关键化学问题得到解答，无需量子计算机
内容总结：
标题：无需量子计算机，经典计算成功破解固氮酶关键化学难题
导语： 长期以来，量子计算机被视为解决某些复杂化学问题的“终极武器”，但一项新成果正在挑战这一观点。加州理工学院的化学家加内特·陈（Garnet Chan）领导的研究团队，利用纯粹的经典计算方法，成功解析了固氮酶（nitrogenase）活性中心的关键电子结构，完成了此前被认为必须依赖量子计算机才能实现的目标。这一成果不仅深化了人类对生命基础过程的理解，也重新点燃了关于经典与量子计算机孰优孰劣的争论。
正文：
一、 经典计算的“逆袭”
固氮酶是自然界中一种能将大气中惰性的氮气转化为生命可利用的氨的酶，对地球生命至关重要。然而，其活性中心“FeMo-co”包含多个铁原子和钼原子，电子之间存在着强烈的量子纠缠效应，导致其可能的状态组合数量超过七万八千种。长期以来，许多量子计算研究者认为，要精确模拟这一系统，必须依赖能够处理量子态的量子计算机。
但加内特·陈教授并不认同。这位以基础科学为毕生追求的化学家认为：“如果经典计算机是解决问题的正确工具，我们就应该用它。我们没有必要非得等到容错量子计算机构建完成。”他的团队在2024年初发表的研究成果，正是这一信念的有力证明。
研究人员采用了两种不同的经典计算方法，通过巧妙的数据压缩和筛选，成功“忽略”了那些对最终能量贡献微乎其微的电子状态，最终精确计算出了FeMo-co的基态能量。这一结果不仅与实验观测数据高度吻合，也为经典计算在解决“电子关联”这一核心难题上开辟了新路径。
二、 争议并未平息
尽管这一成果被视为理论化学的重大胜利，但它并未终结关于量子计算机价值的讨论。
部分研究者指出，经典计算解决这一问题耗时长达近二十年，效率低下。达特茅斯学院的量子计算理论家詹姆斯·惠特菲尔德（James Whitfield）认为，针对单一分子系统的成功案例并不具备普适性，量子计算机在“规模化”进行此类发现方面仍不可替代。此外，他认为量子计算机的真正优势可能并非计算静态的基态能量，而是在于模拟系统随时间的动态演化，而后者正是经典计算的“软肋”。
加内特·陈教授对此保持开放态度。他承认量子计算机未来在化学模拟领域潜力巨大，但他希望自己的成果能纠正一种“化学难题只能等量子计算机来解决”的误解。“科学具有自我纠错的能力，”他表示，“但很多时候，纠错的声音往往不如最初声称的那么响亮。”
三、 未来展望
对于化学界而言，这一结果为最终完整解析固氮酶的全反应机制铺平了道路。麻省理工学院的化学家丹尼尔·苏斯（Daniel Suess）评价道，虽然距离描绘完整的反应路径还有很长距离，但该方法提供了一个可靠的起点。
结语： 这场关于经典与量子计算机孰优孰劣的“世纪辩论”远未结束，但加内特·陈团队的成果无疑给经典计算阵营注入了一剂强心针。它提醒世人，在追求未来量子计算的同时，经典算法的潜力仍不可小觑。
中文翻译：
无需量子计算机，关键化学问题获解
引言
加内特·陈最关心的是基础科学。数十年前，他投身化学领域，旨在理解地球上一些最重要的生化过程。
但自那时起，他在另一个领域成为了核心人物：关于量子计算机是否将比普通“经典”计算机具有决定性优势的辩论。过去十年间，许多量子计算研究人员将陈所研究的特定反应视为量子计算机应能大显身手的领域。然而，陈长期怀疑功能强大的量子计算机——距离问世仍需数年——是否确属必要。
“我的主要兴趣是解决化学问题。如果经典计算机是完成这一目标的正确工具，我们就该用它，”他说。虽然他相信量子计算机最终将在该领域发挥重要作用，“但我不明白我们为何要等待一台容错量子计算机被建造出来。”
如今，一项新成果为他的论点提供了有力支撑。
一月初，陈与另外五名来自加州理工学院的量子化学家在理解固氮酶方面达成了一项关键里程碑。固氮酶能将大气中的氮转化为氨，使地球上的生命成为可能。这对理论化学家而言是一次重大胜利，也是数十年努力的结果。
但多年来，固氮酶也一直是量子计算领域的概念验证目标。要理解这种酶，研究人员必须追踪许多通过量子纠缠联结在一起的电子行为。可能的构型数量呈爆炸式增长。研究人员曾假设，他们或许只能通过能够操控量子态的机器来解析这个系统。
然而，陈和他的同事们纯粹使用了经典方法。这使得他们的成果不仅对于支撑生命的化学过程，而且对于理解该过程是否需要量子计算机，都成为了一项关键性声明。
“我认为有必要澄清，这并非一项‘必须先建造量子计算机才能对该问题发表任何看法’的不可完成的任务，”陈说。
并非所有人都同意这一点。一些研究人员指出，使用经典方法获得这一成果耗费了多年时间。他们说，即便一个化学问题最终被证明可以用经典方法解决，但要想大规模地做出此类发现，仍然需要量子计算机。
“如果我们挑选任何一个优化问题，并投入20年时间，你就能弄清楚那一个系统，”达特茅斯学院的量子计算理论家詹姆斯·惠特菲尔德说，“但这个解决方案是否具有可迁移性？诸如此类的问题不会通过解决一个分子系统的一个实例就能得到回答。”
