2798b02 Δ 87c4af8 Δ 6fbd3ac Δ f6bf760 Δ 56bc728 Δ b1f9bc4 Δ b34550f Δ 763cb89 Δ 7b220a7 Δ 7cf6ddc Δ e70a468
editors: nobody
Let $W$ be a connected closed smooth manifold. A smooth Morse function $F$ on $W\times\mathbf R^k$ is admissible when some nondegenerate quadratic form $Q$ on $\mathbf R^k$, of either signature, satisfies $F(x,z)=Q(z)$ for all $|z|\ge\rho$, with $\rho$ independent of $x$; for $k=0$ the condition is empty. The stable Morse number $\SM(W)$ is the least number of critical points of an admissible Morse function, over all $k$ and $F$. Taking $k=0$ gives the ordinary Morse number $\Morse(W)$, so $\SM(W)\le\Morse(W)$.

For a finitely generated abelian group $A$ let $d(A)$ be its least number of generators. Set $r_i(W)=\rank H_i(W;\mathbf Z)$, $t_i(W)=d(\Tor H_i(W;\mathbf Z))$, $b(W)=\sum_i r_i(W)$, and
$$\lambda(W)=b(W)+2\sum_i t_i(W).$$

On the Hamiltonian side let $(M,\omega)$ be closed symplectic and $H$ smooth and one-periodic, with vector field $\omega(X_{H_t},\cdot)=dH_t$ and time-one map $\phi_H^1$. A fixed point $x$ is nondegenerate when $1$ is not an eigenvalue of $d\phi_H^1(x)$, and contractible when $t\mapsto\phi_H^t(x)$ is null-homotopic in $M$; whether this holds depends only on $x$. The set of contractible fixed points is $\Fix_0(\phi_H^1;H)$.

The main result is the following. For every integer $m\ge1$ there exist a simply connected closed Kaehler manifold $(M_m,\omega_m)$ of real dimension $22$ and a smooth one-periodic Hamiltonian $H_m$, all of whose fixed points are nondegenerate, such that
$$\SM(M_m)=80+1968m,\qquad \#\Fix(\phi_{H_m}^1)=\#\Fix_0(\phi_{H_m}^1;H_m)=80+1952m.$$
Hence, with $\delta=1/128$, for every $R>0$ some member of the family has $\SM(M_m)\ge R$ and $\#\Fix(\phi_{H_m}^1)\le(1-\delta)\,\SM(M_m)$.

The Arnold problem asks how few contractible fixed points a nondegenerate Hamiltonian flow can have relative to the topology of $M$. The torus case was settled by Conley and Zehnder, and the general homological method originates with Floer; the variants are catalogued by Arnold and by Golovko. Refinements using Morava K-theory and integer Floer theory are due to Abouzaid-Blumberg and to Bai-Xu.
./abouzaid-blumberg-2021-morava.pdf
./bai-xu-2022-arnold-integers.pdf
./bai-xu-2026-integral-fh.pdf

The stable Morse number appears as a proposed lower bound in the survey of Golovko. Damian showed that the ordinary and stable Morse numbers can differ. Positive results for generic Hamiltonians, under hypotheses that fail for the examples constructed, include the theorem of Dimitroglou Rizell and Golovko for symplectically aspherical manifolds and an independent proof of Poder Balkestahl through the simple homotopy type of the Hamiltonian Floer complex.
./drg-2017-aspherical.pdf
./poder-balkestahl-2021-simple.pdf

The first ingredient is a chain-level inequality. Let $C_*$ be a finite chain complex of finitely generated free abelian groups, with homology of rank $\rho_i$ and torsion subgroup generated by $\tau_i$ elements. Write $Z_i=\ker\partial_i$ and $B_i=\operatorname{im}\partial_{i+1}$. These are free, and the two defining short exact sequences give
$$\rank C_i=\rho_i+\rank B_i+\rank B_{i-1}.$$
Smith normal form applied to $B_i\subset Z_i$ exhibits $H_i(C_*)=Z_i/B_i$ as a free group plus one cyclic summand per nonunit invariant factor, so $\rank B_i\ge\tau_i$. Substitution yields $\rank C_i\ge\rho_i+\tau_i+\tau_{i-1}$ and, after summation, $\sum_i\rank C_i\ge\sum_i\rho_i+2\sum_i\tau_i$.

The stable Morse number obeys the same bound. Let $F$ be admissible on $W\times\mathbf R^k$. The aim is a finite free integral chain complex with one generator per critical point and homology $H_{j-d}(W;\mathbf Z)$. For $k=0$ the ordinary Morse complex already has this form. Assume $k>0$ and, after a linear change of the auxiliary coordinates, write the form at infinity as $Q(u,v)=-|u|^2+|v|^2$ with $(u,v)\in\mathbf R^d\times\mathbf R^{k-d}$. Choose $R_0$ with $F=Q$ for $|(u,v)|\ge R_0$ and all critical points inside $W\times B_{R_0}$, then constants $A,T$ with
$$A>\max_{W\times\overline B_{R_0}}\{|F|,|Q|\},\qquad T>R_0,\qquad T^2>A.$$
Put $K=W\times\overline B_T$ and $K_\pm=K\cap\{F\le\pm A\}$. Outside the core the two functions coincide; inside it the level $-A$ is empty and the level $A$ contains the whole core. Hence $K_\pm=K\cap\{Q\le\pm A\}$.

