Discriminant
In mathematics, the discriminant of a polynomial is a polynomial in its coefficients that vanishes precisely when the polynomial has a repeated root, thereby serving as a key invariant for analyzing the roots' nature, multiplicity, and reality without explicitly solving the equation.[1] For a quadratic equation of the form ax^2 + bx + c = 0 with a \neq 0, the discriminant is given by D = b^2 - 4ac, which determines the number and type of roots: if D > 0, there are two distinct real roots; if D = 0, there is exactly one real root (a repeated root); and if D < 0, there are two complex conjugate roots.[2] This concept extends to higher-degree polynomials, where the discriminant is defined as \Delta(f) = a^{2n-2} \prod_{i < j} (r_i - r_j)^2 for a polynomial f(x) = a x^n + \cdots of degree n with roots r_i, or more generally as (up to a constant factor) the resultant of f and its derivative f', providing insights into multiple roots and the polynomial's factorization over the reals or complexes.[1] The term "discriminant" was coined by James Joseph Sylvester in 1851 to describe criteria for the number and reality of roots in cubic and higher equations, building on earlier 18th-century work by Joseph-Louis Lagrange on differences of roots and Carl Friedrich Gauss's use of determinants for similar purposes in quadratic forms.[3] Beyond algebra, the discriminant plays a crucial role in algebraic number theory as an invariant of a number field K = \mathbb{Q}(\theta), defined via the determinant of traces of an integral basis and quantifying ramification of primes in the ring of integers, with its absolute value bounding the field's arithmetic complexity.[1] In multivariate settings, such as homogeneous polynomials, the discriminant detects critical points where the gradient vanishes, linking to geometric properties like singularities in algebraic varieties.[4]Definition and Origins
Historical Development
The concept of the discriminant in algebra traces its roots to the late 18th century, when mathematicians began systematically studying expressions that distinguish the nature of polynomial roots. Although the explicit formula for the discriminant of quadratic equations had appeared in earlier algebraic works, Carl Friedrich Gauss formalized its role in 1801 within quadratic forms in his seminal Disquisitiones Arithmeticae, where he used it to classify binary quadratic forms based on their properties. Gauss further extended the definition to general polynomials in his 1815 paper Demonstratio nova theorematis omnem functionem algebraicam resolvibilem in factores reales esse, linking it directly to the separation of roots.[1] In the mid-18th century, Leonhard Euler laid foundational groundwork for higher-degree discriminants through his advancements in elimination theory and the theory of resultants, which are intrinsically connected to discriminants as the resultant of a polynomial and its derivative. Euler's investigations into cubic and quartic equations, detailed in works such as his contributions to the solution methods for these polynomials around 1760–1770, emphasized criteria for multiple roots and root distinctions without yet employing the term "discriminant." These efforts built on earlier algebraic traditions and anticipated the invariant properties later explored.[5][6] The term "discriminant" itself was coined in 1851 by James Joseph Sylvester in his paper "On a Remarkable Discovery in the Theory of Canonical Forms and of Hyperdeterminants," derived from the Latin discriminans (meaning "distinguishing"), reflecting its function in separating cases of distinct versus multiple roots. Sylvester's introduction applied it initially to cubic equations but extended the nomenclature to quadratics and higher degrees, marking a milestone in the standardization of the concept.[7][1][8] During the 19th century, Arthur Cayley and Sylvester advanced the discriminant's theoretical framework within invariant theory, viewing it as a fundamental polynomial invariant under linear transformations. Their collaborative work, including Cayley's explorations of covariants and Sylvester's relations between invariants and discriminants for forms like ternary cubics, integrated the discriminant into broader algebraic structures, influencing subsequent developments in algebraic geometry and number theory. Key publications include Sylvester's 1851 paper and Cayley's contemporaneous treatises on invariants around 1854–1860.[9]General Definition
The discriminant of a polynomial p(x) = a_n \prod_{i=1}^n (x - r_i) of degree n, where a_n is the leading coefficient and r_i are the roots (counted with multiplicity), is defined as \Delta(p) = a_n^{2n-2} \prod_{1 \leq i < j \leq n} (r_i - r_j)^2. This expression quantifies the "discrimination" between the roots by capturing the squared differences between all pairs of distinct roots, scaled by a power of the leading coefficient to ensure consistency under polynomial scaling.[10] The discriminant vanishes if and only if the polynomial has at least one multiple root, thereby serving as an invariant that detects the presence of repeated roots without explicitly solving for them.[10] For monic polynomials, where the leading coefficient a_n = 1, the formula simplifies to \Delta(p) = \prod_{1 \leq i < j \leq n} (r_i - r_j)^2, directly reflecting the product of squared root differences.[10] This definition extends naturally to polynomials over any commutative ring, where the discriminant is alternatively expressed using the resultant of the polynomial and its derivative as \Delta(p) = (-1)^{n(n-1)/2} \operatorname{Res}(p, p') / a_n, ensuring the concept remains well-defined without relying on the existence of roots in the ring.[10]Expression in Terms of Roots
