Pith. Sign in

REVIEW

Collective Bootstrap for Matrix Quantum Mechanics

PaperPDF from arXiv

data files

ReviewThe objections and verdict, as JSON Paper breakdownThe authors' claims, conditions and measurements, as JSON Signed receiptProof this review has not been altered
X Bluesky LinkedIn Reddit Hacker News
Xianlong Liu

High Energy Physics - Theory · Submitted 28 Oct 2026 · 45 references

Pith's verdict: The lower-bound construction is sound and honestly labeled, but the three-matrix 'agreement' is produced by a cage constraint whose physical input is assumed, not derived.

T0 · AI review

How this review was made: An AI referee (deepseek/deepseek-v4-flash) read the paper on October 31, 2026 and answered a fixed list of questions about it. T0 is the starting level every paper gets. The levels describe how deeply Pith has checked the paper, never the authors.

  1. T0 AI review An AI read the paper and answered a fixed list of questions.
  2. T1 Audit Automatic checks that the references and files the paper relies on exist and match.
  3. T2 Compute The paper's code or calculations re-run with the same results.
  4. T3 Formal statement The claims written out in a formal math language a computer can read.
  5. T4 Proof Those claims proved, and the proof checked by computer.

How Pith checks papers →

· 2 major objections · 3 minor · Ask about this paper · The compiled record · Read the full review →
Challenge this review Comment (0) Read on arXiv ↗
Cite ▾
arXiv ID2610.38716 (version 1)
Pith IDMWDK2RWO, Pith's permanent record of this paper and its review
CitationXianlong Liu. "Collective Bootstrap for Matrix Quantum Mechanics." arXiv:2610.38716 (2026). https://pith.science/paper/MWDK2RWO
Cite this reviewBibTeX for Pith's review of this paper
Claim paper

Why Pith says this

Highlight any text to comment on it.

Referee report

How Pith checks papers → · Full audit of this paper →

The paper

The paper claims

Loop-space bootstrap certifies a lower bound on the large-N ground state at every truncation, matching one-matrix energies to six digits.

The reading

This paper aims to make the planar (large-$N$) ground state of matrix quantum mechanics computable with certified error control. It formulates the problem directly in the space of single-trace coordinate moments and recasts the collective energy minimization as a semidefinite program (a convex optimization whose constraints are matrix-positivity conditions), claiming that every truncation yields a rigorous lower bound on the exact ground state energy divided by $N^2$. A sequential rank-minimization search then estimates the energy and loop expectation values by driving the lifted variables toward planar factorization, the property $\langle\phi(C)\phi(C')\rangle = \langle\phi(C)\rangle\langle\phi(C')\rangle$ that characterizes a classical large-$N$ configuration. The reported energies agree with the exact free-fermion solution to six digits in the one-matrix model and with master-field optimization to a few parts in a thousand in two- and three-matrix Yang–Mills models. Because momentum-containing words never appear as variables, longer word-length truncations become affordable, which the paper argues opens a route to the bosonic sectors of the BFSS and BMN models.

What carries the argument

Four objects carry the argument. The loop space of single-trace coordinate moments, $\phi(C) = N^{-\ell(C)/2-1}\,\mathrm{tr}(C)$, replaces the phase-space operator basis so that momentum words never appear as independent variables. Loop joining, $\Omega$, and loop splitting, $\omega$, from the collective field formulation organize the kinetic constraints. The virial theorem together with a Schur complement replaces the inverse joining matrix $\Omega^{-1}$ by the positive-semidefinite block condition $S = \begin{pmatrix} \Omega & \omega^\dagger \\ \omega^T & \mu \end{pmatrix} \succeq 0$, making the energy linear in the loops. Quadratic products of loops in the splitting term are lifted to auxiliary variables $Y(C,C') = \phi(C)^\dagger\phi(C')$ with $Y \succeq 0$, and a physical configuration is recovered when $\mathrm{rank}(Y)=1$, which is the planar factorization condition. The relaxed program with the rank condition dropped produces the certified lower bound $E_R(\ell_{\max})$, while the log-det heuristic and energy bisection search for the rank-one point that yields the estimate $E_g$.

