English

Computation is physical. This is easy to forget, because the theory of computation is so thoroughly mathematical that computation can seem like an abstract activity that happens in a Platonic realm of symbols. But every actual computation is performed by a physical system — electrons moving through silicon, or in principle any physical process that can be made to represent and transform information — and the laws of physics therefore constrain what computation can do and what it costs. The relationship between computer science and physics is built on this fact: information is not an abstraction floating free of the physical world but a physical quantity, subject to physical law, and the deepest connections between the two fields follow from taking the physicality of information seriously.

The relationship runs in both directions, and both directions are profound. Physics constrains computation: there are thermodynamic limits on how little energy a computation can use, physical limits on how fast and how small computers can be, and the strange resource of quantum mechanics that enables forms of computation impossible classically. And computation illuminates physics: information has turned out to be a concept of deep physical significance, entropy connects thermodynamics and information theory at a level that suggests they are aspects of one thing, and the idea that physical processes can be understood as computation has become a productive lens in physics itself. The boundary between the two fields is where some of the deepest questions about the nature of information, computation, and physical reality live.

This section examines that boundary: the physicality of information, the thermodynamics of computation, and quantum information as the fusion of the two fields. It connects to information theory (§3.8), which provides the mathematical concept of information, and to quantum computing (§6.6), which provides the technology; here the concern is the relationship — what it means that information is physical, and what that relationship reveals about both computation and the physical world.

Background: Information theory (§3.8) supplies the formal concepts of information and entropy; quantum computing (§6.6) supplies the computational side of quantum information; thermodynamics, statistical mechanics, and quantum mechanics supply the physical context.

The Relationship: Information as a Physical Quantity

The Physicality of Information and Landauer’s Principle

The foundational insight of the relationship is that information is physical — a slogan associated with Rolf Landauer, the IBM physicist who established it most concretely. The claim is not metaphorical. Information must be represented in some physical system (a configuration of matter or energy), and operations on information are physical processes, so the physics of those systems and processes constrains what can be done with information and at what cost.

Landauer’s principle (1961) makes this precise and surprising: erasing one bit of information has a minimum thermodynamic cost. Specifically, erasing a bit requires dissipating at least kT ln 2 of energy as heat, where k is Boltzmann’s constant and T is the temperature. The reason connects information to thermodynamics at a deep level: erasure is a logically irreversible operation (two distinct states, 0 and 1, both map to the single erased state), and reducing the number of possible states reduces entropy in the information-bearing system, which by the second law of thermodynamics must be compensated by an increase in entropy elsewhere — the dissipated heat. The minimum cost of erasing a bit is set by the laws of thermodynamics, not by the engineering of the device. This was confirmed experimentally in 2012, fifty years after Landauer predicted it.

The principle has a striking consequence explored by Charles Bennett: logically reversible computation — computation that does not erase information, that can in principle be run backward — has no thermodynamic lower bound on its energy cost. If you never erase information, you never pay Landauer’s cost. This suggested that the energy cost of computation is fundamentally about the erasure of information, not about computation itself, and that computation could in principle be made arbitrarily energy-efficient by making it reversible. Reversible computing — designing computations that do not destroy information — became a research direction motivated by this insight, and it connects to quantum computing, where the operations are necessarily reversible (unitary).

Landauer’s principle also resolved a long-standing puzzle in physics: Maxwell’s demon, the hypothetical being that seemed able to violate the second law of thermodynamics by sorting fast and slow molecules using information about their velocities. The resolution, completed by Bennett building on Landauer, is that the demon must store information about the molecules to sort them, and erasing that information to continue operating costs exactly enough energy (by Landauer’s principle) to save the second law. The demon does not get something for nothing; it pays, in the thermodynamic cost of erasing the information it must gather. This resolution — that the information the demon uses has a thermodynamic cost — is one of the clearest demonstrations that information is a physical, thermodynamic quantity, and it was reached by taking the physicality of information seriously.

