2
$\begingroup$

I understand that one method to derive an MPS representation of a quantum state involves applying the Schmidt decomposition $ N−1$ times. While I'm familiar with the diagrammatic notation, I wanted to understand the math behind by working through a simple tripartite state to get its MPS form.

Given a general tripartite state:

$$|\psi\rangle = \sum_{i,j,k} \psi_{i,j,k} |i,j,k\rangle.$$

We follow the steps:

Step 1: Partition the State

Partition the state into systems labeled by $\color{blue}{i}$ and $\color{red}{jk}$ yielding: $$|\psi\rangle = \sum_{i,j,k} \psi_{\color{blue}{i},\color{red}{jk}} |i,j,k\rangle.$$

Step 2: First Schmidt Decomposition

Now, we'll apply the Schmidt decomposition to this partition:

$$|\psi\rangle = \sum_{\alpha}\sum_{i,j,k} U_{i,\alpha} \Lambda_{\alpha,\alpha} V^*_{jk,\alpha} |i,j,k\rangle = \sum_{\alpha} \Lambda_{\alpha,\alpha} \color{blue}{\left(\sum_{i} U_{i,\alpha} |i\rangle \right)}\otimes \color{red}{\left(\sum_{j,k}V^*_{jk,\alpha} |j,k\rangle \right)}.$$

Question 1. I don't understand the term red term. Specifically, the matrix element $V_{jk,\alpha}$. Does this suggest that for each column $\alpha$, there is a corresponding $j \times k$ matrix? My question arises because the next step involves applying a Schmidt decomposition to the term highlighted in red. While I can rewrite this term as $V^{(\alpha)}_{j,k}$, I'm uncertain about its mathematical implications. What should I do? I read about "re-shaping the matrix", but I don't understand either what does that mean and how the dependence of alpha is carried by doing so.

$\endgroup$
6
  • $\begingroup$ The left part of the colorful equation is incorrect; the subscript of V is not $i$. -- Do you understand the "First Schmidt decomposition" step if there would be only two systems, i.e., $i$ and $j$? $\endgroup$ Commented Oct 27, 2023 at 16:19
  • $\begingroup$ Having said that: The question as it is does not work well for this site -- you should try to ask a precise question. (I understand this is difficult, since this requires understanding what you do.) $\endgroup$ Commented Oct 27, 2023 at 16:20
  • $\begingroup$ @NorbertSchuch concerning your first comment, the subscript $i$ on $V$ was a typo. Regarding your second comment, I see that it does not work well for this site, but I don't know how to "split" the state $psi$, such that I can perform the two Schmidt decomposition - I'm not sure if it's clear what I said. $\endgroup$
    – Alex
    Commented Oct 27, 2023 at 17:23
  • $\begingroup$ Do you understand how step 1 works (or rather, the first Schmidt decomposition following the "partitioning" you describe in step 1)? Then, it should be clear. If not, you should focus on that part. (Or maybe you do, but then I don't understand why you start talking about the other partition.) $\endgroup$ Commented Oct 27, 2023 at 17:35
  • 1
    $\begingroup$ But if you understand that you can treat $jk$ as a single (larger) index $j'$, and do an SVD of $i$ vs. $j'\equiv jk$, everything is clear. You just have to change what you treat as the larger index. E.g., in the next step, one index is $\alpha j$, and the other is $k$. $\endgroup$ Commented Oct 27, 2023 at 17:50

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.