betaFeedback

WorksPhysics and Astronomy

Ergotropy as distance from equilibrium

Abstract

We show that the ergotropy (a.k.a available energy) of a thermally isolated system is the 11-Wasserstein distance between two energy probability density functions. Specifically, between the energy profile of the system state, and the energy profile of its associated passive state. We prove this for both classical and quantum systems, and illustrate it with examples. Our result formalises the intuition that ergotropy is a quantifier of distance from equilibrium

Introduction

The ergotropy, a.k.a. available energy, of a thermally isolated system is the maximal amount of energy it can release as a consequence of an external time dependent driving [1, 2, 3, 4, 5]. After the extraction of ergotropy a system remains in the so-called passive state ρ~\tilde{\rho} associated to the initial state ρ\rho [6, 7, 8, 9]. The passive state is energetically inert, namely it is such that it cannot release any energy if externally perturbed: in fact it can only absorb energy. The most prominent example is the thermal equilibrium state of canonical form e−βH/Ze^{-\beta H}/Z. However the family of passive states is much larger and includes all states that are stationary and are characterised, just like the canonical ensemble, by the fact that the probability diminishes as the energy increases. In this regard passive sates can be understood as generalised thermodynamic equilibria.

Intuitively, the higher the ergotropy the more one expects the system to be away from its associated equilibrium state. This intuition was strongly corroborated by the recent work of Ref. [10] highlighting that the available energy of the electrons trapped in a flux tube of a collisionless plasma is highly correlated to the amount of turbolent energy transport.

Can this intuition be formalised? Namely, can ergotropy be expressed, mathematically, as some “distance” from equilibrium? Here we answer in the positive. To do so we single out the appropriate space and the proper notion of distance over said space, which coincides with the notion of ergotropy. The space is that of energy probability density functions, and the distance is the 1-Wasserstein distance over said space. Specifically, we prove that given a state ρ\rho, its ergotropy is the 1-Wasserstein distance between its energy profile, and the energy profile of its passive companion ρ~\tilde{\rho}, namely the according equilibrium.

We prove this result for both classical and quantum systems and illustrate it with the 1D harmonic oscillator, both classical and quantum.

Main result: classical case

Given a classical system with Hamiltonian H(z)H(\mathbf{z}), being in a statistical state described by the probability density ρ(z)\rho(\mathbf{z}) (here z=(q,p)\mathbf{z} = (\mathbf{q},\mathbf{p}) is a point in the ff-dimensional phase space), we ask the maximal amount of energy that can be extracted from the system by means of a cyclic variation of an external parameter.

That is, we seek the quantity

E[ρ]=∫dzρ(z)H(z)−min⁡ϕ∫dzρ(ϕ−1(z))H(z)(1)\mathcal{E}[\rho] = \int\mathrm{d}z\rho(z)H(z) - \min_{\phi}\int\mathrm{d}z\rho(\phi^{-1}(z))H(z) \tag*{(1)}

where the minimisation is carried over all symplectic diffeomorphisms ϕ\phi in phase space. E\mathcal{E} is known as the Ergotropy of ρ\rho, relative to HH. This is the classical analogue of the well-known ergotropy of a quantum system [1]. In atmospheric, plasma- and astro-physics it is known as “available energy” [4, 5, 11].

The state ρ˘\breve{\rho} that reaches the minimal energy, known as the ground state, or the passive companion of ρ\rho, is mathematically speaking a measure preserving rearrangement of ρ\rho, specifically the rearrangement according to which the probability density ρ˘\breve{\rho} does not increase as one moves in phase space from one point to a point of higher energy HH [2, 12]. Such “ergotropic rearrangement” has been shown recently [2, 12], to be given by the compact expression

ρ˘=Σ−1∘Ω∘H(2)\breve{\rho} = \Sigma^{-1} \circ\Omega\circ H \tag*{(2)}

where

Σ(r)=∫dz θ[ρ(z)−r](3)\Sigma(r) = \int dz\,\theta[\rho(z)-r] \tag*{(3)}
Ω(E)=∫dz θ[E−H(z)](4)\Omega(E) = \int dz\,\theta[E-H(z)] \tag*{(4)}

and

Σ−1(y)=∫0∞dx θ[Σ(x)−y](5)\Sigma^{-1}(y) = \int_{0}^{\infty} dx\,\theta[\Sigma(x)-y] \tag*{(5)}

is the (generalized) inverse of Σ\Sigma.1 The symbol θ(⋅)\theta(\cdot) denotes the Heaviside step function. Note that, given HH, ρ˘\breve{\rho} is uniquely determined by ρ\rho, and that ρ˘\breve{\rho} is a stationary state, because its Poisson brakets with HH are trivially null. Note that by definition Ω\Omega is non-decreasing (in fact for any physical Hamiltonian Ω\Omega is strictly increasing). Similarly, Σ\Sigma is non-increasing, so ρ˘\breve{\rho} is non-increasing as HH increases.

