Resource Theories Framework

A resource theory is what you get when you take the most productive sentence in quantum information — "this state is a resource" — and turn it into arithmetic. The move is disarmingly simple. Declare, once and for all, which states and operations are free: cost nothing, available in unlimited supply, ruled out as a source of the thing you care about. Everything the free objects cannot reach is then a resource, and the whole subject becomes an economics of conversion. Two separated labs joined by a telephone but no quantum channel can do any local operation and send any classical message for free; what they cannot do for free is share entanglement, so entanglement is their resource. Fix instead "free = diagonal in a chosen basis" and the resource is coherence — quantum superposition itself. Fix "free = the Gibbs state at temperature T" and the resource is being out of equilibrium, and the framework reproduces thermodynamics. One template, many physics. The power of the abstraction is that a single toolkit serves every instance. Given free states $\mathcal{F}$ and free operations $\mathcal{O}$ obeying the golden rule — a free operation never turns a free state into a resource — three questions organize everything. Interconversion: when can $\rho$ be turned into $\sigma$ for free, exactly, approximately, or asymptotically at some optimal rate? Quantification: which functionals never increase under free operations (the monotones), and which of those track a real operational cost? Structure: is the theory reversible — is a resource poured in fully recoverable — or does it hide bound resources you pay for and can never cash out? The engine that answers all three is the data-processing inequality for the quantum relative entropy: because $D(\rho\|\sigma)$ can only shrink under a channel, the relative-entropy distance from $\rho$ to the free set is automatically a monotone. This article builds that machinery from its axioms and shows why a monotone is a one-line no-go theorem, why relative entropy is the universal currency, and why reversibility pins down a unique measure — a resource-theoretic second law.

Key takeaways

A resource theory is fixed by a set of free states $\mathcal{F}$ and a class of free operations $\mathcal{O}$ obeying the golden rule $\Lambda\in\mathcal{O},\,\rho\in\mathcal{F}\Rightarrow\Lambda[\rho]\in\mathcal{F}$ ; whatever the free objects cannot create is the resource.
A resource monotone is any $M$ that never increases under free operations, $M(\Lambda[\rho])\le M(\rho)$ ; a single monotone with $M(\rho)<M(\sigma)$ certifies that $\rho\to\sigma$ is impossible for free — a no-go proof in one inequality.
The relative entropy of resource $M_{\mathcal{F}}(\rho)=\min_{\sigma\in\mathcal{F}}D(\rho\|\sigma)$ is a universal, faithful monotone, and its monotonicity is inherited directly from the data-processing inequality.
The same skeleton specializes to entanglement (free = separable, LOCC), coherence (free = incoherent), asymmetry (free = symmetric under a group), and thermodynamics (free = Gibbs state, thermal operations), where the monotone becomes a nonequilibrium free energy.
Asymptotically, conversion is governed by rates; distillable resource obeys $E_D\le E_C$ (resource cost), and a strict gap means irreversibility, with bound resources at the extreme $E_D=0<E_C$ .
Reversibility forces a unique measure: when a theory is asymptotically reversible, distillation, cost, and every well-behaved monotone collapse to one number — entropy's role, generalized.
Key idea: Fix what is free; everything else is a resource. Free operations cannot manufacture the resource, so any function non-increasing under them is a currency that upper-bounds what a conversion can produce — and relative-entropy distance to the free set furnishes such a currency universally, via data processing.

The core insight

Fix what is free; everything else is a resource. Free operations cannot create the resource, so any function non-increasing under them — a monotone — bounds how much resource a conversion can produce, and the relative-entropy distance to the free set is a universal monotone handed to you by the data-processing inequality.

Where this lives today

The framework's most consequential modern instance is the resource theory of magic, or non-stabilizerness — and it is not academic, it is the bottleneck of fault-tolerant quantum computing. Clifford operations and stabilizer states are free (they are classically simulable by the Gottesman–Knill theorem), so the resource that must be injected to reach universal quantum computation is exactly magic, supplied through noisy "magic states" like $|T\rangle=\tfrac{1}{\sqrt2}(|0\rangle+e^{i\pi/4}|1\rangle)$ . Magic-state distillation — refining many noisy copies into a few clean ones with stabilizer operations — is a resource distillation protocol in the precise sense of this article, and its overhead dominates the qubit budget of every leading fault-tolerant architecture; magic monotones such as the mana and the stabilizer Rényi entropy now quantify that cost and are measured in current trapped-ion and superconducting experiments. The same machinery is live elsewhere: coherence and asymmetry monotones bound the precision of quantum clocks and metrology below the standard limit; the generalized second laws constrain nanoscale heat engines and quantum batteries where a single free-energy inequality no longer suffices; and the abstract program asks which theories are asymptotically reversible at all — a question with a recent, hard-won general answer for entanglement. A single template, first sharpened on Bell pairs, now sets the price of a fault-tolerant qubit.