解决这个关于固氮酶的特定问题可能还无法就此终结关于量子计算机的辩论，但每朝着理解该酶的完整化学性质前进一步，都能使这场辩论减少一些假设性。
大自然的氨工厂
与光合作用并列，固氮是地球上生命最重要的化学过程之一。而固氮酶正是使其成为可能的关键。
在固氮酶进化出来之前，生物受到可用于合成有机物的氮含量的限制。这是一个讽刺性的障碍，因为地球实际上充满了氮：这种元素约占大气的80%。但大气中的氮以双原子分子N₂的形式存在，它是惰性的，因此无法用于生物过程。只有罕见的高能事件才能将这种分子分解为生命可利用的硝酸盐。
“生物体实际上是在等待闪电击中。这就是让氮可供生物质利用的方式，”麻省理工学院研究固氮酶的化学家丹尼尔·苏斯说。
但30亿年前，当固氮酶在早期原核生物中进化出来时，氮的“闸门”打开了。这种酶完成了其他任何生物过程都无法做到的事情：它打破了将N₂结合在一起的叁键，将惰性气体转化为对生物有用的氨。
这种酶效率很高，但极其复杂。这对从中受益的早期微生物来说无关紧要，但对数十亿年后想要复制其“把戏”以制造肥料的人类来说，却至关重要。
固氮酶在化学上如此困难的部分原因在于它的“活性位点”——一个由铁和钼原子组成的簇，称为FeMo-co。每个铁原子带有四到五个未配对电子，它们的行为彼此依赖。事实上，FeMo-co是整个生物学中关联性最强的系统之一，也是所谓的“电子关联问题”的一个典型例子：由于无法独立处理其电子，要确定整个系统的性质（如其真实的电子结构和能量）极其困难。
在人类历史的大部分时间里，紧迫的问题并非固氮酶如何工作，而是如何获得它所产生的足够多的产物。直到19世纪，最可靠的可利用氮源还是从秘鲁沿海岛屿采集的鸟粪，这种资源如此宝贵和稀缺，以至于国家为之开战。随后，德国化学家弗里茨·哈伯和卡尔·博世在1909年攻克了工业固氮问题，其实际意义随之减弱。
而科学问题——理解藏身于普通土壤细菌体内的固氮酶如何完成哈伯-博世法需要工业熔炉才能做到的事情——依然悬而未决。
这本身就是一个重要问题，并且随着人们就解决它的最佳方式展开辩论，它获得了新的关注。
一场非典型的检验
经典计算机以比特（取值为0或1）的形式处理信息。而量子计算机则使用量子比特，它可以同时处于0和1的叠加态，并且可以以经典计算机无法模拟的方式相互纠缠。这意味着，当（或者说如果）一台大规模量子计算机存在时，它将能够同时探索一个问题的许多可能解，而不是逐一进行运算。
对于具有正确数学结构的某些类型问题，这有望带来比任何经典机器所能达到的速度呈指数级的提升。自20世纪90年代量子计算作为理论研究课题兴起以来，问题一直在于哪些问题符合条件。最有前景的领域之一似乎是模拟化学相互作用：主导分子行为的电子相互作用其核心是量子力学的，这表明量子计算机可能特别适合于模拟它们。
固氮酶作为非正式量子计算基准的地位可追溯到2011年微软组织的一次会议，旨在为其新成立的量子计算小组探索应用。当时已研究固氮酶超过十年的陈就该酶做了报告。
他不清楚这场报告在多大程度上影响了后来的事态发展，但在2017年，微软研究人员在《美国国家科学院院刊》上发表了一篇论文，认为固氮酶的纠缠复杂性使其成为量子计算机的一个引人注目的测试对象。
在陈看来，这从一开始就是一个奇怪的搭配。他对此说法提出异议，仍然相信使用他职业生涯中一直发展的经典方法对固氮酶进行建模是可能的。
在接下来的十年里，他将着手证明这一点。
基态之争
陈和其他研究人员并非旨在从头到尾解释固氮酶的工作原理。相反，他们转向了广泛使用的FeMo-co计算模型，并提出一个更初步的问题：它的基态能量是多少？
基态——FeMo-co能量最低的电子构型——是整个反应的起点。但FeMo-co包含一个由七个铁原子组成的簇，每个铁原子有四个或五个未配对电子，这些电子的量子“自旋”可以向上或向下，其轨道可以移动，并且它们的行为取决于周围电子的行为。
这使得测量FeMo-co的基态能量变得异常复杂。电子可能处于的合理构型有超过78000种；基态是所有这构型的一个叠加态，或者说一种加权组合。原则上，薛定谔方程可以告诉你所有这些不同的构型如何贡献于基态，以及基态的总能量应该是多少。但在实践中，对于像FeMo-co这样拥有众多相互作用电子的系统，直接且精确地求解该方程通常是不可行的。
这对于量子计算机和经典计算机都是如此。在这两种情况下，你都必须从对基态基本结构的一个更简单的近似开始——一个基于充分研究的猜测，通常需要多年的研究才能得出，关于哪些构型对基态的贡献最大。
然后，如果你使用的是经典计算机，你可以尝试逐步考虑其他构型，并证明你可以安全地忽略大量剩余构型，因为它们对基态能量的贡献不大。
另一方面，理论上，量子计算机不需要你在最终估算中忽略某些构型。相反，计算机可以直接将你的初始猜测表示为一个量子态，然后让这个态随时间向前演化，直到它自然地达到正确的基态结构——从而允许你精确地计算能量。
