betaFeedback

WorksPhysics and Astronomy

A No-Go Theorem of Analytical Mechanics for the Second Law Violation

Abstract

We follow the Boltzmann-Clausius-Maxwell (BCM) proposal to solve a long-standing problem of identifying the underlying cause of the second law (SL) of spontaneous irreversibility, a stochastic universal principle, as the mechanical equilibrium (stable or unstable) principle (Mec-EQ-P) of analytical mechanics of an isolated nonequilibrium system of any size. The principle leads to nonnegative system intrinsic (SI) microwork and SI-average macrowork dW during any spontaneous process. In conjuction with the first law, Mec-EQ-P leads to a generalized second law (GSL) dQ=dW>0, where dQ=TdS is the purely stochastic SI-macroheat that corresponds to dS>0 for T>0 and dS<0 for T<0, where T is the temperature. The GSL supercedes the conventional SL formulation that is valid only for a macroscopic system for positive temperatures temperatures, but reformulates it to dS<0 for negative temperatures. It is quite surprising that GSL is not only a direct consequence of intertwined mechanical and stochastic macroquantities through the first law but also remains valid for any arbitrary irreversible process in a system of any size as an identity for positive and negative temperatures. It also becomes a no-go theorem for GSL-violation unless we abandon Mec-EQ-P of analytical mechanics used in the BCM proposal, which will be catastrophic for theoretical physics. In addition, Mec-EQ-P also provides new insights into the roles of spontaneity, nonspontaneity, negative temperatures, instability, and the significance of dS<0 due to nonspontaneity and inserting internal constraints.

Even though Einstein [1] was convinced that classical thermodynamics (Cl-Th) as ” ... the only physical theory of universal content ... will never be overthrown,” and Eddington [2] considered the fundamental axiom of the second law (SL) in terms of entropy SS for an isolated system Σ\Sigma as its cornerstone holds ” ... the supreme position among the laws of Nature,” the recent trend has been to search for SL-violation [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15] [16] in Σ\Sigma, whose size ranges from mesoscopic to cosmological scales, to cast doubt on Cl-Th of a nonequilibrium (NEQ) macrostate M\mathcal{M}, following Maxwell [17]; here, M≔{mk,pk}\mathcal{M}\coloneqq\{m_k,p_k\} is formed by microstates mk\mathbf{m}_k of the Hamiltonian H\mathcal{H} with microenergies EkE_k appearing with probabilities pkp_k, k=1,2,⋯k=1,2,\cdots. These attempts span widely different fields from information to biological thermodynamics. Do positive and negative temperatures TT both satisfy SL as is commonly believed [18, for example]?

There are indirect but strong arguments for a deep connection of SL with thermodynamic stability [19–30], in which a (stable) NEQ macrostate Ms\mathcal{M}^{\mathrm s} with a lower bound in energy, the continuous blue curve in Figure 1, asymptotically converges to a unique and stable equilibrium (SEQ) macrostate Mseq\mathcal{M}_{\mathrm{seq}} [30] along blue arrows. Thus, a SL-violation strongly suggests, but not yet verified, thermodynamic instability for its cause so we also consider rarely studied unstable macrostate Mu\mathcal{M}^{\mathrm u}, the continuous red curve, emerging out of its unstable equilibrium (UEQ) macrostate Mueq\mathcal{M}_{\mathrm{ueq}} along red arrows; however, see [18] as exception. All arrows point towards increasing time tt, during which Ms\mathcal{M}^{\mathrm s} (or Mu\mathcal{M}^{\mathrm u}) becomes less (or more) nonuniform with the entropy increasing (decreasing), with its evolution controlled by its sink Mseq\mathcal{M}_{\mathrm{seq}} (or source Mueq\mathcal{M}_{\mathrm{ueq}}) as succinctly explained in Supplementary Material [31]. We use compact notation Q for (Qs,Qu)(Q^{\mathrm s},Q^{\mathrm u}), and QeqQ_{\mathrm{eq}} for (Qseq,Queq)(Q_{\mathrm{seq}},Q_{\mathrm{ueq}}). Thus, we say that Meq\mathcal{M}_{\mathrm{eq}} controls M:(Ms,Mu)\mathcal{M}:(\mathcal{M}^{\mathrm s},\mathcal{M}^{\mathrm u})-evolution along continuous curves, while Mkeq\mathcal{M}_{\mathrm{keq}} controls mk:(mks,mku)m_k:(m_k^{\mathrm s},m_k^{\mathrm u})-evolution along dashed-dot trajectories. We also let Q denote EkE_k, EkeqE_{\mathrm{keq}}, and FkF_k, the last two defined later.

Figure 1. Schematic forms of microenergy EkE_k (dashed-dot curves) and macrowork function EwE_w (solid curves) as functions of the internal variable ξ\xi, with ξ=0\xi= 0 denoting EQ in SZ\mathcal{S}_{Z}. Alternatively, these curves can be considered as a function of time tt in SX\mathcal{S}_{X}, which increases along the directions of the blue and red arrows. We only consider the case when each curve has a single extremum. The discussion is easily extended to more complex forms. The blue color curves and solid blue arrows represent the evolution controlled by the stable (s) case. The red color curves and solid red arrows represent the evolution controlled by the unstable (u) case. In both cases, the arrows lower the energy. The extrema of all curves occur at ξ=0\xi= 0 and represent a uniform body. For the extremum to denote equilibrium, we must also have ξ=0\xi= 0 there. The green double-arrow is discussed in the text.

All processes in Σ\Sigma are system-intrinsic (SI) and generated internally to change EkE_k. They are commonly called spontaneous [32] and are governed by SL, which is only satisfied by a macroscopic system ΣM\Sigma_{\mathrm M} [20, 24–26]. Any SL-violation, being uncommon and controversial, will most probably result in puzzling outcomes, not all of which are recognized or discussed so far in the current literature. Also, there is no serious inquiry into violation thermodynamics (Viol-Th) supporting it because the root cause of SL and any special stochasticity [33] for its validity are not understood; however, see [18]. A nonspontaneous process is not internally generated and requires external agent Σext\Sigma_{\mathrm{ext}} so cannot occur in Σ\Sigma.