The physical picture

Two labs, Alice's and Bob's, share a classical telephone line but no quantum channel. Anything local — a unitary, a measurement, a discarded qubit — is free, and any classical message down the wire is free. What is not free is a shared Bell pair: to have entanglement they must have been given it in advance. Everything Alice and Bob can build for free is a separable state; entanglement is precisely the obstruction, the thing the free operations leave invariant. Swap the rulebook — "free = no superpositions of a fixed basis" — and the obstruction becomes coherence; swap again — "free = invariant under an unknown rotation of a missing reference frame" — and it becomes asymmetry. The picture is always the same: a convex island of free states, and a resource measured by how far outside the island you sit. The first figure draws that island. The free states form a closed convex set $\mathcal{F}$ ; a resource state $\rho$ lives outside it; and the natural yardstick is the shortest relative-entropy bridge from $\rho$ to the island, $M_{\mathcal{F}}(\rho)=\min_{\sigma\in\mathcal{F}}D(\rho\|\sigma)$ . The golden rule is the wall that makes the yardstick a monotone: a free operation slides free states around within the island, never out of it, so it can only shorten the bridge, never lengthen it.

Figure

The anatomy of a resource theory: a convex closed set of free states 𝓕 in green, a resource state ρ outside it, and its closest free state σ⋆ at relative-entropy distance M(ρ) = min D(ρ‖σ). The golden rule keeps every free operation Λ mapping 𝓕 into 𝓕. For the Bell pair the nearest separable state is the equal mixture of 00 and 11, and the distance is exactly 1 ebit of relative entropy of entanglement.

A monotone is bookkeeping for an accounting law. Free operations are credit-free transactions, so any quantity that free operations cannot increase is a currency: if $M(\rho)<M(\sigma)$ , no free operation can take $\rho$ to $\sigma$ , because that would spend currency it does not have. This is the entire content of the framework's no-go theorems, compressed to one line. The art is choosing an $M$ that is both a genuine monotone and tight — one that not only forbids the impossible but predicts the achievable rate. That is where asymptotics enter: by working with many copies and tolerating a vanishing error, the jagged single-shot conversion problem smooths into rates governed by entropic quantities, and in the best-behaved theories every monotone collapses onto a single number. The second figure is that conversion economy — resources traded at a rate, ordered by $E_D\le E_C$ , and pinned to one measure exactly when the theory is reversible.

Figure

The conversion economy: n copies of ρ are converted by free operations 𝓞 into m copies of σ at asymptotic rate R = m per n with vanishing error. Every monotone is a currency squeezed between the distillable resource E_D and the resource cost E_C, so E_D ≤ M ≤ E_C. A pure bipartite state with Schmidt weights 0.8 and 0.2 is reversible at rate S = 0.722 ebits, while a bound resource sits at the far extreme with E_D = 0 but E_C greater than 0.

The core equations

The axiom that starts everything is the golden rule: the free operations must respect the free set, mapping it into itself. Without it, "free" operations could manufacture resource from nothing and the whole accounting collapses.

\Lambda\in\mathcal{O},\ \rho\in\mathcal{F}\ \Longrightarrow\ \Lambda[\rho]\in\mathcal{F} \qquad (1)

where: $\mathcal{F}$ — the set of free states, taken closed and convex, and usually closed under tensoring and partial trace; $\mathcal{O}$ — the free operations, a class of CPTP maps; $\Lambda[\rho]$ — the channel $\Lambda$ applied to $\rho$ . Reading: free operations never create resource, so the free set is stable under them. A resource measure is any functional that respects this economics — it never increases under free operations. If it also vanishes exactly on the free set, it is called faithful, and it doubles as a resource detector.

M:\ \mathcal{D}(\mathcal{H})\to\mathbb{R}_{\ge 0},\qquad M(\Lambda[\rho])\le M(\rho)\ \ \forall\Lambda\in\mathcal{O},\qquad M(\rho)=0\iff\rho\in\mathcal{F} \qquad (2)

where: $M$ — a resource monotone; $\mathcal{D}(\mathcal{H})$ — the density operators on Hilbert space $\mathcal{H}$ ; the monotonicity inequality is the defining axiom, and the vanishing condition is the extra property of faithfulness. Reading: monotones are the currencies of the theory, and faithful ones certify the presence of resource. The universal way to build a monotone is to measure the relative-entropy distance to the free set. Because this distance is defined by a single, model-independent formula, it exists in every resource theory at once.

