Zariski Tangent Space¶
The residue-field dual of a point's local maximal ideal modulo its square, recording first-order directions of an algebraic variety or scheme.
Core Idea¶
At a point \(x\) of an algebraic variety or scheme, let \(\mathcal O_{X,x}\) be its local ring, \(\mathfrak m_x\) the maximal ideal of functions vanishing at \(x\), and \(\kappa(x)=\mathcal O_{X,x}/\mathfrak m_x\) the residue field. The quotient \(\mathfrak m_x/\mathfrak m_x^2\) keeps the first-order part of those vanishing functions: products of two such functions are zero in the quotient. It is the cotangent space. Its \(\kappa(x)\)-linear dual \(T_xX=\operatorname{Hom}_{\kappa(x)}(\mathfrak m_x/\mathfrak m_x^2,\kappa(x))\) is the pointwise Zariski tangent space under the intrinsic local-ring convention.[1]
One can also define a tangent of a morphism \(X\to S\) relative to a base using lifts of a point to the dual numbers \(\kappa(x)[\epsilon]/(\epsilon^2)\), equivalently duals of relative Kähler differentials. Stacks gives the simple maximal-ideal quotient identification under equal or separably algebraic residue-field hypotheses; it is not an unconditional identification over every field extension. At a \(k\)-rational point of a \(k\)-scheme, the familiar quotient, dual-number and Jacobian calculations align.[2]
Structural Signature¶
Sig role-phrases: chosen algebraic-geometric point → local ring and residue field → maximal ideal of vanishing functions → quotient by its square for first-order cotangent data → residue-field dual for tangent directions; base-relative convention must be declared when used.
- Point and local ring. The object is attached to \(x\), not merely to the global equation or whole variety. \(\mathcal O_{X,x}\) records functions defined near \(x\); absent that local algebra, there is no specified first-order neighborhood to test.[1]
- Vanishing ideal. \(\mathfrak m_x\) collects local functions whose value at \(x\) is zero. Its square records products that vanish to at least second order. Quotienting by \(\mathfrak m_x^2\) is the decisive first-order truncation.[1]
- Residue-field cotangent quotient. \(\mathfrak m_x/\mathfrak m_x^2\) is a vector space over \(\kappa(x)\). Calling this quotient itself the tangent space swaps the tangent and cotangent roles.[1]
- Dual directions. A tangent vector is a \(\kappa(x)\)-linear functional on the cotangent quotient. In an affine presentation at a rational point, it can be computed as a solution of linearized defining equations, but its identity is not tied to that particular embedding.[1]
- Relative-base qualification. For \(X/S\), Stacks defines \(T_{X/S,x}\) from dual-number lifts and \(\Omega_{X/S}\); a quotient expression needs its stated residue-field assumptions. This is a convention/compatibility test, not permission to replace the pointwise object with any Euclidean derivative.[2]
What It Is Not¶
The Zariski tangent space is not a tangent cone. At the cusp \(y^2=x^3\) at the origin, there is no nonzero linear term in the equation, so the Zariski tangent space is the entire two-dimensional ambient plane. The tangent cone, constructed from lowest nonzero homogeneous terms, is the doubled line \(y^2=0\). Both are local, but they retain different orders of information.[1]
It is not a smooth manifold's assembled tangent bundle; the construction still exists at singular scheme points, where its dimension can exceed local dimension. It is not the cotangent sheaf \(\Omega_{X/S}\): the latter is a sheaf of relative differentials, while the pointwise tangent is a dual fibre under specified conventions. Nor do dual numbers themselves equal the tangent: they provide a way to probe first-order directions through maps that reduce to \(x\).[2][1]
Scope of Application¶
For affine equations over a stated field, the Jacobian matrix evaluated at a rational point computes tangent directions by keeping the linear terms. Gallier and Shatz show the characteristic-zero cusp \(y^2=x^3\) at \((0,0)\): both partial derivatives vanish there, producing two independent first-order directions even though the curve's local dimension is one. This dimension gap is a local singularity diagnostic, not a claim that the cusp has two independent smooth branches.[1]
