Let $X_1$ and $X_2$ be two closed spin$^c$ manifolds that are bordant via a spin$^c$ manifold-with-boundary $W$.

Let $Z$ be a closed spin$^c$ manifold with $\dim Z=\dim X_1$ mod $2$. Let $$f_1:X_1\to Z,\qquad f_2:X_2\to Z,\qquad F:W\to Z$$ be smooth maps such that $F|_{X_1}=f_1$ and $F|_{X_2}=f_2$. We can associate to $f_1$ and $f_2$ two wrong-way (or Gysin) maps in $K$-theory:

$$f_{1!}:K^0(X_1)\to K^0(Z),$$ $$f_{2!}:K^0(X_2)\to K^0(Z).$$

Let $E_1\to X_1$ and $E_2\to X_2$ be two $\mathbb{C}$-vector bundles such that there exists a vector bundle $\Omega\to W$ satisfying $\Omega|_{X_1}\cong E_1$ and $\Omega|_{X_2}\cong E_2$. Let $[E_i]\in K^0(X_i)$ denote the $K$-theory classes defined by $E_i$.

Question: Is it true that $f_{1!}[E_1]=f_{2!}[E_2]\in K^0(Z)$?

Added after: I would be most interested in an approach not directly using Poincare duality for K-theory/K-homology.

  • $\begingroup$ I know how to do this in K-homology, which gives me confidence that it also works for K-theory, but I haven't worked it out. In K-homology you can attach cylindrical ends $X_1 \times [0, \infty)$ and $X_2 \times [0, \infty)$ to $W$ and argue that the Gysin maps factor through the Mayer-Vietoris boundary maps on the new object $W'$. You then lift the K-homology class on W to a class in a certain K-homology-like group for W'; this requires care since W' is non-compact. (One way to do it involves C*-algebras and coarse geometry.) Functoriality of Mayer-Vietoris completes the proof. $\endgroup$ Commented Aug 16, 2020 at 17:40
  • $\begingroup$ @PaulSiegel Could you elaborate how the Gysin maps factor through the Mayer-Vietoris boundary maps on $W'$ and why this implies the result in $K$-homology? I would like to adapt your argument to $K$-theory. $\endgroup$
    – geometricK
    Commented Aug 18, 2020 at 2:11

2 Answers 2


Let $N^n=\partial M^{n+1}$, $E\in K^\bullet(M)$ and $f:M\to X$

Choose a smooth embedding $i:X\to \mathbb{R}^N,N>>1$, denote by $\chi$ the normal bundle of $X$ and by $\mu$ the normal bundle of $M$ after suitable small deformation of $i\circ f$.

Let $\nu=\mu|_N$ and $\eta$ be the normal bundle of $N\subset M$ (which is trivial and one-dimensional)

By considering tubular neighborhoods we get the natural map:

$t:Th_\chi X\to Th_{\nu+\eta}N$, where $Th$ denotes a Thom space.

After applying the Thom isomorphism $th$ on $K^\bullet$ we obtain the definition of a Gysin map (going in "right-way" on a $Th$'s). So for $f_!(E|_N)=0$ it's sufficient to prove that $t^* th_{\nu+\eta}(E|_N)=0$

Actually $t^*$ is passing through a connecting homomorphism. Namely, there is a commutative diagram:

$\begin{matrix} Th_{\chi}X&\to& Th_{\mu}M/Th_\nu N&\\ \downarrow{t}&\swarrow{\sigma}&\downarrow{\Sigma}&\\ Th_{\nu+\eta}N&\xrightarrow{\sim}& \Sigma Th_{\nu}N&\\ \end{matrix}$

The top arrow comes from the tubular neighborhoods.

The horizontal isomorphism comes from triviality of $\eta$, while suspension $\Sigma$ from Puppe cofiber sequence:

$Th_\nu N\to Th_\mu M\to Th_\mu M/Th_\nu N\xrightarrow{\Sigma} \Sigma Th_{\nu}N$

The map $\sigma$ explains commutativity and is coming from:

$Th_\mu M/Th_\nu N\sim Th_\mu M/Th_\mu (N\times [0,\varepsilon))\to$ $Th_\mu (N\times(-\varepsilon,\varepsilon))/Th_\mu (N\times [0,\varepsilon))\to Th_{\nu+\eta}N$ where $N\times [0,\varepsilon)\subset M$ is a collar of $N$.

