AI for BMN Strings

arXiv:2609.39769 · hep-th · Submitted 2026-09-30 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: High Energy Physics: Theory. The day in high-energy theory: gravity, strings, fields and the questions left open.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "AI for BMN Strings".

Mira: Here's the gist of the paper. With the assistance of AI (Artificial Intelligence) tools, we revisit some mathematically well-formulated problems from our previous studies of BMN (Berenstein-Maldacena-Nastase) strings on type IIB string theory pp-wave background These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant.

Kai: The paper focuses on applying AI tools to solve mathematically well-formulated problems from previous studies of BMN strings on a type IIB string theory pp-wave background The AdS/CFT correspondence relates this to a 4d N = 4 SU(N) super-Yang-Mills theory, where BMN operators describe closed string states in the Penrose limit.

Mira: The work aims to make progress on conjectures related to higher-genus critical superstring amplitudes and provide a testable proposal for estimating the cosmological constant using BMN strings.

Kai: Proof of Factorization Formulas The factorization formulas relate string diagram contributions (Si) to field-theory correlators (Fj) via the relation Si = X j m ij F j (2.1).

Paper summary: Kai: To recap, this paper "AI for BMN Strings" revisits some known mathematical problems from previous studies of BMN strings on a type IIB string theory pp-wave background using AI tools. The main thesis is that they are making progress on conjectures related to higher genus critical superstring amplitudes and possible estimates for the cosmological constant.

Mira: They show that the factorization formulas connect string diagram contributions Si to field theory correlators Fj through a relation Si equals X j m i j F j, and they derive results for genus one showing that this means S equals two F.

Lev: The derivation for genus one involves a constrained Fourier completeness identity which leads to a simplex integral representation of the torus two-point function, and it confirms that the string diagram contribution is exactly twice the torus two-point function.

Kai: Beyond those initial results, they investigate the non-negativity of properly normalized all genus BMN two point functions pm n, which they are checking against their conjecture three three.

Mira: For two string modes, they prove that these functions are non-negative at each genus and made a stronger conjecture that it should be true for three string modes separately at each genus.

Lev: They mention tests at genus three and higher support positivity for selected mode pairs, which suggests some positive signs but not necessarily the whole story yet.

Kai: Then they tackle the entropy of BMN strings, Sm n, defined as the von Neumann entropy of the resulting mixed state related to pm n. They establish an upper bound in the strong coupling limit where g goes to infinity, showing it is less than two plus epsilon times log g plus something small.

Mira: They also find a lower bound by looking at pm m which equals about logarithm of g over g to the power of one quarter, leading them to conclude that as g goes to infinity, the entropy Sm n grows at least as fast as log g minus some small constant.

Lev: The paper notes that this entropy analysis gives us an upper bound and a lower bound on how fast these string states can grow in terms of the genus.

Kai: But the real tension comes when they consider conjecture A one, which claims Si m n should be non-negative for every genus h greater than or equal to one and every string diagram i with two or three distinct modes.

Mira: They provide a counterexample at asymptotically large genus where they show that the ratio of SGL from (one two minus three) to (two three minus five) divided by zero to zero is negative, which disproves conjecture A one <ref:2609.39769#pg2>.

Lev: This counterexample at large genus shows that the overall conjecture A one is ultimately disproved because some string diagram contributions can be negative after considering all intermediate states and impurity routings.

Kai: So, to summarize, this paper uses AI to test conjectures about higher genus critical superstring amplitudes and the cosmological constant estimation using BMN strings in a type IIB pp-wave background.

Mira: The work makes progress on some things, but it also clearly flags that the non-negativity conjecture three three and the all coupling entropy monotonicity conjecture in section four point three are still open questions.

Conclusion: Kai: So looking at the title "AI for BMN Strings," this paper is about using computational tools to tackle some deep mathematical problems arising from string theory on this specific background.

Mira: It really sits at the intersection of quantum information and theoretical physics, bringing in the ideas from gauge theory and string theory to see what we can learn together.

Lev: It’s an attempt to connect these very different areas by using AI to explore mathematical structures that have physical meaning for things like the cosmological constant.

Kai: The authors are clearly pushing for a deeper understanding of how these string calculations map onto physics in a way that we can test later.

Mira: Ultimately, they are showing that while they have made some headway, the non-negativity conjecture three three and the all coupling entropy monotonicity conjecture in section four point three remain open questions.

Lev: So for someone listening to this, it means there's still a lot of analysis needed to fully settle those questions.