The dual-number interpretation serves a different use: at zero of the affine line \(\mathbb A^1_k\), a lift from \(\operatorname{Spec}k[\epsilon]/(\epsilon^2)\) is specified by \(x\mapsto a\epsilon\), with \(a\in k\). This one parameter is its one-dimensional tangent direction. For a general morphism of schemes, Stacks defines the relative tangent by analogous lifts and relative differentials; the base and residue-field hypotheses must remain visible when transferring this elementary calculation.[2]
Clarity¶
The phrase “first-order” refers to the quotient by \(\mathfrak m_x^2\) or to \(\epsilon^2=0\). It is not a promise that every tangent vector extends to an actual curve or higher-order deformation. A direction may pass the linear equations and fail nonlinear or obstruction conditions later. Conversely, an algebraic singularity can have a large tangent space precisely because first-order equations lose higher-order constraints.[1][2]
For a Noetherian local ring, \(\dim\mathcal O_{X,x}\leq\dim_{\kappa(x)}\mathfrak m_x/\mathfrak m_x^2\); equality characterizes a regular local ring. One may relate this to smoothness over a field under additional assumptions. Stacks proves a clean equivalence for finite-type algebras over an algebraically closed field, so the seed's unconditional statement “singular exactly when tangent dimension is larger” must be read with an explicit regularity/smoothness convention. Inseparable-field examples require more care.[3][4]
Manages Complexity¶
The quotient compresses infinitely many local functions and polynomial terms into a finite-dimensional first-order vector space when the local algebra is finite type. That lets one test a difficult local question with linear algebra: find the Jacobian kernel or compute \(\mathfrak m/\mathfrak m^2\), compare dimensions, and identify possible infinitesimal directions. It preserves enough information to detect a cusp's excess tangent dimension while discarding higher-order curve shape.[1][3]
This compression is deliberately lossy. A cusp and other singular plane curves may have the same full two-dimensional tangent space at a point yet different tangent cones or later local behavior. A local tangent vector is a candidate first-order motion, not a whole deformation family. Applying the quick Jacobian picture beyond a rational point without checking the relative/base convention can introduce a second, subtler loss: it may compute a different tangent notion.[1][2]
Abstract Reasoning¶
The multiplication map sends \(\mathfrak m_x\times\mathfrak m_x\) into \(\mathfrak m_x^2\), so the quotient forgets quadratic and higher products. A functional \(v:\mathfrak m_x/\mathfrak m_x^2\to\kappa(x)\) assigns a first-order value to each vanishing function, compatibly with linear combinations. For a \(k\)-rational affine point, write a defining equation as \(f(p+h)=f(p)+L_p(h)+\) terms of order at least two. Its allowed tangent directions satisfy \(L_p(h)=0\), equivalently the Jacobian linear equations.[1][2]
At the cusp \(f=y^2-x^3\), \(L_{(0,0)}=0\); every \((u,v)\in k^2\) passes the first-order equation, yielding \(\dim T=2\). For \(\mathbb A^1_k\) at zero, \(\mathfrak m=(x)\) and \(\mathfrak m/\mathfrak m^2\) has basis \([x]\), so \(T\) has dimension one. The map \(x\mapsto a\epsilon\) in the dual-number algebra expresses the same one-dimensional direction when the stated rational-point convention applies.[1][2]
Knowledge Transfer¶
At another algebraic-geometric point, first identify whether the question is about the intrinsic local tangent or a tangent relative to \(S\). Then determine \(\mathcal O_{X,x}\), \(\mathfrak m_x\), \(\kappa(x)\) and \(\mathfrak m_x/\mathfrak m_x^2\); take its dual, or use the corresponding relative-differential/dual-number construction. In coordinates, compute linearized equations only after documenting the field and point. Comparing tangent and Krull dimensions for regularity is a separate qualified step, not part of the tangent definition.[2][3]
Transfer the local-first-order roles from a singular cusp to a smooth affine line without transferring their dimensions or geometries. A moduli problem may use the same dual-number Idea, but the seed's proposed \(H^1\) of a tangent sheaf is a separate deformation-theory claim and was not accepted as a universal instance here. The new setting must supply its own representing object and obstruction analysis.[2]