We prove a generalized SL (GSL), not to be confused with the one proposed by Bekenstein [34, 35] for black holes, by considering M\mathcal{M} in an isolated NEQ discrete system Σ\Sigma [36, 37] of any size, not necessarily ΣM\Sigma_{\mathrm M}. It unravels the mystery of and provides deeper insight into the root cause of SL dS≥0dS\geq0 and its violation dS<0dS<0 [38]. The equality occurs in the equilibrium (EQ) macrostate Meq\mathcal{M}_{\mathrm{eq}} having maximum entropy [39], and SS is always given by Gibbs’ formulation [19, 40] S≔−∑kpkln⁡pkS\coloneqq-\sum_k p_k\ln p_k. There is another system ΣE\Sigma_{\mathrm E} of intermediate size determined by the interaction range λE\lambda_{\mathrm E} to ensure EkE_k-additivity in M\mathcal{M}, which is uniform in the state space SX\mathcal{S}_{\mathrm X}, while Meq\mathcal{M}_{\mathrm{eq}} is uniform as expressed by H\mathcal{H} [31]; here, X≔(E,w)\mathbf X\coloneqq(E,\mathbf w) is the fixed observable set containing energy EE and work parameter w\mathbf{w} that uniquely specifies Meq\mathcal{M}_{\mathrm{eq}} and H(w)\mathcal{H}(\mathbf{w}). There are two distinct ways to specify M\mathcal{M}. The most common approach is to use SX\mathcal{S}_{X} to nonuniquely specify M\mathcal{M} so that its various quantities such as S(X,t)S(X,t) or H(w,t)\mathcal{H}(w,t) have explicit tt dependence. In SX\mathcal{S}_{X}, we can focus on a Σ\Sigma of any size [31]. The alternative is to uniquely specify M=M(w,ξ)=M(ξ)\mathcal{M}=\mathcal{M}(w,\xi)=\mathcal{M}(\xi) and H(w,ξ)=H(ξ)\mathcal{H}(w,\xi)=\mathcal{H}(\xi) in an extended state space SZ\mathcal{S}_{Z} spanned by Z≔(E,W)\mathbf{Z}\coloneqq(E,\mathbf{W}), W=(w,ξ)\mathbf{W}=(w,\xi) the extended work parameter including the internal variable ξ\xi [41–46] that is determined by internal structures generated by nonuniformity in ΣE\Sigma_E and ΣM\Sigma_M. It contains ι\iota components, with ι\iota increasing and SS decreasing with nonuniformity of M(ξ)\mathcal{M}(\xi). There is no explicit time dependence in S(Z)S(Z) and H(ξ)\mathcal{H}(\xi) in SZ\mathcal{S}_{Z} that provides some benefit. We define ξ\xi so that ξ=0\xi=0 in Meq\mathcal{M}_{\mathrm{eq}} [31, (B)] for which ι=0\iota=0. We also assume that ξ\xi is the same for all mkm_k's; extension to different ξk\xi_k is easily done. As explicit time dependence in Σ\Sigma in SX\mathcal{S}_{X} is equivalent to the implicit time dependence in ΣE\Sigma_E and ΣM\Sigma_M in SZ\mathcal{S}_{Z}, we can adopt either approach. We mostly use Σ\Sigma and the first approach for its simplicity. We use Σ\Sigma for any system when the state space is not specified.

The four steps below form the core of our analysis:

(S1) Describe Σˉ\bar{\Sigma} mechanically by its deterministic H\mathcal{H} by specifying {mk,Ek}\{m_k,E_k\} and SI-microwork {dWk}≔−{dEk}\{dW_k\}\coloneqq-\{dE_k\}.

(S2) Introduce stochasticity by appending a probability pkp_k to mkm_k as was first proposed by Clausius, Maxwell, and Boltzmann (the BCM proposal) [17, 19, 20, 25, 30, 47, 48] to capture dissipation in M\mathcal{M}, and to identify various ensemble averages ⟨∙⟩\langle\bullet\rangle such as its energy

E≔∑kpkEk,E\coloneqq\sum_k p_kE_k,

and SS above. In general, SS is a state function only for ΣM\Sigma_M in SZ\mathcal{S}_{Z}, but not for Σ\Sigma and ΣE\Sigma_E. Different choices for {pk}\{p_k\} result in different macrostates and averages for the same {mk}\{m_k\}. A constant pkp_k, ∀k\forall k, describes a pure mechanical system [33] with constant EE and SS.

(S3) Introduce the first law dE=dQ−dWdE=dQ-dW [31, (D)] during any infinitesimal change of Σˉ\bar{\Sigma} to determine the allowed change d{pk}d\{p_k\} and to identify the physics behind

dQ≔∑kEk dpk,dQ\coloneqq\sum_k E_k\,dp_k,

a stochastic quantity [33], and

dW≔−∑kpkdEk=∑kpkdWk,dW\coloneqq-\sum_kp_kdE_k=\sum_kp_kdW_k,

a mechanical quantity. The law immediately leads to the irreversibility principle (Irr-P) [44, 45] expressed by the identity

dQ=dW≥0(1)dQ=dW\geq0 \tag*{(1)}

that intertwines these seemingly unrelated macroquantities similar to intertwining of electric and magnetic fields in the Maxwell theory. While they have the same sign, their sign is fixed yet. We use (S1), (S2), the first law involving dQdQ and dWdW, and dSdS to formulate a generic version of NEQ statistical thermodynamics to be called General Thermodynamics (Gen-Th) for any possible M\mathcal{M} by taking arbitrary {pk}\{p_k\} for Σ\Sigma; see Eq. (3).

(S4) Invoke the mostly overlooked fundamental principle Mec-EQ-P [49, 50] introduced below that controls (spontaneous) evolution of mkm_k, ∀k\forall k, in Σ\Sigma [32] to formulate the BCM proposal for mkm_k (MicroBCM) that fixes the nonnegative signature of dWkdW_k in Lemma 1, and to directly establish the generalized second law (GSL) that controls dissipation [33] in Σˉ\bar{\Sigma} in Theorem 2:

dQ=dW≥0;dWk≥0, ∀k.(2)dQ=dW\geq0; dW_k\geq0,\ \forall k. \tag*{(2)}

It requires {dpk}\{dp_k\} to take a special form {dpkGSL}\{dp_k^{\mathrm{GSL}}\} that not only determines dSdS but also ensures dQ≥0dQ\geq0; any form different from {dpkGSL}\{dp_k^{\mathrm{GSL}}\} such as {−dpkGSL}\{-dp_k^{\mathrm{GSL}}\} does not satisfy GSL and must be rejected under MicroBCM proposal. The nonnegative signature of dWdW after including (S4) in Gen-Th determines the (spontaneous) evolution of M\mathcal{M}, and the resulting thermodynamics is denoted by Gen-GSL-Th. We combine Eqs. (1) and (2) as

TdS=dW{≥0in Gen-Th≥0in Gen-GSL-Th,T≔dQdS∣E,(3)TdS=dW \begin{cases} \geq0 & \text{in Gen-Th}\\ \geq0 & \text{in Gen-GSL-Th} \end{cases} ,\qquad T\coloneqq\left.\frac{dQ}{dS}\right|_{E}, \tag*{(3)}

in terms of the SI-temperature TT for any Σˉ\bar{\Sigma}, which for ΣM\Sigma_M [31, (E)] becomes the standard definition (∂E/∂S)ξ(\partial E/\partial S)_{\xi}.

