Rigorous Frontiers

Rigorous quantum mechanics offers several distinct kinds of research: proving existence of a dynamics, obtaining a quantitative approximation bound, classifying an operational resource, and certifying a computation. A useful project states which implication is missing, on which class of systems, and in which topology. “Make the argument rigorous” is too broad until the limiting operation and required uniformity are identified.

Helpful background. Dense Domains distinguishes an operator domain from an arbitrary dense test space; Projection-Valued Measures supplies the event-based formulation of sharp observables; Functional Analysis Essentials identifies the estimates and convergence notions used in proofs. The Mathematical Quantum Mechanics volume connects these capabilities. Read the stable theory before treating a failure of one proof technique as an open physical question.

Choose the mathematical obstruction

The following entry points separate finite computations from statements about an infinite system or a controlled physical approximation.

Starting strength	Research route	A concrete first result
Probability and operator inequalities	Measurement theory	A robustness bound for a specified instrument or incompatibility witness
Semigroups and differential equations	Open-system rigour	A domain or energy estimate that survives a stated cutoff
Numerical analysis and spectral theory	Spectral problems	Certified spectral enclosures with explicit error sources
Lattice models and many-body analysis	Many-body spectral gaps	A finite certificate whose hypotheses imply a size-independent bound
Convexity and representation theory	Quantum marginals	An exact compatibility statement or a certified obstruction
Information theory	Rigorous quantum information	A finite-resource converse with matching or quantified achievable performance
Infinite-dimensional analysis	Infinite-dimensional information	An energy-constrained continuity or truncation estimate
PDE and asymptotic analysis	Scattering	A propagation estimate for an explicitly delimited interaction class
Operator algebras and nonlocal games	Operator-algebraic frontiers	A result distinguishing tensor-product, commuting, or finite-dimensional models

The common decision is whether the project is structural or quantitative. A structural theorem might establish existence without a usable convergence rate. A quantitative theorem might improve constants for a restricted model without changing general existence theory. Both can be valuable; their deliverables and comparison literature differ.

Match the proof route to the evidence available

If a finite convex description is available, start with primal constructions and dual obstructions: the two-measurement parent POVM and overlapping singlet marginals give fully solvable calibrations. A numerical optimizer can suggest a theorem, but the independent certificate is the durable deliverable. The 2024 review by Tavakoli and colleagues explains which relaxations certify which correlation models.

If the main obstacle is an infinite limit, select the topology and the uniformity requirement before selecting a solver. Bosonic truncation needs an energy budget; spectral approximation needs information about the full operator; open dynamics needs both a well-defined evolution and a justified approximation. These projects favor functional analysis and numerical error analysis over raw optimization speed.

If an operational quantity is the target, start with a task whose resources can be enumerated. The finite-resource erasure example distinguishes heralded perfect-pair distribution from approximate quantum transmission. The operator-algebraic guide distinguishes a finite-dimensional strategy from an unrestricted upper bound. A mismatch here can invalidate an apparent improvement even when every inequality is correct.

Choose the first route if you can already manipulate positive matrices and want a short feedback loop. Choose an infinite-limit problem when you can work with operator domains or estimates and have supervision in the relevant analysis. Do not choose an undecidability theorem as a starter merely because its headline is compelling: it may require complexity-theoretic machinery largely independent of the physical model that first interested you.

Write the quantifiers before the proposal

A finite-volume calculation produces numbers $\Delta_L$. A thermodynamic gap statement requires an appropriate uniform lower bound as $L$ grows. Exchanging these assertions is a scientific error even when every diagonalization is accurate. Similarly, convergence of each finite-energy input does not automatically give an unconstrained channel norm bound.

Make a proposal begin with one sentence of the form: for every model in class $\mathcal C$, under assumptions $A$, establish conclusion $B$ with a bound uniform in parameters $P$. Follow it with a known theorem whose hypotheses almost apply. The difference between its assumptions and the proposed class is the research task.

For example, the hypothetical finite-size data $\Delta_L(s)=s+1/L$ are positive for every finite $L$ and $s\ge0$. They give a thermodynamic gap at each fixed $s>0$, but no positive uniform lower bound on the whole interval $s\in[0,1]$, or even on $s\in(0,1]$. On $[s_0,1]$ with $s_0>0$, the uniform bound is $s_0$. This elementary example explains why “gapped at every sampled parameter” is weaker than a theorem on a phase region. It is a quantifier diagnostic, not spectral data for a proposed Hamiltonian.