Examples¶
Cuspidal plane curve. Over a characteristic-zero field, take \(X=V(y^2-x^3)\) and \(p=(0,0)\). Mapped back: local algebra = \(k[x,y]/(y^2-x^3)\) localized at \((x,y)\); vanishing ideal = \(\mathfrak m=(x,y)\); first-order quotient = classes \([x],[y]\) survive because the defining relation has no linear term; tangent dual = \(k^2\). The one-dimensional local curve has two-dimensional Zariski tangent, a regularity failure under the stated finite-type conditions. Its doubled-line tangent cone is not this two-dimensional space.[1][3]
Affine line via dual numbers. Let \(X=\mathbb A^1_k\) and \(p=0\) rational. Mapped back: local algebra = \(k[x]_{(x)}\); \(\mathfrak m/\mathfrak m^2\) is generated by \([x]\); tangent dual has one basis direction. A lift over \(k[\epsilon]/(\epsilon^2)\) sends \(x\) to \(a\epsilon\), and \(a\in k\) selects the tangent vector. This is a transparent application of Stacks' dual-number definition, not a claim that every scheme tangent has this coordinate form.[2]
Negative boundary. The tangent cone of the cusp remembers the quadratic initial relation \(y^2=0\), while \(T_pX\) discards it in first order. Calling these identical would wrongly turn a doubled line into the full tangent plane, or vice versa.[1]
Structural Tensions¶
Intrinsic quotient versus Jacobian chart. \(\mathfrak m/\mathfrak m^2\) is invariantly local, but can be opaque to compute. A Jacobian in selected coordinates is efficient, yet can make an ambient kernel appear to depend on the embedding or hide residue-field hypotheses. Using only the chart can misstate the invariant; refusing it misses a sharp cusp diagnostic. Diagnostic: What local-ring quotient does this coordinate kernel actually represent?[1][2]
First-order directions versus higher-order realization. Keeping every dual-number direction faithfully describes the tangent space; demanding a true curve or unobstructed deformation asks a stronger question. Declaring every direction realizable overclaims; rejecting tangent vectors until integration is proved discards useful local information. Diagnostic: Is the conclusion about order \(\epsilon\) only, or about an extension beyond \(\epsilon^2=0\)?[2]
Absolute local versus relative-to-base tangent. The compact dual of \(\mathfrak m/\mathfrak m^2\) illuminates pointwise geometry. Relative differentials make base-fixed motion explicit, but add notation and residue-field conditions. Applying the first formula universally can give the wrong object; overcomplicating a rational point obscures an easy computation. Diagnostic: What is the morphism to the base, and what permits the quotient identification?[2]
Autonomous tangent identity versus generic linearization. The construction has a portable “discard higher order, inspect a linear remainder” skeleton. Reducing it to any linear approximation loses local rings, scheme points and residue-field duality; refusing the skeleton hides why the cusp and affine-line computations compare. Diagnostic: Does the alleged transfer still have a maximal-ideal quotient and its dual?[1]
Structural–Framed Character¶
Evaluative weight. “Large” tangent dimension can diagnose nonregularity under specified hypotheses, but no dimension is intrinsically good or bad. The object exists before a researcher uses it as a quality test.
Human-practice dependence. Mathematicians choose a presentation and base convention. Once \(X\), \(x\) and the convention are fixed, \(\mathfrak m_x/\mathfrak m_x^2\) and its dual are determined, independent of one observer's picture of a tangent line.
Institutional origin. Algebraic-geometry notation and pedagogical choices have a history, but a university text or theorem registry does not create the first-order local relation. The only discretionary component here is which object and base are under study.
Vocabulary travel. “Tangent” literally travels among algebraic varieties and schemes through the local or relative construction. Smooth-manifold and numerical tangent language can guide intuition, but they do not automatically supply \(\mathfrak m/\mathfrak m^2\) at a singular scheme point.
