SciPost Submission Page
Many bounded versions of undecidable problems are NPhard
by Andreas Klingler, Mirte van der Eyden, Sebastian Stengele, Tobias Reinhart, Gemma De las Cuevas
This is not the latest submitted version.
This Submission thread is now published as
Submission summary
Authors (as registered SciPost users):  Andreas Klingler · Sebastian Stengele 
Submission information  

Preprint Link:  https://arxiv.org/abs/2211.13532v2 (pdf) 
Date submitted:  20221209 09:47 
Submitted by:  Klingler, Andreas 
Submitted to:  SciPost Physics 
Ontological classification  

Academic field:  Physics 
Specialties: 

Approaches:  Theoretical, Computational 
Abstract
Several physically inspired problems have been proven undecidable; examples are the spectral gap problem and the membership problem for quantum correlations. Most of these results rely on reductions from a handful of undecidable problems, such as the halting problem, the tiling problem, the Post correspondence problem or the matrix mortality problem. All these problems have a common property: they have an NPhard bounded version. This work establishes a relation between undecidable unbounded problems and their bounded NPhard versions. Specifically, we show that NPhardness of a bounded version follows easily from the reduction of the unbounded problems. This leads to new and simpler proofs of the NPhardness of bounded version of the Post correspondence problem, the matrix mortality problem, the positivity of matrix product operators, the reachability problem, the tiling problem, and the ground state energy problem. This work sheds light on the intractability of problems in theoretical physics and on the computational consequences of bounding a parameter.
Current status:
Reports on this Submission
Anonymous Report 2 on 2023214 (Invited Report)
 Cite as: Anonymous, Report on arXiv:2211.13532v2, delivered 20230214, doi: 10.21468/SciPost.Report.6742
Strengths
1Very good overview for some NPhard problems and their corresponding undecidable formulations , which were analyzed in (quantum) information science literature.
2The authors show a rigorous relationship between the undecidable problems and their "bounded" versions.
3Generally well written with good figures and a good, succinct introduction for the considered problems.
Weaknesses
1 The overall ordering of the paper is sometimes confusing. Concepts are discussed, but defined only later.
2 The authors could spend more time explaining how their main result, theorem 2, fits within the broader complexity theory literature.
Report
The paper is well written and I could not spot any technical errors. For physicists wanting to show complexity theoretic hardness for their respective problems, this paper provides a good summary of different proof techniques.
The main theoretical contribution is showing that a reduction using NHalt (NHaltAll) or PCP also gives NP (coNP)hardness for the bounded version when the requirements of theorem 2 are met. While this is not a surprising result for complexity theorists, it might proof to be a useful tool in future hardness proofs.
Overall , I think if the authors perform some minor changes, the work should be published SciPost Physics.
Requested changes
It is not always clear which aspects and ideas are the authors own contribution, and which originate from literature. In particular, have there already been studies on "bounded" languages (as in def. 1) in literature?
In physics, the approximate version of problems (eg. $\epsilon$closeness to positive MPOs in some norm) might be more relevant than the undecidability for infinite system size. Could the authors comment, if their work can also be used to show approximate NPhardness in the finite case?
Small comments:
2.A:
x refers to two different concepts: an element in a language and an initial state of a Turing machine. Maybe use different symbols?
 BHalt seems to be the natural bounded version of Halt. (which is Pcomplete). I think it would improve readibility, if the authors made clear that NHalt is used because it is essential for later reductions, as BHalt would not work.
2.B:
Eq. 1: does n need to be an exact polynomial, or just upperbounded by one?
3.:
The final paragraph can only be understood after having read the paper. Even when one is familiar with the problems, the (coNP) formulation of the tiling problem was not mentioned thus far, making the NHalt vs NHaltAll discussion hard to follow.
4A:
For PCP, n is the number of words (effectively number of dominos)..In BPCP it is the maximal size of the matching. Maybe use k as the number of input words (dominos) for both?
Spelling: "It is to be NPcomplete" > "Is known to be NPcomplete"
4C:
Why are the MPO are chosen to describe diagonal operators? This does not seem to be the canonical definition.
4G:
Could the authors elaborate, why NHaltAll needed to be used for Wang tiling problems in particular?
Anonymous Report 1 on 202318 (Invited Report)
 Cite as: Anonymous, Report on arXiv:2211.13532v2, delivered 20230108, doi: 10.21468/SciPost.Report.6477
Strengths
1. The paper is very well written, I think it contains sufficient background on computational complexity for readers with a physics background to understand the results and their significance.
2. The authors shed light on the relationship between bounded and unbounded problems, and in particular on the relationship between undecidability and hardness. This is an interesting area which to the best of my knowledge has not been rigorously explored before.
Weaknesses
1. The paper in its entirety contains a very good overview of computational complexity for the non expert, and thorough descriptions of all the computational problems used. However sometimes computational problems / computational complexity terms are referenced prior to their definition. This may put off some readers (a possible solution would be to update Appendix A to also include descriptions of all computational problems used in the manuscript, and reference it in the introduction for any readers who aren't familiar with complexity theory).
2. There isn't much discussion of why the distinction between undecidability and hardness is interesting from a physical point of view  I think this would be nice to include (particularly since the submission is being considered for SciPost Physics).
Report
This paper tackles the question of whether the observation that many undecidable problems have NPhard bounded versions can be rigorously understood. The authors derive a method that leverages the standard techniques for proving undecidability in order to derive hardness results for bounded versions of certain types of undecidable problems.
The paper begins by explaining what a bound version of a problem is, and deriving the relationship between reductions between unbound problems and reductions between bound problems. It then goes on to introduce two bound versions of the halting problem, and finally it uses reductions from the halting problem to a number of other problems to demonstrate NP / coNP hardness of a number of bound problems that are known to be undecidable in the unbounded case.
The question that is tackled by these authors is an extremely interesting one at the intersection of physics and computational complexity. I believe that the approach taken by the authors to tackling it is elegant, and the results in the paper are well motivated and insightful. As detailed in section 5 there are a number of other research questions which could now be tackled using this paper as a starting point. I believe this paper meets the criteria for publication in SciPost Physics.
Given the very short turnaround time for the review I have not been able to check all the proofs rigorously, but to the best of my understanding the proofs in the paper are correct.
Requested changes
1. In theorem 2 it's not completely clear to me why p has to be strictly increasing as opposed to just nondecreasing. Could the authors clarify this point.
2. In the proof of theorem 4 it might be helpful to note in a footnote that a deterministic Turing machine is a special case of a nondeterministic Turing machine to help explain the reduction from Halt to NHalt.
3. Including a bit more detail about the complexity class coNP in appendix A could help readers when they come to the proof of theorem 8 (in particular including a sentence about why the definition of coNP implies acceptance on all computational paths).
4. In section 4A the abbreviation MPO is used for the first time but not explained until section 4C.
5. Page 4, bottom of column 1, where it says "it is to be NPcomplete" I think it should be "it is known to be NPcomplete"
6. Page 9, bottom of column one, where it says "of the latter use the dimension" should this be "of the latter uses the dimension"
7. In section 5 could the others include some discussion of whether there is any hope of understanding an implication in the opposite direction (maybe there are obvious counterexamples showing not  if so could the authors reference them).
8&9. The points from the weaknesses section could also be tackled.