ConceptComplete

Nonsingular Varieties

A variety is nonsingular (or smooth) if it has a well-defined tangent space of the expected dimension at every point. Singularities โ€” nodes, cusps, and worse โ€” are the points where the geometry degenerates. Understanding where and how singularities arise is one of the central concerns of algebraic geometry.

Throughout, kk denotes an algebraically closed field and varieties are as in Hartshorne, Chapter I.


The Zariski tangent space

Definition5.1Zariski tangent space

Let YY be a variety over kk and let PโˆˆYP \in Y be a point. Let mPโІOY,P\mathfrak{m}_P \subseteq \mathcal{O}_{Y,P} be the maximal ideal of the local ring at PP. The Zariski tangent space of YY at PP is the kk-vector space

TPY=(mP/mP2)โˆ—.T_P Y = (\mathfrak{m}_P / \mathfrak{m}_P^2)^*.

Equivalently, TPY=Derโกk(OY,P,k)T_P Y = \operatorname{Der}_k(\mathcal{O}_{Y,P}, k), the space of kk-derivations OY,Pโ†’k\mathcal{O}_{Y,P} \to k.

RemarkIntrinsic definition

The Zariski tangent space is defined intrinsically from the local ring, without reference to an embedding. This is crucial: it means the tangent space is an invariant of the variety, not of a particular presentation by equations.

ExampleZariski tangent space of affine space

Let Y=AnY = \mathbb{A}^n and P=(a1,โ€ฆ,an)P = (a_1, \ldots, a_n). The local ring is OAn,P=k[x1,โ€ฆ,xn]mP\mathcal{O}_{\mathbb{A}^n, P} = k[x_1, \ldots, x_n]_{\mathfrak{m}_P} where mP=(x1โˆ’a1,โ€ฆ,xnโˆ’an)\mathfrak{m}_P = (x_1 - a_1, \ldots, x_n - a_n). Then

mP/mP2โ‰…โจi=1nkโ‹…(xiโˆ’ai)โ€พ,\mathfrak{m}_P / \mathfrak{m}_P^2 \cong \bigoplus_{i=1}^n k \cdot \overline{(x_i - a_i)},

so dimโกkTPAn=n\dim_k T_P \mathbb{A}^n = n. The tangent space at every point of An\mathbb{A}^n is nn-dimensional, as expected.

ExampleTangent space for an embedded variety

Let Y=V(f1,โ€ฆ,fr)โІAnY = V(f_1, \ldots, f_r) \subseteq \mathbb{A}^n and PโˆˆYP \in Y. For each fif_i, consider the differential

dPfi=โˆ‘j=1nโˆ‚fiโˆ‚xj(P)โ‹…(xjโˆ’aj).d_P f_i = \sum_{j=1}^n \frac{\partial f_i}{\partial x_j}(P) \cdot (x_j - a_j).

These are linear forms on knk^n. The Zariski tangent space is

TPY=V(dPf1,โ€ฆ,dPfr)={vโˆˆknโ€…โ€Š|โ€…โ€Šโˆ‘j=1nโˆ‚fiโˆ‚xj(P)โ‹…vj=0ย forย allย i}.T_P Y = V(d_P f_1, \ldots, d_P f_r) = \left\{ v \in k^n \;\middle|\; \sum_{j=1}^n \frac{\partial f_i}{\partial x_j}(P) \cdot v_j = 0 \text{ for all } i \right\}.

In other words, TPY=kerโกJ(P)T_P Y = \ker J(P) where JJ is the Jacobian matrix. We always have dimโกTPYโ‰ฅdimโกY\dim T_P Y \geq \dim Y, with equality characterizing nonsingularity.


The Jacobian criterion

Definition5.2Jacobian matrix

Let Y=V(f1,โ€ฆ,fr)โІAnY = V(f_1, \ldots, f_r) \subseteq \mathbb{A}^n. The Jacobian matrix of YY at a point PโˆˆYP \in Y is the rร—nr \times n matrix

J(P)=(โˆ‚f1โˆ‚x1(P)โ‹ฏโˆ‚f1โˆ‚xn(P)โ‹ฎโ‹ฑโ‹ฎโˆ‚frโˆ‚x1(P)โ‹ฏโˆ‚frโˆ‚xn(P)).J(P) = \begin{pmatrix} \dfrac{\partial f_1}{\partial x_1}(P) & \cdots & \dfrac{\partial f_1}{\partial x_n}(P) \\[6pt] \vdots & \ddots & \vdots \\[6pt] \dfrac{\partial f_r}{\partial x_1}(P) & \cdots & \dfrac{\partial f_r}{\partial x_n}(P) \end{pmatrix}.

Theorem5.1Jacobian criterion for nonsingularity

Let YY be a variety of dimension dd in An\mathbb{A}^n, and write Y=V(f1,โ€ฆ,fr)Y = V(f_1, \ldots, f_r) where (f1,โ€ฆ,fr)=I(Y)(f_1, \ldots, f_r) = I(Y) (or any set of generators). A point PโˆˆYP \in Y is nonsingular if and only if

rankโกJ(P)=nโˆ’d.\operatorname{rank} J(P) = n - d.