Import versus recognition. Recognize a new instance by finding its point, local algebra, residue field, first-order quotient and dual, or the correctly qualified relative lift. Importing a drawn slope from a familiar plane curve without those roles is only analogy.
Its character: a formal algebraic-geometric first-order object, with a portable linearization intuition and indispensable local-ring, residue-field and base-convention constraints.
Structural Core vs. Domain Accent¶
Portable skeleton. Live Truncation names the boundary-retain-discard idea: here the degree-two boundary removes products in \(\mathfrak m_x^2\) and retains first-order classes. That prime explains the quotient's portable cutoff aspect, not the subsequent residue-field dual or the complete tangent identity; no immediate typed edge is asserted. A more exact cross-domain prime for first-order local probing remains a future-prime question.
Domain-bound residual. The maximal ideal's square, quotient over \(\kappa(x)\), duality and relative dual-number conditions are the mathematics that make the Zariski identity exact. Without them, “linearization” could mean a calculus derivative, a numerical approximation or a design simplification rather than tangent directions at a scheme point.[2][1]
Why not prime. The cusp and affine-line cases share these roles but both live within algebraic geometry. Their different outputs do not show literal transfer of the whole local-ring construction to unrelated substrates. Promoting the name to a prime on the basis of “tangents are everywhere” would conflate distinct tangent notions.
Instantiates / Related Primes¶
This workspace stages approved unparented placement. Live Tangent bundle assembles tangent spaces of a smooth manifold, while the present identity is a pointwise algebraic construction that remains meaningful at singular scheme points. Live Cotangent sheaf and Dual number are mathematically related—the former helps express relative tangent and the latter probes it—but neither is a strict genus of the dual \(T_xX\). Live Truncation captures one cutoff aspect without supplying an immediate genus of the tangent object. No canonical link is asserted.[2][1]
Neighborhood in Abstraction Space¶
Zariski Tangent Space sits in a sparse region of the domain-specific corpus (64th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Sheaves & Birational Geometry (22 abstractions)
Nearest neighbors
- Algebraic Variety — 0.86
- Ringed Space — 0.85
- Sheaf of Modules — 0.85
- Local Tate Duality — 0.85
- Picard Group — 0.84
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
Cotangent space: \(\mathfrak m_x/\mathfrak m_x^2\), before taking the dual. Tangent cone: retains the lowest nonzero homogeneous equations, which can be quadratic at a cusp. Smooth tangent bundle: varies smooth tangent spaces over a manifold and has different scope. Relative tangent: uses a declared morphism to a base and may need \(\Omega_{X/S}\) rather than the simple quotient. Regular point: an additional local-ring dimension condition, not the definition of the tangent object.[1][2][3]
References¶
[1] Jean Gallier and Stephen S. Shatz, Algebraic Geometry, University of Pennsylvania work-in-progress manuscript dated 15 June 2016, §2.2, especially printed pp. 86–89 (Jacobian and characteristic-zero cusp Example 2.3) and pp. 98–99, including Definition 2.10 on the cotangent quotient and dual. Its unfinished status is noted by the authors. https://www.cis.upenn.edu/~jean/algeoms.pdf registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t
[2] The Stacks Project, §33.16 “Tangent spaces,” Definitions 33.16.1 and 33.16.3; Lemmas 33.16.4–33.16.5 (dual-number lifts, relative differentials, and residue-field qualifications). https://stacks.math.columbia.edu/tag/0B28 registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q
[3] The Stacks Project, §10.60 “Dimension,” discussion preceding Definition 10.60.10 (Noetherian local-ring inequality and regularity equality for \(\dim_{\kappa}\mathfrak m/\mathfrak m^2\)). https://stacks.math.columbia.edu/tag/00KD registry ↩a ↩b ↩c ↩d ↩e
[4] The Stacks Project, Lemma 10.140.2 (finite-type algebra over an algebraically closed field: regular local and smooth at maximal ideal equivalence). https://stacks.math.columbia.edu/tag/00TS registry ↩