The traditional approach in CI-Th is to supplement (S1), (S2), (S3) by the SL axiom for ΣM\Sigma_M [24] instead of (S4) and then prove its thermodynamic stability [25]. However, by reversing this approach, we succeed not only to prove GSL for any spontaneous process in Σˉ\bar{\Sigma} and for any {pk}\{p_k\} as a direct consequence of mkm_k-evolution in (S4), but also proclaim a No-Go Theorem of theoretical physics for its violation; see Theorem 2. Our proof should be contrasted with the current situation of proving dS≥0dS \ge0 without ever mentioning TT; one usually recourses to ad-hoc assumptions like molecular chaos or master equations [57, 58]. The real significance of our reverse approach becomes very transparent and unravels many mysteries of SL by recognizing fluctuating dWkdW_k as the primitive mechanical concept that completely captures the stochasticity in dQdQ and the importance of TT for SL. Their intertwining provides the basis for the famous fluctuation-dissipation theorem [51–53]; see also [54–56]. To our knowledge, our reverse approach with MicroBCM proposal has never been used before to directly prove SL (dS≥0dS \ge0) for T≥0T \ge0 and its extension requiring a reformulation of dS<0dS < 0 for T<0T < 0; the latter contradicts the conventional wisdom [18]. In addition, GSL also disproves that instability causes SL violation.

Setup: We describe the salient aspects [31] to prove Theorem 2 for Σˉ\bar{\Sigma} in Gen-GSL-Th before justifying them. We assume [59] for simplicity that EkE_k has only a single extremum at EkeqE_{k\mathrm{eq}} as shown by the dashed-dotted curves in Figure 1 for mksm_k^{s} and mkum_k^{u}, each spontaneously evolving to lower microenergies as tt increases along the blue and red arrows, respectively. From the Uniformity Theorem of mkeqm_k^{\mathrm{eq}} in [31, (C)], the extremum represents a uniform microstate mkeqm_{k\mathrm{eq}} so the rest of the dashed-dotted curves denote nonuniform microstates mkm_k (mksm_k^{s} or mkum_k^{u}). Then, mkseqm_{k\mathrm{seq}} is the (asymptotically) stable equilibrium of mksm_k^{s}, where EkseqE_{k\mathrm{seq}} is minimum. Any perturbation away from mkseqm_{k\mathrm{seq}} always restores mksm_k^{s} back to mkseqm_{k\mathrm{seq}} so the latter acts as a sink for which Eks→EkseqsE_k^{s} \to E_{k\mathrm{seq}}^{s} spontaneously as shown by the blue arrow; we say that mkseqm_{k\mathrm{seq}} controls mksm_k^{s}-evolution [52] during which mksm_k^{s} becomes more uniform with dS>0dS > 0. At a mechanically unstable equilibrium mkueqm_{k\mathrm{ueq}} of mkum_k^{u}, EkueqE_{k\mathrm{ueq}} is maximum; any perturbation from it to mkum_k^{u} spontaneously repels it further away from mkueqm_{k\mathrm{ueq}}. As mkum_k^{u} never returns to mkueqm_{k\mathrm{ueq}} with EkuE_k^{u} running away from EkueqE_{k\mathrm{ueq}}, mkueqm_{k\mathrm{ueq}} becomes a source controlling mkum_k^{u}-evolution along the red arrow during which mkum_k^{u} becomes more nonuniform with dS<0dS < 0. Both are deterministic spontaneous mkm_k-evolutions, whose directions are controlled by the principle of mechanical equilibrium (Mec-EQ-P) [31, (C)].

In Gen-GSL-Th for Σˉ\bar{\Sigma}, the directional mkm_k-evolution performs dWk=−dEk≥0dW_k=-dE_k\ge0 (Lemma 1), whose average dW≥0dW\ge0 is due to spontaneous irreversible processes [60] and from Irr-P is dissipated or wasted [33] as macroheat dQdQ to ensure unchanging EE. Without (S4), Gen-Th also describes GSL-violation (dQ=dW<0dQ=dW<0) and SL-violation (dS<0dS<0), resulting in violation thermodynamics, Viol-GSL-Th of GSL and Viol-Th of SL, for Σˉ\bar{\Sigma}; see [31, (C)] for various possible thermodynamics.

To make SS a state function for ΣM\Sigma_M in GZ\mathfrak{G}_Z requires an important restriction on {pk}\{p_k\}, but not on {dpk}\{dp_k\}. The resulting form of Gen-Th is called restriction thermodynamics (Rest-Th) for ΣM\Sigma_M [31, (E)}, which still allows dSdS of either sign so dQ=TdS⋛0dQ=T dS\gtreqless0 in Rest-Th, unless supplemented by (S4); see Eq. (3).

JUSTIFICATION: We justify the above claims for Σˉ\bar{\Sigma} in GX\mathfrak{G}_X within Gen-GSL-Th without any restriction on possible {pk}\{p_k\}. The claims also apply to ΣE\Sigma_E and ΣM\Sigma_M in GZ\mathfrak{G}_Z without any change. A simple calculation at the end clarifies the claims for a one-particle system in GX\mathfrak{G}_X.

The extremum and the form of {Ek}\{E_k\} determine the extremum and the form of EwE_w (Ew=Ews,EwuE_w=E_w^s,E_w^u) after averaging [31, (D)] for M\mathfrak{M} as shown by continuous curves in Figure 1. The averaging also generalizes microstate Mec-EQ-P to the thermodynamical principle of (stability and instability) equilibrium (Th-EQ-P) for M\mathfrak{M} in Gen-GSL-Th, according to which the extremum EeqE_{\mathrm{eq}} of EwE_w controls M\mathfrak{M}-evolution. We must not confuse this evolution with that produced by intervention required to prepare nonuniform M\mathfrak{M} out of uniform Meq\mathfrak{M}_{\mathrm{eq}} by internal constraints as discusses below and in [31]; see also [24, 61]. As mkueqm_{k\mathrm{ueq}} is, in effect, physically equivalent to a “nonexistent” microstate because of its instability, Mueq\mathfrak{M}_{\mathrm{ueq}} is also physically nonexistent [22, 62, 63]. Because of this, CI-Th only deals with Ms\mathfrak{M}^{s} having the sink Mseq\mathfrak{M}_{\mathrm{seq}} to which it asymptotically converges. Accordingly any mks∈Msm_k^{s}\in\mathfrak{M}^{s} converges to mkseq∈Mseqm_{k\mathrm{seq}}\in\mathfrak{M}_{\mathrm{seq}}. Despite this, we also consider Mu\mathfrak{M}^{u} to obtain additional and surprising information at the microstate level that is not available in CI-Th, and allows for a reformulation of SL for negative TT.

We first prove the following lemma for an equilibrium point of a mechanical mkm_k and a thermodynamic M\mathfrak{M}.