On the side boundary $W\times\partial B_T$ the function equals $Q$, whose restricted differential vanishes only where $u=0$ or $v=0$, at the values $\pm T^2$. Since $T^2>A$, there are no side-boundary critical points in $[-A,A]$. A gradient-like field for $F$ on that band may be chosen tangent to the side boundary: take such a field on the boundary where $dF\ne0$, extend to a collar, and combine with an interior field by a partition of unity, the inequality $dF(V)>0$ away from critical points surviving the combination. The Morse attachment theorem then applies on the compact manifold with corners and, relative to $K_-$, attaches one $j$-cell for each index-$j$ critical point. Separating coincident critical values by a small perturbation supported in the Morse charts of the critical points does not alter the endpoint pair, the critical points, or the indices. A finite free complex $C_*$ results, with
$$\sum_j\rank C_j=\#\Crit(F),\qquad H_j(C_*)\cong H_j(K_+,K_-;\mathbf Z).$$

The homotopy $(u,v)\mapsto(u,(1-s)v)$ decreases both $Q$ and the Euclidean norm, so by the endpoint identity it deformation retracts the pair onto $W\times(D_T^d,\{u\in D_T^d:|u|\ge\sqrt A\})$. The disk-annulus pair has relative integral homology $\mathbf Z$ in degree $d$ and zero elsewhere, with $d=0$ giving the pair $(\{0\},\varnothing)$; both signatures of $Q$ are covered. The relative Kunneth formula, the second factor being torsion-free, then gives $H_j(C_*)\cong H_{j-d}(W;\mathbf Z)$. The chain inequality forces $\#\Crit(F)\ge\lambda(W)$, and minimising over $F$ gives $\SM(W)\ge\lambda(W)$.

The reverse direction for the manifolds at hand uses Smale's handle theorem. Let $W$ be simply connected closed smooth of dimension $D\ge9$. Then $W$ carries an ordinary Morse function with exactly $r_i(W)+t_i(W)+t_{i-1}(W)$ critical points of index $i$ for every $i$. The existence is the torsion-inclusive form of Smale's Morse theorem; a self-contained handle-cancellation proof, including the delicate indices $2$ and $3$, is supplied as part of the proof of the equality theorem. Summing the index counts gives $\lambda(W)$ critical points, and since ordinary Morse functions are admissible,
$$\lambda(W)\le\SM(W)\le\Morse(W)\le\lambda(W).$$
./smale-1963-survey.pdf

Torsion counts combine distinct primes by a maximum rather than a sum. For distinct primes $p,q$ and nonnegative integers $a,b$,
$$d\bigl((\mathbf Z/p)^a\oplus(\mathbf Z/q)^b\bigr)=\max(a,b).$$
Projection to each primary summand gives the lower bound, and the Chinese remainder theorem pairs cyclic summands to give the upper bound. Placing the two prime-primary parts across degrees by a maximum, not a sum, is the source of the strict inequality in the construction.

Two bielliptic surfaces supply the torsion. Put $E=\mathbf C/(\mathbf Z+i\mathbf Z)$, $\zeta_p=\exp(2\pi i/p)$, $\Lambda_2=\mathbf Z+i\mathbf Z$, $\Lambda_3=\mathbf Z+\zeta_3\mathbf Z$, and $E_p=\mathbf C/\Lambda_p$. The map $\sigma_p(z,w)=(z+1/p,\zeta_p w)$ on $E\times E_p$ has order $p$ and is free, because each nonidentity power translates the first factor by a nonzero multiple of $1/p$. The quotient $S_p=(E\times E_p)/\langle\sigma_p\rangle$ is a smooth compact connected complex surface, bielliptic in the sense of Serrano.

For each $p$ the integral homology is
$$H_i(S_p;\mathbf Z)\cong\begin{cases}\mathbf Z,&i=0,4,\\ \mathbf Z^2\oplus T_p,&i=1,2,\\ \mathbf Z^2,&i=3,\\ 0,&\text{otherwise},\end{cases}$$
with $T_2=(\mathbf Z/2)^2$ and $T_3=\mathbf Z/3$. Each surface has total Betti number $8$.

Projectivity is seen as follows. A degree-three line bundle on an elliptic curve has three independent sections, and Riemann-Roch, applied after subtracting one or two points with multiplicity, shows the complete linear system is base-point free and separates points and tangent directions; the curve therefore embeds in $\mathbf{CP}^2$. Taking the two curve embeddings and the Segre product embeds $E\times E_p$ in $\mathbf{CP}^8$. To an orbit under $\sigma_p$ assign the product of the $p$ linear forms defining the orbit points; unique factorisation makes the descended map $S_p\to\mathbf P(\operatorname{Sym}^p\mathbf C^9)$ separate orbits. For immersion, choose local forms $L_0,\dots,L_{p-1}$ of an orbit. A tangent vector in the kernel of the derivative would satisfy $\sum_j\dot L_j\prod_{k\ne j}L_k=a\prod_k L_k$ for some scalar $a$. The factors are pairwise nonproportional because the action is free, so restricting to $L_j=0$ forces $\dot L_j$ to vanish there, hence $\dot L_j$ is a scalar multiple of $L_j$. The projective derivative of the curve embedding then vanishes, and the tangent vector is zero. The descended map is an injective holomorphic immersion, hence an embedding by compactness; projectivity follows.