Entropy: The Concept That Unifies

The deepest connection between the fields is entropy, a single concept that appears in thermodynamics, in statistical mechanics, and in information theory, in forms that turned out to be the same thing. That the entropy of physics and the entropy of information theory are connected — that they are, at a deep level, one concept — is among the most remarkable unifications in science, and it sits exactly on the boundary between physics and computer science.

Thermodynamic entropy was defined in the nineteenth century by Clausius and given a statistical interpretation by Boltzmann: the entropy of a physical system is (proportional to) the logarithm of the number of microscopic states consistent with its macroscopic description. A gas with more accessible microstates has higher entropy. Boltzmann’s formula, S = k log W (W being the number of microstates), engraved on his tombstone, defines entropy as a count of possibilities.

Information entropy was defined by Shannon in 1948 (§3.8) as a measure of the uncertainty in a random variable, or equivalently the average information content of its outcomes: H = −Σ p log p. Shannon’s entropy measures, in bits, how much information is needed on average to specify an outcome, or how much uncertainty there is about it. Shannon arrived at this formula from the requirements of communication theory, with no reference to physics, and the story (possibly apocryphal) is that von Neumann suggested he call it “entropy” partly because the formula matched Boltzmann’s and partly because “no one knows what entropy really is, so in a debate you will always have the advantage.”

That the two entropies have the same mathematical form is not a coincidence: they are the same concept, applied in different contexts. Statistical mechanics can be reformulated entirely in information-theoretic terms (Jaynes’s maximum entropy formulation, 1957), in which thermodynamic entropy is the Shannon entropy of the probability distribution over microstates, and the laws of thermodynamics become statements about information. In this view, the second law of thermodynamics — entropy increases — is a statement about information: an isolated system’s microstate becomes progressively less determined by its initial macroscopic description as it evolves, which is an increase in the information needed to specify it. The unification of thermodynamic and information entropy is the deepest single connection between physics and computer science, and it suggests that information is not merely analogous to a physical quantity but is one.

Quantum Information: The Fusion of the Fields

Quantum information is where physics and computer science fuse most completely, into a discipline that is simultaneously both and neither. Quantum mechanics is a physical theory; quantum information and quantum computation (§6.6) take the physical theory and ask what it means for information and computation — and the answers reveal that the physical theory permits forms of information processing that classical physics does not.

The qubit — the quantum unit of information — is a physical system (a superposition of states), and its properties as an information carrier follow from quantum mechanics. Superposition allows a qubit to be in a combination of 0 and 1; entanglement allows qubits to have correlations with no classical analog; the no-cloning theorem (quantum information cannot be copied) is a consequence of the linearity of quantum mechanics that has no classical counterpart and profound implications for what can be done with quantum information. These are physical facts — features of how quantum systems behave — that are simultaneously facts about information, because the quantum systems are information carriers. The field studies them as both at once.

The computational consequences (the quantum algorithms of §6.6, Shor’s and Grover’s and the rest) are where quantum information becomes quantum computation, and they show that the physical theory permits computation that classical physics does not. That a quantum computer can factor integers efficiently while no classical computer is known to, is a statement about both physics (quantum systems can be made to compute this way) and computer science (the complexity of factoring on a quantum machine differs from its complexity on a classical one). The quantum complexity classes (BQP and the rest, §6.6) are computer science defined relative to a physical theory; they classify problems by their difficulty on a machine that exploits quantum physics. The boundary between the fields has dissolved here: quantum information is irreducibly both physics and computer science, and the questions it asks (what can be computed, what can be communicated, what can be secured, given the actual laws of physics) belong to both.

The connection runs deeper still, into foundational physics. The idea that information is fundamental to physical reality — Wheeler’s “it from bit,” the proposal that physical existence is at bottom informational — is a speculative but serious position in the foundations of physics, motivated partly by the role information has turned out to play. The black hole information paradox (whether information that falls into a black hole is destroyed, which quantum mechanics forbids) is a central problem in theoretical physics that is fundamentally about information. The holographic principle and the relationship between entanglement and spacetime geometry in recent theoretical physics treat information as a fundamental physical quantity. These are physics, but they are physics in which information — the computer scientist’s central concept — has become fundamental, which is the dialogue between the fields at its deepest and most speculative.

