Flip (algebraic geometry)¶

A small birational replacement over one base that turns a negative relative canonical or log-canonical direction into a positive one without contracting a divisor.

Version: v1 · 2026-10-17 · History
Domain-specific #: 14155
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomains: Birational Algebraic Geometry, Mori Theory → Mathematics
Aliases: Mori flip, Log flip

Core Idea¶

A flip replaces a small birational contraction that is negative for a canonical or log-canonical divisor with another small model over the same base on which the transformed divisor is positive. In a log pair \((X,\Delta)\), the input map \(f:X\to Z\) contracts curves on which \(K_X+\Delta\) is negative. The replacement \(f^+:X^+\to Z\) has \(K_{X^+}+\Delta^+\) positive on its contracted curves. Both maps are small: neither contracts a divisor. Thus \(X\dashrightarrow X^+\) changes the exceptional locus while agreeing in codimension one.^[1]^[2] When \(\Delta=0\), this is an ordinary \(K\)-flip; a nonzero boundary gives a log-pair instance.

The small input contraction alone is a flipping contraction, not the completed flip. The replacement supplies the positive side of the diagram. A flip is one local move in a minimal model program; whether a whole sequence of moves exists and terminates involves additional theorems with their own hypotheses.^[3] It does not follow from the definition that the replacement is smooth or improves every possible measure of singularity. The defining change is the relative sign of the selected divisor.

Structural Signature¶

Eligible input and directional divisor. A normal variety or suitable log pair supplies \(X\) and a canonical or log-canonical divisor whose intersection with contracted curves can be tested. The boundary \(\Delta\) may be zero.^[1]^[2]
Negative small contraction. The input \(f:X\to Z\) contracts a locus of codimension at least two, with \(K_X+\Delta\) negative on the contracted curves. A divisorial contraction or a crepant, zero-sign move does not fill this role.^[1]^[2]^[4]
Common base. The input and replacement both map to the same \(Z\), so positivity is judged relative to one contraction target. Two unrelated birational models do not form this diagram merely because they look similar.^[1]^[2]
Positive small replacement. A model \(X^+\) with \(f^+:X^+\to Z\) contracts no divisor and has \(K_{X^+}+\Delta^+\) positive on the replacement curves. Without it, the proposal has not supplied a flip.^[1]^[2]
Codimension-one continuity. The birational comparison preserves divisorial data while exchanging the lower-dimensional exceptional loci. A move that removes a divisor is of a different kind.^[1]^[2]

These roles separate the operation from its local picture. In Fujino and Sato's toric family the exceptional loci are projective spaces. In Hassett and Hyeon's moduli-space flip the affected loci parameterize different kinds of curves. Neither geometry is compulsory for every flip.^[1]^[2]

What It Is Not¶

An Atiyah flop is a close visual neighbor: it also replaces a small locus while preserving divisors, but the canonical divisor is trivial on the contracted curves. The absent negative-to-positive sign change makes it a flop rather than this strict flip.^[4] Calling every small birational replacement a flip erases the condition that directs the minimal model program.

A divisorial contraction fails smallness because it contracts a divisor. Hassett and Hyeon distinguish an earlier divisorial contraction of elliptic tails from their later small contraction of elliptic bridges.^[2] Likewise, constructing only the negative contraction does not prove that a positive replacement exists. A birational map over a different base, with no fixed divisor used on both sides, is not enough to pass the identity test.

Nor is a flip the entire Minimal Model Program. A program chooses and sequences possible extremal steps and asks about endpoints; a flip is one permitted step. Existence of klt flips is proved under the hypotheses of BCHM's Corollary 1.4.1, but that result must not be read as unrestricted termination of all flip sequences.^[3]

Scope of Application¶

The operation belongs literally to birational algebraic geometry when a small extremal contraction requires a positive model over its target. Fujino and Sato give an explicit toric construction: with their parameter inequality \(a>k/(n-k+1)\), \(X\to W\) is the negative side and \(X^+\to W\) the positive side. The specialization \((n,k)=(3,2)\), \(a\geq2\), is a three-dimensional flipping contraction. Their small maps are projective over \(W\); this does not assert that the source varieties are projective over a point.^[1]

The same structure occurs in log birational models of moduli spaces. For \(g\geq4\), Hassett and Hyeon construct \(\Psi:\overline M_g(7/10+\epsilon)\to\overline M_g(7/10)\), contracting an elliptic-bridge locus, and its small Mori flip \(\Psi^+:\overline M_g(7/10-\epsilon)\to\overline M_g(7/10)\). The input is the pseudostable model; the common base is a Chow GIT quotient, and the flipped model is a Hilbert GIT quotient.^[2] The relative sign comparison uses the same fixed log divisor \(K+(7/10-\epsilon)\Delta\) on both sides. The \(+\epsilon\) in the input model label must not be substituted as its sign-testing divisor.^[2]

Clarity¶

The flip diagram resolves three easy conflations. First, small says what is not removed: a divisor survives the birational comparison, even though curves or other codimension-at-least-two loci can be exchanged. Second, negative then positive describes the chosen canonical or log-canonical direction, not a promise of global smoothness or improvement by every metric. Third, over the same base is what makes the two models the two sides of one move.^[1]^[2]

The moduli example is a useful test of the language. An elliptic-bridge Curve can be represented on one side and a tacnodal curve on the other, but the ambient birational transformation acts on moduli-space models, not directly on one curve as a standalone variety.^[2] The labels of those models tell where the spaces occur in a log-MMP sequence; they do not replace the fixed divisor needed to test the flip's sign.

