Manifolds modeled on locally convex spaces

19 Counterexamples

Counterexamples that separate notions which agree in finite dimensions. This chapter collects those about immersions and Finsler structures: a weak immersion that is not an immersion (\(\ell ^1 \to \ell ^2\)), an injective immersion that is not initial (the figure eight), an immersion between Banach spaces that is not split (\(c_0 \subseteq \ell ^\infty \), by Phillips’ lemma), and a Finsler length structure that is not a Finsler metric. Three more are stated next to the results they limit: McAlpin’s complete Riemannian manifold without shortest curves (Example 12.13), a bounded Finsler structure on an incomplete model whose distance does not give the manifold topology (Theorem 13.51), and a weak Riemannian metric without Levi-Civita connection (Theorem 13.37).

19.1 Weak immersions, injective immersions and split immersions

The notions of immersion form a chain: split immersions are immersions, immersions are weak immersions, and injective immersions need not be initial. In finite dimensions the first two notions coincide. The three examples of this section show that every implication is strict, and that the first two separations occur already between Banach spaces.

19.1.1 A weak immersion that is not an immersion

Let \(\iota \colon \ell ^1 \to \ell ^2\) be the inclusion of real sequences. Both spaces are Banach spaces, hence manifolds with a single chart.

The inclusion \(\iota \) is continuous, linear and injective, hence a weak immersion. It is not bounded below, hence it is not an immersion.

Proof ▶

Continuity follows from the closed graph theorem: the graph of \(\iota \) is closed because the coordinate evaluations are continuous on both spaces. The differential of \(\iota \) at every point is \(\iota \), which is injective. For \(x_n = e_0 + \dots + e_{n-1}\) we have \(\| x_n\| _1 = n\) and \(\| x_n\| _2 = \sqrt n\), so no inequality \(m\| x\| _1 \le \| \iota x\| _2\) with \(m {\gt} 0\) can hold, and \(\iota \) is not a topological isomorphism onto its range.

Remark 19.2
#

The range of \(\iota \) is dense in \(\ell ^2\) and not closed. In finite dimensions an injective linear map is automatically a topological isomorphism onto a closed subspace; in infinite dimensions the condition that the differential be an embedding with closed range is a genuine additional requirement, and this is the reason to distinguish weak immersions from immersions.

19.1.2 An injective immersion that is not initial: the figure eight

Let \(\theta (t) = 2\arctan t \in (-\pi , \pi )\) and

\[ j \colon \mathbb {R}\to \mathbb {R}^2, \qquad j(t) = (\sin \theta (t), \sin 2\theta (t)). \]

As \(t\) runs over \(\mathbb {R}\), \(j(t)\) traces the figure eight (a lemniscate), passing through the origin at \(t = 0\) and tending to the origin again as \(t \to \pm \infty \).

The map \(j\) is smooth, injective and has nonvanishing velocity; hence it is an injective immersion.

Proof ▶

On \((-\pi , \pi )\) the pair \((\sin a, \sin 2a)\) determines \(a\), which gives injectivity. If \(\dot j(t) = 0\), then \(\cos \theta = 0\) and \(\cos 2\theta = 0\) at the same time, which is impossible. A nonzero vector spans a closed line, so \(v \mapsto v\, \dot j(t)\) is a topological isomorphism onto a closed subspace.

Let \(\sigma (s) = \tfrac 12\arctan s\) and \(f(s) = (\sin \sigma , -\sin 2\sigma )\). Then \(f\) is a smooth curve in \(\mathbb {R}^2\) with values in \(j(\mathbb {R})\), but its unique lift \(g\) with \(j \circ g = f\) is not continuous at \(0\). Hence \(j\) is not an initial submanifold.

Proof ▶

The curve \(f\) crosses the origin along the branch that \(j\) reaches only as \(t \to \pm \infty \). Explicitly, \(g = \tan (\lambda /2)\) with \(\lambda (s) = \pi - \sigma (s)\) for \(s {\gt} 0\), \(\lambda (s) = -\pi - \sigma (s)\) for \(s {\lt} 0\) and \(\lambda (0) = 0\); trigonometric identities give \(j \circ g = f\). For \(s {\gt} 0\) we have \(g(s) {\gt} 1\), while \(g(0) = 0\). If \(j\) were initial, \(g\) would be smooth, in particular continuous.