What This Perspective Changes

Seeing the relationship between computer science and physics changes how a practitioner understands both information and computation.

The first change is understanding that information is physical, with the consequences that follow. The practitioner who grasps Landauer’s principle understands that computation has a thermodynamic cost grounded in physical law, that this cost is about the erasure of information specifically, and that the energy consumption of computing — a major practical concern as computing scales — is constrained by physics in ways that connect to the deepest principles of thermodynamics. Information is not free and not abstract; it is a physical quantity with physical costs, and seeing this connects the practical concern of energy-efficient computing to fundamental physics.

The second change is appreciation of entropy as a unifying concept. The practitioner who sees that the entropy of thermodynamics and the entropy of information theory are the same concept has access to one of the deepest unifications in science, and it changes how they understand both. Information-theoretic reasoning applies to physical systems; thermodynamic reasoning applies to information; and the second law of thermodynamics is, in a precise sense, a statement about information. This connection enriches the understanding of entropy in both its information-theoretic use (§3.8) and its physical meaning.

The third change is the correct framing of quantum computing. The practitioner who understands quantum information as the fusion of physics and computer science sees quantum computing correctly: not as a faster classical computer but as computation that exploits the actual laws of physics, which permit information processing that classical physics does not. This framing clarifies what quantum computing can and cannot do (it exploits specific quantum phenomena for specific problems, §6.6) and why it is a genuinely different kind of computation rather than merely a faster one.

The fourth change is a glimpse of the deepest questions. The practitioner who sees that information has become fundamental to physics — in the black hole information paradox, in “it from bit,” in the relationship between entanglement and spacetime — glimpses questions at the frontier of both fields, where the nature of information, computation, and physical reality are entangled in ways that neither field can address alone. These are speculative frontiers, but seeing them reveals that the relationship between computation and physics reaches to the foundations of what reality is.

Resources

Foundational and Bridging Texts

Feynman’s Feynman Lectures on Computation (Addison-Wesley, 1996, based on lectures from the 1980s) is the bridge text: the great physicist’s lectures on the physics of computation, covering the thermodynamics of computing, reversible computation, the limits set by physics, and an early treatment of quantum computation. Feynman approached computation as a physicist, and the lectures show what it means to take the physicality of computation seriously, from someone who helped initiate the field of quantum computing. It is the single best entry to the relationship.

Charles Bennett’s and Rolf Landauer’s papers — Landauer’s “Irreversibility and Heat Generation in the Computing Process” (1961) and Bennett’s work on reversible computation and the resolution of Maxwell’s demon (“The Thermodynamics of Computation—A Review,” 1982) — are the primary sources for the thermodynamics of information, and they are readable by anyone with the relevant background. Bennett’s review article is the best single source on the thermodynamics of computation.

For entropy as the unifying concept, the information-theoretic formulation of statistical mechanics — Jaynes’s papers (“Information Theory and Statistical Mechanics,” 1957, free) — establish the connection at its source. Cover and Thomas’s Elements of Information Theory (§3.8 reference) covers the information-theoretic side with the rigor that makes the connection precise.

Resource	Role	Type
Feynman, Feynman Lectures on Computation	The bridge text; physics of computation	Depth
Landauer, “Irreversibility and Heat Generation” (1961)	Landauer’s principle; foundational paper	Depth
Bennett, “The Thermodynamics of Computation” (1982)	Reversible computation; Maxwell’s demon resolved	Depth
Jaynes, “Information Theory and Statistical Mechanics” (1957, free)	Entropy unification at its source	Depth
MacKay, Information Theory, Inference, and Learning Algorithms (free)	Information theory, physics, inference, and learning bridge	Depth
SEP, “Computation in Physical Systems” (free)	Physical realization of computation	Reference

Quantum Information and the Deeper Connections

Nielsen and Chuang’s Quantum Computation and Quantum Information (§6.6 reference) is the canonical text for the fusion of the fields, treating quantum information as both physics and computer science with the rigor both demand. Its early chapters establish quantum information theory; the relationship between the physical theory and the information-theoretic consequences is developed throughout.

