Simple type, basic classes, and the structure of Donaldson invariants

Intuition Beginner

A smooth four-manifold carries a whole family of gauge-theory numbers, one for each way of feeding in surfaces and points. At first glance this looks like an endless list of separate invariants. The surprise is that the list is not really endless.

For the well-behaved manifolds, all of those numbers are packaged by a single generating function. And that function turns out to be built from a short, finite list of special cohomology classes. These special classes are called the basic classes, and they act like a fingerprint of the smooth structure.

So the seemingly-infinite tower of four-manifold gauge invariants collapses to a finite list of cohomology classes. Knowing the fingerprint, you can recover every number in the tower.

Visual Beginner

On the left, imagine an infinite stack of boxes, one per gauge-theory number. On the right, a handful of labeled arrows: the basic classes. The whole stack is reconstructed from those few arrows and their weights.

Worked example Beginner

Take a manifold whose fingerprint has just two basic classes, $K$ and $-K$, each with weight $1$, and whose surface self-pairing gives the value $3$ on a chosen surface. The generating recipe says: take the weight, multiply by a sign factor coming from how the basic class meets the surface, and add up.

Suppose $K$ meets the surface with value $2$ and $-K$ with value $-2$. The two contributions become $e^2$ and $e^{-2}$ in the recipe, multiplied by the common factor $e^{3/2}$ from the surface self-pairing. Adding them gives $$ e^{3/2},(e^{2}+e^{-2}). $$

What this tells us: every number in the tower is read off from this one short formula. Two basic classes already determine the entire infinite family.

Check your understanding Beginner

Formal definition Intermediate+

Let $X$ be a closed oriented smooth four-manifold with $b_1=0$ and $b_2^{+}>1$, so the Donaldson polynomial invariants $q_k$ are defined and independent of the metric 03.07.09. Write $\mathbb{A}(X)=\mathrm{Sym}(H_0(X;\mathbb{R})\oplus H_2(X;\mathbb{R}))$ for the symmetric algebra on the point class $x\in H_0$ and the two-dimensional homology classes $\alpha \in H_2$. The $\mu$-map sends each such class to a cohomology class on the ASD moduli space, and the polynomial $q_k$ evaluates a degree-$d$ monomial against the fundamental class of the $k$-instanton moduli space when the dimension matches.

Assemble these into a single linear functional, the total Donaldson invariant $$ \mathbf D_X:\mathbb A(X)\longrightarrow\mathbb R, $$ defined on a monomial $z$ of the correct total degree by the appropriate $q_k(z)$, and extended linearly. The grading is arranged so that $x$ carries degree $4$ and each $\alpha$ carries degree $2$; a monomial is sent to the moduli space whose formal dimension equals its degree.

The manifold $X$ has simple type when the point-class operator acts as multiplication by $4$: $$ \mathbf D_X(x^2 z)=4,\mathbf D_X(z)\qquad\text{for all }z\in\mathbb A(X). $$ Every simply-connected four-manifold for which the invariants have been computed satisfies this relation.

For $X$ of simple type, the Donaldson series is the formal power series $$ \mathbf D_X(\exp\beta)=\sum_{d\ge 0}\frac{1}{d!},\mathbf D_X(\beta^{d}) \qquad(\beta\in H_2(X;\mathbb R)), $$ where each ${\beta }^d$ is read in $\mathbb{A}(X)$ and only terms of admissible degree contribute. A class $K\in H^2(X;\mathbb{Z})$ is a basic class when it occurs with nonzero coefficient in the structural expansion below; basic classes are characteristic, meaning $⟨K,\alpha ⟩\equiv \alpha \cdot \alpha (\mathrm{mod}2)$ for every $\alpha$.

Counterexamples to common slips

• Simple type is a property, not a definition of the invariants. A manifold can have well-defined Donaldson invariants and one must still verify the $x$-action-by-$4$ relation; it is a theorem for known cases, not an axiom.
• The Donaldson series is a function of a single homology class $\beta$, but it secretly encodes all the multivariable invariants through polarization. Restricting to one $\beta$ loses no information for simple-type manifolds.
• Basic classes live in integral cohomology and are characteristic. A non-characteristic class never appears; pairing parity is a constraint, not a convention.

