Stable rank-2 bundles on projective space and Barth's theorem

Intuition Beginner

On the projective line there is essentially one kind of vector bundle: every one splits into a sum of line bundles, so nothing interesting hides there. Move up to projective 3-space and the story changes. There are vector bundles of rank 2 that refuse to split into two line bundles. These are the simplest genuinely new objects, and the question of this unit is: how many are there, and how do we tell them apart?

The answer is that two integers do almost all of the work. A rank-2 bundle carries two numbers called Chern classes, written $c_1$ and $c_2$. The first records a kind of overall twist, the second records a finer curvature-like quantity. Once you fix these two numbers, the bundles that satisfy a balance condition form a single well-organised family.

Visual Beginner

Picture all rank-2 bundles laid out in a grid, one cell for each pair of Chern numbers. Most cells are empty; only certain pairs occur. Inside each occupied cell sits a smooth family, the moduli space, and one famous real slice of it holds the instanton bundles from physics.

The balance condition is the key idea. A bundle is called stable when every smaller piece sitting inside it is, in a precise averaged sense, less twisted than the whole. Stable bundles are the well-behaved ones: they do not break apart, and they assemble into clean families rather than tangled heaps.

Worked example Beginner

The simplest rank-2 bundle on projective 3-space that does not split is the null-correlation bundle. Build it like this. Pick a fixed pattern that pairs up directions in space with one another, a sort of swirl with no preferred axis. This pairing carves out a rank-2 sub-bundle of the tangent directions, and that sub-bundle is the null-correlation bundle.

Its Chern numbers are $c_1=0$ and $c_2=1$. The pair $(0,1)$ is the smallest occupied cell in the grid, and the bundle living there is unique once you account for the symmetries of space. This single bundle is the seed example for everything that follows, and it is also the first instanton bundle.

Check your understanding Beginner

Formal definition Intermediate+

Let $E$ be a torsion-free coherent sheaf 04.06.02 of rank $r$ on ${\mathbb{P}}^n$ over an algebraically closed field. The degree of $E$ is $\mathrm{deg}E=c_1(E)\cdot H^{n-1}$, the intersection of its first Chern class with a hyperplane, and the slope is

$$\mu (E):=\frac{\mathrm{deg}E}{\mathrm{rk}E}.$$

Definition (Mumford-Takemoto stability). The sheaf $E$ is slope-stable (resp. slope-semistable) if for every coherent subsheaf $\mathcal{F}\subset E$ with $0<\mathrm{rk}\mathcal{F}<\mathrm{rk}E$,

$$\mu (\mathcal{F})<\mu (E)(\text{resp.}\le ).$$

For a rank-2 bundle $E$ on ${\mathbb{P}}^n$ it suffices to test rank-1 subsheaves, and after saturation these are line bundles $\mathcal{O}(d)\hookrightarrow E$. Stability then reads: $E$ has no section twisting up to a line sub-bundle of slope $\ge \mu (E)$.

Normalisation. Tensoring $E$ by $\mathcal{O}(m)$ shifts $c_1(E)$ by $2m$ without changing stability, so one normalises $c_1\in \{0,-1\}$. A normalised rank-2 bundle $E$ has $H^0(E)=0$ exactly when it is stable (for $c_1=0$) or semistable, so the vanishing of sections is the practical stability test. The pair

$$(c_1,c_2)\in \{0,-1\}\times \mathbb{Z}$$

is the basic discrete invariant: $c_1$ is the normalised determinant class and $c_2$ measures the second-order twist.

