Suppose that we have a vector space $X$ over the field $\mathbb F \in \{ \mathbb R, \mathbb C \}$.

Question: Does there exist a Hilbert space $\widehat X$ over $\mathbb F$ such that $\widehat X$, when we forget the Hilbert space structure, is the same as $X$ in the category of vector spaces?

Which criteria hold for the existence of such a Hilbert space, on which choices do we have? My motivation is to better understand which structure we gain from the Hilbert space setting (and topological vector spaces in general).

  • 16
    $\begingroup$ For instance, a Hilbert space, like every Banach space, never has a countable infinite dimension, where dimension is the cardinality of a Hamel basis. This is a conseqeunce of the Baire category theorem. $\endgroup$
    – Zero
    Jan 10 '15 at 15:31
  • 6
    $\begingroup$ Since vector spaces are distinguished up to isomorphism by their dimension, your question boils down to: what are the possible cardinalities of Hamel bases for Hilbert spaces? I suspect this has aleady been answered somewhere on MO $\endgroup$
    – Yemon Choi
    Jan 10 '15 at 15:54
  • 12
    $\begingroup$ Some simple cardinal arithmetic shows that the algebraic dimension of an infinite dimensional Hilbert space of analytic dimension $\kappa$ is $\kappa^{\aleph_0}$. $\endgroup$ Jan 10 '15 at 15:54
  • 1
    $\begingroup$ Which is equal to $\max(\kappa, 2^{\aleph_0})$. $\endgroup$
    – Adam P. Goucher
    Jan 10 '15 at 18:06
  • 5
    $\begingroup$ @AdamP.Goucher I don't think that's necessarily true. If $\kappa$ is a cardinal greater than the continuum but of countable cofinality, then $\kappa^{\aleph_0} > \kappa > 2^{\aleph_0}$. See en.wikipedia.org/wiki/… $\endgroup$
    – user43208
    Jan 10 '15 at 18:10

Let's go ahead and prove Eric Wofsey's claim (see his comment under the question), that if $H$ is an infinite-dimensional Hilbert space with an orthonormal basis $U$ of cardinality $\kappa$, then any Hamel basis $B$ of $H$ has cardinality $|B| = \kappa^{\aleph_0}$. Notice this implies that some (real or complex) vector spaces do not admit a Hilbert space structure; e.g., any whose dimension is the limit $\lambda$ of cardinals $\alpha_n$ defined by $\alpha_0 = \aleph_0$ and $\alpha_{n+1} = 2^{\alpha_n}$ (this $\lambda$ has countable cofinality and hence can't be of the form $\lambda = \kappa^{\aleph_0}$, else $\lambda = \lambda^{\aleph_0}$ which would contradict König's theorem, as mentioned in an earlier comment).

First remark: $|B| = |H|$. Since $H$ is infinite-dimensional, we have $|B| \geq 2^{\aleph_0} = |\mathbb{C}|$ (see for example the last two paragraphs of this M.SE post: https://math.stackexchange.com/a/547888/43208), and we have $|H| = |B|\cdot |\mathbb{C}|$ by the lemma at this MO post: https://mathoverflow.net/a/49572/2926). Combining these two, $|B| = |B| \cdot |B| \geq |B| \cdot |\mathbb{C}| = |H|$, and obviously $|B| \leq |H|$, so $|B| = |H|$.

Let $\langle -, -\rangle$ be the inner product. For each $h \in H$, let $N_h = \{e \in U: \langle h, e\rangle \neq 0\}$; notice this is at most countable. Consider $H_{fin} = \{h \in H: |N_h| < \infty\}$ and its complement, which I will denote as $H_\infty$.

Each $h \in H_\infty$ determines and is uniquely determined by the countably infinite set $N_h \subseteq U$ together with the corresponding element $\sum_{e \in N_h} \langle h, e\rangle e$ in $l^2(N_h)$. In this way we have a bijection

$$H_\infty \to \bigcup_{N \in X} L(N) \qquad (1)$$

where $X$ is the collection of countably infinite subsets $N$ of $U$ and $L(N)$ is the subset of $l^2(N)$ whose elements $h$ satisfy $\langle h, e\rangle \neq 0$ for all $e \in N$.

