A gerbe is locally inhabited and locally connected by isomorphisms, yet it need not possess one global object. Its local objects can all have isomorphic automorphism groups while their overlap data carries a global obstruction. That makes a gerbe a natural next step after constructible stacks and descent.
The band records the automorphism type in a way that survives changing local objects, including the conjugacy ambiguity that appears for non-abelian groups. But the band is not the whole gerbe. Two gerbes can share the same local stabiliser type and band while representing different global twisting classes.
The Gold Standard is to choose local objects, write torsors of isomorphisms, track double- and triple-overlap data, identify the band and equivalence notion, and test whether the cocycle can be removed by re-trivialisation.
A practical orientation
Use this guide after the constructible-stack descent article. The original worked examples compare trivial and twisted U(1)-banded gerbes on a good cover and use a finite-group model to expose non-abelian conjugacy limits.
Choose a reading route
Build the gerbe locally
Chapters 1–3 define local non-emptiness, connectedness and band data.
Work and transfer examples
Chapters 7–9 compare twisting, refinement and non-abelian changes.
State classification limits
Chapters 10–11 give cohomological boundaries and evidence checks.
Chapter contents
- 1. Start with local objects and local isomorphisms
- 2. Understand what the band remembers
- 3. Neutral gerbes have a global object
- 4. Choose objects on opens and arrows on overlaps
- 5. Compare with line bundles and ordinary descent
- 6. Test equivalence through re-trivialisation
- 7. Worked example: one band, two global classes
- 8. Transfer check: refine the cover and change trivialisation
- 9. Non-abelian bands require conjugacy-aware bookkeeping
- 10. State the classification theorem boundary
- 11. Limits, misconceptions and practical evidence check
- 12. FAQs and the next reading route
1. Start with local objects and local isomorphisms
Over every point there is a neighborhood on which the stack has an object, and any two local objects become isomorphic after passing to a suitable cover. Global existence is not assumed. This local non-emptiness and local connectedness distinguishes a gerbe from an arbitrary stack of categories.
Automorphisms of a chosen local object form a sheaf of groups. Choosing another local object identifies these groups only up to conjugacy in the non-abelian case, which is why a band is subtler than one globally fixed group label.
- Locally non-empty.
- Locally any two objects are isomorphic.
- Automorphism sheaves attached to objects.
- Change of object creates conjugacy.
- Global object may not exist.
2. Understand what the band remembers
For an abelian band A, local automorphism sheaves can be identified coherently with A. In a non-abelian setting, inner automorphisms intervene, so the band records group data modulo the appropriate conjugacy ambiguity.
The band tells us the local symmetry type and how it is transported. It does not say whether the gerbe is neutral, nor does it alone determine the global cocycle.
| Record | Remembers | Can miss |
|---|---|---|
| Local stabiliser group | Automorphisms in one chart | Change of object |
| Band | Transported symmetry type | Global twisting class |
| Gerbe class | Higher gluing obstruction | Extra connection data |
| Gerbe with connection | Differential refinement | Unsupported geometric claims |
3. Neutral gerbes have a global object
A gerbe is neutral when it has a global object. Relative to that object, the gerbe is equivalent to the stack of torsors for its automorphism sheaf. Neutral does not mean empty of structure; it means the obstruction to a global object has vanished.
Local objects can exist everywhere without gluing to a global one. The failure is measured by higher overlap data, not by a missing chart.
| Condition | Local objects | Global object | Interpretation |
|---|---|---|---|
| Neutral gerbe | Yes | Yes | Equivalent to torsors for band after choice |
| Non-neutral gerbe | Yes locally | No | Descent obstruction persists |
| Empty stack region | No | No | Not a gerbe there |
4. Choose objects on opens and arrows on overlaps
On a cover {Ui}, choose objects xi. On Uij choose isomorphisms fij: xj→xi. Their composite on a triple overlap need not equal fik; the difference is an automorphism gijk of xi.
The elements gijk are not arbitrary. On fourfold overlaps they satisfy a coherence condition, modified by the action of the transition isomorphisms in the non-abelian case. This is the higher descent content that local stabilisers alone omit.
| Cover level | Datum | Role |
|---|---|---|
| Ui | Object xi | Local trivialisation |
| Uij | Isomorphism fij | Compare objects |
| Uijk | Automorphism gijk | Measure composition defect |
| Uijkl | Coherence equation | Make defect globally consistent |
5. Compare with line bundles and ordinary descent
A line bundle can be described by functions on double overlaps satisfying a condition on triple overlaps. A U(1)-gerbe can be described by U(1)-valued functions on triple overlaps satisfying a condition on fourfold overlaps.
This degree shift explains why bundle gerbes naturally represent degree-three integral cohomology classes under suitable hypotheses, while circle bundles represent degree-two classes.
| Geometric object | Local transition data | Characteristic degree |
|---|---|---|
| Function | Values on opens | 0 |
| Circle bundle | Functions on double overlaps | 2 integral class |
| U(1)-gerbe | Functions on triple overlaps | 3 integral class |
| 2-gerbe | Higher data | 4 integral class in suitable models |
6. Test equivalence through re-trivialisation
Changing local objects or overlap isomorphisms changes the displayed cocycle. Two cocycles can nevertheless define equivalent gerbes when their difference is a coboundary produced by that change of trivialisation.
The invariant is therefore a class, not a preferred list of numbers. A calculation that declares a gerbe nontrivial from one nonzero gijk without testing re-trivialisation is incomplete.