The moduli space. Fix $(c_1,c_2)$. The set of isomorphism classes of stable rank-2 bundles with those Chern classes carries the structure of a quasi-projective scheme $M_{{\mathbb{P}}^n}(c_1,c_2)$, the moduli space of stable rank-2 bundles. On ${\mathbb{P}}^3$ it is described, following 04.07.03, by monads: a stable rank-2 bundle is the cohomology bundle $\ker \beta /\mathrm{im}\alpha$ of a complex

$$\mathcal{O}(-1{)}^a\stackrel{\alpha }{\to }{\mathcal{O}}^b\stackrel{\beta }{\to }\mathcal{O}(1{)}^a,\beta \alpha =0,$$

with $\alpha$ a bundle injection, $\beta$ surjective. The integers $a,b$ are read off from $c_2$, and the moduli space is cut out of the linear-algebra data $(\alpha ,\beta )$ modulo the natural symmetry group. The instanton moduli space of 03.07.10 is the real, symplectic-quotient slice of this: an instanton bundle is a $c_1=0$ monad bundle carrying a quaternionic (ADHM) reality structure.

The Serre construction. Stable rank-2 bundles are produced from codimension-2 subvarieties. Given a local complete intersection curve $Y\subset {\mathbb{P}}^n$ of codimension 2 with ${\omega }_Y\cong {\omega }_{{\mathbb{P}}^n}(\ell ){|}_Y$ for some $\ell$, Serre duality 04.08.03 supplies a class in ${\mathrm{Ext}}^1({\mathcal{I}}_Y(\ell ),\mathcal{O})$, hence an extension

$$0\to \mathcal{O}\to E\to {\mathcal{I}}_Y(\ell )\to 0$$

whose middle term $E$ is a rank-2 bundle precisely when the extension class generates the relevant local groups (the Cayley-Bacharach condition). The zero locus of a section of $E$ recovers $Y$. This is the dictionary "rank-2 bundles $\leftrightarrow$ codimension-2 cycles."

Key theorem with proof Intermediate+

Theorem (Barth's restriction theorem, 1977). Let $E$ be a slope-stable rank-2 bundle on ${\mathbb{P}}^n$, $n\ge 2$, with $c_1\in \{0,-1\}$. Then for a general line $L\subset {\mathbb{P}}^n$ the restriction $E{|}_L$ is balanced, i.e.

$$E{|}_L\cong {\mathcal{O}}_L(d)\oplus {\mathcal{O}}_L(d^{\prime })\text{with }|d-d^{\prime }|\le 1,$$

and more generally the restriction of $E$ to a general hyperplane ${\mathbb{P}}^{n-1}$ remains semistable.

Proof. Every bundle on a line $L\cong {\mathbb{P}}^1$ splits as ${\mathcal{O}}_L(d)\oplus {\mathcal{O}}_L(d^{\prime })$ by Grothendieck's splitting theorem, with $d+d^{\prime }=c_1$ fixed. Lines on which the gap $|d-d^{\prime }|$ exceeds $1$ are the jumping lines; call their union $S(E)$. Suppose, for contradiction, that every line jumps, so $E{|}_L\cong {\mathcal{O}}_L(d)\oplus {\mathcal{O}}_L(d^{\prime })$ with $d\ge d^{\prime }+2$ for all $L$. The sub-line-bundle ${\mathcal{O}}_L(d)\subset E{|}_L$ of top degree is then defined on every line and varies algebraically; it glues to a coherent subsheaf $\mathcal{F}\subset E$ whose restriction to a general line has degree $d>c_1/2=\mu (E)$. After saturating, $\mathcal{F}$ is a rank-1 reflexive subsheaf, hence a line bundle $\mathcal{O}(e)$ with $e\ge \mu (E)$, contradicting the stability inequality $\mu (\mathcal{F})<\mu (E)$.

Therefore the generic line cannot jump: $S(E)$ is a proper closed subset of the variety of lines, and for $L\notin S(E)$ the restriction is balanced. The hyperplane statement follows by the same monodromy argument applied to the restriction map $H^0(E(k){|}_H)\to H^1(E(k-1))$: stability forces the relevant intermediate cohomology to be concentrated, so a general $H$ inherits semistability. $\square$

Bridge. This restriction theorem builds toward the monad classification of 04.07.03: the foundational reason a stable bundle is recoverable from linear algebra is that its restriction to a general line is balanced, so the line-by-line splitting data is constant in codimension 1 and the jumping lines record exactly the defect. The central insight is that stability is the geometric mechanism converting a global sheaf into the intermediate-cohomology bookkeeping a monad encodes. This is exactly the algebraic input that the ADHM construction of 03.07.10 consumes, and it generalises Grothendieck's splitting on ${\mathbb{P}}^1$ to a generic-splitting statement on ${\mathbb{P}}^n$; putting these together, Barth's theorem appears again in the proof that the instanton moduli space is the real slice of $M_{{\mathbb{P}}^3}(0,c_2)$.