The discriminant of a polynomial p(x) = a_n \prod_{i=1}^n (x - r_i), where a_n is the leading coefficient and r_1, \dots, r_n are the roots (counted with multiplicity), is explicitly given by \Delta(p) = a_n^{2n-2} \prod_{1 \leq i < j \leq n} (r_i - r_j)^2. This formula expresses the discriminant as a symmetric function of the roots, scaled by the leading coefficient to the power $2n-2, reflecting its homogeneity under scaling of the polynomial coefficients.[11] This root-based expression can be derived from the Vandermonde determinant, which arises naturally when expressing the roots in terms of power sums or symmetric functions. Consider the Vandermonde matrix V with entries V_{k,i} = r_i^{k-1} for k,i = 1, \dots, n. Its determinant is \det V = \prod_{1 \leq i < j \leq n} (r_j - r_i) = (-1)^{n(n-1)/2} \prod_{1 \leq i < j \leq n} (r_i - r_j). Squaring this determinant yields (\det V)^2 = \prod_{1 \leq i < j \leq n} (r_i - r_j)^2, and for the monic case (a_n = 1), the discriminant is precisely this square; in general, the factor a_n^{2n-2} accounts for the leading coefficient in the polynomial's factorization.[12][11] An equivalent derivation links the discriminant to the resultant of p and its derivative p'. The resultant \Res(p, p') vanishes if and only if p and p' share a common root, i.e., if p has a multiple root. Specifically, \Delta(p) = (-1)^{n(n-1)/2} a_n^{-1} \Res(p, p'), where the sign alternates based on the degree, and the inverse leading coefficient normalizes the expression. This relation follows from evaluating p'(r_k) = a_n \prod_{j \neq k} (r_k - r_j) and incorporating the product over all roots into the resultant definition via the Sylvester matrix.[11] The sign convention in the root product formula incorporates (-1)^{n(n-1)/2} to align with the resultant expression, ensuring consistency across definitions; however, since the discriminant is squared, it is always non-negative for real coefficients with distinct real roots, and the absolute value |\Delta(p)| often interprets the scale of root separation without sign concerns.[11] For illustration, consider a generic cubic polynomial p(x) = a_3 (x - r)(x - s)(x - t) with distinct roots r, s, t. The discriminant is \Delta(p) = a_3^{4} (r - s)^2 (r - t)^2 (s - t)^2, highlighting the pairwise squared differences scaled by the leading coefficient to the fourth power, which quantifies the "spread" of the roots.[11]Discriminants for Low-Degree Polynomials
Quadratic Case
The discriminant of a quadratic equation ax^2 + bx + c = 0, where a \neq 0, is defined as \Delta = b^2 - 4ac.[13] This expression arises directly from the quadratic formula for the roots, x = \frac{-b \pm \sqrt{b^2 - 4ac}}{2a}, where the discriminant determines the existence and nature of real solutions.[13] In the 16th century, Gerolamo Cardano incorporated the solution of quadratic equations into his influential treatise Ars Magna (1545), reviewing earlier geometric methods and presenting the algebraic approach that implicitly relies on the discriminant to resolve the roots.[5] Cardano's compilation built on works by Al-Khwarizmi and others, standardizing the quadratic formula in Europe and highlighting the role of the square root term in distinguishing solvable cases.[5] Geometrically, the quadratic equation y = ax^2 + bx + c represents a parabola, which is a conic section formed by intersecting a plane with a cone; the discriminant \Delta then indicates the number of real intersection points between this parabola and the x-axis (the line y = 0).[14] Specifically, if \Delta > 0, the parabola crosses the x-axis at two distinct points, corresponding to two real roots.[14] For example, consider the equation x^2 - 3x + 2 = 0. Here, a = 1, b = -3, c = 2, so \Delta = (-3)^2 - 4(1)(2) = 9 - 8 = 1 > 0, implying two distinct real roots.[13] Factoring gives (x - 1)(x - 2) = 0, with roots x = 1 and x = 2.[13]Cubic Case
For a cubic polynomial ax^3 + bx^2 + cx + d = 0 with a \neq 0, the discriminant \Delta is given by \Delta = 18abcd - 4b^3 d + b^2 c^2 - 4a c^3 - 27 a^2 d^2. [10] This expression arises as the resultant of the polynomial and its derivative, scaled appropriately, and vanishes precisely when the polynomial has a repeated root.[10] To simplify computations, one often reduces the general cubic to the depressed form x^3 + p x + q = 0 via the substitution x = y - \frac{b}{3a}, which eliminates the quadratic term. For this form, the discriminant simplifies to \Delta = -4 p^3 - 27 q^2. [10] This version is particularly useful in Cardano's method for solving cubics, where the sign of \Delta determines the nature of the roots. For a cubic polynomial with real coefficients, the sign of the discriminant provides key information about the roots: if \Delta > 0, there are three distinct real roots; if \Delta = 0, there is at least one multiple root (and all roots real); if \Delta < 0, there is one real root and two complex conjugate roots.[10] As an illustrative example, consider the polynomial x^3 - 3x + 2 = 0. Here, a = 1, b = 0, c = -3, d = 2, so \Delta = 18(1)(0)(-3)(2) - 4(0)^3(2) + (0)^2(-3)^2 - 4(1)(-3)^3 - 27(1)^2(2)^2 = 108 - 108 = 0. [10] The zero discriminant confirms a multiple root, and indeed the roots are $1 (with multiplicity 2) and -2, all real.Quartic Case
