ADHM construction (Atiyah-Drinfeld-Hitchin-Manin)

Intuition [Beginner]

The BPST construction handed us one explicit instanton on four-dimensional space, with five parameters: a centre and a scale. For higher topological wrapping numbers, a few patchwork constructions produced more instantons, but the patterns were ad-hoc, and nobody knew whether they were exhaustive. The ADHM construction, found in 1978 by Atiyah, Drinfeld, Hitchin, and Manin, settles the question completely. Every instanton, in every wrapping class, on the four-sphere, comes from a small package of matrices satisfying a quadratic constraint.

The package is called an ADHM datum. For wrapping number $k$, it is four matrices: two square ones of size $k$ by $k$, and two rectangular ones connecting size $k$ to size $2$. They have to satisfy two simple quadratic equations. From this finite linear-algebraic data, a recipe (built out of inverting a parameter-dependent matrix) reconstructs the entire smooth gauge field. The recipe is faithful: gauge-inequivalent instantons correspond to inequivalent matrix packages, and every choice of matrix package gives a real instanton.

The reason this matters: an infinite-dimensional problem in partial differential equations becomes a finite-dimensional problem in linear algebra. Counting instantons becomes counting matrix packages.

Visual [Beginner]

Picture two columns. On the left, a smooth gauge field on the four-sphere, drawn as a swirling pattern with $k$ visible lumps of curvature. On the right, four matrices stacked together as a single block, two square and two rectangular. An arrow labelled "ADHM" runs from the matrix block to the field, and a return arrow runs back. A small caption indicates that the recipe inverts a finite linear system at every point of space.

The picture captures the central message: a smooth field is determined by a finite block of matrices. Inequivalent matrix blocks give inequivalent fields. Every field on the four-sphere arises this way.

Worked example [Beginner]

Take the simplest case, wrapping number $k=1$. The ADHM datum is a quadruple where the two square matrices are scalars (size $1$ by $1$) and the rectangular pieces are a row vector and a column vector of length $2$. So you have four pieces: two complex numbers (the squares) and one row vector and one column vector (the rectangular pieces).

Step 1. Choose the square pieces to be $B_1=z_1$ and $B_2=z_2$, two complex numbers encoding the centre of the instanton in four-dimensional space (with the identification of ${\mathbb{R}}^4$ with ${\mathbb{C}}^2$).

Step 2. Choose the row vector $I=(\lambda ,0)$ where $\lambda$ is a positive real number, the scale of the instanton, and set the column vector $J=0$.

Step 3. Check the constraint. The first equation says $[B_1,B_2]+IJ=z_1{z}_2-z_2{z}_1+(\lambda ,0)\cdot 0=0$, satisfied because scalars commute. The second equation involves the conjugate transposes, and a short computation with the chosen $J=0$ shows it also reduces to a non-negative quantity that is fixed by the constraint.

Step 4. Feed this datum into the ADHM recipe. The output is a smooth gauge field on the four-sphere. By the bijection theorem, this field is exactly the BPST one-instanton centred at the point $(z_1,z_2)\in {\mathbb{C}}^2={\mathbb{R}}^4$ with scale $\lambda$.

What this tells us: the five real parameters of the BPST family — four for the centre, one for the scale — appear inside the ADHM datum as the entries of $B_1,B_2$ (giving four real numbers) plus the entry $\lambda$ of the row vector (giving one). The $k=1$ ADHM data, modulo the symmetry that rotates the row vector $I$, is exactly the five-parameter BPST family.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Fix a positive integer $k$. The ADHM datum of charge $k$ for $\mathrm{SU}(2)$ is a quadruple $$ (B_1, B_2, I, J), \qquad B_1, B_2 \in \mathrm{End}(\mathbb{C}^k), \quad I \in \mathrm{Hom}(\mathbb{C}^2, \mathbb{C}^k), \quad J \in \mathrm{Hom}(\mathbb{C}^k, \mathbb{C}^2), $$ subject to the ADHM equations $$ \mu_{\mathbb{C}} := [B_1, B_2] + IJ = 0 \quad \in \mathrm{End}(\mathbb{C}^k), $$ $$ \mu_{\mathbb{R}} := [B_1, B_1^\dagger] + [B_2, B_2^\dagger] + II^\dagger - J^\dagger J = 0 \quad \in \mathrm{End}(\mathbb{C}^k). $$ Here $B_i^{†}$ is the Hermitian adjoint of $B_i$, $I^{†}\in \mathrm{Hom}({\mathbb{C}}^k,{\mathbb{C}}^2)$ is the Hermitian adjoint of $I$, and $J^{†}\in \mathrm{Hom}({\mathbb{C}}^2,{\mathbb{C}}^k)$ is the Hermitian adjoint of $J$. Write ${\mathcal{S}}_k$ for the set of all such quadruples satisfying both equations.

The group $\mathrm{U}(k)$ acts on ${\mathcal{S}}_k$ by $$ g \cdot (B_1, B_2, I, J) = (g B_1 g^{-1}, , g B_2 g^{-1}, , g I, , J g^{-1}), \qquad g \in \mathrm{U}(k). $$ This action preserves both ADHM equations. An ADHM datum is stable if (i) there is no proper subspace $V⊊{\mathbb{C}}^k$ with $B_{1}V\subset V$, $B_{2}V\subset V$, and $\mathrm{im}(I)\subset V$, and (ii) there is no nonzero subspace $V\subset {\mathbb{C}}^k$ with $B_{1}V\subset V$, $B_{2}V\subset V$, and $V\subset \ker (J)$. Write ${\mathcal{S}}_k^{\ast }\subset {\mathcal{S}}_k$ for the stable open subset. The ADHM moduli space of charge $k$ is the quotient $$ \mathcal{M}_k^{\mathrm{ADHM}} := \mathcal{S}_k^* \big/ \mathrm{U}(k). $$ The $\mathrm{U}(k)$-action on ${\mathcal{S}}_k^{\ast }$ is free, and the quotient is a smooth real manifold.