M_{\mathcal{F}}(\rho)\;=\;\min_{\sigma\in\mathcal{F}} D(\rho\|\sigma),\qquad D(\rho\|\sigma)=\mathrm{Tr}\,\rho\big(\log\rho-\log\sigma\big) \qquad (3)

where: $D(\rho\|\sigma)$ — the quantum (Umegaki) relative entropy; the minimum is over free states $\sigma$ and is attained because $\mathcal{F}$ is compact; $\log$ is base 2, so $M_{\mathcal{F}}$ is measured in bits (ebits for entanglement). Reading: the closest free state sets the resource content; the minimization is what makes it a distance-to-freeness. Away from single copies, conversion is asymptotic: feed in $n$ copies, aim to output $m$ copies of the target, and demand only that the error vanish in the limit. The best achievable ratio is the conversion rate.

R_{\rho\to\sigma}=\sup\Big\{\tfrac{m}{n}:\ \lim_{n\to\infty}\ \inf_{\Lambda\in\mathcal{O}}\ \tfrac12\big\|\Lambda[\rho^{\otimes n}]-\sigma^{\otimes m}\big\|_1=0\Big\} \qquad (4)

where: $R_{\rho\to\sigma}$ — optimal copies of $\sigma$ produced per copy of $\rho$ ; $\tfrac12\|\cdot\|_1$ — the trace distance, the error tolerated; $m,n$ — output and input copy counts. Reading: the rate is the steepest exchange ratio at which near-perfect conversion is still possible. Specialize the target to the standard unit resource $\phi^+$ and you get the two operational numbers that bracket every monotone: how much unit resource you can distill out of $\rho$ , and how much it costs to make $\rho$ . They are never in the wrong order.

E_D(\rho)\;\le\;E_C(\rho),\qquad E_D(\rho)=R_{\rho\to\phi^+},\qquad E_C(\rho)=\frac{1}{R_{\phi^+\to\rho}} \qquad (5)

where: $E_D$ — distillable resource, the yield of pure unit resource; $E_C$ — resource cost, the unit resource needed to form $\rho$ ; $\phi^+$ — the unit resource (a Bell pair for entanglement, a maximally coherent qubit for coherence). Reading: you can never distill more than you paid; a strict gap is irreversibility, and $E_D=0<E_C$ is a bound resource. For coherence the abstract measure becomes a concrete, computable identity — the relative entropy of coherence is just the entropy the dephasing channel injects, which is what lets the worked examples below return exact numbers.

C_R(\rho)=\min_{\sigma\in\mathcal{I}}D(\rho\|\sigma)=S\big(\Delta[\rho]\big)-S(\rho),\qquad \Delta[\rho]=\sum_i \langle i|\rho|i\rangle\,|i\rangle\langle i| \qquad (6)

where: $\mathcal{I}$ — the incoherent (diagonal) states; $\Delta$ — the full dephasing channel that deletes off-diagonal elements in the fixed basis; $S$ — the von Neumann entropy. Reading: the closest incoherent state to $\rho$ is its own dephased version $\Delta[\rho]$ , so the coherence is the entropy gap opened by dephasing. For thermodynamics the same distance becomes a physical free energy, and its monotonicity is the second law written in the resource-theory dialect.

F_\beta(\rho)=\beta^{-1}\,D(\rho\|\gamma_\beta)-\beta^{-1}\log Z,\qquad \gamma_\beta=\frac{e^{-\beta H}}{Z},\ \ Z=\mathrm{Tr}\,e^{-\beta H} \qquad (7)

where: $\gamma_\beta$ — the Gibbs state at inverse temperature $\beta$ , the single free state; $H$ — the system Hamiltonian; $Z$ — the partition function. Reading: distance-from-equilibrium in relative entropy is nonequilibrium free energy, and thermal operations cannot increase it.

Derivation AThe relative entropy of resource is a faithful monotone

The goal is to prove that the single formula $M_{\mathcal{F}}(\rho)=\min_{\sigma\in\mathcal{F}}D(\rho\|\sigma)$ is simultaneously non-negative, faithful (zero exactly on the free set), and non-increasing under every free operation — so that one construction hands a valid, resource-detecting monotone to every resource theory at once. The whole proof rests on just two standard facts about the quantum relative entropy: Klein's inequality, which controls its sign and equality case, and the data-processing inequality, which says a channel can only decrease it. The golden rule then does the essential work of guaranteeing that the image of the optimal free state is still free.

Step 1. Non-negativity. Klein's inequality states $D(\rho\|\sigma)\ge 0$ for all states, with equality if and only if $\rho=\sigma$ ; taking the minimum over $\sigma\in\mathcal{F}$ preserves the bound, so $M_{\mathcal{F}}(\rho)\ge 0$ . Why: a distance built from a non-negative divergence cannot itself go negative, so the measure starts life on the right side of zero.

