TeNPy Forum

Hi, and sorry for the slow response; I hope it's still useful, given that I added extensive examples and further functions! Let me first explain why your examples do or don't work: When you write the MPO grid as W_0 = (S^z, S^x); W_1 = (N, N^2)^T or as in the last example, The [tenpyclass=tenpy.netw...

You can also do SpIdSm = tenpy.networks.site.kron(s.Sp, s.Id, S.Sm, group=False) The tenpy.networks.site.kron function with n input onsite operators generates an operator acting on n consecutive sites at once, with n input and n output legs, e.g for n=3, p0*, p1*, p2*, p0, p1, p2 . The expectation_v...

Since you fully re-initialize the TwoSiteDMRGEngine in the example you sent without passing the environment, you actually don't reuse the environment here. You define a new (random) product_state , but don't use it to initialize psi itself, I assume you also do that within the loop? And what exactly...

Really, the proper functions is to adjust the *correlation_function methods of the MPS to the MPSEnvironment, see the new 173 . If you really just need the nearest-neighbor correlations, the expectation_value actually supports this: import tenpy s = tenpy.networks.site.SpinHalfSite() psi = tenpy.net...

The first example adds terms b^\dagger_{x,0} b_{x,1} n_{x+1,1} + b^\dagger_{x,1} b_{x,0} n_{x+1,0} , which is not hermitian due to the switch in the n index. DMRG (or more precisely the Lanczos called by DMRG) always assumes a hermitian Hamiltonian! The "poorly conditioned Lanczos" is like...

If you expect translation invariance along the cylinder direction (and you Hamiltonian itself is also translation invariant in x-direction), Lx=1 is certainly a natural choice. Using Lx=1 does not hinder the calculation of correlation lengths - in general, if you have an infinite MPS, it just implie...

TEBD breaks translation invariance by performing the Trotterization of the evolution operator onto even/odd bonds. The direct way to reduce this effect is to do more, but smaller timesteps. You could also try to do higher-order evolution, but of course that doesn't make your algorithm higher-order i...

It's correct behaviour that the state doesn't change: You start from the fully polarized all-up state, which is an eigenstate of the Hamiltonian! Try to start from another initial state, e.g. the Neel state: product_state = ["up", "down"] psi = MPS.from_product_state(M.lat.mps_si...

Actually, I have a few more questions for you about lanczos in TeNPy. I notice that in TeNPy you set the default maximum number of iterations for lanczos (i.e. the attribute N_max) to be 20. However, when I tried to realize DMRG and lanczos algorithm on my own, I found that for a Hermitian matrix w...

What bond dimensions did you try for the timing benchmarks? I'd expect it to become relevant only around chi >~500 or so, maybe even just with chi on the order of a thousands.

It's documented in the conda-forge docs here . I think that's actually fine and properly linked against mkl. I've checked, and when I start from a fresh conda environment defined by TeNPy's environment.yml, I also get the # All requested packages already installed. Indeed, when I checked the install...

Of course, both approaches can work and should give similar results eventually; but I'd usually recommend not grouping sites if you can manage to write the model without grouping. Let me elaborate: When using the tenpy.networks.site.GroupedSite , you will double your local Hilbert space dimension d ...

I'm confused - what state do you start with? Some ground state, or other pure state, represented as an MPS? What do you want to calculate afterwards? You can use TeNPy to get the single-site reduced density matrix, and then apply the dephasing channel, but you can not apply it on the whole MPS: it's...

With the random unitary evolution, you get an ensemble of Haar-distributed states, i.e. each normalized state has the same probability. Since there *are* states with less than maximal, but no states with more-than-maximal entropy, it is clear that the average entropy in the Haar distribution is less...

Hi Umberto, there is a tenpy.models.lattice.TrivialLattice , which is exactly the approach 1) you described - it just takes a list of sites, and defines them as a single, big unit cell. Internally in TeNPy, this is used if you group sites together, since you then no longer have the original lattice ...

I don't see the issue here, I checked that it works for an even number of couplings , where the couplings are not incommensurable. For nearest neighbor couplings A_i B_{i+1}, this means it works for (finite, odd # sites) or (infinite, even # sites), as you can easily check with the example below. No...

Hi,
yes, that is correct. It works if the number of couplings is even (ie. odd number of sites for finite systems, even number of sites for infinite MPS) , otherwise it complains about incommensurate arrays for the expansion.

Yes, you can do better. The explicit loops you have make this scale as O(L^3 \chi^3) , since each call to psi.expectation_value_term is O(|j+1 - i| \chi^3) . You can get this down to O(L^2 \chi^3) by using the term_correlation_function_right or ..._left, with term_L = [('Cd', 0), ('Cd', 1)], term_R=...

TeNPy tries to preserve local, diagonal symmetries, in your case the Z3 symmetry (\sum_i Q_i \mod 3) where Q_i = diag(0, 1, 2) are the charge values you define. However, the Hamiltonian does not preserve these symmetries, since the \Gamma^{(1,2)}_i don't commute with this symmetry. Instead, you need...

1) These additional sweeps are meant to (i) bring the MPS into canonical form and (ii) converge the environments. Effectively, it's mostly SVDs and environment updates, and usually cheap compared to the Lanczos optimization, so it should be "cheap" to converge things a bit better, even if ...

The export OMP_NUM_THREADS=1 is a bash command. You should have a slurm/sge/bash script which you submit to the cluster, include this line there. For example, if you use the setup of the TeNPy project repository , you can explicitly add the corresponding line in the file cluster_templates/slurm.sh (...

Indeed, the get_probability is not working; it even gives you wrong dimensions of the array when you go beyond 2 sites. Using the onsite reduced density matrix is as pointed out in the post you linked to is indeed one way. Another and maybe even clearer solution would be to define projection operato...

I think you're on the right track. I presume your PhtantasticLattice is derived from the base lattice class. What you should do is override the ordering method, similar as is done in the the tenpy.models.lattice.Honeycomb lattice - take a look at the source there. You want the structure def ordering...

Yes, the main idea is that you *never* have the Schmidt states as whole, but only as (partial) MPS, recursivley adding tensors to get the Schmidt states on the next bond; see also Figure 4 c) of our paper 1805.00055 . Indeed, the formula you wrote gives the Schmidt states |a_l> in the basis of |a_{l...

You're referring to the update_LP_RP defined in get_sweep_schedule for n=2 , i.e. a two-site update. As you can see from the i0s, move_right variables, we do 2L updates in total, each updating 2 sites, in the order (0, 1), ..., (L-2, L-1), (L-1, L), (L, L+1), (L-1, L), ..., (1, 2) . Note that this i...