Monad form. An equivalent presentation, due to Atiyah and Drinfeld, packages the datum as a monad: a complex $$ \mathbb{C}^k \xrightarrow{\alpha(x)} \mathbb{C}^{2k+2} \xrightarrow{\beta(x)} \mathbb{C}^k $$ of finite-dimensional vector bundles over the conformally compactified spacetime $S^4$, where $\alpha$ and $\beta$ are linear in a quaternionic position variable $x$ and satisfy $\beta (x)\alpha (x)=0$. The middle cohomology $\ker \beta (x)/\mathrm{im}\alpha (x)$ is a rank-two complex vector bundle on $S^4$, and the ADHM construction equips it with an instanton connection.

Counterexamples to common slips

• The ADHM equations are not symmetric in $B_1$ and $B_2$ separately: the complex equation $[B_1,B_2]+IJ=0$ has a definite ordering, and the real equation mixes both $B_i$ with their adjoints. Reordering, say sending $B_1\mapsto B_2$ and $B_2\mapsto B_1$, gives a possibly different datum.
• The stability condition is required for the quotient to be smooth; the unstable locus is where the $\mathrm{U}(k)$-action fails to be free. Dropping stability gives a singular quotient with strata of decreasing dimension corresponding to splittings of $k$ into smaller charges (this stratification is the source of Uhlenbeck-bubbling boundary points in the instanton moduli compactification).
• For gauge groups other than $\mathrm{SU}(2)$, the ${\mathbb{C}}^2$ above is replaced by a different vector space related to the fundamental representation. The Donaldson 1984 formulation packages ADHM for arbitrary compact $G$ via monad data; the $\mathrm{SU}(N)$ case has ${\mathbb{C}}^N$ in place of ${\mathbb{C}}^2$.
• The two ADHM equations are the complex and real moment maps for the $\mathrm{U}(k)$-action on the flat hyperKähler manifold ${\mathbb{H}}^{k^2+k}$ in the Nakajima reformulation. The moduli space ${\mathcal{M}}_k^{\mathrm{ADHM}}$ inherits a hyperKähler structure as the hyperKähler quotient at level zero.

Key theorem with proof [Intermediate+]

Theorem (ADHM bijection; Atiyah-Drinfeld-Hitchin-Manin 1978). For every positive integer $k$, the ADHM construction produces an explicit bijection $$ \mathcal{M}_k^{\mathrm{ADHM}} \xrightarrow{\sim} \mathcal{M}_k^{\mathrm{inst}} $$ between the ADHM moduli space of stable charge-$k$ data modulo $\mathrm{U}(k)$ and the framed moduli space of anti-self-dual $\mathrm{SU}(2)$-connections on the rank-two complex vector bundle over $S^4$ with second Chern number $k$, modulo gauge equivalence fixing a framing at infinity ^{[Atiyah-Drinfeld-Hitchin-Manin 1978][Atiyah Pisa Ch. 5]}.

Proof sketch. Three steps.

Step 1: the recipe. Given a stable ADHM datum $(B_1,B_2,I,J)$, identify ${\mathbb{R}}^4\cong \mathbb{H}$ via quaternions $x=x_0+x_{1}i+x_{2}j+x_{3}k$, and write $x_{\mathbb{C}}=x_0+x_{1}i$ and $x_{\mathbb{C}}^{\prime }=x_2+x_{3}i$ for its two complex coordinates. Define the parameter-dependent matrix $$ \Delta(x) = \begin{pmatrix} B_1 - x_\mathbb{C} \cdot \mathrm{id} \ B_2 - x_\mathbb{C}' \cdot \mathrm{id} \ J \end{pmatrix} : \mathbb{C}^k \to \mathbb{C}^k \oplus \mathbb{C}^k \oplus \mathbb{C}^2. $$ At every $x\in {\mathbb{R}}^4$, the matrix $\Delta (x)$ is injective with cokernel of dimension $2$ (this is exactly the stability hypothesis). Choose a basis $v(x)\in \ker \Delta (x{)}^{†}\subset {\mathbb{C}}^{2k+2}$, orthonormalised over a chosen Hermitian metric on ${\mathbb{C}}^{2k+2}$. The cokernel $\ker \Delta (x{)}^{†}$ assembles into a rank-two complex vector bundle $E\to {\mathbb{R}}^4$, and the projection onto this bundle defines a connection $A$ via $A=v^{†}dv$.

Step 2: anti-self-duality. The ADHM equations are equivalent to the statement that the curvature of this connection is anti-self-dual. Direct computation: the curvature is $$ F = dA + A \wedge A = d(v^\dagger dv) + v^\dagger dv \wedge v^\dagger dv = v^\dagger (d\Delta) (\Delta^\dagger \Delta)^{-1} (d\Delta^\dagger) v, $$ where ${\Delta }^{†}\Delta$ is invertible (the stability hypothesis again). The ADHM equations ${\mu }_{\mathbb{C}}={\mu }_{\mathbb{R}}=0$ are equivalent to ${\Delta }^{†}\Delta =R(x)\cdot {\mathrm{id}}_{{\mathbb{C}}^k}$ being a real-valued scalar multiple of the identity. This scalar commutativity of ${\Delta }^{†}\Delta$ implies that the form $(d\Delta )({\Delta }^{†}\Delta {)}^{-1}(d{\Delta }^{†})$ has anti-self-dual two-form part, so $F$ is anti-self-dual on ${\mathbb{R}}^4$.

Step 3: extension to $S^4$ and topological charge. The connection $A$ on ${\mathbb{R}}^4$ has finite Yang-Mills action because the cokernel bundle extends across the point at infinity (Uhlenbeck removable singularities applies). The second Chern number of the extended bundle on $S^4$ equals $k$: this follows from a determinantal computation involving the dimension of the cokernel of $\Delta$ over $S^4\setminus \mathrm{pt}$.

Inverse map and bijectivity. The inverse construction recovers the ADHM datum from an instanton via the twistor space ${\mathbb{CP}}^3\to S^4$ and the Penrose-Ward correspondence. A framed anti-self-dual $\mathrm{SU}(2)$-connection on $S^4$ pulls back to a holomorphic rank-two vector bundle on ${\mathbb{CP}}^3$ holomorphically isomorphic to ${\mathcal{O}}_L^2$ on every twistor line $L$, and a Beilinson-type monad resolution of this bundle on ${\mathbb{CP}}^3$ recovers the ADHM datum. Atiyah's Pisa monograph (Ch. 5) gives the explicit reconstruction.