Step 2. Faithfulness, easy direction. If $\rho\in\mathcal{F}$ , choose $\sigma=\rho$ in the minimization; then $D(\rho\|\rho)=0$ forces $M_{\mathcal{F}}(\rho)=0$ . Why: a free state is zero distance from itself, so every free state reads exactly zero — no false positives.

Step 3. Faithfulness, hard direction. If $M_{\mathcal{F}}(\rho)=0$ , compactness of $\mathcal{F}$ gives a minimizer $\sigma^\star\in\mathcal{F}$ with $D(\rho\|\sigma^\star)=0$ ; Klein's equality case then forces $\rho=\sigma^\star$ , which lies in $\mathcal{F}$ . Why: only genuinely free states can read zero, so the measure never mistakes a resource for free — it is a faithful detector.

Step 4. Monotonicity. Let $\Lambda\in\mathcal{O}$ and let $\sigma^\star$ achieve the minimum for $\rho$ . By the golden rule $\Lambda[\sigma^\star]\in\mathcal{F}$ , so it is a feasible free state for $\Lambda[\rho]$ ; data processing gives $M_{\mathcal{F}}(\Lambda[\rho])\le D(\Lambda[\rho]\|\Lambda[\sigma^\star])\le D(\rho\|\sigma^\star)=M_{\mathcal{F}}(\rho)$ . Why: the golden rule keeps the optimal comparison state inside the free island after the operation, and data processing shrinks the distance — the two facts combine into monotonicity. *Verified: each step uses only Klein's inequality and data processing for $D$ (Wilde Ch. 11; Watrous Ch. 5); the construction is generic, so $M_{\mathcal{F}}$ is a faithful monotone in every resource theory whose free set is closed under the free operations.*

Go deeper · Derivation B · 1 minA single monotone is a no-go theorem

Derivation BA single monotone is a no-go theorem

The goal is to show that any monotone whatsoever, evaluated on two states, can certify that no free operation converts one into the other — turning a hard existence question about protocols into a single scalar inequality. This is the most-used deduction in the entire subject: it is why a computed negativity forbids LOCC from raising entanglement, why a coherence measure forbids incoherent operations from creating a superposition, and why an athermality measure forbids a thermal machine from lowering a system's free energy. The argument is three lines of substitution and one contrapositive, and it assumes nothing beyond the monotone axiom.

Step 1. Suppose a free operation achieves the conversion exactly: $\Lambda[\rho]=\sigma$ with $\Lambda\in\mathcal{O}$ . Take this as the hypothesis to be tested. Why: we want to know what any valid monotone must say if the conversion were possible, so we assume possibility and follow the consequences.

Step 2. Apply the monotone axiom to $\Lambda$ : $M(\Lambda[\rho])\le M(\rho)$ for every monotone $M$ . Why: this is the one property every monotone has by definition — free operations cannot raise it.

Step 3. Substitute the hypothesis $\Lambda[\rho]=\sigma$ to get $M(\sigma)\le M(\rho)$ . Why: the value of the target is bounded by the value of the source, so a conversion can only preserve or spend currency, never mint it.

Step 4. Take the contrapositive: if $M(\rho)<M(\sigma)$ for even one monotone $M$ , then no $\Lambda\in\mathcal{O}$ with $\Lambda[\rho]=\sigma$ can exist. Why: a single monotone that goes the wrong way is an insurmountable obstruction, so one inequality certifies impossibility — no search over protocols is needed. *Verified: the deduction holds for any function satisfying the monotone axiom, with no further assumptions; note the converse fails — equality of all known monotones does not by itself certify that a conversion is possible.*

Go deeper · Derivation C · 1 minReversibility forces a unique asymptotic measure

Derivation CReversibility forces a unique asymptotic measure

The goal is to prove that in an asymptotically reversible resource theory the distillable resource, the resource cost, and every additive, asymptotically continuous, normalized monotone are all the same number — the resource-theoretic echo of entropy being the unique asymptotic rate in Shannon's coding theorems. The strategy is a squeeze: monotonicity in the asymptotic limit sandwiches any such measure between $E_D$ and $E_C$ , and reversibility clamps the two ends together, leaving the measure no freedom. This is why "nice" theories, like pure-state entanglement, have a single canonical measure while messy ones do not.

Step 1. Distillation bound. Converting $\rho^{\otimes n}\to(\phi^+)^{\otimes m}$ with vanishing error cannot increase a regularized, asymptotically continuous monotone in the limit, so $E_D(\rho)\,M^{\infty}(\phi^+)\le M^{\infty}(\rho)$ . Why: the monotone axiom, carried into the many-copy limit, says the resource you distill out cannot exceed the resource you started with.