Remark 19.5
#

This is the classical finite dimensional example; it shows that the condition defining initial submanifolds is not a consequence of being an injective immersion, and has nothing to do with infinite dimensions. The topology of \(\mathbb {R}\) transported by \(j\) is strictly finer than the topology induced by \(\mathbb {R}^2\) on the figure eight.

19.1.3 An immersion that is not split: \(c_0 \subseteq \ell ^\infty \)

We realize \(\ell ^\infty \) as the Banach space of bounded real sequences and \(c_0\) as the closed subspace of sequences tending to \(0\). The classical obstruction is the following result of Phillips [ Phi40 , 7.5, pp. 539–540 ] , stated there for the space \(c\) of convergent sequences, which contains \(c_0\) with codimension one; we formalize the short proof of Whitley [ Whi66 ] .

There is no continuous linear projection of \(\ell ^\infty \) onto \(c_0\).

The proof rests on an uncountable family of almost disjoint subsets of \(\mathbb {N}\), and on the fact that a functional vanishing on \(c_0\) can only see countably many of them.

For each \(A \subseteq \mathbb {N}\) there is an infinite set \(N_A \subseteq \mathbb {N}\) such that \(N_A \cap N_B\) is finite whenever \(A \ne B\). The indicator \(\chi _A \in \ell ^\infty \) of \(N_A\) does not belong to \(c_0\).

Proof ▶

Index the finite binary words by \(\mathbb {N}\), and let \(N_A\) be the set of indices of the initial segments of the indicator function of \(A\). If \(A\) and \(B\) differ at \(i_0\), their initial segments of length larger than \(i_0\) are different, so \(N_A \cap N_B\) is finite. Since \(N_A\) is infinite, \(\chi _A\) takes the value \(1\) infinitely often.

Let \(f\) be a continuous linear functional on \(\ell ^\infty \) that vanishes on \(c_0\). If \(|x_k| \le 1\) for all but finitely many \(k\), then \(|f(x)| \le \| f\| \). Consequently, \(f(\chi _A) \ne 0\) only for countably many \(A \subseteq \mathbb {N}\).

Proof ▶

For the first claim, modify \(x\) on a finite set to obtain \(y\) with \(\| y\| \le 1\); then \(x - y \in c_0\) and \(f(x) = f(y)\). For the second, suppose \(|f(\chi _A)| {\gt} 1/m\) for every \(A\) in a finite set \(\mathcal F\), and let \(x = \sum _{A \in \mathcal F} \pm \chi _A\), with the signs of \(f(\chi _A)\). Off the finite set where two of the \(N_A\) meet, \(|x_k| \le 1\); hence \(\# \mathcal F/m {\lt} f(x) \le \| f\| \). So each set \(\{ A : |f(\chi _A)| {\gt} 1/m\} \) is finite.

Proof of Theorem 19.6 ▶

Suppose \(P\) is a continuous projection onto \(c_0\), and apply Lemma 19.8 to the functionals \(f_n = e_n \circ (1 - P)\), where \(e_n\) is the \(n\)-th coordinate. The subsets of \(\mathbb {N}\) are uncountable, so some \(A\) avoids the countable union of the exceptional sets. Then \((1 - P)\chi _A = 0\), that is, \(\chi _A = P\chi _A \in c_0\), a contradiction.

The subspace \(c_0\) is an embedded submanifold of \(\ell ^\infty \) that is not split, and the inclusion \(c_0 \to \ell ^\infty \) is an immersion that is not a split immersion.

Proof ▶

Combine Propositions 18.1 and 18.2 with Theorem 19.6.

Remark 19.10
#

In Hilbert spaces every closed subspace is complemented, so the distinction between embedded and split submanifolds, and between immersions and split immersions, is invisible there. For Banach spaces it is not, and Corollary 19.9 gives the standard witness.