许多研究人员认为量子计算机在这方面具有优势，因为用经典方法排除不重要的构型这一过程可能会变得极其困难。然而，陈和其他人不同意。一方面，他们认为，量子计算机仍然会遇到需要那个合理的初始猜测的相同瓶颈，并且没有明显理由表明量子方法在突破这一瓶颈方面有任何优势。此外，经典技术正在迅速成熟。
但对陈来说，断言可能终究不需要量子计算机，“就像试图抵抗潮汐一样，”他说。
筛选出解决方案
自2000年从剑桥大学获得博士学位以来，陈一直致力于开发和改进通过只关注最重要的构型来压缩复杂量子态的方法。他和他的团队现在希望将这些方法应用于FeMo-co。
他们使用了两种不同的技术来筛选需要关注的构型。使用第一种方法，他们从猜测开始，逐步调整少量电子的行为。然后他们证明，调整更多数量的电子并不会导致显著的能量变化，这为他们提供了清晰的规则，指明哪些构型可以忽略，哪些不能忽略。
第二种方法是陈职业生涯中一直致力研究的。它涉及将初始态分解成多个部分，并只允许有限数量的信息在这些部分之间流动。然后他们证明，只需要考虑信息流变化达到某个特定极限的情况。“意识到可以通过‘更简单’的方法实现描述，并极力推动这些方法（因为该问题在计算上仍然具有挑战性），这是关键，”陈在一封电子邮件中写道。
两种方法都对FeMo-co的基态能量得出了相同的估算值（并且与科学家的实验结果相符），这让研究人员相信他们找到了真正的基态。
辩论的转向
陈希望，他的团队所取得的技术突破现在可以扩展到对整个固氮酶及其反应进行建模。“我希望所有那些主张‘我们需要建造一台量子计算机来解决固氮酶问题’的人，现在既然我们有了解决它的途径，能够加入这项使命，”他说。
但是，从基态到对该反应的完整数学描述，将会困难得多，这涉及计算一整系列中间化学态的能量。“我们离实现这一终极目标还差得很远，”苏斯说。“我们仍然只是描述了静息态。但这种方法很有前景，因为它表明我们可以有一些信心继续推进。”
目前也不清楚这一结果对研究人员在量子计算方面的期望可能意味着什么。惠特菲尔德认为，计算单个基态能量值从来就不是量子计算机预期能超越经典计算机的地方。他说，量子计算机可能拥有的优势反而在于接下来的问题：对系统随时间演化的方式进行建模。这很可能会展示经典方法会变得多么低效，以及量子计算机可能有多强大。
在与量子计算界进行了多年的友好切磋之后，陈并不期望这一新结果能改变许多人的想法。毕竟，他说，通过量子计算机进行量子化学模拟仍然前景广阔：如果明天就有一台量子计算机可用，他会很乐意使用它。但他希望他团队的新成果能有助于纠正一种误解，即认为最困难的化学问题在量子硬件问世之前根本无法触及。
“科学是自我修正的，”他在一封电子邮件中写道，“但很多时候，修正案得到的关注远不及最初的声明，因为该领域已经转向了其他声明。”
英文来源：
Key Chemistry Question Answered, No Quantum Computer Required
Introduction
What Garnet Chan cares most about is basic science. He entered chemistry decades ago to understand some of the most consequential biochemical processes on Earth.
But since then, he’s become a central figure in a different arena: the debate over whether quantum computers will have a decisive advantage over ordinary “classical” ones. Over the past decade, many quantum computing researchers have identified the very reactions Chan studies as an area in which quantum computers should excel. Chan, however, has long doubted that powerful quantum computers — which are still years away — will be necessary.
“My main interest is in solving chemical problems. If classical computers are the right tool to do it, we should,” he said. While he believes quantum computers will eventually play an important role in the field, “I don’t see why we should wait for a fault-tolerant quantum computer to be built.”