Step 2. Normalize the unit. Fix the normalization $M^{\infty}(\phi^+)=1$ so the unit resource is worth one; then Step 1 reads $E_D(\rho)\le M^{\infty}(\rho)$ . Why: choosing the unit resource as the yardstick turns the abstract inequality into a statement in the natural currency.

Step 3. Cost bound. Forming $\rho^{\otimes n}$ from $(\phi^+)^{\otimes k}$ likewise cannot increase the monotone, giving $M^{\infty}(\rho)\le E_C(\rho)\,M^{\infty}(\phi^+)=E_C(\rho)$ . Why: the same axiom run in the reverse direction says making $\rho$ costs at least its measured resource content.

Step 4. Squeeze. Combining, $E_D(\rho)\le M^{\infty}(\rho)\le E_C(\rho)$ ; reversibility asserts $E_D=E_C$ , collapsing the chain and forcing all three equal. Why: two inequalities pointing through a common quantity, with the outer ends pinned together, leave exactly one value — the measure is unique. *Verified: mirrors the pure-state entanglement reversibility argument (Bennett et al. 1996) and the general treatment of Chitambar & Gour (RMP 2019); rigorous given additivity and asymptotic continuity of $M$ .*

Limiting cases

The single axiomatic skeleton becomes a specific, named physics the moment you fix the pair $(\mathcal{F},\mathcal{O})$ . The rows below are not different theories — they are one theory read in different currencies.

Case	Result	Meaning
Entanglement	$\mathcal{F}=\{\sum_i p_i\,\rho_i^A\otimes\sigma_i^B\}$ , $\ \mathcal{O}=\text{LOCC}$	Free states are separable; free operations are local operations and classical communication. Reversible for pure states, irreversible for mixed — bound entanglement has $E_D=0<E_C$ .
Coherence	$\mathcal{F}=\{\sum_i p_i\,\|i\rangle\langle i\|\}$ , $\ \mathcal{O}=\text{incoherent ops}$	Free states are diagonal in a fixed basis; $C_R(\rho)=S(\Delta[\rho])-S(\rho)$ . Several inequivalent free-operation classes share these free states.
Asymmetry	$\mathcal{F}=\{\rho:\,U_g\rho U_g^\dagger=\rho\ \forall g\}$ , $\ \mathcal{O}=G\text{-covariant}$	Free states are invariant under a group representation; resourceful states break the symmetry and serve as reference frames or clocks.
Thermodynamics	$\mathcal{F}=\{\gamma_\beta\}$ , $\ \mathcal{O}=\text{thermal ops}$	A single free state, the Gibbs state; the monotone is the nonequilibrium free energy $F_\beta$ , and its monotonicity is the second law.
Magic / purity	$\mathcal{F}=\text{stabilizer (or maximally mixed)}$ , $\ \mathcal{O}=\text{stabilizer (unital) ops}$	Non-stabilizerness is the resource powering fault-tolerant speed-up; purity is the resource for work extraction and Landauer erasure.

Common trapThere is no such thing as "the" resource measure until you name the free operations

The free set and the free operations are independent design choices, and the same free states can be paired with genuinely different operation classes. Coherence alone has at least three standard classes — incoherent, strictly incoherent, and dephasing-covariant operations — all fixing the same diagonal free states but giving different conversion rates and different lists of valid monotones. So a phrase like "the coherence of $\rho$ " is ambiguous until the operation class is specified. A second, subtler trap: a monotone is only an upper bound on what conversions forbid, not a certificate of what they allow. Finding $M(\rho)\ge M(\sigma)$ for every monotone you can compute does not prove $\rho\to\sigma$ is possible — impossibility is certified by one monotone going the wrong way, but possibility requires an explicit protocol. Monotones are one-directional evidence.

Worked examples

1 · The coherence of a pure qubit, exactly (fully worked)

In the resource theory of coherence with free basis $\{|0\rangle,|1\rangle\}$ , compute the relative entropy of coherence $C_R$ and the $\ell_1$ coherence $C_{\ell_1}$ of the plus state $|+\rangle=\tfrac{1}{\sqrt2}(|0\rangle+|1\rangle)$ , and then of a general pure qubit $|\psi\rangle=\cos\theta\,|0\rangle+\sin\theta\,|1\rangle$ . Interpret the answer as a currency.

