On approximation of maps into real algebraic homogeneous spaces

By Proposition 2.7, the variety Y is malleable. Hence the conclusion follows from Theorems 4.2 and 4.3.

Since each \mathcal{C}^{\infty} map \mathbb{S}^n \to \mathbb{S}^n is homotopic to a regular one (see [bib64, Theorem 1] for n odd, and [bib4, Corollary 4.2] for n arbitrary), the conclusion follows at once from Corollary 1.5.

By [bib5, Theorem 2.4], each regular map from X \times \mathbb{S}^{2n-k} into \mathbb{S}^{2n} is null homotopic, and hence the conclusion follows from Corollary 1.5.

By [bib4, Corollary 2.5], the set

is equal either to \mathbb{Z} or 2\mathbb{Z}, so the statement follows by combining Hopf’s theorem with Corollary 1.5.

By [bib4, Proposition 4.5], there exists an algebraic model X of M such that each \mathcal{C}^{\infty} map from X into \mathbb{S}^m is homotopic to a regular map. Hence the conclusion follows from Corollary 1.5.

The first assertion is proved in [bib41, Corollary 2.3], and the second follows from Corollary 1.5.

Suppose that (\tilde E, \tilde p, \tilde s) is a dominating spray for Y. Choose an algebraic vector bundle \tilde p' \colon \tilde E' \to Y such that the direct sum p \colon E \coloneqq \tilde E \oplus \tilde E' \to Y is an algebraically trivial vector bundle. Defining s \colon E \to Y by s(v,v') = \tilde s(v), we get a dominating spray (E,p,s) for Y.

Let \alpha \colon E \to TY be the morphism of algebraic vector bundles over Y defined by

Since \alpha is a surjective morphism, its kernel \operatorname{Ker} \alpha is an algebraic vector subbundle of E. It follows that E = \hat E \oplus \operatorname{Ker} \alpha is the direct sum for some algebraic vector subbundle \hat p \colon \hat E \to Y of p \colon E \to Y. This completes the proof.

With notation as in Lemma 2.3, the vector bundle \hat E is isomorphic to the tangent bundle TY, which completes the proof.

Let d = \dim G and let G^0 be the irreducible component of the variety G that contains the identity element 1 of G. By a theorem of Chevalley [bib24, Theorem 2] or [bib19, Corollary 7.12], G^0 is a unirational variety, and hence there exist a Zariski open subset U of \mathbb{R}^d and a regular map \psi \colon U \to G^0 such that the image \psi(U) is Zariski dense in G^0. Clearly, at some point a \in U the derivative d_a \psi \colon \mathbb{R}^d \to T_{\psi(a)} G^0 is a linear isomorphism. Using a translation in \mathbb{R}^d we may assume that a=0 \in U. The map \varphi \colon U \to G, v \mapsto \psi(v)\psi(0)^{-1} is regular, \varphi(0) = 1, and the derivative of \varphi at 0 is a linear isomorphism. As noted by János Kollár in no. A.2 of the Appendix, taking the map

where c is a small positive real number, we obtain a regular map g \coloneqq \varphi \circ h \colon \mathbb{R}^d \to G such that g(0)=1 and the derivative of g at 0 is a linear isomorphism. Now we obtain a dominating spray (E,p,s) for Y, where p \colon E \coloneqq Y \times \mathbb{R}^d \to Y is the product vector bundle and s \colon E \to Y is defined by s(y,v) = g(v) \cdot y for all (y,v) \in E.

The same as that of Proposition 2.7.

Suppose that (\tilde E, \tilde p, \tilde s) is a dominating spray for h \colon Z \to X. Choose an algebraic vector bundle \tilde p' \colon \tilde E' \to Z such that the direct sum p \colon E \coloneqq \tilde E \oplus \tilde E' \to Z is an algebraically trivial vector bundle. Defining s \colon E \to Z by s(v,v') = \tilde s(v), we get a dominating spray (E,p,s) for h \colon Z \to X.

Let \alpha \colon E \to V(h) be a morphism of algebraic vector bundles over Z defined by

Since \alpha is a surjective morphism, its kernel \operatorname{Ker} \alpha is an algebraic vector subbundle of E. It follows that E = \hat E \oplus \operatorname{Ker} \alpha is the direct sum for some algebraic vector subbundle \hat p \colon \hat E \to Z of p \colon E \to Z. This completes the proof.