For the deepest and most speculative connections — information in foundational physics — the literature is largely in physics rather than in accessible books, but Leonard Susskind’s writing and lectures on the black hole information paradox and the holographic principle (his The Black Hole War, 2008, for the accessible account) convey what it means for information to have become fundamental to physics. Wheeler’s “it from bit” essay (“Information, Physics, Quantum: The Search for Links,” 1989) is the primary source for the proposal that information is physically fundamental.

Seth Lloyd’s Programming the Universe (2006) presents, accessibly and speculatively, the view of the universe as a quantum computer — a perspective that takes the computational view of physics to its limit. It is a popular treatment of a serious idea, useful for seeing how far the computational lens on physics can be pushed.

Resource	Role	Type
Nielsen & Chuang, Quantum Computation and Quantum Information	The fused field, rigorously	Depth
Susskind, The Black Hole War	Information in foundational physics, accessibly	Depth
Wheeler, “Information, Physics, Quantum” (1989)	“It from bit”; foundational essay	Depth
Lloyd, Programming the Universe	The universe as computer; speculative	Auxiliary

Traps

Trap	Why it misleads	Better response
Treating information as purely abstract	The mathematical theory of computation and information is so abstract that it is easy to think of information as a Platonic abstraction with no physical character — bits as pure mathematical objects rather than physical states. This misses the deepest insight of the relationship: information is physical, subject to thermodynamic law, with physical costs (Landauer’s principle) and physical limits. The abstract view cannot explain why erasing a bit costs energy or why computation has thermodynamic limits.	Take the physicality of information seriously. Information must be physically represented, operations on it are physical processes, and physics constrains what can be done with it and at what cost. Landauer’s principle is the entry point: understanding why erasing a bit has a minimum thermodynamic cost reveals that information is a physical, thermodynamic quantity, not a free abstraction.
Confusing the two entropies as merely analogous	The entropy of thermodynamics and the entropy of information theory have the same mathematical form, and it is tempting to treat this as a suggestive analogy rather than an identity. This undersells the connection: they are, at a deep level, the same concept, and statistical mechanics can be reformulated entirely in information-theoretic terms. Treating them as merely analogous misses one of the deepest unifications in science.	Understand the connection as identity rather than analogy. The Jaynes formulation shows that thermodynamic entropy is the Shannon entropy of the distribution over microstates, and the second law becomes a statement about information. The two entropies are one concept applied in two contexts, which is why information-theoretic reasoning applies to physical systems and thermodynamic reasoning applies to information.
Thinking quantum computing is just faster classical computing	The framing of quantum computers as faster classical computers misunderstands the relationship between physics and computation that quantum information embodies. A quantum computer is not a faster classical machine; it is computation that exploits quantum physics — superposition, entanglement, interference — which classical physics does not provide, enabling specific computations (not all computations) that have no efficient classical equivalent.	Understand quantum computing as computation that exploits the actual laws of physics, which permit information processing classical physics does not. This framing (developed in §6.6) clarifies that quantum advantage is specific to problems that can exploit the quantum phenomena, not a general speedup, and that quantum computing is a genuinely different kind of computation grounded in a different physics, not a faster version of the same thing.