Step 1. Write the density matrix and dephase it. For $|+\rangle$ , $\rho=\tfrac12\left(\begin{smallmatrix}1&1\\1&1\end{smallmatrix}\right)$ ; the full dephasing channel deletes the off-diagonal entries, giving $\Delta[\rho]=\tfrac12\left(\begin{smallmatrix}1&0\\0&1\end{smallmatrix}\right)=I/2$ . By equation (6), $\Delta[\rho]$ is the closest incoherent state, so it does the work.

Step 2. Evaluate the entropy gap. The pure state has $S(\rho)=0$ ; the dephased state is maximally mixed, $S(I/2)=1$ bit. Hence $C_R(|+\rangle)=S(\Delta[\rho])-S(\rho)=1-0=\mathbf{1\ bit}$ — the plus state is a maximally coherent qubit, worth exactly one unit of coherence.

Step 3. The $\ell_1$ measure. Summing the magnitudes of the off-diagonal entries gives $C_{\ell_1}(|+\rangle)=|{\tfrac12}|+|{\tfrac12}|=\mathbf{1}$ . Both measures agree that $|+\rangle$ carries one unit, which is why it is the natural coherence analogue of the Bell pair.

Step 4. The general angle. For $|\psi\rangle=\cos\theta|0\rangle+\sin\theta|1\rangle$ , dephasing yields $\Delta[\rho]=\mathrm{diag}(\cos^2\theta,\sin^2\theta)$ , so $C_R=H_2(\cos^2\theta)$ (the binary entropy) while the off-diagonal $\cos\theta\sin\theta=\tfrac12\sin2\theta$ gives $C_{\ell_1}=|\sin2\theta|$ . At $\theta=30^\circ$ : $\cos^2\theta=0.75$ , so $C_R=H_2(0.75)=\mathbf{0.811\ bits}$ and $C_{\ell_1}=\sin60^\circ=\mathbf{0.866}$ . A monotone that reads 1 bit at $|+\rangle$ and $0.811$ bits at $30^\circ$ is the currency: no incoherent operation can turn the $30^\circ$ state into $|+\rangle$ , because $0.811<1$ .

2 · The Bell pair costs one ebit — and LOCC cannot mint it (fully worked)

In the entanglement theory ( $\mathcal{F}=\text{separable}$ , $\mathcal{O}=\text{LOCC}$ ), compute the relative entropy of entanglement of the Bell pair $\Phi^+=\tfrac{1}{\sqrt2}(|00\rangle+|11\rangle)$ , and use the monotone no-go to prove that no LOCC protocol can create $\Phi^+$ from the product state $|00\rangle$ .

Step 1. Guess the closest separable state and compute the distance. Take $\sigma^\star=\tfrac12(|00\rangle\langle00|+|11\rangle\langle11|)=\mathrm{diag}(\tfrac12,0,0,\tfrac12)$ , which is separable. Since $\Phi^+$ is pure, $S(\Phi^+)=0$ , and $D(\Phi^+\|\sigma^\star)=-S(\Phi^+)-\mathrm{Tr}\,\Phi^+\log\sigma^\star=-\mathrm{Tr}\,\Phi^+\log\sigma^\star$ .

Step 2. Evaluate the cross term. The Bell state lives entirely in the $\{|00\rangle,|11\rangle\}$ subspace, where $\log\sigma^\star=\log(\tfrac12)\,I=-1\cdot I$ . So $\mathrm{Tr}\,\Phi^+\log\sigma^\star=-1$ , giving $D(\Phi^+\|\sigma^\star)=1$ . This $\sigma^\star$ is in fact optimal, so $E_R(\Phi^+)=\mathbf{1\ ebit}$ — the Bell pair sits exactly one ebit outside the separable island (the number anchored in Figure 1).

Step 3. Evaluate the source. The product state $|00\rangle$ is separable, so it is free: $E_R(|00\rangle)=0$ . Every faithful entanglement monotone reads zero on it.

Step 4. Apply the no-go. We have a monotone with $E_R(|00\rangle)=0<1=E_R(\Phi^+)$ . By Derivation 2, $M(\rho)<M(\sigma)$ forbids any free operation from taking $\rho$ to $\sigma$ . Therefore no LOCC protocol, on any number of copies, can produce a Bell pair from $|00\rangle$ — the one-line proof of the intuition that separated labs cannot manufacture entanglement over a telephone. The gap $1-0=1$ ebit is precisely the entanglement that must be supplied, not created.

Physics of the portrait

The framework is not the work of a single person but of a thirty-year abstraction, in which a specific trick for entanglement was slowly recognized as a template for all of quantum information — and then written down as pure axioms.

