# SVD for non-contiguous bipartitions

**URL:** <https://itensor.discourse.group/t/svd-for-non-contiguous-bipartitions/206>\
**Category:** ITensor C++ Questions\
**Created:** [June 13, 2022, 12:39pm UTC](https://itensor.discourse.group/t/svd-for-non-contiguous-bipartitions/206 "2022-06-13T12:39:01Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![SSR](https://avatars.discourse-cdn.com/v4/letter/s/dec6dc/32.png) [@SSR](https://itensor.discourse.group/u/SSR)\
**Post date:** [June 13, 2022, 12:39pm UTC](https://itensor.discourse.group/t/svd-for-non-contiguous-bipartitions/206/1 "2022-06-13T12:39:01Z")

</div>

Hi,

1. I am looking for SVD for non-contiguous bipartitions of an MPS. However, even before that if I wish to compute entropy for 2:rest bipartition (i.e., 2:1,3,4,5…N), one way to do that is by applying a swap gate that exchanges the sites (1,2)–\> (2,1) and then uses the usual SVD along 1:rest. First I constructed the swapped MPS phi from psi. And then used the SVD for phi along 1:rest. When I am trying to do so, it is showing some problem.  
INPUT:  
psi.position(1);  
auto phi=MPS(N);  
phi.ref(1) = psi(2);  
phi.ref(2) = psi(1);

for(int kkk= 3; kkk \<= N; ++kkk)  
{  
phi.ref(kkk) = psi(kkk);  
}

phi.ref(1) = phi(1) \* delta(dag(sites(2)),sites(1));  
phi.ref(2) = phi(2) \* delta(dag(sites(1)),sites(2));

auto l1 = leftLinkIndex(phi,1);  
auto s1 = siteIndex(phi,1);

auto [UU,S,V] = svd(phi(1),{l1,s1});

OUT PUT:  
From line 568, file indexset.cc

In findIndex: more than one Index found, consider using findInds instead

In findIndex: more than one Index found, consider using findInds instead  
Aborted (core dumped)

Any input will be highly appreciated.

1. Also, do you think finding the Schmidt coefficients for non-contiguous bipartitions SWAP operation is the only way out? I tried first constructing the reduced density matrix and then diagonalizing it but even for moderately large system size, I am having memory problems when I am using diagHermitian.

Thanks