OpenAI publishes science blog: GPT-6 Astra assists physicists in proposing a candidate lattice definition of the Standard Model
OpenAI published the first guest post on its science blog. Nathaniel Craig, head of the theoretical particle physics group at the University of California, Santa Barbara, and a visiting scientist at OpenAI, described his collaboration with GPT-6 Astra to propose a candidate lattice definition of the Standard Model. In a preprint completed in October and prepared with the help of GPT-6 Astra, Craig progressively mapped the gauge fields on the original lattice to a finer lattice to compensate for phases; Astra proposed the mapping method, proved that the comparison would converge to a definite answer, and led the establishment of the detailed proof. Lattice field theory experts Maarten Golterman, Andreas Kronfeld, and Yigal Shamir reviewed the proof and raised key questions; the scheme is currently more of an answer in principle, and Craig is investigating whether it satisfies reflection positivity at finite lattice spacing.
Why read this
SelectedAI score 76
The report outlines the technical obstacles of chiral fermion doubling and phase compensation in lattice field theory, as well as the specific work GPT-6 Astra undertook in the construction and proof stages.
Topics
Body · AI translation
Smart Things reported on November 16 that today, OpenAI published the first guest post on its science blog: GPT-6 Astra participated in a Standard Model study, helping physicists propose a candidate scheme for defining the Standard Model on a "lattice."
In the blog, Nathaniel Craig, head of the theoretical particle physics group at the University of California, Santa Barbara, and a visiting scientist at OpenAI, described how he worked with GPT-6 Astra to propose a candidate lattice definition of the Standard Model.
Simply put, physicists divide continuous spacetime into many tiny grids, called "lattices," and then use computer simulations to model the interactions of particles and fundamental forces.
Although the Standard Model can accurately describe fundamental particles and the three fundamental interactions—strong, weak, and electromagnetic—for the past more than 50 years, the physics community has lacked a mathematical definition that can directly handle the complete interactions.
Existing methods usually require first simplifying the interactions into a series of "small corrections," and therefore cannot cover complex phenomena dominated by the strong interaction, such as the proton mass.
In a preprint completed in October and prepared with the help of GPT-6 Astra, Nathaniel Craig proposed a candidate lattice-based definition. Craig had previously had ideas for solving this problem, but had been unable to prove that they could cover all necessary cases. Astra helped him turn the research idea into practice and led the subsequent proof stage.
On the social platform X, a developer said: "We have witnessed another OpenAI scientific breakthrough, but GPT-6.2 still has not been released." He added that GPT-6 Astra helped physicists transform the original research idea into a concrete mathematical construction and completed some key proofs. If this scheme is ultimately validated, it could in the future be used to simulate some particle interactions that are currently difficult to calculate directly. This shows that AI is moving further from "explaining scientific knowledge" toward "participating in scientific research."
▲Comments on X about this research (Source: X)
Some researchers also believe that this research reflects a new division of labor in human-machine collaboration: human physicists are responsible for determining the research direction, while GPT-6 Astra advances topological corrections and mathematical proofs by comparing lattices of different levels of fineness. He said: "This is a striking example of AI beginning to change the way scientific research is done."
▲Comments on X about this research (Source: X)
01. Interactions cannot rely only on small corrections; the Standard Model needs a non-perturbative definition
Quantum mechanics and special relativity describe the interactions of particles and fundamental forces in the Standard Model. But with these interactions alone, physicists still cannot make complete predictions for all phenomena.
Usually, researchers first assume that there are no interactions. For example, they assume that electrons no longer repel one another, and then gradually add interactions in the form of a series of small corrections. As these corrections become smaller and smaller, only a few terms need to be added to obtain very precise answers. This method has enabled physicists to predict the electron magnetic moment to an accuracy of a few parts per trillion.
This method is called perturbation theory.
But many complex phenomena in the Standard Model lie beyond the applicable range of perturbation theory.
Take the proton mass as an example. A proton is composed of three quarks, but its mass mainly comes from the motion and interactions between quarks and the gluons that carry the strong interaction. If the strong interaction is turned off, the proton would completely fall apart. Therefore, perturbation theory cannot effectively explain the magnitude of the proton mass.
Physicists need more powerful methods to study the non-perturbative regime, that is, to retain the complete interactions from the outset rather than treating interactions as small corrections.
