Bordism and Tangential Structures: a Primer

Bordism adds an equivalence relation on closed smooth manifolds: two structured $n$-manifolds are equivalent when they are the boundary pieces of one compact structured $(n+1)$-manifold. With the outward-normal-first orientation convention, a bordism from $M_0$ to $M_1$ satisfies

$$\partial W\cong (-M_0)\sqcup M_1.$$

Every structure-preserving diffeomorphism gives a cylinder bordism, but bordism can also identify manifolds of different homotopy type. It preserves every datum declared to be part of the structure: an orientation, spin structure, framing, map to a background space, or bundle must extend across $W$ and restrict to the specified data at both ends. The dimension, smooth category, tangential structure, and background data are therefore part of the question, not optional labels.

Helpful background. Homotopy, Degree, Winding, and Covering Spaces supplies the comparison with equivalence of maps, while Characteristic Classes and Chern–Weil Theory supplies characteristic-class obstructions to filling a structured manifold.

Bordism data and smooth setting

The setting is finite-dimensional smooth topology. Every representative $M^n$ is closed, meaning compact and without boundary; a bordism $W$ is compact and may have boundary. Manifolds may be disconnected. Boundary collars are understood, so structure-preserving gluing is smooth. The page uses stable tangent structures and the outward-normal-first boundary convention. Topological, piecewise-linear, singular, noncompact, and manifolds-with-corners variants require their own definitions.

This is a primer on the equivalence relation and its input data. It does not construct bordism categories, prove the Pontryagin–Thom theorem, compute general bordism groups, or classify topological field theories, invertible phases, or anomalies.

The bordism relation

First ignore additional structure. A smooth bordism from closed $n$-manifolds $M_0$ to $M_1$ consists of a compact smooth $(n+1)$-manifold $W$, a decomposition of its boundary into incoming and outgoing components, and identifications of those components with $M_0$ and $M_1$. Product collars

$$[0,\epsilon)\times M_0, \qquad (-\epsilon,0]\times M_1$$

are either included in the data or chosen before gluing. In the unoriented theory one may summarize the boundary as $\partial W\cong M_0\sqcup M_1$. In the oriented theory, the boundary identification is

$$\partial W\cong (-M_0)\sqcup M_1,$$

where $-M_0$ is $M_0$ with the opposite orientation. A structured version also requires the boundary identifications to preserve the declared structure. The older word cobordism is often used for the same geometric object; this page uses bordism for both the object and the induced equivalence relation.

A closed structured manifold $M$ is null-bordant when it is the entire structured boundary of a compact manifold one dimension higher. Equivalently, it is bordant to the empty manifold. Writing $M=\partial W$ is therefore a strong existence statement: an actual smooth $W$ and all required extensions must exist.

Freed 2013, Lecture 1, Definition 1.19 and Lemma 1.25, PDF gives a collared definition and the following proof that bordism is an equivalence relation.

Reflexivity. The cylinder $[0,1]\times M$ is a bordism from $M$ to $M$. Its product structure extends any stable tangential or background data on $M$.

Symmetry. Apply stable reflection in one trivial line to the $\theta$-structure on $W$, producing its opposite structured manifold $W^\vee$, and then exchange the incoming and outgoing collars. For the unoriented structure this reflection is invisible. For the oriented structure it reverses the orientation of $W$: if $\partial W=(-M_0)\sqcup M_1$, then

$$\partial(-W)=(-M_1)\sqcup M_0,$$

so $-W$ is a bordism from $M_1$ to $M_0$. Exchanging collars without also applying the opposite operation to the interior structure does not prove symmetry for a general $\theta$.

Transitivity. If $W_{01}$ runs from $M_0$ to $M_1$ and $W_{12}$ runs from $M_1$ to $M_2$, identify their $M_1$ boundary components. The collars combine into a bicollar $(-\epsilon,\epsilon)\times M_1$, and compatible structures glue across it. The result

$$W_{02}=W_{01}\mathbin{\cup_{M_1}}W_{12}$$

is a smooth structured bordism from $M_0$ to $M_2$. Without collars and agreement of the restricted structures, this gluing step has not been justified.