Let x \in U. The zero vector in the fiber (E_f)_x is (x, 0_{f(x)}), where 0_{f(x)} is the zero vector in the fiber E_{f(x)}. Since s_f(x,0_{f(x)}) = s(0_{f(x)}) = f(x), it follows that s_f induces a diffeomorphism between Z(E_f) and f(U). Moreover, the derivative

is an isomorphism because

is an isomorphism for all z \in Z. Consequently, s_f is a local diffeomorphism at the point (x,0_{f(x)}). Now the lemma follows from [bib23, (12.7)].

By Lemma 3.4, we may assume without loss of generality that

is an isomorphism for all z \in Z. Defining f \colon U \to Z by f(x) = F(x, t_0), in view of Lemma 3.6, there exist an open neighborhood M \subseteq E_f of the zero section Z(E_f) and an open neighborhood N \subseteq Z of f(U) such that the restriction \sigma \colon M \to N of s_f \colon E_f \to Z is a diffeomorphism. Since \overline{U_0} is a compact subset of U, we can choose an open neighborhood I_0 of t_0 in [0,1] such that F_t(U_0) \subseteq N for all t \in I_0. Therefore, for each t \in I_0, there exists a unique \mathcal{C}^{\infty} map \zeta_t \colon U_0 \to E_f satisfying \zeta_t(U_0) \subseteq M and F_t(x) = \sigma(\zeta_t(x)) for all x \in U_0. Writing \zeta_t(x) as \zeta_t(x) = (\alpha_t(x), \xi_t(x)), where \alpha_t \colon U_0 \to U and \xi_t \colon U_0 \to E are \mathcal{C}^{\infty} maps, we get

By Definition 3.2?, s(\xi_t(x)) \in h^{-1}(h(p(\xi_t(x)))), and hence

On the other hand,

Consequently, \alpha_t(x)=x. It follows that \zeta_t \colon U_0 \to E_f is a \mathcal{C}^{\infty} section, over U_0, of the vector bundle p_f \colon E_f \to U. Clearly, \zeta_{t_0}(U_0) \subseteq Z(E_f). Furthermore, the map

is continuous. Note that \zeta(x,t) = (x,\xi(x,t)), where

is a continuous map with p(\xi(x,t)) = f(x) = F(x,t_0) for all (x,t) \in U_0 \times I_0. By construction, the map \xi satisfies conditions ?–?.

By Lemma 3.3, the submersion h \colon Z \to X admits a dominating spray (\tilde E, \tilde p, \tilde s) such that \tilde p \colon \tilde E = Z \times \mathbb{R}^m \to Z is the product vector bundle. In view of Lemma 3.7 and the compactness of the interval [0,1] (see the Lebesgue lemma for compact metric spaces [bib22, p. 28, Lemma 9.11]), there exists a partition 0=t_0 < t_1 < \cdots < t_k =1 of [0,1] such that for each i=1,\ldots,k there exists a continuous map \xi^i \colon U_0 \times [t_{i-1},t_i] \to \tilde E with the following properties:

• \xi^i(x,t) = (F(x,t_{i-1}), \eta^i(x,t)) for all (x,t) \in U_0 \times [t_{i-1},t_i],

• \eta^i(x, t_{i-1}) = 0 for all x \in U_0,

• \tilde s(F(x,t_{i-1}), \eta^i(x,t)) = F(x,t) for all (x,t) \in U_0 \times [t_{i-1}, t_i],

• for every t \in [t_{i-1}, t_i] the map U_0 \to \mathbb{R}^m, x \mapsto \eta^i(x,t) is of class \mathcal{C}^{\infty}.

For i=1,\ldots,k we define recursively a dominating spray (E(i), p^{(i)}, s^{(i)}) for h \colon Z \to X by

while for i \geq 2 we require

to be the product vector bundle and set

where z \in Z and v_1,\ldots,v_i \in \mathbb{R}^m.