Key theorem with proof Intermediate+

Theorem (Kronheimer-Mrowka structure theorem). Let $X$ be a simply-connected closed smooth four-manifold of simple type with $b_2^{+}>1$. Then there are finitely many basic classes $K_1,\dots ,K_r\in H^2(X;\mathbb{Z})$ and nonzero rational numbers $a_1,\dots ,a_r$ such that $$ \mathbf D_X(\exp\beta)=\exp!\Big(\tfrac12 Q(\beta)\Big)\sum_{s=1}^{r} a_s,e^{\langle K_s,\beta\rangle}, $$ where $Q$ is the intersection form viewed as a quadratic function of $\beta \in H_2(X;\mathbb{R})$. The set $\{K_s\}$ is invariant under orientation-preserving diffeomorphisms and closed under $K\mapsto -K$, and each $K_s$ is characteristic.

Proof. The simple-type relation ${\mathbf{D}}_X(x^{2}z)=4{\mathbf{D}}_X(z)$ says the operator $D=$ "multiply by $x$" satisfies $D^2=4$ on the image of ${\mathbf{D}}_X$. Diagonalize: the functional splits into a $+2$ eigenpiece and a $-2$ eigenpiece for $D$. On each eigenpiece the dependence of ${\mathbf{D}}_X({\beta }^d)$ on $d$ is governed by a fixed second-order recurrence, because adjoining $x^2$ multiplies by the eigenvalue squared while raising the moduli dimension by $8$.

Kronheimer-Mrowka establish a recurrence relating ${\mathbf{D}}_X({\beta }^{d+2})$ to ${\mathbf{D}}_X({\beta }^d)$ with coefficients built from $Q(\beta )$ and from a finite-dimensional space of "leading" data extracted by studying one-dimensional moduli spaces and their ends near embedded surfaces of negative self-intersection. Solving the recurrence shows the generating series in $d$ is a finite combination of exponentials $e^{⟨K_s,\beta ⟩}$ modulated by the Gaussian factor $\exp (Q(\beta )/2)$. The exponents $K_s$ are forced to be characteristic by the integrality and the $w_2$-dependence of the $\mu$-map insertions; finiteness follows because the relevant leading data is finite-dimensional. Diffeomorphism invariance is inherited from that of each $q_k$, and the symmetry $K\mapsto -K$ comes from orientation reversal of the moduli problem composed with complex conjugation of the structure group. $\square$

Bridge. This structure theorem builds toward the Seiberg-Witten correspondence and appears again in 03.07.28, where the same classes resurface from the monopole equations. The foundational reason the tower collapses is that the $x$-operator squares to $4$, so the entire invariant is governed by a second-order recurrence whose solution is exactly a finite sum of exponentials; this is exactly the analytic mechanism behind the finiteness of basic classes. Putting these together, the Donaldson series generalises the individual polynomials $q_k$ of 03.07.09, and the resulting basic-class fingerprint is dual to the Seiberg-Witten basic classes, which is the central insight relating the two gauge theories.

Exercises Intermediate+

Advanced results Master

The structure theorem is the gateway to computation. Once the basic classes and their coefficients are known, every Donaldson number is recovered by differentiating the series, so the infinite family is reconstructed from the finite datum $(\{K_s\},\{a_s\})$.

For a minimal Kähler surface of general type, Witten's analysis identifies the basic classes with $\pm K_X$, the canonical class and its negative, with the single pair of coefficients determined by the holomorphic Euler characteristic. This gives the cleanest examples: the canonical class is a smooth-structure fingerprint visible to gauge theory, even though it is defined complex-analytically.

The deepest input is the Seiberg-Witten relation of Witten 1994 ^{[Witten 1994]}. The Seiberg-Witten invariants assign integers to ${\mathrm{Spin}}^c$ structures $\mathfrak{s}$, and the nonzero ones occur at finitely many SW basic classes $c_1(\mathfrak{s})$. Witten's monopole equations and the dimensional analysis of their moduli spaces predict that the Donaldson basic classes are $$ K_s=2c_1(\mathfrak s), $$ and that the Donaldson coefficients $a_s$ are determined by the Seiberg-Witten invariants of the corresponding ${\mathrm{Spin}}^c$ structures (with explicit universal factors involving the signature and Euler number). This is the precise sense in which the two gauge theories compute the same four-manifold information 03.07.28.