Disjoint union gives a bordism group

Fix a stable tangential structure $\theta$ and a dimension $n$. Let $\Omega_n^\theta$ denote the set of bordism classes of closed $\theta$-manifolds. Disjoint union defines

$$[M]+[N]=[M\sqcup N].$$

The empty $n$-manifold is zero, and swapping components proves commutativity. For a general $\theta$-manifold $M$, let $M^\vee$ denote the opposite boundary structure appearing at the other end of the structured cylinder. Then

$$\partial([0,1]\times M)\cong M^\vee\sqcup M,$$

so $[M^\vee]=-[M]$. In oriented bordism, $M^\vee=-M$. In unoriented bordism, the opposite structure is indistinguishable from the original one, so every class is its own inverse and $2[M]=0$. One should not identify the inverse with orientation reversal when no orientation is part of the structure.

The first groups can be computed from elementary manifold classification; the completeness of the degree-two calculation is proved in Freed 2013, Lecture 1, Proposition 1.32, PDF.

Dimension zero. Unoriented bordism gives $\Omega_0^O\cong\mathbb Z/2$, while oriented bordism gives $\Omega_0^{SO}\cong\mathbb Z$. A compact $1$-manifold has an even number of boundary points; orientation replaces parity by signed count.

Dimension one. Both groups vanish: $\Omega_1^O=\Omega_1^{SO}=0$. Every closed $1$-manifold is a union of circles, and $S^1=\partial D^2$.

Dimension two. Here $\Omega_2^O\cong\mathbb Z/2$, generated by $\mathbb{RP}^2$, whereas $\Omega_2^{SO}=0$. Every closed oriented surface bounds a handlebody; $\mathbb{RP}^2$ has a mod-$2$ obstruction to bounding.

These examples expose what the quotient forgets. Both $S^2$ and $T^2$ are oriented boundaries, of $D^3$ and a solid torus respectively. They represent the same class in $\Omega_2^{SO}$ although they are neither diffeomorphic nor homotopy equivalent. Thus bordism can identify distinct homotopy types. Structure-preserving diffeomorphism always implies bordism through a cylinder, but there is no blanket ordering between structured bordism and ordinary homotopy equivalence.

Tangential structures are chosen lifts

Let $BO=\operatorname*{colim}_r BO(r)$ classify stable real vector bundles, and let

$$\tau_M:M\longrightarrow BO$$

classify the stable tangent bundle of $M$. A stable tangential structure is represented by a fibration

$$\theta:B\longrightarrow BO.$$

A $\theta$-structure on $M$ is a chosen lift $\widetilde\tau_M$ together with the indicated homotopy:

$$\begin{array}{ccc} &B&\\ &\mathrel{\nearrow}^{\widetilde\tau_M}&\mathrel{\downarrow}^{\theta}\\ M&\mathrel{\underset{\tau_M}{\longrightarrow}}&BO. \end{array}$$

Equivalently, for sufficiently large $r$, it is compatible structure on

$$TM\oplus\varepsilon^r.$$

The word chosen matters. Saying that a lift exists is not the same as specifying one, and distinct lifts on the same underlying manifold may represent distinct bordism classes. Stabilization also matters: a stable framing is a trivialization after adding trivial line bundles, not necessarily a framing of $TM$ itself.

Common examples are:

• $BO\xrightarrow{\mathrm{id}}BO$. There is no reduction of the stable tangent bundle; this is unoriented bordism.
• $BSO\to BO$. The structure is an orientation, with obstruction $w_1(TM)$.
• $B\mathrm{Spin}\to BO$. The structure is an orientation and spin lift. Existence requires $w_1(TM)=w_2(TM)=0$. For each fixed orientation, spin lifts—when they exist—form an $H^1(M;\mathbb Z/2)$-torsor.
• $EO\to BO$. Since $EO$ is contractible, a lift chooses a null-homotopy of the stable tangent classifier: equivalently, a stable framing or stable trivialization of $TM$.
• $B\times X\to B\xrightarrow{\theta}BO$. The structure is a $\theta$-structure together with an independent map $f:M\to X$.