Exercises Intermediate+

Advanced results Master

The spectrum of a stable rank-2 bundle (Barth-Elencwajg). To a stable rank-2 bundle $E$ on ${\mathbb{P}}^3$ with $c_1=0$ one attaches its spectrum $k_E=(k_1\le k_2\le \cdots \le k_{c_2})$, a sequence of $c_2$ integers extracted from the restriction of $E$ to lines and the intermediate cohomology $H^1(E(j))$. The spectrum satisfies a symmetry $k_i=-k_{c_2+1-i}$ (a reflection forced by Serre duality 04.08.03) and a connectedness constraint: if an integer $k$ appears with $|k|\ge 2$ then $k-\mathrm{sign}(k)$ also appears. The dimensions of the cohomology groups $H^1(E(j))$ are recovered from the spectrum, so it is a near-complete cohomological invariant refining $(c_1,c_2)$. Bundles with the constant spectrum $(0,0,\dots ,0)$ are exactly the ones with $H^1(E(-1))=0$, the mathematical instanton bundles.

Barth's bounds. Stability is not vacuous: for a stable rank-2 bundle on ${\mathbb{P}}^n$ with $c_1=0$, Barth's discriminant inequality forces $c_2\ge 1$, and the symmetry of the spectrum forces $c_2\ge 0$ with equality only for the split bundle (which is not stable). More refined bounds — the Bogomolov inequality $c_1^2\le 4c_2$ in the rank-2 normalisation $c_1^2-4c_2\le 0$ — constrain which cells $(c_1,c_2)$ are occupied at all. On ${\mathbb{P}}^2$ every semistable bundle is a monad cohomology bundle and the moduli space $M_{{\mathbb{P}}^2}(0,c_2)$ is irreducible of dimension $4c_2-3$; on ${\mathbb{P}}^3$ the analogous count is $8c_2-3$, matching the instanton-moduli dimension of 03.07.10.

Barth's theorem on the moduli of rank-2 bundles on ${\mathbb{P}}^3$. Barth's monad normal form (refined with Hulek) states that every stable rank-2 bundle on ${\mathbb{P}}^3$ with $c_1=0$ arises as the cohomology of a self-dual monad

$$\mathcal{O}(-1{)}^{c_2}\stackrel{\alpha }{\to }{\mathcal{O}}^{2c_2+2}\stackrel{{\alpha }^{\ast }}{\to }\mathcal{O}(1{)}^{c_2},$$

where the second map is, up to a symmetric form, the transpose of the first. The moduli space $M_{{\mathbb{P}}^3}(0,c_2)$ is thereby identified with a space of such $\alpha$ modulo ${\mathrm{GL}}_{c_2}\times \mathrm{Sp}$ symmetry; it is smooth at every point where $H^2(\mathrm{End}E)=0$ and, in low $c_2$, irreducible of the expected dimension. The instanton locus is the subspace where $\alpha$ carries a compatible quaternionic structure — the reality constraint that cuts the real $S^4$ instantons out of the complex moduli.

