Iyama–Wemyss Mutations

• Iyama–Wemyss mutations are operations that replace a chosen summand in a modifying module with an exchange kernel derived from approximation sequences, preserving the NCCR property.
• They apply across various contexts including toric singularities, GIT wall-crossing, and cDV settings by ensuring derived equivalences through tilting and mutation.
• These mutations underpin the connectivity of noncommutative crepant resolutions, influencing stability conditions and wall-and-chamber structures in modern algebraic geometry.

Iyama–Wemyss mutations are operations on modifying or maximal modifying modules over Gorenstein rings that replace a chosen summand by an exchange kernel obtained from approximation theory. If $M$ gives a non-commutative crepant resolution (NCCR), then the mutated module again gives an NCCR, and the corresponding endomorphism rings are derived equivalent. In recent work, these mutations appear as the module-theoretic mechanism behind connectivity of toric NCCRs, wall-crossing for magic windows in geometric invariant theory, symmetries of BPS and Gopakumar–Vafa invariants for cDV singularities, wall-and-chamber structures in $K$-theory and Bridgeland stability, and mutation networks for NCCRs of anticanonical cones over del Pezzo surfaces (Tomonaga, 30 Oct 2025, Hara et al., 2023, 2207.13540, Hara et al., 5 Mar 2026, Nordskova et al., 13 Apr 2026).

1. Modifying modules and the formal mutation operation

Let $R$ be a normal Gorenstein $\Bbbk$-algebra, or more generally a Gorenstein normal domain of dimension at least $2$, and let $M\in \mathrm{refl}\,R$. A reflexive $R$-module $M$ is called modifying if $\End_R(M)\in \mathrm{CM}\,R$. It gives an NCCR if, in addition, $\End_R(M)$ has finite global dimension. In the toric setting, the summands of $K$0 may be rank-one divisorial modules $K$1 indexed by the divisor class group (Hara et al., 2023, Tomonaga, 30 Oct 2025).

Write

$K$2

and fix a nonzero summand $K$3. With $K$4, one forms two approximation sequences. For a right $K$5-approximation of $K$6,

$K$7

the right mutation is

$K$8

Dually, for

$K$9

the left mutation is

$R$0

Iyama–Wemyss show that if $R$1 is modifying, then so are $R$2, and if $R$3 gives an NCCR, then so do $R$4; moreover $R$5 and $R$6 are derived equivalent (Tomonaga, 30 Oct 2025).

2. Exchange sequences, tilting realizations, and the three-dimensional case

For a $R$7-dimensional basic NCCR

$R$8

with $R$9 indecomposable and non-projective, one chooses a right $\Bbbk$0-approximation

$\Bbbk$1

fitting into

$\Bbbk$2

The right mutation at $\Bbbk$3 is then

$\Bbbk$4

Dually, one takes a right $\Bbbk$5-approximation of $\Bbbk$6 and dualizes to obtain $\Bbbk$7. In dimension $\Bbbk$8, these two operations agree up to grading shift, so one writes simply

$\Bbbk$9

In the $2$0-dimensional isolated Gorenstein case, if

$2$1

with $2$2, there is a unique minimal right $2$3-approximation

$2$4

and an exchange sequence

$2$5

dually one obtains $2$6, and in this isolated case one shows $2$7, so the right and left mutations coincide (Nordskova et al., 13 Apr 2026, Hara et al., 5 Mar 2026).

The categorical realization is by tilting. If $2$8 gives an NCCR, then there are canonical derived equivalences

$2$9

Concretely, one constructs a tilting bimodule $M\in \mathrm{refl}\,R$0 with

$M\in \mathrm{refl}\,R$1

In the cDV setting, the mutation can also be represented by a $M\in \mathrm{refl}\,R$2-term tilting complex

$M\in \mathrm{refl}\,R$3

with $M\in \mathrm{refl}\,R$4, producing inverse equivalences by $M\in \mathrm{refl}\,R$5 (Hara et al., 2023, 2207.13540).

A recurrent compression in the literature is to suppress the sign and write $M\in \mathrm{refl}\,R$6. That shorthand is justified only in the $M\in \mathrm{refl}\,R$7-dimensional isolated or graded setting described above; in general, right and left Iyama–Wemyss mutations are defined separately.

3. Rank-one toric singularities and upper-set combinatorics

For a Gorenstein toric singularity

$M\in \mathrm{refl}\,R$8

with class group $M\in \mathrm{refl}\,R$9 of rank one, Tomonaga gives a classification of toric NCCRs in terms of upper sets. After choosing generators $R$0 of positive class and $R$1 of negative class, one sets

$R$2

and defines a partial order on $R$3 by

$R$4

An upper set $R$5 is a subset such that $R$6 and $R$7; it is non-trivial if $R$8. To such an $R$9 one associates

$M$0

The key theorem is that

$M$1

gives a bijection between non-trivial upper sets $M$2 and splitting modules $M$3 which are NCCRs (Tomonaga, 30 Oct 2025).

