Let $\mathcal{H}_1$ and $\mathcal{H}_2$ be Hilbert spaces with inner products $\langle \cdot, \cdot \rangle_1$ and $\langle \cdot, \cdot \rangle_2$, respectively. As described in many references (and Wikipedia as well), one can define a corresponding pre-Hilbert tensor product space by taking the (algebraic) tensor product $\mathcal{H}_1 \times \mathcal{H}_2$ equipped with the inner product $$ \langle x_1 \otimes x_2, y_1 \otimes y_2 \rangle \triangleq \langle x_1, y_1 \rangle_1 \langle x_2, y_2 \rangle_2 \quad \forall x_1,y_1 \in \mathcal{H}_1, \; x_2,y_2 \in \mathcal{H}_1, $$ which extends bilinearly to finite linear combinations. The completion of this construction then yields a Hilbert space, which is a topological tensor product denoted by $\mathcal{H}_1 \hat{\otimes} \mathcal{H}_2$.

Now, my question regards the computation of the inner product for a given pair of tensors from $\mathcal{H}_1 \hat{\otimes} \mathcal{H}_2$. Let two arbitrary tensors in that space be $$ x = \sum_{n=1}^\infty u_n \otimes v_n, \quad \text{and} \quad y = \sum_{n=1}^\infty w_n \otimes z_n. $$ Now, the inner product was defined so as to extend bilinearly for finite linear combinations. Hence, how can one compute the inner product $\langle x, y \rangle$? Does it make sense to write $$ \langle x, y \rangle = \langle \sum_{n=1}^\infty u_n \otimes v_n, \sum_{m=1}^\infty w_m \otimes z_m \rangle = \sum_{n=1}^\infty \sum_{m=1}^\infty \langle u_n \otimes v_n, w_m \otimes z_m \rangle, $$ as if bilinearity holds also for infinite linear combinations?

  • $\begingroup$ Yes, of course, these things are constructed so that this kind of linear combinations are continuous operations. All necessary continuous estimates are supplied by Cauchy-Schwarz. $\endgroup$ Jun 5, 2019 at 21:19
  • $\begingroup$ Thank you @GiuseppeNegro for your comment. But could you please elaborate what do you mean by "all necessary continuous estimates"? $\endgroup$ Jun 8, 2019 at 17:38
  • $\begingroup$ Anyway, let me add that there is nothing deep here, do not fear these things. $\endgroup$ Jun 9, 2019 at 21:35
  • $\begingroup$ Okay, thanks in advance @GiuseppeNegro. $\endgroup$ Jun 13, 2019 at 9:42

1 Answer 1


$\newcommand{\Hcal}{\mathcal{H}}$Your question boils down to the following.

If $(h_1), (h_2)$ are sequences in $\Hcal_1, \Hcal_2$ respectively, is it true that $$\tag{1} h_1(n)\to h_1 \ \text{ and }\ h_2(n)\to h_2\ \quad \Longrightarrow\quad h_1(n)\otimes h_2(n)\to h_1\otimes h_2?$$

Indeed, once (1) is proven, you can just apply it to the sequences $$ h_1(n):=\sum_{k=1}^n u_k, \qquad h_2(n):=\sum_{k=1}^n v_k,$$ which you defined in the main text.

The proof of (1) is an immediate consequence of the obvious inequality $$ \lVert h_1\otimes h_2\rVert\le \lVert h_1\rVert\lVert h_2\rVert$$ (actually, this is even an identity), and of the bilinearity of $\otimes$.

  • $\begingroup$ Sorry, I still do not see exactly how this suffices. If h_1(n) and h_2(n) are defined as the partial sums like you did, then h_1(n) \otimes h_2(n) is going to be a sum of n^2 tensor products (with all cross-products), instead of a sum of tensor products u_k \otimes v_k (with same k) as I had written. Anyway, I wonder whether the answer to my question is yes simply because the scalar product defined for finite linear combinations of elementary tensors is continuous? $\endgroup$ Feb 5, 2020 at 16:52

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.