Equivalently, PP is nonsingular iff dimโกkTPY=d=dimโกY\dim_k T_P Y = d = \dim Y.

RemarkIndependence of presentation

The rank condition rankโกJ(P)=nโˆ’d\operatorname{rank} J(P) = n - d does not depend on the choice of generators of I(Y)I(Y). If we add redundant generators, the Jacobian gets more rows but the rank at a nonsingular point remains nโˆ’dn - d.

ExampleJacobian criterion for a plane curve

Let C=V(f)โІA2C = V(f) \subseteq \mathbb{A}^2 be an irreducible plane curve (dimโกC=1\dim C = 1). Here n=2n = 2, d=1d = 1, so the Jacobian criterion requires rankโกJ(P)=1\operatorname{rank} J(P) = 1, i.e.,

(โˆ‚fโˆ‚x(P),โ€…โ€Šโˆ‚fโˆ‚y(P))โ‰ (0,0).\left( \frac{\partial f}{\partial x}(P),\; \frac{\partial f}{\partial y}(P) \right) \neq (0, 0).

A point PP is singular iff f(P)=0f(P) = 0 and both partial derivatives vanish at PP.

ExampleThe circle is smooth

Let C=V(x2+y2โˆ’1)โІA2C = V(x^2 + y^2 - 1) \subseteq \mathbb{A}^2 over kk with charโก(k)โ‰ 2\operatorname{char}(k) \neq 2. The Jacobian is

J=(2x2y).J = \begin{pmatrix} 2x & 2y \end{pmatrix}.

At a point P=(a,b)P = (a, b) on CC: J(P)=(2a,โ€…โ€Š2b)J(P) = (2a,\; 2b). This vanishes iff a=b=0a = b = 0, but (0,0)(0, 0) does not lie on CC (since 0+0โˆ’1โ‰ 00 + 0 - 1 \neq 0). So CC is nonsingular at every point.

ExampleJacobian for a surface in Aยณ

Let S=V(x2+y2+z2โˆ’1)โІA3S = V(x^2 + y^2 + z^2 - 1) \subseteq \mathbb{A}^3. Here n=3n = 3, d=2d = 2, so we need rankโกJ(P)=1\operatorname{rank} J(P) = 1. The Jacobian is

J=(2x2y2z).J = \begin{pmatrix} 2x & 2y & 2z \end{pmatrix}.

This has rank 0 only at the origin, which does not lie on SS. Hence SS is smooth (when charโก(k)โ‰ 2\operatorname{char}(k) \neq 2).


Singular vs. nonsingular points

Definition5.3Nonsingular point, singular point

Let YY be a variety of dimension dd. A point PโˆˆYP \in Y is nonsingular (or smooth) if dimโกkTPY=d\dim_k T_P Y = d. Otherwise (dimโกkTPY>d\dim_k T_P Y > d), PP is a singular point of YY.

A variety is nonsingular (or smooth) if every point is nonsingular.

ExampleThe node: yยฒ = xยฒ(x+1)

Let C=V(y2โˆ’x2(x+1))=V(y2โˆ’x3โˆ’x2)โІA2C = V(y^2 - x^2(x+1)) = V(y^2 - x^3 - x^2) \subseteq \mathbb{A}^2. Set f=y2โˆ’x3โˆ’x2f = y^2 - x^3 - x^2. The partial derivatives are

โˆ‚fโˆ‚x=โˆ’3x2โˆ’2x,โˆ‚fโˆ‚y=2y.\frac{\partial f}{\partial x} = -3x^2 - 2x, \qquad \frac{\partial f}{\partial y} = 2y.

At the origin P=(0,0)P = (0, 0): f(0,0)=0f(0,0) = 0, โˆ‚fโˆ‚x(0,0)=0\frac{\partial f}{\partial x}(0,0) = 0, โˆ‚fโˆ‚y(0,0)=0\frac{\partial f}{\partial y}(0,0) = 0. Both partial derivatives vanish, so the origin is a singular point (a node).

The tangent space is T(0,0)C=kerโก(0,0)=k2T_{(0,0)} C = \ker(0, 0) = k^2, which is 22-dimensional -- strictly larger than dimโกC=1\dim C = 1.

At a generic point, say P=(0,โˆ’1)P = (0, -1): we check f(0,โˆ’1)=1โˆ’0โˆ’0=1โ‰ 0f(0, -1) = 1 - 0 - 0 = 1 \neq 0, so this point is not on CC. Try P=(3,6)P = (3, 6): f(3,6)=36โˆ’27โˆ’9=0f(3, 6) = 36 - 27 - 9 = 0 and โˆ‚fโˆ‚x(3,6)=โˆ’27โˆ’6=โˆ’33โ‰ 0\frac{\partial f}{\partial x}(3, 6) = -27 - 6 = -33 \neq 0. So (3,6)(3, 6) is a smooth point.

ExampleThe cusp: yยฒ = xยณ

Let C=V(y2โˆ’x3)โІA2C = V(y^2 - x^3) \subseteq \mathbb{A}^2. Set f=y2โˆ’x3f = y^2 - x^3.