In particular, (E,p,s) \coloneqq (E(k), p^{(k)}, s^{(k)}) is a dominating spray for h \colon Z \to X. Note that p \colon E = Z \times \mathbb{R}^n \to Z is the product vector bundle with \mathbb{R}^n = (\mathbb{R}^m)^k. Now, consider a map

where \eta \colon U_0 \times [0,1] \to \mathbb{R}^n = (\mathbb{R}^m)^k is defined by

for all (x,t) \in U_0 \times [t_0,t_1], and

for all (x,t) \in U_0 \times [t_{i-1}, t_i] with i=2,\ldots,k. One readily checks that \eta is a well-defined continuous map satisfying ?–?.

Let F \colon U \times [0,1] \to Z be a homotopy of \mathcal{C}^{\infty} sections such that F_0 is a regular section and F_1 = f. Let U_0 be an open subset of X such that its closure \overline{U_0} is compact and contained in U, and let (E,p,s), \xi \colon U_0 \times [0,1] \to E, \xi(x,t) = (F(x,0), \eta(x,t)), {\eta \colon U_0 \times [0,1] \to \mathbb{R}^n} be as in Lemma 3.8. In particular, we have

By the Weierstrass approximation theorem, there exists a regular map \beta \colon X \to \mathbb{R}^n such that the restriction \beta|_{U_0} is arbitrarily close to the \mathcal{C}^{\infty} map \eta_1 \colon U_0 \to \mathbb{R}^n, x \mapsto \eta(x,1) in the \mathcal{C}^{\infty} topology. Then

is a regular map such that g|_{U_0} is close to f|_{U_0} in the \mathcal{C}^{\infty} topology. Moreover, in view of Definition 3.2?, g \colon U \to Z is a section of h \colon Z \to X. The proof is complete since U_0 is chosen in an arbitrary way.

Let (E,p,s) be a dominating spray for Y. We obtain a dominating spray (\tilde E, \tilde p, \tilde s) for h \colon X \times Y \to X setting

It suffices to prove ?\Rightarrow?. To this end let \Phi \colon X \times [0,1] \to Y be a homotopy such that \Phi_0 is a regular map and \Phi_1=f. We may assume that \Phi is a \mathcal{C}^{\infty} map, see [bib49a, Proposition 10.22]. By Lemma 4.1, the canonical projection h \colon X \times Y \to X is a malleable submersion. Clearly,

is a homotopy of \mathcal{C}^{\infty} sections of h \colon X \times Y \to X. Therefore, according to Corollary 3.10, the \mathcal{C}^{\infty} section

can be approximated by regular sections in the \mathcal{C}^{\infty} topology, which implies that ? holds.

It suffices to prove ?\Rightarrow?. Suppose that ? holds, and let \gamma \colon X \to Y be a regular map homotopic to f. We may assume that X,Y are Zariski closed subsets of \mathbb{R}^k,\mathbb{R}^l, respectively. Then there exists a Zariski open neighborhood \Omega \subseteq \mathbb{R}^k of X and a regular map \tilde \gamma \colon \Omega \to \mathbb{R}^l with \tilde \gamma|_X = \gamma, see [bib2, p. 62]. By the Tietze extension theorem, there exists a continuous map \tilde f \colon \Omega \to \mathbb{R}^l with \tilde f|_X = f. Now, let \rho \colon T \to Y be a \mathcal{C}^{\infty} tubular neighborhood of Y in \mathbb{R}^l, where T is an open neighborhood of Y in \mathbb{R}^l and \rho is a \mathcal{C}^{\infty} retraction. Choose an open neighborhood U \subseteq \Omega of X such that \tilde f(U) \subseteq T and \tilde \gamma(U) \subseteq T. By [bib32, p. 44, Theorem 2.2], we can also choose a \mathcal{C}^{\infty} map U \to T arbitrarily close to \tilde f|_U \colon U \to T in the strong \mathcal{C}^0 topology; such a map is homotopic to \tilde f|_U \colon U \to T. Therefore, for the proof of ?, we may assume that the map \tilde f|_U \colon U \to T is of class \mathcal{C}^{\infty}. Clearly, the maps \varphi, \psi \colon U \to Y, defined by

are of class \mathcal{C}^{\infty} and satisfy \varphi|_X=\gamma, \psi|_X=f. Shrinking U, the variety X is a continuous retract of U (X is locally contractible [bib2, Theorem 9.3.6], so it is a Euclidean neighborhood retract by Borsuk’s theorem [bib22, p. 537, Theorem E.3]). Consequently, the \mathcal{C}^{\infty} maps \varphi, \psi are homotopic, and hence there exists a \mathcal{C}^{\infty} homotopy \Phi \colon U \times [0,1] \to Y with \Phi_0 = \varphi and \Phi_1 = \psi [bib49a, Proposition 10.22]. Let U_0 \subseteq \mathbb{R}^k be an open subset whose closure \overline{U_0} is compact and contained in U.

