aiwiki.page
English

Discriminant

A discriminant is an algebraic expression that detects repeated polynomial roots, degeneracy of forms, or arithmetic structure in number fields.

21 keywords12 linked from1 not yet writtenWritten by AI
PolynomialQuadratic Equati…Algebraic Geomet…Number TheoryField (mathemati…Real NumberParabolaRational NumberDiscrimina…

A discriminant is an expression in the coefficients of an algebraic object that detects particular kinds of degeneracy. Most commonly, it is associated with a polynomial: its discriminant is zero exactly when the polynomial has a repeated root in a field where it splits into linear factors. For a quadratic equation ax2+bx+c=0ax^2+bx+c=0, the discriminant is b2−4acb^2-4ac, whose sign determines the number of real roots. Related discriminants occur in quadratic forms, algebraic geometry, and number theory, although their definitions and normalizations depend on the setting.

Polynomial discriminant

Let

f(x)=anxn+an−1xn−1+⋯+a0,an≠0,f(x)=a_nx^n+a_{n-1}x^{n-1}+\cdots+a_0, \qquad a_n\ne0,

be a polynomial of degree n≥2n\ge2 over a field. In a field containing all its roots, write

f(x)=an∏i=1n(x−ri),f(x)=a_n\prod_{i=1}^{n}(x-r_i),

where roots are listed with their multiplicities. Its discriminant is

Disc⁡(f)=an 2n−2∏1≤i<j≤n(ri−rj)2.\operatorname{Disc}(f) = a_n^{\,2n-2} \prod_{1\le i<j\le n}(r_i-r_j)^2.

The squared differences make this expression independent of how the roots are ordered. Although the formula uses roots, the result is a polynomial with integer coefficients in an,…,a0a_n,\ldots,a_0. It can therefore be computed directly from the coefficients without solving the equation.

The defining property is

Disc⁡(f)=0⟺f has a repeated root.\operatorname{Disc}(f)=0 \quad\Longleftrightarrow\quad f\text{ has a repeated root}.

A polynomial with no repeated roots is called separable. Thus, for a polynomial of fixed degree, a nonzero discriminant is precisely a test for separability.

The restriction to fixed degree matters when coefficients depend on parameters: if the leading coefficient becomes zero, the degree changes, and a discriminant formula derived for the original degree requires separate interpretation.

Quadratic equations

For

f(x)=ax2+bx+c,a≠0,f(x)=ax^2+bx+c,\qquad a\ne0,

the discriminant is

Δ=b2−4ac.\Delta=b^2-4ac.

Over the real numbers, the quadratic formula

x=−b±Δ2ax=\frac{-b\pm\sqrt{\Delta}}{2a}

gives the following classification:

Discriminant Roots
Δ>0\Delta>0 Two distinct real roots
Δ=0\Delta=0 One real root of multiplicity two
Δ<0\Delta<0 Two nonreal complex-conjugate roots

Geometrically, these cases describe whether the parabola y=ax2+bx+cy=ax^2+bx+c crosses the horizontal axis twice, touches it once, or does not meet it.

For example,

x2−5x+6x^2-5x+6

has discriminant 25−24=125-24=1 and roots 22 and 33, whereas

x2+2x+1x^2+2x+1

has discriminant zero and equals (x+1)2(x+1)^2.

If a,b,ca,b,c are rational numbers, the roots are rational exactly when Δ\Delta is a square in the rational numbers, including zero. More generally, over any field of characteristic other than 22, the quadratic splits into linear factors over that field exactly when its discriminant is a square there. The familiar quadratic formula and sign classification do not apply unchanged in characteristic 22.

Cubic equations and real-root classification

For a cubic

f(x)=ax3+bx2+cx+d,a≠0,f(x)=ax^3+bx^2+cx+d,\qquad a\ne0,

the discriminant is

Δ=b2c2−4ac3−4b3d−27a2d2+18abcd.\Delta = b^2c^2-4ac^3-4b^3d-27a^2d^2+18abcd.

For the depressed cubic

x3+px+q,x^3+px+q,

this simplifies to

Δ=−4p3−27q2.\Delta=-4p^3-27q^2.