Kai: It’s just a reminder that even with AI assistance, we need more analytical work and more computational resources to get those specific conjectures resolved.

University of Science and Technology of China

hep-th

Submitted: 2026-09-30

Updated: 2026-11-11

Comments: 41 pages, 3 figures. v2: minor improvements, two appendices added

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 71/100

The gist: With the assistance of AI (Artificial Intelligence) tools, we revisit some mathematically well-formulated problems from our previous studies of BMN (Berenstein-Maldacena-Nastase) strings on type IIB…

Key concepts

BMN Strings
These are specific types of strings studied in type IIB string theory when considering the Penrose limit. They describe closed string states and are used to relate string theory calculations to 4D N=4 super-Yang-Mills theory.
Factorization Formulas
These formulas connect contributions from string diagrams (Si) to field-theory correlators (Fj). The paper showed that for genus one, the planar three-point function is exactly twice the torus two-point function, confirming this relationship.
Non-negativity Conjecture
This conjecture posits that certain BMN two-point functions (pm_n) must be non-negative. While proven for two string modes at each genus, a counterexample was found for three modes at asymptotically large genus.

Terminology

Summary

With the assistance of AI (Artificial Intelligence) tools, we revisit some mathematically well-formulated problems from our previous studies of BMN (Berenstein-Maldacena-Nastase) strings on type IIB string theory pp-wave background These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant. p. 0

Where the paper says it

USTC-ICTS/PCFT-26-65 AI for BMN Strings Min-xin Huang∗ Interdisciplinary Center for Theoretical Study, University of Science and Technology of China, Hefei, Anhui 230026, China Peng Huanwu Center for Fundamental Theory, Hefei, Anhui 230026, China Abstract With the assistance of AI (Artificial Intelligence) tools, we revisit some mathematically well-formulated problems from our previous studies of BMN (Berenstein-Maldacena-Nastase) strings on type IIB string theory pp-wave background. These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant.

Introduction and Motivation

The paper focuses on applying AI tools to solve mathematically well-formulated problems from previous studies of BMN strings on a type IIB string theory pp-wave background The AdS/CFT correspondence relates this to a 4d N = 4 SU(N) super-Yang-Mills theory, where BMN operators describe closed string states in the Penrose limit. The work aims to make progress on conjectures related to higher-genus critical superstring amplitudes and provide a testable proposal for estimating the cosmological constant using BMN strings. p. 0

Where the paper says it

These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant. We make significant progress on some problems, while others remain unresolved.

Proof of Factorization Formulas

The factorization formulas relate string diagram contributions (Si) to field-theory correlators (Fj) via the relation Si = X j m ij F j (2.1). The derivation for genus one shows that the planar three-point function is related to the factorization formula as S = 2F (2. p. 6

Where the paper says it

So we have S = 2 OmOn torus, which is exactly the factorization statement (2. 2 General arguments We consider a fixed genus h ≥ 1, with a single-trace operator at each endpoint; the planar two-point function is the identity.

3). The general derivation involves a constrained Fourier-completeness identity leading to a simplex integral representation of the torus two-point function. p. 6
Where the paper says it

So we have S = 2 OmOn torus, which is exactly the factorization statement (2. 2 General arguments We consider a fixed genus h ≥ 1, with a single-trace operator at each endpoint; the planar two-point function is the identity.

The final result shows that the string diagram contribution is exactly twice the torus two-point function, which matches the factorization statement (2. p. 6
Where the paper says it

This matches exactly twice the torus two-point function, see e. So we have S = 2 OmOn torus, which is exactly the factorization statement (2.

3).

Non-negativity of BMN Two-Point Functions

The paper investigates the non-negativity of properly normalized all-genus BMN two-point functions pm n:= g squared sinh(g/2) X h>=0 OmOn h, (3.1). A consequence of the proposal is the non-negativity of BMN two-point functions, which is easy to prove at each genus for the case of two string modes. p. 1

Where the paper says it

2 General arguments . 6 3 Non-negativity of BMN two-point functions 11 4 Entropy of BMN strings 13 4.

The authors made a stronger conjecture that this was also true for the case of three string modes separately at each genus, i. p. 3
Where the paper says it

However, the mathematical structures are stronger that the same combinatorial structure applies to any number of string modes. 1 Genus one For illustration we first consider the simple case of genus one.

e., Om1,m2,m3On1,n2,n3 h >= 0 (3.3). Extensive tests at genus three and higher support positivity for selected mode pairs.