Synthesis. The classification of stable rank-2 bundles on projective space is the bridge between three worlds, and putting these together explains why one theorem from 1977 sits under so much later geometry. The foundational reason the moduli space is tractable is Barth's restriction theorem: stability forces a general line to split in balanced fashion, which is exactly the hypothesis that lets the Beilinson monad of 04.07.03 reconstruct the bundle from finite linear-algebra data, and this is dual to the statement that the jumping locus is a proper divisor recording the bundle's spectrum. The central insight is that the discrete invariants $(c_1,c_2)$ and the spectrum together generalise Grothendieck's complete splitting invariant on ${\mathbb{P}}^1$ to higher projective space, where completeness is lost but stability restores enough rigidity to build a smooth moduli scheme. This pattern appears again in the ADHM construction of 03.07.10, where the real slice of $M_{{\mathbb{P}}^3}(0,c_2)$ is the instanton moduli space of dimension $8c_2-3$: the algebraic geometry of Barth and the gauge theory of Yang-Mills are the same object read in two languages.

Full proof set Master

Proposition (uniqueness of the null-correlation bundle). Up to isomorphism and the action of ${\mathrm{PGL}}_4$, there is a unique stable rank-2 bundle $N$ on ${\mathbb{P}}^3$ with $c_1=0$ and $c_2=1$, and it fits in $0\to \mathcal{O}(-1)\to {\Omega }_{{\mathbb{P}}^3}^1(1)\to N\to 0$.

Proof. Let $E$ be stable with $(c_1,c_2)=(0,1)$. By Riemann-Roch on ${\mathbb{P}}^3$ and the vanishing $H^0(E)=0$ (stability, $c_1=0$), the twist $E(1)$ has $\chi (E(1))=4$ and $H^0(E(1))\ne 0$; choose a nonzero section $s\in H^0(E(1))$. Its zero locus $Y=(s{)}_0$ is a curve of degree $c_2(E(1))=c_2+c_1+1=2$ — a conic or a pair of lines — and the section realises the Serre extension

$$0\to \mathcal{O}\to E(1)\to {\mathcal{I}}_Y(2)\to 0.$$

Computing ${\mathrm{Ext}}^1({\mathcal{I}}_Y(2),\mathcal{O})$ via Serre duality 04.08.03 for a degree-2 codimension-2 locus gives a one-dimensional space of extension classes, so $E(1)$ — hence $E$ — is determined by $Y$ up to scalar. The variety of such $Y$ is a single ${\mathrm{PGL}}_4$-orbit (all smooth conics are projectively equivalent, and the degenerate loci do not yield bundles), so $E$ is unique up to ${\mathrm{PGL}}_4$. Identifying the universal such extension with the contact sub-bundle of a symplectic form on $H^0(\mathcal{O}(1){)}^{\vee }$ gives the displayed presentation with ${\Omega }_{{\mathbb{P}}^3}^1(1)$, whose stability and Chern classes $(0,1)$ are checked directly from the Euler sequence. $\square$

Proposition (jumping-line curve has degree $c_2$). For a stable rank-2 bundle $E$ on ${\mathbb{P}}^3$ with $c_1=0$ and a general plane $P\subset {\mathbb{P}}^3$, the jumping lines of $E$ contained in $P$ form a plane curve of degree $c_2$ in the dual plane $P^{\vee }$.

Proof. Restrict $E$ to $P\cong {\mathbb{P}}^2$; by Barth's restriction theorem $E{|}_P$ is semistable with the same $c_2$. A line $L\subset P$ jumps iff $h^0(E{|}_L(-1))>0$, i.e. iff $L$ lies in the support of the first direct image $R^1{\pi }_{\ast }(q^{\ast }E(-1))$ along the incidence correspondence $\{(x,L):x\in L\}\subset P\times P^{\vee }$ with projections $q,\pi$. This direct image is a sheaf on $P^{\vee }$ whose support is the degeneracy locus of a map of vector bundles obtained from the Beilinson monad 04.07.03 of $E{|}_P$; a Chern-class computation on the incidence variety identifies the class of this degeneracy locus with $c_2\cdot [\text{line}]$ in $P^{\vee }$. So the jumping-line curve has degree $c_2$. Equality of the spectrum-symmetric count with $c_2$ confirms the bound. $\square$

Connections Master

• Monads on projective space and the Beilinson resolution 04.07.03 supply the engine of this unit: every stable rank-2 bundle is the cohomology bundle of a monad, and Barth's restriction theorem is precisely the hypothesis that makes the monad reconstruction work, so the two units are the two halves of the same classification.

