立即打开
量子计算取得两大突破,将实现一些“不可能”

量子计算取得两大突破,将实现一些“不可能”

 JEREMY KAHN 2021年10月09日
钢铁制造和金融这两个截然不同的行业或许即将能用量子计算机做一些直到现在还不可能实现的事情。

对于许多行业来说,量子计算正越来越接近于实现其作为革命性技术的潜力。半月前,有两家公司的声明让我们看到,钢铁制造和金融这两个截然不同的行业或许即将能用量子计算机做一些直到现在还不可能实现的事情。

总部位于英国的剑桥量子计算公司(Cambridge Quantum Computing)最近同意与工业巨头霍尼韦尔(Honeywell)的量子计算部门合并,分拆为一家新的上市公司。该公司称,通过与世界领先的钢铁生产商日本新日铁(Nippon Steel Corporation)合作,它们成功模拟了两种不同构型铁晶体的行为。

这种化学模拟十分复杂,传统计算机无法准确实现。因此,新日铁和剑桥量子计算通过互联网访问了IBM量子计算机,利用剑桥量子计算公司开发的专门算法运行了上述模拟。

参与研究的科学家们表示,该技术最终或许可以创造新型钢铁,还有助于解释地球固体铁核中金属可以承受极端高温高压的原理。

同一时期,来自高盛(Goldman Sachs)、IonQ(一家量子计算机制造公司)和QC Ware(一家专注于量子计算算法的初创公司)的研究人员称,他们已经证明,一种支撑金融风险定价的基础数学技术在量子计算机上比在传统计算机上运行得更好也更快。

蒙特卡罗模拟和铁晶体模拟

研究人员此前就已提出理论,认为这种被称为蒙特卡罗模拟的数学方法应该存在“量子优势”。但这是科学家首次在真正的量子计算硬件上使用专门的量子算法,证明了这种优势的存在。

高盛量子研究主管威尔·曾表示,上述实验表明,如果拥有足够强大的量子计算机,金融风险定价可以得到极大优化。

不过,他提醒道,现有的量子计算机还不足以运行大型蒙特卡罗模拟;投行如果需要更好地为复杂的衍生品合约定价,或为资产组合计算隔夜风险价值,则需要运行大型蒙特卡罗模拟,高盛希望量子计算机最终能在这两个领域充分发挥优势。

目前,高盛使用的是传统计算技术为衍生品定价,根据金融工具的复杂程度,计算时间从一秒钟到几分钟不等。但结果可能没有量子计算机那么精确。而且,曾指出,在处理高杠杆的金融产品时,哪怕风险定价只改进了一点点,盈利能力也能出现巨大差异。

对资产组合进行风险评估时,关键在于准确性和计算的时间成本,这种计算十分复杂,需要一个超级计算集群成夜运转。如果计算结果更准确,高盛就可以减少为防范投资组合价值突然下跌而储备的资本。而一台功能强大的量子计算机也许在几分钟内就能得出更准确的答案。

高盛、IonQ和QC Ware进行的小型实验只涉及4个量子处理单元,即4个量子位元,能够执行约100次逻辑操作。曾说,高盛估计,要想在单一复杂衍生品合约的定价上超越传统计算机,需要一台拥有约8000个量子位元、能够执行约5400万次操作的量子计算机。

尽管如此,这项研究仍然意义重大,因为蒙特卡罗模拟可以解决的问题千千万万,无论是确定价格变化的潜在影响,还是创建更有弹性的供应链。该技术对机器学习应用程序也具有重要意义。比如,使用蒙特卡罗模拟进行结果预测时,如存在大量不同可能出现的结果,该模拟可以建立全部可能场景的概率分布图。

为了模拟铁晶体,剑桥量子计算公司和新日铁的科学家使用的是IBM拥有7个量子位元的量子处理器。同样,研究人员也注意到,想要更精确地模拟铁晶体的能态,需要比现有量子计算机更强大、更不容易出错的量子设备。