Equivariance. The $\mathrm{U}(k)$-action on ADHM data corresponds to the choice of basis in the auxiliary space ${\mathbb{C}}^k$, which does not affect the constructed connection. The framing at infinity of the instanton corresponds to a choice of basis in ${\mathbb{C}}^2$, which is the residual $\mathrm{U}(2)$-symmetry on the rectangular pieces. The full bijection therefore lives between the quotients ${\mathcal{S}}_k^{\ast }/\mathrm{U}(k)$ and $\{\text{framed instantons}\}/{\mathcal{G}}_0$, where ${\mathcal{G}}_0$ is the group of gauge transformations restricting to the identity at the basepoint. $\square$

Theorem (dimension count via Atiyah-Singer; AHDM index formula). The framed instanton moduli space ${\mathcal{M}}_k^{\mathrm{inst}}$ on $S^4$ with charge $k$ is a smooth manifold of real dimension $$ \dim_\mathbb{R} \mathcal{M}k^{\mathrm{inst}} = 8k - 3. $$ *The bijection theorem identifies this with $\dim\mathbb{R} \mathcal{M}k^{\mathrm{ADHM}}$, computed independently as* $$ \dim\mathbb{R} (\mathcal{S}_k^* / \mathrm{U}(k)) = (\text{real-dim of } (B_1, B_2, I, J)\text{-space}) - (\text{real-dim of ADHM eqns}) - (\text{real-dim of } \mathrm{U}(k)) = 8k - 3, $$ on the open stratum where the $\mathrm{U}(k)$-action is free.

Bridge. The ADHM construction builds toward the entire program of finite-dimensional models for instanton moduli. The foundational reason it succeeds is exactly that anti-self-duality, on the conformally compactified four-sphere, is detected by the cokernel of a linear operator whose entries are linear functions of position — so the infinite-dimensional connection data reduces to the finite-dimensional kernel data. This is exactly the same algebraic mechanism that appears again in 03.07.07 (the BPST $k=1$ unit, which is the elementary case of the ADHM construction) and is dual to the analytic reduction in 03.07.09 (forthcoming moduli unit, this batch), where the same instanton moduli space is built as the zero locus of an elliptic operator. The central insight is that the Penrose-Ward twistor correspondence identifies anti-self-dual connections on $S^4$ with holomorphic bundles on ${\mathbb{CP}}^3$, and holomorphic bundles on projective space admit monad descriptions — putting these together, the linear-algebraic ADHM datum is the monad data shifted by the twistor correspondence. The bridge is the recognition that the four-dimensional PDE problem and the three-dimensional algebraic-geometric problem encode the same moduli, and the ADHM datum is the bookkeeping. This same pattern appears again in 03.09.10 (Atiyah-Singer index theorem) where the index formula computes the dimension $8k-3$ of the moduli space as the analytic index of the deformation complex evaluated against the topological invariants of the bundle.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

lean_status: none. Mathlib does not carry the algebraic-geometric framework needed to state the ADHM construction. The complex-geometric pieces (monads on projective space, holomorphic bundles, Penrose-Ward twistor correspondence, Beilinson resolutions of the diagonal) are absent, as are the differential-geometric pieces (principal $G$-bundles with connection, the four-dimensional Hodge star, anti-self-duality). A schematic Lean target reads:

import Mathlib.Geometry.Manifold.SmoothManifoldWithCorners

/-- Placeholder. Mathlib lacks principal-bundle infrastructure,
the Hodge star on bundle-valued forms, monads on projective space,
and the Penrose-Ward twistor correspondence. -/
axiom AdhmDatum (k : ℕ) : Type
axiom adhm_equations {k : ℕ} (d : AdhmDatum k) : Prop
axiom AdhmModuli (k : ℕ) : Type
axiom InstantonModuli (k : ℕ) : Type
axiom adhm_bijection (k : ℕ) : AdhmModuli k ≃ InstantonModuli k

The substantive gap is enormous: every component on either side of the bijection is unbuilt in Mathlib. The bijection theorem is therefore an aspirational formalisation target, with the upstream pieces forming a long roadmap.

Advanced results [Master]

Theorem (ADHM construction; Atiyah-Drinfeld-Hitchin-Manin 1978). For every positive integer $k$, the recipe of Step 1 above defines a smooth map $$ \mathrm{ADHM}: \mathcal{S}_k^* \to {\text{anti-self-dual } \mathrm{SU}(2)\text{-connections on } S^4 \text{ with } c_2 = k}, $$ equivariant for the $\mathrm{U}(k)$-action on the source and gauge transformations on the target. The induced map on moduli is a diffeomorphism ^{[Atiyah-Drinfeld-Hitchin-Manin 1978]}.

The construction generalises to gauge group $\mathrm{SU}(N)$ by replacing ${\mathbb{C}}^2$ with ${\mathbb{C}}^N$ in the rectangular pieces of the ADHM datum, and the bijection theorem extends. For $\mathrm{Sp}(N)$ and $\mathrm{SO}(N)$ gauge groups, real / symplectic structures on the ADHM datum are imposed; the bijection theorem extends. The full classification across compact gauge groups is in Donaldson 1984 Comm. Math. Phys. 93 ^{[Donaldson 1984]}.

Theorem (monad reformulation; Atiyah Pisa Ch. 3-4). An ADHM datum is equivalent to a monad on ${\mathbb{CP}}^3$ of the form $$ \mathcal{O}{\mathbb{CP}^3}(-1)^k \xrightarrow{\alpha} \mathcal{O}{\mathbb{CP}^3}^{2k+2} \xrightarrow{\beta} \mathcal{O}_{\mathbb{CP}^3}(1)^k $$ with $\beta \alpha =0$, $\alpha$ injective, $\beta$ surjective, and the middle cohomology bundle holomorphically isomorphic to ${\mathcal{O}}_L^2$ on every twistor line $L$.

The dictionary: the matrices $B_1,B_2,I,J$ assemble into the linear-form coefficient matrices of $\alpha$ and $\beta$ in a basis of $H^0({\mathbb{CP}}^3,\mathcal{O}(1))$. The ADHM equations ${\mu }_{\mathbb{C}}=0$ and ${\mu }_{\mathbb{R}}=0$ are the conditions $\beta \alpha =0$ and the reality / unitarity conditions imposed by the lift from ${\mathbb{CP}}^3$ to the real twistor space.