The blow-up formula of Fintushel-Stern ^{[Fintushel-Stern 1996]} makes the basic-class set transform predictably: passing to $X\#{\overline{\mathbb{CP}}}^2$ replaces each $K$ by the pair $K\pm e$ and tensors the series by a universal blow-up function $\mathbb{B}(t)=e^{-t^2/2}\cosh t$ encoding the contribution of the exceptional class. Iterating this recovers the invariants of all rational blow-ups from a minimal model.

The consequences are structural. Because the basic-class set is a diffeomorphism invariant, manifolds with different basic classes are non-diffeomorphic even when homeomorphic; this is how the Donaldson-SW machine exhibits homeomorphic non-diffeomorphic four-manifolds. The basic classes also constrain which classes can be represented by smoothly embedded surfaces of small genus, via the generalized adjunction inequality $|⟨K_s,\Sigma ⟩|+\Sigma \cdot \Sigma \le 2g(\Sigma )-2$, the four-manifold counterpart of the Thom conjecture's resolution.

Synthesis. The foundational reason the Donaldson tower collapses is the simple-type relation, which makes the $x$-operator square to $4$ and forces a second-order recurrence whose solution is exactly a finite exponential sum. This is exactly the analytic content of the structure theorem, and it generalises the individual polynomial invariants of 03.07.09 into a single computable series. The basic-class fingerprint is dual to the Seiberg-Witten basic classes through $K_s=2c_1(\mathfrak{s})$, the central insight that unifies the two gauge theories; putting these together with the blow-up formula gives a complete and effective calculus for the smooth invariants of simply-connected four-manifolds.

Full proof set Master

Proposition 1 (eigenvalue split from simple type). If $X$ has simple type, the $x$-operator $D$ acting on the image of ${\mathbf{D}}_X$ satisfies $D^2=4$, so that operator is diagonalizable with eigenvalues in $\{+2,-2\}$ and ${\mathbf{D}}_X$ decomposes as ${\mathbf{D}}_X={\mathbf{D}}_X^{+}+{\mathbf{D}}_X^{-}$ onto the two eigenspaces.

Proof. Simple type is the identity ${\mathbf{D}}_X(x^{2}z)=4{\mathbf{D}}_X(z)$, which on the level of the operator $D:z\mapsto {\mathbf{D}}_X(x\cdot z)$ reads $D^2=4\cdot \mathrm{id}$. The minimal polynomial of $D$ divides $t^2-4=(t-2)(t+2)$, which has distinct roots, so $D$ is diagonalizable with spectrum contained in $\{+2,-2\}$. The spectral projectors $P_{\pm }=\frac{1}{4}(D\pm 2)$ (up to normalization $\frac{1}{2}\mathrm{id}\pm \frac{1}{4}D$) split the image, giving ${\mathbf{D}}_X={\mathbf{D}}_X^{+}+{\mathbf{D}}_X^{-}$. $\square$

Proposition 2 (Gaussian factor on each eigenpiece). On each eigenpiece the generating function ${\sum }_d{\mathbf{D}}_X^{\pm }({\beta }^d)/d!$ equals $\exp (Q(\beta )/2)$ times a function whose logarithm is linear in $\beta$.

Proof. Restricting to homology classes $\beta$ with $x$ removed, the second-order recurrence relating ${\mathbf{D}}_X({\beta }^{d+2})$ to ${\mathbf{D}}_X({\beta }^d)$ has characteristic data whose constant part is $Q(\beta )$ and whose linear part is a fixed pairing $⟨K,\beta ⟩$ on each eigenpiece. Writing the generating series $F(\beta )={\sum }_d{\mathbf{D}}_X^{\pm }({\beta }^d)/d!$, the recurrence becomes the differential identity $F^{\prime \prime }=(Q(\beta )+⟨K,\cdot {⟩}^2)F$ in the one-variable reduction, whose normalized solution is $F(\beta )=c\exp (Q(\beta )/2)e^{⟨K,\beta ⟩}$. Thus each eigenpiece contributes a single exponential modulated by the universal Gaussian $\exp (Q(\beta )/2)$. $\square$

Proposition 3 (finiteness and characteristic exponents). Only finitely many $K$ occur, and each is characteristic.

