天体物理学家如何利用Codex模拟黑洞

黑洞周围的引力极其强大，任何物质一旦靠近到一定距离，就连光也无法逃脱。像陈志均这样的天体物理学家通过计算机模拟和观测来研究黑洞，但当前的算法和计算能力限制了这些模拟的真实性。

作为亚利桑那大学和斯图尔德天文台的研究员，陈志均正借助 Codex 攻克这一难题。

他表示，黑洞是检验爱因斯坦广义相对论的最佳场所之一。该理论目前是我们对引力的最佳解释：引力并非将物体拉拢的力，而是质量与能量弯曲时空结构的结果。

陈志均是国际事件视界望远镜（EHT）合作项目的成员，该项目于2019年发布了首张黑洞图像。该团队目前正在收集观测数据，旨在制作首部超大质量黑洞视频，重点关注M87星系中心的那个黑洞。

然而，将观测转化为科学理解需要海量数据处理、大规模计算工作流以及能够模拟宇宙中最极端物理现象的仿真程序。

由于光无法逃离黑洞，科学家转而研究其周围被称为事件视界的区域——这是一个物质无法逃逸的边界。“这是一道有去无回的表面，”陈志均说。在边界外旋转的物质会发出光线，天体物理学家可以观测、测量并模拟这些光。

EHT于2019年发布的图像显示，黑洞的阴影嵌在事件视界附近炽热的等离子体中。陈志均协助开发了团队用于解读观测数据的模拟和计算工具。此后，随着团队从静态图像转向视频研究，陈志均和同事们持续改进仪器和观测能力。

一段由超级计算机模拟生成的短视频，展示了银河系中心黑洞周围等离子体的运动。

图片来源： EHT理论工作组 / CK Chan

应对螺旋难题

陈志均团队面临的最大障碍之一是模拟黑洞周围的等离子体。等离子体是由带电电子和离子组成的超高温物质。

在许多模拟中，科学家将等离子体简化为流体，利用经典方程模拟其绕黑洞的运动。这种方法在密度较高、电子与离子频繁碰撞的等离子体中效果尚可。

但在陈志均团队研究的超大质量黑洞附近，某些区域温度极高且物质稀薄，粒子几乎不会相遇。“它们实际上不会相互碰撞，”他说。相反，粒子主要沿磁力线螺旋运动。

要正确模拟这种行为，研究人员需要追踪数万亿个电子和离子绕黑洞快速螺旋运动的轨迹。标准模拟必须计算每个微小转弯，迫使计算机采用极短的时间步长。

结果，即便是全球最快的超级计算机，大部分时间也耗费在计算这些微观粒子运动上，而非模拟科学家真正想研究的宏观行为。

“几十年来，这限制了我们对黑洞等离子体模拟的真实性，”陈志均说。

用AI构建更优的数字孪生

陈志均推测，新的数学技术或许能突破部分限制。基本思路是通过数学变换改变模拟追踪粒子运动的方式，使计算机无需直接追踪每个微小螺旋。

“但手动探索所有数学可能性将耗费大量时间，”陈志均说。于是他借助 Codex 推导候选算法，并用已知解进行验证。

Codex 生成了许多潜在方案——并非全部正确。“但这没关系，”他说，“大多数科学想法都会失败。关键在于这些算法是可检验的。一旦找到有效方案，就可能解锁此前无法实现的模拟。”

某些AI系统能直接返回结果，却不展示推导步骤。但陈志均的团队利用 Codex 提出并实现数值方案，以便他们从物理角度检查、测试和理解。

大型语言模型仍会出错，许多科学家对在研究中应用AI持谨慎态度。但陈志均认为，科学恰恰可能是当今AI系统的最佳应用场景之一，因为科学思想可以严格检验。

“我们不会因为某个想法来自爱因斯坦、优秀学生或AI模型就接受它，”他说，“只有经过反复验证后才会接受。”

陈志均将AI视为一种工具，能帮助研究人员探索更多想法、加速测试，并在坚持验证与可重复性的前提下推动发现。

如果陈志均用 Codex 测试的方法成功，新算法最终将让科学家模拟黑洞周围数万亿个粒子。这将使研究人员得以研究数十年来难以触及的物理现象。

The gravity around a black hole is so extreme that nothing, not even light, can escape once it gets close enough. Astrophysicists like Chi-kwan Chan study black holes with computer simulations and observations. But current algorithms and computing power limit how realistic those simulations can be.

