Decomposition of the Diagonal

• Decomposition of the diagonal is a concept in algebraic geometry that expresses the diagonal class as a sum of cycles with support restrictions, linking to CH₀-triviality and stable rationality.
• It extends to various fields such as K3 surface theory, derived categories, and homotopy theory, where it simplifies complex global structures into manageable building blocks.
• Applications range from detecting rationality obstructions in cubic threefolds using intermediate Jacobians to optimizing quantum circuit synthesis through exact diagonal operator factorizations.

“Decomposition of the diagonal” is a polysemous term whose central meaning depends on context. In algebraic geometry, it denotes Chow-theoretic or cohomological identities for the class of the diagonal $\Delta_X\subset X\times X$, often with support constraints on the correcting terms, and it is closely tied to universal $\mathrm{CH}_0$-triviality, Abel–Jacobi maps, integral Hodge classes, and obstructions to stable rationality (Voisin, 2010, Achter et al., 2020). In the geometry of K3 surfaces it also refers to the Beauville–Voisin decomposition of the small diagonal in $S^3$ (Bazhov, 2016). In derived algebraic geometry it appears as an explicit resolution of the diagonal on root stacks (Zhao, 2023). In homotopy theory it names decompositions of diagonal arrangements (Kishimoto et al., 2014). In quantum information, by contrast, it refers to exact circuit factorizations of diagonal unitary operators (Beer et al., 2015, Tułowiecki et al., 2024).

1. Algebraic-geometric core: Chow and cohomological decompositions

For a smooth projective variety $Y$ of dimension $n$, with diagonal $\Delta_Y\subset Y\times Y$, the basic Bloch–Srinivas pattern is an equality

$N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$

with $\operatorname{Supp}(Z_1)\subset D\times Y$, $\operatorname{Supp}(Z_2)\subset Y\times W$, where $D\subset Y$ is a proper closed subset, $\mathrm{CH}_0$0 is a lower-dimensional subset, and $\mathrm{CH}_0$1 (Voisin, 2010). The corresponding cohomological form is

$\mathrm{CH}_0$2

with the same support conditions, and an integral cohomological decomposition of the diagonal is the case $\mathrm{CH}_0$3 (Voisin, 2010).

A more general formalism allows an $\mathrm{CH}_0$4-decomposition of type $\mathrm{CH}_0$5 for a cycle class $\mathrm{CH}_0$6, meaning

$\mathrm{CH}_0$7

with $\mathrm{CH}_0$8 supported on $\mathrm{CH}_0$9 and $S^3$0 supported on $S^3$1 (Achter et al., 2020). A strict $S^3$2-decomposition is the special case where $S^3$3 is supported on $S^3$4 for a divisor $S^3$5 and $S^3$6 is supported on $S^3$7 for a $S^3$8-dimensional $S^3$9 (Achter et al., 2020).

This formalism is tightly linked to $Y$0-theory. For a smooth proper $Y$1, $Y$2 is universally $Y$3-trivial if and only if $Y$4 admits a strict $Y$5-decomposition (Achter et al., 2020). Stable rationality implies universally $Y$6-triviality for smooth projective varieties; thus, for stably rational $Y$7, $Y$8 admits a strict Chow decomposition (Achter et al., 2020). If $Y$9 is rationally chain connected over a perfect field, then $n$0 is universally trivial, hence some nonzero multiple $n$1 admits a strict Chow decomposition (Achter et al., 2020).

These statements underlie a standard distinction. A decomposition of the diagonal is not merely an equality in a Chow group: the support conditions are the decisive part of the notion, because they force the induced correspondences to factor through lower-dimensional varieties. That factorization is the mechanism behind consequences for torsion, algebraicity, and birational geometry.