关于上述两个实验的论文都发表于非同行评议的研究资源库arxiv.org上。点击相关链接即可阅读。

量子处理能力

理论上,量子计算机的处理能力要比传统计算机强得多,因为它们可以利用量子力学的现象进行计算。在传统计算机中,信息以二进制格式存储,称为一个比特,可以是0,也可以是1。在量子计算机中,量子位元可以以一种“叠加”的状态存在,在这种状态下,它们可以同时表示0和1。传统计算机中,每个比特都是相互独立的。在量子计算机中,一种被称为纠缠的特性使得量子位元可以相互影响,理论上加速了计算时间。

这两篇研究论文中使用的量子计算机运用了不同的方法来产生量子效应。IBM的机器使用了由铌和铝等超导材料制成的量子位元,将其固定在硅芯片上,并冷却到极低的温度。IonQ处理器使用强大的激光捕获来自稀土金属镱的离子,并利用这些离子形成量子位元。

现有量子位元的一个主要问题是,它们只能在相对较短的时间内保持量子态,超导量子位元为120微秒(微秒为百万分之一秒),而捕获离子的维持时间为10分钟。当量子位元脱离量子态时,就会产生错误,必须通过使用更多的量子位元或软件算法来纠正这些错误。

这两个实验都用了一些算法来“减少错误”,以改善运算结果。两个实验使用的系统中,都是部分计算在传统半导体计算机芯片上运行,部分在量子处理器上运行。(财富中文网)

译者:Agatha

对于许多行业来说,量子计算正越来越接近于实现其作为革命性技术的潜力。半月前,有两家公司的声明让我们看到,钢铁制造和金融这两个截然不同的行业或许即将能用量子计算机做一些直到现在还不可能实现的事情。

总部位于英国的剑桥量子计算公司(Cambridge Quantum Computing)最近同意与工业巨头霍尼韦尔(Honeywell)的量子计算部门合并,分拆为一家新的上市公司。该公司称,通过与世界领先的钢铁生产商日本新日铁(Nippon Steel Corporation)合作,它们成功模拟了两种不同构型铁晶体的行为。

这种化学模拟十分复杂,传统计算机无法准确实现。因此,新日铁和剑桥量子计算通过互联网访问了IBM量子计算机,利用剑桥量子计算公司开发的专门算法运行了上述模拟。

参与研究的科学家们表示,该技术最终或许可以创造新型钢铁,还有助于解释地球固体铁核中金属可以承受极端高温高压的原理。

同一时期,来自高盛(Goldman Sachs)、IonQ(一家量子计算机制造公司)和QC Ware(一家专注于量子计算算法的初创公司)的研究人员称,他们已经证明,一种支撑金融风险定价的基础数学技术在量子计算机上比在传统计算机上运行得更好也更快。

蒙特卡罗模拟和铁晶体模拟

研究人员此前就已提出理论,认为这种被称为蒙特卡罗模拟的数学方法应该存在“量子优势”。但这是科学家首次在真正的量子计算硬件上使用专门的量子算法,证明了这种优势的存在。

高盛量子研究主管威尔·曾表示,上述实验表明,如果拥有足够强大的量子计算机,金融风险定价可以得到极大优化。

不过,他提醒道,现有的量子计算机还不足以运行大型蒙特卡罗模拟;投行如果需要更好地为复杂的衍生品合约定价,或为资产组合计算隔夜风险价值,则需要运行大型蒙特卡罗模拟,高盛希望量子计算机最终能在这两个领域充分发挥优势。

目前,高盛使用的是传统计算技术为衍生品定价,根据金融工具的复杂程度,计算时间从一秒钟到几分钟不等。但结果可能没有量子计算机那么精确。而且,曾指出,在处理高杠杆的金融产品时,哪怕风险定价只改进了一点点,盈利能力也能出现巨大差异。