19.2 A Finsler length structure that is not a Finsler metric

The difference between the two notions is not an infinite-dimensional phenomenon. On \(\mathbb {R}^2\) with the sup norm, let

\[ \beta (x, y) = \frac{\max (0, x^2 - |y|)}{x^2 + |y|} \quad (\beta (0,0) = 0), \qquad \mu _p(v) = \| v\| + \beta (p)\, |v_2| . \]

The weight \(\beta \) takes values in \([0,1]\), is supported in the cusp \(\{ |y| {\lt} x^2\} \), equals \(1\) on the \(x\)-axis minus the origin, and is continuous away from the origin.

Proof ▶

It is the sum of the norm and a nonnegative multiple of the seminorm \(|v_2|\).

Proof ▶

It is not continuous on \(T\mathbb {R}^2\) at \((0, e_2)\): \(\mu _{(t,0)}(e_2) = 2\) for \(t \neq 0\), while \(\mu _0(e_2) = 1\). Let now \(\gamma \) be a \(C^{1}\) curve. Away from the times where \(\gamma \) passes through the origin, \(\beta \circ \gamma \) is continuous. Suppose that \(\gamma (t_0) = 0\). If \(\dot\gamma _2(t_0) \neq 0\), then \(|\gamma _2(t)| \ge c\, |t - t_0|\) while \(\gamma _1(t)^2 = O(|t-t_0|^2)\), so near \(t_0\) the curve stays outside the cusp and \(\beta (\gamma (t)) = 0\). If \(\dot\gamma _2(t_0) = 0\), then \(0 \le \beta (\gamma )\, |\dot\gamma _2| \le |\dot\gamma _2| \to 0\).

The example shows what a \(C^{1}\) curve can and cannot see: through a point \(p\) it only meets the pairs \((p, v)\) that are compatible with the way it approaches \(p\). The pairs \(((1/n, 0), e_2)\), which converge to \((0, e_2)\) in \(T\mathbb {R}^2\), are not of this kind.

AI89

T. Ya. Azizov and I. S. Iokhvidov, Linear Operators in Spaces with an Indefinite Metric, Wiley, 1989.

Atk75

C. J. Atkin, The Hopf–Rinow theorem is false in infinite dimensions, Bull. London Math. Soc. 7 (1975), 261–266.

BBI01

D. Burago, Y. Burago and S. Ivanov, A Course in Metric Geometry, Graduate Studies in Mathematics 33, American Mathematical Society, 2001.

Bel05

D. Beltiță, Integrability of analytic almost complex structures on Banach manifolds, Ann. Global Anal. Geom. 28 (2005), 59–73.

BL23

D. Beltiţă and G. Larotonda, Unitary group orbits versus groupoid orbits of normal operators, J. Geom. Anal. 33 (2023), no. 3, Paper No. 95, doi:10.1007/s12220-022-01187-5.

Bog74

J. Bognár, Indefinite Inner Product Spaces, Ergebnisse der Mathematik und ihrer Grenzgebiete 78, Springer, 1974.

Car57

H. Cartan, Variétés analytiques réelles et variétés analytiques complexes, Bull. Soc. Math. France 85 (1957), 77–99, doi:10.24033/bsmf.1481.

dMU21

L. de Moura and S. Ullrich, The Lean 4 theorem prover and programming language, in Automated Deduction – CADE 28, Lecture Notes in Comput. Sci. 12699, Springer, 2021, 625–635.

Eke78

I. Ekeland, The Hopf–Rinow theorem in infinite dimension, J. Differential Geom. 13 (1978), 287–301.

GN26

H. Glöckner and K.-H. Neeb, Infinite-Dimensional Lie Groups, book manuscript, arXiv:2602.12362, 2026.

GLT

T. Goliński, G. Larotonda, A. B. Tumpach, Nijenhuis operators on Banach homogeneous spaces, Rend. Lincei Mat. Appl. 35 (2024), 713–739, doi:10.4171/RLM/1057.

Gra58

H. Grauert, On Levi’s problem and the imbedding of real-analytic manifolds, Ann. of Math. (2) 68 (1958), 460–472, doi:10.2307/1970257.

Ham82

R. S. Hamilton, The inverse function theorem of Nash and Moser, Bull. Amer. Math. Soc. 7 (1982), 65–222.

Hat02

A. Hatcher, Algebraic Topology, Cambridge University Press, 2002.

Her79

M. R. Herman, Sur la conjugaison différentiable des difféomorphismes du cercle à des rotations, Publ. Math. IHÉS 49 (1979), 5–233.