The discriminant \Delta of the general quartic polynomial ax^4 + bx^3 + cx^2 + dx + e = 0 with a \neq 0 is given by \Delta = 256a^3 e^3 - 192 a^2 b d e^2 - 128 a^2 c^2 e^2 + 144 a^2 c d^2 e - 27 a^2 d^4 + 144 a b^2 c e^2 - 6 a b^2 d^2 e - 80 a b c^2 d e + 18 a b c d^3 + 16 a c^4 e - 4 a c^3 d^2 - 27 b^4 e^2 + 18 b^3 c d e - 4 b^3 d^3 - 4 b^2 c^3 e + b^2 c^2 d^2. [10] In Ferrari's method for solving the quartic, this discriminant coincides with that of the associated resolvent cubic up to a positive constant factor, linking the root structures of both polynomials.[15] Specifically, \Delta > 0 implies the resolvent cubic has three distinct real roots, while \Delta < 0 implies it has one real root and two complex conjugate roots; \Delta = 0 indicates a multiple root in the resolvent, corresponding to multiple roots in the quartic.[16] For a quartic with real coefficients, the sign of \Delta determines possible root configurations as follows: if \Delta > 0, there are either four distinct real roots or no real roots (two complex conjugate pairs); if \Delta < 0, there are exactly two distinct real roots and one complex conjugate pair; if \Delta = 0, there is at least one multiple root.[17] Distinguishing four real roots from zero real roots when \Delta > 0 requires additional checks, such as verifying that the quadratic factors derived from a suitable real root of the resolvent cubic have positive discriminants, ensuring all roots are real.[15] As an illustrative example, consider the biquadratic equation x^4 + 5x^2 - 4 = 0, where a=1, b=0, c=5, d=0, e=-4. Substituting these coefficients into the discriminant formula yields \Delta = -107584 < 0, indicating two real roots and two complex roots.[10] To verify, substitute z = x^2 to obtain the quadratic z^2 + 5z - 4 = 0, with roots z = \frac{-5 \pm \sqrt{41}}{2} \approx 0.702, -5.702; the positive root gives two real x = \pm \sqrt{0.702} \approx \pm 0.838, while the negative root yields no real x.[17]Fundamental Properties
Zero Discriminant and Multiple Roots
A fundamental property of the discriminant of a polynomial p(x) = a_n x^n + \cdots + a_0 of degree n \geq 1 over a field is that \Delta(p) = 0 if and only if p has a multiple root in some algebraic closure of the field, meaning p shares a common root with its formal derivative p'(x).[18] This equivalence follows from the definition of the discriminant as \Delta(p) = a_n^{2n-2} \prod_{1 \leq i < j \leq n} (r_i - r_j)^2, where r_1, \dots, r_n are the roots of p (counted with multiplicity); the product vanishes precisely when at least two roots coincide.[18] Equivalently, via the resultant, the discriminant satisfies \Delta(p) = (-1)^{n(n-1)/2} a_n^{-1} \operatorname{Res}(p, p'), where \operatorname{Res}(p, p') is the resultant of p and p'.[18] The resultant \operatorname{Res}(p, p') = 0 if and only if p and p' have a common root, which occurs exactly when p has a multiple root, as a simple root r satisfies p'(r) \neq 0 by the product rule applied to the factorization of p.[18] Thus, \Delta(p) = 0 precisely under this condition.[18] Over fields of characteristic zero (or more generally, where the characteristic does not divide n), a zero discriminant implies that p is not square-free, as \gcd(p, p') \neq 1, so p factors non-trivially into polynomials of positive degree and is therefore reducible if n \geq 2.[18] In positive characteristic, additional care is needed, but the presence of multiple roots still indicates inseparability and reducibility in the sense of sharing factors with the derivative.[19] For example, consider p(x) = (x-1)^2 (x-2) = x^3 - 4x^2 + 5x - 2, which has a multiple root at x=1. The discriminant \Delta(p) = 0, confirming the multiple root.[18]Invariances Under Transformations
The discriminant of a polynomial exhibits invariance under certain transformations of the variable, preserving its value or scaling it predictably. In particular, it is invariant under translations of the variable. If q(x) = p(x + h) for a constant h, then \Delta(q) = \Delta(p). This property arises from the root formulation of the discriminant, \Delta(p) = a_n^{2n-2} \prod_{1 \leq i < j \leq n} (r_i - r_j)^2, where a_n is the leading coefficient and r_1, \dots, r_n are the roots of p; translating the variable shifts all roots by -h, leaving the differences r_i - r_j unchanged and thus the product unaltered.[10] To illustrate, consider a quadratic polynomial p(x) = x^2 - (r + s)x + rs with roots r and s, where \Delta(p) = (r - s)^2. The translated polynomial q(x) = p(x + h) = (x + h - r)(x + h - s) has roots r - h and s - h, so \Delta(q) = ((r - h) - (s - h))^2 = (r - s)^2 = \Delta(p). This confirms the invariance, as the root separation is preserved despite the shift.[10] Under more general linear transformations, the discriminant scales in a specific manner. For q(x) = a \, p(bx + c) with constants a \neq 0, b \neq 0, the roots of q are s_k = (r_k - c)/b, leading to differences s_i - s_j = (r_i - r_j)/b. The leading coefficient of q becomes a b^n a_n, yielding \Delta(q) = a^{2n-2} b^{n(n-1)} \Delta(p). This scaling reflects the combined effects of multiplying the polynomial by a (which raises the leading coefficient power) and the affine change in the variable (which compresses root differences by $1/b). The core qualitative invariance up to this explicit factor underscores the discriminant's role as a robust measure of root configuration. The discriminant also respects ring homomorphisms between integral domains. If \phi: R \to S is a ring homomorphism and p \in R, then \Delta(\phi(p)) = \phi(\Delta(p)). This holds because \Delta(p) is itself a polynomial in the coefficients of p with integer coefficients, so applying \phi to the coefficients of p and then computing the discriminant is equivalent to applying \phi to the value \Delta(p). This preservation ensures the discriminant behaves consistently under base ring extensions or specializations, facilitating its use in algebraic contexts like number fields.Homogeneity and Scaling