For the homology, the universal cover $\mathbf C^2\to S_p$ has deck group generated by $a(z,w)=(z+1/p,\zeta_p w)$, $d(z,w)=(z+i,w)$, and $t_\lambda(z,w)=(z,w+\lambda)$ for $\lambda\in\Lambda_p$. The power $a^p$ translates the first factor by $1$, and $a,d$ commute and generate $\mathbf Z^2$; conjugation acts on $\Lambda_p$ by $\zeta_p$ and by the identity. The deck group is thus $\Lambda_p\rtimes\mathbf Z^2$, with abelianisation
$$H_1(S_p;\mathbf Z)\cong\mathbf Z^2\oplus\Lambda_p/(\zeta_p-1)\Lambda_p.$$
For $p=2$ the multiplication $\zeta_p-1$ is $-2I$, giving $T_2=(\mathbf Z/2)^2$. For $p=3$ the matrix in the basis $(1,\zeta_3)$ is $\begin{pmatrix}-1&-1\\1&-2\end{pmatrix}$, of determinant three and coprime entries, so its Smith normal form has diagonal $(1,3)$ and the finite summand is $\mathbf Z/3=T_3$. Complex orientation makes $S_p$ a closed oriented four-manifold; Poincare duality and the universal coefficient theorem give $\Tor H_2(S_p;\mathbf Z)\cong\Tor H_1(S_p;\mathbf Z)=T_p$ and $H_3(S_p;\mathbf Z)\cong\mathbf Z^2$. The degree-$p$ covering gives $p\chi(S_p)=\chi(E\times E_p)=0$, and the ranks already found in degrees $0,1,3,4$ are $1,2,2,1$, so the rank in degree two is two.

The blow-up centers are $Z_2=S_2$ of complex dimension $2$ and $Z_3=S_3\times\mathbf{CP}^1\times\mathbf{CP}^1$ of complex dimension $4$, with projectivity of $Z_3$ from the Segre embedding. The projective-line factors spread the $3$-primary torsion across shifts by $0,2,4$ with multiplicities $1,2,1$, and the different codimensions make the larger primary multiplicity depend on the degree.

Two tools install the centers. First a simultaneous embedding, in the spirit of the secant-and-tangent argument of Hartshorne. Let $d\ge0$ and let $V_1,\dots,V_s$ be smooth complex projective varieties of complex dimension at most $d$, split into connected components. The disjoint union embeds holomorphically in $\mathbf{CP}^{2d+1}$. To see this, put the components in pairwise disjoint linear subspaces of some $\mathbf{CP}^D$ with $D>2d+1$. For each ordered pair the secant locus is swept by the projectivisation of a rank-two bundle over the pair space, of complex dimension $\dim V_\alpha+\dim V_\beta+1\le2d+1<D$, hence of measure zero by Sard's theorem; each embedded projective tangent space is swept in complex dimension $2\dim V\le2d<D$. Finitely many such sets leave a point $q$, and projection from $q$ is injective off the secants and has injective derivative off the tangents, so restricts to an embedding. Iterating the projection down to $2d+1$ completes the argument.

Second, the integral blow-up formula of Duan-Li. Let $A$ be compact connected Kaehler and $Z\subset A$ a nonempty closed connected complex submanifold of complex codimension $c\ge2$. The blow-up $\pi:\widehat A\to A$ is compact connected Kaehler, and simply connected when $A$ is, and for every $i$
$$H_i(\widehat A;\mathbf Z)\cong H_i(A;\mathbf Z)\oplus\bigoplus_{j=1}^{c-1}H_{i-2j}(Z;\mathbf Z).$$
The exceptional divisor is $E_Z=\mathbf P(\nu)$ for the complex normal bundle $\nu$, with tautological bundle $\mathcal O(-1)$. For the Kaehler form take $\pi^*\omega_A+\varepsilon\eta$ with $\varepsilon$ small, where $\eta$ is the Chern curvature of a positively curved metric on the fibrewise $\mathcal O(1)$, positive on the kernel directions of $\pi^*\omega_A$ and bounded on the compact remainder. For simple connectivity, $A\setminus Z$ is simply connected by general position in real codimension $2c\ge4$, and a loop in $\widehat A$ can be pushed off $E_Z$ (real codimension two) and contracted in $A\setminus Z$. For the splitting, the Leray-Hirsch theorem gives $H^k(E_Z;\mathbf Z)=\bigoplus_{j=0}^{c-1}h^j\rho^*H^{k-2j}(Z;\mathbf Z)$ with $h=c_1(\mathcal O(1))$ restricting to an integral basis on each fibre; the Gysin pushforward satisfies $\iota^*\iota_*(v)=-hv$, so $s_j(u)=-\iota_*(h^{j-1}\rho^*u)$ lifts the positive powers $h^j$. Classes restricting to an $h^0$ term are recognised through the isomorphism $(\widehat A,E_Z)\to(A,Z)$ of relative cohomology, coming from the homeomorphism $\widehat A/E_Z\cong A/Z$, which shows each such class lies in $\pi^*H^k(A;\mathbf Z)$. The map $\pi^*+\sum_j s_j$ is a bijection over $\mathbf Z$: restriction to $E_Z$ kills the positive powers, and $\pi^*$ is injective because $\pi$ has degree one. Poincare duality converts $H^{k-2j}(Z;\mathbf Z)$ to $H_{i-2c+2j}(Z;\mathbf Z)$, and replacing $j$ by $c-j$ gives the homology splitting, torsion included.
./duan-li-2009-blowups.pdf