For the complete Standard Model in continuous spacetime, the academic community currently has no non-perturbative definition, and it is not even clear whether this interacting theory is mathematically well-defined. One alternative is to discretize spacetime into a lattice, that is, a lattice of points. The lattice spacing determines the shortest distance that the theory can resolve. Studying a finite lattice can provide a rigorous non-perturbative definition of the theory at a specific spacing.
Thus, researchers can put the theory into a computer and directly calculate its predictions without relying on perturbation theory.
This method is called lattice field theory. Over the past few decades, lattice field theory has greatly advanced the physics community's understanding of the strong interaction, including quantitative first-principles calculations of the proton mass.
02. The weak interaction distinguishes left- and right-handed particles; putting it on a lattice produces extra particles
Since lattice field theory has been so successful in strong interaction research, why has this theory not been extended into a non-perturbative lattice definition of the complete Standard Model?
One key obstacle is that the Standard Model belongs to a class of theories known as "chiral theories."
Chiral theories take different forms under spatial reflection. A glove analogy can help explain this concept: a left-hand glove becomes a right-hand glove after mirroring. If you have both left and right gloves, the overall image looks unchanged after mirroring. But if you have only a left-hand glove, the right-hand glove in the mirror will be different from the original.
This "left-right chirality" also exists in many fundamental particles in the Standard Model, namely fermions, including quarks and electrons.
Most interactions in the Standard Model are not chiral. For example, the strong interaction and the electromagnetic interaction treat left- and right-handed fermions in the same way, so they are easier to put on a lattice.
But the third fundamental interaction in the Standard Model, the weak interaction, lacks this reflection symmetry. It treats left-handed and right-handed fermions differently, and therefore does not remain invariant under a mirror reflection that swaps left and right.
It is precisely this differential treatment of particles of different chirality that makes the Standard Model a chiral theory.
And putting a chiral theory on the lattice is very difficult. If left-handed fermions are placed on the lattice, the most natural rules tend to simultaneously produce an unwanted right-handed partner.
This means that directly latticizing the Standard Model yields the wrong particle content for the weak interaction. The "doubling problem" is one of the reasons why a complete non-perturbative definition of the Standard Model has been missing ever since the theory was born.
03. Ordinary mirrors create "right-handed partners"; chiral mirrors offer a solution for latticization
To understand the doubling problem, one can draw an analogy with light. As a light wave propagates forward, the electric field within it oscillates along some direction. Depending on the direction of oscillation, light can be divided into different polarization states. Although ordinary polarized sunglasses typically distinguish between horizontal and vertical polarization, in this analogy it is more appropriate to use left- and right-circular polarization.
Because the electromagnetic interaction has reflection symmetry, left- and right-circular polarization obey the same physical laws.
But if one wants to construct a world with only left-circular light, one can analogize it to the weak interaction selecting only left-handed fermions: researchers can place a laser beam that produces left-circular light into a box with mirrors at both ends, letting the light bounce back and forth inside the box for observation and study. This process is analogous to putting a chiral theory on the lattice.
If ordinary mirrors are used, something unexpected happens: the left-circular light emitted by the laser becomes right-circular light after reflecting off the mirror at the other end, so both polarizations appear in the box at the same time. This is because the mirror changes the direction of propagation of the light but does not change the way the electric field rotates, so when observed from the same end, the polarization direction has flipped.
This is similar to the situation where extra right-handed fermions appear on the lattice: you only put in left-handed light, but the box automatically produces the other chirality.
The solution is to use chiral mirrors. In certain special thin films, molecules arrange themselves layer by layer, forming a helical structure along the film. When the wavelength and chirality of the light match the helical structure, the tiny reflections produced by different layers reinforce one another, causing the rotation direction of the electric field of the reflected light to change. Such thin films can allow left-circular light to remain left-circular after reflection while letting right-circular light pass through. Similar helical structures also exist in nature; for example, the shells of certain scarab beetles also have a helical structure.
On the lattice, a similar trick can also solve the problem. Decades ago, Paul Ginsparg and Nobel laureate Ken Wilson proposed the corresponding lattice rules. Rajamani Narayanan and Herbert Neuberger constructed the corresponding overlap fermions for defining fundamental particles on the lattice.
