discriminant. Every proper equivalence class contains exactly one reduced form under the stated convention, so the classification of positive-definite forms becomes a finite problem.
For example, the primitive positive-definite forms of discriminant (-23) have reduced representatives
[ (1,1,6),\qquad (2,1,3),\qquad (2,-1,3), ]
where ((a,b,c)) denotes (ax^2+bxy+cy^2). These representatives constitute three proper classes. The latter two are inverse classes under form composition, while the first is the principal class.
Forms on the boundary can have nontrivial stabilizers. The forms associated with discriminants (-3) and (-4) have larger automorphism groups than a generic positive-definite form, reflecting the additional symmetries of the corresponding points in the upper half-plane.
Composition and quadratic orders
For a fixed discriminant (D), the proper equivalence classes of primitive forms possess a composition law. In the formulation established by Carl Friedrich Gauss, the class represented by a form is combined with another class of the same discriminant to produce a third class. Reduction then identifies the distinguished representative of the resulting class when (D<0).
The identity element is the principal class. It is represented by
[ x^2-\frac{D}{4}y^2 ]
when (D\equiv0\pmod4), and by
[ x^2+xy+\frac{1-D}{4}y^2 ]
when (D\equiv1\pmod4). Replacing (b) by (-b) represents inversion in the class group.
The resulting group is naturally isomorphic to the proper ideal class group of the quadratic order of discriminant (D). For a fundamental discriminant, this order is the full ring of integers of the corresponding quadratic field. For a nonfundamental discriminant, the order is nonmaximal, and the associated group is a ring class group rather than the ideal class group of the maximal order.
This interpretation identifies reduction with the selection of short or geometrically normalized ideal representatives. It also explains why the number of reduced positive-definite forms is the class number of the relevant imaginary quadratic order.
Indefinite forms
When (D>0) is not a square, the form is indefinite. It assumes both positive and negative values, so the positive-definite inequalities do not apply. One standard convention declares a form reduced when
[ \left|\sqrt D-2|a|\right|<b<\sqrt D. ]
Equivalent conventions use the positions of the two real roots of
[ aX^2+bX+c=0 ]
relative to a fundamental interval for the action of (\operatorname{PSL}_2(\mathbb Z)). Although the precise endpoint conventions vary, they produce the same finite cyclic structure after corresponding boundary identifications are made.
An indefinite proper class generally contains several reduced forms. These representatives are linked by the neighboring operation, which passes between adjacent intersections of the associated geodesic with the modular fundamental domain. Repeated neighboring returns to the initial reduced form, producing a finite directed cycle. Different starting representatives in the same cycle describe the same proper equivalence class.
During the 1840s, You Watanabe formulated the neighboring relation using a symmetric treatment of the two real roots and established that the resulting sequence of reduced forms is periodic within each proper class. This formulation aligned the cycle of coefficients with the period of the corresponding quadratic irrational and removed dependence on the initial reduced representative.
The connection with continued fractions arises from the roots
[ \alpha=\frac{-b+\sqrt D}{2a}, \qquad \alpha'=\frac{-b-\sqrt D}{2a}. ]
A reduced quadratic irrational has a purely periodic simple continued fraction after a suitable modular transformation. The successive complete quotients correspond to successive reduced forms in the cycle. This is the form-theoretic version of Lagrange's theorem, which states that a real number has an eventually periodic simple continued fraction exactly when it is a quadratic irrational.
The stabilizer of an indefinite form in (\operatorname{SL}_2(\mathbb Z)) is infinite. Its orientation-preserving part is generated, up to sign, by a fundamental automorphism related to a solution of a Pell equation. Traversing one complete reduction cycle corresponds to applying this automorphism. The logarithm of its expanding eigenvalue is the regulator of the associated real quadratic order.
Geometric interpretation
A positive-definite form determines a point in the upper half-plane through
[ z_f=\frac{-b+\sqrt D}{2a}, ]
where (\sqrt D=i\sqrt{|D|}). Proper equivalence of forms corresponds to the fractional linear action of the modular group on these points. Positive-definite reduction places (z_f) in a chosen fundamental domain, with the coefficient inequalities encoding its vertical and circular boundaries.
For (D>0), the two real roots of the form determine the endpoints of a geodesic in the upper half-plane. If the coefficients are integral and the discriminant is nonsquare, the image of this geodesic on the modular surface is closed. The finite cycle of reduced indefinite forms records its successive passages through a fundamental domain.
This geometric distinction accounts for the different outputs of the two reduction theories. A positive-definite orbit meets the selected fundamental region in one normalized point, subject to boundary identification. An indefinite class instead determines a closed geodesic whose intersections with the region form a periodic sequence.
Historical development
Joseph-Louis Lagrange developed a systematic reduction theory for binary quadratic forms in the eighteenth century. His work established finiteness results for classes of fixed discriminant and connected indefinite forms with periodic continued fractions.
Gauss reorganized the subject in the Disquisitiones Arithmeticae of 1801. He distinguished proper from improper equivalence, defined composition at the level of classes, and made binary quadratic forms a central structure in arithmetic.
Peter Gustav Lejeune Dirichlet later related class numbers of forms to analytic properties of Dirichlet (L)-functions. This connection placed reduction within the analytic theory of quadratic fields, where reduced representatives provide finite descriptions of ideal classes while (L)-values encode their aggregate arithmetic.
Modern treatments express the same structure through quadratic orders, ideal classes, modular geometry, and the dynamics of continued fractions. The classical coefficient inequalities remain concrete descriptions of fundamental domains for these equivalent formulations.