โˆ‚fโˆ‚x=โˆ’3x2,โˆ‚fโˆ‚y=2y.\frac{\partial f}{\partial x} = -3x^2, \qquad \frac{\partial f}{\partial y} = 2y.

At P=(0,0)P = (0, 0): f=0f = 0, โˆ‚fโˆ‚x=0\frac{\partial f}{\partial x} = 0, โˆ‚fโˆ‚y=0\frac{\partial f}{\partial y} = 0. The origin is singular.

At P=(1,1)P = (1, 1): f=0f = 0, โˆ‚fโˆ‚x=โˆ’3โ‰ 0\frac{\partial f}{\partial x} = -3 \neq 0. So (1,1)(1, 1) is smooth. In fact, the origin is the only singular point.

The cusp is "worse" than the node: the local ring OC,0\mathcal{O}_{C, 0} has m/m2\mathfrak{m}/\mathfrak{m}^2 two-dimensional (same as the node), but the tangent cone is different (see below).

ExampleThe tacnode: yยฒ = xโด

Let C=V(y2โˆ’x4)โІA2C = V(y^2 - x^4) \subseteq \mathbb{A}^2. Set f=y2โˆ’x4f = y^2 - x^4.

โˆ‚fโˆ‚x=โˆ’4x3,โˆ‚fโˆ‚y=2y.\frac{\partial f}{\partial x} = -4x^3, \qquad \frac{\partial f}{\partial y} = 2y.

At P=(0,0)P = (0, 0): both vanish, so the origin is singular. The curve factors as (yโˆ’x2)(y+x2)=0(y - x^2)(y + x^2) = 0, giving two smooth branches (the parabolas y=x2y = x^2 and y=โˆ’x2y = -x^2) that are tangent to each other at the origin. This is a tacnode: the two branches share the same tangent line y=0y = 0 (with contact of order 2).

Compare with the node y2=x2(x+1)y^2 = x^2(x+1), where the two branches have distinct tangent directions at the singular point.

ExampleOrdinary multiple point: xยณ + yยณ = 0 and higher

Consider C=V(x3+y3)โІA2C = V(x^3 + y^3) \subseteq \mathbb{A}^2 (over a field where charโก(k)โ‰ 3\operatorname{char}(k) \neq 3). Since x3+y3=(x+y)(x+ฯ‰y)(x+ฯ‰2y)x^3 + y^3 = (x + y)(x + \omega y)(x + \omega^2 y) where ฯ‰\omega is a primitive cube root of unity, the curve is actually reducible -- three lines through the origin.

For a genuinely irreducible example, take C=V(x4+y4+x2y2โˆ’x2โˆ’y2)C = V(x^4 + y^4 + x^2 y^2 - x^2 - y^2). The origin need not be singular here. Instead, consider the curve germ y3=x3(x+1)y^3 = x^3(x+1) at the origin. Set f=y3โˆ’x3(x+1)f = y^3 - x^3(x+1):

โˆ‚fโˆ‚x(0,0)=0,โˆ‚fโˆ‚y(0,0)=0.\frac{\partial f}{\partial x}(0,0) = 0, \qquad \frac{\partial f}{\partial y}(0,0) = 0.

The origin is a singular point. The lowest-degree terms are y3โˆ’x3=(yโˆ’x)(yโˆ’ฯ‰x)(yโˆ’ฯ‰2x)y^3 - x^3 = (y - x)(y - \omega x)(y - \omega^2 x), giving three distinct tangent lines. This is an ordinary triple point (or ordinary 3-fold point): a point of multiplicity 3 with 3 distinct tangent directions.

More generally, a singular point of multiplicity mm with mm distinct tangent directions is called an ordinary mm-fold point.


The singular locus

Theorem5.2The singular locus is closed

Let YY be a variety. The singular locus

Singโก(Y)={PโˆˆYโˆฃPย isย singular}\operatorname{Sing}(Y) = \{ P \in Y \mid P \text{ is singular} \}

is a proper closed subset of YY. In particular, the set of nonsingular points is a dense open subset.

ExampleSingular locus of the nodal cubic

For C=V(y2โˆ’x3โˆ’x2)โІA2C = V(y^2 - x^3 - x^2) \subseteq \mathbb{A}^2, we computed that Singโก(C)\operatorname{Sing}(C) is determined by

f=0,โˆ‚fโˆ‚x=โˆ’3x2โˆ’2x=0,โˆ‚fโˆ‚y=2y=0.f = 0, \quad \frac{\partial f}{\partial x} = -3x^2 - 2x = 0, \quad \frac{\partial f}{\partial y} = 2y = 0.

From the partials: y=0y = 0 and x(3x+2)=0x(3x + 2) = 0, so x=0x = 0 or x=โˆ’2/3x = -2/3. Check on CC: (0,0)(0, 0) gives f=0f = 0 (singular); (โˆ’2/3,0)(-2/3, 0) gives f=โˆ’4/27+4/9=8/27โ‰ 0f = -4/27 + 4/9 = 8/27 \neq 0 (not on CC). Thus Singโก(C)={(0,0)}\operatorname{Sing}(C) = \{(0,0)\}, a single point -- a proper closed subset of CC.