Lemma 1 During mkeqm_{k\mathrm{eq}}-controlled spontaneous evolution of mkm_k of Σˉ\bar{\Sigma} in Gen-GSL-Th, mkm_k performs nonnegative microwork ΔWk\Delta W_k as mk→mk′m_k\to m'_k. Performing ensemble average with arbitrary {pk}\{p_k\} then determines thermodynamic stability (instability) of the resulting macrostate M=(Ms,Mu)\mathfrak{M}=(\mathfrak{M}^{s},\mathfrak{M}^{u}) in Σˉ\bar{\Sigma}, which spontaneously performs nonnegative macrowork ΔW\Delta W in Gen-GSL-Th as M→M′\mathfrak{M}\to\mathfrak{M}'.

Proof. (a) Microstate Evolution: The extremum of EkE_k at EkeqE_{k\mathrm{eq}} represents mkeq=(mkseq,mkueq)m_{k\mathrm{eq}}=(m_{k\mathrm{seq}},m_{k\mathrm{ueq}}) that is uniform in GX\mathfrak{G}_X; see Uniformity Theorem in [31, (C)]. We now consider a nonuniform mk=(mks,mku)m_k=(m_k^{s},m_k^{u}) away from mkeqm_{k\mathrm{eq}} as it spontaneously evolves towards mk′=(mk′s,mk′u)m'_k=(m_k^{\prime s},m_k^{\prime u}), see red and blue arrows in Figure 1, during which EkE_k spontaneously decreases to Ek′E'_k so that ΔEk=−ΔWk≔Ek′−Ek≤0\Delta E_k=-\Delta W_k\coloneqq E'_k-E_k\le0. Thus, mkm_k performs spontaneous nonnegative microwork during its evolution towards mk′m'_k given by

ΔWk=−ΔEk≥0.(4)\Delta W_k=-\Delta E_k\ge0. \tag*{(4)}

(b) Macrostate Evolution: We now average over mksm_k^{s} and mkum_k^{u} using arbitrary pkp_k to obtain macroquantities of Ms\mathfrak{M}^{s} and Mu\mathfrak{M}^{u}, respectively, in Gen-GSL-Th. The mechanical equilibrium microstate mkeqm_{k\mathrm{eq}} determines the thermodynamic EQ macrostate Meq=(Mseq,Mueq)\mathfrak{M}_{\mathrm{eq}}=(\mathfrak{M}_{\mathrm{seq}},\mathfrak{M}_{\mathrm{ueq}}). The macrowork function EwE_w gives Eeq≔⟨Eeq⟩E_{\mathrm{eq}}\coloneqq\langle E_{\mathrm{eq}}\rangle for thermodynamic EQ and uniform macrostate Meq\mathfrak{M}_{\mathrm{eq}} in GX\mathfrak{G}_X [31, (C)]. The spontaneous process M→M′\mathfrak{M}\to\mathfrak{M}' results in the nonnegative spontaneous macrowork

ΔW≔⟨ΔWk⟩=−⟨ΔEk⟩≥0.(5)\Delta W\coloneqq\langle\Delta W_k\rangle=-\langle\Delta E_k\rangle\ge0. \tag*{(5)}

This completes the proof. ■

We recall that there is no sign restriction on ΔWk\Delta W_k and ΔW\Delta W in Gen-Th of Σ\Sigma. We now prove the main Theorem.

Theorem 2 We consider Σ\Sigma in SX\mathcal{S}_{X}. The spontaneous Meq\mathcal{M}_{eq}-evolution of any M\mathcal{M} in Gen-GSL-Th during which it either converges to the sink Mseq\mathcal{M}_{seq} for Ms\mathcal{M}^{s} or runs away from the source Mueq\mathcal{M}_{ueq} for Mu\mathcal{M}^{u} directly leads to GSL in Eq. (2), making it internally consistent with analytical mechanics. A violation of GSL requires ΔW<0\Delta W < 0 that cannot occur in Σ\Sigma so the theorem is a No-Go Theorem for GSL violation. The ΔS<0\Delta S < 0 due to instability in the spontaneous evolution in Mu\mathcal{M}^{u} is not a violation of SL because of its negative TT. This makes Viol-GSL-Th and Viol-Th inconsistent with analytical mechanics and the BCM proposal.

Proof. In Gen-GSL-Th, the isolated Σ\Sigma in M≠Meq\mathcal{M} \ne\mathcal{M}_{eq} either spontaneously relaxes as tt increases to Mseq\mathcal{M}_{seq} during which dS≥0dS \ge0 or runs off from Mueq\mathcal{M}_{ueq} during which dS≤0dS \le0. As M→M′\mathcal{M} \to\mathcal{M}', we have ΔW≥0\Delta W \ge0 from Lemma 1. It then follows from Eq. (2) that the corresponding spontaneous macroheat ΔQ=∫MM′T dS\Delta Q = \int_{\mathcal{M}}^{\mathcal{M}'} T\,dS is nonnegative, which proves GSL in Eq. (3) for any M\mathcal{M} in Σ\Sigma, and establishes its consistency with Mech-EQ-P of analytical mechanics. A GSL-violation dQ<0dQ < 0 can only happen if dW<0dW < 0 such as along the double green arrow in Figure 1 near Mseq\mathcal{M}_{seq}. As this evolution violates Lemma 1, it is nonspontaneous and cannot occur in Σ\Sigma. This makes any GSL violation impossible and turns the theorem into a No-Go theorem for GSL-violation. The requirement dQ≥0dQ \ge0 is consistent with T≥0T \ge0 and dS≥0dS \ge0 for Ms\mathcal{M}^{s}, and T<0T < 0 and dS<0dS < 0 for Mu\mathcal{M}^{u}. The instability in Mu\mathcal{M}^{u} and ΔS<0\Delta S < 0 during its spontaneous evolution does not imply violating SL, contrary to the comment at the start [18, for example]. It is also clear that Viol-GSL-Th and Viol-Th can only hold for nonspontaneous processes. As they cannot occur in Σ\Sigma, they are inconsistent with analytical mechanics and the BCM proposal. This completes the proof. ■