Subsequently, Martin Lüscher showed how to distinguish left- and right-handed overlap fermions on the lattice. Yoshio Kikukawa and his collaborators studied how to use these methods to describe the weak interaction in the Standard Model.
However, one key part remained unsolved: how to determine the correct phase.
04. Gauge field changes alter the phase; refining the lattice level by level finds a compensation method
Return to the chiral mirror box from earlier. The rotation direction of circularly polarized light can distinguish left-circular from right-circular, but light has another important property: optical phase.
The relative phase of two waves determines whether they reinforce or cancel each other. Waves in phase have their peaks add together, while a peak meeting a trough cancels out. In the chiral mirror box, left-circular light bounces back and forth inside the box. When the reflected light returns, it meets new light emitted by the laser. If the two are in phase, the light is reinforced; if they differ by half a period, the light cancels out.
Therefore, to form a strong beam in the box, the returning light must remain synchronized with the light source each time.
On the lattice, overlap fermions can retain the needed left-handed fermions and eliminate doublers, but they themselves do not tell physicists how to set the phase.
The problem is that the phase on the lattice depends on the state of the Standard Model gauge field at each lattice site. Researchers can start from a set of gauge fields, gradually change these fields through many small steps, and finally return them to their original state.
Although the final gauge field is the same, the process of moving along this closed path may cause the phase to change. Therefore, researchers must find a way to compensate the phase for every allowed path.
Nathaniel Craig's approach is to progressively map the gauge fields on the original lattice onto a series of finer lattices. By comparing the fermion states on adjacent levels of lattices, a portion of the phase correction can be obtained. By accumulating these corrections, the phase rule on the original lattice can be restored.
Astra helped Craig develop this strategy into a complete construction and proof. It proposed a method for mapping gauge fields onto finer lattices, proved that these comparisons converge to a definite answer, and led the establishment of the detailed proof.
Human experts also played a key role. Lattice field theory experts Maarten Golterman, Andreas Kronfeld, and Yigal Shamir reviewed Astra's proof and raised key questions that helped Craig further verify this construction.
05. Progress toward a non-perturbative definition; a continuum spacetime theory still needs refinement
This work proposes a candidate non-perturbative lattice definition of the Standard Model and may answer a question that has been open for more than 50 years. But like all scientific progress, this construction is more of a beginning than an endpoint.
The physics community still needs to judge through discussion and review whether this proposal possesses all the properties we require. The method Craig proposed with Astra's help is currently more of an answer in principle than a complete scheme that can be immediately used for realistic lattice simulations.
Currently, Craig is focusing on whether this construction satisfies reflection positivity at finite lattice spacing. Reflection positivity helps connect the computational results of lattice models with a quantum theory in which probability is conserved. In addition, how to define a complete interacting theory in continuum spacetime remains an open question.
OpenAI's recent proof regarding the existence of Yang-Mills theory has supplied a key missing piece for this problem. Starting from overlap fermions on the lattice, that proof constructed a unified Yang-Mills theory containing the Standard Model gauge fields and fermions.
Craig said these advances are expected to push the physics community toward a deeper understanding of the Standard Model.
06. Conclusion: AI participates in scientific research, moving from assisting computation to theoretical construction
From proposing research ideas to completing complex proofs and then undergoing review by physics experts, the role of GPT-6 Astra in this work has gone beyond that of a traditional scientific research assistant. In this research, scientists were responsible for posing questions, judging direction, and testing results, while AI helped advance the heavy work of derivation, construction, and proof.
Sources: OpenAI, X
This article comes from the WeChat public account "Zhidongxi" (ID: zhidxcom), author: Yang Jingli, published by 36Kr with authorization.
spatial reflection. A glove analogy can help explain this concept: a left-hand glove becomes a right-hand glove after mirroring. If you have both left and right gloves, the overall image looks unchanged after mirroring. But if you have only a left-hand glove, the right-hand glove in the mirror will be different from the original.
This "left-right chirality" also exists in many fundamental particles in the Standard Model, namely fermions, including quarks and electrons.
Most interactions in the Standard Model are not chiral. For example, the strong interaction and the electromagnetic interaction treat left- and right-handed fermions in the same way, so they are easier to put on a lattice.
AI News