With Codex, Chan—a researcher at the University of Arizona and Steward Observatory—is tackling this problem.

Black holes are among the best places to test Einstein’s general theory of relativity, he said. The theory is currently our best explanation of gravity: instead of a force pulling objects together, gravity is the result of mass and energy bending the fabric of space and time.

Chan is part of the international Event Horizon Telescope (EHT) collaboration, which published the first image of a black hole in 2019. The team is currently gathering observations to produce the first video of a supermassive black hole, focusing on the one at the center of the M87 galaxy.

But turning observations into scientific understanding requires enormous amounts of data processing, large-scale computing workflows, and simulations capable of modeling some of the most extreme physics in the universe.

Since light can’t escape a black hole, scientists instead study the region around it called the event horizon, a boundary beyond which matter can’t escape. “It’s a surface of no return,” said Chan. Matter swirling just outside this boundary emits light that astrophysicists can see, measure, and simulate.

The 2019 image released by the EHT showed a black hole’s shadow embedded in glowing plasma near the event horizon. Chan helped develop the simulation and computing tools the team used to interpret the observations. Since then, Chan and his colleagues have continued improving their instruments and observing capabilities as the team moves from still images toward videos.

A short video generated by a supercomputer simulation showing the movement of plasma around the black hole at the center of the Milky Way galaxy.

Credit: EHT Theory Working Group / CK Chan

Addressing a spiraling problem

One of the biggest roadblocks for Chan and his team is modeling the plasma around black holes. Plasma is superheated matter made up of electrically charged electrons and ions.

In many simulations, scientists simplify plasma by treating it like a fluid, using well-known equations to model its movement around a black hole. That works reasonably well in denser plasma where the electrons and ions constantly collide with each other.

But near the supermassive black holes that Chan and his colleagues are studying, some regions become so hot and diffuse that particles rarely encounter each other. “They don’t really collide with each other,” he said. Instead, the particles mostly spiral around magnetic field lines.

To model that behavior correctly, researchers need to follow trillions of electrons and ions as they rapidly corkscrew around a black hole. Standard simulations must calculate every tiny turn, forcing computers to take extremely small timesteps.

As a result, even the world’s fastest supercomputers can spend most of their time calculating these minuscule particle motions instead of simulating the larger behavior scientists actually want to study.

“For decades, this has limited how realistically we can simulate black hole plasma,” Chan said.

Using AI to build a better digital twin

Chan suspected that new mathematical techniques could help work around some of these limitations. The basic idea was to change, mathematically, how the simulation tracked particle motion so the computer no longer had to follow every tiny spiral directly.

“But exploring all the mathematical possibilities by hand would have taken an enormous amount of time,” Chan said. So he turned to Codex to help derive candidate algorithms and test them against known solutions.

Codex generated many potential approaches—not all of them correct. “But that’s okay,” Chan said. “Most scientific ideas fail. What matters is that these algorithms are testable. Once you find one that works, it can potentially unlock simulations that were previously impossible.”

Some AI systems can return results without showing the steps they used to produce their conclusions. But Chan’s group uses Codex to propose and implement numerical schemes that they can inspect, test, and understand physically.

Large language models still make mistakes, and many scientists remain cautious about using AI in research. But Chan believes science may be one of the best uses for today’s AI systems precisely because scientific ideas can be tested rigorously.

“We don’t accept an idea because it came from Einstein, from a bright student, or from an AI model,” he said. “We accept it only after repeated testing.”

Chan sees AI as a tool that can help researchers explore more ideas, test them faster, and accelerate discovery while remaining grounded in verification and reproducibility.

If the approaches that Chan is testing with Codex succeed, the new algorithms could eventually allow scientists to simulate trillions of particles around black holes. That would enable researchers to study physics that has remained out of reach for decades.