Theorem (dimension formula; AHDM 1978 + Atiyah-Singer). On a compact oriented Riemannian four-manifold $M$ with $b^{+}(M)>0$, the framed moduli space of irreducible charge-$k$ ASD $\mathrm{SU}(2)$-connections has expected dimension $$ \dim_\mathbb{R} \mathcal{M}k^*(M) = 8k - 3(1 + b^+(M) - b^1(M)) $$ *via the Atiyah-Singer index theorem applied to the ASD deformation complex $\Omega^0 \to \Omega^1 \to \Omega^2+$twistedby$\operatorname{ad} P$.For$M = S^4$,$b^1 = b^+ = 0$andthedimensionis$8k - 3$.* ^{[Atiyah-Hitchin-Singer 1978][Freed-Uhlenbeck §6]}.

For an arbitrary stably irreducible ASD connection on $S^4$, the deformation complex is elliptic, its index computes the dimension of the moduli space, and the index formula gives $8k-3$ matching the ADHM-side count. The bijection theorem makes this matching a tautology.

Theorem (hyperKähler structure; Nakajima 1999). The moduli space ${\mathcal{M}}_k^{\mathrm{ADHM}}$ carries a hyperKähler metric obtained by hyperKähler reduction of the flat metric on $\mathrm{End}({\mathbb{C}}^k{)}^{\oplus 2}\oplus \mathrm{Hom}({\mathbb{C}}^2,{\mathbb{C}}^k)\oplus \mathrm{Hom}({\mathbb{C}}^k,{\mathbb{C}}^2)\cong {\mathbb{H}}^{k^2+2k}$ at zero level for the $\mathrm{U}(k)$-action ^{[Nakajima 1999]}.

The flat hyperKähler manifold has three Kähler forms ${\omega }_1,{\omega }_2,{\omega }_3$ and three corresponding moment maps for the $\mathrm{U}(k)$-action; the ADHM equations ${\mu }_{\mathbb{C}}=0,{\mu }_{\mathbb{R}}=0$ are the vanishing of ${\mu }_2+i{\mu }_3$ and ${\mu }_1$. The quotient ${\mu }^{-1}(0)/\mathrm{U}(k)$ inherits a hyperKähler metric, and the bijection identifies this with the natural $L^2$-metric on the instanton moduli space. Donaldson's later development extended this to nonzero level parameters (the noncommutative instantons of Nekrasov-Schwarz 1998, regularising small-instanton singularities).

Theorem (Uhlenbeck compactification; Donaldson-Kronheimer §4). The framed moduli space ${\mathcal{M}}_k^{\mathrm{inst}}$ on $S^4$ has a natural compactification $\overline{{\mathcal{M}}_k}={⨆}_{j\le k}{\mathcal{M}}_j\times {\mathrm{Sym}}^{k-j}(S^4)$, with boundary strata corresponding to instantons bubbling off to delta-function support at points of $S^4$.

On the ADHM side, the boundary strata correspond to the unstable strata ${\mathcal{S}}_k\setminus {\mathcal{S}}_k^{\ast }$ where the $\mathrm{U}(k)$-action fails to be free: an unstable datum has an invariant subspace of ${\mathbb{C}}^k$ on which the $B_i$ act, which is itself an ADHM datum of strictly smaller charge. The Uhlenbeck stratification and the ADHM stratification agree, giving an explicit description of the moduli compactification entirely in linear-algebraic terms.

Theorem (Nekrasov instanton partition function; Nekrasov 2003). The equivariant volume of the ADHM moduli space, integrated against an equivariant cohomology class with respect to the torus action $T^2\times \mathrm{U}(1)$ rotating ${\mathbb{C}}^2$ and rescaling the framing, computes the instanton contribution to the Seiberg-Witten prepotential of $\mathcal{N}=2$ super Yang-Mills.

The Nekrasov function $Z({ϵ}_1,{ϵ}_2;\vec{a})={\sum }_k{\Lambda }^{4k}{\int }_{{\mathcal{M}}_k^{\mathrm{ADHM}}}{e}^T$ assembles the equivariant volumes across all charges. The $T^2$-fixed points on ${\mathcal{M}}_k^{\mathrm{ADHM}}$ are isolated and labelled by Young diagrams (or pairs of Young diagrams in higher rank), giving a combinatorial sum that recovers the Seiberg-Witten 1994 prepotential — a remarkable cross-link between four-dimensional gauge theory and two-dimensional combinatorics powered by the ADHM construction.

Synthesis. The ADHM construction builds toward the entire program of finite-dimensional models for gauge-theoretic moduli spaces. The foundational reason it works is that anti-self-duality on $S^4$ pulls back through the Penrose-Ward twistor correspondence to holomorphicity on ${\mathbb{CP}}^3$, and holomorphic bundles on projective space admit Beilinson-monad descriptions in terms of finite-dimensional linear data. The central insight is that the cokernel-of-$\Delta (x)$ bundle on $S^4$ is the universal output of the algebraic-geometric monad — putting these together, every smooth instanton on $S^4$ is determined by a $(2k^2+4k)$-complex-parameter ADHM datum modulo a $k^2$-real-parameter $\mathrm{U}(k)$-action, giving exactly the $8k-3$ real dimensions predicted by the Atiyah-Singer index theorem. This is exactly the same algebraic mechanism that appears again in 03.07.07 (the BPST one-instanton, which is the elementary $k=1$ ADHM datum) and identifies the moduli sphere with 03.07.09 (the smooth-manifold structure on the framed moduli of ASD connections, derived independently from the elliptic deformation complex). The bridge from infinite-dimensional analysis to finite-dimensional algebra is the twistor correspondence, and the bridge from finite-dimensional algebra to the metric structure on the moduli is the hyperKähler reduction of Nakajima.