对资产组合进行风险评估时,关键在于准确性和计算的时间成本,这种计算十分复杂,需要一个超级计算集群成夜运转。如果计算结果更准确,高盛就可以减少为防范投资组合价值突然下跌而储备的资本。而一台功能强大的量子计算机也许在几分钟内就能得出更准确的答案。

高盛、IonQ和QC Ware进行的小型实验只涉及4个量子处理单元,即4个量子位元,能够执行约100次逻辑操作。曾说,高盛估计,要想在单一复杂衍生品合约的定价上超越传统计算机,需要一台拥有约8000个量子位元、能够执行约5400万次操作的量子计算机。

尽管如此,这项研究仍然意义重大,因为蒙特卡罗模拟可以解决的问题千千万万,无论是确定价格变化的潜在影响,还是创建更有弹性的供应链。该技术对机器学习应用程序也具有重要意义。比如,使用蒙特卡罗模拟进行结果预测时,如存在大量不同可能出现的结果,该模拟可以建立全部可能场景的概率分布图。

为了模拟铁晶体,剑桥量子计算公司和新日铁的科学家使用的是IBM拥有7个量子位元的量子处理器。同样,研究人员也注意到,想要更精确地模拟铁晶体的能态,需要比现有量子计算机更强大、更不容易出错的量子设备。

关于上述两个实验的论文都发表于非同行评议的研究资源库arxiv.org上。点击相关链接即可阅读。

量子处理能力

理论上,量子计算机的处理能力要比传统计算机强得多,因为它们可以利用量子力学的现象进行计算。在传统计算机中,信息以二进制格式存储,称为一个比特,可以是0,也可以是1。在量子计算机中,量子位元可以以一种“叠加”的状态存在,在这种状态下,它们可以同时表示0和1。传统计算机中,每个比特都是相互独立的。在量子计算机中,一种被称为纠缠的特性使得量子位元可以相互影响,理论上加速了计算时间。

这两篇研究论文中使用的量子计算机运用了不同的方法来产生量子效应。IBM的机器使用了由铌和铝等超导材料制成的量子位元,将其固定在硅芯片上,并冷却到极低的温度。IonQ处理器使用强大的激光捕获来自稀土金属镱的离子,并利用这些离子形成量子位元。

现有量子位元的一个主要问题是,它们只能在相对较短的时间内保持量子态,超导量子位元为120微秒(微秒为百万分之一秒),而捕获离子的维持时间为10分钟。当量子位元脱离量子态时,就会产生错误,必须通过使用更多的量子位元或软件算法来纠正这些错误。

这两个实验都用了一些算法来“减少错误”,以改善运算结果。两个实验使用的系统中,都是部分计算在传统半导体计算机芯片上运行,部分在量子处理器上运行。(财富中文网)

译者:Agatha

Quantum computing is getting ever closer to realizing its potential as a transformative technology for many businesses. This past week a pair of announcements provided a glimpse of how two diverse sectors, steel manufacturing and finance, may be on the cusp of being able to do things with quantum computers that were until now impossible.

Cambridge Quantum Computing, a U.K.-based company that recently agreed to merge with the quantum computing arm of industrial giant Honeywell and spin out as a new publicly traded company, said it had worked with Japan’s Nippon Steel Corporation, one of the world’s leading steel producers, to simulate the behavior of iron crystals in two different configurations.

This chemical simulation is so complex scientists cannot perform it accurately on a conventional computer. In this case, Nippon Steel and Cambridge Quantum Computing used an IBM quantum computer, accessed over the Internet, and specialized algorithms, developed by Cambridge Quantum Computing, to run the simulation.

Scientists involved in the research said the techniques could eventually aid in the creation of new types of steel as well as help answer fundamental questions about what happens in the earth’s solid iron core, where the metal is subjected to extreme heat and pressure.

Also on Tuesday, researchers from Goldman Sachs, IonQ (a company that builds quantum computers), and QC Ware, a startup that specializes in quantum computing algorithms, said they had demonstrated how a fundamental mathematical technique that underpins the pricing of financial risk can be run better and faster on a quantum computer than on conventional ones.