Because the set of non-trivial upper sets is a finite lattice under inclusion, any two such upper sets are connected by elementary moves

$M$4

where $M$5 is minimal. In the toric module picture this operation corresponds exactly to right or left Iyama–Wemyss mutation at the summand $M$6. More precisely, if $M$7 and

$M$8

then

$M$9

By induction, any two splitting NCCRs are related by iterated Iyama–Wemyss mutations and hence are derived equivalent. In rank one, the only criterion for mutation is that $\End_R(M)\in \mathrm{CM}\,R$0 be minimal; geometrically one may think of $\End_R(M)\in \mathrm{CM}\,R$1 as the lattice points in the line $\End_R(M)\in \mathrm{CM}\,R$2, and upper sets are rays of the form $\End_R(M)\in \mathrm{CM}\,R$3 (Tomonaga, 30 Oct 2025).

The basic example is the $\End_R(M)\in \mathrm{CM}\,R$4-dimensional $\End_R(M)\in \mathrm{CM}\,R$5 toric singularity, where $\End_R(M)\in \mathrm{CM}\,R$6, the positive generators $\End_R(M)\in \mathrm{CM}\,R$7 have weight $\End_R(M)\in \mathrm{CM}\,R$8, the negative generators $\End_R(M)\in \mathrm{CM}\,R$9 have weight $\End_R(M)$0, and $\End_R(M)$1. Up to translation, the only non-trivial upper set is

$\End_R(M)$2

so

$\End_R(M)$3

Its unique minimal element is $\End_R(M)$4. Removing $\End_R(M)$5 gives $\End_R(M)$6, hence

$\End_R(M)$7

On the module side, taking $\End_R(M)$8, the left mutation produces $\End_R(M)$9. Concretely, the exact sequence

$K$00

resolves $K$01, and dualizing shows that the new summand is $K$02. Thus a single Iyama–Wemyss mutation replaces $K$03 by $K$04, exactly matching the combinatorics of $K$05 (Tomonaga, 30 Oct 2025).

4. Wall-crossing in geometric invariant theory

In the GIT framework of Hara–Hirano, let $K$06 be a generic quasi-symmetric representation of a connected reductive group $K$07, with character lattice $K$08. For a generic parameter $K$09 avoiding a finite $K$10-periodic hyperplane arrangement $K$11, one defines the magic window subcategory

$K$12

generated by the vector bundles

$K$13

Halpern–Leistner and Sam’s theorem identifies $K$14 with the derived category of the GIT quotient stack and, under the induced algebra description, with

$K$15

For adjacent generic parameters $K$16, the window-shift functor

$K$17

is induced by the tilting module

$K$18

and under the identifications above one has

$K$19

Thus wall-crossing equivalences between magic windows coincide with derived equivalences between NCCRs induced by tilting modules (Hara et al., 2023).

The same paper identifies these wall-crossings with Iyama–Wemyss mutation. Writing

$K$20

the matching $K$21 reproduces the exchange kernel, and one gets

$K$22

When $K$23 is a torus, exactly one summand crosses the wall, say $K$24. In that case, if

$K$25

then

$K$26

so the mutation is periodic of period $K$27. In the one-parameter Calabi–Yau complete-intersection setting, the corresponding window-shifts generate an action of

$K$28

on $K$29, with generators identified with appropriate compositions of Iyama–Wemyss mutations (Hara et al., 2023).

A concrete torus example is

$K$30

for which the affine quotient is the cone over the Segre cubic threefold. For any chamber between $K$31 and $K$32,

$K$33

Crossing the wall at $K$34 exchanges $K$35 with $K$36, and iterating the corresponding mutation six times returns to the original module, matching the general period-$K$37 statement (Hara et al., 2023).

5. cDV singularities, restricted roots, and BPS/GV wall-crossing

For a cDV singularity admitting an NCCR

$K$38

of affine Dynkin type $K$39, write

$K$40

and fix a vertex $K$41. Iyama–Wemyss define exchange sequences

$K$42

in $K$43-mod, where $K$44, and after dualizing,

$K$45

One then sets

$K$46

At the derived level, if $K$47, the object

$K$48

is a $K$49-term tilting complex with $K$50, giving derived equivalences

$K$51

$K$52

At the level of modules, the short exact sequences

$K$53

realize the two one-step mutations (2207.13540).

The same affine Dynkin data governs a restricted-root combinatorics. If

$K$54

is the restriction map, then

$K$55

is the set of restricted roots. A key vanishing theorem states that if $K$56 has $K$57 and

$K$58

then the noncommutative BPS invariant $K$59 vanishes. The associated hyperplane arrangement has walls

$K$60

and the chambers are labelled by the $K$61-term tilting complexes of $K$62 (2207.13540).