中文

计算是物理的。这一点很容易被忘记，因为计算理论已经被数学化得如此彻底，以至于计算看起来像是一种发生在柏拉图式符号领域中的抽象活动。但每一次实际计算，都由一个物理系统执行——电子在硅中运动，或者原则上任何能够被用来表示和变换信息的物理过程——因此，物理定律会约束计算能做什么，以及它的代价是什么。计算机科学与物理学之间的关系就建立在这个事实之上：信息不是脱离物理世界而漂浮的抽象物，而是一个物理量，受物理定律支配；两个领域之间最深的连接，都来自认真对待信息的物理性。

这种关系是双向的，而且两个方向都很深刻。物理约束计算：计算能消耗多低能量存在热力学极限，计算机能有多快、多小也存在物理极限，而量子力学这种奇异资源使某些经典上不可能的计算形式成为可能。反过来，计算也照亮物理：信息已经被证明是具有深层物理意义的概念，熵在热力学和信息论之间建立了连接，其深度暗示二者可能是一件事的两个方面；而物理过程可以被理解为计算这一想法，也已经成为物理学自身中一种有成效的视角。两个领域的边界，正是关于信息、计算和物理现实本性的某些最深问题所在之处。

本节考察的就是这个边界：信息的物理性、计算的热力学，以及作为两个领域融合的量子信息。它连接到信息论（§3.8），后者提供信息的数学概念；也连接到量子计算（§6.6），后者提供相关技术。这里关心的是二者之间的关系——信息是物理的这件事意味着什么，以及这种关系揭示了计算和物理世界的什么。

背景知识：信息论（§3.8）提供信息和熵的形式概念；量子计算（§6.6）提供量子信息的计算侧；热力学、统计力学和量子力学提供物理语境。

这种关系：信息作为物理量

信息的物理性与 Landauer 原理

这种关系的基础洞见是：信息是物理的。这一口号与 IBM 物理学家 Rolf Landauer 联系在一起，他最具体地确立了这一点。这个主张不是隐喻。信息必须被表示在某个物理系统中，也就是物质或能量的某种构型中；而对信息的操作都是物理过程，因此这些系统和过程的物理性质，会约束我们能对信息做什么，以及以什么代价去做。

Landauer 原理（1961）以一种精确且令人惊讶的方式说明了这一点：擦除一 bit 信息具有最低热力学成本。具体来说，擦除一 bit 至少需要以热的形式耗散 kT ln 2 的能量，其中 k 是 Boltzmann 常数，T 是温度。原因在深层上把信息与热力学连接起来：擦除是逻辑不可逆操作（两个不同状态 0 和 1 都映射到同一个被擦除状态），而减少可能状态的数量，会降低承载信息系统中的熵；根据热力学第二定律，这必须由其他地方的熵增加来补偿——也就是耗散出的热。擦除一 bit 的最低成本由热力学定律决定，而不是由设备工程决定。Landauer 预测五十年后，这一点在 2012 年得到了实验证实。

这一原理产生了一个由 Charles Bennett 探索出的惊人后果：逻辑可逆计算——不擦除信息、原则上可以反向运行的计算——在能量成本上没有热力学下界。如果你从不擦除信息，就不需要支付 Landauer 成本。这表明，计算的能量成本根本上不是关于计算本身，而是关于信息擦除；原则上，计算可以通过使其可逆而变得任意节能。可逆计算——设计不破坏信息的计算——正是由这一洞见推动而成为研究方向，并且它连接到量子计算，因为量子计算中的操作必然是可逆的（酉的）。

Landauer 原理还解决了物理学中的一个长期谜题：Maxwell demon。这个假想存在似乎能够通过利用分子速度信息来分拣快慢分子，从而违反热力学第二定律。建立在 Landauer 基础上的 Bennett 最终完成了解决：这个 demon 必须存储关于分子的信息，才能对它们进行分拣；而要持续运行，它必须擦除这些信息；根据 Landauer 原理，擦除这些信息所花费的能量，恰好足以拯救第二定律。这个 demon 并没有无中生有地获得什么；它付出了代价，代价就是擦除它必须收集的信息所需的热力学成本。这个解决方案——demon 使用的信息具有热力学成本——是信息作为物理量、热力学量的最清楚展示之一，而它正是通过认真对待信息的物理性达成的。