The detailed spin construction belongs to Spin Structures and Dirac Operators. The lift formulation and these examples are developed in Freed 2013, Lecture 9, Definition 9.45 and Examples 9.49–9.54, PDF.

Not every physical symmetry type is a direct product $B\times X$, but this product is the clean model for independent background data. For example, a map $f:M\to BG$ classifies a topological principal $G$-bundle. It does not by itself specify a connection. A metric, connection, differential cocycle, or boundary condition is additional geometric data and needs its own extension rule.

How structure reaches the boundary

Along the boundary of a smooth manifold $W$ there is a stable splitting

$$TW\big|_{\partial W} \cong T(\partial W)\oplus\varepsilon^1_{\mathrm{out}},$$

where the last line is trivialized by the outward normal. Restricting a stable tangent lift on $W$ and using this trivial line induces a lift on $\partial W$. This is why the stable formulation passes naturally between dimensions $n+1$ and $n$.

For an oriented $W$, outward-normal-first means

$$o(TW) = o(\varepsilon^1_{\mathrm{out}})\wedge o(T\partial W).$$

Give $[0,1]\times M$ the product orientation $\mathrm dt\wedge o(TM)$. At $t=1$ the outward normal is $+\partial_t$, so the induced orientation is $+M$. At $t=0$ the outward normal is $-\partial_t$, so it is $-M$. Hence

$$\partial([0,1]\times M)=(-M)\sqcup M,$$

which simultaneously checks the sign in the definition and the oriented group inverse.

One circle, two spin structures

Fix an orientation on the circle. It has two compatible spin structures because their isomorphism classes form a torsor for

$$H^1(S^1;\mathbb Z/2)\cong\mathbb Z/2.$$

Only one is induced from the unique compatible spin structure on the oriented disk. In the standard fermionic terminology, the bounding structure is antiperiodic, or Neveu–Schwarz, while the periodic, or Ramond, structure is nonbounding. Thus the same oriented circle is zero with one spin structure and nonzero with the other. Forgetting the chosen lift loses information that structured bordism is designed to retain. The terminology and the mod-$2$ index detecting the nonbounding circle are reviewed by Gaiotto and Johnson-Freyd 2023, § 3.1.

Stable tangent and stable normal conventions

Some sources formulate bordism using stable normal bundles. If $M\hookrightarrow\mathbb R^N$ is an embedding with normal bundle $\nu_M$, then

$$TM\oplus\nu_M\cong\varepsilon^N, \qquad [\nu_M]=[\varepsilon^N]-[TM]$$

in $KO^0(M)$. After subtracting ranks, the reduced classes satisfy

$$\widetilde{[\nu_M]}=-\widetilde{[TM]} \quad\text{in }\widetilde{KO}^0(M).$$

The stable normal class is independent of the chosen high-dimensional embedding, but it is the complement of the tangent class, not the same class.

Let $\iota:BO\to BO$ classify stable complementation. From $\theta:B\to BO$ form the pullback structure

$$\begin{array}{ccc} B^\perp&\longrightarrow&B\\ \downarrow&&\downarrow\theta\\ BO&\xrightarrow{\iota}&BO. \end{array}$$

A tangential $B$-lift of $TM$ corresponds to a normal $B^\perp$-lift of $\nu_M$. For $O$, $SO$, and $\mathrm{Spin}$ there are familiar identifications between the two descriptions. In general there need not be; in particular, tangent-to-normal translation can exchange the two Pin conventions. This page always uses stable tangent data.

The characteristic classes provide a useful check. From $TM\oplus\nu_M\cong\varepsilon^N$,

$$w(TM)w(\nu_M)=1.$$

Comparing degrees one and two gives

$$\begin{aligned} w_1(\nu_M)&=w_1(TM),\\ w_2(\nu_M)&=w_2(TM)+w_1(TM)^2. \end{aligned}$$

On an oriented manifold $w_1=0$, so the tangent and normal spin obstructions agree. Outside that case, silently using the same symbol for both structures can change the problem. Debray 2019, Remark 2.14, PDF and Freed 2013, Lecture 9, §§ 9.62–9.69, PDF make this complement operation explicit.