ExampleSingular locus of the Whitney umbrella

Let S=V(x2โˆ’y2z)โІA3S = V(x^2 - y^2 z) \subseteq \mathbb{A}^3 (the Whitney umbrella). Set f=x2โˆ’y2zf = x^2 - y^2 z:

โˆ‚fโˆ‚x=2x,โˆ‚fโˆ‚y=โˆ’2yz,โˆ‚fโˆ‚z=โˆ’y2.\frac{\partial f}{\partial x} = 2x, \quad \frac{\partial f}{\partial y} = -2yz, \quad \frac{\partial f}{\partial z} = -y^2.

The singular locus is {f=0,โ€…โ€Š2x=0,โ€…โ€Šโˆ’2yz=0,โ€…โ€Šโˆ’y2=0}\{f = 0,\; 2x = 0,\; -2yz = 0,\; -y^2 = 0\}, which gives x=0x = 0, y=0y = 0, and then f=0f = 0 is automatic. So Singโก(S)=V(x,y)={(0,0,t)โˆฃtโˆˆk}\operatorname{Sing}(S) = V(x, y) = \{(0, 0, t) \mid t \in k\}, the entire zz-axis.

Here the singular locus is a line sitting inside a surface -- a closed subset of dimension 1 inside a variety of dimension 2.


Smooth varieties

Definition5.4Smooth (nonsingular) variety

A variety YY is smooth (or nonsingular) if Singโก(Y)=โˆ…\operatorname{Sing}(Y) = \varnothing, i.e., every point of YY is nonsingular.

ExampleAffine and projective spaces are smooth

An\mathbb{A}^n is smooth: the local ring at every point is k[x1,โ€ฆ,xn]mk[x_1, \ldots, x_n]_{\mathfrak{m}}, which is a regular local ring of dimension nn, and dimโกTPAn=n=dimโกAn\dim T_P \mathbb{A}^n = n = \dim \mathbb{A}^n.

Pn\mathbb{P}^n is smooth: it is covered by affine charts Uiโ‰…AnU_i \cong \mathbb{A}^n, each of which is smooth. Smoothness is a local property, so Pn\mathbb{P}^n is smooth.

ExampleSmooth quadric hypersurface

Let Q=V(x02+x12+โ‹ฏ+xn2)โІPnQ = V(x_0^2 + x_1^2 + \cdots + x_n^2) \subseteq \mathbb{P}^n (assuming charโก(k)โ‰ 2\operatorname{char}(k) \neq 2). In the affine chart x0=1x_0 = 1, this is V(1+x12+โ‹ฏ+xn2)V(1 + x_1^2 + \cdots + x_n^2), i.e., V(x12+โ‹ฏ+xn2+1)V(x_1^2 + \cdots + x_n^2 + 1). The Jacobian is (2x1,โ€ฆ,2xn)(2x_1, \ldots, 2x_n), which vanishes only at the origin, but 0+โ‹ฏ+0+1โ‰ 00 + \cdots + 0 + 1 \neq 0, so no singular point exists in this chart. By symmetry, the same holds in every chart. So QQ is smooth.

More generally, a quadric V(โˆ‘aijxixj)V(\sum a_{ij} x_i x_j) in Pn\mathbb{P}^n is smooth if and only if the associated symmetric matrix (aij)(a_{ij}) is nonsingular (i.e., has nonzero determinant).

ExampleSmooth cubic surface in Pยณ

The Fermat cubic surface S=V(x03+x13+x23+x33)โІP3S = V(x_0^3 + x_1^3 + x_2^3 + x_3^3) \subseteq \mathbb{P}^3 (with charโก(k)โ‰ 3\operatorname{char}(k) \neq 3). In the chart x0=1x_0 = 1:

f=1+x13+x23+x33,J=(3x12,โ€…โ€Š3x22,โ€…โ€Š3x32).f = 1 + x_1^3 + x_2^3 + x_3^3, \quad J = (3x_1^2,\; 3x_2^2,\; 3x_3^2).

J=0J = 0 requires x1=x2=x3=0x_1 = x_2 = x_3 = 0, but then f=1โ‰ 0f = 1 \neq 0. By symmetry, all charts give the same conclusion: SS is smooth.

A classical result: every smooth cubic surface over C\mathbb{C} contains exactly 27 lines. This is one of the most famous theorems in 19th-century algebraic geometry (Cayley-Salmon theorem, 1849).

ExampleSmooth cubic curve (elliptic curve)

Let E=V(y2โˆ’x3โˆ’axโˆ’b)โІA2E = V(y^2 - x^3 - ax - b) \subseteq \mathbb{A}^2 where 4a3+27b2โ‰ 04a^3 + 27b^2 \neq 0 (the discriminant condition). Set f=y2โˆ’x3โˆ’axโˆ’bf = y^2 - x^3 - ax - b:

โˆ‚fโˆ‚x=โˆ’3x2โˆ’a,โˆ‚fโˆ‚y=2y.\frac{\partial f}{\partial x} = -3x^2 - a, \qquad \frac{\partial f}{\partial y} = 2y.