2. Intermediate Jacobians, Abel–Jacobi maps, and threefolds

For a smooth projective threefold $n$2 with $n$3, the intermediate Jacobian is

$n$4

and the Abel–Jacobi map on homologically trivial codimension-$n$5 cycles is

$n$6

Voisin’s analysis organizes the relation between diagonal decompositions and families of $n$7-cycles through two properties: () asks for a smooth projective variety $n$8 and a codimension-$n$9 cycle $\Delta_Y\subset Y\times Y$0 such that the induced morphism $\Delta_Y\subset Y\times Y$1 is surjective with rationally connected general fiber; (*) is the corresponding statement for families of cycles in a fixed cohomology class $\Delta_Y\subset Y\times Y$2, where the target is the Deligne-twisted torsor $\Delta_Y\subset Y\times Y$3 arising from

$\Delta_Y\subset Y\times Y$4

These constructions make the Abel–Jacobi map the intermediary between diagonal decompositions and the geometry of $\Delta_Y\subset Y\times Y$5-cycles (Voisin, 2010).

A central theorem states that if $\Delta_Y\subset Y\times Y$6 is a smooth projective threefold with $\Delta_Y\subset Y\times Y$7 and $\Delta_Y\subset Y\times Y$8 admits an integral cohomological decomposition of the diagonal, then $\Delta_Y\subset Y\times Y$9 is generated by classes of algebraic cycles, $N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$0 has no torsion for any $N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$1, and $N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$2 satisfies condition (*) (Voisin, 2010). Conversely, assuming those three properties and the existence on $N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$3 of a $N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$4-cycle $N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$5 of class

$N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$6

one recovers an integral cohomological decomposition of the diagonal (Voisin, 2010). In the same circle of ideas, a cohomological decomposition

$N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$7

with $N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$8 and $N\Delta_Y=Z_1+Z_2\in CH^n(Y\times Y),$9 for the support of $\operatorname{Supp}(Z_1)\subset D\times Y$0, implies that $\operatorname{Supp}(Z_1)\subset D\times Y$1 annihilates the torsion of $\operatorname{Supp}(Z_1)\subset D\times Y$2 for all $\operatorname{Supp}(Z_1)\subset D\times Y$3, annihilates $\operatorname{Supp}(Z_1)\subset D\times Y$4, forces $\operatorname{Supp}(Z_1)\subset D\times Y$5 for $\operatorname{Supp}(Z_1)\subset D\times Y$6, and yields a cycle $\operatorname{Supp}(Z_1)\subset D\times Y$7 with $\operatorname{Supp}(Z_1)\subset D\times Y$8 (Voisin, 2010).

Cubic threefolds provide the main explicit class of examples. For a smooth cubic threefold $\operatorname{Supp}(Z_1)\subset D\times Y$9, Iliev–Markushevich and Tikhomirov proved that the Abel–Jacobi map

$\operatorname{Supp}(Z_2)\subset Y\times W$0

from a desingularization of the Hilbert scheme of degree $\operatorname{Supp}(Z_2)\subset Y\times W$1, genus $\operatorname{Supp}(Z_2)\subset Y\times W$2 curves is surjective with general fiber $\operatorname{Supp}(Z_2)\subset Y\times W$3, and Voisin proved that for a general cubic threefold the analogous map

$\operatorname{Supp}(Z_2)\subset Y\times W$4

for degree $\operatorname{Supp}(Z_2)\subset Y\times W$5 elliptic curves is surjective with rationally connected general fiber (Voisin, 2010). These results supply the families required for () and (*), and they feed directly into the proof that degree $\operatorname{Supp}(Z_2)\subset Y\times W$6 integral Hodge classes on suitable fibrations into cubic threefolds are algebraic (Voisin, 2010).