The family of ninefolds follows. Let $m\ge1$. The projection with $d=4$ places $m$ copies of $Z_2$ and $m$ copies of $Z_3$ disjointly in $\mathbf{CP}^9$. Blowing up successively, the codimensions are always $7$ and $5$, and each step preserves compactness, connectedness, a Kaehler form, and simple connectivity, while the complex dimension stays $9$. The blow-up formula gives, for the resulting ninefold $X_m$,
$$H_i(X_m;\mathbf Z)\cong H_i(\mathbf{CP}^9;\mathbf Z)\oplus\left(\bigoplus_{r=1}^{6}H_{i-2r}(Z_2;\mathbf Z)\right)^{\oplus m}\oplus\left(\bigoplus_{r=1}^{4}H_{i-2r}(Z_3;\mathbf Z)\right)^{\oplus m}.$$

A short-period estimate excludes spurious fixed points. Let $V$ be a smooth vector field on a closed manifold $P$ with flow $\psi^s$. There is $\tau>0$ such that $0<T<\tau$ and $\psi^T(x)=x$ force $V(x)=0$; this is the elementary lower bound on nonconstant periods of Yorke. Cover $P$ by finitely many coordinate balls, each closing inside a larger ball of the same chart; bounds on the coordinate speed and the distances from the smaller balls to the boundaries of the larger ones give $\eta>0$ so that a trajectory leaving a smaller ball stays in its larger ball for time $\eta$, while a common Lipschitz constant $L$ bounds the coordinate field on the larger balls. Suppose $\psi^T(x)=x$ with $0<T<\eta$; write the orbit as $y:[0,T]\to\mathbf R^d$ and the coordinate field as $v$, and let $m=|v(y(t_0))|$ be the maximum. Since $y(T)=y(0)$, $\int_0^T v(y(t))\,dt=0$ and $|y(t)-y(t_0)|\le Tm$, whence
$$m=\Big|\frac1T\int_0^T\big(v(y(t_0))-v(y(t))\big)\,dt\Big|\le\frac LT\int_0^T|y(t_0)-y(t)|\,dt\le LTm.$$
For $T<\tau:=\min\{\eta,1/L\}$ this forces $m=0$.
./yorke-1969-periods.pdf

The estimate controls a perturbed Hamiltonian involution. Let $(P,\omega)$ be closed symplectic and $g$ a Hamiltonian diffeomorphism with $g^2=\mathrm{id}$. Write $\Fix(g)=F_1\sqcup\dots\sqcup F_r$ for the connected fixed components, and let $h_j$ be a smooth Morse function on $F_j$. Then there is a smooth $g$-invariant $f:P\to\mathbf R$ restricting to $h_j$ on $F_j$ such that, for all small enough $\epsilon>0$, the flow $\psi^s$ of $f$ satisfies
$$\Fix(\psi^\epsilon\circ g)=\bigsqcup_{j=1}^r\Crit(h_j),$$
every point of the set nondegenerate; moreover $\psi^\epsilon\circ g$ is the time-one map of a smooth one-periodic Hamiltonian, and when $P$ is simply connected the fixed-point loops are contractible.