Consider the canonical projection h \colon \mathbb{R}^k \times Y \to \mathbb{R}^k and the homotopy of its \mathcal{C}^{\infty} sections

According to Lemma 4.1, h \colon \mathbb{R}^k \times Y \to \mathbb{R}^k is a malleable submersion. Hence, by Lemma 3.8, there exist a dominating spray (E,p,s) for h \colon \mathbb{R}^k \times Y \to \mathbb{R}^k and a continuous map {\xi \colon U_0 \times [0,1] \to E} such that p \colon E = (\mathbb{R}^k \times Y) \times \mathbb{R}^n \to \mathbb{R}^k \times Y is the product vector bundle and \xi(x,t) = (F(x,0), \eta(x,t)), where the continuous map \eta \colon U_0 \times [0,1] \to \mathbb{R}^n satisfies

In particular,

By the Weierstrass approximation theorem, there exists a regular map \beta \colon \mathbb{R}^k \to \mathbb{R}^n such that the restriction \beta|_{U_0} is arbitrarily close to the \mathcal{C}^{\infty} map \eta_1 \colon U_0 \to \mathbb{R}^n, x \mapsto \eta(x,1) in the \mathcal{C}^0 topology. Then

is a \mathcal{C}^{\infty} section of h \colon \mathbb{R}^k \times Y \to \mathbb{R}^k such that \alpha|_{U_0} is close in the \mathcal{C}^0 topology to the section

We have \alpha(x) = (x, \theta(x)), where \theta \colon U \to Y is a \mathcal{C}^{\infty} map. Since F(x,0) = (x, \varphi(x)) and \varphi|_X = \gamma is a regular map, the restriction g \coloneqq \theta|_X \colon X \to Y is a regular map. By construction, g|_{X \cap U_0} is close to the continuous map \psi|_{X \cap U_0} = f|_{X \cap U_0}. The proof is complete since U_0 is chosen in an arbitrary way.

By Theorem 1.1, ? implies ?. To prove the reversed implication suppose that ? does not hold. Then, by Chevalley’s theorem [bib25] (see [bib26] for a modern treatment), there exists a surjective morphism of algebraic \mathbb{R}-groups \varphi \colon \Gamma \to A, where A is an Abelian \mathbb{R}-variety, \dim A \geq 1. Since \varphi is defined over \mathbb{R}, its restriction \varphi(\mathbb{R}) \colon \Gamma(\mathbb{R}) \to A(\mathbb{R}) is a regular map of real algebraic varieties. The image \varphi(\mathbb{R})(\Gamma(\mathbb{R})) is Zariski dense in A(\mathbb{R}), the map \varphi being surjective. Hence there exists a point a \in \Gamma(\mathbb{R}) at which the derivative d_a \varphi(\mathbb{R}) \colon T_a\Gamma(\mathbb{R}) \to T_{\varphi(\mathbb{R})(a)}A(\mathbb{R}) is surjective. Therefore, in view of the rank theorem for \mathcal{C}^{\infty} maps, we can find a continuous null homotopic map f \colon \mathbb{S}^k \to \Gamma(\mathbb{R}) such that the composite map \varphi(\mathbb{R}) \circ f \colon \mathbb{S}^k \to A(\mathbb{R}) is not constant. It follows that \varphi(\mathbb{R}) \circ f cannot be approximated by regular maps in the \mathcal{C}^0 topology because each regular map from \mathbb{S}^k into A(\mathbb{R}) is constant, see [bib52, Corollary 3.9]. Consequently, ? does not hold. In other words, ? implies ?.