3. Stable rationality, degeneration methods, and positive characteristic

In later work the integral decomposition of the diagonal became a birational invariant and an obstruction to stable rationality. Over algebraically closed fields of positive characteristic, this obstruction has been extended from Hodge-theoretic intermediate Jacobians to $\operatorname{Supp}(Z_2)\subset Y\times W$7-adic cohomology and algebraic representatives. For a smooth projective threefold $\operatorname{Supp}(Z_2)\subset Y\times W$8 over an algebraically closed field $\operatorname{Supp}(Z_2)\subset Y\times W$9, if $D\subset Y$0 admits a strict cohomological $D\subset Y$1-decomposition with respect to $D\subset Y$2, then $D\subset Y$3, $D\subset Y$4 is $D\subset Y$5-algebraic for all $D\subset Y$6, $D\subset Y$7 and $D\subset Y$8 are isomorphisms, $D\subset Y$9, $\mathrm{CH}_0$00 admits a universal codimension-$\mathrm{CH}_0$01 cycle, and the minimal class

$\mathrm{CH}_0$02

is $\mathrm{CH}_0$03-algebraic (Achter et al., 2020). A partial converse gives necessary and sufficient conditions for a strict cohomological $\mathrm{CH}_0$04-decomposition in terms of precisely these data (Achter et al., 2020).

A key replacement for the complex intermediate Jacobian is the second algebraic representative $\mathrm{CH}_0$05. Under geometric rational chain connectedness or stable rationality, there exists a canonical symmetric $\mathrm{CH}_0$06-isogeny

$\mathrm{CH}_0$07

constructed from miniversal or universal codimension-$\mathrm{CH}_0$08 cycles; in characteristic $\mathrm{CH}_0$09 it agrees with the principal polarization on $\mathrm{CH}_0$10 induced by Hodge theory (Achter et al., 2020). This transfer of the polarization package is what allows Voisin-type obstructions to be reformulated in positive characteristic.

The quartic double solid with nodes is the main test case. In characteristic greater than two, a desingularization of a very general quartic double solid with seven nodes satisfies conditions (1)–(4) and (6) of the $\mathrm{CH}_0$11-adic criterion but fails condition (5): it has no universal codimension-$\mathrm{CH}_0$12 cycle class (Achter et al., 2020). This produces stably irrational unirational examples detected by the diagonal obstruction even when classical invariants do not detect irrationality (Achter et al., 2020).

Recent degeneration arguments have pushed the obstruction further. For a very general degree-$\mathrm{CH}_0$13 hypersurface or a very general $\mathrm{CH}_0$14 complete intersection, Fiammengo–Lüders reduce the existence of a decomposition of the diagonal to the corresponding question for cubic hypersurfaces together with Voisin’s essential dimension condition. In particular, if a very general degree-$\mathrm{CH}_0$15 hypersurface $\mathrm{CH}_0$16 does not admit a decomposition of the diagonal and satisfies $\mathrm{CH}_0$17, then a very general $\mathrm{CH}_0$18 complete intersection in $\mathrm{CH}_0$19 does not admit a decomposition of the diagonal; similarly, under the cubic input, a very general degree-$\mathrm{CH}_0$20 hypersurface in $\mathrm{CH}_0$21 does not admit a decomposition of the diagonal (Fiammengo et al., 8 Oct 2025). Using a recent result on cubic threefolds, they also obtain a new proof that a very general complex quartic $\mathrm{CH}_0$22-fold and a very general complex $\mathrm{CH}_0$23-complete intersection $\mathrm{CH}_0$24-fold do not admit a decomposition of the diagonal and are therefore not retract rational (Fiammengo et al., 8 Oct 2025).

4. The small diagonal of a K3 surface

For a smooth projective K3 surface $\mathrm{CH}_0$25 over $\mathrm{CH}_0$26, the small diagonal is

$\mathrm{CH}_0$27

viewed as a class in $\mathrm{CH}_0$28. Beauville and Voisin proved the exact identity

$\mathrm{CH}_0$29

where $\mathrm{CH}_0$30 is the canonical $\mathrm{CH}_0$31-cycle of degree $\mathrm{CH}_0$32 given by the class of any point lying on a rational curve on $\mathrm{CH}_0$33 (Bazhov, 2016).