Monte Carlo and iron crystal simulation

Researchers had previously theorized this kind of “quantum advantage” should exist for this mathematical method, called a Monte Carlo simulation. But this is the first time that scientists have demonstrated clear evidence of this improved performance using a specialized quantum algorithm on real quantum computing hardware.

Will Zeng, the head of quantum research at Goldman Sachs, said that the experiment was able to show that with a sufficiently powerful enough quantum computer, there should be a significant performance improvement in pricing financial risk.

He cautioned, however, that current quantum computers are not powerful enough to run the large Monte Carlo simulations the investment bank would need to better price complex derivative contracts or calculate overnight value-at-risk calculations for asset portfolios, two areas in which Goldman hopes quantum computers will eventually offer a major advantage.

Currently, Goldman uses conventional computing techniques to price derivatives, with a calculation taking anywhere from less than a second to a several minutes, depending on the financial instrument’s complexity. But the results may not be as accurate as what can be achieved with a quantum computer. And, as Zeng notes, when dealing with financial products that are highly leveraged, even a small percentage improvement in risk pricing can result in a huge difference in profitability.

In the case of valuing the risk of an entire asset portfolio, the issue is both accuracy and the cost of computing time—the calculations are so complex that it literally takes a supercomputing cluster all night to run them. The more accurate the result of the calculation, potentially the less capital Goldman needs to hold in reserve to guard against sudden drops in the value of its portfolio. A powerful quantum computer might be able to achieve more accurate answers in just minutes.

The small experiment Goldman, IonQ, and QC Ware conducted involved just four quantum processing units, known as qubits, with the ability to carry out about 100 logical operations. Zeng said that Goldman has estimated that outperforming a conventional computer in pricing a single complex derivatives contract would require a quantum computer with about 8,000 qubits and the ability to carry out about 54 million operations.

The research is nonetheless significant because of the vast array of problems that can be addressed using Monte Carlo simulations, from determining the potential effects of price changes to creating more resilient supply chains. The technique is also important for many machine-learning applications. Used in cases where there are many different possible outcomes, a Monte Carlo simulation builds up a picture of the probability distribution of the possible scenarios.

For the simulation of the iron crystals, the Cambridge Quantum Computing and Nippon Steel scientists used a IBM quantum processor with seven qubits. Here, too, the researchers noted that achieving a more accurate simulation of the energy states of the iron crystals would require a much more powerful, and less error prone, quantum device than what currently exists.

Papers about both experiments were published on the non–peer reviewed research repository arxiv.org. You can see the finance research here and the quantum chemistry research paper here.

Quantum processing power

Quantum computers have theoretically exponentially greater processing power than conventional computers because they harness phenomenon from quantum mechanics to help perform calculations. In a conventional computer, information is stored in a binary format, called a bit, that can be either a 0 or 1. In a quantum computer, qubits can exist in a state called superposition, in which they can represent both 0 and 1 simultaneously. In a traditional computer, each bit functions independently. In a quantum computer, a property called entanglement allows qubits to influence one another, in theory speeding up calculation times.

The quantum computers used in the two research papers each utilize a different method to create quantum effects. The IBM machine has qubits made from superconducting materials, such as niobium and aluminum, anchored on a silicon chip, and cooled to extremely low temperatures. The IonQ processor uses powerful lasers to trap ions from a rare earth metal, ytterbium, and uses these to form its qubits.

A major problem with today’s qubits is that they can be held in a quantum state only for a relatively short period of time, ranging from about 120 microseconds (or millionths of a second) for superconducting qubits, to up to 10 minutes for trapped ions. And when the qubits fall out of a quantum state, they produce errors that must then be corrected, either by using more qubits or by using software algorithms.

Both experiments employed some of these “error reduction” algorithms to try to improve the results. And both involved systems in which part of the calculation is run on conventional semiconductor computer chips and some on quantum processors.

最新:
  • 热读文章
  • 热门视频
活动
扫码打开财富Plus App