Average a Riemannian metric over $\{\mathrm{id},g\}$ so that the exponential chart at a fixed point intertwines $g$ with $d_xg$; the fixed locus is then locally a submanifold with tangent $\ker(d_xg-\mathrm{id})$, and compactness gives finitely many closed components. The eigenvalues of $d_xg$ are $\pm1$, so $T_xP=E_x^+\oplus E_x^-$ with $E_x^\pm=\ker(d_xg\mp \mathrm{id})$. For $u\in E_x^+$ and $v\in E_x^-$ symplecticity gives $\omega_x(u,v)=-\omega_x(u,v)=0$, so the summands are symplectically orthogonal and each is nondegenerate; thus each $F_j$ is symplectic with normal space $E_x^-$ at $x$. Extend the $h_j$ over disjoint tubular neighbourhoods by cut-offs and set $f=\tfrac12(\tilde f+\tilde f\circ g)$ for the extension $\tilde f$; then $f$ is $g$-invariant, restricts to $h_j$, and $g$ commutes with the flow $\psi^s$. At $x\in F_j$ invariance gives $d_xf|_{E_x^-}=0$, and orthogonality makes $X_f$ tangent to $F_j$, where it is the field of $h_j$; hence $X_f(x)=0$ if and only if $d_xh_j=0$. With $\Phi_\epsilon=\psi^\epsilon\circ g$ there holds $\Phi_\epsilon^2=\psi^{2\epsilon}$, so for $2\epsilon<\tau$ any fixed point of $\Phi_\epsilon$ is an equilibrium of $X_f$ fixed by $g$, that is, a prescribed critical point; conversely each prescribed critical point is fixed. For nondegeneracy let $A_x=d_xX_f$, which commutes with $d_xg$ and preserves $E_x^\pm$. On $E_x^+=T_xF_j$ the identity $\omega_x(A_x^+u,v)=\operatorname{Hess}_xh_j(u,v)$ makes $A_x^+$ invertible by the Morse condition, and the tangent block $(\exp(\epsilon A_x^+)-\mathrm{id})/\epsilon\to A_x^+$ is invertible for small $\epsilon$; the normal block $-\exp(\epsilon A_x^-)-\mathrm{id}\to-2\,\mathrm{id}$ is likewise invertible. Finitely many critical points admit one common bound on $\epsilon$. Finally, write $g$ as the time-one map of a smooth path $K_s$, choose nonnegative $a,b$ integrating to one with supports in two successive disjoint subintervals of $(0,1)$, and set $A(t)=\int_0^t a$ and $H_\epsilon(t,x)=a(t)K_{A(t)}(x)+\epsilon b(t)f(x)$. The time-one map is $\psi^\epsilon\circ g$, the periodic extension is smooth because $H_\epsilon$ vanishes near the endpoints, and simple connectivity of $P$ makes every fixed-point loop of the path contractible.

A specific surface carries four fixed curves. Let $Y$ be the blow-up of $\mathbf{CP}^1\times\mathbf{CP}^1$ at the four points $\{0,\infty\}\times\{0,\infty\}$. The lifted diagonal circle action preserves a Kaehler form $\omega_Y$ and is Hamiltonian, and its half-turn $g_Y$ has fixed locus exactly four exceptional curves $E_1,\dots,E_4$, each a $\mathbf{CP}^1$; along each, $dg_Y$ is $+\mathrm{id}$ on the tangent bundle and $-\mathrm{id}$ on the symplectic normal bundle. The surface $Y$ is simply connected, with
$$H_i(Y;\mathbf Z)\cong\begin{cases}\mathbf Z,&i=0,4,\\ \mathbf Z^6,&i=2,\\ 0,&\text{otherwise}.\end{cases}$$

In affine coordinates the diagonal action is $t\cdot(z,w)=(e^{2\pi it}z,e^{2\pi it}w)$, which fixes the four centres and lifts holomorphically. Averaging a Kaehler form over the lifted action makes it invariant; Cartan's formula $d(\omega_Y(V,\cdot))=\mathcal L_V\omega_Y=0$ together with $H^1(Y;\mathbf R)=0$ gives a function $k$ with $dk=\omega_Y(V,\cdot)$, so $V=X_k$ and the autonomous Hamiltonian $k/2$ has time-one map $g_Y$. Near a centre write $z$ near $0$ or $1/z$ near $\infty$ in each factor; the four circle weights are $(1,1),(1,-1),(-1,1),(-1,-1)$, and the induced action on the exceptional line sends $[z_1:z_2]\mapsto[e^{2\pi i\alpha t}z_1:e^{2\pi i\beta t}z_2]$, whose value at $t=1/2$ is the identity in all four cases. In the local model $\{(\ell,v):\ell\in\mathbf{CP}^1,\ v\in\ell\subset\mathbf C^2\}$ the downstairs half-turn is $v\mapsto-v$, so the lift is $(\ell,v)\mapsto(\ell,-v)$, fixing exactly the zero section, with derivative $+\mathrm{id}$ on the tangent space and $-\mathrm{id}$ on the normal line; that normal space is the negative eigenspace, by orthogonality of the two eigenspaces. The blow-down is an equivariant isomorphism away from the exceptional curves, so no further fixed points occur. Since each blown-up point has complex codimension two, four blow-ups add one free generator in degree two to the ranks $(1,2,1)$ of $\mathbf{CP}^1\times\mathbf{CP}^1$, giving the stated homology.

These pieces realise four copies of a Morse problem on $C=X\times\mathbf{CP}^1$. Let $X$ be connected closed simply connected Kaehler, let $\omega_X$ be any Kaehler form, and let $(Y,\omega_Y)$ be the surface just described; give $M=X\times Y$ the product form and put $C=X\times\mathbf{CP}^1$. For every smooth Morse function $h:C\to\mathbf R$ there is a smooth one-periodic Hamiltonian $H$ on $M$ whose time-one map has only nondegenerate fixed points and
$$\#\Fix(\phi_H^1)=\#\Fix_0(\phi_H^1;H)=4\,\#\Crit(h).$$
The pullback of $k/2$ generates the involution $g=\mathrm{id}_X\times g_Y$ for the product form, with fixed components $X\times E_1,\dots,X\times E_4$. Pulling $h$ back along isomorphisms $E_j\cong\mathbf{CP}^1$ gives Morse functions $h_j$, each with $\#\Crit(h)$ critical points; the involution perturbation yields exactly the fixed points counted, all nondegenerate, and simple connectivity of $M$ makes the fixed-point loops contractible.