Now he has a result that strengthens his case.
In early January, Chan and five other quantum chemists based out of the California Institute of Technology reached a key milestone in understanding the enzyme nitrogenase, which converts atmospheric nitrogen into ammonia and makes life on our planet possible. It was a major triumph for theoretical chemists, the outcome of decades of effort.
But for years, nitrogenase had also served as a proof-of-concept target in the realm of quantum computing. To understand the enzyme, researchers must follow the behavior of many electrons that are all linked together via quantum entanglement. The number of possible configurations grows explosively large. Researchers hypothesized that they would likely only be able to decipher the system via a machine that could manipulate quantum states.
But Chan and his colleagues used purely classical methods. That makes their result a pivotal statement not only about the chemistry that supports life, but also about whether quantum computers are needed to understand it.
“I think it’s important to clarify that this is not an impossible task where you have to first build a quantum computer to say anything about the problem,” Chan said.
Not everyone agrees. Some researchers cite the many years it took to obtain the result classically. Even if one chemistry problem has ultimately proved tractable with classical methods, they say, quantum computers are still needed to make these kinds of discoveries at scale.
“If we pick any optimization problem and you put 20 years into it, you can figure out that one system,” said James Whitfield, a quantum computing theorist at Dartmouth College. “But whether that solution is transferable? Questions like that won’t be answered by solving one instance of one molecular system.”
Solving this particular problem about nitrogenase may not settle the debate over quantum computers just yet, but each step toward understanding the enzyme’s full chemistry makes the debate less hypothetical.