Suppose that f \colon X \to \mathbb{S}^n is a regular map with \deg(f)=1. By [bib4, Corollary 4.2], for every integer d there exists a regular map \varphi_d \colon \mathbb{S}^n \to \mathbb{S}^n with \deg(\varphi_d)=d. Since the composite map \varphi_d \circ f is regular and \deg(\varphi_d \circ f) = d, it follows from Hopf’s theorem that each \mathcal{C}^{\infty} map from X into \mathbb{S}^n is homotopic to a regular map. Therefore the set \mathcal{R}(X, \mathbb{S}^n) is dense in the space \mathcal{C}^{\infty}(X, \mathbb{S}^n) by Corollary 1.5. The converse is obvious.

By Hopf’s theorem, there are exactly two homotopic classes of \mathcal{C}^{\infty} maps from X into \mathbb{S}^n. One of them is represented by a constant map (which obviously is a regular map). Therefore the proof is complete in view of Corollary 1.5.

Let f \colon \mathbb{S}^n \to Y be a \mathcal{C}^{\infty} map. Choose an element a \in G such that a \cdot f(s_n)=y_0. Since \pi^{\mathrm{alg}}_n(Y,y_0) = \pi_n(Y,y_0), the \mathcal{C}^{\infty} map g \colon \mathbb{S}^n \to Y, x \mapsto a \cdot f(x) is homotopic to a regular map, and hence, by Theorem 1.1, it can be approximated by regular maps in the \mathcal{C}^{\infty} topology. Consequently, the map f can be approximated by regular maps in the \mathcal{C}^{\infty} topology because f(x) = a^{-1} \cdot g(x) for all x \in \mathbb{S}^n.

? This is proved in [bib3, Theorem 1.1] and has already been mentioned in Section ?.

? The case n=p is contained in Theorem 1.6. For 1 \leq n-p \leq 3, one has \pi^{\mathrm{alg}}_n(\mathbb{S}^p) = \pi_n(\mathbb{S}^p) by [bib56, Corollaries 1.2, 1.3 and 1.4]. The case n<p is trivial.

? One has \pi_{p+4}(\mathbb{S}^p)=0 for p \geq 6 and \pi_{p+5}(\mathbb{S}^p)=0 for p \geq 7 (see [bib35, pp. 331, 332]), which together with ? completes the proof.

? In this case, \pi^{\mathrm{alg}}_n(\mathbb{S}^p) = \pi_n(\mathbb{S}^p) by [bib4, Theorem 2.1 and p. 163].

? One has \pi_{p+13}(\mathbb{S}^p) = \mathbb{Z}/3 if p \geq 15 (see [bib59, p. 188]), so the assertion follows from ?.

In the case under consideration, the H-space multiplication Y \times Y \to Y, (a,b) \mapsto ab and the inverse operation Y \to Y, a \mapsto a^{-1} are regular maps. The assertion follows since the group operation in \pi_n(Y) is induced by the H-space multiplication.

According to Bott’s periodicity theorem [bib21] (see also [bib36, Chap. 8, Remark 4.2, and 12.2]) the nth homotopy group \pi_n(Y) is trivial in cases ?, ? and ?. Furthermore, by [bib36, Chap. 8, Theorem 5.1], the homotopy group \pi_n(Y) is trivial in case ?. Therefore the proof is complete in view of Corollary 1.2.

By Proposition 5.5, it is sufficient to show that \pi^{\mathrm{alg}}_{2k-1}(G(m)) = \pi_{2k-1}(G(m)).

Suppose that (n,m)\neq(5,3) and (n,m)\neq(6,3). Then the only cases not already covered by Proposition 6.1 and Theorem 6.2 are (n,m) equal to (4,2), (5,2) and (6,2). Hence, by Proposition 5.5, it is sufficient to show that \pi^{\mathrm{alg}}_n(G(2)) = \pi_n(G(2)) for n=4,5 and 6. This is proved in [bib56, Corollaries 1.2, 1.3 and 1.4] for G(2)=\mathrm{SU}(2)=\mathbb{S}^3. The case G(2)=\mathrm{U}(2) follows since the inclusion map \mathrm{SU}(m) \hookrightarrow \mathrm{U}(m) induces for all i \geq 2 an isomorphism between the corresponding ith homotopy groups, see [bib36, Chap. 8, 12.2].