Extended reading notes

Core claim

The central claim is inequality (3.23): at every truncation $\ell_{\max}$, the optimum $E_R(\ell_{\max})$ of the relaxed semidefinite program and the optimum $E_P(\ell_{\max})$ of the rank-constrained problem satisfy $E_R(\ell_{\max}) \le E_P(\ell_{\max}) \le E_0$, where $E_0$ is the exact large-$N$ ground state energy divided by $N^2$. Each truncated computation is therefore a genuine lower bound, and both optima are nondecreasing in $\ell_{\max}$, so the bound tightens monotonically toward $E_0$. The numerical estimate $E_g$ is produced by bisecting in the energy and using a log-det rank-minimization heuristic to test whether the feasible region contains a point whose lifted matrix $Y$ has rank one, which is exactly the planar factorization condition; the paper acknowledges that this search is a local heuristic, although every accepted energy level is certified by an approximate rank-one point. On this basis the reported relative energy deviations run from below $10^{-6}$ in the one-matrix model to a few times $10^{-3}$ in the Yang–Mills models, with the three-matrix commutator angle $\theta \approx 0.7895$ against a master-field value of $0.791(2)$. The paper emphasizes that the rigorous bound relies only on constraints satisfied by the exact solution, while the three-matrix estimates further depend on artificial cage constraints stated in Section 7.

Load-bearing premise

The load-bearing premise behind the three-matrix estimates is that the true eigenvalue distribution of $R=\sum_a M_a^2$ is supported below the artificial cage bound $S=3$; the paper acknowledges the cages are artificial and that the bound (3.23) remains valid only if $S$ exceeds the true spectral edge, a condition the paper does not independently verify, and if the edge lies above $S$ the cage excludes the physical solution and the reported agreement with master-field values does not test the method.

Editorial extensions

If this is right

  • Every truncated computation returns a certified lower bound on the planar ground state energy, and the bound is nondecreasing in the truncation length, so systematic improvement is guaranteed rather than assumed.
  • Because only coordinate moments appear, the single-trace variable count grows like $d^\ell$ instead of $(2d)^\ell$, making longer word lengths and more matrices affordable than in phase-space bootstraps.
  • The decoupled two-matrix benchmark reproduces the free-independence factorization rules of the large-$N$ limit, confirming that the rank-one search finds the correct factorization structure.
  • At a factorized solution the collective Hamiltonian determines the $O(1)$ fluctuation spectrum around the large-$N$ background, so the same computation can yield excitation energies alongside the ground state.
  • The framework is aimed at the bosonic sectors of the BFSS and BMN nine-matrix models, where the coordinate-only formulation is expected to control the growth of variables.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The three-matrix numbers are only as strong as the cage assumption: without $\langle\phi(R^{n+1})\rangle < 3\langle\phi(R^n)\rangle$ the search yields $\langle\phi^2\rangle \approx 31$ and $\langle\phi^3\rangle \approx 30$, an order of magnitude above master-field values, so the 0.022% energy agreement partly measures the cage's influence; computing the spectral edge of $R$ at the master-field sol
  • The accuracy asymmetry visible in the tables, energies at $10^{-6}$ to $10^{-3}$ relative error while long loops sit at the percent level, suggests that adding the higher dynamical equations would improve loop values before the energy; this is the paper's own suggested next step, and my reading is that the energy accuracy is carried almost entirely by the virial constraint.
  • The certified lower bound $E_R$ and the heuristic estimate $E_g$ are logically independent outputs: a failure of the rank-one search raises $E_g$ but never breaks the bound, so the two numbers should carry different trust levels in future applications.
  • The quadratic growth of lifted variables with the number of short loops is the practical bottleneck toward nine-matrix models; the symmetry reductions used here suggest that flavor and rotation symmetry, not loop counting alone, will decide whether BFSS is reachable.
Original abstract

We develop a collective bootstrap framework for large $N$ matrix quantum mechanics formulated directly in the loop space of single trace coordinate moments. Loop joining and splitting operations organize the kinetic constraints systematically and avoid treating momentum containing moments as independent variables. Employing the virial theorem, a Schur complement, and a relaxation of quadratic products of loops, we recast the nonlinear constrained optimization of the collective energy into a semidefinite program, which yields a lower bound on the planar ground state energy at every truncation. A sequential semidefinite optimization heuristic then searches for configurations satisfying planar factorization and estimates the ground state energy and loop expectation values. Applications to one-, two-, and three-matrix models exhibit close agreement with analytic solutions and master field optimization. By working exclusively with coordinate moments, the framework permits truncations at higher word length and offers a promising approach to models with more matrices, including the bosonic sectors of the BFSS and BMN models.

The compiled record

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-10-30: 1 request, 4,964 input tokens.

What the abstract itself states, as probabilities: claims first 0.05 · states limitation 0.19 · human participants 0.01 · controlled comparison 0.11 · randomized assignment 0.01 · states quantitative result 0.24 · compares with prior results 0.93 · mentions code or data availability 0.02.

The abstract, sentence by sentence

arXiv listing abstract v1. Results and the main claim are underlined; .