熵：统一性的概念

两个领域之间最深的连接是熵。熵是一个同时出现在热力学、统计力学和信息论中的概念，而这些形式最终被证明是同一件事。物理中的熵与信息论中的熵是相互连接的——更深地说，它们是同一个概念——这是科学中最惊人的统一之一，而它正位于物理学和计算机科学的边界上。

热力学熵由 Clausius 在十九世纪定义，并由 Boltzmann 赋予统计解释：一个物理系统的熵，与符合其宏观描述的微观状态数量的对数成正比。一个拥有更多可达微观状态的气体，熵更高。Boltzmann 的公式 S = k log W（W 是微观状态数量）刻在他的墓碑上，它把熵定义为对可能性的计数。

信息熵由 Shannon 在 1948 年定义（§3.8），用来度量随机变量中的不确定性，或者等价地说，度量其结果的平均信息含量：H = −Σ p log p。Shannon 熵以 bit 为单位，度量平均需要多少信息才能指定一个结果，或者结果中有多少不确定性。Shannon 是从通信理论的要求出发得到这个公式的，并没有参考物理学。一个可能是轶闻的说法是，von Neumann 建议他把它称为 “entropy”，部分是因为这个公式与 Boltzmann 的公式相匹配，部分是因为“没人真正知道熵是什么，所以在辩论中你总会占优势”。

这两种熵具有相同数学形式，并不是巧合：它们是同一个概念，被应用在不同语境中。统计力学可以完全用信息论术语重新表述（Jaynes 的最大熵表述，1957），其中热力学熵就是微观状态概率分布的 Shannon 熵，而热力学定律则变成关于信息的陈述。在这个视角下，热力学第二定律——熵增加——就是关于信息的陈述：一个孤立系统的微观状态，随着演化，会越来越少地由其初始宏观描述所决定，因此指定它所需的信息会增加。热力学熵与信息熵的统一，是物理学和计算机科学之间最深的单一连接；它暗示信息不仅仅类似于一个物理量，而就是一个物理量。

量子信息：两个领域的融合

量子信息是物理学和计算机科学最完整融合的地方，形成了一门同时属于二者、又不能还原为其中任何一方的学科。量子力学是一种物理理论；量子信息和量子计算（§6.6）则拿这套物理理论来追问：它对信息和计算意味着什么？答案揭示出，物理理论允许一些经典物理不允许的信息处理形式。

Qubit——量子信息单位——是一个物理系统（状态的叠加），而它作为信息载体的性质来自量子力学。叠加允许一个 qubit 处在 0 和 1 的组合中；纠缠允许 qubit 之间具有经典上没有对应物的关联；不可克隆定理（量子信息不能被复制）是量子力学线性性的结果，没有经典对应物，并且对量子信息能做什么具有深远影响。这些都是物理事实——关于量子系统如何行为的事实——同时也是关于信息的事实，因为这些量子系统本身就是信息载体。这个领域把它们同时作为二者来研究。

计算后果（§6.6 中的量子算法，Shor、Grover 以及其他算法）是量子信息变成量子计算的地方，它们表明这套物理理论允许经典物理不允许的计算。量子计算机能够高效分解整数，而目前没有已知经典计算机可以做到这一点，这既是关于物理的陈述（量子系统可以被构造成以这种方式计算），也是关于计算机科学的陈述（因数分解在量子机器上的复杂度不同于它在经典机器上的复杂度）。量子复杂性类（BQP 及其他，§6.6）是相对于一种物理理论定义的计算机科学；它们按照利用量子物理的机器上的困难程度来分类问题。两个领域的边界在这里已经消解：量子信息不可还原地同时是物理学和计算机科学，它提出的问题——在真实物理定律给定时，什么可以被计算，什么可以被通信，什么可以被保护——属于二者共同的范围。