The discriminant \Delta of a polynomial p(x) of degree n is a homogeneous polynomial of degree $2n-2 in the coefficients of p(x).[20] This homogeneity manifests in the scaling behavior: if the coefficients of p(x) are multiplied by a scalar \lambda, yielding the polynomial \lambda p(x), then \Delta(\lambda p) = \lambda^{2n-2} \Delta(p). To see this from the root expression, recall that \Delta(p) = a_n^{2n-2} \prod_{1 \leq i < j \leq n} (r_i - r_j)^2, where a_n is the leading coefficient of p(x) and r_1, \dots, r_n are its roots (counted with multiplicity). For \lambda p(x), the leading coefficient scales to \lambda a_n while the roots remain unchanged, so the product of squared differences is invariant, and the overall scaling factor arises solely from (\lambda a_n)^{2n-2} = \lambda^{2n-2} a_n^{2n-2}. The homogeneous nature of \Delta ensures that its vanishing locus in the coefficient space defines a projective hypersurface, which parameterizes polynomials with multiple roots within the framework of projective varieties.[21]Discriminant of Polynomial Products
The discriminant exhibits a multiplicative property when a polynomial is expressed as the product of two coprime polynomials. Specifically, if p(x) = f(x) g(x), where f and g are coprime polynomials over a field with \deg f = m and \deg g = k, then the discriminant satisfies \Delta(p) = \Delta(f) \, \Delta(g) \, \operatorname{Res}(f, g)^2. This relation holds regardless of the leading coefficients of f and g, as the contributions from these coefficients cancel out in the derivation.[18] To outline the proof, assume a splitting field where f(x) = a \prod_{i=1}^m (x - \alpha_i) and g(x) = b \prod_{j=1}^k (x - \beta_j), with a = \operatorname{lc}(f) and b = \operatorname{lc}(g). The discriminant \Delta(f) = a^{2m-2} \prod_{1 \leq i < l \leq m} (\alpha_i - \alpha_l)^2 and similarly for \Delta(g). For p(x) = ab \prod_{i=1}^m \prod_{j=1}^k (x - \alpha_i)(x - \beta_j), the discriminant \Delta(p) = (ab)^{2(m+k)-2} \prod_{\text{all distinct root pairs}} (r_s - r_t)^2. The product over all root differences factors into the intra-f pairs, intra-g pairs, and cross terms \prod_{i=1}^m \prod_{j=1}^k (\alpha_i - \beta_j)^2. The cross term equals \left[ \operatorname{Res}(f, g) / (a^k b^m) \right]^2, and substituting the expressions for the intra terms yields \Delta(p) = \Delta(f) \, \Delta(g) \, \operatorname{Res}(f, g)^2 after simplification, with the coprimality ensuring no shared roots that would introduce additional multiplicity factors.[18] In the special case where the polynomial factors into distinct irreducible factors p(x) = f_1(x) \cdots f_r(x) over the base field, with each f_i irreducible and pairwise coprime, the discriminant is the product of the individual discriminants adjusted by pairwise resultants: \Delta(p) = \left( \prod_{i=1}^r \Delta(f_i) \right) \left( \prod_{1 \leq i < j \leq r} \operatorname{Res}(f_i, f_j)^2 \right). This follows by iterative application of the two-factor formula, leveraging the coprimality to avoid zero resultants from shared factors.[18] For an illustrative example, consider p(x) = (x^2 - 1)(x - 2), where f(x) = x^2 - 1 (degree 2, roots \pm 1, \Delta(f) = 4) and g(x) = x - 2 (degree 1, \Delta(g) = 1). The factors are coprime, and \operatorname{Res}(f, g) = f(2) = 3. Thus, \Delta(p) = 4 \cdot 1 \cdot 3^2 = 36. Direct computation for the monic cubic p(x) = x^3 - 2x^2 - x + 2 confirms this value using the cubic discriminant formula.[18]Applications to Roots and Fields
Real and Complex Roots
For the quadratic equation ax^2 + bx + c = 0 with real coefficients a \neq 0, b, and c, the discriminant \Delta = b^2 - 4ac determines the nature of the roots: if \Delta > 0, there are two distinct real roots; if \Delta = 0, there is one real root of multiplicity two; and if \Delta < 0, the two roots are complex conjugates.[22] This classification arises because the roots are given by x = \frac{-b \pm \sqrt{\Delta}}{2a}, where a negative \Delta yields imaginary values under the real square root.[22] For polynomials of higher degree with real coefficients, the discriminant provides analogous information about the distribution of real and complex roots, though the interpretation becomes more nuanced. The discriminant is zero if and only if the polynomial has at least one multiple root over the complex numbers.[23] For real coefficients, multiple roots that are non-real occur in conjugate pairs with the same multiplicity, but a real multiple root reduces the count of distinct real roots. Sturm's theorem connects to this by providing a method to count the number of distinct real roots via sign variations in the Sturm sequence (generated from the polynomial and its derivatives, akin to the Euclidean algorithm), allowing verification of whether a zero discriminant corresponds to a real multiple root when the observed number of distinct real roots is less than expected.[24] For polynomials with real coefficients and no multiple roots, the sign of the discriminant indicates the configuration of non-real roots: it is positive if and only if the number of non-real roots is a multiple of 4, and negative if the number of non-real roots is congruent to 2 modulo 4.[25] This property arises because non-real roots occur in conjugate pairs, and the sign of the discriminant is given by (-1)^s, where s is the number of such pairs.[26] In the cubic case, the sign of the discriminant further specifies the root configuration. For the general cubic ax^3 + bx^2 + cx + d = 0 with real coefficients, the discriminant \Delta = 18abcd - 4b^3 d + b^2 c^2 - 4a c^3 - 27 a^2 d^2; if \Delta > 0, there are three distinct real roots, while \Delta < 0 indicates one real root and two complex conjugate roots (with \Delta = 0 signaling a multiple root).[27] For the depressed cubic x^3 + p x + q = 0, this simplifies to \Delta = - (4 p^3 + 27 q^2 ). Consider the example x^3 + x + 1 = 0, where p = 1, q = 1, so \Delta = - (4 \cdot 1^3 + 27 \cdot 1^2) = -31 < 0. The graph of this cubic decreases overall with a local maximum and minimum, crossing the x-axis only once (approximately at x \approx -0.68), confirming the single real root and two complex ones.[28]Discriminant in Algebraic Number Fields