The ergotropy of the state ρ\rho is thus given by the difference between the energy expectation of ρ\rho and that of ρ˘\breve{\rho}

E[ρ]=∫dzH(z)[ρ(z)−ρ˘(z)](6)\mathcal{E}[\rho] = \int dz H(z)[\rho(z)-\breve{\rho}(z)] \tag*{(6)}

In order to state our main result we need the following two ingredients. First, let us introduce the energy profile P(E)\mathcal{P}(E) of ρ\rho, namely the probability density that the system lies on the hyper-surface H(z)=EH(z)=E, given that it is distributed in phase space according to ρ\rho, namely,

P(E)=∫dzδ[E−H(z)]ρ(z)(7)\mathcal{P}(E) = \int dz\delta[E-H(z)]\rho(z) \tag*{(7)}

Similarly, let P˘(E)\breve{\mathcal{P}}(E) be the energy profile of ρ˘\breve{\rho}

P˘(E)=∫dzδ[E−H(z)]ρ˘(z)(8)\breve{\mathcal{P}}(E) = \int dz\delta[E-H(z)]\breve{\rho}(z) \tag*{(8)}

Second, let us introduce the 1-Wasserstein distance W1[Q1,Q2]W_1[Q_1,Q_2] between two normalized probability distributions, Qi:RN→R+Q_i:\mathbb{R}^{N}\to\mathbb{R}_+, i=1,2i=1,2

W1[Q1,Q2]=inf⁡γ∈Γ∫dxdy γ(x,y)∣x−y∣(9)W_1[Q_1,Q_2] = \inf_{\gamma\in\Gamma}\int dxdy\,\gamma(x,y)|x-y| \tag*{(9)}

where Γ\Gamma is the set of all joint probability distributions γ(x,y)\gamma(x,y) (a.k.a. “transport plans” in this context) whose marginals are Q1(x)Q_1(x) and Q2(y)Q_2(y), namely ∫dxγ(x,y)=Q2(y)\int dx\gamma(x,y)=Q_2(y), ∫dyγ(x,y)=Q1(x)\int dy\gamma(x,y)=Q_1(x). DD is a proper metric on the space of distributions.

The minimisation problem (9) has a simple analytical solution in the case of probability distributions on the line (N=1N=1) [14]:

W1[Q1,Q2]=∫dx∣c1(x)−c2(x)∣(10)W_1[Q_1,Q_2]=\int dx\lvert c_1(x)-c_2(x)\rvert \tag*{(10)}

where

ci(x)=∫−∞xds Qi(s)(11)c_i(x)=\int_{-\infty}^{x}ds\,Q_i(s) \tag*{(11)}

denote the cumulative distribution functions.

Our main result is the following

Theorem 1 The ergotropy of ρ\rho is the 1-Wasserstein distance between the energy profile of ρ\rho, and the energy profile of ρ~\tilde{\rho}

E[ρ]=W1[P,P~](12)\mathcal{E}[\rho]=W_1[\mathcal{P},\tilde{\mathcal{P}}] \tag*{(12)}

Using the energy profiles P\mathcal{P}, P~\tilde{\mathcal{P}}, the ergotropy can be readily written as

E=∫0∞dE [P(E)−P~(E)]E.(13)\mathcal{E}=\int_{0}^{\infty}dE\,[\mathcal{P}(E)-\tilde{\mathcal{P}}(E)]E. \tag*{(13)}

where we have conventionally set the minimum of HH to 0. Introducing the cumulative probability functions

C(x)=∫0xdE P(E)(14)C(x)=\int_{0}^{x}dE\,\mathcal{P}(E) \tag*{(14)}
C~(x)=∫0xdE P~(E)(15)\tilde{C}(x)=\int_{0}^{x}dE\,\tilde{\mathcal{P}}(E) \tag*{(15)}

and integrating by parts, we get:

E=[(C(x)−C~(x))x]0∞−∫0∞dE [C(E)−C~(E)](16)\mathcal{E}=\left[(C(x)-\tilde{C}(x))x\right]_{0}^{\infty}-\int_{0}^{\infty}dE\,[C(E)-\tilde{C}(E)] \tag*{(16)}

Note that, in the limit x→∞x\to\infty, it is C(x)→1C(x)\to1, C~(x)→1\tilde{C}(x)\to1, therefore C(x)−C~(x)→0C(x)-\tilde{C}(x)\to0, but the product (C(x)−C~(x))x(C(x)-\tilde{C}(x))x is indeterminate. However, because P,P~\mathcal{P},\tilde{\mathcal{P}} have finite first moment (because we assumed from the start that ρ,ρ~\rho,\tilde{\rho} have a well defined energy expectation), it means they go to zero faster than 1/x21/x^2 for large xx. Thus, using de l’Hôpital rule, we get

lim⁡x→∞[C(x)−C~(x)]x=lim⁡x→∞C~(x)−C(x)1/x(17)\lim_{x\to\infty}[C(x)-\tilde{C}(x)]x=\lim_{x\to\infty}\frac{\tilde{C}(x)-C(x)}{1/x} \tag*{(17)}
=lim⁡x→∞P(x)−P~(x)1/x2=lim⁡x→∞P(x)1/x2−lim⁡x→∞P~(x)1/x2=0(18)=\lim_{x\to\infty}\frac{\mathcal{P}(x)-\tilde{\mathcal{P}}(x)}{1/x^2} =\lim_{x\to\infty}\frac{\mathcal{P}(x)}{1/x^2}-\lim_{x\to\infty}\frac{\tilde{\mathcal{P}}(x)}{1/x^2}=0 \tag*{(18)}