We develop a collective bootstrap framework for large $N$ matrix quantum mechanics formulated directly in the loop space of single trace coordinate moments. Loop joining and splitting operations organize the kinetic constraints systematically and avoid treating momentum containing moments as independent variables. Employing the virial theorem, a Schur complement, and a relaxation of quadratic products of loops, we recast the nonlinear constrained optimization of the collective energy into a semidefinite program, which yields a lower bound on the planar ground state energy at every truncation. A sequential semidefinite optimization heuristic then searches for configurations satisfying planar factorization and estimates the ground state energy and loop expectation values. Applications to one-, two-, and three-matrix models exhibit close agreement with analytic solutions and master field optimization. By working exclusively with coordinate moments, the framework permits truncations at higher word length and offers a promising approach to models with more matrices, including the bosonic sectors of the BFSS and BMN models.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#sentencerolethe authors' own workstates the main conclusionstrength
s0 We develop a collective bootstrap framework for large $N$ matrix quantum mechanics formulated directly in the loop space of single trace coordinate moments. contribution 0.99 0.96 0.12 asserted 1.00
s1 Loop joining and splitting operations organize the kinetic constraints systematically and avoid treating momentum containing moments as independent variables. method 0.98 0.83 0.05 asserted 1.00
s2 Employing the virial theorem, a Schur complement, and a relaxation of quadratic products of loops, we recast the nonlinear constrained optimization of the collective energy into a semidefinite program, which yields a lower bound on the planar ground state energy at every truncation. method 0.96 0.95 0.26 asserted 1.00
s3 A sequential semidefinite optimization heuristic then searches for configurations satisfying planar factorization and estimates the ground state energy and loop expectation values. method 0.99 0.91 0.09 asserted 1.00
s4 Applications to one-, two-, and three-matrix models exhibit close agreement with analytic solutions and master field optimization. result 1.00 0.93 0.28 asserted 0.98
s5 By working exclusively with coordinate moments, the framework permits truncations at higher word length and offers a promising approach to models with more matrices, including the bosonic sectors of the BFSS and BMN models. interpretation 0.63 0.71 0.29 hedged possibility 0.71

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from simulation or computation 0.91.

What was examined, and what was not

sources read
  • arXiv listing abstract v1
not examined
  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers
levels run
  • P abstract
not run
  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