In algebraic number theory, the discriminant of a number field K of degree n over \mathbb{Q} is the integer \Delta_K = \det\left( \operatorname{Tr}_{K/\mathbb{Q}}(e_i e_j) \right), where \{e_1, \dots, e_n\} is a \mathbb{Z}-basis for the ring of integers \mathcal{O}_K of K.[29] This quantity is independent of the choice of basis and serves as a fundamental invariant of the field.[29] For a number field K = \mathbb{Q}(\alpha) generated by an algebraic integer \alpha with minimal polynomial f(x) \in \mathbb{Z}, the discriminant \Delta_K relates to the polynomial discriminant \operatorname{disc}(f) by the formula \Delta_K = \operatorname{disc}(f) / m^2, where m = [\mathcal{O}_K : \mathbb{Z}[\alpha]] is the index of the order \mathbb{Z}[\alpha] in \mathcal{O}_K.[30] The primes dividing \Delta_K are precisely the ramified primes in the extension K/\mathbb{Q}, providing a measure of how the prime ideals of \mathbb{Z} factor in \mathcal{O}_K.[29] For instance, in quadratic fields, the ramification occurs exactly at the primes dividing \Delta_K. In the quadratic case, for K = \mathbb{Q}(\sqrt{d}) with d a squarefree integer not congruent to 0 or 1 modulo 4, the ring of integers is \mathcal{O}_K = \mathbb{Z}[\sqrt{d}] and \Delta_K = 4d; if d \equiv 1 \pmod{4}, then \mathcal{O}_K = \mathbb{Z}\left[\frac{1 + \sqrt{d}}{2}\right] and \Delta_K = d.[29] Here, \Delta_K coincides with the discriminant \Delta of the minimal polynomial x^2 - \Delta x + \cdots = 0 for a primitive element, adjusted for the fundamental discriminant D underlying the field. For example, in K = \mathbb{Q}(\sqrt{5}), where $5 \equiv 1 \pmod{4}, the minimal polynomial of \frac{1 + \sqrt{5}}{2} is x^2 - x - 1 with \operatorname{disc}(x^2 - x - 1) = 5, and since the index is 1, \Delta_K = 5.[31] The prime 5 ramifies in this field as (5) = \mathfrak{p}^2 for the prime ideal \mathfrak{p} = \left( \frac{1 + \sqrt{5}}{2}, 2 \right).[29]Fundamental Discriminants
In the context of quadratic number fields, a fundamental discriminant is defined as an integer \Delta that is either a square-free integer congruent to 1 modulo 4, or four times a square-free integer congruent to 2 or 3 modulo 4.[32] This ensures \Delta is the discriminant of the ring of integers of a quadratic field and distinguishes the maximal order from non-maximal ones.[33] For a quadratic field K = \mathbb{Q}(\sqrt{d}), where d is a square-free integer not equal to 1, the fundamental discriminant \Delta_d is given by \Delta_d = d if d \equiv 1 \pmod{4}, and \Delta_d = 4d otherwise.[34] This \Delta_d serves as the field discriminant, determining the structure of the ring of integers \mathcal{O}_K: \mathcal{O}_K = \mathbb{Z}[\sqrt{d}] when d \equiv 2 or $3 \pmod{4}, and \mathcal{O}_K = \mathbb{Z}\left[\frac{1 + \sqrt{d}}{2}\right] when d \equiv 1 \pmod{4}.[34] Each quadratic field has a unique fundamental discriminant, classifying quadratic fields up to isomorphism.[33] The prime factorization of the ideal (p) in \mathcal{O}_K is governed by \Delta_d: an odd prime p ramifies if and only if p divides \Delta_d, in which case (p) = \mathfrak{p}^2 for some prime ideal \mathfrak{p} of norm p.[35] For the prime 2, ramification occurs precisely when d \equiv 2 or $3 \pmod{4} (equivalently, when \Delta_d is even), with the specific behavior depending on \Delta_d modulo 8: if \Delta_d \equiv 4 \pmod{8}, then 2 ramifies as (2) = \mathfrak{p}^2 where \mathfrak{p} has norm 2; if \Delta_d \equiv 0 \pmod{8}, the ramification index is also 2 but with adjusted norm considerations.[34] When \Delta_d is odd (i.e., d \equiv 1 \pmod{4}), 2 does not ramify and either splits or remains inert based on whether \Delta_d \equiv 1 or $5 \pmod{8}.[35] Fundamental discriminants play a key role in the arithmetic of quadratic fields by classifying the orders within \mathcal{O}_K: every quadratic order has a discriminant that is the fundamental discriminant multiplied by the square of the conductor, with the class number of the order relating to that of the maximal order via this structure.[32] This classification aids in computing ideal class groups and understanding the distribution of primes. A representative example is the Gaussian field K = \mathbb{Q}(i), where d = -1 \equiv 3 \pmod{4}, so \Delta_d = -4. Here, the ring of integers is \mathbb{Z}, and the prime 2 ramifies as (2) = (1 + i)^2, with no odd primes dividing -4.[34] The class number of this field is 1, illustrating a principal ideal domain.[35]Geometric and Homogeneous Interpretations
Homogeneous Bivariate Polynomials