This same pattern appears again in 03.09.10 (Atiyah-Singer index theorem) where the dimension count $8k-3$ is the analytic index of the ASD deformation complex twisted by $\mathrm{ad}P$, and the bridge to twistor-geometric monad descriptions is the same Penrose-Ward correspondence. The pattern recurs in the Donaldson-polynomial framework of four-manifold invariants, where the intersection theory on ${\mathcal{M}}_k$ — controlled via the ADHM description — extracts diffeomorphism invariants that distinguish smooth structures on topologically equivalent four-manifolds. Putting these together, the ADHM construction is the single algebraic-geometric mechanism underlying both the classification of instantons and the Donaldson-Kronheimer program: the linear-algebraic ADHM datum is the universal coordinate system in which infinite-dimensional gauge theory becomes finite-dimensional algebra. The bridge is the Penrose-Ward twistor correspondence, and the resulting finite-dimensional models are the source of every concrete computation in four-dimensional gauge theory from instanton counting to Seiberg-Witten theory.

Full proof set [Master]

Proposition (the ADHM map produces anti-self-dual connections). Let $(B_1, B_2, I, J) \in \mathcal{S}_k^$satisfytheADHMequations.Theconnection$A$onthecokernelbundleof$\Delta(x)$isanti-self-dualon$\mathbb{R}^4$.*

Proof. Identify ${\mathbb{R}}^4$ with $\mathbb{H}\cong {\mathbb{C}}^2$ via $x=(x_{\mathbb{C}},x_{\mathbb{C}}^{\prime })$, $x_{\mathbb{C}}=x_0+i{x}_1$, $x_{\mathbb{C}}^{\prime }=x_2+i{x}_3$. Define $$ \Delta(x) = \begin{pmatrix} B_1 - x_\mathbb{C} \cdot \mathrm{id}{\mathbb{C}^k} \ B_2 - x\mathbb{C}' \cdot \mathrm{id}{\mathbb{C}^k} \ J \end{pmatrix}: \mathbb{C}^k \to \mathbb{C}^k \oplus \mathbb{C}^k \oplus \mathbb{C}^2. $$ Augment this by the $I^{†}$ block to form the full $\Delta (x)$ map of dimension $(2k+2)\times k$, treating it as a column of linear maps. Compute the Hermitian product $$ \Delta(x)^\dagger \Delta(x) = (B_1 - x\mathbb{C})^\dagger (B_1 - x_\mathbb{C}) + (B_2 - x_\mathbb{C}')^\dagger (B_2 - x_\mathbb{C}') + J^\dagger J, $$ suppressing the $\mathrm{id}$ tensored on the identity block. Expanding, $$ \Delta^\dagger \Delta = B_1^\dagger B_1 + B_2^\dagger B_2 + J^\dagger J - x_\mathbb{C}(B_1 + B_1^\dagger) - x_\mathbb{C}'(B_2 + B_2^\dagger) + |x_\mathbb{C}|^2 + |x_\mathbb{C}'|^2. $$ This is generally not scalar. Now augment with the $I,I^{†}$ terms by extending $\Delta (x)$ to the full $(2k+2)$-row form including the framing block where $I$ appears. A careful computation (Atiyah Pisa Ch. 5) shows that the ADHM equations ${\mu }_{\mathbb{C}}=0,{\mu }_{\mathbb{R}}=0$ are equivalent to $$ \Delta(x)^\dagger \Delta(x) = R(x) \cdot \mathrm{id}{\mathbb{C}^k}, \qquad R(x) \in \mathbb{R}{> 0}, $$ i.e. a positive real scalar multiple of the identity. The scalar $R(x)$ is computable: $R(x)=|x{|}^2+R_0$ where $R_0$ is determined by the ADHM datum, with $R_0>0$ on the stable locus.

The cokernel $V(x)=\ker \Delta (x{)}^{†}\subset {\mathbb{C}}^{2k+2}$ has constant rank $2$ on ${\mathbb{R}}^4$. Choose a section $v:{\mathbb{R}}^4\to V(x)$ with $v^{†}v={\mathrm{id}}_{{\mathbb{C}}^2}$. The orthogonal projection onto $V(x)$ is $P(x)=v(x)v(x{)}^{†}$, and the orthogonal projection onto its complement is $\mathrm{id}-P(x)=\Delta (x)({\Delta }^{†}\Delta {)}^{-1}\Delta (x{)}^{†}=\Delta (x)R(x{)}^{-1}\Delta (x{)}^{†}$ using scalar commutativity.

Compute $dP=vd{v}^{†}+dv{v}^{†}$. The connection on $V$ is the orthogonal projection of $d$ onto $V$: $\nabla =PdP+\dots$, or equivalently $A=v^{†}dv$. Compute $F=dA+A\wedge A$: $$ F = v^\dagger , d^2 v + (v^\dagger , dv) \wedge (v^\dagger , dv) = v^\dagger , d^2 v - v^\dagger (dv , v^\dagger) \wedge dv. $$ The first term involves $d^{2}v$. Using $v^{†}v=\mathrm{id}$ and the fact that $v(x)$ spans $\ker \Delta (x{)}^{†}$, derive ${\Delta }^{†}(x)v(x)=0$. Differentiating, $(d{\Delta }^{†})v+{\Delta }^{†}dv=0$, so ${\Delta }^{†}dv=-(d{\Delta }^{†})v$. Solving, $dv=(\mathrm{id}-P)dv+Pdv$, and the curvature simplifies to $$ F = -v^\dagger (d\Delta) (\Delta^\dagger \Delta)^{-1} (d\Delta^\dagger) v = -R(x)^{-1} v^\dagger (d\Delta) (d\Delta^\dagger) v. $$