The integral counts now follow from the homology. Since $b(S_2)=b(S_3)=8$ and $\mathbf{CP}^1\times\mathbf{CP}^1$ has ranks $(1,2,1)$ in degrees $(0,2,4)$, the center $Z_3$ has $b(Z_3)=32$, and the blow-up formula gives
$$b(X_m)=10+m(6\cdot8+4\cdot32)=10+176m.$$
All homology shifts are even, so torsion lies in two parallel lists, in degrees $1+2j$ and $2+2j$. In either list
$$\Tor H_{1+2j}(X_m;\mathbf Z)\cong\Tor H_{2+2j}(X_m;\mathbf Z)\cong(\mathbf Z/2)^{ma_j}\oplus(\mathbf Z/3)^{mq_j}$$
with $a=(2,2,2,2,2,2)$ and $q=(1,3,4,4,3,1)$, indexed from $j=1$, all other entries zero. The first sequence records the six blow-up shifts of the two copies of $\mathbf Z/2$; the second is the coefficientwise sequence of
$$(u+u^2+u^3+u^4)(1+2u+u^2)=u+3u^2+4u^3+4u^4+3u^5+u^6,$$
where $u$ records a shift by two in homological degree, the four blow-up shifts give $u+\dots+u^4$, and the product with $\mathbf{CP}^1\times\mathbf{CP}^1$ multiplies by $1+2u+u^2$.

Since $d((\mathbf Z/2)^a\oplus(\mathbf Z/3)^q)=\max(a,q)$, generator counts are taken in each degree before summation. If $B$ has torsion-free homology in even degrees with $e_k=\rank H_{2k}(B;\mathbf Z)$, the integral Kunneth theorem replaces $a,q$ by the convolutions $(e*a)_j=\sum_k e_k a_{j-k}$ and $(e*q)_j=\sum_k e_k q_{j-k}$, with no additional torsion. For $C_m$ take $e=(1,1)$; for $M_m$ take $e=(1,6,1)$ from the homology of $Y$. The resulting per-list profiles, every entry multiplied by $m$, are:
C_m 2-primary $(2,4,4,4,4,4,2)$, sum $24$; C_m 3-primary $(1,4,7,8,7,4,1)$, sum $32$; C_m generators $(2,4,7,8,7,4,2)$, sum $34$; M_m 2-primary $(2,14,16,16,16,16,14,2)$, sum $96$; M_m 3-primary $(1,9,23,31,31,23,9,1)$, sum $128$; M_m generators $(2,14,23,31,31,23,14,2)$, sum $140$.

Each list occurs once in odd and once in even degrees, so
$$\sum_i t_i(C_m)=2\cdot34m=68m,\qquad \sum_i t_i(M_m)=2\cdot140m=280m.$$
The free ranks are $b(C_m)=2b(X_m)=20+352m$ and $b(M_m)=8b(X_m)=80+1408m$, whence
$$\lambda(C_m)=20+488m,\qquad \lambda(M_m)=80+1968m.$$
The total multiplicity at each prime is four times larger for $M_m$, but the degreewise maximum adds $2m$ generators per parity for $C_m$ and $12m$ for $M_m$, a surplus of $(12-4\cdot2)m=4m$ per parity. With the two parities and the factor two in $\lambda$,
$$\lambda(M_m)-4\lambda(C_m)=2(280-4\cdot68)m=16m.$$

To finish, fix $m\ge1$, choose any Kaehler form on $X_m$, give $M_m=X_m\times Y$ the product with the invariant form on $Y$, and put $C_m=X_m\times\mathbf{CP}^1$. Both $C_m$ and $M_m$ are closed, simply connected, of real dimension $20$ and $22$, so Smale's theorem gives $\SM(M_m)=\lambda(M_m)=80+1968m$ and supplies an ordinary Morse function $h_m$ on $C_m$ with $\lambda(C_m)=20+488m$ critical points. The product realisation applied to $X_m$ and $h_m$ yields a smooth one-periodic $H_m$ on $M_m$ with exactly $4\#\Crit(h_m)=80+1952m$ fixed points, all nondegenerate, and contractible fixed-point loops, so that $\#\Fix(\phi_{H_m}^1)=\#\Fix_0(\phi_{H_m}^1;H_m)=80+1952m$. Then
$$\frac{\SM(M_m)-\#\Fix(\phi_{H_m}^1)}{\SM(M_m)}=\frac{16m}{80+1968m}=\frac{m}{5+123m}\ge\frac1{128},$$
the last inequality being equivalent to $m\ge1$. For a prescribed $R>0$ take $m=\max\{1,\lceil(R-80)/1968\rceil\}$; then $\SM(M_m)\ge R$ while the dimension $22$ and $\delta=1/128$ stay fixed. This proves the main assertion.