Nature’s Ammonia Factory
Alongside photosynthesis, nitrogen fixation is one of the most essential chemical processes for life on Earth. Nitrogenase is what makes it possible.
Before nitrogenase evolved, living things were limited by the amount of nitrogen available to be incorporated into organic matter. It was an ironic obstacle, given that the planet was in fact suffused with nitrogen: The element accounts for about 80% of the atmosphere. But atmospheric nitrogen exists as the diatomic molecule N2, which is inert and therefore unusable in biological processes. Only rare high-energy events could break the molecule into nitrates that life could use.
“Organisms were literally waiting for lightning to strike. That’s how you’d get nitrogen to be available for biomass,” said Daniel Suess, a chemist at the Massachusetts Institute of Technology who studies nitrogenase.
But 3 billion years ago, the nitrogen floodgates opened when nitrogenase evolved in early prokaryotes. The enzyme accomplished what no other biological process could do: It broke the triple bond holding N2 together and converted the inert gas into biologically useful ammonia.
The enzyme was effective but extraordinarily complicated. That was beside the point to the early microbes that benefited from it, but it would come to matter enormously to the humans who, billions of years later, wanted to replicate its trick in order to make fertilizer.
Part of what makes nitrogenase so chemically difficult is its “active site” — a cluster of iron and molybdenum atoms called FeMo-co. Each iron atom carries four or five unpaired electrons whose behavior depends on that of the others. In fact, FeMo-co is one of the most correlated systems in all of biology, and a prime example of what’s known as the electron correlation problem: Because its electrons can’t be treated independently, it’s extremely hard to determine properties of the overall system, such as its true electronic structure and energy.
For most of human history, the pressing question wasn’t how nitrogenase worked — it was how to get enough of what it produced. As late as the 19th century, the most reliable source of usable nitrogen was guano harvested from islands off the coast of Peru, a resource so valuable and rare that nations went to war over it. Then the German chemists Fritz Haber and Carl Bosch cracked industrial nitrogen fixation in 1909, and the practical significance of the problem receded.
The scientific one — understanding how nitrogenase, tucked inside an ordinary soil bacterium, accomplishes what the Haber-Bosch process requires an industrial furnace to do — remained open.
It was an important question in its own right — and one that would achieve new prominence as people debated the best way to solve it.
An Unlikely Test
A classical computer processes information as bits, which take one of two values: either 0 or 1. A quantum computer instead uses qubits, which can exist in a superposition of 0 and 1 simultaneously and can become entangled with one another in ways that have no classical analogue. That means that when (or if) a large-scale quantum computer exists, it will be able to explore many possible solutions to a problem at once, rather than grinding through them in sequence.
For certain kinds of problems with the right mathematical structure, this promises an exponential speedup over anything a classical machine could achieve. The question, ever since quantum computing took off as a subject of theoretical study in the 1990s, has been which problems qualify. One of the most promising domains seems to be simulating chemical interactions: The electron interactions that govern how molecules behave are quantum mechanical at their core, which suggests that a quantum computer might be uniquely suited to modeling them.
The status of nitrogenase as an informal quantum computing benchmark traces back to a 2011 meeting Microsoft organized to explore applications for its nascent quantum computing group. Chan, who’d already been studying nitrogenase for more than a decade at the time, gave a talk on the enzyme.
He doesn’t know to what extent that talk influenced later events, but in 2017, Microsoft researchers published a paper in the Proceedings of the National Academy of Sciences arguing that the entangled complexity of nitrogenase made it a compelling test for quantum computers.
In Chan’s view, it was a strange fit from the start. He disputed the claim, continuing to believe that it was possible to model nitrogenase using classical methods like the ones he’d spent his career developing.
Over the next decade, he would get to work proving it.
Ground-State Debates
Chan and other researchers didn’t set out to explain how nitrogenase works end to end. Rather, they turned to a widely used computational model of FeMo-co and asked a more preliminary question: What is its ground-state energy?
The ground state — FeMo-co’s lowest-energy electronic configuration — is the starting point for the whole reaction. But FeMo-co contains a cluster of seven iron atoms, each with four or five unpaired electrons whose quantum “spins” can point up or down, whose orbitals can shift, and whose behavior depends on what the electrons around them are doing.