Background spaces and bordism over a target

For an independent background space $X$, a cycle for $\Omega_n^\theta(X)$ is a triple

$$(M^n,\widetilde\tau_M,f), \qquad f:M\longrightarrow X.$$

A bordism between $(M_0,\widetilde\tau_0,f_0)$ and $(M_1,\widetilde\tau_1,f_1)$ consists of a compact $\theta$-manifold $W$ and a map

$$F:W\longrightarrow X$$

whose restrictions agree with $f_0$ and $f_1$. It is not enough for the underlying manifolds to be bordant. The background map must extend too.

Take $X=BU(1)$. A map $f:\Sigma\to BU(1)$ classifies a complex line bundle $L\to\Sigma$. If $(W,\widetilde L)$ is an oriented bordism from $(\Sigma_0,L_0)$ to $(\Sigma_1,L_1)$, then $c_1(\widetilde L)$ restricts to the two boundary classes, and

$$\begin{aligned} 0 &= \left\langle c_1(\widetilde L)\big|_{\partial W}, [\partial W] \right\rangle \\ &= -\left\langle c_1(L_0),[\Sigma_0]\right\rangle +\left\langle c_1(L_1),[\Sigma_1]\right\rangle. \end{aligned}$$

The first Chern number is therefore preserved by bordism over $BU(1)$. Although the underlying $S^2$ bounds $D^3$, a pair $(S^2,L_n)$ with

$$\left\langle c_1(L_n),[S^2]\right\rangle=n\ne0$$

cannot be a boundary over $BU(1)$. This is a topological bundle statement; specifying a connection and requiring it to extend would be a stronger geometric problem.

Characteristic numbers obstruct fillings

The preceding calculation is the general mechanism. Suppose a closed $n$-form $\widetilde\omega$ or a degree-$n$ characteristic class on $W$ restricts to $\omega_i$ on $M_i$. Then Stokes’ theorem, or the relative fundamental-class pairing, gives

$$0 = \int_{\partial W}\widetilde\omega = -\int_{M_0}\omega_0 +\int_{M_1}\omega_1.$$

Thus characteristic numbers built naturally from data that extends across $W$ are constant on bordism classes. A nonzero value on a proposed boundary is an obstruction. Real differential forms see only the real image of these classes, so torsion and mod-$2$ obstructions require integral or $\mathbb Z/2$ methods.

For example, if $a\in H^1(\mathbb{RP}^2;\mathbb Z/2)$ is the generator, then the stable tangent-bundle identity gives

$$w(T\mathbb{RP}^2) =(1+a)^3 =1+a+a^2.$$

Consequently

$$\left\langle w_1^2,[\mathbb{RP}^2]_2\right\rangle=1,$$

so $\mathbb{RP}^2$ is not an unoriented boundary. In the oriented theory, the signature is a bordism invariant and $\sigma(\mathbb{CP}^2)=1$, so $\mathbb{CP}^2$ is not an oriented boundary. These are obstructions, not a claim that one chosen number classifies every structured bordism problem. Milnor 1962, pp. 16–23 places these characteristic-number tests in the broader cobordism theory.

Nor is every familiar topological number a bordism invariant. The sphere $S^2$ bounds $D^3$ but has Euler characteristic $2$. The integer Euler characteristic therefore does not descend to oriented bordism in dimension $2$.

Bordism is not ordinary homology

There is a natural map

$$\Omega_n^{SO}(X) \longrightarrow H_n(X;\mathbb Z), \qquad [M,f] \longmapsto f_*[M].$$

An oriented bordism over $X$ gives a homology between the two pushed-forward fundamental cycles, so this map is well defined. The converse is false: homological equality need not produce a smooth structured filling. At $X=\mathrm{pt}$,

$$H_4(\mathrm{pt};\mathbb Z)=0,$$

while $\mathbb{CP}^2$ is nonzero in oriented bordism, as its signature shows. Ordinary homology records chains modulo chain boundaries; bordism requires those chains to be represented by smooth manifolds carrying all the declared extensions.