Manages Complexity¶

An explicit flip may be presented through fans, equations, GIT quotients, singularity conditions, or a parameterized family. The common diagram reduces this variety of presentations to five checks: eligible input, negative small map, shared base, positive small map, and codimension-one agreement. A proposed example that supplies only the first two checks is still waiting for its replacement model.

In the toric construction, fan subdivisions and a numerical inequality certify the two relative signs; the exceptional loci are \(\mathbb P^{n-k}\) and \(\mathbb P^{k-1}\), each contracted to a point of \(W\).^[1] In the moduli construction, stability conditions and GIT quotients identify the common base and flipped model, while the log polarization proves the required positive side.^[2] The role map lets one compare these otherwise unlike proofs without pretending their technical tools are interchangeable.

Abstract Reasoning¶

Given a claimed flip, ask which single divisor directs it. Is the input contraction small and negative for that divisor on the contracted curves? Is there a second small map to the same base for which the transformed divisor is positive? Are the two models identified outside their exceptional loci? If any answer is missing, the claim may name a flipping contraction, a flop, a divisorial move, or a different birational construction.

The sign test also prevents a subtle error in a variation-of-models example. Hassett and Hyeon's source model is named with \(7/10+\epsilon\), but its negative relative direction is tested using \(K+(7/10-\epsilon)\Delta\); the flipped \(7/10-\epsilon\) model is positive for that same log direction.^[2] Comparing a \(+\epsilon\) divisor on the source with a \(-\epsilon\) divisor on the target would be changing the measuring rule mid-comparison.

Knowledge Transfer¶

The toric example and the moduli-space example transfer the identity test, not the construction method. The former uses explicit fan changes and an ordinary \(K\)-direction. The latter uses a boundary divisor, stability conditions, and GIT to build a log flip. In both, a negative small contraction and positive small replacement share a base and preserve divisors.^[1]^[2]

This transfer has limits. A projective-space exceptional locus is not required in moduli geometry; a Chow/Hilbert quotient is not required in toric geometry. Existence for one family does not prove existence under arbitrary singularity assumptions. BCHM supplies a general klt existence theorem under its stated contraction hypotheses, while questions about termination concern sequences, not the defining roles of an individual flip.^[3]

Examples¶

Canonical — a threefold toric flip¶

Fujino and Sato's Example 3.3.2 builds toric varieties \(X\) and \(X^+\) mapping to \(W\). For \(a>k/(n-k+1)\), the map from \(X\) is a negative flipping contraction and the map from \(X^+\) is positive. Their \((n,k)=(3,2), a\geq2\) case gives a three-dimensional instance. The two exceptional loci are projective lines in that specialization, but the specific fan rays and parameter are features of the example, not of every flip.^[1]

Mapped back: toric \(X\) and \(K_X\) → negative small \(X\to W\) → shared \(W\) → positive small \(X^+\to W\) and \(K_{X^+}\) → codimension-one identification away from the exchanged curves.

Applied — the first log flip of stable-curve moduli¶

Hassett and Hyeon contract elliptic-bridge loci in a pseudostable moduli-space model and construct a Hilbert-quotient model mapping to the same Chow-quotient base. Their Theorem 2.12 proves that the second small map is the flip of the first. The affected represented curves can shift from elliptic bridges to tacnodal curves, while the flip itself transforms the ambient moduli-space models.^[2]

Mapped back: moduli-space log pair and fixed \(K+(7/10-\epsilon)\Delta\) → negative small \(\Psi\) → shared \(\overline M_g(7/10)\) → positive small \(\Psi^+\) → agreement off the small exceptional loci.

Structural Tensions¶

No opposed design objectives are constitutive of a flip. Its negative and positive sides are defining conditions, not values to balance. A separate proof distinction often matters: constructing one local flip does not establish that a sequence of such flips terminates. BCHM proves flip existence in a specified klt setting and gives scoped results about running the program; the two obligations should be stated separately.^[3] Diagnostic: Is a claim proving this particular positive replacement, or proving that every relevant sequence eventually stops?

Structural–Framed Character¶

The flip sits near the structural end of the spectrum within algebraic geometry. Its two small morphisms, common target, divisor signs and codimension-one comparison can be checked in a toric fan construction and in a moduli-space construction without copying either example's appearance.^[1]^[2] The sign is a mathematical test, so the name carries little evaluative weight: “positive” here is a relative intersection property, not an endorsement of the resulting variety.