• Coherent sheaves 04.06.02 are the category in which slope stability is defined; the destabilising objects are coherent subsheaves, and the saturation operation that turns a rank-1 subsheaf into a line bundle is a coherent-sheaf construction used throughout the restriction-theorem proof.

• Serre duality 04.08.03 underlies both the Serre construction (the ${\mathrm{Ext}}^1$ class building a bundle from a codimension-2 cycle) and the reflection symmetry $k_i=-k_{c_2+1-i}$ of the spectrum, so it appears at both the construction and the invariant-theory ends of the story.

• The ADHM construction 03.07.10 is the consumer: the instanton moduli space of charge $c_2$ is the real, quaternionic slice of the complex moduli space $M_{{\mathbb{P}}^3}(0,c_2)$ of this unit, and its dimension $8c_2-3$ is read off from Barth's monad normal form.

Historical & philosophical context Master

Wolf Barth's 1977 paper Some properties of stable rank-2 vector bundles on ${\mathbb{P}}_n$ ^{[Barth 1977]} opened the systematic study of higher-rank bundles on projective space at a moment when algebraic geometers had mastered line bundles and divisors but had almost no tools for bundles that fail to split. Barth's restriction theorem and his introduction of jumping lines gave the first handle: a stable bundle could be probed line by line, and its failure to split generically — recorded by a divisor of jumping lines — became a computable invariant. The companion work with Elencwajg defined the spectrum, turning the intermediate cohomology into a combinatorial gadget.

The subject acquired sudden urgency when Atiyah, in his 1979 Pisa lectures Geometry of Yang-Mills Fields ^{[Atiyah 1979]}, showed that the self-dual Yang-Mills instantons of mathematical physics are, via the Penrose-Ward twistor correspondence, exactly the stable rank-2 bundles on ${\mathbb{P}}^3$ with a reality structure. Barth's classification, written for purely algebro-geometric reasons, turned out to encode the moduli of instantons on the four-sphere; the ADHM construction then made the linear-algebra data fully explicit. The episode is a textbook case of mathematics developed in one field becoming the indispensable language of another, and the philosophical lesson is that a good discrete invariant — here the pair $(c_1,c_2)$ refined by the spectrum — can outlive the motivations that produced it.

Bibliography Master

@article{Barth1977,
  author  = {Barth, Wolf},
  title   = {Some properties of stable rank-2 vector bundles on $\mathbb{P}_n$},
  journal = {Mathematische Annalen},
  volume  = {226},
  pages   = {125--150},
  year    = {1977}
}

@article{BarthHulek1978,
  author  = {Barth, Wolf and Hulek, Klaus},
  title   = {Monads and moduli of vector bundles},
  journal = {Manuscripta Mathematica},
  volume  = {25},
  pages   = {323--347},
  year    = {1978}
}

@incollection{BarthElencwajg1978,
  author    = {Barth, Wolf and Elencwajg, Georges},
  title     = {Concernant la cohomologie des fibr\'es alg\'ebriques stables sur $\mathbb{P}_n(\mathbb{C})$},
  booktitle = {Vari\'et\'es analytiques compactes},
  series    = {Lecture Notes in Mathematics},
  volume    = {683},
  publisher = {Springer},
  year      = {1978}
}

@book{OSS1980,
  author    = {Okonek, Christian and Schneider, Michael and Spindler, Heinz},
  title     = {Vector Bundles on Complex Projective Spaces},
  publisher = {Birkh\"auser},
  series    = {Progress in Mathematics},
  volume    = {3},
  year      = {1980}
}

@book{Atiyah1979,
  author    = {Atiyah, Michael F.},
  title     = {Geometry of Yang-Mills Fields},
  publisher = {Scuola Normale Superiore},
  address   = {Pisa},
  year      = {1979}
}