Entropy of BMN Strings

The entropy Sm(g) is defined as the von Neumann entropy of the resulting mixed state, which is related to the two-point functions pm n(g) via Sm(g):= -X n∈Z pm n(g) log pm n(g). The authors proved an upper bound for the entropy Sm(g) < (2 +ϵ) log(g) +o(1) in the strong coupling limit g → ∞ in [8]. p. 2

Where the paper says it

Unfortunately, in Section 4. 2, we prove with AI assistance that the entropy of BMN strings is unbounded in the strong coupling limit, rendering this proposal nonviable.

A lower bound is established by showing that pm m(g) = Om((log g/g) 1/4). p. 1
Where the paper says it

1 An improved upper bound . 2 Lower bounds .

This leads to the conclusion that as g → ∞, Sm(g)(g) ≥ log g − O(1) −→ ∞. p. 1
Where the paper says it

3 Monotonicity . 25 5 Conclusion 26 A A stronger conjecture 27 A.

Stronger Conjecture and Counterexample

The paper considers a stronger conjecture (A.1) that Si(m, n) ≥ 0 for every genus h ≥ 1 and every string diagram i from the factorization formulas with two or three distinct string modes. p. 3

Where the paper says it

So the multiplicity is 2 and the factorization formula is just S = 2F. The relation was checked for the cases of two [5], three [6], four [9] string modes.

The authors provide an ingenious counterexample at asymptotically large genus where lim L→∞ SGL ((1, 2, −3),(2, 3, −5)) SGL ((0, 0, 0),(0, 0, 0)) = −256/1215π6 < 0 (A.6). The counterexample shows that the conjecture (A. p. 1
Where the paper says it

3 Monotonicity . 25 5 Conclusion 26 A A stronger conjecture 27 A.

1) is ultimately disproved by an ingenious counterexample at large genus, as discussed in the Appendix A. p. 1
Where the paper says it

2 Lower bounds . 1 Large initial mode .

Conclusion

The work makes progress on several mathematically well-defined problems in the study of BMN strings. However, despite many prompts, the non-negativity conjecture (3. p. 1

Where the paper says it

2 General arguments . 6 3 Non-negativity of BMN two-point functions 11 4 Entropy of BMN strings 13 4.

p. 1
Where the paper says it

2 General arguments . 6 3 Non-negativity of BMN two-point functions 11 4 Entropy of BMN strings 13 4.

3) in Section 3 and the all-coupling entropy monotonicity conjecture in Section 4. p. 2
Where the paper says it

Unfortunately, in Section 4. 2, we prove with AI assistance that the entropy of BMN strings is unbounded in the strong coupling limit, rendering this proposal nonviable.

p. 2
Where the paper says it

Unfortunately, in Section 4. 2, we prove with AI assistance that the entropy of BMN strings is unbounded in the strong coupling limit, rendering this proposal nonviable.

3 remain unresolved. p. 0
Where the paper says it

These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant. We make significant progress on some problems, while others remain unresolved.

p. 0
Where the paper says it

These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant. We make significant progress on some problems, while others remain unresolved.

Further analytical works, aided by better AI tools and more computational resources, may help resolve these questions. p. 1
Where the paper says it

In particular, a historic achievement is the solution of a Millennium prize problem on Navier-Stokes equations by an advanced AI model, based on some earlier human works. At present, the AI tools are particularly suited for studying mathematically well-formulated problems.

p. 1
Where the paper says it

In particular, a historic achievement is the solution of a Millennium prize problem on Navier-Stokes equations by an advanced AI model, based on some earlier human works. At present, the AI tools are particularly suited for studying mathematically well-formulated problems.

The derivations and proofs in this paper were developed with assistance from GPT-6 Astra. p. 2
Where the paper says it

In our previous works [4, 5, 6, 7, 8, 9, 10, 11], we proposed and studied such mathematically well-defined relations and their physical implications. In this paper, with the assistance of public tools from OpenAI ChatGPT, we will make progress on some previous conjectures.

p. 2
Where the paper says it

In our previous works [4, 5, 6, 7, 8, 9, 10, 11], we proposed and studied such mathematically well-defined relations and their physical implications. In this paper, with the assistance of public tools from OpenAI ChatGPT, we will make progress on some previous conjectures.