Origins · 2 minresource theory and how this was discovered

resource theory

the free-state and free-operation template, abstracted 1996–2016 · the generalized second laws and the economics of quantum information

The resource theory did not begin as a theory of resources; it began as a theory of one resource. In 1996 Charles Bennett and collaborators, studying how many Bell pairs could be concentrated from or diluted into partially entangled pure states, discovered that a single number — the entropy of entanglement — governed a reversible exchange, and that "local operations and classical communication" was the natural free class. The word that reorganized the subject was free: once you name what costs nothing, the resource defines itself. Through the late 1990s the Horodecki family turned entanglement from a yes/no property into a graded, computable hierarchy, discovering bound entanglement — states you pay to make and can never cash out — the first proof that a resource theory can be irreversible. In the 2000s the same grammar was ported wholesale: to reference frames and superselection rules (asymmetry), to quantum thermodynamics, where Brandão, Horodecki, Oppenheim and others showed the second law fractures into a family of free-energy monotones for small systems, and to coherence, axiomatized by Baumgratz, Cramer and Plenio in 2014. The final move was to erase the physics entirely: in 2016 Coecke, Fritz and Spekkens wrote down a mathematical theory of resources — free states, free operations, monotones, conversion rates — with no mention of entanglement or heat, and Chitambar and Gour's 2019 Reviews of Modern Physics survey codified it. The through-line is one idea held with discipline: an entanglement measure, a coherence measure, a free energy are the same kind of object — a monotone controlling a conversion rate — and the deepest theorems (data processing gives you a monotone; reversibility gives you a unique one) are properties of the template, not of any particular resource.

History

1996

— Bennett, Bernstein, Popescu and Schumacher establish reversible pure-state entanglement concentration and dilution

1998

— the Horodecki family proves bound entanglement, the first irreversible resource

2013

— Brandao, Horodecki, Oppenheim and others derive the generalized second laws of quantum thermodynamics

2016

— Coecke, Fritz and Spekkens write down the abstract mathematical theory of resources

2019

— Chitambar and Gour survey and codify quantum resource theories

Going further

Texts. Nielsen & Chuang, Quantum Computation and Quantum Information (2010), Ch. 11–12 · Wilde, Quantum Information Theory (2nd ed., 2017), Ch. 11, 21 · Watrous, The Theory of Quantum Information (2018), Ch. 5–6 · Preskill, Lecture Notes on Quantum Computation, Ch. 10 · Chitambar & Gour, Quantum resource theories, Rev. Mod. Phys. 91, 025001 (2019) · Horodecki & Oppenheim, (Quantumness in the context of) resource theories, Int. J. Mod. Phys. B 27, 1345019 (2013).

Provenance. A verifier should recompute, all confirmed here: (1) the data-processing proof that $M_{\mathcal{F}}$ is a faithful monotone, using Klein's inequality and monotonicity of $D$ (Wilde Ch. 11; Watrous Ch. 5); (2) the coherence identity $C_R(\rho)=S(\Delta[\rho])-S(\rho)$ , giving $C_R(|+\rangle)=1$ bit, $C_R(30^\circ)=H_2(0.75)=0.811$ bits, and $C_{\ell_1}(|+\rangle)=1$ , $C_{\ell_1}(30^\circ)=\sin60^\circ=0.866$ ; (3) the relative entropy of entanglement of the Bell pair, $E_R(\Phi^+)=D(\Phi^+\|\tfrac12(|00\rangle\langle00|+|11\rangle\langle11|))=1$ ebit, and the no-go $E_R(|00\rangle)=0<1$ forbidding LOCC creation; (4) the reversibility squeeze $E_D\le M^\infty\le E_C$ (Bennett et al. 1996; Chitambar & Gour, RMP 2019); (5) the pure-state reversible rate $S=H_2(0.8)=0.722$ ebits for Schmidt weights $\{0.8,0.2\}$ . No empirical constants are asserted; all statements are textbook structural results cited above.

Prerequisites. Quantum Channels and CPTP Maps · Quantum Relative Entropy · Data-Processing Inequality · Von Neumann Entropy. Leads to. Strong Subadditivity of Quantum Entropy · Holevo Bound · Complementary Channels and Degradability.

Ask the tutor: Ask me to specialize the framework to any resource you name — entanglement, coherence, magic, or athermality — and compute its monotone on a state of your choice.

What actually makes something a "resource" rather than just a property of a state?

A resource is defined relative to a rulebook, not in isolation. You first declare which states and operations are free — cost nothing, available without limit — and a property becomes a resource precisely when the free operations cannot create it. Entanglement is a resource under LOCC because separated labs cannot make it with local operations and phone calls; it is not a resource if you allow global operations, which can make it trivially. So the same physical quantity can be a precious resource in one theory and free in another. The content of a resource theory is entirely in the choice of what is free — everything else, including which functions count as monotones and what the conversion rates are, follows from that single decision.

Why does the quantum relative entropy show up in every resource theory?