这种连接还更深地进入基础物理。信息是物理现实的根本，这一思想——Wheeler 的 “it from bit”，即物理存在在底层上是信息性的提议——是物理学基础中一种推测性但严肃的立场，部分动机来自信息所扮演的角色。黑洞信息悖论（落入黑洞的信息是否会被毁灭，而量子力学禁止这种毁灭）是理论物理中的核心问题，并且根本上是关于信息的问题。全息原理，以及近期理论物理中纠缠与时空几何之间的关系，都把信息作为基本物理量来处理。这些都是物理学，但它们是把信息——计算机科学家的核心概念——变成基础概念的物理学；这正是两个领域之间最深、也最具推测性的对话。

这种视角会改变什么

看清计算机科学与物理学之间的关系，会改变实践者理解信息和计算的方式。

第一个变化，是理解信息是物理的，以及由此产生的后果。一个掌握 Landauer 原理的实践者会明白，计算具有扎根于物理定律的热力学成本；这种成本具体关乎信息擦除；而随着计算规模扩大，计算能耗这一重大实践问题，会以某种方式受物理约束，并连接到热力学最深的原则。信息不是免费的，也不是纯抽象的；它是一个具有物理成本的物理量。看见这一点，会把节能计算这一实践关切，与基础物理连接起来。

第二个变化，是理解熵作为统一概念的价值。一个看见热力学熵和信息论熵是同一个概念的实践者，就接触到了科学中最深的统一之一，而这会改变他对二者的理解。信息论推理可以应用于物理系统；热力学推理可以应用于信息；热力学第二定律在精确意义上就是关于信息的陈述。这一连接会丰富人们对熵的理解，无论是在其信息论用法（§3.8）中，还是在其物理意义中。

第三个变化，是正确定位量子计算。一个把量子信息理解为物理学和计算机科学融合的实践者，会正确看待量子计算：它不是更快的经典计算机，而是利用真实物理定律的计算；这些物理定律允许经典物理不允许的信息处理。这一框架会澄清量子计算能做什么、不能做什么（它利用特定量子现象处理特定问题，§6.6），也会说明为什么它是真正不同类型的计算，而不仅仅是更快的计算。

第四个变化，是瞥见最深层的问题。一个看见信息已经在物理学中成为基础概念的实践者——无论是在黑洞信息悖论、“it from bit”，还是纠缠与时空关系中——都会瞥见两个领域前沿的问题。在那里，信息、计算与物理现实的本性彼此纠缠，无法由任何一个领域单独处理。这些是推测性前沿，但看见它们会说明，计算与物理的关系延伸到了“现实是什么”的基础层面。

资源

基础性与桥梁性文本

Feynman 的 Feynman Lectures on Computation（Addison-Wesley，1996，基于 1980 年代讲座）是桥梁文本：这位伟大物理学家关于计算物理学的讲座，覆盖计算的热力学、可逆计算、物理设定的极限，以及对量子计算的早期处理。Feynman 是以物理学家的方式处理计算的，而这些讲座展示了认真对待计算的物理性意味着什么；况且他本人也参与开启了量子计算这一领域。它是进入这种关系的最佳单一入口。

Charles Bennett 和 Rolf Landauer 的论文——Landauer 的 “Irreversibility and Heat Generation in the Computing Process”（1961），以及 Bennett 关于可逆计算和 Maxwell demon 解决方案的工作（“The Thermodynamics of Computation—A Review”，1982）——是信息热力学的一手来源，只要具备相关背景，就可以阅读。Bennett 的综述文章是关于计算热力学的最佳单一来源。

对于作为统一概念的熵，统计力学的信息论表述——Jaynes 的论文（“Information Theory and Statistical Mechanics”，1957，免费）——在源头上建立了这种连接。Cover 和 Thomas 的 Elements of Information Theory（§3.8 参考）则以使这种连接变得精确所需的严谨性，覆盖信息论一侧。

