6
$\begingroup$

Let $๐‘‹$ be an infinite-dimensional Banach space (complex or real). A subspace of $๐‘‹$ means a closed linear submanifold. Subspaces $M$ and $N$ of $X$ are quasi-complementary if $M\cap N=\{0\}$ and $M+N$ is dense in $X$. In the case $M+N=X$ we say that $M$ and $N$ are complementary.

Question. If $n\geq 2$, do there exist subspaces $M_1,\ldots, M_n$ of $X$ which are pairwise quasi-complementary but not complementary, that is, $$ M_i\cap M_j=\{0\},\quad \overline{M_i+M_j}=X,\quad \text{but}\quad M_i+M_j\ne X\quad (1\leq i< j\leq n).$$

My question is related to the problem of realization of an abstract lattice as a subspace lattice. For instance, the pentagon cannot be realized in a finite-dimensional Banach space, however, it can be realized as a subspace lattice in the infinite-dimensional separable Hilbert space (Halmos: Reflexive lattices of subspaces). I would like to know if the pentagon (and some other abstract lattices) can be realized in a (infinite-dimensional) Banach space that is not isomorphic to a Hilbert space.

$\endgroup$
8
  • $\begingroup$ I'm not sure what do you mean exactly by your last question: you can take a lattice of subspaces of a Hilbert space, and just take direct sum with anything non-Hilbert. One way to make sense of this is to ask whether a pentagon realisable by non-complemented subspaces. But this is still trivially true... $\endgroup$ Commented Sep 3, 2024 at 17:06
  • $\begingroup$ Denis T, can you please explain your construction in more detail? A realization of the pentagon in a given Banach space $X$ would be a subspace lattice with non-trivial subspaces $M$, $N$, and $K$ of $X$ such that $M\wedge K=\{0\}$, $M\vee N=X$, and $N\subsetneq K$. Here $\wedge$ is the intersection and $\vee$ is closure of the sum. Of course, the pentagon can be realizes is some Banach spaces, but can be it realized in any infinite-dimensional Banach space? $\endgroup$ Commented Sep 4, 2024 at 4:46
  • 1
    $\begingroup$ Regarding the last paragraph: A lattice is modular iff it doesn't contain a pentagon (e.g. p.80 in Grätzer's book books.google.ca/books?id=SoGLVCPuOz0C&pg=PA80) A closed subspace lattice of a Banach space $E$ is modular iff $E$ is finite dimensional (e.g. p.144 in Maeda & Maeda 's book books.google.ca/books?id=p8DqCAAAQBAJ&pg=PA144 ) $\endgroup$ Commented Sep 4, 2024 at 19:38
  • 1
    $\begingroup$ Together; If $E$ is infinite dimensional Banach space, then its closed subspace lattice contains a pentagon. $\endgroup$ Commented Sep 4, 2024 at 19:40
  • 1
    $\begingroup$ Why, @JankoBracic? The join of the pentagon need not be the entire space. $\endgroup$ Commented Sep 5, 2024 at 17:49

1 Answer 1

7
$\begingroup$

It is well known that every subspace $Y$ of a separable Banach space $X$ is quasi-complemented.We denote its quasi-complement by $Z$. This is a classical result due to F. J. Murray and G. Mackey.The proof is based on the existence of $\{(y_n, f_n)_{n} \cup (z_n, g_n)_n \}$ which is a biorthogonal system of $X$ with $(y_n)_n, (z_n)_n $ generate the subspaces $Y,Z$ respectively and $(\bigcap_n \operatorname{Ker} f_n) \cap(\bigcap_n \operatorname{Ker} g_n) = \{ 0 \}, $

Theorem : Let $ X $ be a separable Banach space, $Y$ be an infinite dimensional subspace with infinite dimension quasi-complement $Z$. Choose $(y_k,f_k)_{k\in N}$, $(z_k, g_k)_{k\in N}$ the biorthogonal system associated to $Y, Z$.

For a given $n\in N$ choose $(a_i, b_i)_{i=1}^n$ in the unit sphere of $R^2$ with $a_ib_j - b_ia_j \neq 0$ and for each $i= 1\dots n$ set $M_i$ be the closed linear span of the sequence $\{a_iy_k+b_iz_k\}_k$.

Then for every $i \neq j$ the subspaces $M_i , M_j $ satisfy

$M_i \cap M_j = \{0\}$ and $M_i + M_j$ dense in $X$.

Proof : Fix $i \neq j$ and consider the following biorthogonal system

$\{ (a_iy_k + b_iz_k, b_jf_k - a_jg_k)_k\cup (a_jy_k + b_jz_k, b_if_k - a_ig_k)_k$.

This corresponds to the pair $M_i M_j$. It is easy to see that $ (y_k)_k \cup (z_k)_k \subset M_i + M_j $ hence $M_i + M_j$ is dense in $X$.

For similar reasons

$\langle \{ b_jf_k-a_jg_k \}_k \cup \{ b_if_k-a_ig_k \}_k \rangle = \langle \{f_k\}_k \cup \{g_k\}_k \rangle $

which is $ w^* $ dense in $X^*$.Hence

$(\bigcap_k \operatorname{Ker} b_jf_k- a_jg_k) \cap(\bigcap_k \operatorname{Ker} b_if_k-a_ig_k) = \{ 0 \}$

which yields that $M_i \cap M_j = \{0\}$.

Remark 1 : Clearly there exists a sequence $\{M_i\}_{i\in N}$ satisfying the conclusion of the theorem.

Remark 2 : The theorem remains valid for non separable reflexive Banach spaces and more generally for WCG spaces.