This identity has two classical consequences. For divisor classes $\mathrm{CH}_0$34,

$\mathrm{CH}_0$35

and

$\mathrm{CH}_0$36

Thus, although $\mathrm{CH}_0$37 is not equal to $\mathrm{CH}_0$38 for a K3 surface, the subring generated by divisor classes behaves as though all $\mathrm{CH}_0$39-cycles coming from divisor intersections collapse to the canonical class (Bazhov, 2016).

Bazhov gave a new proof of this decomposition under the hypotheses $\mathrm{CH}_0$40, $\mathrm{CH}_0$41 very ample, and $\mathrm{CH}_0$42, replacing Beauville–Voisin’s use of one-parameter families of elliptic curves by an explicit projective-geometric construction in the embedding $\mathrm{CH}_0$43 (Bazhov, 2016). The proof uses codimension-$\mathrm{CH}_0$44 linear sections, special multiplicity loci, universal incidence cycles $\mathrm{CH}_0$45 and $\mathrm{CH}_0$46 in $\mathrm{CH}_0$47 and $\mathrm{CH}_0$48, and a decomposition of their restrictions to $\mathrm{CH}_0$49 and $\mathrm{CH}_0$50. In this setting, “decomposition of the diagonal” refers not to the big diagonal in $\mathrm{CH}_0$51 that appears in rationality questions, but to the small diagonal in $\mathrm{CH}_0$52 and the multiplicative structure of the Chow ring.

5. Categorical and homotopical variants

A different usage occurs for root stacks. For the $\mathrm{CH}_0$53-th root stack

$\mathrm{CH}_0$54

of a line bundle with section cutting out a divisor $\mathrm{CH}_0$55, an explicit resolution of the diagonal on

$\mathrm{CH}_0$56

is constructed from equivariant kernels $\mathrm{CH}_0$57 and the associated Fourier–Mukai endofunctors $\mathrm{CH}_0$58 (Zhao, 2023). The key identity

$\mathrm{CH}_0$59

identifies the projector onto the “bulk” subcategory, while the successive differences are built from functors $\mathrm{CH}_0$60 corresponding to $\mathrm{CH}_0$61-weight pieces (Zhao, 2023). The resulting semi-orthogonal decomposition is

$\mathrm{CH}_0$62

where the $\mathrm{CH}_0$63 are the images of fully faithful functors from $\mathrm{CH}_0$64 (Zhao, 2023). Here the phrase denotes a Fourier–Mukai resolution of $\mathrm{CH}_0$65, not a Bloch–Srinivas decomposition in the Chow group.

An analogous shift of meaning appears in topology. For a simplicial complex $\mathrm{CH}_0$66 on $\mathrm{CH}_0$67 and a space $\mathrm{CH}_0$68, the partially diagonal subspaces

$\mathrm{CH}_0$69

assemble into the diagonal arrangement

$\mathrm{CH}_0$70

If $\mathrm{CH}_0$71, then for a connected CW complex $\mathrm{CH}_0$72 there is a homotopy equivalence

$\mathrm{CH}_0$73

where $\mathrm{CH}_0$74 is the polyhedral product associated to $\mathrm{CH}_0$75 and $\mathrm{CH}_0$76 (Kishimoto et al., 2014). Combined with the Bahri–Bendersky–Cohen–Gitler decomposition of $\mathrm{CH}_0$77, this yields

$\mathrm{CH}_0$78

For a closed connected manifold $\mathrm{CH}_0$79, this gives the Euler characteristic formula

$\mathrm{CH}_0$80

where $\mathrm{CH}_0$81 is the complement of the arrangement (Kishimoto et al., 2014).

These variants show that, outside birational geometry, “decomposition of the diagonal” often means a decomposition of an object supported on a diagonal locus—such as $\mathrm{CH}_0$82 or a diagonal arrangement—rather than an equality of algebraic cycles on $\mathrm{CH}_0$83.