It is instructive to consider, as an example, a two-level particle as our system Σ\Sigma. The entropy is S=−p1ln⁡p1−p2ln⁡p2S = -p_1 \ln p_1 - p_2 \ln p_2, where p1p_1 and p2p_2 are the probabilities of the two levels. Then, dS=ln⁡[(1−p1)/p1]dp1dS = \ln[(1-p_1)/p_1]dp_1. The maximum of SS occurs at p1=p2=1/2p_1 = p_2 = 1/2, which represents the EQ point. For any other value of p1p_1, the behavior is different for dS>0dS > 0 and dS<0dS < 0. In the former case, dSdS brings p1p_1 and p2p_2 closer to the EQ point making it a sink, while in the latter case, p1p_1 and p2p_2 move away from the EQ point making it a source. This is consistent with our discussion above and brings out the difference of Ms\mathcal{M}^{s} to Mu\mathcal{M}^{u} vividly. To make the example more interesting, we need to associate energies to the two level. We capture nonuniformity of the system by adding small contributions ϵ1\epsilon_1 and ϵ2\epsilon_2 to the equilibrium energies e1e_1 and e2>e1e_2 > e_1, respectively, so that E1=e1+ϵ1E_1 = e_1 + \epsilon_1 and E2=e2+ϵ2E_2 = e_2 + \epsilon_2. We also take p1=1/2−δp_1 = 1/2-\delta and p2=1/2+δp_2 = 1/2+\delta, with δ\delta a small quantity. We keep only leading order terms. For ϵ1>0\epsilon_1 > 0 and ϵ2>0\epsilon_2 > 0, we are considering Ms\mathcal{M}^{s}; see the blue solid curve. Along the blue arrow, we have dW≈(ϵ1+ϵ2)/2dW \approx(\epsilon_1+\epsilon_2)/2, and dQ≈(ϵ1−ϵ2)δdQ \approx(\epsilon_1-\epsilon_2)\delta. From Irr-P, we have (e2−e1)δ+(ϵ1+ϵ2)/2=0(e_2-e_1)\delta+(\epsilon_1+\epsilon_2)/2 = 0, which is the condition for dE=0dE = 0, as is easily seen from equating Eeq=(e1+e2)/2E_{eq} = (e_1+e_2)/2, and E≈Eeq+(e2−e1)δ+(ϵ1+ϵ2)/2E \approx E_{eq}+(e_2-e_1)\delta+(\epsilon_1+\epsilon_2)/2, and δ<0\delta< 0, dS>0dS > 0, making the system more uniform and T>0T > 0. For ϵ1<0\epsilon_1 < 0 and ϵ2<0\epsilon_2 < 0, we are considering Mu\mathcal{M}^{u}; see the red solid curve. Along the red arrow, we again have dW>0dW > 0 but dS<0dS < 0 as system becomes more nonuniform. In both cases, dQ>0dQ > 0 so GSL remains intact, but the behavior of dSdS shows that T<0T < 0 as expected.

DISCUSSION: We are ready to highlight our findings and discuss their tantalizing consequences. Gen-Th for Σ\Sigma always satisfies the first law dE=0dE = 0 [31, (D)] (from which follows Irr-P in (S2)) for any {pk}\{p_k\} that determines dWdW. Irr-P intertwines dWdW and dQdQ, but without imposing restrictions on dpkdp_k [33] so dQdQ has either sign but always determined by dWdW. It is here the relevance of microscopic Mec-EQ-P for {mk}\{m_k\}, which has not been properly recognized so far, becomes central in our reverse approach to finally identify Mec-EQ-P as the root cause of the stochastic principle of GSL/SL that is surprisingly and completely determined by mechanical works {dWk≥0}\{dW_k \ge0\} and dWdW alone in Gen-GSL-Th for Σ\Sigma. Thus, GSL is a consequence of analytical mechanics, the foundation of the BCM proposal, with dQ≥0dQ \ge0 directly related to dS⋛0dS \gtreqless0 depending on T⋛0T \gtreqless0, respectively. This provides a direct proof of SL for Ms\mathcal{M}^{s} for T≥0T \ge0 and dS≥0dS \ge0, and extends it to Mu\mathcal{M}^{u} for T<0T < 0 and dS≤0dS \le0, a very tantalizing extension, which corrects a common misconception of SL about negative TT [18]. As negative temperatures are not physically impossible [18, 25, 64, 65], it is quite surprising to realize that dS<0dS < 0 is not a violation of SL, if T<0T < 0. This has been recognized before; however, see [66]. Thus, Gen-GSL-Th supersedes Cl-Th for Σ\Sigma. In addition, Viol-Th and Viol-GSL-Th cannot be taken as a viable possibility within the BCM proposal.

It is clear from the above theorem that dS⋛0dS \gtreqless0 alone without any reference to TT cannot be used to describe SL or its violation, a point that has not been appreciated so far and highlights the role of TT for SL.

For a genuine GSL/SL violation in Σ\Sigma, we need nonspontaneous evolution requiring T≥0T \ge0 and dS<0dS < 0 for Ms\mathcal{M}^{s}, and T<0T < 0 and dS>0dS > 0 for Mu\mathcal{M}^{u}. During this evolution, Ms\mathcal{M}^{s} becomes less and less uniform (dS<0dS < 0) and Mu\mathcal{M}^{u} becomes more and more uniform (dS>0dS > 0). Thus, Ms\mathcal{M}^{s} runs off from the uniform sink Mseq\mathcal{M}_{seq} catastrophically to asymptotically approach an extremely nonuniform macrostates Mcatss\mathcal{M}^{s}_{cats} so that Scatss≪SseqS_{cats}^{s} \ll S_{seq} as if Mseq\mathcal{M}_{seq} is unstable. Similarly, Mu\mathcal{M}^{u} runs towards and terminates in the uniform source Mueq\mathcal{M}_{ueq} along the direction opposite to red arrows with SS increasing to SueqS_{ueq} as if Mueq\mathcal{M}_{ueq} is stable. Both possibilities require some external agent Eext\mathcal{E}_{ext} that, for example, inserts an internal partition for T>0T > 0 (T<0T < 0) to obtain dS<(>)0dS < (>)0 by manipulating Mseq\mathcal{M}_{seq} (Mueq\mathcal{M}_{ueq}) to drive Ms\mathcal{M}^{s} (Mu\mathcal{M}^{u}) run away from (towards) the sink Mseq\mathcal{M}_{seq} (Mueq\mathcal{M}_{ueq}) in a nonspontaneous manner; see [31, (E)], where various possibilities are discussed. If Σˉ\bar{\Sigma} is now detached from Σext\Sigma_{ext}, the result will be an internal constraint discussed by Callen [24], removal of which will initiate spontaneous processes to increase the entropy in accordance with GSL/SL as it must. A similar situation occurs in the demon paradox [17, 67–69], in which the demon starts manipulating Mseq\mathcal{M}_{\mathrm{seq}} in a nonspontaneous manner as discussed recently [61, 69].