Homogeneous bivariate polynomials, also known as binary forms, offer a projective perspective on the discriminant by homogenizing univariate polynomials. Consider a univariate polynomial p(t) = \sum_{k=0}^n a_k t^k of degree n. This can be viewed as the dehomogenization of a binary form f(x, y) = \sum_{k=0}^n a_k x^k y^{n-k}, which is homogeneous of degree n in the variables x and y. The roots of p(t) = 0 correspond to the points [t : 1] in the projective line \mathbb{P}^1 where f(x, y) = 0, and multiple roots of p arise precisely when f has a singular point, meaning there exists [x : y] \in \mathbb{P}^1 such that f(x, y) = \frac{\partial f}{\partial x}(x, y) = \frac{\partial f}{\partial y}(x, y) = 0. In this bivariate setting, the discriminant \Delta(f) serves as an invariant that detects such singularities and remains unchanged under projective transformations induced by \mathrm{SL}(2).[36] The discriminant of the binary form f is closely related to the resultant of its partial derivatives. Specifically, \Delta(f) = (-1)^{n(n-1)/2} \frac{1}{n^{n-2}} \Res\left( \frac{\partial f}{\partial x}, \frac{\partial f}{\partial y} \right), where the resultant is taken with respect to one of the variables, say y. The partial derivatives \frac{\partial f}{\partial x} and \frac{\partial f}{\partial y} are themselves binary forms of degree n-1, so their resultant is a determinant of a (2n-2) \times (2n-2) Sylvester-type matrix whose entries are linear in the coefficients of f. This expression for \Delta(f) generalizes the univariate discriminant formula \Delta(p) = (-1)^{n(n-1)/2} a_n^{-1} \Res(p, p'), and \Delta(f) equals \Delta(p) up to normalization by powers of the leading coefficient.[37] As a polynomial in the coefficients a_0, \dots, a_n, the discriminant \Delta(f) is homogeneous of degree $2n-2. Under the natural action of \mathrm{GL}(2) on binary forms, defined by f \circ g (x, y) = f((x, y) g) for g \in \mathrm{GL}(2), the discriminant transforms as a relative invariant: \Delta(f \circ g) = (\det g)^{2n-2} \Delta(f). This scaling ensures that \Delta(f) detects multiple roots invariantly under projective equivalence, emphasizing its role in classical invariant theory for binary forms.[11] A concrete example illustrates these properties for n=2. The quadratic binary form f(x, y) = a x^2 + 2 h x y + b y^2 has partial derivatives \frac{\partial f}{\partial x} = 2 a x + 2 h y and \frac{\partial f}{\partial y} = 2 h x + 2 b y. The resultant of these linear forms is $4 a b - 4 h^2, and applying the general formula yields \Delta(f) = 4 a b - 4 h^2. This expression is homogeneous of degree 2 in a, h, b, vanishes precisely when the associated conic has a singular point (repeated root), and scales by (\det g)^2 under \mathrm{GL}(2) transformations.[38]Quadratic Forms and Conics
In the context of quadratic forms, the discriminant serves as a measure of degeneracy for the associated bilinear structure. A quadratic form over the real or complex numbers can be expressed as Q(\mathbf{x}) = \sum_{i,j} a_{ij} x_i x_j = \mathbf{x}^T A \mathbf{x}, where A = (a_{ij}) is a symmetric matrix known as the coefficient matrix. The discriminant of this quadratic form is defined as the determinant of A, denoted \det(A). If \det(A) = 0, the quadratic form is singular or degenerate, meaning the matrix A has a non-trivial kernel, and the form factors into linear terms, representing a cone or pair of hyperplanes in the appropriate dimension.[39] This condition arises because the Hessian matrix of second partial derivatives for the quadratic function is twice the matrix A (up to scaling), and its determinant vanishing indicates a loss of non-degeneracy in the curvature. The Hessian perspective connects directly to geometric interpretations, where the discriminant assesses whether the quadratic form defines a non-degenerate quadric hypersurface. For instance, in two variables, the binary quadratic form ax^2 + bxy + cy^2 has associated matrix \begin{pmatrix} a & b/2 \\ b/2 & c \end{pmatrix}, and $4\det(A) = 4ac - b^2, with zero value (i.e., b^2 = 4ac) implying the form degenerates into linear factors. The conic type is classified by the sign of b^2 - 4ac = -(4ac - b^2): b^2 - 4ac < 0 for ellipses (including circles), =0 for parabolas, and >0 for hyperbolas, influencing applications in optimization and conic classification.[39] Quadratic forms extend naturally to the study of conic sections in the projective plane, where the discriminant determines both the type and degeneracy of the curve defined by the general equation ax^2 + bxy + cy^2 + dx + ey + f = 0. In homogeneous coordinates [x : y : z], this corresponds to the quadratic form ax^2 + bxy + cy^2 + dxz + eyz + fz^2 = 0, represented by the symmetric 3×3 matrix \begin{pmatrix} a & b/2 & d/2 \\ b/2 & c & e/2 \\ d/2 & e/2 & f \end{pmatrix}. The discriminant \Delta of the conic is the determinant of this matrix. If \Delta \neq 0, the conic is non-degenerate; if \Delta = 0, it degenerates into a pair of lines (possibly coincident or parallel).[40] Additionally, the sign of the quadratic part's discriminant b^2 - 4ac classifies non-degenerate conics: negative for ellipses (including circles), zero for parabolas, and positive for hyperbolas.[41] The full discriminant \Delta can be computed explicitly as the expanded form \Delta = abc + 2fgh - af^2 - bg^2 - ch^2, where the linear coefficients are adjusted for symmetry (often written with 2d, 2e for convenience), providing a simplified invariant without matrix computation.[41] This 3×3 determinant encapsulates the degeneracy condition, distinguishing proper conics from reducible curves like intersecting lines. In projective geometry, non-degeneracy ensures the conic is irreducible and smooth, foundational for intersection theory and duality.[40] A representative example is the equation xy = 1, or xy - 1 = 0, which rewrites as $0 \cdot x^2 + 1 \cdot xy + 0 \cdot y^2 + 0 \cdot x + 0 \cdot y - 1 = 0. The associated matrix is \begin{pmatrix} 0 & 1/2 & 0 \\ 1/2 & 0 & 0 \\ 0 & 0 & -1 \end{pmatrix}, with determinant \Delta = 0 \cdot (0 \cdot (-1) - 0 \cdot 0) - (1/2) \cdot ((1/2) \cdot (-1) - 0 \cdot 0) + 0 \cdot ((1/2) \cdot 0 - 0 \cdot 0) = -(1/2) \cdot (-1/2) = 1/4 \neq 0, confirming a non-degenerate conic.[40] Furthermore, b^2 - 4ac = 1 > 0 identifies it as a hyperbola, rotated by 45 degrees relative to the axes.[41]Quadric Surfaces and Higher Geometry