Finally, $\Sigma^*$ is the connecting homorphism and it follows that $\Sigma^* th_{\nu}(F)=0$ for all $F\in Im( K^\bullet(M)\to K^\bullet(N))$, so $t^* th_{\nu+\eta}(E|_N)=0$

  • $\begingroup$ I have a couple of questions. How is the map $t$ defined? Also, it seems to me that $\nu\oplus\eta$ is just the normal bundle of $N$ in $\mathbb{R}^N$ - is this correct? $\endgroup$
    – geometricK
    Commented Aug 21, 2020 at 8:30
  • $\begingroup$ @geometricK, correct. Let $D$ and $S$ denote a small disk and sphere bundles respectively. I assume that $D_{\nu+\eta}N\subset D_\chi X$. This inclusion automatically gives a "wrong-way" map $t:D_\chi/S_\chi\to D_{\nu+\eta}/S_{\nu+\eta}$ The horizontal map is obtained in the same way $\endgroup$ Commented Aug 21, 2020 at 20:16
  • $\begingroup$ Am I correct in understanding that your argument in fact shows a stronger result, namely that the Gysin map $i_!$ for the inclusion $i:\partial M\hookrightarrow M$ is always $0$? (I assume $i_!$ is the $K$-map induced by the composition $Th_{\mu}M\to Th_{\mu}M/Th_{\nu}N\xrightarrow{\Sigma}\Sigma Th_{\nu}N\sim Th_{\nu+\eta}N)$. Then the fact that $f_!$ factors through $i_!$ gives the result I was after originally. $\endgroup$
    – geometricK
    Commented Aug 22, 2020 at 2:12
  • $\begingroup$ @geometricK, of course we can forget about $X$ and prove that $i_!(E|_N)$ is zero. Always in a sense that we have to know that $\Sigma^*(E_N)=0$, which is coming from $K^\bullet(M/N\xrightarrow{\Sigma} \Sigma N\to \Sigma M)$, then the same holds after applying $th$-isomorphism, so $\Sigma^* th_\nu (E|_N)=0$. I don't know how to prove that the two long sequences naturally isomorphic at $K^\bullet(M/N)$ and $K^\bullet(Th_\mu(M)/Th_\mu(N))$, but for required vanishing it's not essential $\endgroup$ Commented Aug 22, 2020 at 8:47
  • $\begingroup$ Thanks. There should be a relative version of the Thom isomorphism, which I'm trying to locate and which should give the isomorphism $K^*(M/N)\to K^*(Th_{\mu}M/Th_{\nu}N)$. As you say this isn't strictly necessary but would still be nice to have. $\endgroup$
    – geometricK
    Commented Aug 25, 2020 at 6:29

The answer is yes, using general properties of orientations and fundamental classes.

Let $X_1$ and $X_2$ be $n$--dimensional. Then $f_{!i}$ is the composite $$K^0(X_i) \xrightarrow[\sim]{\cap [X_i]} K_n(X_i) \xrightarrow{f_{i*}} K_n(Z) \xleftarrow[\sim]{\cap [Z]} K^0(Z).$$

Meanwhile Poincare duality for $W$ has the form $K^0(W) \xrightarrow{\cap [W]} K_{n+1}(W, X_1 \coprod X_2)$, and $d([W]) = [X_1]-[X_2]$. Thus $ d(\Omega \cap [W]) = (E_1 \cap [X_1], -E_2 \cap [X_2])$, and so

$$ (f_{1*})(E_1 \cap [X_1]) - (f_{2*}(E_2 \cap [X_2]) = F_* i_* (d(\Omega \cap [W])) = 0,$$

since the composite

$$K_{n+1}(W,X_1\coprod X_2) \xrightarrow{d} K_n(X_1 \coprod X_2) \xrightarrow{i_*} K_n(W)$$

is zero.

  • $\begingroup$ Although I completely agree with your answer, I was secretly hoping for an approach without Poincare duality, as (although I didn't mention this) Poincare duality may not be available to me in the setting I'm actually working in. $\endgroup$
    – geometricK
    Commented Aug 16, 2020 at 20:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.