Its definition depends on proof and the hypotheses of birational geometry, rather than a discretionary human practice such as a locally chosen convention.^[3] Its MMP vocabulary arose in a mathematical research community, but no particular institution or authority creates a flip by designating one. The generic words “replacement” and “preservation” travel widely; the technical vocabulary of small contractions and canonical divisors does not travel to unrelated domains with the same identity. A new algebraic-geometric instance is recognized by its diagram and sign tests, not imported by analogy from Fujino and Sato's toric construction or the moduli example. A visually similar exchange with zero canonical sign fails and is a flop.^[4] Its character: a reusable, formally recognizable birational move whose field-specific conditions remain decisive.

Structural Core vs. Domain Accent¶

The most general skeletal relation is a constrained replacement of one model by another over a shared reference, preserving some data while changing a localized part. This satisfies the broad Prime Transformation: \(X\) is the input, the small-contraction and sign conditions govern the restructuring, \(X^+\) is the output, and birational/divisorial data persist while the exceptional locus changes. The flip's domain-bound mechanism is constitutive: normal algebraic varieties or log pairs, canonical divisors, contracted curves, a common base, small morphisms, and the negative-to-positive relative sign. Remove those and the name no longer identifies an algebraic-geometric flip.^[1]^[2]

A proposed future cross-domain Prime for the more specific paired-replacement skeleton would need at least two unlike non-algebraic systems with a single fixed directional test that reverses across two models over the same reference while specified invariants survive. If a proposed example merely replaces one thing with another, or changes its success measure between the two sides, it fails that test. Software rewriting and organizational reconfiguration can suggest such questions, but the present sources establish no matching cross-domain cases. The two positive examples demonstrate transfer within birational geometry, which is why the named flip remains domain-specific.

This entry is a kind of Transformation.

The flip strictly instantiates Transformation: its rule-governed birational comparison changes the small exceptional locus while preserving the function field and codimension-one divisorial agreement. The Transformation Prime also covers many mappings without a canonical divisor or small contraction, so the flip is a narrower kind. Minimal Model Program is the larger procedure in which a flip can be one move; it is not synonymous with that move. Mori dream space names a class of varieties, not this operation. The stricter Equivalence-Preserving Rewriting requires an additional cost-selection architecture that an individual flip need not have.

Relationships to Other Abstractions¶

Current abstraction Flip (algebraic geometry) Domain-specific

Parents (1) — more general patterns this builds on

Flip (algebraic geometry) is a kind of Transformation Prime

Every flip is a rule-governed birational transformation that preserves codimension-one data while replacing a small exceptional locus and reversing the relative canonical or log-canonical sign.

Hierarchy path (1) — routes to 1 parentless root

Flip (algebraic geometry) → Transformation → Function (Mapping)

Neighborhood in Abstraction Space¶

Flip (algebraic geometry) sits in a sparse region of the domain-specific corpus (92^nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2721 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-17

Not to Be Confused With¶

Flop: a small crepant birational replacement with canonical divisor trivial on the contracted curves, not the negative-to-positive flip.^[4]
Divisorial contraction: a birational contraction that removes a divisor and fails the smallness requirement.^[2]
Flipping contraction: the negative small input map before its positive replacement has been constructed.^[1]
Minimal Model Program: the larger sequence of possible contractions, flips and endpoints, rather than one local move.^[3]
All-dimensional termination theorem: neither the definition of a flip nor the cited existence result asserts unrestricted termination of every sequence.^[3]

References¶

[1] Osamu Fujino and Hiroshi Sato, Notes on toric varieties from Mori theoretic viewpoint, II, Nagoya Mathematical Journal 239 (2020), 42–75, §3.3 Example 3.3.2 and Remark 3.3.3. Original full-text construction; the small maps are projective over the common base. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o

[2] Brendan Hassett and Donghoon Hyeon, Log minimal model program for the moduli space of stable curves: The first flip, arXiv:0806.3444v1 (2008), Abstract, §1, Proposition 2.9, Theorem 2.12, Corollary 2.15 and §9 proof of Theorem 2.12. Original full-text preprint; for \(g\geq4\) and the fixed log divisor \(K+(7/10-\epsilon)\Delta\). registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s

[3] Caucher Birkar, Paolo Cascini, Christopher D. Hacon and James McKernan, Existence of Minimal Models for Varieties of Log General Type, arXiv:math/0610203, §1.4 Corollary 1.4.1 and following qualification on termination. Original full-text theorem for the stated klt setting; not a universal termination claim. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g

[4] Masayuki Kawakita, Complex Algebraic Threefolds, Chapter 1, Example 1.3.18, publisher full-chapter preview pp. 38–39. Specialist account of the K-trivial Atiyah flop, used here only for the negative boundary. registry ↩a ↩b ↩c ↩d