In three-dimensional space, quadric surfaces are defined by the general equation ax^2 + by^2 + cz^2 + 2fyz + 2gzx + 2hxy + 2px + 2qy + 2rz + d = 0, which can be represented in matrix form as \mathbf{X}^T E \mathbf{X} = 0, where \mathbf{X} = (x, y, z, 1)^T and E is the symmetric 4×4 coefficient matrix: E = \begin{pmatrix} a & h & g & p \\ h & b & f & q \\ g & f & c & r \\ p & q & r & d \end{pmatrix}. The discriminant of this quadric surface is given by \Delta = \det(E), up to a nonzero scalar multiple.[42] A nonvanishing discriminant (\Delta \neq 0) indicates a smooth quadric surface, such as an ellipsoid or hyperboloid, while \Delta = 0 signals degeneracy, often resulting in ruled surfaces like cones or cylinders.[42] The classification further relies on the eigenvalues of the 3×3 submatrix e (corresponding to the quadratic terms) and the sign of \Delta: for full rank (rank 4 for E, rank 3 for e), an ellipsoid arises when all eigenvalues have the same sign and \Delta < 0, whereas hyperboloids occur with mixed eigenvalue signs (one sheet if \Delta > 0, two sheets if \Delta < 0).[42] This determinant-based discriminant extends naturally to quadrics in higher dimensions. For a quadric hypersurface in projective space \mathbb{P}^n defined by a homogeneous quadratic form f(\mathbf{x}) = \mathbf{x}^T A \mathbf{x} = 0, where A is the symmetric (n+1) \times (n+1) matrix and \mathbf{x} \in \mathbb{P}^n, the discriminant is \Delta(f) = (-1)^{(n+1)(n+2)/2} 2^{n+1} \det(A) if n+1 is even, or \Delta(f) = (-1)^{(n+1)n/2} \det(A) if n+1 is odd (arising from the determinant of the Hessian matrix, which is twice A).[21] In general, \Delta(f) serves as an invariant under projective transformations and vanishes precisely when the hypersurface is singular, i.e., when \det(A) = 0, leading to a singular locus of positive dimension.[21] For even-dimensional projective spaces (odd number of variables), the discriminant may also involve Pfaffian invariants in certain contexts, but the determinant remains the primary measure for singularity detection.[43] In higher geometry, this framework allows the discriminant to detect singularities in quadric hypersurfaces embedded in \mathbb{P}^n for n \geq 3, confirming smoothness for nondegenerate cases essential in applications like intersection theory. For instance, the equation x^2 + y^2 - z^2 - w^2 = 0 in \mathbb{P}^3 defines a smooth hyperboloid of one sheet, with matrix \operatorname{diag}(1,1,-1,-1) yielding \det(A) = 1 \neq 0, thus \Delta \neq 0.[21] In contrast, degeneracy to a ruled surface, such as a cone, occurs when \Delta = 0, as in x^2 + y^2 - z^2 = 0 (homogenized with zero w^2 term), where the matrix has determinant zero and a singular point at the origin.[42] This analogy to the two-dimensional conic discriminant underscores the role of such invariants in classifying projective varieties by their geometric type.[21]Broader Generalizations and Uses
Relation to Resultants and Resolvents
The discriminant of a polynomial p(x) = a_n x^n + \cdots + a_0 of degree n is intimately connected to the resultant through the formula \Delta(p) = (-1)^{n(n-1)/2} \frac{\operatorname{Res}(p, p')}{a_n}, where p'(x) is the derivative of p(x) and \operatorname{Res}(p, p') denotes the resultant of p and p'.[18] This relation arises because the resultant \operatorname{Res}(p, p') vanishes if and only if p and p' share a common root, which occurs precisely when p has a multiple root, mirroring the condition for \Delta(p) = 0.[44] More generally, the discriminant can be viewed as a specialized resultant encoding differences between roots. Specifically, \Delta(p) = a_n^{2n-2} \prod_{i < j} (r_i - r_j)^2, where r_1, \dots, r_n are the roots of p, and this product of squared root differences aligns with the resultant's structure as a determinant that captures root separations between two polynomials.[18] In this sense, the discriminant emerges as the resultant evaluated in a manner that isolates pairwise root distinctions, providing a symmetric measure of the polynomial's root configuration.[44] In the context of Galois theory, the discriminant relates to resolvents, which are auxiliary polynomials used to probe the structure of Galois groups. The discriminant itself functions as a resolvent corresponding to the sign homomorphism, determining whether the Galois group is contained in the alternating group by checking if \Delta(p) is a square in the base field.[45] Its square root effectively resolves the parity of permutations in the group, aiding in solvability assessments for equations.[45] For example, consider an irreducible cubic polynomial f(x) = x^3 + a x^2 + b x + c over a field K of characteristic not 2 or 3. The associated quadratic resolvent R_2(x) has discriminant equal to that of f, and the Galois group of f over K is the alternating group A_3 if R_2(x) is reducible over K (equivalently, if \Delta(f) is a square in K), and the symmetric group S_3 otherwise.[45] This connection highlights how the discriminant, via the resolvent, distinguishes cyclic from full symmetric extensions in the cubic case.[45]Applications in Algebraic Geometry