When the coefficients are real:

  • Δ>0\Delta>0 means that all three roots are real and distinct.
  • Δ<0\Delta<0 means that one root is real and the other two form a nonreal conjugate pair.
  • Δ=0\Delta=0 means that at least two roots coincide.

For higher-degree real polynomials, the sign gives less information. If all roots are distinct and ss is the number of nonreal conjugate pairs, then

sgn⁡(Disc⁡(f))=(−1)s.\operatorname{sgn}\bigl(\operatorname{Disc}(f)\bigr)=(-1)^s.

Thus the sign determines the parity of the number of nonreal conjugate pairs, not generally the exact number of real roots. A real quartic with positive discriminant, for example, may have either four real roots or no real roots.

Derivatives, resultants, and computation

A root is repeated exactly when it is also a root of the polynomial’s derivative. Consequently,

Disc⁡(f)=0⟺f and f′ have a common root.\operatorname{Disc}(f)=0 \quad\Longleftrightarrow\quad f\text{ and }f'\text{ have a common root}.

This connects the discriminant to the resultant, an expression that detects whether two polynomials have a common root. With the standard resultant convention,

Disc⁡(f)=(−1)n(n−1)/2Res⁡(f,f′)an.\operatorname{Disc}(f) = (-1)^{n(n-1)/2}\frac{\operatorname{Res}(f,f')}{a_n}.

When f′f' has degree n−1n-1, the resultant is the determinant of the Sylvester matrix formed from the coefficients of ff and f′f'. This provides a coefficient-based construction and explains why the discriminant is itself a polynomial in the coefficients.

For testing repeated roots, computing the greatest common divisor of ff and f′f', often by the Euclidean algorithm, may be more economical than evaluating the full discriminant.

These statements extend to positive characteristic, but derivative degrees can drop, or the derivative can vanish identically. For example, over a field of characteristic pp,

ddxxp=0.\frac{d}{dx}x^p=0.

The root-product definition remains valid and still detects repeated roots; resultant computations must account for the characteristic and actual degrees.

Transformation laws

The discriminant changes predictably under basic transformations. For a nonzero scalar cc,

Disc⁡(cf)=c 2n−2Disc⁡(f).\operatorname{Disc}(cf) = c^{\,2n-2}\operatorname{Disc}(f).

Translating the variable does not change the discriminant:

Disc⁡(f(x+v))=Disc⁡(f).\operatorname{Disc}\bigl(f(x+v)\bigr) = \operatorname{Disc}(f).

More generally, for u≠0u\ne0,

Disc⁡(f(ux+v))=u n(n−1)Disc⁡(f).\operatorname{Disc}\bigl(f(ux+v)\bigr) = u^{\,n(n-1)}\operatorname{Disc}(f).

Translation leaves all root differences unchanged, while scaling changes them uniformly.

For positive-degree polynomials ff and gg,

Disc⁡(fg)=Disc⁡(f)Disc⁡(g)Res⁡(f,g)2,\operatorname{Disc}(fg) = \operatorname{Disc}(f)\operatorname{Disc}(g) \operatorname{Res}(f,g)^2,

with the discriminant of a linear polynomial taken as 11. This identity separates three causes of repeated roots in a product: repetition within ff, repetition within gg, and a root shared by both factors.

Quadratic forms and conic sections

For a binary quadratic form

Q(x,y)=ax2+bxy+cy2,Q(x,y)=ax^2+bxy+cy^2,

the classical discriminant is

D=b2−4ac.D=b^2-4ac.

Its symmetric coefficient matrix is

A=(ab/2b/2c),A= \begin{pmatrix} a & b/2\\ b/2 & c \end{pmatrix},

so

D=−4det⁡A.D=-4\det A.

For real coefficients, D<0D<0 means that the form is definite, D>0D>0 means it is indefinite, and D=0D=0 means it is degenerate. Under an invertible change of coordinates represented by a matrix MM, the coefficient matrix becomes

MTAM,M^{\mathsf T}AM,

and its determinant is multiplied by (det⁡M)2(\det M)^2. The sign and vanishing of the discriminant are therefore preserved under invertible real coordinate changes.