Indeed let $X$ be a WCG space. From Amir Lindenstrauss theorem there exists a family $(P_{\alpha})_{\alpha\in A}$ mutually orthogonal bounded projections such that for each $\alpha\in A$ we have that $P_{\alpha}(X)$ is separable and $\langle \bigcup_{\alpha}P_{\alpha}(X) \rangle$ is dense in $X$. For a fixed $ n\in N$ and every $ \alpha\in A$ choose a family $(M_i^{\alpha})_{i=1}^n$ of closed subspaces of $P_{\alpha}(X)$ which are pairwise quasi-complemented in $P_{\alpha}(X)$.

For $ i= 1\dots n$ set $M_i = \overline{\langle\bigcup_{\alpha}M_i^{\alpha}\rangle}$. The family $(M_i)_{i=1}^n$ has the desided property.Indeed it is clear that for $i \neq j$ $ M_i + M_j $ is dense in $X$. To see that $M_i \cap M_j =\{ 0 \}$ assume on the contrary that there exists $x \in M_i \cap M_j$ with $x\neq 0$.Then there exists $ \alpha\in A$ such that $P_{\alpha}(x) \neq 0$.Since $P_{\alpha}(M_q) = M_q^{\alpha}$ for all $q= 1\dots n$ we derive a contradiction.

I also found interesting the question, stated by Janko Bracic, if the spaces $(M_i)_i$ could be selected to be pairwise isomorphic. It has an affirmative answer and is explained below.

Remark 3 : Assume that for all $k\in N$ $\|f_k\| = \|g_k\|=1$ and $f_k(y_k) = g_k(z_k) =1$.For $i= 1 \dots n $ and $k\in N$ choose $a_i^k, b_i^k$ and $\epsilon_k > 0$ such that the following hold.

  1. For $i \neq j$ $\frac{1}{|a_i^k|}(|a_i^k - a_j^k|\|y_k\| + |b_i^k - b_j^k|\|z_k\|) < \epsilon_k$ and $\sum_k \epsilon_k < \frac{1}{2}$.

  2. For every $k\in N$ and $i\neq j$ we have $a_i^k, a_j^k \neq0$ and $ a_i^k b_j^k- b_i^ka_j^k\neq 0$.

Define

$M_i = \overline { \langle (a_i^ky_k + b_i^kz_k)_k \rangle}$

It follows easily that the new family $(M_i)_{i=1}^n$ satisfies the conclusion of the theorem.We claim that the spaces are pairwise isomorphic. Indeed let $ y_i = \sum_{k=1}^l c_k(a_i^ky_k + b_i^kz_k)$ be a normalized vector in $M_i$. Then $|f_k (y_i)| =|c_ka_i^k|\leq 1$ hence $|c_k|\leq \frac{1}{|a_i^k|}$.

This yields that $\|y_i -y_j\|\leq \frac{1}{2}$ which shows that $M_i, M_j$ are isomorphic.

Remark 4 : There is something which I could not understand. If $M_i, M_j$ are spaces as in Remark 3 then $\operatorname{dist} (B_{M_i}, B_{M_j}) = 0$. Hence $M_i, M_j$ are not a complemented pair. But if the initial subspaces $Y, Z$ are a complemented pair then $M_i + M_j = Y+Z= X$ hence $M_i, M_j$ is a complemented pair. Any help?

Remark 5 : There is no problem concerning the proof of Remark 3. My false was that I consider that $M_i + M_j = Y+Z$. What actually we have is that $\langle (y_k)_k \cup (z_k)_k \rangle \subset M_i +M_j$ which does not imply that $\overline {\langle (y_k)_k \rangle} + \overline{ \langle (z_k)_k \rangle} \subset M_i +M_j$.

Therefore Remark 3 yields that independently to what is the initial quasi-complemented pair $Y,Z$( complemented or not) the $(M_i)_i$ resulting from Remark 3, since $\operatorname{dist} (B_{M_i}, B_{M_j}) = 0$, are pairwise quasi-complemented but not complemented. Hence Remark 3 answers completely the posed question for separable and WCG spaces.

$\endgroup$
7
  • $\begingroup$ @S Argyros Thank you for the answer! Is there any hope for the existence of subspaces $M_j$ in the case when $X$ is not separable? I know that Lindenstrauss proved that every subspace of a reflexive Banach space is quasi-complemented. Hence, if $X$ is reflexive and not isomorphic to a Hilbert space there exists a subspace which is quasi-complemented but not complemented (for a Hilbert space we can construct such a subspaces using an orthogonal basis). $\endgroup$ Commented Sep 5, 2024 at 10:23
  • $\begingroup$ Thus, this answers my question for a reflexive space $X$ and two subspaces. But I don't see how to prove the existence of three or more subspaces which are pairwise quasi-complemented but not complemented (should they be isomorphic subspaces?). $\endgroup$ Commented Sep 5, 2024 at 10:24
  • 2
    $\begingroup$ The theorem states that for every separable Banach space $X$ (reflexive or not) and every $n\in N$ there exists $(M_i)_{i=1}^n$ closed subspaces which are mutually quasi-complement.The remark states that this family could be an infinite sequence. $\endgroup$ Commented Sep 5, 2024 at 10:38
  • 2
    $\begingroup$ Concerning non separable spaces this approach has some difficulties to be applied. $\endgroup$ Commented Sep 5, 2024 at 10:40
  • 2
    $\begingroup$ A proof for WCG spaces has been added in Remark 2. $\endgroup$ Commented Sep 5, 2024 at 12:00

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.