On the other hand, trivially (C(x)−C~(x))x→0(C(x)-\tilde{C}(x))x\to0, for x→0x\to0, therefore

E=∫0∞dE [C~(E)−C(E)](19)\mathcal{E}=\int_{0}^{\infty}dE\,[\tilde{C}(E)-C(E)] \tag*{(19)}

Next, let’s consider the probability density P(Φ)P(\Phi) (P˘(Φ)\breve{P}(\Phi)) that the system is on the surface of constant energy that encloses the volume Φ\Phi, provided it is distributed in phase space according to ρ(ρ˘)\rho(\breve{\rho}), namely [2]

P(Φ)=∫dzδ[Φ−Ω(H(z))]ρ(z)(20)P(\Phi)=\int dz\delta[\Phi-\Omega(H(z))]\rho(z) \tag*{(20)}
P˘(Φ)=∫dzδ[Φ−Ω(H(z))]ρ˘(z)(21)\breve{P}(\Phi)=\int dz\delta[\Phi-\Omega(H(z))]\breve{\rho}(z) \tag*{(21)}

For simplicity we shall call P,P˘P,\breve{P}, the phase-volume-profiles of ρ,ρ˘\rho,\breve{\rho}, respectively. We recall that, as shown in Ref. [2],

P(Φ)=∫dΘG[Θ∣Φ]P˘(Θ),(22)P(\Phi)=\int d\Theta G[\Theta|\Phi]\breve{P}(\Theta), \tag*{(22)}

where GG is the overlap function

G[Θ∣Φ]=∫dzδ[Φ−Ω(H(z))]δ[Θ−Σ(ρ(z))](23)G[\Theta|\Phi]=\int dz\delta[\Phi-\Omega(H(z))]\delta[\Theta-\Sigma(\rho(z))] \tag*{(23)}

We remark that GG is a “doubly stochastic” conditional probability density [13], namely

∫dΘG[Θ∣Φ]=∫dΦG[Θ∣Φ]=1,G≥0(24)\int d\Theta G[\Theta|\Phi]=\int d\Phi G[\Theta|\Phi]=1,\quad G\geq0 \tag*{(24)}

This implies that (see Ref. [13], pag 99)

∫0xdΦP(Φ)≤∫0xdΦP˘(Φ)∀x≥0(25)\int_{0}^{x}d\Phi P(\Phi)\leq\int_{0}^{x}d\Phi\breve{P}(\Phi)\quad\forall x\geq0 \tag*{(25)}

It is not difficult to see that energy and phase-volume profiles are linked by the change of variable E→Φ=Ω(E)E\to\Phi=\Omega(E):

P(E)=P(Ω(E))ω(E)(26)P(E)=P(\Omega(E))\omega(E) \tag*{(26)}

where

ω(E)=ddEΩ(E)(27)\omega(E)=\frac{d}{dE}\Omega(E) \tag*{(27)}

is the density of states [13, 15]. In fact, we have:

P(Ω(E))=∫dzδ[Ω(E)−Ω(H(z))]ρ(z)(28)P(\Omega(E))=\int dz\delta[\Omega(E)-\Omega(H(z))]\rho(z) \tag*{(28)}
=∫dzδ[E−H(z)]ω(E)ρ(z)=P(E)ω(E)(29)=\int dz\frac{\delta[E-H(z)]}{\omega(E)}\rho(z)=\frac{P(E)}{\omega(E)} \tag*{(29)}

where we used δ(f(x)−a)=δ(x−f−1(a))/∣f′(f−1(a))∣\delta(f(x)-a)=\delta(x-f^{-1}(a))/|f'(f^{-1}(a))| for a strictly increasing function ff. Similarly,

P˘(E)=P˘(Ω(E))ω(E)(30)\breve{P}(E)=\breve{P}(\Omega(E))\omega(E) \tag*{(30)}

Therefore, choosing x=Ω(y)x=\Omega(y) as the upper integration limit in Eq. (25), and making the usual change of variable Φ→E\Phi\to E, we get:

∫0ydEP(E)≤∫0ydEP˘(E)∀y≥0(31)\int_{0}^{y}dE P(E)\leq\int_{0}^{y}dE\breve{P}(E)\quad\forall y\geq0 \tag*{(31)}

that is

C(y)≤C˘(y).(32)C(y) \leq\breve{C}(y). \tag*{(32)}

This allows to replace [C˘(E)−C(E)][\breve{C}(E)-C(E)] with ∣C˘(E)−C(E)∣|\breve{C}(E)-C(E)| in Eq. (16), and obtain

E=∫0∞dE∣C˘(E)−C(E)∣(33)\mathcal{E}=\int_{0}^{\infty} dE|\breve{C}(E)-C(E)| \tag*{(33)}

Noting that the r.h.s. of the above equation is the 1-Wasserstein distance between P\mathcal{P} and P˘\breve{\mathcal{P}}, Eq. (10), concludes the proof.