Th-EQ-P for Ms\mathcal{M}^{\mathrm{s}} becomes a consequence of SL in Cl-Th for ΣM\Sigma_{\mathrm{M}}. Instead, we prove not only Th-EQ-P for Ms\mathcal{M}^{\mathrm{s}} and Mu\mathcal{M}^{\mathrm{u}} but also GSL for Σ‾\overline{\Sigma} by using (S4), i.e., Mec-EQ-P, of analytic mechanics (instead of SL) in Gen-GSL-Th. This shows the strength of our reverse approach. As mks\mathfrak{m}^{\mathrm{s}}_{k} and Ms\mathcal{M}^{\mathrm{s}} asymptotically converge to mkseq\mathfrak{m}_{k\mathrm{seq}} and Mseq\mathcal{M}_{\mathrm{seq}}, respectively, by removing nonuniformity gradually, S(t)S(t) continuously increases and reaches its maximum. As mku\mathfrak{m}^{\mathrm{u}}_{k} and Mu\mathcal{M}^{\mathrm{u}} runs away from mkueq\mathfrak{m}_{k\mathrm{ueq}} and Mueq\mathcal{M}_{\mathrm{ueq}}, respectively, by increasing nonuniformity gradually, S(t)S(t) continuously decreases and reaches some minimum ScatsS_{\mathrm{cats}}, even though GSL remains valid so this case corresponds to T<0T<0. With (S4) in Gen-GSL-Th, BCM proposal imposes a very strong constraint of nonnegativity on dWkdW_k for Σ‾\overline{\Sigma}. The nonnegative average dWdW through Irr-P determines dQdQ, which imposes a restriction on {dpk}\{dp_k\} to have a particular form {dpkGSL}\{dp_k^{\mathrm{GSL}}\} to ensure dQ≥0dQ\geq0. Another important finding is that Theorem 2 also becomes the No-Go theorem so that that GSL-violation requires rejecting (S4), i.e., Mec-EQ-P and, therefore, the BCM proposal, a prospect that has not been recognized so far. Indeed, the violation cannot be taken as a viable possibility as it defies analytic mechanics. The conclusion provides not only a tantalizing insight into SL-violation in ΣM\Sigma_{\mathrm{M}} for the first time by recognizing that Viol-Th requires rejecting analytical mechanics, but is also a testament to the robustness of the BCM proposal. This elevates Mec-EQ-P to be of primary relevance for thermodynamic foundation of GSL/SL in Gen-GSL-Th, and clarifies the significance of ”asymptotic approach to stable equilibrium” for a thermodynamic system in Ms\mathcal{M}^{\mathrm{s}} in time after being momentarily disturbed [22, 24, 62, 70] from Mseq\mathcal{M}_{\mathrm{seq}}, which forms the cornerstone of Cl-Th. We also extend GSL/SL now to cover Mu\mathcal{M}^{\mathrm{u}}, where negative TT plays an important role. Thus, SL must always be stated along with TT, a fact that has not been appreciated.

Thermodynamic behavior of EwE_{\mathrm{w}} in Figure 1 can also be used to obtain GSL/SL [31, (D)], but we have used EkE_{\mathrm{k}} to emphasize their mechanical nature.

References

it will always refer to nonvanishing {dpk}\{\mathrm{d}p_k\} and nonvanishing dQ\mathrm{d}Q as discussed in the text. We use dissipation to always mean converting dW≥0\mathrm{d}W \ge0 to dQ≥0\mathrm{d}Q \ge0, but use irreversibility to imply any signature of dW\mathrm{d}W.

Supplementary Material

We consider an isolated mechanical system Σ\Sigma containing N=1,2,⋯N = 1, 2, \cdots particles, which is then treated thermodynamically following the BCM proposal introduced in the main text. Its deterministic Hamiltonian H(x∣w)\mathcal{H}(\mathbf{x}|\mathbf{w}) with no explicit time dependence is a function of x≔{(ri,pi)}\mathbf{x} \coloneqq\{(\mathbf{r}_i,\mathbf{p}_i)\}, i=1,2,⋯ ,Ni = 1, 2, \cdots, N denoting the set of positions and momenta of NN particles and specified by the fixed set w\mathbf{w} of extensive observable work parameters (including NN, volume VV, etc. but not energy EE) of Σ\Sigma. Its microstates {mk(w)}\{m_k(\mathbf{w})\} of energies {Ek(w)}\{E_k(\mathbf{w})\} are also time-independent and depend only on w\mathbf{w} of the entire Σ\Sigma and not of its various parts. We express this fact of stationarity by saying that Heq≔H(x∣w)\mathcal{H}_{\mathrm{eq}} \coloneqq\mathcal{H}(\mathbf{x}|\mathbf{w}), mkeq≔mk(w)m_{k\mathrm{eq}} \coloneqq m_k(\mathbf{w}) and Ekeq≔Ek(w)E_{k\mathrm{eq}} \coloneqq E_k(\mathbf{w}) are uniform in w\mathbf{w}, which is in the spirit of and consistent with the well-known result of equilibrium (EQ) thermodynamics [1] that an EQ macrostate Meq\mathcal{M}_{\mathrm{eq}} is uniform in the set X≔(E,w)\mathbf{X} \coloneqq(E,\mathbf{w}) of its state variables that defines the EQ state space SX\mathcal{S}_{\mathbf{X}}. As a consequence of uniformity, Meq\mathcal{M}_{\mathrm{eq}} and mkeqm_{k\mathrm{eq}} can neither perform any mechanical work nor generate any mechanical power. Various system-intrinsic (SI) EQ functions including entropy Seq=S(X)S_{\mathrm{eq}} = S(\mathbf{X}) of Meq\mathcal{M}_{\mathrm{eq}} remain time independent. (From now on, we will mostly suppress w\mathbf{w} for Σ\Sigma unless necessary as it is fixed.) In a nonequilibrium (NEQ) macrostate M\mathcal{M} of Σ\Sigma that we are interested in, all SI-quantities such as H(x∣t)\mathcal{H}(\mathbf{x}|t), {mk(t)}\{m_k(t)\}, {Ek(t)}\{E_k(t)\} and S(t)S(t) in SX\mathcal{S}_{\mathbf{X}} must change in time tt so they acquire an explicit time dependence; see, for example, Eq. (9) that captures this observation for microwork dWk(t)dW_k(t) and micropower Pk(t)P_k(t) by microstate mk(t)m_k(t) [2]. This time-dependence is a result of nonuniformity among various disjoined but mutually interacting uniform subsystems {Σl}\{\Sigma_l\} with Hamiltonians {Hleq≔Hl(xl∣wl)}\{\mathcal{H}_{l\mathrm{eq}} \coloneqq\mathcal{H}_l(\mathbf{x}_l|\mathbf{w}_l)\} in terms of work variables wl\mathbf{w}_l, l=1,2,⋯l = 1, 2, \cdots; nonuniformity results in internal flows among {Σl}\{\Sigma_l\} that affect Xl≔(El,wl)\mathbf{X}_l \coloneqq(E_l,\mathbf{w}_l) so a NEQ M(t)\mathcal{M}(t) and mk(t)m_k(t) are individually nonuniform in {Xl}\{\mathbf{X}_l\} and {wl}\{\mathbf{w}_l\}, respectively. For the isolated system, all SI-changes in it such as dWk(t)dW_k(t) and Pk(t)P_k(t) are generated internally and are commonly identified as spontaneous. They are governed by the second law (SL) in classical thermodynamics (Cl-Th) [3]. A nonspontaneous process, which is never under the purview of SL, cannot occur in Σ\Sigma. For it to occur, intervention from an outside agent Σext\Sigma_{\mathrm{ext}} that may also include Maxwell’s demon [4] is required to disturb Σ\Sigma. The intervention only creates an internal constraint as discussed by Callen [3], whose removal, however, is controlled by SL.