Lemma: If $A$ is an infinite set and $\alpha$ a cardinal such that $\alpha \leq |A|$, then the number of subsets of $A$ of size $\alpha$ is $|A|^\alpha$.

Proof: For each inclusion of a subset of size $\alpha$, we can choose an injective function $\alpha \to A$ with the same image, belonging to the set of all such functions which has cardinality $|A|^\alpha$, so $|A|^\alpha$ is an upper bound. On the other hand, a function $\alpha \to A$ can be identified with its graph, a subset of $\alpha \times A$ of size $\alpha$, and $|\alpha \times A| = |A|$ (excepting $\alpha = 0$ which is trivial), so $|A|^\alpha$ is also a lower bound. $\Box$

Applying this lemma to the bijection (1), we have $|X| = \kappa^{\aleph_0}$. Also we have $|L(N)| = 2^{\aleph_0}$ for each $N$, so $|H_\infty| = \kappa^{\aleph_0} \cdot 2^{\aleph_0} = \kappa^{\aleph_0}$.

Similarly, we have an injection

$$H_{fin} \to \bigcup_{n \geq 0} X_n \times \mathbb{C}^n \qquad (2)$$

where $X_n$ is the collection of subsets of $U$ of finite cardinality $n$. Since $|X_n| = \kappa^n = \kappa$ by the lemma, we have $|H_{fin}| \leq \aleph_0 \cdot \kappa \cdot 2^{\aleph_0} = \max\{\kappa, 2^{\aleph_0}\} \leq \kappa^{\aleph_0}$.

Finally, $\kappa^{\aleph_0} = |H_\infty| \leq |H| = |H_\infty| + |H_{fin}| \leq \kappa^{\aleph_0} + \kappa^{\aleph_0} = \kappa^{\aleph_0}$, so $|H| = \kappa^{\aleph_0}$ as claimed.

Added in response to Sushil's comment below: The converse, that if $\kappa = \kappa^{\aleph_0}$ then there exists a Hilbert space of algebraic dimension $\kappa$, is fairly immediate. Indeed, suppose $B$ is a set of cardinality $\kappa$, and let $V$ be the $\mathbb{C}$-vector space $V$ consisting of formal $\mathbb{C}$-linear combinations of elements of $B$. Assign to $V$ the unique inner product that makes $B$ an orthonormal set of $V$. Then the completion of $V$ with respect to the norm of the inner product is a Hilbert space $H$ whose algebraic dimension is $\kappa$. This is because $B$ is an orthonormal basis of $H$ (by construction of $H$), so the claim of the very first paragraph of this answer shows that any Hamel basis of $H$ has cardinality $\kappa^{\aleph_0} = \kappa$.

  • $\begingroup$ How will prove converse implication for Hilbert space atleast in fact stated by Goldstern. One sided implication for Hilbert space you have proved $\endgroup$
    – Sushil
    Jun 27 '15 at 8:06
  • $\begingroup$ How it follows |A|^α is also a lower bound from identifying function as subset of α×A $\endgroup$
    – Sushil
    Jun 27 '15 at 10:02
  • 1
    $\begingroup$ @Sushil I'll consider the converse a little later, when I have more time and leisure. For the other question, if $\binom{B}{\alpha}$ denotes the collection of subsets of $B$ of size $\alpha$, then by taking graphs of functions we have an injection $A^\alpha \to \binom{\alpha \times A}{\alpha}$, and since $|\alpha \times A| = |A|$ when $\alpha \leq |A|$, there is a bijection $\alpha \times A \to A$, and by composition along this bijection we get an injection $A^\alpha \to \binom{A}{\alpha}$. $\endgroup$
    – user43208
    Jun 27 '15 at 10:22

Arthur Kruse showed in "Badly incomplete normed linear spaces" (Math. Z. 83 (1964) 314--320, DOI:10.1007/BF01111164, also freely available at GDZ) that a Banach space of (Hamel) dimension $\kappa$ exists if and only if $\kappa^{\aleph_0}=\kappa$.

(This has been reproved several times since then, and perhaps there are even older references.)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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