Main result: quantum case

In the quantum case, the ergotropy is given by

E[ρ^]=Tr⁡ρ^H^−min⁡U^Tr⁡U^ρ^U^†(34)\mathcal{E}[\hat{\rho}]=\operatorname{Tr}\hat{\rho}\hat{H}-\min_{\hat{U}}\operatorname{Tr}\hat{U}\hat{\rho}\hat{U}^{\dagger} \tag*{(34)}

where the minimisation is over all unitary operators UU. The passive companion of ρ^\hat{\rho}, ρ˘\breve{\rho}, is known to be given by the expression [1]:

ρ˘=∑iri∣Ei⟩⟨Ei∣(35)\breve{\rho}=\sum_i r_i|E_i\rangle\langle E_i| \tag*{(35)}

where ∣Ei⟩|E_i\rangle denote the eigenvectors of H^\hat{H}, and we adopted the convention that the eigenvalues rir_i of ρ\rho are ordered in non-increasing fashion while the eigenvalues, EnE_n, of HH are ordered in non-decreasing fashion,

E0≤E1≤⋯(36)E_0\leq E_1\leq\cdots \tag*{(36)}
r0≥r1≥⋯(37)r_0\geq r_1\geq\cdots \tag*{(37)}

Accordingly, the ergotropy can be swiftly expressed as:

E[ρ^]=∑i(pi−ri)Ei(38)\mathcal{E}[\hat{\rho}]=\sum_i(p_i-r_i)E_i \tag*{(38)}

where

pi=∑j∣⟨Ei∣rj⟩∣2rj(39)p_i=\sum_j|\langle E_i|r_j\rangle|^2r_j \tag*{(39)}

with ∣ri⟩|r_i\rangle denoting the eigenvectors of ρ^\hat{\rho}.

The energy profiles of ρ^\hat{\rho}, ρ˘\breve{\rho} are given, respectively, by

P(E)=∑ipiδ(E−Ei)(40)\mathcal{P}(E)=\sum_i p_i\delta(E-E_i) \tag*{(40)}
P˘(E)=∑iriδ(E−Ei)(41)\breve{\mathcal{P}}(E)=\sum_i r_i\delta(E-E_i) \tag*{(41)}

They represent the probability density that a projective measurement of the observable HH has outcome EE provided the system is, respectively, in the state ρ^\hat{\rho}, ρ˘\breve{\rho}. Note the analogy with the classical case, where an integral of Dirac deltas gets replaced by a discrete sum of Dirac deltas.

The following holds:

Theorem 2 The ergotropy of ρ^\hat{\rho} is the 1-Wasserstein distance between the energy profile of ρ^\hat{\rho}, and the energy profile of ρ˘\breve{\rho}

E[ρ^]=W1[P,P˘](42)\mathcal{E}[\hat{\rho}] = W_1[\mathcal{P},\breve{\mathcal{P}}] \tag*{(42)}

Using the energy profiles, Eqs. (40),(41), the ergotropy reads exactly like in the classical case, i.e., as:

E=∫0∞dE [P(E)−P˘(E)]E.(43)\mathcal{E} = \int_{0}^{\infty} dE\,[\mathcal{P}(E)-\breve{\mathcal{P}}(E)]E . \tag*{(43)}

Performing the integration by parts as in the classical case, one gets to

E=∫0∞dE [C˘(E)−C(E)](44)\mathcal{E} = \int_{0}^{\infty} dE\,[\breve{C}(E)-C(E)] \tag*{(44)}

where

C(E)=∑ipiθ(E−Ei)(45)C(E) = \sum_i p_i\theta(E-E_i) \tag*{(45)}
C˘(E)=∑iriθ(E−Ei)(46)\breve{C}(E) = \sum_i r_i\theta(E-E_i) \tag*{(46)}

are the cumulative probability functions associated to P\mathcal{P} and P˘\breve{\mathcal{P}}.

Since the probabilities pip_i are linked to the probabilities rjr_j by a bi-stochastic matrix, whose elements are the quantum overlaps ∣⟨Ei∣rj⟩∣2|\langle E_i|r_j\rangle|^2, see Eq. (39), one can prove (see appendix) that C˘(E)≥C(E)\breve{C}(E) \ge C(E). This allows to replace C˘(E)−C(E)\breve{C}(E)-C(E) with ∣C˘(E)−C(E)∣|\breve{C}(E)-C(E)|, in Eq. (43), which concludes the proof.