- Local-object change recorded.
- Overlap-arrow change recorded.
- Cocycle recomputed.
- Coboundary relation tested.
- Equivalence category stated.
7. Worked example: one band, two global classes
Take a good cover of a space with a nonzero degree-three integral class. Put equivalent local categories of U(1)-torsors on every open. In the trivial case choose triple-overlap functions whose class vanishes. In the twisted case choose coherent triple-overlap functions representing the nonzero class.
Every open sees the same U(1) automorphism group, and every double overlap admits the required comparison. The difference appears only after the triple data are assembled and tested modulo re-trivialisation.
The trivial gerbe has a global object. The twisted gerbe does not, even though its local stabiliser type and band are the same. That is precisely the information lost by a local isotropy inventory.
| Feature | Trivial gerbe | Twisted gerbe |
|---|---|---|
| Local objects | Exist | Exist |
| Band | U(1) | U(1) |
| Double-overlap comparison | Exists | Exists |
| Cocycle class | Zero | Nonzero |
| Global object | Exists | Obstructed |
8. Transfer check: refine the cover and change trivialisation
Pull both cocycles to a common refinement. Their classes should persist even though the number and labels of overlaps change. Then replace the chosen local objects and overlap arrows; the representative changes but the class should remain.
A conclusion that disappears under a valid refinement was tied to bookkeeping. A conclusion that changes under re-trivialisation confused a representative with the gerbe.
| Changed context | Expected invariant | Failure exposed |
|---|---|---|
| Cover refinement | Gerbe class | Cover-dependent calculation |
| Local object choice | Equivalence class | Presentation dependence |
| Overlap arrow choice | Class modulo coboundary | Representative mistaken for invariant |
| Pullback along map | Functorial class change | Untracked base-space change |
9. Non-abelian bands require conjugacy-aware bookkeeping
For a non-abelian group G, automorphism identifications depend on chosen local objects and are well defined only up to inner automorphism. Overlap data acts on G, and the triple defect is twisted by that action.
Do not reuse an abelian cocycle formula by commuting terms. A finite-group teaching model should write the order of multiplication, action and conjugation explicitly, then state which weaker invariant survives if full classification is out of scope.
| Abelian case | Non-abelian change | Practical consequence |
|---|---|---|
| Automorphisms commute | Order matters | Write composition order |
| Fixed group sheaf | Band up to inner action | Track conjugacy |
| Ordinary cocycle addition | Twisted product law | Use correct coherence |
| H²-style class notation | Pointed/non-abelian classification | State theorem precisely |
10. State the classification theorem boundary
For U(1)-bundle gerbes on a manifold, stable isomorphism classes correspond to degree-three integral cohomology under the standard bundle-gerbe theory. Breen’s cocycle descriptions treat non-abelian 1- and 2-gerbes with explicit local trivialisations and higher coherence.
These statements do not imply that every stack with local automorphisms is classified by H³, nor that a band alone supplies a class. Name the site, band, regularity, equivalence notion and connective data before invoking a classification.
| Claim | Needed context | Safe conclusion |
|---|---|---|
| U(1)-bundle gerbe class | Manifold and stable isomorphism | Degree-three integral class |
| Neutrality | Global object or zero obstruction | Equivalent to torsor stack after choice |
| Non-abelian gerbe | Band and non-abelian descent theorem | Context-specific classification |
| Connection refinement | Connective structure and curvature | Differential data retained |
11. Limits, misconceptions and practical evidence check
The same band does not imply the same gerbe, and local equivalence does not imply a global object. A nonzero representative does not prove nontriviality until coboundaries and allowed equivalences are tested.
Record the site and cover, local objects, automorphism sheaves, band identifications, overlap arrows, triple defects, fourfold coherence, re-trivialisations, refinements and theorem hypotheses. Recompute the class on a common refinement.
- Band is not the full class.
- Local objects need not globalise.
- Representatives tested modulo coboundary.
- Non-abelian order explicit.
- Theorem hypotheses preserved.
12. FAQs and the next reading route
Gerbes make higher gluing concrete: the same local symmetry can support different global twisting because the decisive information lives in coherent overlap data.
Read the Topology root, the exit-path article and the constructible-stack descent article before this guide. The Gold Standard parent and How X Works hub connect the same discipline across subjects: preserve comparison maps, test changed contexts and state the theorem boundary.
What is the simplest difference between a bundle and a gerbe?
A bundle uses transition data on double overlaps with a triple-overlap condition; a gerbe shifts the decisive data and coherence one level higher.
Does knowing the band determine the gerbe?
No. The band records transported local automorphism type, while distinct global twisting classes can share that band.
Why can a gerbe have local objects but no global object?
The local objects may fail to satisfy effective global descent because their higher overlap cocycle is nontrivial.
Continue through the verified ecosystem
Sources and scope note
- Breen — Notes on 1- and 2-Gerbes — used for explicit non-abelian cocycles, local trivialisations and higher gerbe coherence.
- Murray and Stevenson — Bundle Gerbes: Stable Isomorphism and Local Theory — used for stable isomorphism, local theory and degree-three classification of bundle gerbes.
- Hitchin — On the paper “Bundle gerbes” by Michael Murray — used for the local-overlap hierarchy, bundle-gerbe definition and modern scope of the theory.
The cover calculations and finite-group comparisons are original teaching examples. The sources support gerbe, band and stable-isomorphism concepts; the article does not claim a universal classification outside their stated settings.