资源	作用	类型
Feynman, Feynman Lectures on Computation	桥梁文本；计算的物理学	深入
Landauer, “Irreversibility and Heat Generation”（1961）	Landauer 原理；基础论文	深入
Bennett, “The Thermodynamics of Computation”（1982）	可逆计算；Maxwell demon 的解决	深入
Jaynes, “Information Theory and Statistical Mechanics”（1957，免费）	熵统一的源头	深入
MacKay, Information Theory, Inference, and Learning Algorithms（免费）	信息论、物理、推断和学习之间的桥梁	深入
SEP, “Computation in Physical Systems”（免费）	计算的物理实现	参考

量子信息与更深层连接

Nielsen 和 Chuang 的 Quantum Computation and Quantum Information（§6.6 参考）是两个领域融合的经典教材，它以物理学和计算机科学都要求的严谨性，处理量子信息。其前几章建立量子信息理论；物理理论与信息论后果之间的关系贯穿全书。

对于最深也最具推测性的连接——基础物理中的信息——相关文献主要在物理学中，而不是在易读书籍中。不过，Leonard Susskind 关于黑洞信息悖论和全息原理的写作与讲座（他的 The Black Hole War，2008，是可进入叙述）能传达“信息已经成为物理学基础概念”意味着什么。Wheeler 的 “it from bit” 论文（“Information, Physics, Quantum: The Search for Links”，1989）是“信息在物理上是根本的”这一提议的一手来源。

Seth Lloyd 的 Programming the Universe（2006）以可进入且带有推测性的方式，呈现了宇宙作为量子计算机的视角——这是一种把物理学的计算视角推到极限的观点。它是一个严肃思想的通俗处理，有助于看见用计算透镜理解物理可以走多远。

资源	作用	类型
Nielsen & Chuang, Quantum Computation and Quantum Information	严谨处理融合领域	深入
Susskind, The Black Hole War	基础物理中的信息，可进入	深入
Wheeler, “Information, Physics, Quantum”（1989）	“It from bit”；基础论文	深入
Lloyd, Programming the Universe	宇宙作为计算机；推测性	辅助

陷阱

陷阱	为什么会误导	更好的回应
把信息当成纯抽象	计算和信息的数学理论如此抽象，以至于人们很容易把信息想成一种没有物理性质的柏拉图式抽象——把 bit 当成纯数学对象，而不是物理状态。这会错过这种关系中最深的洞见：信息是物理的，受热力学定律支配，具有物理成本（Landauer 原理）和物理极限。纯抽象视角无法解释为什么擦除一 bit 需要能量，也无法解释为什么计算具有热力学极限。	认真对待信息的物理性。信息必须被物理表示，对信息的操作是物理过程，而物理学约束我们能对它做什么，以及以什么代价去做。Landauer 原理是入口：理解为什么擦除一 bit 存在最低热力学成本，会揭示信息是物理量、热力学量，而不是免费的抽象物。
把两种熵误认为只是类比	热力学熵和信息论熵具有相同的数学形式，因此人们很容易把这看成一种有启发性的类比，而不是同一性。这会低估这种连接：它们在深层上是同一个概念，而统计力学可以完全以信息论术语重新表述。把二者仅仅看成类比，会错过科学中最深的统一之一。	把这种连接理解为同一性，而不是类比。Jaynes 的表述显示，热力学熵就是微观状态分布上的 Shannon 熵，而第二定律变成关于信息的陈述。这两种熵是同一个概念在两个语境中的应用，因此信息论推理可以应用于物理系统，热力学推理也可以应用于信息。
认为量子计算只是更快的经典计算	把量子计算机描述成更快的经典计算机，会误解量子信息所体现的物理与计算之间的关系。量子计算机不是更快的经典机器；它是利用量子物理——叠加、纠缠、干涉——的计算，而这些资源在经典物理中不存在，从而允许特定计算（不是所有计算）具有经典上没有高效对应物的能力。	把量子计算理解为利用真实物理定律的计算，而这些物理定律允许经典物理不允许的信息处理。这一框架（§6.6 中展开）会澄清量子优势只适用于能够利用量子现象的问题，而不是普遍加速；也会说明量子计算是一种基于不同物理的真正不同计算类型，而不是同一事物的更快版本。