Figures
Figure 1. Illustration of the bisection sequential SDP. The relaxed feasible region
Figure 2. Optimization history in the (ϕ2, ϕ4) plane of the bisection sequential SDP for the one-matrix model at g = 4 with truncation ℓmax = 20. ⟨𝜙2⟩ 0.50 0.75 1.00 1.25 ⟨𝜙 4⟩ 0.3 0.6 0.9 1.2 1.5 g = gc ⟨𝜙2⟩ 0.50 0.75 1.00 1.25 ⟨𝜙 4⟩ 0.3 0.6 0.9 1.2 1.5 g = 0.0 ⟨𝜙2⟩ 0.50 0.75 1.00 1.25 ⟨𝜙 4⟩ 0.3 0.6 0.9 1.2 1.5 g = 0.1 ⟨𝜙2⟩ 0.00 0.25 0.50 0.75 1.00 ⟨𝜙 4⟩ 0.0 0.2 0.4 0.6 0.8 g = 4.0 ⟨𝜙2⟩ 0.00 0.25 0.50 0.75 1.00 ⟨𝜙 4⟩ 0.0 0.2 0.4 0.6 0.8 g = 10.0 ⟨𝜙2⟩ 0.00 0.25 0.50 0.75 1.00 ⟨𝜙 4⟩ 0.0 0.2 0.4 0.6 0.8 g = 100.0 ℓmax = 4 ℓmax = 8 ℓmax = 10 ℓmax = 14 ℓmax = 16 Ground State
Figure 3. Feasible regions in the projected (ϕ2, ϕ4) subspace of the one-matrix model for various g and different truncations, obtained with the ray search of Appendix C. The exact ground states are marked in red stars. and the collective Hamiltonian is Hcol = 1 2N2 ΠmΩmnΠn + N 2Vcol(ϕ), (4.5) Vcol(ϕ) = 1 8 ωm(Ω−1 )mnωn + 1 2 ϕ2 + g 4 ϕ4 . (4.6) 11
Figure 4. Ground state energy of the one-matrix model versus the coupling
Figure 5. Loop expectation values of the one-matrix model versus the coupling
Figure 6. Diagrammatic illustration of the factorization rule
Figure 7. Ground state energy of the decoupled two-matrix model (
Figure 8. Loops ϕ(Mn 1 ) of the decoupled two-matrix model (k = 0) at truncation ℓmax = 8. Markers are bootstrap results and lines are the exact solution. g 0 2 4 6 8 10 ⟨𝜙(C)⟩ 10−2.0 10−1.5 10−1.0 10−0.5 100.0 [1122] [111122] [11111122] [11112222] g 0 5 10 ⟨𝜙(C)⟩ 10−2.0 10−1.5 10−1.0 10−0.5 100.0 [112112] [11112112] [11122122] [11221122]
Figure 9. Mixed loops of the types ϕ(C1C2) (left) and ϕ(C1C2C ′ 1C ′ 2 ) (right) for the decoupled two-matrix model (k = 0) at ℓmax = 8. Markers are bootstrap results and lines are the exact solution. diagonalizes the Hamiltonian, H = tr 1 2 P 2 + + 1 2 P 2 − + k+ 2 M2 + + k− 2 M2 −  , (5.10) where k± = 1 ± k. The ground state energy is E0 = 1 2 q k+ + q k−  , (5.11) and the lowest loops are ⟨ϕ(M2 1 )⟩ = ⟨ϕ(M2 2 )⟩ = 1 4 q 1/k+ + q 1/k−  , (5.12) ⟨ϕ(M1M2)⟩ = 1 4 q 1/k+ − q 1/k−  . (5.13) Hence k = ±1 are critical values beyond which the
Figure 10. Ground state energy of the coupled two-matrix model versus
Figure 11. Loops of the types ϕ(C1C2) (left) and ϕ(C1C2C ′ 1C ′ 2 ) (right) for the coupled two-matrix model at k = 1 and ℓmax = 8. Markers and dashed lines are bootstrap results; solid lines are master field results with ℓmax = 6 and N = 16. 6 Two-matrix Yang–Mills model 6.1 Symmetries We next consider two Hermitian matrices with a Yang–Mills type interaction, studied with phase space bootstrap in [27] and with master field optimization in [18], H = tr 1 2 P 2 1 + 1 2 P 2 2 + m2 2 M2 1 + m2 2 M2 2 − λ N [M1, M2] 2 ! , (6.1) where λ ≡ g 2 YM is the ’t Hooft coupli
Figure 12. Fitted scaling exponents pC of the loops for the massless two-matrix Yang–Mills model, compared with the exact value ℓ(C)/3. 20
Figure 13. Ground state energy of the massless two-matrix Yang–Mills model versus
Figure 14. Loop expectation values of the massless two-matrix Yang–Mills model. Markers are
Figure 15. Ground state energy of the massive two-matrix Yang–Mills model versus
Figure 16. Loop expectation values of the two-matrix Yang–Mills model at
Figure 17. Feasible region search in the (ϕ1, E) plane for various cages S. The inset plot exhibits a refinement of the boundary lines separating the feasible and infeasible regions. To prevent the high-lying loops from becoming too large in the bootstrap process, we also add the cages ⟨ϕ(Rn+1)⟩ < S⟨ϕ(Rn )⟩ for a positive number S. These artificial constraints are not implied by the dynamics. They are equivalent to assuming that the large N eigenvalue distribution of the operator R in the normalization of the loops is supported below S, and they are satisfied by the true ground state only if…
Figure 18. Ground state energy of the three-matrix Yang–Mills model versus
Figure 19. The lowest loops of the three-matrix Yang–Mills model versus
Figure 20. Collective bootstrap of one-dimensional vector model.
Figure 21. Physically allowed regions for the one-dimensional vector model for fixed
Figure 22. The distribution function u(x, p) in phase space. Left: g = gc. Right: g = 4. where {·, ·} denotes the Poisson bracket. For O = tr(MnP) we have O(x, p) = x np, and therefore 0 = Z dx dp 2π u(x, p)[nxn−1 p 2 − x n (x + gx3 )] . (B.4) The second term in the bracket gives ϕn+1 + gϕn+3, while the first term becomes n Z dxdp 2π u(x, p)x n−1 p 2 = n Z dx xn−1 Z pF −pF dp 2π p 2 = n 3π Z dx xn−1 p 3 F = n 3 Z dx xn−1 p 2 F ρ(x) = n 3 Z dx xn−1 (2ϵF − x 2 − gx4 /2)ρ(x) = n 3  2ϵFϕn−1 − ϕn+1 − g 2 ϕn+3 . Collecting terms, we obtain n 3
Figure 23. Feasible regions with recursion relations

Figures reproduced from arXiv:2610.38716 by the authors.

References

Discussion

Disagree with the review, or know something it missed? Leave a comment, highlight any passage above to comment on it, or challenge this review formally.