Illustrative examples

Classical harmonic oscillator

Consider a 1D classical harmonic oscillator

H(q,p)=p22+q22(47)H(q,p)=\frac{p^2}{2}+\frac{q^2}{2} \tag*{(47)}

being in the Gaussian state centered at (q,p)=(a,0)(q,p)=(a,0):

ρ(q,p)=12πσe−(q−a)2+p22σ(48)\rho(q,p)=\frac{1}{2\pi\sigma}e^{-\frac{(q-a)^2+p^2}{2\sigma}} \tag*{(48)}

Its ergotropic rearrangement is

ρ˘(q,p)=12πσe−q2+p22σ(49)\breve{\rho}(q,p)=\frac{1}{2\pi\sigma}e^{-\frac{q^2+p^2}{2\sigma}} \tag*{(49)}

To see that note that ρ\rho can be mapped onto ρ˘\breve{\rho} via a translation, which is a measure preserving transformation. Specifically, ρ˘(q,p)=ρ(M−1(q,p))\breve{\rho}(q,p)=\rho(\mathcal{M}^{-1}(q,p)), where M(q,p)=(q−a,p)\mathcal{M}(q,p)=(q-a,p). Furthermore, ρ˘(q,p)\breve{\rho}(q,p) is a decreasing function of the energy, thus it is the passive companion of ρ\rho.

The system energy is easily calculated by means of Gaussian integrals:

e=12πσ∫dqdp(p22+q22)e−(q−a)2+p22σ=a22+σ(50)e=\frac{1}{2\pi\sigma}\int dqdp\left(\frac{p^2}{2}+\frac{q^2}{2}\right)e^{-\frac{(q-a)^2+p^2}{2\sigma}}=\frac{a^2}{2}+\sigma \tag*{(50)}

Similarly the passive state energy expectation is given by

e~=12πσ∫dqdp(p22+q22)e−q2+p22σ=σ(51)\tilde{e} = \frac{1}{2\pi\sigma}\int dqdp\left(\frac{p^2}{2}+\frac{q^2}{2}\right)e^{-\frac{q^2+p^2}{2\sigma}}=\sigma \tag*{(51)}

Their difference e−e~e-\tilde{e} gives the ergotropy:

E=a22(52)\mathcal{E}=\frac{a^2}{2} \tag*{(52)}

Note the interesting fact that it does not depend on the width σ\sigma of the Gaussian state. Thus the ergotropy is, so to speak, simply the energy H(q,p)H(q,p) of the center of mass (q,p)=(a,0)(q,p)=(a,0) of the Gaussian distribution.

Let us calculate the energy profile P\mathcal{P} of the state ρ\rho. Using the definition in Eq. (7), we have:

P(E)=∫dqdpδ(E−p2/2−q2/2)12πσe−(q−a)2+p22σ(53)\mathcal{P}(E)=\int dqdp\delta(E-p^2/2-q^2/2)\frac{1}{2\pi\sigma}e^{-\frac{(q-a)^2+p^2}{2\sigma}} \tag*{(53)}

To perform the integral we use the celebrated formula [15, 13]:

∫dzδ(E−H(z))f(z)=∫E=HdS∣∇H∣f(z)(54)\int dz\delta(E-H(z))f(z)=\int_{E=H}\frac{dS}{\lvert\nabla H\rvert}f(z) \tag*{(54)}

where ∫E=HdS\int_{E=H}dS denotes integration over the surface H=EH=E. In the case of the harmonic oscillator the surface H=EH=E is simply the circle of radius R=2ER=\sqrt{2E}. Thus dS=Rdϕ=2EdϕdS=Rd\phi=\sqrt{2E}d\phi, where ϕ\phi is the angular coordinate. The gradient of HH is ∇H=(q,p)\nabla H=(q,p), thus ∣∇H∣=q2+p2=2E\lvert\nabla H\rvert=\sqrt{q^2+p^2}=\sqrt{2E}, on the H=EH=E surface. Using q=2Ecos⁡ϕq=\sqrt{2E}\cos\phi, p=2Esin⁡ϕp=\sqrt{2E}\sin\phi we then obtain:

P(E)=e−E/σ2πσe−a22σ∫02πdϕea2Eσcos⁡ϕ(55)\mathcal{P}(E)=\frac{e^{-E/\sigma}}{2\pi\sigma}e^{-\frac{a^2}{2\sigma}}\int_{0}^{2\pi}d\phi e^{\frac{a\sqrt{2E}}{\sigma}\cos\phi} \tag*{(55)}
=e−E/σσe−a22σI0(a2E/σ)(56)=\frac{e^{-E/\sigma}}{\sigma}e^{-\frac{a^2}{2\sigma}}I_0(a\sqrt{2E}/\sigma) \tag*{(56)}

where I0(x)I_0(x) is the modified Bessel function of the first kind [16].