In (B) below, we show that nonuniformity in mk(t)m_k(t) can be described precisely in terms of a set ξ\boldsymbol{\xi} of internal variables [5, 6, 7, 8] if we restrict the size of subsystems to satisfy quasi-additivity of their energies. This requires imposing size restriction on Σ\Sigma, which is denoted by ΣE\Sigma_{\mathrm{E}}. We introduce W=(w,ξ)\mathbf{W} = (\mathbf{w},\boldsymbol{\xi}) as the work variable and Z=(E,W)\mathbf{Z} = (E,\mathbf{W}) as the state variable that determines a NEQ state space SZ\mathcal{S}_{\mathbf{Z}} for ΣE\Sigma_{\mathrm{E}}. The explicit time dependence of M(t)\mathcal{M}(t) in SX\mathcal{S}_{\mathbf{X}} is fully equivalent to the implicit dependence only through ξ(t)\boldsymbol{\xi}(t) in SZ\mathcal{S}_{\mathbf{Z}}; see Eq. (13).

For NEQ statistical mechanics, we follow the BCM proposal and consider an ensemble in which mkm_k appears with probability pkp_k, see (D) below; they together define various macrostates M:{mk,pk}\mathcal{M} : \{m_k,p_k\} for the same set {mk}\{m_k\} but different sets {pk}\{p_k\}, which define the entropy SS by its Gibbs-Shannon formulation [9] for Σ\Sigma as the ensemble average of microentropy Sk≔−ηkS_k \coloneqq-\eta_k,

S=⟨S⟩=∑kpkSk,(6)S = \langle S\rangle= \sum_k p_k S_k, \tag*{(6)}

which is easy to verify, see for example [10], for any arbitrary collection {pk}\{p_k\}. The validity of the above SS is based on the use of Stirling’s approximation. We will not be concerned about the error in the approximation and accept the formula to be the definition of SS. To make the subsystems entropies additive, we require SS to be a state function S(Z)S(\mathbf{Z}) in SZ\mathcal{S}_{\mathbf{Z}}, which requires a further restriction on the size as macroscopic (M) to capture quasi-independence of subsystems. We denote this macroscopic system by ΣM\Sigma_{\mathrm{M}} to be uniquely specified in SZ\mathcal{S}_{\mathbf{Z}}; in contrast, Σ\Sigma and ΣM\Sigma_{\mathrm{M}} are not uniquely specified in SX\mathcal{S}_{\mathbf{X}} and SZ\mathcal{S}_{\mathbf{Z}}, respectively. We use Σ\Sigma to denote any of the three systems and any possible subsystems when we are not interested in specifying the state space.

Table of Various Forms of Thermodynamics in the text
Gen-Th(S1), (S2), (S3) for Σ\Sigma, dW=dQ≥0dW = dQ \ge0
Cl-ThGen-Th and SL axiom for ΣM\Sigma_{\mathrm{M}}
Rest-ThGen-Th for ΣM\Sigma_{\mathrm{M}} with state function SS
Viol-ThGen-Th for Σ\Sigma with dS<0dS < 0
Gen-GSL-ThGen-Th and (S4) for Σ\Sigma with dQ≥0dQ \ge0
Viol-GSL-ThGen-Th and (S4) for Σ\Sigma with dQ<0dQ < 0

Table 1.

In NEQ thermodynamics, SX\mathcal{S}_{\mathbf{X}} is the most convenient state space as it requires no knowledge of any internal nonuniformity that is hidden from the observer and may not always be known, but requires an explicit time dependence such as in Ek(t)E_k(t); see Figure 1 in main text. Despite this, the choice of SZ\mathcal{S}_{\mathbf{Z}} helps to explain not only the origin of explicit time dependence resulting from nonuniformity but also to justify the nonstandard definition of TT in Eq. (7). The first law and macroheat dQdQ, see (D), for Σ\Sigma play a central role in proving the generalized second law (GSL); both can be directly introduced without identifying TT. But we need TT to relate GSL to dSdS to also directly prove the second law (SL) dS≥0dS \ge0, and to shed light on the possible significance of dS<0dS < 0 in its violation, which is one of our goals. In most cases of interest shown by the solid blue curve M\mathcal{M} [1], Σˉ\bar{\Sigma} attempts to thermodynamically minimize nonuniformity in {Xl}\{\mathbf{X}_l\} by reducing internal flows and become uniform as SS continue to increase as explained below. As EE is fixed, we introduce a NEQ temperature as the ratio dQ/dSdQ/dS, which we also simply write as

T=dQdS∣E(7)T = \left.\frac{dQ}{dS}\right|_E \tag*{(7)}

in any infinitesimal internal process within Σ‾\overline{\Sigma}. While this definition is unconventional, it is indeed the conventional definition (∂E/∂S)ξ(\partial E/\partial S)_{\xi} for ΣM\Sigma_M as seen later from Eqs. (28) and (37). Thus, we take TT above to represent the temperature for any number of particles in Σ‾\overline{\Sigma}.

(A), (B) and (C) below are based on step (S1), see main text, only and deal with microstates {mk(t)}\{m_k(t)\} of the nonuniform deterministic Hamiltonian H(x∣t)\mathcal{H}(\mathbf{x}\mid t). (A) deals with it in SX\mathcal{S}_X to show the explicit time dependence in {Ek(t)}\{E_k(t)\}, which is related to the implicit time-dependence of ξ(t)\boldsymbol{\xi}(t) in SZ\mathcal{S}_Z introduced in (B). (C) introduces the Mechanical Equilibrium Principle (Mec-EQ-P) of Energy [11], which forms step (S4). This step is required to complete the BCM proposal by connecting it to analytical mechanics. M(t)\mathcal{M}(t) and their thermodynamics in SX\mathcal{S}_X and SZ\mathcal{S}_Z are considered next. (D) requires step (S2) with any arbitrary pkp_k to identify a general NEQ macrostates M\mathcal{M}, and develop its first law for any Σ‾\overline{\Sigma}. This results in (S3) and the General thermodynamics (Gen-Th) but without invoking Mec-EQ-P for any possible M\mathcal{M}. (E) puts macroscopic size restriction and considers ΣM\Sigma_M to ensure SS is additive and a state function in SZ\mathcal{S}_Z.

Various forms of thermodynamics that we consider are listed in the table and in the main text.