6. Diagonal operator decomposition in quantum information

In quantum circuit synthesis, “decomposition of the diagonal” refers to factoring diagonal unitary operators into native gates. For diagonal Hermitian $\mathrm{CH}_0$84-qubit gates, every diagonal entry is $\mathrm{CH}_0$85, so the operator has a binary representation. The multiple-controlled $\mathrm{CH}_0$86 gates $\mathrm{CH}_0$87 define binary vectors $\mathrm{CH}_0$88, and the set $\mathrm{CH}_0$89 forms a basis of $\mathrm{CH}_0$90. Consequently every diagonal Hermitian gate has a unique decomposition into a product of $\mathrm{CH}_0$91 gates, obtained by solving a linear system over $\mathrm{CH}_0$92 (Houshmand et al., 2014). In the reported experiments on all $\mathrm{CH}_0$93-, $\mathrm{CH}_0$94-, and $\mathrm{CH}_0$95-qubit diagonal Hermitian gates, the proposed synthesis achieved average reductions in CZ gate count of $\mathrm{CH}_0$96, $\mathrm{CH}_0$97, and $\mathrm{CH}_0$98, and average reductions in single-qubit gate count of $\mathrm{CH}_0$99, $S^3$00, and $S^3$01, respectively (Houshmand et al., 2014).

For general diagonal qudit unitaries,

$S^3$02

Beer and Dziemba generalized Welch et al.’s phase-context method from qubits to qudits (Beer et al., 2015). If the number of distinct phases is $S^3$03, the operator is decomposed into $S^3$04 blocks, each realized by a compute–phase–uncompute pattern using a cascaded entangler $S^3$05 and a single-qudit phase gate on an ancilla. The entangler itself is decomposed into multi-controlled INC gates through a signed base-$S^3$06 expansion

$S^3$07

and the resulting circuit size is $S^3$08, generalizing the qubit bound $S^3$09 and improving on earlier qubit methods with $S^3$10 scaling when $S^3$11 is small (Beer et al., 2015).

When the only native entangling gate is CX and the only single-qubit parametric gate is

$S^3$12

a parity-network framework gives exact resource counts on several topologies (Tułowiecki et al., 2024). Any generic diagonal construction uses exactly $S^3$13 phase gates, one for each nonzero signature in $S^3$14 (Tułowiecki et al., 2024). On a fully connected architecture, the optimal counts are

$S^3$15

where NPA means “No Permutation Allowed,” WPA means “Wire Permutation Allowed,” and SPA means “State Permutation Allowed” (Tułowiecki et al., 2024). On a line, the paper proves

$S^3$16

and on a ring, for many $S^3$17 admitting primitive trinomials,

$S^3$18

(Tułowiecki et al., 2024).

A recent mathematical formalization treats arbitrary diagonal operators in $S^3$19 through a recurrence

$S^3$20

where the tail is assembled from single-qubit diagonal factors $S^3$21 and commuting control operators $S^3$22 (Fedin et al., 10 Oct 2025). The parameter transformation is governed by matrices $S^3$23 satisfying

$S^3$24

and exact tensor-product factorization is characterized by log-phase separability,

$S^3$25

This quantum-information usage is structurally analogous to the geometric one in that both study how a “diagonal” object can be reconstructed from simpler building blocks, but the underlying categories, invariants, and complexity questions are entirely different (Fedin et al., 10 Oct 2025).

Across these literatures, the phrase therefore names a family of techniques rather than a single theorem. In algebraic geometry it is a support-sensitive identity for diagonal cycle classes; in K3 theory it governs the small diagonal and the Beauville–Voisin ring; in derived and homotopical settings it becomes a resolution or homotopy splitting of diagonal loci; and in quantum information it denotes explicit circuit factorizations of diagonal operators. This suggests that the unifying idea is formal rather than domain-specific: diagonal objects encode global structure, and decomposing them exposes that structure in a form usable for birational, categorical, homotopical, or algorithmic analysis.