A singular point requires 2y=02y = 0 (so y=0y = 0) and 3x2+a=03x^2 + a = 0 and x3+ax+b=0x^3 + ax + b = 0. From 3x2=โˆ’a3x^2 = -a, substitute into x3+ax+b=0x^3 + ax + b = 0: one can show this system has a solution iff 4a3+27b2=04a^3 + 27b^2 = 0. So the condition 4a3+27b2โ‰ 04a^3 + 27b^2 \neq 0 is precisely what ensures smoothness.

For instance, y2=x3โˆ’xy^2 = x^3 - x (where a=โˆ’1a = -1, b=0b = 0, discriminant =โˆ’4+0=โˆ’4โ‰ 0= -4 + 0 = -4 \neq 0) is smooth. But y2=x3y^2 = x^3 (where a=0a = 0, b=0b = 0, discriminant =0= 0) has a cusp at the origin.

ExampleGrassmannians are smooth

The Grassmannian G(r,n)\mathbb{G}(r, n) is smooth. This can be seen from its local description: around any point WโˆˆG(r,n)W \in \mathbb{G}(r, n), choose a complementary subspace Wโ€ฒW' of dimension nโˆ’rn - r. Then the open set {UโˆˆG(r,n)โˆฃUโˆฉWโ€ฒ=0}\{U \in \mathbb{G}(r, n) \mid U \cap W' = 0\} is isomorphic to Ar(nโˆ’r)\mathbb{A}^{r(n-r)} (the space of rร—(nโˆ’r)r \times (n-r) matrices, representing UU as the graph of a linear map Wโ†’Wโ€ฒW \to W'). Since Ar(nโˆ’r)\mathbb{A}^{r(n-r)} is smooth and smoothness is local, G(r,n)\mathbb{G}(r, n) is smooth.


The tangent cone

Definition5.5Tangent cone

Let Y=V(f)โІAnY = V(f) \subseteq \mathbb{A}^n be a hypersurface and PโˆˆYP \in Y. Write f=fm+fm+1+โ‹ฏf = f_m + f_{m+1} + \cdots where fif_i is the homogeneous component of degree ii (in the variables xjโˆ’ajx_j - a_j, centered at PP) and fmโ‰ 0f_m \neq 0. The tangent cone to YY at PP is

CPY=V(fm)โІAn,C_P Y = V(f_m) \subseteq \mathbb{A}^n,

the zero set of the leading form fmf_m. The integer mm is the multiplicity of YY at PP, denoted multโกP(Y)\operatorname{mult}_P(Y).

RemarkTangent cone vs. tangent space

If PP is a smooth point (multiplicity 1), then f1=dPff_1 = d_P f is a nonzero linear form, and the tangent cone V(f1)V(f_1) is a hyperplane -- this is the tangent space. For singular points, the tangent cone carries more refined information than the tangent space. The tangent space is always a linear subspace; the tangent cone is a (possibly reducible) cone.

ExampleTangent cone of the node

For C=V(y2โˆ’x3โˆ’x2)C = V(y^2 - x^3 - x^2), the lowest-degree terms at the origin are y2โˆ’x2=(yโˆ’x)(y+x)y^2 - x^2 = (y - x)(y + x). So

C(0,0)C=V(y2โˆ’x2)=V(yโˆ’x)โˆชV(y+x),C_{(0,0)} C = V(y^2 - x^2) = V(y - x) \cup V(y + x),

two distinct lines through the origin. The multiplicity is multโก(0,0)(C)=2\operatorname{mult}_{(0,0)}(C) = 2.

Geometrically, the two lines y=xy = x and y=โˆ’xy = -x are the two tangent directions of the two branches of the curve at the node.

ExampleTangent cone of the cusp

For C=V(y2โˆ’x3)C = V(y^2 - x^3), the lowest-degree terms at the origin are y2y^2 (degree 2). So

C(0,0)C=V(y2),C_{(0,0)} C = V(y^2),

a "double line" along the xx-axis. The multiplicity is multโก(0,0)(C)=2\operatorname{mult}_{(0,0)}(C) = 2.

Although the cusp and the node both have multiplicity 2, their tangent cones differ: the node has two distinct tangent lines, while the cusp has a single tangent line with multiplicity 2. The cusp is "more singular" in this sense.

ExampleTangent cone of the tacnode

For C=V(y2โˆ’x4)C = V(y^2 - x^4), the lowest-degree terms at the origin are y2y^2 (degree 2, since x4x^4 has degree 4). So

C(0,0)C=V(y2),C_{(0,0)} C = V(y^2),

the same double line as the cusp. The multiplicity is 2. But the tacnode and the cusp are genuinely different singularities (the tacnode is analytically equivalent to two smooth branches tangent to each other; the cusp is an irreducible singular germ). Distinguishing them requires looking beyond the tangent cone, e.g., at higher-order jet spaces or the resolution.


