# Absence of a feedback in case of non hermitian hamiltonian

**URL:** https://itensor.discourse.group/t/absence-of-a-feedback-in-case-of-non-hermitian-hamiltonian/2279
**Category:** DMRG and Numerical Methods
**Tags:** dmrg, julia
**Created:** [March 4, 2025, 3:55pm UTC](https://itensor.discourse.group/t/absence-of-a-feedback-in-case-of-non-hermitian-hamiltonian/2279 "2025-03-04T15:55:42Z")
**Posts on this page:** 2
**Page:** 2

<div class="post-metadata">

### Author: ![mtfishman](https://yyz2.discourse-cdn.com/free1/user_avatar/itensor.discourse.group/mtfishman/32/11_2.png) [@mtfishman](https://itensor.discourse.group/u/mtfishman)
#### Post date: [March 7, 2025, 2:57am UTC](https://itensor.discourse.group/t/absence-of-a-feedback-in-case-of-non-hermitian-hamiltonian/2279/21 "2025-03-07T02:57:30Z")

</div>

@miles maybe with a large enough setting of `tol` it normally just builds two or three Krylov vectors anyway, but then maybe builds more for trickier cases like criticality “automatically” (obviously we would have to experiment with that).

@lkdvos that’s an interesting suggestion to default to `ishermitian=false`. My first reaction would be that it may just hide real user issues, like if they input an incorrect Hamiltonian that is accidentally non-Hermitian, but probably they would find out anyway in one way or another (like with complex energies…). Also, users should really be testing their Hamiltonians/MPOs out on small systems with ED.

---

<div class="post-metadata">

### Author: ![VinceNeede](https://yyz2.discourse-cdn.com/free1/user_avatar/itensor.discourse.group/vinceneede/32/731_2.png) [@VinceNeede](https://itensor.discourse.group/u/VinceNeede)
#### Post date: [March 7, 2025, 1:35pm UTC](https://itensor.discourse.group/t/absence-of-a-feedback-in-case-of-non-hermitian-hamiltonian/2279/22 "2025-03-07T13:35:24Z")

</div>

What about using vectors of values for the `eigsolve` parameters, like it is done with `maxdim`, `noise` and `cutoff`? This would allow to set a lower `maxiter` and higher `tol` at the beginning, optimizing first on the global aspect of the `dmrg`, and then set a higher `maxiter` and `krylovdim` and a lower `tol`, obtaining a more fine result on each step of the sweep

[Previous page](https://itensor.discourse.group/t/absence-of-a-feedback-in-case-of-non-hermitian-hamiltonian/2279.md?page=1)