Two remarks locate the example among existing bounds. Over a field of characteristic $2$ the total Betti number of $M_m$ is $b(M_m)$ plus twice the $2$-primary multiplicity, namely $80+1408m+384m=80+1792m$; over characteristic $3$ it is $80+1408m+512m=80+1920m$; over any other characteristic it is the free rank $80+1408m$. All three are below the fixed-point count $80+1952m$, so no single field recovers the count, because the prime giving more generators changes with degree.

The integral Floer bound of Bai-Xu, using homology graded modulo twice the minimal Chern number, gives a different number. The minimal Chern number of $M_m$ is $1$: an exceptional line $E\subset Y$ has $c_1(TY)|_E=2+(-1)=1$, so $c_1(TM_m)$ evaluates to $1$ on $\{x\}\times E$. With grading modulo $2$ each parity combines the odd or the even degrees, and per parity the primary torsion multiplicities are $96m$ and $128m$, whose maximum $128m$ is the least number of torsion generators. The cyclically graded integral bound is then $b(M_m)+2(128m+128m)=80+1920m$, smaller than the fixed-point count by $32m$. The strict gap relies on keeping the integer homological degrees separate in the stable Morse number.
./bai-xu-2022-arnold-integers.pdf
./bai-xu-2026-integral-fh.pdf

It remains to supply the handle-cancellation proof behind Smale's theorem. Throughout, $W$ is a simply connected closed smooth $D$-manifold with $D\ge9$, given with an ordered handle decomposition. A handle of index $i$ is a copy of $D^i\times D^{D-i}$ attached along $S^{i-1}\times D^{D-i}$, with belt sphere $\{0\}\times S^{D-i-1}$ in the new boundary; a slide changes the integral basis of the handles of one index by an elementary addition; consecutive handles of indices $i$ and $i+1$ cancel when the attaching sphere of the latter meets the belt sphere of the former transversely in one point. Attaching maps of later handles are carried along by these operations, so the represented manifold is unchanged. These are the standard facts of Morse theory and of the $h$-cobordism theorem (Milnor, Morse Theory, 1963; Milnor, Lectures on the $h$-Cobordism Theorem, 1965).

First remove handles of index one and coindex one. Connectedness cancels all but one $0$-handle along a tree of $1$-handles, and the dual operation leaves one $D$-handle. To remove the remaining $1$-handles, insert cancelling $2/3$ pairs as follows. Let $P$ be the union of the $0$- and $1$-handles; it retracts onto a graph whose free generators correspond to the $1$-handles, and $\pi_1(\partial P)\to\pi_1(P)$ is an isomorphism because the handles of $P$ relative to its boundary have indices $D-1$ and $D$. Since $\pi_1(W)=0$ and handles of index at least three do not change the fundamental group, the attaching words of the $2$-handles normally generate this free group. Fix a generator and insert a cancelling $2/3$ pair whose $2$-handle attaches along a trivial circle in a small ball of $\partial P$. Express the chosen generator as a product of conjugates of the old attaching words and the inverse words; sliding the new $2$-handle over the old ones realises this product, a slide being a band sum whose band path gives the conjugation and whose orientation gives the choice of word or inverse. General position lets the bands avoid the other attaching circles. The new attaching circle is now homotopic to a standard circle running once over the chosen $1$-handle and avoiding the others, and a generic homotopy of circles in the $(D-1)$-manifold is an isotopy avoiding the other attaching circles, since the expected dimensions of self-intersections, or of intersections with another attaching circle in a one-parameter family, are at most $3-(D-1)<0$. Transporting the framing along this isotopy, the new $2$-handle cancels the chosen $1$-handle. Repetition removes all $1$-handles; the reversed decomposition removes the $(D-1)$-handles, its inserted pairs having original indices $D-3,D-2$ that introduce no $1$-handles. The remainder has handles only in indices $0$, $D$, and $2,\dots,D-2$.

Suppose the decomposition carries more than $\lambda(W)$ handles. Orient $W$ and its levels, and let $C_*$ be the integral handle complex, whose boundary coefficients are algebraic intersection numbers of attaching spheres with belt spheres between consecutive indices. With $B_i=\operatorname{im}\partial_{i+1}$, the chain-rank identity gives
$$\sum_i\rank C_i-\lambda(W)=2\sum_i\big(\rank B_i-t_i(W)\big).$$
Each summand is nonnegative, so some $\rank B_i>t_i(W)$ with $2\le i\le D-3$; the vanishing of the groups in degrees $1$ and $D-1$ excludes the remaining indices. Because $C_i/Z_i\cong B_{i-1}$ is free, $Z_i=\ker\partial_i$ is a direct summand of $C_i$, so the nonzero Smith invariant factors of $\partial_{i+1}$ are those of the inclusion $B_i\subset Z_i$. Exactly $t_i(W)$ of these are nonunits, and the strict inequality yields a unit invariant factor. Row and column operations, realised by handle slides, relabelling, and orientation changes, give a boundary coefficient equal to $1$. The slides occur in connected levels: if $P_i$ is the union of the handles of index at most $i$ and $L=\partial P_i$, then $P_i$ is simply connected (it has no $1$-handles), the reversed handles relative to $L$ have indices at least $D-i\ge3$, and $L\to P_i$ is an isomorphism on components and fundamental groups; the same holds before the $i$-handles are attached, and slide paths avoid the other attaching spheres because those have codimension at least two. One thus obtains an attaching sphere $A\cong S^i$ and a belt sphere $B\cong S^{D-1-i}$ in a simply connected level, with algebraic intersection number $1$.