The ground state energy profile P~\tilde{\mathcal{P}} is obtained by setting a=0a=0 in the above expression of P\mathcal{P}, which gives:

P~(E)=e−E/σσ(57)\tilde{\mathcal{P}}(E)=\frac{e^{-E/\sigma}}{\sigma} \tag*{(57)}

Figure 1, left panel, shows plot of P(E)\mathcal{P}(E) for a=2a=2, a=22a=2\sqrt{2} and the according plot of P~(E)\tilde{\mathcal{P}}(E), that is the a=0a=0 energy profile. The chosen value of σ\sigma is σ=1\sigma=1. In the right panel are displayed the according cumulative functions C(E)\mathcal{C}(E). The total shaded area is the ergotropy of the a=22a=2\sqrt{2} state, amounting to E=4\mathcal{E}=4. The darker shaded area, similarly is the ergotropy of the a=2a=2 state, amounting to E=2\mathcal{E}=2. Accordingly, the light shaded area is the energy difference ΔE\Delta E between the states with a=22a=2\sqrt{2} and a=2a=2, which amounts to ΔE=2\Delta E=2.

Figure 1. Left Panel: Energy profiles, P(E)\mathcal{P}(E), of Gaussian states, Eq. (48), and of their passive companion (a=0a = 0), for a classical 1D harmonic oscillator. Right panel: according cumulative functions, C(E)C(E), Eq. (14). All graphs plotted for σ=1\sigma= 1.

Quantum harmonic oscillator

Consider a 1D quantum harmonic oscillator:

H^=p^22+q^22(58)\hat{H} = \frac{\hat{p}^{2}}{2} + \frac{\hat{q}^{2}}{2} \tag*{(58)}

being in the coherent state

∣α⟩=e−∣α∣2/2∑n=0∞αnn!∣En⟩(59)|\alpha\rangle= e^{-|\alpha|^{2}/2} \sum_{n=0}^{\infty} \frac{\alpha^{n}}{\sqrt{n!}} |E_{n}\rangle \tag*{(59)}

where ∣En⟩|E_{n}\rangle denote the eigenstates of H^\hat{H}. The system’s density operator is:

ρ^=∣α⟩⟨α∣(60)\hat{\rho} = |\alpha\rangle\langle\alpha| \tag*{(60)}

Since ρ^\hat{\rho} is a pure state, its passive companion is simply the quantum mechanical ground state

ρ˘=∣E0⟩⟨E0∣(61)\breve{\rho} = |E_{0}\rangle\langle E_{0}| \tag*{(61)}

The probability pnp_n of finding the eigenenergy En=(n+12)E_n = \left(n + \frac{1}{2}\right) is given by the Poisson distribution:

pn=e−∣α∣2∣α∣2nn!(62)p_n = e^{-|\alpha|^{2}} \frac{|\alpha|^{2n}}{n!} \tag*{(62)}

The according energy expectation is

e=∑npnEn=∣α∣2+12(63)e = \sum_n p_n E_n = |\alpha|^{2} + \frac{1}{2} \tag*{(63)}

The probability rnr_n of finding the eigenenergy EnE_n in the ground state ρ˘\breve{\rho} is trivially 1 if n=0n = 0, and null otherwise:

rn=δ0,n(64)r_n = \delta_{0,n} \tag*{(64)}

where δa,b\delta_{a,b} is the Kronecker symbol. The according energy expectation is simply the ground state energy

e˘=12(65)\breve{e} = \frac{1}{2} \tag*{(65)}

Thus the ergotropy amounts to

E=∣α∣2(66)\mathcal{E} = |\alpha|^2 \tag*{(66)}

As shown in Figure 2 the cumulative functions (45,46) are, in this specific case, histograms made of bars of width 1 (except for the n=0n = 0 bar, which is of width 1/2). Accordingly, the 1-Wasserstein distance is easily written as

Figure 2. Left Panel: Energy eigenstates populations, pnp_n, of coherent states, Eq. (62), of a 1D quantum harmonic oscillator. Right panel: according cumulative functions, C(E)C(E), Eq. (45), and the cumulative function C˘(E)\breve{C}(E) of the ground state (i.e. the α=0\alpha= 0 curve).

W1[P,P˘]=∑n=0∞∑k=0n(rk−pk)=∑n=0∞[1−e−∣α∣2∑k=0n∣α∣2kk!](67)W_1[\mathcal{P},\breve{\mathcal{P}}] = \sum_{n=0}^{\infty}\sum_{k=0}^{n}(r_k-p_k) = \sum_{n=0}^{\infty}\left[1-\mathrm{e}^{-|\alpha|^2}\sum_{k=0}^{n}\frac{|\alpha|^{2k}}{k!}\right] \tag*{(67)}

As shown in the appendix, the above sum amounts to ∣α∣2|\alpha|^2, i.e., the ergotropy, in agreement with our main result.

Remarks

Given that the ergotropy is the 1-Wasserstein distance between the two energy profiles P,P˘\mathcal{P},\breve{\mathcal{P}}, one might wonder whether ergotropy extraction protocols realise optimal transport plans in energy space. The answer is in the negative.