This makes measuring FeMo-co’s ground-state energy extraordinarily complex. There are more than 78,000 plausible configurations the electrons might be in; the ground state is a superposition, or a sort of weighted combination, of all these configurations. In principle, the Schrödinger equation tells you how all these different configurations contribute to the ground state and what its overall energy should be. But in practice, solving that equation directly and exactly for a system with as many interacting electrons as FeMo-co has is often impossible.
This is true for both quantum and classical computers. In both cases, you have to start with a simpler approximation of the ground state’s basic structure — an educated guess, often reached only after years of research, about which configurations are contributing the most to the ground state.
Sueddeutsche Zeitung Photo/Alamy
Then, if you’re using a classical computer, you can try to progressively account for other configurations and show that you can safely ignore the huge number of remaining configurations because they don’t add much to the ground-state energy.
On the other hand, in theory, a quantum computer won’t require you to leave configurations out of your final estimate. Instead, the computer can represent your initial guess directly as a quantum state, and then evolve that state forward in time until it naturally reaches the right ground-state structure — allowing you to calculate the energy precisely.
Many researchers think quantum computers are at an advantage here, because the process of classically ruling out insignificant configurations can get prohibitively difficult. Chan and others, however, disagree. For one thing, they argue, quantum computers still encounter the same bottleneck of needing that reasonable initial guess, and there’s no obvious reason why quantum methods should have any advantage at clearing that bottleneck. Moreover, classical techniques have been rapidly maturing.
But for Chan, asserting that quantum computers might not be needed after all was “like trying to resist the ocean tide,” he said.
Sifting Out the Solution
Since receiving his doctorate from the University of Cambridge in 2000, Chan had been developing and refining ways to compress complicated quantum states by focusing only on their most important configurations. He and his team now hoped to apply these approaches to FeMo-co.
They used two different techniques to winnow down the configurations they needed to look at. Using one method, they started with their guess and incrementally adjusted the behavior of small numbers of electrons. They then showed that adjusting larger numbers of electrons didn’t lead to significant energy changes, giving them a clear recipe for which configurations they could ignore and which they couldn’t.
Their second method was the one that Chan had spent his career working on. It involved breaking their initial state into pieces and allowing only a limited amount of information to flow between those pieces. They then showed that they only needed to consider changes in that information flow up to a particular limit. “Realizing that the description could be achieved by ‘simpler’ methods and pushing these methods extremely hard (as the problem is still computationally challenging) was the key,” Chan wrote in an email.
Both methods produced the same energy estimate for FeMo-co’s ground state (and matched what scientists had observed experimentally), giving the researchers confidence that they had found the true ground state.
The Debate Shifts
Chan hopes that the technical breakthroughs his team made can now be extended to model the full nitrogenase enzyme and its reaction. “My hope is that all these people advocating ‘We need to build a quantum computer to solve the nitrogenase problem’ will join this mission now that we have a route to doing it,” he said.
But getting from the ground state to a full mathematical description of the reaction will be far more difficult, involving calculating energies for a whole sequence of intermediate chemical states. “We’re not even close to achieving the holy grail of this,” Suess said. “We’ve still just described the resting state. But the method is promising in that it suggests we can proceed with some confidence.”
It’s also unclear what the result might mean for researchers’ hopes for quantum computing. Whitfield argues that calculating a single ground-state energy value was never where quantum computers were expected to best their classical counterparts. Their likely advantage, he said, instead lies in that next question on the table: modeling how the system evolves over time. That’s likely to showcase how inefficient classical methods can get — and how much more powerful quantum computers can be.
After years of friendly sparring with the quantum computing community, Chan does not expect the new result to change many minds. After all, he said, quantum chemistry simulation via quantum computers still holds great promise: If a quantum computer were to become available tomorrow, he would gladly use it. But he hopes his team’s new result will help correct the misconception that the hardest chemical problems are simply out of reach until quantum hardware arrives.
“Science is self-correcting,” he wrote in an email, “but quite often, the corrections do not receive the same attention as the initial claim, because the field has moved on to other claims.”