General impossibility results help delimit that task. Cubitt, Pérez-García, and Wolf establish undecidability for a carefully constructed family of spectral-gap problems; this does not prevent proofs for a particular experimentally motivated Hamiltonian. Likewise, the negative resolution of Connes’ embedding problem through MIP*=RE is established background for selecting new operator-algebra questions, rather than an invitation to propose solving the original conjecture.

A graduate project with an early exit

A suitable first project is a certificate audit for a small family of local Hamiltonians. Begin with the Knabe criterion and the commuting Ising check in many-body spectral gaps, then reproduce one published AKLT certificate. The first calculation checks normalization, degeneracy, and boundary conventions without an expensive solver. Use exact arithmetic or interval-certified numerical bounds at the next stage; ordinary floating-point eigenvalues alone are not the certificate.

A workstation and mathematical programming package suffice for the first stage. The deliverable is a short proof document identifying every hypothesis, a reproducible computation, and a table of parameter intervals that the criterion certifies or leaves undecided. Record boundary conditions and numerical enclosure widths.

The failure criterion is explicit: if the certificate depends on an unchecked positivity assumption, loses validity under the chosen boundary condition, or becomes inconclusive throughout the proposed interval, it has not established a gap. Inconclusiveness is a result about the method, not evidence of gaplessness. A research extension changes the local decomposition or strengthens an inequality and proves that the revised certificate covers a previously inaccessible interval. Agree that extension with an adviser before scaling expensive calculations.

For a lighter alternative, reproduce the exact unbiased-qubit compatibility region and its calibration-error band in rigorous measurement theory. A laptop and linear algebra suffice. Deliver the parent effects, an incompatibility witness, a numerical comparison, and the interval where calibration uncertainty prevents either conclusion. Stop if a purported improvement is smaller than the certified calibration or arithmetic error. A research extension needs a motivated uncertainty model or implementation restriction and a stronger bound than the applicable existing work; rotating two qubit axes and plotting the already known boundary is training.

In either path, keep a short research contract containing the model, target theorem, baseline source, affordable computation size, and a negative-result criterion. A failed approach can be useful when it identifies a precise obstruction. It becomes a weak project when every negative result is answered only by requesting a larger computation. Before investing in a new proof, check that the desired statement is neither already covered by a theorem under different notation nor ruled out by a counterexample within the proposed class.

Evidence and literature horizon

Reviewed 11 October 2026. These guides distinguish proved results, algorithms with stated guarantees, and numerical evidence. The linked pages identify their individual literature horizons; recent certificates discussed there include 2026 preprints. No claim of an exhaustive classification of open mathematical problems is intended. Use the direction-selection guide to compare supervision, prerequisites, and the cost of a negative result.

Established describes the cited theorems and the exactly solved examples, with their hypotheses. Active describes bounded methodological directions where a new instance still needs a literature and feasibility check. A numerical pattern or unproved extension is conjectural, even when it fits all tested sizes. Update this guide when a theorem changes the admissible model class, a certifiable algorithm changes the realistic resource cost, or a counterexample invalidates a proposed implication. More citations or larger simulations alone do not establish that such a change occurred.

References

• Cubitt, T. S., Pérez-García, D., and Wolf, M. M. “Undecidability of the spectral gap.” Nature 528, 207–211 (2015). Full proof and versions.
• Ji, Z., Natarajan, A., Vidick, T., Wright, J., and Yuen, H. “MIP*=RE.” Preprint (2020; revised 2022). Theorem and proof.
• Rai, K. S., Kull, I., Emonts, P., Tura, J., Schuch, N., and Baccari, F. “A Hierarchy of Spectral Gap Certificates for Frustration-Free Spin Systems.” Quantum 10, 2065 (2026). Concrete certificate project source.
• Tavakoli, A., Pozas-Kerstjens, A., Brown, P., and Araújo, M. “Semidefinite programming relaxations for quantum correlations.” Reviews of Modern Physics 96, 045006 (2024). Methods and model distinctions.
• van Luijk, L. “Energy-Limited Quantum Dynamics.” Communications in Mathematical Physics 406, 120 (2025). Article.