This also locates the difference from homotopy. A homotopy deforms one map through an admissible family on a fixed source. A bordism replaces the source manifold itself by the boundary of a manifold one dimension higher. Neither word should be used without naming the objects and admissible structures being compared.

Recognizing the input language of modern anomaly and invertible-phase classifications

Consider a $d$-dimensional Euclidean fermionic theory whose admissible backgrounds require a spin structure and a topological $U(1)$ bundle. The background data have the type

$$(M^d,s,f:M\to BU(1)).$$

The notation

$$\Omega_d^{\mathrm{Spin}}(BU(1))$$

then says exactly what is being quotiented: closed smooth $d$-manifolds with chosen spin structure and line bundle, modulo compact spin $(d+1)$-manifolds carrying an extending map to $BU(1)$. Before using such a symbol, identify five inputs:

1. the dimension;
2. the manifold category, smooth here;
3. the tangential or symmetry structure;
4. any target map, bundle, twist, or differential refinement;
5. whether the source uses tangent or stable-normal conventions.

The surface flux calculation above is the smallest demonstration of why this parsing matters. Forgetting the $BU(1)$ map would declare the underlying sphere null-bordant and erase its nonextendable flux. Forgetting the spin lift similarly identifies the bounding and nonbounding spin circles.

This recognition step is not a universal classification theorem. A bare bordism group does not by itself classify anomalies or invertible phases, and a formula such as $\operatorname{Hom}(\Omega_d^\theta,U(1))$ cannot be applied without additional hypotheses. Precise results may use a dimension shift, generalized-cohomology or dual spectra, differential data, locality, reflection positivity or unitarity, counterterms, and boundary conditions. Recent broad lecture notes such as Moore and Saxena 2025, TASI Lectures on Topological Field Theories and Differential Cohomology make those added layers visible; they are deliberately outside this primer.

Limits and common pitfalls

Forgetting compactness or the smooth category. The group above uses compact smooth bordisms. Noncompact ends, singularities, and topological or piecewise-linear manifolds define different problems.

Checking only the underlying boundary. If $M=\partial W$ after all structure is forgotten, $(M,\widetilde\tau,f)$ may still fail to bound because the lift, bundle, or map does not extend.

Treating existence as a choice. Vanishing $w_2$ says a spin lift exists after orientation is fixed. It does not select one of the possible lifts.

Dropping the incoming sign. With outward-normal-first orientation, the incoming component is $-M_0$. This sign is what makes gluing and the group inverse consistent.

Mixing stable tangent and normal structures. The normal bundle is the stable complement of the tangent bundle. Translate $B$ to $B^\perp$ before borrowing a normal-convention result.

Replacing characteristic numbers by classes. A characteristic number pairs a top-degree polynomial with a fundamental class. Equality of raw classes on different manifolds is not a meaningful comparison without a specified map.

Using one vanishing invariant as a converse. A characteristic number can obstruct a bordism. Its vanishing need not construct one or classify the structured bordism class.

Turning mathematical input into a physical conclusion. A bordism class is not automatically an observable, anomaly, phase, or cancellation condition. Those conclusions require the later field-theoretic framework.

Exercises

1. Check the cylinder sign

Give $[0,1]\times M$ the orientation $\mathrm dt\wedge o(TM)$. Derive the orientations on its two boundary components and explain how the result gives both reflexivity and the inverse in $\Omega_n^{SO}$.

Solution

At $t=1$, the outward normal is $+\partial_t$, so outward-normal-first induces $o(TM)$. At $t=0$, the outward normal is $-\partial_t$, so the boundary orientation must be $-o(TM)$. Therefore

$$\partial([0,1]\times M)=(-M)\sqcup M.$$

Viewed with the two ends incoming and outgoing, the cylinder is a bordism from $M$ to itself, proving reflexivity. Viewed as a filling of the disjoint union, it proves $[-M]+[M]=0$, so $[-M]$ is the inverse of $[M]$.

2. Compute the zero-dimensional groups

Why are $\Omega_0^O\cong\mathbb Z/2$ and $\Omega_0^{SO}\cong\mathbb Z$?