The text was written by the author with assistance from AI tools. p. 0
Where the paper says it

USTC-ICTS/PCFT-26-65 AI for BMN Strings Min-xin Huang∗ Interdisciplinary Center for Theoretical Study, University of Science and Technology of China, Hefei, Anhui 230026, China Peng Huanwu Center for Fundamental Theory, Hefei, Anhui 230026, China Abstract With the assistance of AI (Artificial Intelligence) tools, we revisit some mathematically well-formulated problems from our previous studies of BMN (Berenstein-Maldacena-Nastase) strings on type IIB string theory pp-wave background. These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant.

p. 0
Where the paper says it

USTC-ICTS/PCFT-26-65 AI for BMN Strings Min-xin Huang∗ Interdisciplinary Center for Theoretical Study, University of Science and Technology of China, Hefei, Anhui 230026, China Peng Huanwu Center for Fundamental Theory, Hefei, Anhui 230026, China Abstract With the assistance of AI (Artificial Intelligence) tools, we revisit some mathematically well-formulated problems from our previous studies of BMN (Berenstein-Maldacena-Nastase) strings on type IIB string theory pp-wave background. These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant.

A Stronger Conjecture

The authors consider an even stronger conjecture that, for every genus h ≥ 1 and every string diagram i from the factorization formulas with two or three distinct string modes, Si(m, n) ≥ 0? (A.1). The tests at higher genus show that individual non-reducible string-diagram contributions are non-negative for all mode numbers. p. 2

Where the paper says it

The situation is also easier in low dimensional non-critical strings, see e. If our conjectures for the “pp-wave holographic dictionary” are correct, we have provided the critical superstring amplitudes at any higher genus, and even exact results at any finite string coupling due to the convergence of the genus expansion, in the extreme pp-wave background with infinite spacetime curvature and RamondRamond flux.

However, a counterexample is constructed at asymptotically large genus where lim L→∞ SGL ((1, 2, −3),(2, 3, −5)) SGL ((0, 0, 0),(0, 0, 0)) = −256/1215π6 < 0 (A.6).

A Counterexample

For a positive integer L, let GL be the diagram shown in Figure 2, obtained by splitting the incoming string into two daughters, inserting L consecutive genus-one two-point diagrams on the first daughter and 3L on the second, and finally joining the daughters. Its genus is h = 4L + 1. p. 0

Where the paper says it

USTC-ICTS/PCFT-26-65 AI for BMN Strings Min-xin Huang∗ Interdisciplinary Center for Theoretical Study, University of Science and Technology of China, Hefei, Anhui 230026, China Peng Huanwu Center for Fundamental Theory, Hefei, Anhui 230026, China Abstract With the assistance of AI (Artificial Intelligence) tools, we revisit some mathematically well-formulated problems from our previous studies of BMN (Berenstein-Maldacena-Nastase) strings on type IIB string theory pp-wave background. These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant.

Both branches persist between the outer splitting and joining vertices, so GL is non-reducible in the sense defined above. p. 7
Where the paper says it

the splitting-and-joining histories that produce a given field-theory diagram. Then we show that the sewing variables for each history can be replaced by the lengths of its 4h segments, giving the same simplex integral as on the field-theory side.

The counterexample shows that, with all intermediate states and impurity routings included, lim L→∞ SGL ((1, 2, −3),(2, 3, −5)) SGL ((0, 0, 0),(0, 0, 0)) = −256/1215π6 < 0 (A. p. 8
Where the paper says it

Consequently the number of sewings is the same mij at every such configuration. Here Si includes all allowed impurity routings on its internal strings, as in its definition; a routing is fixed once the history and impurity positions are given.

p. 8
Where the paper says it

Consequently the number of sewings is the same mij at every such configuration. Here Si includes all allowed impurity routings on its internal strings, as in its definition; a routing is fixed once the history and impurity positions are given.

6).

The Tests

For the case of two string modes, the common phase from shifting the circle origins cancels between the two impurity factors. Averaging (2.21) over these 4h choices therefore gives Si(m, n) = g 2h(4h)! p. 5

Where the paper says it

4) cancel the powers produced by (2. 6), and the sum over intermediate modes therefore gives S = g 2 2 Z 1 0 dx Z x 0 dt Z 1−x 0 du X A⊆{1,.

X j m ij Z∆4h−1 X 4h r=1 Aj dµh = g 2h(4h)! X j m ij Z∆4h−1 Aj squared dµh ≥ 0. This proves non-negativity for each string diagram separately, with all intermediate states and impurity routings summed, for the case of two string modes. p. 3
Where the paper says it

, nk) ∈ Z k be two sets of string modes, satisfying the level matching condition X k i=1 mi = 0, X k i=1 ni = 0. Then the factorization formula is X two-string intermediate states Om OpOq planar OpOq On planar = 2 OmOn torus.