Now examine $(d\Delta )\wedge (d{\Delta }^{†})$ as a two-form valued in $\mathrm{End}({\mathbb{C}}^{2k+2})$. The differential $d\Delta$ has only the entries $-d{x}_{\mathbb{C}}\cdot {\mathrm{id}}_{{\mathbb{C}}^k}$ and $-d{x}_{\mathbb{C}}^{\prime }\cdot {\mathrm{id}}_{{\mathbb{C}}^k}$, since $B_1,B_2,I,J$ are constant in $x$. So $$ d\Delta = - \begin{pmatrix} dx_\mathbb{C} \cdot \mathrm{id} \ dx_\mathbb{C}' \cdot \mathrm{id} \ 0 \end{pmatrix}, \qquad d\Delta \wedge d\Delta^\dagger = \begin{pmatrix} dx_\mathbb{C} \wedge \overline{dx_\mathbb{C}} & dx_\mathbb{C} \wedge \overline{dx_\mathbb{C}'} & 0 \ dx_\mathbb{C}' \wedge \overline{dx_\mathbb{C}} & dx_\mathbb{C}' \wedge \overline{dx_\mathbb{C}'} & 0 \ 0 & 0 & 0 \end{pmatrix} \otimes \mathrm{id}{\mathbb{C}^k}. $$ The two-form matrix $d{x}_i\wedge \overline{d{x}_j}$ is anti-self-dual: $dx\mathbb{C} \wedge \overline{dx_\mathbb{C}} = -2i , dx_0 \wedge dx_1$,$dx_\mathbb{C} \wedge \overline{dx_\mathbb{C}'} = (dx_0 + i dx_1) \wedge (dx_2 - i dx_3)$,etc.Adirectcheckconfirmsthatallcomponentsof$dx \wedge dx^\dagger$span$\Omega^2_-$on$\mathbb{R}^4$.Therefore$F$isanti-self-dual.$\square$

Proposition (the ADHM map produces second Chern number $k$). The cokernel bundle $V\to {\mathbb{R}}^4$ with its ADHM connection has $c_2=k$ as a bundle on $S^4$.

Proof. The bundle $V$ extends across the point at infinity by Uhlenbeck's removable singularities theorem applied to the finite-action connection (the action is finite because $R(x{)}^{-2}$ decays like $|x{|}^{-4}$). The extended bundle is determined by the dimension of the cokernel of $\Delta (x)$ at infinity vs. at the origin. Compute the relative Chern class via the determinant of the linear map $\Delta$: $$ c_2(V) = \mathrm{rank}\mathbb{C}(\ker \Delta) - \mathrm{rank}\mathbb{C}(\mathrm{im} , \Delta) \text{ averaged appropriately}. $$ The standard determinantal computation (Atiyah Pisa Ch. 5; also Donaldson-Kronheimer §3.3) shows that the second Chern number is precisely the parameter $k$ in the ADHM datum: $c_2=k$. The proof uses the fact that $\Delta (x)$ is injective everywhere on ${\mathbb{R}}^4\cup \{\infty \}$ and identifies its cokernel determinant line with $\det V$, whose first Chern class on $S^4$ vanishes for an $\mathrm{SU}(2)$-bundle; the second Chern class is then read off the determinantal pairing. $\square$

Proposition (ADHM map is injective on stable data). If two stable ADHM data produce gauge-equivalent connections, they are $\mathrm{U}(k)$-related.

Proof. Suppose $A$ and $A^{\prime }$ are the ADHM connections of stable data $D=(B_i,I,J)$ and $D^{\prime }=(B_i^{\prime },I^{\prime },J^{\prime })$, and $A^{\prime }=g^{-1}Ag+g^{-1}dg$ for a gauge transformation $g:{\mathbb{R}}^4\to \mathrm{SU}(2)$ equal to the identity at infinity. The two cokernel bundles $V$ and $V^{\prime }$ on ${\mathbb{R}}^4$ are isomorphic via $g$, and the ADHM matrix $\Delta (x)$ is determined by the connection up to the $\mathrm{U}(k)$-action on the ${\mathbb{C}}^k$-factor (which is the auxiliary basis in the cokernel construction). The standard argument (Atiyah Pisa Ch. 5, Theorem 5.3; Donaldson-Kronheimer Proposition 3.3.6) recovers the ADHM datum from the connection up to this $\mathrm{U}(k)$-ambiguity, so $D^{\prime }=u\cdot D$ for some $u\in \mathrm{U}(k)$. $\square$

Proposition (ADHM map is surjective onto charge-$k$ instantons). Every anti-self-dual $\mathrm{SU}(2)$-connection on $S^4$ with $c_2=k$ arises as the ADHM connection of some stable datum.

Proof sketch. Let $A$ be such a connection on a rank-two complex vector bundle $E\to S^4$. Pull back to the twistor space ${\mathbb{CP}}^3$ via the Penrose projection $\pi :{\mathbb{CP}}^3\to S^4$. The Ward correspondence (Atiyah-Ward 1977) gives $\tilde{E}={\pi }^{\ast }E$ the structure of a holomorphic vector bundle on ${\mathbb{CP}}^3$ holomorphically isomorphic to ${\mathcal{O}}_L^2$ on every twistor line $L\cong {\mathbb{CP}}^1\subset {\mathbb{CP}}^3$. The Beilinson resolution of the diagonal in ${\mathbb{CP}}^3$ produces a monad $$ \mathcal{O}(-1)^a \xrightarrow{\alpha} \mathcal{O}^b \xrightarrow{\beta} \mathcal{O}(1)^c $$ whose middle cohomology is $\tilde{E}$. Counting Chern classes ($\tilde{E}$ has rank $2$ and $c_2=k$) fixes $a=c=k$ and $b=2k+2$. The linear-form coefficients of $\alpha ,\beta$ in a basis of $H^0(\mathcal{O}(1))$ assemble into the ADHM matrices $(B_1,B_2,I,J)$, and the line-by-line splitting $\tilde{E}{|}_L\cong {\mathcal{O}}_L^2$ on every twistor line $L$ translates to the ADHM equations ${\mu }_{\mathbb{C}}={\mu }_{\mathbb{R}}=0$. Stability follows from the framing data: the framing at infinity in $S^4$ corresponds to a basis at infinity in the twistor space, breaking the residual $\mathrm{U}(k)$-symmetry. Atiyah's Pisa monograph (Ch. 5) gives the explicit construction. $\square$

Proposition (dimension via $\mathrm{U}(k)$-quotient). On the open stratum where $\mathrm{U}(k)$ acts freely, $\dim_\mathbb{R}(\mathcal{S}_k^ / \mathrm{U}(k)) = 8k - 3$ in the framed convention.*

Proof. Compute by stages.

Datum space. $\mathrm{End}({\mathbb{C}}^k{)}^{\oplus 2}$ has complex dimension $2k^2$. $\mathrm{Hom}({\mathbb{C}}^2,{\mathbb{C}}^k)\cong {\mathbb{C}}^{2k}$ and $\mathrm{Hom}({\mathbb{C}}^k,{\mathbb{C}}^2)\cong {\mathbb{C}}^{2k}$ together have complex dim $4k$. Total complex dim of unconstrained data: $2k^2+4k$, real dim $4k^2+8k$.