Because it is the divergence that contracts under channels. The data-processing inequality says $D(\rho\|\sigma)\ge D(\Lambda[\rho]\|\Lambda[\sigma])$ for any CPTP map $\Lambda$ , and that single property is exactly what a monotone needs: measure the distance from a state to the free set, and any free operation can only shrink it, because the golden rule keeps the comparison state free. So $\min_{\sigma\in\mathcal{F}}D(\rho\|\sigma)$ is a monotone in every resource theory at once, without any theory-specific work. Relative entropy also carries operational meaning — through quantum Stein's lemma it is the optimal error exponent for distinguishing a resource from the free states — so the monotone it builds is not just mathematically convenient but tracks a real hypothesis-testing task.

What is the difference between a monotone and an operationally meaningful measure?

Any function that never increases under free operations is a monotone, and there are infinitely many — sums, maxima, and convex combinations of monotones are again monotones. That makes monotones cheap and abundant, and most of them mean nothing operationally. A measure earns operational meaning when it equals a physical rate: the distillable resource $E_D$ is the yield of pure resource you can extract, the cost $E_C$ is what you must spend to build the state, and these bracket every monotone. The relative entropy of resource is monotone by construction but only sometimes equals a rate — for pure-state entanglement it does, coinciding with distillation and cost, while for bound entanglement it sits strictly above the zero distillation rate. Being a monotone is necessary; tracking a rate is the prize.

Why is reversibility special, and what breaks it?

Reversibility means every ebit or bit you pour into a state can be poured back out — distillation equals cost, $E_D=E_C$ , and a single measure governs all conversions, exactly as entropy governs reversible heat exchange. Pure-state entanglement is reversible: the entropy of entanglement is that unique measure. What breaks it is wasted classical uncertainty. Mixed states carry ignorance that free operations can irreversibly burn but never recover, opening the gap $E_D<E_C$ ; at the extreme lies bound entanglement, $E_D=0<E_C$ , resource that is real (you paid for it) yet permanently locked. Irreversibility is therefore a genuine physical statement — a resource-theoretic arrow of time — not a failure of the theory. Whether broad reversibility can be restored by enlarging the free operations is a deep and only recently settled question.

Is "magic" really a resource theory, or just a metaphor for quantum computing?

It is a resource theory in the full technical sense, and one of the most important. Fix the free states to be stabilizer states and the free operations to be stabilizer (Clifford) operations — by the Gottesman–Knill theorem these are efficiently simulable on a classical computer, so they cannot supply any quantum computational advantage. The resource that must be injected to reach universal, classically-hard quantum computation is non-stabilizerness, or magic, carried by states like $|T\rangle$ . All the standard machinery applies: magic monotones (mana, robustness of magic, stabilizer Rényi entropy) obey the no-go theorem, and magic-state distillation is literally resource distillation with stabilizer operations as the free class. This is why the framework is not decorative for quantum computing — it quantifies the exact ingredient fault tolerance is most expensive to produce.

What is catalysis, and can a catalyst be secretly used up?

A catalyst is an auxiliary resource state $c$ that assists a conversion $\rho\to\sigma$ and must be returned unchanged at the end, so it is genuinely reused rather than consumed. Catalysis matters because it can enlarge what free operations achieve: some conversions impossible with $\rho$ alone become possible with $\rho\otimes c\to\sigma\otimes c$ , even though $c$ emerges intact. The catch, and the reason the definition is subtle, is that allowing the catalyst to change by a tiny amount — approximate or "embezzling" catalysis — can unlock dramatically more, sometimes trivializing the theory, so the exact-return condition is doing real work. Catalysis also reshapes the second law: in quantum thermodynamics, the full family of free-energy constraints is exactly what governs which catalytic transformations are allowed, a refinement invisible in the macroscopic limit where a single free energy suffices.

How does the framework connect to the ordinary second law of thermodynamics?

Directly and generalizingly. Choose the free state to be the Gibbs state $\gamma_\beta$ and the free operations to be thermal operations (energy-conserving interactions with a heat bath at temperature $T$ ). Then the relative entropy of resource becomes the nonequilibrium free energy $F_\beta(\rho)=\beta^{-1}D(\rho\|\gamma_\beta)-\beta^{-1}\log Z$ , and its monotonicity under free operations is the statement that free energy cannot increase — the second law. For large systems this single monotone captures everything, recovering textbook thermodynamics. For small quantum systems, though, the framework reveals more: a whole family of Rényi free energies must each be non-increasing, so a single inequality is replaced by many. These are the generalized second laws, and they are a genuine prediction of the resource-theoretic viewpoint, testable in nanoscale engines and quantum batteries where fluctuations matter.