Now we consider the non-trivial case of three string modes. The authors use exact computations to establish that every individual string-diagram contribution at genus two is non-negative for all mode numbers. p. 2

Where the paper says it

The situation is also easier in low dimensional non-critical strings, see e. If our conjectures for the “pp-wave holographic dictionary” are correct, we have provided the critical superstring amplitudes at any higher genus, and even exact results at any finite string coupling due to the convergence of the genus expansion, in the extreme pp-wave background with infinite spacetime curvature and RamondRamond flux.

The tests at higher genus show that the total computational time is less than an hour with AI tools. p. 0
Where the paper says it

USTC-ICTS/PCFT-26-65 AI for BMN Strings Min-xin Huang∗ Interdisciplinary Center for Theoretical Study, University of Science and Technology of China, Hefei, Anhui 230026, China Peng Huanwu Center for Fundamental Theory, Hefei, Anhui 230026, China Abstract With the assistance of AI (Artificial Intelligence) tools, we revisit some mathematically well-formulated problems from our previous studies of BMN (Berenstein-Maldacena-Nastase) strings on type IIB string theory pp-wave background. These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant.

The counterexample shows that, with all intermediate states and impurity routings included, lim L→∞ SGL ((1, 2, −3),(2, 3, −5)) SGL ((0, 0, 0),(0, 0, 0)) = −256/1215π6 < 0 (A.6).

For all-zero external modes, the vertex vector has only its all-zero daughter component, equal to x(a+1)/2(1−x)(b+1)/2. The normalized weight in (A. p. 8

Where the paper says it

The finite position sums above provide the same measure without requiring a mode sum on a vacuum string. Sewing the normalized planar Wick-contraction vertices through a complete intermediate-state basis reproduces the usual Wick weight, including cyclic and identical-string factors.

13) concentrates at x = 1/4, the unique maximum of x(1 − x) cubed. The local uniform convergence just established and the bound HL ≤ 1 thus imply that the ratio tends to H∞(1/4). Finally, C(1,2,−3)(1/4) = 1/12π3, C(2,3,−5)(1/4) = -1/60π3. The counterexample shows that the unequal numbers of insertions select a size ratio where the two external cubic couplings have opposite signs. p. 3
Where the paper says it

Then the factorization formula is X two-string intermediate states Om OpOq planar OpOq On planar = 2 OmOn torus. 3) The intermediate sum includes the choice of size and a subset of k string modes for the daughter strings, and we also need to sum over all intermediate mode numbers p and q satisfying the level matching condition.

The paper makes progress on several mathematically well-defined problems in the study of BMN strings. However, despite many prompts, the non-negativity conjecture (3.3) in Section 3 and the all-coupling entropy monotonicity conjecture in Section 4.3 remain unresolved. Further analytical works, aided by better AI tools and more computational resources, may help resolve these questions. The derivations and proofs in this paper were developed with assistance from GPT-6 Astra. The text was written by the author with assistance from AI tools.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved AI can perform rigorous mathematical proof verification for complex string theory conjectures by leveraging exact genus-three computational results and AI-assisted exhaustive testing of rational functions, as shown in Appendix A of Section 5.

  2. The system can execute sophisticated Monte Carlo evaluations on complex correlators, sampling admissible diagrams uniformly and integrating impurity positions analytically to provide unbiased statistical tests for non-negativity of matrix elements (3.3) across various string modes and genera.

  3. The AI can rigorously check the combinatorial identity (4.26) by counting genus-h diagrams with ordered pairs of distinct intervals having equal translations, thereby proving the essential relationship between field theory and string diagram multiplicities.

  4. The system can implement a finite-average form of Azuma's inequality to bound the deviation between individual matching probabilities and their averages, providing rigorous control over the statistical fluctuations in BMN two-point functions at fixed coupling.

  5. The AI can perform symbolic manipulation of complex rational functions derived from genus-two string diagrams to exhaust all possible cases and prove non-negativity for individual contributions (A.4), which is crucial for verifying the stronger conjecture (A.1).

Abstract

With the assistance of AI (Artificial Intelligence) tools, we revisit some mathematically well-formulated problems from our previous studies of BMN (Berenstein-Maldacena-Nastase) strings on type IIB string theory pp-wave background. These mathematical problems have important physical implications, relating to the computations of higher-genus critical superstring amplitudes, possible estimate of the cosmological constant. We make significant progress on some problems, while others remain unresolved.

Sources

Related papers