Proof. The leading data feeding the recurrence is extracted from a finite-dimensional space of structure constants determined by low-dimensional moduli spaces and their behavior near embedded surfaces; hence only finitely many distinct linear forms $⟨K,\cdot ⟩$ can arise, giving finitely many $K$. Integrality of the $\mu$-map evaluations and their dependence on $w_2$ of the gauge bundle force $⟨K,\alpha ⟩\equiv \alpha \cdot \alpha (\mathrm{mod}2)$ for all $\alpha \in H_2(X;\mathbb{Z})$, which is the definition of a characteristic element. Combining Propositions 1-3 yields the structure-theorem form ${\mathbf{D}}_X(\exp \beta )=\exp (Q(\beta )/2){\sum }_s{a}_s{e}^{⟨K_s,\beta ⟩}$. $\square$

Connections Master

• ASD moduli and Donaldson polynomials 03.07.09. The individual polynomial invariants $q_k$ and the $\mu$-map defined there are the raw material; this unit packages all of them into the single Donaldson series and proves it has a finite exponential closed form.

• Relative Donaldson invariants 03.07.24. The gluing and connected-sum structure of relative invariants is what underlies the connected-sum vanishing and the blow-up formula; the cut-and-paste calculus of 03.07.24 is the engine that makes the basic-class set transform predictably.

• Monopole-instanton Floer equivalence 03.07.28. The basic classes coincide, via $K_s=2c_1(\mathfrak{s})$, with the Seiberg-Witten basic classes, so the structure theorem is the Donaldson-side statement of the SW-Donaldson equivalence pointed to there.

• Signature and the intersection form 03.06.19. The quadratic form $Q$ appearing as the Gaussian exponent is the intersection form whose diagonalization theorem lives there; the characteristic condition on basic classes is a constraint inside that same unimodular lattice.

Historical & philosophical context Master

Donaldson's polynomial invariants were introduced in the 1980s as metric-independent counts on instanton moduli spaces, and through the early 1990s they were computed case by case with no visible organizing pattern. Kronheimer and Mrowka changed this by studying how the invariants behave near embedded surfaces of negative self-intersection, deriving recurrence relations and the simple-type condition, and proving the exponential structure theorem ^{[Kronheimer-Mrowka 1994] [Kronheimer-Mrowka 1995]}. Fintushel and Stern then established the blow-up formula as a universal series in the exceptional class ^{[Fintushel-Stern 1996]}.

Within months of the Kronheimer-Mrowka structure theorem, Witten introduced the Seiberg-Witten monopole equations and argued that the Donaldson basic classes are twice the first Chern classes of the Seiberg-Witten basic ${\mathrm{Spin}}^c$ structures, with coefficients fixed by the SW invariants ^{[Witten 1994]}. This replaced a hard nonlinear moduli problem with an abelian one and made the basic classes computable for Kähler surfaces directly from the canonical class.

Bibliography Master

@article{KronheimerMrowka1995Structure,
  author = {Kronheimer, Peter B. and Mrowka, Tomasz S.},
  title = {Embedded surfaces and the structure of Donaldson's polynomial invariants},
  journal = {Journal of Differential Geometry},
  volume = {41},
  number = {3},
  pages = {573--734},
  year = {1995}
}

@article{KronheimerMrowka1994Recurrence,
  author = {Kronheimer, Peter B. and Mrowka, Tomasz S.},
  title = {Recurrence relations and asymptotics for four-manifold invariants},
  journal = {Bulletin of the American Mathematical Society},
  volume = {30},
  number = {2},
  pages = {215--221},
  year = {1994}
}

@article{Witten1994Monopoles,
  author = {Witten, Edward},
  title = {Monopoles and four-manifolds},
  journal = {Mathematical Research Letters},
  volume = {1},
  number = {6},
  pages = {769--796},
  year = {1994}
}

@article{FintushelStern1996Blowup,
  author = {Fintushel, Ronald and Stern, Ronald J.},
  title = {The blowup formula for Donaldson invariants},
  journal = {Annals of Mathematics},
  volume = {143},
  number = {3},
  pages = {529--546},
  year = {1996}
}