In algebraic geometry, the discriminant plays a crucial role in identifying singularities of hypersurfaces within families parameterized by a space of coefficients. For a hypersurface defined by a polynomial equation f(x_1, \dots, x_n; a_1, \dots, a_m) = 0, where the a_i are parameters, the discriminant locus is the subvariety in the parameter space where the hypersurface develops singularities, such as nodes on curves. This locus is itself a hypersurface, and its vanishing detects points where the partial derivatives \partial f / \partial x_i fail to be independent, corresponding to singular points on the fiber. For instance, in families of plane curves, the discriminant captures nodal singularities, distinguishing smooth curves from those with transverse self-intersections.[46] In the context of moduli spaces, the discriminant arises as a canonical section of a line bundle over the moduli space of curves, encoding the geometry of degenerations. Specifically, on the Deligne-Mumford compactification \overline{\mathcal{M}}_g of the moduli space of genus-g curves, the discriminant divisor \Delta is the closure of the locus of smooth curves degenerating to singular ones, and the associated section vanishes precisely along the boundary components parameterizing stable curves with nodes. This structure allows the discriminant to measure obstructions to smoothness and facilitates the study of compactifications. The discriminant thus provides a tool for analyzing the birational geometry and cohomology of these moduli spaces. A prominent application is to elliptic curves, where the discriminant distinguishes smooth models from singular ones in Weierstrass form. For an elliptic curve given by y^2 = x^3 + A x + B over a field of characteristic not 2 or 3, the discriminant is \Delta = -16(4A^3 + 27B^2), and the curve is singular if and only if \Delta = 0. The j-invariant, j = 1728 \frac{(4A)^3}{\Delta}, classifies isomorphism classes of elliptic curves, with the relation linking \Delta to the modular discriminant and ensuring that \Delta \neq 0 corresponds to non-singular fibers in families. This formulation is essential for studying the moduli space \mathcal{M}_{1,1} of elliptic curves, where the discriminant hypersurface parameterizes cuspidal or nodal degenerations. An illustrative example is the space of plane cubic curves in \mathbb{P}^2, parameterized by \mathbb{P}^9. The discriminant is a hypersurface of degree 12 in this projective space, whose vanishing classifies singular cubics, such as those with a node or cusp. This discriminant locus stratifies the parameter space, with the generic singular cubic exhibiting a node, and it serves as the dual variety to the Veronese surface of lines, highlighting the interplay between singularities and projective duality.[46]Computational Aspects
The discriminant of a univariate polynomial f(x) of degree n with leading coefficient a_n can be computed symbolically using the formula \Delta(f) = (-1)^{n(n-1)/2} a_n^{-1} \operatorname{Res}(f, f'), where f' is the derivative of f and \operatorname{Res} denotes the resultant.[18] The resultant \operatorname{Res}(f, f') is the determinant of the Sylvester matrix \operatorname{Syl}_{n,n-1}(f, f'), a (2n-1) \times (2n-1) matrix constructed from the coefficients of f and f' with shifted rows.[18] For multivariate polynomials or more complex cases, Gröbner bases can be employed to eliminate variables and reduce to univariate resultants, though for standard univariate discriminants, the Sylvester approach suffices.[47] Computing the determinant of the Sylvester matrix via Gaussian elimination requires O(n^3) arithmetic operations in the coefficient field.[18] More efficient methods use subresultant polynomial remainder sequences (PRS), which compute the resultant—and thus the discriminant—in O(n^2) time by avoiding intermediate coefficient swell and leveraging structured elimination.[48] Numerical computation of discriminants for high-degree polynomials is prone to instability, particularly when using floating-point arithmetic to approximate roots and evaluate \Delta(f) \approx a_n^{2n-2} \prod_{i < j} (r_i - r_j)^2, as small perturbations in coefficients amplify errors exponentially with n. The condition number of the Sylvester matrix grows rapidly with degree, leading to severe loss of precision in resultant-based methods for n \gtrsim 20 over floating-point fields. In SageMath, thediscriminant() method for univariate polynomials implements the resultant formula \Delta(f) = (-1)^{n(n-1)/2} a_n^{-1} \operatorname{Res}(f, f'), calling the internal resultant() function which uses PARI's polresultant for supported base rings or falls back to Sylvester determinant evaluation.[49] Mathematica's Discriminant[poly, var] function similarly computes the discriminant via the resultant of poly and its derivative, with options for modular arithmetic but defaulting to symbolic evaluation over exact fields.[50] Both systems support subresultant-based optimizations for efficiency in higher degrees.
For a degree-5 example, consider the monic quintic f(x) = x^5 + p x + q. The derivative is f'(x) = 5x^4 + p. Construct the 9×9 Sylvester matrix with the first four rows from coefficients of f (shifted: [1,0,0,0,p,0,q,0,0] etc.) and the next five from f' (shifted: [5,0,0,0,p,0,0,0,0] etc.). The resultant is the determinant of this matrix, and \Delta(f) = \operatorname{Res}(f, f') since n(n-1)/2 = 10 is even and a_5 = 1. For p=0, q=-1 (i.e., f(x) = x^5 - 1), this yields \Delta(f) = 3125.[51]