Arrange that the pair meets in exactly one geometric point. Suppose first that $i\ge3$ and $D-1-i\ge3$. Join two opposite-sign intersections by arcs in $A$ and in $B$ avoiding the others; the resulting Whitney circle bounds a disk because $L$ is simply connected, and the high-dimensional Whitney trick of the $h$-cobordism theorem removes the pair (Milnor, Lectures on the $h$-Cobordism Theorem, 1965, Theorem 6.6). Both sphere dimensions are at least three, so a general-position choice keeps the interior of the Whitney disk off all attaching spheres of $(i+1)$-handles and belt spheres of $i$-handles, and the move preserves the other intersections at this level. Repetition leaves a single intersection of $A$ and $B$.

When one sphere has dimension two, reverse the decomposition if necessary so that $i=2$; the level $L$ is then the boundary after attaching the $2$-handles to the single $0$-handle. Write the belt spheres as $B_1,\dots,B_h$ and put $V=L\setminus\bigcup_j B_j$. A pair of opposite-sign intersections is removed by moving a small disk of $A$ into $V$. The boundary $L$ comes from $S^{D-1}$ by surgery on the attaching circles: each surgery replaces $S^1\times D^{D-2}$ by $D^2\times S^{D-3}$, and removing its belt sphere leaves $(D^2\setminus\{0\})\times S^{D-3}$, which retracts onto $S^1\times S^{D-3}$. Thus $V$ has the homotopy type of the exterior of the original circles in $S^{D-1}$, and that exterior is simply connected because loops and disks can be pushed off the codimension-two circles; hence $\pi_1(V)=\pi_1(L)=0$.

Choose two opposite-sign intersections of the attaching sphere $A\cong S^2$ with one belt sphere. A thin disk $\Delta\subset A$ around an arc between the two points contains just these two intersections and misses the other belt spheres; its boundary and an exterior collar in $A$ lie in $V$. The oriented normal bundle of each belt sphere has rank two, so excision and the Thom isomorphism give
$$H_2(L,V;\mathbf Z)\cong\bigoplus_{j=1}^h\mathbf Z,$$
with coordinates the intersection numbers with the $B_j$. All coordinates of $[\Delta]$ vanish, so $[\Delta]=0$. This gives a relative homotopy: the exact sequences of the pair, together with $\pi_1(V)=\pi_1(L)=0$ and the degree-two Hurewicz isomorphisms for $V$ and $L$, identify
$$\pi_2(L,V)\cong\operatorname{coker}(\pi_2(V)\to\pi_2(L)),\qquad H_2(L,V;\mathbf Z)\cong\operatorname{coker}(H_2(V;\mathbf Z)\to H_2(L;\mathbf Z)).$$
Hence $\Delta$ can be homotoped into $V$ through maps whose boundary stays in $V$. Extending the boundary motion across the exterior collar and fixing its outer edge produces a homotopy of $A$, supported in the disk and collar, that removes the chosen pair and preserves the other intersections with the belt spheres.

That homotopy may be taken to be an isotopy of the attaching sphere. Smooth it relative to the initial sphere and its fixed part, and apply parametric general position: for a one-parameter family of maps of a $2$-sphere into the $(D-1)$-manifold $L$, the expected dimension of self-intersections, or of intersections with another attaching $2$-sphere, is at most $2+2+1-(D-1)<0$. Since $D-1\ge8$, the relative version leaves room to spare and each map in the family can be made an immersion. At the endpoint the altered disk and collar form a compact subset of the open set $V$, so a small approximation keeps the altered disk and collar off the belt spheres, while keeping small neighbourhoods of the other transverse intersections fixed preserves those intersections. Isotopy extension transports the framing. Repetition leaves exactly one intersection with the chosen belt sphere.

Finally the chosen $i$-handle is ordered last among the $i$-handles and the chosen $(i+1)$-handle first among the $(i+1)$-handles; the two are consecutive and cancel, and intersections with other handles do not obstruct the operation because the attaching maps are carried along by the cancellation. No handles of index $1$ or $D-1$ are introduced. Each cancellation lowers the total count by two, and while the count exceeds $\lambda(W)$ the excess identity supplies another cancellable pair. The process ends with exactly $\lambda(W)$ handles; equality in the excess identity forces $\rank B_i=t_i(W)$ for every $i$, and the chain-rank identity then gives $\rank C_i=r_i(W)+t_i(W)+t_{i-1}(W)$. The Morse function of this decomposition has exactly these index counts, which is the existence statement of Smale's theorem and completes the proof of $\SM(W)=\lambda(W)$.