Multiplicity of a point

Definition5.6Multiplicity

Let YY be a variety and PโˆˆYP \in Y. The multiplicity multโกP(Y)\operatorname{mult}_P(Y) is defined as follows. If Y=V(f)Y = V(f) is a hypersurface, it is the order of vanishing of ff at PP, i.e., the degree of the leading form in the Taylor expansion at PP.

More generally, if m=mP\mathfrak{m} = \mathfrak{m}_P is the maximal ideal of OY,P\mathcal{O}_{Y,P} and d=dimโกOY,Pd = \dim \mathcal{O}_{Y,P}, the Hilbert-Samuel multiplicity is

e(OY,P)=d!โ‹…limโกnโ†’โˆžโ„“(OY,P/mn)nde(\mathcal{O}_{Y,P}) = d! \cdot \lim_{n \to \infty} \frac{\ell(\mathcal{O}_{Y,P}/\mathfrak{m}^n)}{n^d}

where โ„“\ell denotes the length. For a hypersurface, this agrees with the order of vanishing.

ExampleMultiplicities of plane curve singularities

| Curve | Equation | multโก(0,0)\operatorname{mult}_{(0,0)} | Tangent cone | |---|---|---|---| | Smooth point | y=x2y = x^2 (at origin: not on curve) | -- | -- | | Node | y2=x3+x2y^2 = x^3 + x^2 | 2 | y2โˆ’x2y^2 - x^2: two lines | | Cusp | y2=x3y^2 = x^3 | 2 | y2y^2: double line | | Tacnode | y2=x4y^2 = x^4 | 2 | y2y^2: double line | | Triple point | y3=x3(x+1)y^3 = x^3(x+1) | 3 | y3โˆ’x3y^3 - x^3: three lines | | A4A_4 singularity | y2=x5y^2 = x^5 | 2 | y2y^2: double line |

The multiplicity determines the "order" of the singularity, while the tangent cone captures the tangent geometry. Two singularities can have the same multiplicity and tangent cone (cusp vs. tacnode) but be analytically different.


Regular local rings

Definition5.7Regular local ring

A Noetherian local ring (R,m,k)(R, \mathfrak{m}, k) is regular if

dimโกk(m/m2)=dimโกR\dim_k(\mathfrak{m}/\mathfrak{m}^2) = \dim R

where dimโกR\dim R is the Krull dimension. In general, dimโกk(m/m2)โ‰ฅdimโกR\dim_k(\mathfrak{m}/\mathfrak{m}^2) \geq \dim R (this is a consequence of the Krull principal ideal theorem). Equality is the condition of regularity.

Theorem5.3Characterization of nonsingularity

Let YY be a variety and PโˆˆYP \in Y. The following are equivalent:

  1. PP is a nonsingular point of YY.
  2. dimโกkTPY=dimโกY\dim_k T_P Y = \dim Y.
  3. The local ring OY,P\mathcal{O}_{Y,P} is a regular local ring.
  4. (For Y=V(f1,โ€ฆ,fr)โІAnY = V(f_1, \ldots, f_r) \subseteq \mathbb{A}^n) The Jacobian matrix J(P)J(P) has rank nโˆ’dimโกYn - \dim Y.

Moreover, every regular local ring is an integral domain, a UFD (Auslander-Buchsbaum theorem), and in particular is normal (integrally closed).

ExampleThe local ring at a cusp is not regular

For C=V(y2โˆ’x3)C = V(y^2 - x^3) and P=(0,0)P = (0,0):

OC,P=(k[x,y]/(y2โˆ’x3))(x,y)=k[t2,t3](t2,t3).\mathcal{O}_{C, P} = (k[x, y]/(y^2 - x^3))_{(x, y)} = k[t^2, t^3]_{(t^2, t^3)}.

The maximal ideal m=(t2,t3)\mathfrak{m} = (t^2, t^3) satisfies m/m2=kโ‹…tห‰2โŠ•kโ‹…tห‰3\mathfrak{m}/\mathfrak{m}^2 = k \cdot \bar{t}^2 \oplus k \cdot \bar{t}^3 (2-dimensional), but dimโกOC,P=1\dim \mathcal{O}_{C, P} = 1 (since CC is a curve). Since 2โ‰ 12 \neq 1, the local ring is not regular, confirming that the origin is a singular point.

Furthermore, OC,P\mathcal{O}_{C, P} is not a UFD: the element t2t^2 is irreducible but not prime (since t2โ‹…t2=t4=tโ‹…t3t^2 \cdot t^2 = t^4 = t \cdot t^3, and tโˆ‰k[t2,t3]t \notin k[t^2, t^3]). It is also not integrally closed: t=t3/t2t = t^3/t^2 is integral over k[t2,t3]k[t^2, t^3] but not in it.

ExampleRegular local ring of a smooth curve

For C=V(yโˆ’x2)โІA2C = V(y - x^2) \subseteq \mathbb{A}^2 and P=(0,0)P = (0, 0):

OC,P=(k[x,y]/(yโˆ’x2))(x,y)โ‰…k[x](x).\mathcal{O}_{C, P} = (k[x, y]/(y - x^2))_{(x, y)} \cong k[x]_{(x)}.