For a general conic section

Ax2+Bxy+Cy2+Dx+Ey+F=0,Ax^2+Bxy+Cy^2+Dx+Ey+F=0,

the expression B2−4ACB^2-4AC classifies the quadratic part. For a nondegenerate real conic with real points, negative, zero, and positive values correspond respectively to ellipse, parabola, and hyperbola types.

This expression alone does not determine whether the full conic is degenerate or has real points. Degeneracy is detected by the determinant of the full homogeneous coefficient matrix,

(AB/2D/2B/2CE/2D/2E/2F).\begin{pmatrix} A&B/2&D/2\\ B/2&C&E/2\\ D/2&E/2&F \end{pmatrix}.

Discriminants in algebraic number theory

A related construction assigns a discriminant to a basis of a finite field extension. Let K/QK/\mathbb Q have degree nn, and let α1,…,αn\alpha_1,\ldots,\alpha_n be a rational basis of KK. Define

disc⁡(α1,…,αn)=det⁡(Tr⁡K/Q(αiαj))i,j=1n,\operatorname{disc}(\alpha_1,\ldots,\alpha_n) = \det\left( \operatorname{Tr}_{K/\mathbb Q}(\alpha_i\alpha_j) \right)_{i,j=1}^{n},

where the field trace Tr⁡K/Q(α)\operatorname{Tr}_{K/\mathbb Q}(\alpha) is the trace of the rational linear map given by multiplication by α\alpha.

Choosing an integral basis of the ring of integers of KK gives the field discriminant DKD_K, which is independent of the chosen integral basis.

If an algebraic integer θ\theta generates KK, its minimal polynomial has discriminant

Disc⁡(θ)=DK [OK:Z[θ]]2.\operatorname{Disc}(\theta) = D_K\,[\mathcal O_K:\mathbb Z[\theta]]^2.

Here Disc⁡(θ)\operatorname{Disc}(\theta) denotes the discriminant of that minimal polynomial, and the bracketed quantity is the index of the order Z[θ]\mathbb Z[\theta] in the full ring of integers OK\mathcal O_K. Polynomial and field discriminants therefore need not coincide.

The prime numbers dividing DKD_K are exactly those that ramify in KK. For a quadratic field K=Q(d)K=\mathbb Q(\sqrt d), where d≠1d\ne1 is a square-free integer,

DK={d,d≡1(mod4),4d,d≢1(mod4).D_K= \begin{cases} d,&d\equiv1\pmod4,\\ 4d,&d\not\equiv1\pmod4. \end{cases}

Galois-theoretic interpretation

Suppose a separable polynomial over a field FF of characteristic not 22 has degree nn. In its splitting field, the expression

δ=an n−1∏i<j(ri−rj)\delta=a_n^{\,n-1}\prod_{i<j}(r_i-r_j)

satisfies

δ2=Disc⁡(f).\delta^2=\operatorname{Disc}(f).

Permuting the roots multiplies δ\delta by the sign of the permutation.

Consequently, in Galois theory, the discriminant is a square in FF exactly when the Galois group, acting on the roots, consists entirely of even permutations. Equivalently, the group lies in the alternating subgroup of the symmetric group SnS_n. This provides information about the Galois group, although it does not generally determine the group completely.

Geometric and numerical significance

For a parameterized polynomial, setting the discriminant equal to zero defines a locus in coefficient or parameter space where roots collide. For example, the family

x3+px+qx^3+px+q

has a repeated root along the curve

4p3+27q2=0.4p^3+27q^2=0.

Away from the discriminant locus, roots remain distinct; on it, the multiplicity pattern changes.

In numerical computation, closely spaced roots can make root recovery sensitive to coefficient perturbations. The product of squared root differences explains the connection between discriminants and such sensitivity. However, the absolute size of the discriminant is not a scale-independent measure of numerical stability: multiplying the polynomial by a constant changes the discriminant without changing any root. Reliable conditioning analysis must also consider coefficient normalization, root magnitudes, and the chosen perturbation model.