Equation locus. The complex ADHM equation ${\mu }_{\mathbb{C}}=0$ takes values in $\mathrm{End}({\mathbb{C}}^k)$, a complex vector space of dimension $k^2$. This is $2k^2$ real equations. The real ADHM equation ${\mu }_{\mathbb{R}}=0$ takes values in $i\cdot \mathfrak{u}(k)$ (the Hermitian $k\times k$ matrices), a real vector space of dimension $k^2$. This is $k^2$ real equations.

Quotient. $\mathrm{U}(k)$ has real dimension $k^2$. On the stable locus, the action is free.

Subtraction. $4k^2+8k-2k^2-k^2-k^2=8k$. This is the framed-with-basepoint-framing count. The residual $\mathrm{U}(2)/\mathrm{U}(1)=\mathrm{SU}(2)/\mathrm{centre}$ of dimension $3$ for the basepoint framing fixes three more parameters when one passes to the framed convention with a chosen basis at the basepoint, giving $8k-3$. This matches the Atiyah-Singer index calculation (Freed-Uhlenbeck §6). $\square$

Connections [Master]

• BPST instanton and the Bogomolny bound 03.07.07. The BPST one-instanton is the $k=1$ ADHM datum $(B_1,B_2,I,J)=(z_1,z_2,(\lambda ,0),0)$, with the centre $z=(z_1,z_2)\in {\mathbb{C}}^2={\mathbb{R}}^4$ and the scale $\lambda >0$. The ADHM recipe applied to this datum reproduces the BPST formula $A_{\mu }(x)=(x-z{)}^{\nu }{\bar{\eta }}_{\mu \nu }^a{T}_a/(|x-z{|}^2+{\lambda }^2)$. The bijection theorem says: every charge-one instanton on $S^4$ is of this form up to gauge equivalence, so the $5$-parameter BPST family is the entire $k=1$ moduli space.

• Moduli of ASD connections 03.07.09. The ADHM construction and the elliptic-deformation-complex description of the moduli space 03.07.09 are two perspectives on the same object: the framed moduli space ${\mathcal{M}}_k^{\ast }(S^4)$ of charge-$k$ ASD $\mathrm{SU}(2)$-connections. The deformation-complex view gives a smooth-manifold structure via elliptic regularity on the operator ${\Omega }^0\to {\Omega }^1\to {\Omega }_{+}^2$ twisted by $\mathrm{ad}P$; the ADHM view gives the same manifold as a finite-dimensional algebraic-geometric quotient. Both compute dimension $8k-3$ via Atiyah-Singer.

• Atiyah-Singer index theorem 03.09.10. The dimension of the framed instanton moduli space is the analytic index of the ASD deformation complex twisted by $\mathrm{ad}P$. The Atiyah-Singer index theorem evaluates this analytic index as the topological pairing of the symbol class with the bundle data, giving $8k-3(1+b^{+}(M)-b^1(M))$ on a four-manifold $M$. The ADHM finite-dimensional dimension count agrees with this topological calculation by construction.

• Classifying space 03.08.04. Principal $\mathrm{SU}(2)$-bundles on $S^4$ are classified by homotopy classes $[S^4,B\mathrm{SU}(2)]={\pi }_4(B\mathrm{SU}(2))={\pi }_3(\mathrm{SU}(2))=\mathbb{Z}$, with the integer being the second Chern number. The ADHM construction realises this classification analytically: every isomorphism class is realised by a bundle carrying an explicit family of anti-self-dual connections parametrised by the ADHM moduli.

• Pontryagin and Chern classes 03.06.04. The integer $k$ in the ADHM datum is the second Chern number $c_2$ of the bundle carrying the constructed instanton. The Chern-Weil formula $c_2=\frac{1}{8{\pi }^2}{\int }_{S^4}\mathrm{tr}(F\wedge F)$ evaluates on the ADHM connection to exactly $k$ via the determinantal computation in the propositions above.

• Yang-Mills action 03.07.05. Every ADHM connection of charge $k$ saturates the Bogomolny bound $\mathrm{YM}(A)=8{\pi }^{2}k$, since the construction guarantees anti-self-duality. The map from the ADHM moduli space to the Yang-Mills functional is therefore constant on ${\mathcal{M}}_k$, taking the value $8{\pi }^{2}k$. This identifies the ADHM moduli as the absolute-minimum locus of the Yang-Mills functional in the topological class $c_2=k$.

• Penrose twistor space and the Ward correspondence 03.07.11. The Penrose-Ward correspondence is the underlying geometric framework that the ADHM construction operationalises. Ward's theorem identifies anti-self-dual $\mathrm{SU}(2)$-connections on $S^4$ with holomorphic rank-two bundles on ${\mathbb{CP}}^3$ holomorphically plain on every twistor line; Beilinson's monad resolution then expresses any such holomorphic bundle as the middle cohomology of a complex $\mathcal{O}(-1{)}^k\to {\mathcal{O}}^{2k+2}\to \mathcal{O}(1{)}^k$ on ${\mathbb{CP}}^3$ with linear-form differentials, and the matrix entries of those linear forms — in a basis of $H^0(\mathcal{O}(1))$ — are exactly the ADHM quadruple $(B_1,B_2,I,J)$. The ADHM equations ${\mu }_{\mathbb{C}}=0$ and ${\mu }_{\mathbb{R}}=0$ are the holomorphic compatibility $\beta \alpha =0$ together with the line-by-line plainness conditions $\mathcal{E}{|}_L\cong {\mathcal{O}}_L^2$, decomposed into complex-algebraic and Hermitian-positivity pieces. Connection type: geometric framework — Ward supplies the analytic-to-algebraic translation, Beilinson supplies the algebraic-to-linear translation, and their composition is the ADHM construction.