The maximal ideal is (x)(x), which is principal: m/m2โ‰…k\mathfrak{m}/\mathfrak{m}^2 \cong k (1-dimensional), and dimโกOC,P=1\dim \mathcal{O}_{C, P} = 1. Since 1=11 = 1, this is a regular local ring -- a discrete valuation ring (DVR). Every smooth curve has DVRs as its local rings.


Normalization and resolution of singularities

RemarkNormalization

The normalization of a variety YY is the unique (up to isomorphism) normal variety Y~\tilde{Y} with a finite birational morphism ฮฝ:Y~โ†’Y\nu: \tilde{Y} \to Y. For curves, normalization resolves all singularities:

  • The normalization of the cuspidal cubic V(y2โˆ’x3)V(y^2 - x^3) is A1\mathbb{A}^1, via tโ†ฆ(t2,t3)t \mapsto (t^2, t^3).
  • The normalization of the nodal cubic V(y2โˆ’x2(x+1))V(y^2 - x^2(x+1)) is A1\mathbb{A}^1, via tโ†ฆ(t2โˆ’1,t(t2โˆ’1))t \mapsto (t^2 - 1, t(t^2 - 1)).

In both cases, the normalization "unravels" the singularity by pulling apart branches (node) or "smoothing" the parametrization (cusp). For higher-dimensional varieties, normalization resolves singularities in codimension 1 but may leave higher-codimension singularities.

RemarkResolution of singularities

A resolution of singularities of a variety YY is a proper birational morphism ฯ€:Y~โ†’Y\pi: \tilde{Y} \to Y where Y~\tilde{Y} is smooth. Hironaka's celebrated theorem (1964) guarantees that resolutions exist for varieties over fields of characteristic zero.

Resolution is typically achieved by blowing up along the singular locus (or subvarieties thereof), repeatedly, until the result is smooth.


Blowing up singularities

ExampleBlowing up the node

Consider C=V(y2โˆ’x2โˆ’x3)โІA2C = V(y^2 - x^2 - x^3) \subseteq \mathbb{A}^2 with a node at the origin. The blowup of A2\mathbb{A}^2 at the origin replaces (0,0)(0,0) by a P1\mathbb{P}^1 (the exceptional divisor EE). In local coordinates:

Chart 1 (y=txy = tx): Substituting y=txy = tx into f=y2โˆ’x2โˆ’x3f = y^2 - x^2 - x^3:

t2x2โˆ’x2โˆ’x3=x2(t2โˆ’1โˆ’x).t^2 x^2 - x^2 - x^3 = x^2(t^2 - 1 - x).

The factor x2x^2 corresponds to the exceptional divisor (with multiplicity 2). The strict transform is C~1=V(t2โˆ’1โˆ’x)\tilde{C}_1 = V(t^2 - 1 - x), a smooth curve. On EE (where x=0x = 0), the strict transform meets EE at t2=1t^2 = 1, i.e., t=ยฑ1t = \pm 1. These are two distinct points, corresponding to the two tangent directions of the node.

Chart 2 (x=syx = sy): Substituting x=syx = sy into ff:

y2โˆ’s2y2โˆ’s3y3=y2(1โˆ’s2โˆ’s3y).y^2 - s^2 y^2 - s^3 y^3 = y^2(1 - s^2 - s^3 y).

The strict transform is V(1โˆ’s2โˆ’s3y)V(1 - s^2 - s^3 y), also smooth.

The blowup resolves the node in a single step: the strict transform C~\tilde{C} is smooth, and ฯ€:C~โ†’C\pi: \tilde{C} \to C is an isomorphism away from the origin. Above the origin, C~\tilde{C} has two points (the two branches are separated).

ExampleBlowing up the cusp

Consider C=V(y2โˆ’x3)โІA2C = V(y^2 - x^3) \subseteq \mathbb{A}^2 with a cusp at the origin.

First blowup (chart y=txy = tx): Substitute into f=y2โˆ’x3f = y^2 - x^3:

t2x2โˆ’x3=x2(t2โˆ’x).t^2 x^2 - x^3 = x^2(t^2 - x).

The strict transform C~1=V(t2โˆ’x)\tilde{C}_1 = V(t^2 - x) is smooth (it is a parabola in the (x,t)(x, t)-plane). On the exceptional divisor (x=0x = 0): t2=0t^2 = 0, so t=0t = 0. The strict transform meets EE at a single point with multiplicity 2 (tangency).

But wait -- C~1=V(t2โˆ’x)\tilde{C}_1 = V(t^2 - x) is actually smooth. The intersection with EE is transverse after a coordinate change: set u=tu = t, then x=u2x = u^2, and C~1โ‰…A1\tilde{C}_1 \cong \mathbb{A}^1 via uโ†ฆ(u2,u)u \mapsto (u^2, u). The cusp is resolved after a single blowup.

However, note that the strict transform meets the exceptional divisor with contact order 2, not transversally. If we want normal crossings (for the purpose of embedded resolution), we would need a second blowup. For the abstract resolution of CC alone, one blowup suffices.