Solution

A compact $0$-manifold is a finite set of points. Every compact $1$-manifold is a disjoint union of circles and intervals, so its boundary contains an even number of points. In the unoriented theory, two points bound an interval and parity is the only invariant. Hence $\Omega_0^O\cong\mathbb Z/2$.

An oriented point has sign $+$ or $-$. The two ends of an oriented interval have opposite induced signs, so signed point count is unchanged by oriented bordism. Pairing opposite signs with intervals shows that this count is complete, giving $\Omega_0^{SO}\cong\mathbb Z$.

3. Translate a tangent obstruction to the normal bundle

From $TM\oplus\nu_M\cong\varepsilon^N$, derive $w_1(\nu_M)$ and $w_2(\nu_M)$. Why do tangent and normal spin obstructions agree when $M$ is oriented?

Solution

Write

$$(1+w_1(TM)+w_2(TM)+\cdots) (1+w_1(\nu)+w_2(\nu)+\cdots)=1.$$

The degree-one term gives $w_1(\nu)=w_1(TM)$. The degree-two term gives

$$w_2(\nu) =w_2(TM)+w_1(TM)w_1(\nu) =w_2(TM)+w_1(TM)^2.$$

If $M$ is oriented, $w_1(TM)=0$, so $w_2(\nu)=w_2(TM)$. The spin obstruction therefore agrees in the two conventions, even though the structured lifts still need to be translated.

4. Decode a QFT-facing bordism symbol

Interpret $\Omega_2^{\mathrm{Spin}}(BU(1))$. If $(S^2,L_n)$ and $(S^2,L_m)$ are bordant over $BU(1)$, what condition follows? What physical conclusion does that condition not establish?

Solution

A representative is a closed smooth spin surface with a chosen complex line bundle. A bordism is a compact smooth spin $3$-manifold with a line bundle restricting to the two boundary bundles. Because the extending first Chern class pairs trivially with the total boundary,

$$m-n=0.$$

Thus $m=n$ is necessary. In particular, $(S^2,L_n)$ with $n\ne0$ is not null-bordant over $BU(1)$. This does not by itself compute a partition function, classify an anomaly or phase, or prove anomaly cancellation; those claims require a specified field-theory framework and its hypotheses.

Synthesis and continuation

Bordism asks whether two closed manifolds become the structured boundary of one compact manifold a dimension higher. Cylinders, reversal, and collared gluing make this an equivalence relation, while disjoint union makes its classes an abelian group. A stable tangential structure is a chosen lift of the stable tangent classifier, and every background map or bundle named in the problem must extend across the bordism. Characteristic numbers can obstruct such an extension, but a few vanishing invariants do not construct or classify a bordism.

Continue to Bordism Categories and Symmetric Monoidal TQFTs for the categorical structure deliberately omitted here: objects and bordisms, disjoint-union monoidal structure, duals, symmetric monoidal functors, state spaces and maps, diffeomorphism invariance, and gluing.

References

• Arun Debray, Bordism and Invertible Field Theories — Open PDF, lecture notes, 2019, §§ 1–2. This gives a compact account of stable tangential structures, boundary signs, characteristic-number obstructions, and the $B^\perp$ normal convention.
• Daniel S. Freed, Bordism: Old and New — Open PDF, course notes, 2013, especially Lectures 1, 2, and 9. These support the collared definition, equivalence and group laws, low-dimensional examples, stable tangential lifts, induced boundary structures, and tangent-to-normal translation.
• Davide Gaiotto and Theo Johnson-Freyd, “Mock Modularity and a Secondary Elliptic Genus”, JHEP 08 (2023) 094, § 3.1. This source fixes the bounding/antiperiodic and nonbounding/periodic terminology for the two spin circles and gives an index test.
• John Milnor, “A Survey of Cobordism Theory”, L’Enseignement Mathématique 8 (1962), 16–23. This concise structural source develops unoriented, oriented, and structured bordism and characteristic-number tests for boundaries.
• Gregory W. Moore and Vivek Saxena, TASI Lectures on Topological Field Theories and Differential Cohomology, arXiv:2510.07408 (2025), with an appendix by Daniel S. Freed. This current review supplies context for the QFT handoff and for why bordism data alone are only one layer of a field-theory formulation.