In 1D the optimal transport plan γˉ\bar{\gamma} that solves the optimisation problem in Eq. (9) is given by the Monge map T=c1−1∘c2T=c_1^{-1}\circ c_2, meaning that the two random variables x,yx,y are delta-correlated [14]:

γˉ(x,y)=Q2(y)δ[x−c1−1(c2(y))](68)\bar{\gamma}(x,y)=Q_2(y)\delta[x-c_1^{-1}(c_2(y))] \tag*{(68)}

In our specific case it means that the optimal transport plan Jˉ(E1,E2)\bar{\mathcal{J}}(E_1,E_2) reads

Jˉ(E1,E2)=P˘(E2)δ[E1−C−1(C˘(E2))](69)\bar{\mathcal{J}}(E_1,E_2)=\breve{\mathcal{P}}(E_2)\delta[E_1-C^{-1}(\breve{C}(E_2))] \tag*{(69)}

However, in the present case the two random quantities E1,E2E_1,E_2 are generally not delta correlated. In the classical case, the joint probability density J(E1,E2)\mathcal{J}(E_1,E_2) can be inferred from the conditional probability density G[Θ∣Φ]G[\Theta|\Phi], Eq. (23), that the system was on the energy hypersurface that encloses the volume Φ\Phi at the beginning, given that it is found on the energy hypersurface that encloses the volume Θ\Theta after ergotropy extraction. Thus the joint probability density of Θ,Φ\Theta,\Phi reads:

J(Θ,Φ)=P~(Θ)G[Θ∣Φ](70)\mathcal{J}(\Theta,\Phi)=\widetilde{P}(\Theta)G[\Theta|\Phi] \tag*{(70)}

Note that its marginals are, correctly, P(Θ)P(\Theta) and P~(Φ)\widetilde{P}(\Phi), see Eqs. (22),(24). Making the usual change of variables Φ=Ω(E1),Θ=Ω(E2)\Phi=\Omega(E_1),\Theta=\Omega(E_2), we get:

J(E2,E1)=P~(Ω(E2))G[Ω(E1)∣Ω(E2)]ω(E1)(71)\mathcal{J}(E_2,E_1)=\widetilde{P}(\Omega(E_2))G[\Omega(E_1)|\Omega(E_2)]\omega(E_1) \tag*{(71)}

where we note that GG is typically not a δ\delta function (see [17] for an example). Thus the optimal transport plan J‾\overline{\mathcal{J}} does not, in general, coincide with the actual ergotropy extraction “plan” J\mathcal{J}. This is so because the ergotropic transport plan J\mathcal{J} is subject to the constraint that is induced by a symplectic diffeomorphism in phase space, whereas the optimisation in Eq. (9) is merely carried over all transport plans that are compatible with the marginals P,P~\mathcal{P},\widetilde{\mathcal{P}}.

A similar argument applies as well in the quantum case.

Acknowledgements

The sparkle that ignited the idea of this work was lit at the “Quantum Hike conference 2026”, Val D’Ayas, Italy. The author gratefully thanks the organisers (Sabrina Maniscalco, Lorenzo Maccone, Paola Verrucchi, Paolo Zanardi, Karol Życzkowski) and the sponsor (Algorithmiq) for the opportunity to participate. The author thanks Karen V. Hovhannisyan for his valuable remarks on the manuscript.

Use of artificial intelligence

This work was solely produced by the author, without any aid of artificial intelligence.

Proof Eq. (32) in the quantum case

In the quantum case the function C(E)\mathcal{C}(E) is a histogram made of bars of variable widths and heights. The ii-th bar has width Δi≡(Ei+1−Ei)\Delta_i\equiv(E_{i+1}-E_i) and height si=∑j=0ipjs_i=\sum_{j=0}^{i}p_j. The bars that compose the histogram C~(E)\widetilde{\mathcal{C}}(E) have the same widths Δi≡(Ei+1−Ei)\Delta_i\equiv(E_{i+1}-E_i) and their heights are s~i=∑j=1irj\widetilde{s}_i=\sum_{j=1}^{i}r_j. Because the overlaps ∣⟨Ei∣rj⟩∣2|\langle E_i|r_j\rangle|^2 form a bistochastic matrix, Eq. (39) implies that the distribution rir_i majorises the distribution pip_i [18], namely

∑j=0ipj↓≤∑j=0irj↓(72)\sum_{j=0}^{i}p_j^\downarrow\le\sum_{j=0}^{i}r_j^\downarrow \tag*{(72)}

where pj↓p_j^\downarrow denotes the decreasing rearrangement of the distribution pjp_j. Note that by definition of decreasing rearrangement, it is pj≤pj↓p_j\le p_j^\downarrow. Furthermore, from Eq. (37) we have rj↓=rjr_j^\downarrow=r_j. Therefore

si=∑j=0ipj≤∑j=0ipj↓≤∑j=0irj↓=∑j=0irj=s~i(73)s_i=\sum_{j=0}^{i}p_j\le\sum_{j=0}^{i}p_j^\downarrow\le\sum_{j=0}^{i}r_j^\downarrow=\sum_{j=0}^{i}r_j=\widetilde{s}_i \tag*{(73)}

Thus each bar of the histogram of CC is not higher than the corresponding bar of C˘\breve{C}, meaning that C(E)≤C˘(E)C(E) \le\breve{C}(E).