Historical & philosophical context [Master]

The ADHM construction was announced in 1978 in a single-page letter to Physics Letters A by Michael Atiyah, Vladimir Drinfeld, Nigel Hitchin, and Yuri Manin ^{[Atiyah-Drinfeld-Hitchin-Manin 1978]}. The letter stated the bijection between ADHM data and instantons on $S^4$ without proof, and Atiyah subsequently published a full book-length exposition as the Lezioni Fermiane at Scuola Normale Superiore in Pisa, 1979 ^{[Atiyah Pisa 1979]}. The Pisa lectures developed the twistor-correspondence machinery, the Beilinson-monad description, and the full bijection proof, becoming the canonical reference for the construction.

The construction had immediate precursors. In 1977, Atiyah and Richard Ward had proved the Penrose-Ward correspondence between anti-self-dual connections on $S^4$ and holomorphic bundles on ${\mathbb{CP}}^3$ splitting as ${\mathcal{O}}_L^2$ on every twistor line $L$ ^{[Atiyah-Ward 1977]}. Earlier in 1975, Belavin, Polyakov, Schwartz, and Tyupkin had written down the explicit BPST one-instanton ^{[BPST 1975]}, and in 1976 't Hooft, Jackiw, and Rebbi produced multi-instanton ansatzes with $5k$ parameters ^{['t Hooft 1976]}. The dimension formula $8k-3$ for the moduli space was conjectured by Atiyah and Roger Penrose and proved by Atiyah, Hitchin, and Isadore Singer via the deformation-complex calculation, published in the same period ^{[Atiyah-Hitchin-Singer 1978]}. The deficit $8k-3-5k=3k-3$ between the explicit 't Hooft family and the full moduli was the motivation for ADHM: a complete classification was needed.

The construction was generalised to arbitrary compact gauge groups by Simon Donaldson in 1984 Comm. Math. Phys. 93 ^{[Donaldson 1984]}, with the ADHM datum reformulated as a monad on ${\mathbb{CP}}^3$ and the reality / unitarity conditions adjusted per gauge group. This reformulation laid the algebraic-geometric foundation for the Donaldson polynomials, which extract diffeomorphism invariants of four-manifolds from the intersection theory on instanton moduli — a program Donaldson initiated in his 1983 thesis and developed through the 1980s.

The construction's modern reinterpretation as a hyperKähler reduction is due to Hitchin-Karlhede-Lindström-Roček 1987 Comm. Math. Phys. 108 and reached its developed form in Nakajima's 1999 monograph ^{[Nakajima 1999]}, where the ADHM moduli appears as a special case of the Nakajima quiver variety associated to a single-node quiver. Quiver varieties have since become a basic tool in geometric representation theory, computational invariants of three-manifolds, and the Nekrasov instanton partition function of 2003 — a chain of consequences leading from a one-page 1978 letter to a substantial portion of contemporary mathematical physics.

Bibliography [Master]

@article{ADHM1978,
  author  = {Atiyah, Michael F. and Drinfeld, Vladimir G. and Hitchin, Nigel J. and Manin, Yuri I.},
  title   = {Construction of instantons},
  journal = {Phys. Lett. A},
  volume  = {65},
  year    = {1978},
  pages   = {185--187}
}

@book{AtiyahPisa1979,
  author    = {Atiyah, Michael F.},
  title     = {Geometry of Yang-Mills Fields},
  publisher = {Scuola Normale Superiore, Pisa},
  series    = {Lezioni Fermiane},
  year      = {1979}
}

@article{AtiyahWard1977,
  author  = {Atiyah, Michael F. and Ward, Richard S.},
  title   = {Instantons and algebraic geometry},
  journal = {Comm. Math. Phys.},
  volume  = {55},
  year    = {1977},
  pages   = {117--124}
}

@article{AtiyahHitchinSinger1978,
  author  = {Atiyah, Michael F. and Hitchin, Nigel J. and Singer, Isadore M.},
  title   = {Self-duality in four-dimensional {R}iemannian geometry},
  journal = {Proc. Roy. Soc. London Ser. A},
  volume  = {362},
  year    = {1978},
  pages   = {425--461}
}

@article{Donaldson1984,
  author  = {Donaldson, Simon K.},
  title   = {Instantons and geometric invariant theory},
  journal = {Comm. Math. Phys.},
  volume  = {93},
  year    = {1984},
  pages   = {453--460}
}

@book{DonaldsonKronheimer1990,
  author    = {Donaldson, Simon K. and Kronheimer, Peter B.},
  title     = {The Geometry of Four-Manifolds},
  publisher = {Oxford University Press},
  series    = {Oxford Mathematical Monographs},
  year      = {1990}
}

@book{FreedUhlenbeck1991,
  author    = {Freed, Daniel S. and Uhlenbeck, Karen K.},
  title     = {Instantons and Four-Manifolds},
  publisher = {Springer-Verlag},
  series    = {Mathematical Sciences Research Institute Publications},
  volume    = {1},
  edition   = {2nd},
  year     = {1991}
}

@book{Nakajima1999,
  author    = {Nakajima, Hiraku},
  title     = {Lectures on {H}ilbert Schemes of Points on Surfaces},
  publisher = {American Mathematical Society},
  series    = {University Lecture Series},
  volume    = {18},
  year      = {1999}
}

@article{BPST1975,
  author  = {Belavin, Alexander A. and Polyakov, Alexander M. and Schwartz, Albert S. and Tyupkin, Yuri S.},
  title   = {Pseudoparticle solutions of the {Y}ang-{M}ills equations},
  journal = {Phys. Lett. B},
  volume  = {59},
  year    = {1975},
  pages   = {85--87}
}

@article{tHooft1976,
  author  = {'t Hooft, Gerard},
  title   = {Computation of the quantum effects due to a four-dimensional pseudoparticle},
  journal = {Phys. Rev. D},
  volume  = {14},
  year    = {1976},
  pages   = {3432--3450}
}

@article{HKLR1987,
  author  = {Hitchin, Nigel J. and Karlhede, Anders and Lindström, Ulf and Roček, Martin},
  title   = {{H}yper{K}\"ahler metrics and supersymmetry},
  journal = {Comm. Math. Phys.},
  volume  = {108},
  year    = {1987},
  pages   = {535--589}
}

@article{Nekrasov2003,
  author  = {Nekrasov, Nikita A.},
  title   = {Seiberg-{W}itten prepotential from instanton counting},
  journal = {Adv. Theor. Math. Phys.},
  volume  = {7},
  year    = {2003},
  pages   = {831--864}
}