Mutation at $K$63 defines an element $K$64 in the wall-crossing groupoid. For adjacent $K$65-term tilting complexes, there is a derived equivalence between the corresponding stability-theoretic hearts,

$K$66

Passing to BPS invariants, if $K$67 is again given by a quiver with potential and $K$68 is indivisible and not colinear with $K$69 or $K$70, then

$K$71

In the geometric crepant-resolution picture, with $K$72 the genus-zero Gopakumar–Vafa numbers of a crepant resolution $K$73, one obtains

$K$74

and

$K$75

for every effective class $K$76. The example $K$77 realizes mutation at a vertex $K$78 as the corresponding finite-Weyl reflection on restricted roots, yielding explicit identifications among nonzero invariants (2207.13540).

6. Mutation cones and Bridgeland stability conditions

For a $K$79-dimensional complete local Gorenstein isolated singularity $K$80 and a basic maximal modifying $K$81-module

$K$82

Hara–Hirano define a wall-and-chamber structure in real $K$83-theory called the mutation cone. For any basic modifying module $K$84, write $K$85 and

$K$86

If $K$87 are the indecomposable projectives of $K$88, the open positive cone is

$K$89

For a basic tilting $K$90-module $K$91,

$K$92

is an open simplicial cone whose closure is polyhedral. Mutation functors induce isomorphisms on $K$93-groups; composing along a path $K$94 of simple mutations gives $K$95, and one shows that $K$96 depends only on the endpoints. Fixing $K$97, the mutation cone is

$K$98

The open chambers $K$99 are disjoint, their closures meet along codimension-one faces exactly when the corresponding modules differ by a single simple mutation, and there is a bijection

$R$00

Thus crossing a single wall is exactly applying one simple mutation (Hara et al., 5 Mar 2026).

The same paper introduces the tilting-noetherian property for $R$01, meaning that there are no infinite ascending chains in the poset of basic tilting $R$02-modules. It proves that, for maximal modifying modules, $R$03 is tilting-noetherian if and only if all maximal modifying $R$04-modules are connected by iterated mutations. This makes mutation connectivity an intrinsic finiteness property of the tilting theory (Hara et al., 5 Mar 2026).

On the triangulated side, let $R$05 be the full subcategory whose cohomology modules have support $R$06, and let $R$07 be its standard finite-length heart. A heart $R$08 is called modifying if it is obtained from a heart $R$09 by a composition of mutation functors. The resulting space $R$10 of modifying Bridgeland stability conditions is a disjoint union of chambers $R$11, and the forgetful map gives a regular covering

$R$12

where

$R$13

The subgroup $R$14 generated by mutation functors acts freely and transitively on the set of chambers and is precisely the Galois group of this covering. The same framework also describes the subgroup of autoequivalences preserving $R$15 in terms of mutation functors and class-group twists (Hara et al., 5 Mar 2026).

7. Anticanonical cones over del Pezzo surfaces and polygonal models

For a del Pezzo surface $R$16, let

$R$17

A very strong exceptional collection $R$18 of vector bundles on $R$19 is a full exceptional collection satisfying

$R$20

equivalently, with slopes increasing in the interval $R$21. Such a collection generates a geometric helix

$R$22

Any thread of $R$23 is very strong, and the rolled-up helix algebra

$R$24

is a $R$25-Calabi–Yau NCCR of $R$26. Conversely, every graded NCCR of $R$27 is Morita equivalent to such a rolled-up helix algebra (Nordskova et al., 13 Apr 2026).

In this setting, Iyama–Wemyss mutation can be read directly on the helix. A mutation of $R$28 at the summand corresponding to $R$29 is realized by a sequence of left or right braid mutations on a thread until the result is again a very strong thread, yielding a mutated helix $R$30. On the toric-system–polygon side of Hille–Perling, the helix is encoded by line segments

$R$31

whose midpoints trace a convex polygon $R$32 containing the origin. Mutation at index $R$33 is then a simple affine shear in the plane replacing $R$34 by a new segment parallel to $R$35, with the rest of the polygon shifted accordingly. Algebraically, this shear is the DWZ quiver-with-potential mutation, and it agrees with the Iyama–Wemyss mutation on the NCCR (Nordskova et al., 13 Apr 2026).

The polygon also encodes the quiver. If

$R$36

then convexity of $R$37 is equivalent to the very-strong condition, and the number of arrows $R$38 is

$R$39

with respect to the normalized volume form $R$40. A quiver mutation at vertex $R$41 is precisely the shear

$R$42

Up to tensoring the helix by a line bundle, shifting all objects in degree, permuting orthogonal blocks, and rotating the labeling, every helix can be reached from any other by a finite sequence of such quiver mutations. Accordingly, all NCCRs of anticanonical cones over del Pezzo surfaces are connected by a finite chain of Iyama–Wemyss mutations. This is presented as the non-commutative analogue of the statement that crepant resolutions are connected by flops (Nordskova et al., 13 Apr 2026).