Hir76

M. W. Hirsch, Differential Topology, Graduate Texts in Mathematics 33, Springer, 1976.

Jor82

H. Joris, Une \(C^\infty \)-application non-immersive qui possède la propriété universelle des immersions, Arch. Math. 39 (1982).

KH95

A. Katok and B. Hasselblatt, Introduction to the Modern Theory of Dynamical Systems, Encyclopedia of Mathematics and its Applications 54, Cambridge University Press, 1995.

KMW24

Y. Karshon, D. Miyamoto and J. Watts, Diffeological submanifolds and their friends, Differential Geom. Appl. 96 (2024), Paper No. 102170, doi:10.1016/j.difgeo.2024.102170.

Lan99

S. Lang, Fundamentals of Differential Geometry, Graduate Texts in Mathematics 191, Springer, 1999.

Lar19

G. Larotonda, The metric geometry of infinite dimensional Lie groups and their homogeneous spaces, Forum Math. 31 (2019), 1567–1605.

Lar26

G. Larotonda, Estructuras geométricas para las variedades de Banach, 2nd ed., unpublished book manuscript, version 3 of 15 August 2026, https://mate.dm.uba.ar/~glaroton/Estructuras_v3.pdf.

LCA

@glaroton, lc-analysis: Michal–Bastiani calculus on locally convex spaces, Lean library, https://github.com/glaroton/lc-analysis.

Lem98

L. Lempert, The Dolbeault complex in infinite dimensions I, J. Amer. Math. Soc. 11 (1998), 485–520.

Mai63

B. Maissen, Über Topologien im Endomorphismenraum eines topologischen Vektorraumes, Math. Ann. 151 (1963), 283–285.

Mar72

J. E. Marsden, Darboux’s theorem fails for weak symplectic forms, Proc. Amer. Math. Soc. 32 (1972), 590–592, doi:10.1090/S0002-9939-1972-0293678-X.

Men13

A. C. G. Mennucci, On asymmetric distances, Anal. Geom. Metr. Spaces 1 (2013), 200–231.

mC20

The mathlib Community, The Lean mathematical library, in Proceedings of the 9th ACM SIGPLAN International Conference on Certified Programs and Proofs (CPP 2020), 2020, 367–381.

Mos65

J. Moser, On the volume elements on a manifold, Trans. Amer. Math. Soc. 120 (1965), 286–294.

MM05

P. W. Michor and D. Mumford, Vanishing geodesic distance on spaces of submanifolds and diffeomorphisms, Doc. Math. 10 (2005), 217–245.

Nee06

K.-H. Neeb, Towards a Lie theory of locally convex groups, Jpn. J. Math. 1 (2006), 291–468.

NN57

A. Newlander, L. Nirenberg, Complex analytic coordinates in almost complex manifolds, Ann. of Math. 65 (1957), 391–404.

Pal66

R. S. Palais, Lusternik–Schnirelman theory on Banach manifolds, Topology 5 (1966), 115–132.

Pen70

J.-P. Penot, Sur le théorème de Frobenius, Bull. Soc. Math. France 98 (1970), 47–80.

Pat99

I. Patyi, On the \(\bar\partial \)-equation in a Banach space, Bull. Soc. Math. France 128 (2000), 391–406.

Phi40

R. S. Phillips, On linear transformations, Trans. Amer. Math. Soc. 48 (1940), 516–541.

Sch23

A. Schmeding, An Introduction to Infinite-Dimensional Differential Geometry, Cambridge Studies in Advanced Mathematics 202, Cambridge University Press, 2023, doi:10.1017/9781009091251.

Wei69

A. Weinstein, Symplectic structures on Banach manifolds, Bull. Amer. Math. Soc. 75 (1969), 1040–1041, doi:10.1090/S0002-9904-1969-12353-0.

Wei71

A. Weinstein, Symplectic manifolds and their Lagrangian submanifolds, Adv. Math. 6 (1971), 329–346.

Whi66

R. Whitley, Projecting \(m\) onto \(c_0\), Amer. Math. Monthly 73 (1966), 285–286.