Proof of Eq. (67)

Let

f(x)=∑n=0∞[1−e−x∑k=0nxkk!](74)f(x) = \sum_{n=0}^{\infty} \left[1-e^{-x}\sum_{k=0}^{n}\frac{x^k}{k!}\right] \tag*{(74)}

We want to prove that f(x)=xf(x)=x for x≥0x \ge0. In order to do so we first show that f′(x)=1f'(x)=1. Noting that

e−x∑k=0∞xkk!=1(75)e^{-x}\sum_{k=0}^{\infty}\frac{x^k}{k!}=1 \tag*{(75)}

we can rewrite ff as:

f(x)=e−x∑n=0∞∑k=n+1∞xkk!(76)f(x)=e^{-x}\sum_{n=0}^{\infty}\sum_{k=n+1}^{\infty}\frac{x^k}{k!} \tag*{(76)}

Let us take its derivative:

f′(x)=−e−x∑n=0∞∑k=n+1∞xkk!+e−x∑n=0∞∑k=n+1∞xk−1(k−1)!=−f(x)+e−x∑n=0∞∑k=n∞xkk!=−f(x)+e−x∑n=0∞[xnn!+∑k=n+1∞xkk!]=−f(x)+1+f(x)=1(77)\begin{aligned} f'(x)&=-e^{-x}\sum_{n=0}^{\infty}\sum_{k=n+1}^{\infty}\frac{x^k}{k!}+e^{-x}\sum_{n=0}^{\infty}\sum_{k=n+1}^{\infty}\frac{x^{k-1}}{(k-1)!}\\ &=-f(x)+e^{-x}\sum_{n=0}^{\infty}\sum_{k=n}^{\infty}\frac{x^k}{k!}\\ &=-f(x)+e^{-x}\sum_{n=0}^{\infty}\left[\frac{x^n}{n!}+\sum_{k=n+1}^{\infty}\frac{x^k}{k!}\right]\\ &=-f(x)+1+f(x)\\ &=1 \tag*{(77)} \end{aligned}

Furthermore, we have:

lim⁡x→0f(x)=0(78)\lim_{x\to0}f(x)=0 \tag*{(78)}

To prove it, note that xk/k!→δ0,kx^k/k! \to\delta_{0,k} for x→0x\to0, thus e−x∑k=0nxk/k!→1e^{-x}\sum_{k=0}^{n}x^k/k! \to1 in Eq. (B.1). Integrating the equation f′(s)=1f'(s)=1 between 00 and xx, with the condition lim⁡x→0f(x)=0\lim_{x\to0}f(x)=0, we obtain f(x)=xf(x)=x.

References

  1. [1]A E Allahverdyan, R Balian and Th M Nieuwenhuizen 2004 Europhys. Lett. 67 565–571
  2. [2]Campisi M 2026 Europhysics Letters 154 50002
  3. [3]Hatsopoulos G N and Gyftopoulos E P 1976 Found. Phys. 6 127–141
  4. [4]Gardner C S 1963 The Physics of Fluids 6 839–840 URL https://doi.org/10.1063/1.1706823
  5. [5]Helander P 2017 Journal of Plasma Physics 83 715830401
  6. [6]Lenard A 1978 J. Stat. Phys. 19 575–586
  7. [7]Pusz W and Woronowicz S L 1978 Comm. Math. Phys. 58 273–290 URL https://doi.org/10.1007/BF01614224
  8. [8]Górecki J and Pusz W 1980 Letters in Mathematical Physics 4 433–443 URL https://doi.org/10.1007/BF00943428
  9. [9]Daniëls H A M 1981 J. Math. Phys. 22 843–846 ISSN 0022-2488 URL https://doi.org/10.1063/1.524949
  10. [10]Mackenbach R J J, Proll J H E and Helander P 2022 Phys. Rev. Lett. 128 175001
  11. [11]Lemou M, Méhats F and Raphaël P 2012 Inventiones mathematicae 187 145–194 URL https://doi.org/10.1007/s00222–011–0332–9
  12. [12]Campisi M 2026 arXiv:2603.28388arxiv.org/abs/2603.28388
  13. [13]Campisi M 2025 Lectures on the Mechanical Foundations of Thermodynamics 2nd ed SpringerBriefs in Physics (Cham: Springer)
  14. [14]Vallender S S 1974 Theory of Probability and Its Applications 18 784–786 URL https://doi.org/10.1137/1118101
  15. [15]Khinchin A 1949 Mathematical Foundations of Statistical Mechanics (New York: Dover)
  16. [16]IS Gradshteyn I R 1980 Table of Integrals, Series, and Products (Elsevier Academic Press)
  17. [17]Campisi M 2008 Phys. Rev. E 78 051123
  18. [18]Sagawa T 2020 Entropy, Divergence, and Majorization in Classical and Quantum Thermodynamics SpringerBriefs in Mathematical Physics (Springer Singapore)

Paper details

vlnc:W_fa6f9mgy
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

None listed by the authors

View statement
Looking for related papers…
Contents