ExampleComparing node and cusp blowups

| Singularity | Blowups needed | Strict transform meets EE at | Behavior | |---|---|---|---| | Node y2=x2(x+1)y^2 = x^2(x+1) | 1 | 2 distinct points | Two branches separated | | Cusp y2=x3y^2 = x^3 | 1 | 1 point (tangent to EE) | Branch smoothed | | Tacnode y2=x4y^2 = x^4 | 2 | After 1st: cusp; after 2nd: smooth | Two tangent branches separated | | A4A_4: y2=x5y^2 = x^5 | 2 | Successive blowups resolve | Higher-order contact resolved iteratively |


Nonsingularity in the projective setting

ExampleSmoothness check for projective hypersurfaces

For a projective hypersurface Y=V(F)โІPnY = V(F) \subseteq \mathbb{P}^n with FF homogeneous of degree dd, the singular points are

Singโก(Y)=Vโ€‰โฃ(F,โ€…โ€Šโˆ‚Fโˆ‚x0,โ€…โ€Šโ€ฆ,โ€…โ€Šโˆ‚Fโˆ‚xn)โІPn.\operatorname{Sing}(Y) = V\!\left(F,\; \frac{\partial F}{\partial x_0},\; \ldots,\; \frac{\partial F}{\partial x_n}\right) \subseteq \mathbb{P}^n.

By Euler's formula for homogeneous polynomials (โˆ‘xiโˆ‚Fโˆ‚xi=dโ‹…F\sum x_i \frac{\partial F}{\partial x_i} = d \cdot F), the condition F=0F = 0 is redundant when charโก(k)โˆคd\operatorname{char}(k) \nmid d: it follows from the vanishing of all partials at a point of Pn\mathbb{P}^n.

Concretely: the Klein quartic V(x3y+y3z+z3x)โІP2V(x^3 y + y^3 z + z^3 x) \subseteq \mathbb{P}^2 (over C\mathbb{C}) is smooth. Its partials are:

โˆ‚Fโˆ‚x=3x2y+z3,โˆ‚Fโˆ‚y=x3+3y2z,โˆ‚Fโˆ‚z=y3+3z2x.\frac{\partial F}{\partial x} = 3x^2 y + z^3, \quad \frac{\partial F}{\partial y} = x^3 + 3y^2 z, \quad \frac{\partial F}{\partial z} = y^3 + 3z^2 x.

One can verify (e.g., by Grobner basis or direct computation) that the system F=โˆ‚Fโˆ‚x=โˆ‚Fโˆ‚y=โˆ‚Fโˆ‚z=0F = \frac{\partial F}{\partial x} = \frac{\partial F}{\partial y} = \frac{\partial F}{\partial z} = 0 has no solution in P2\mathbb{P}^2. The Klein quartic is a smooth curve of genus 3, and it has the largest automorphism group (168 elements) among genus-3 curves.

ExampleGeneric hypersurfaces are smooth

For "most" choices of coefficients, a degree-dd hypersurface in Pn\mathbb{P}^n is smooth. More precisely, the locus of singular hypersurfaces in the parameter space PN\mathbb{P}^N (where N=(n+dd)โˆ’1N = \binom{n+d}{d} - 1) is a proper closed subset -- the discriminant hypersurface ฮ”\Delta.

For plane curves (n=2n = 2):

  • Degree 1 (lines): always smooth. ฮ”=โˆ…\Delta = \varnothing.
  • Degree 2 (conics): singular iff the 3ร—33 \times 3 matrix has detโก=0\det = 0. ฮ”\Delta is a degree-3 hypersurface in P5\mathbb{P}^5.
  • Degree 3 (cubics): the discriminant ฮ”โІP9\Delta \subseteq \mathbb{P}^9 has degree 12.

This is a manifestation of Bertini's theorem: a "general" member of a linear system is smooth.


Summary: a taxonomy of plane curve singularities

NameEquationMult.Tangent coneฮด\delta-invariantBlowups
A1A_1 (node)y2=x2+x3y^2 = x^2 + x^32Two distinct lines11
A2A_2 (cusp)y2=x3y^2 = x^32Double line11
A3A_3 (tacnode)y2=x4y^2 = x^42Double line22
A4A_4y2=x5y^2 = x^52Double line22
D4D_4 (ordinary triple pt)x3+y3=0x^3 + y^3 = 03Three distinct lines31
E6E_6y3=x4y^3 = x^43Triple line33

The ADE classification of simple curve singularities (du Val singularities for surfaces) connects algebraic geometry to Lie theory, representation theory, and mathematical physics via the McKay correspondence.

RemarkLooking ahead

Nonsingularity is preserved by many natural operations: smooth subvarieties of smooth varieties are smooth, products of smooth varieties are smooth, and a general hyperplane section of a smooth projective variety is smooth (Bertini's theorem). The theory of nonsingular varieties is the foundation for intersection theory, differential forms, sheaf cohomology, and the Riemann-Roch theorem in later chapters.

For a more general framework that handles singularities on equal footing -- including nilpotent elements, non-reduced structures, and arithmetic geometry -- one passes to the language of schemes and the notion of regular schemes.