Mechanical Systems

(A) MECHANICAL SYSTEMS : We first consider the mechanical system Σ‾\overline{\Sigma} in SX\mathcal{S}_X that is specified by H(xk∣t)\mathcal{H}(\mathbf{x}_k\mid t) given in terms of xk≔{(rki,pki)}\mathbf{x}_k \coloneqq\{(\mathbf{r}_{ki},\mathbf{p}_{ki})\} and with explicit time dependence. Let mk(t)≔xkm_k(t) \coloneqq\mathbf{x}_k denote one of its microstates with microenergy Ek≔H(xk∣t)E_k \coloneqq\mathcal{H}(\mathbf{x}_k\mid t) so that

dEk=∂Ek∂xk⋅dxk+∂Ek∂tdt=∂Ek∂tdt(8)dE_k=\frac{\partial E_k}{\partial\mathbf{x}_k}\cdot d\mathbf{x}_k+\frac{\partial E_k}{\partial t}dt=\frac{\partial E_k}{\partial t}dt \tag*{(8)}

as the first term vanishes due to Hamilton’s equations of motion. As this term includes the effect of interparticle potentials, we are not interested in those potentials so we will no longer exhibit xk\mathbf{x}_k, which is explicitly identified by the suffix kk. Instead, we are interested in dEkdE_k that is determined solely by varying the parameter tt [12]. We write H(xk∣t)\mathcal{H}(\mathbf{x}_k\mid t) as Ek(t)E_k(t) or simply EkE_k; similarly, we simply write mkm_k for mk(t)m_k(t) unless clarity is needed. The SI-microwork dWk(t)dW_k(t) can be expressed in terms of instantaneous power Pk(t)P_k(t) in SX\mathcal{S}_X as

dWk(t)=−dEk(t)=Pk(t)dt.(9)dW_k(t)=-dE_k(t)=P_k(t)dt. \tag*{(9)}

Similarly, the mechanical system ΣE\Sigma_E in a microstate mk(ξ)m_k(\boldsymbol{\xi}) in SZ\mathcal{S}_Z is specified by H(xk∣ξ)\mathcal{H}(\mathbf{x}_k\mid\boldsymbol{\xi}), which now depends on the set ξ\boldsymbol{\xi} of work parameters of ΣE\Sigma_E, see (B) below, that determines the microwork dWk(ξ)dW_k(\boldsymbol{\xi}) as we now explain. We remark that ξ\boldsymbol{\xi} is assumed to be the same for all microstates so it does not carry the suffix kk, while the microforce Fk\mathbf{F}_k in Eq. (11) and dWkdW_k do so the latter are fluctuating (over mkm_k). We will not consider here the case when ξ\boldsymbol{\xi} is fluctuating (over mkm_k), to which the present method is easily extended [13, 14]. The significance of ξ\boldsymbol{\xi} becomes clear when we consider the microenergy Ek(ξ)≡H(xk∣ξ)E_k(\boldsymbol{\xi})\equiv\mathcal{H}(\mathbf{x}_k\mid\boldsymbol{\xi}) of mk(ξ)m_k(\boldsymbol{\xi}), for which

dEk=∂Ek∂ξ⋅dξ(10)dE_k=\frac{\partial E_k}{\partial\boldsymbol{\xi}}\cdot d\boldsymbol{\xi} \tag*{(10)}

by using the same argument as used in Eq. (8). As above, we simply denote Ek(ξ)E_k(\boldsymbol{\xi}) by EkE_k, and mk(ξ)m_k(\boldsymbol{\xi}) by mkm_k unless clarity is needed. The SI-microforce, also known as the generalized microforce, in mkm_k is given by

Fk≔−∂Ek∂ξ,(11)\mathbf{F}_k\coloneqq-\frac{\partial E_k}{\partial\boldsymbol{\xi}}, \tag*{(11)}

and the SI-microwork done by it is given by

dWk(ξ)=Fk⋅dξ=−dEk,(12)dW_k(\boldsymbol{\xi})=\mathbf{F}_k\cdot d\boldsymbol{\xi}=-dE_k, \tag*{(12)}

which clarifies the significance of ξ\boldsymbol{\xi} as the work parameter for ΣE\Sigma_E. In NEQ thermodynamics, the macroscopic analog (ensemble average ⟨∙⟩\langle\bullet\rangle, see (D) below) F=⟨F⟩\mathbf{F}=\langle\mathbf{F}\rangle of Fk\mathbf{F}_k determines the affinity [5, 6] associated with F\mathbf{F}. For us, ξ\boldsymbol{\xi} describes nonuniformity and changes spontaneously with time in accordance with the principle of mechanical equilibrium, called Mec-EQ-P; see (C) below, the main text and [11 p. 99]. Thus, Ek(ξ)E_k(\boldsymbol{\xi}) continues to change with ξ\boldsymbol{\xi} and results in the above microwork. The corresponding instantaneous microstate power in SZ\mathcal{S}_Z is

Pk(t)=Fk⋅ξ˙,(13)P_k(t)=\mathbf{F}_k\cdot\dot{\boldsymbol{\xi}}, \tag*{(13)}

which can be used in Eq. (9) to obtain dWk(t)dW_k(t). We thus see that implicit time-dependence in ξ\boldsymbol{\xi} for ΣE\Sigma_E in SZ\mathcal{S}_Z is equivalent to the explicit time dependence in SX\mathcal{S}_X without any ξ\boldsymbol{\xi} for Σ‾\overline{\Sigma}. The variation of Ek(ξ)E_k(\boldsymbol{\xi}) for an isolated system in SZ\mathcal{S}_Z as shown in Figure 1 in the main text with ξ\boldsymbol{\xi} can also be treated as time variation of Ek(t)E_k(t) in SX\mathcal{S}_X. This is also consistent with the way a time-dependent Hamiltonian is treated in mechanics [12] by considering tt as a parameter in the Hamiltonian so we can use tt instead of ξ\boldsymbol{\xi}, in which case Eq. (12) reduces to Eq. (9) obtained in SX\mathcal{S}_X, where Σ‾\overline{\Sigma} has no restriction on its size. We will not be concerned with the actual time dependence of ξ\boldsymbol{\xi} in SZ\mathcal{S}_Z in this investigation; all we need to remember that it can be accounted for by an explicit time dependence in Pk(t)P_k(t) and other quantities for Σ‾\overline{\Sigma} in SX\mathcal{S}_X.

According to the discussion above, we can consider either ξ\boldsymbol{\xi} explicitly (to be specified in (B) below) and use the state space SZ\mathcal{S}_Z, or consider Σ‾\overline{\Sigma} instead in the state space SX\mathcal{S}_X without any specification of ξ\boldsymbol{\xi}. In the former case, microstates are specified in SZ\mathcal{S}_Z, which restricts Σ‾\overline{\Sigma} to ΣE\Sigma_E. NEQ temperature in Eq. (7) is always defined.

Internal Variables for Mechanical Microstates

(B) INTERNAL VARIABLES FOR MECHANICAL MICROSTATES {mk}\{m_k\} OF ΣE\Sigma_E: We now turn to the definition of internal variable [5, 6, 7, 8] for a deterministic nonuniform mechanical system ΣE\Sigma_E, but denote it by Σ‾\overline{\Sigma} to simplify notation unless clarity is needed. For simplicity, we consider it to be formed by two different and disjoined but mutually interacting subsystems Σ1\Sigma_1 and Σ2\Sigma_2 that are each uniform in SX\mathcal{S}_X. The internal variables are required to uniquely specify {mk}\{m_k\} in SZ\mathcal{S}_Z; they are not specified uniquely in SX\mathcal{S}_X. We use suffixes for Σ1,Σ2\Sigma_1,\Sigma_2 and no suffix for Σ‾\overline{\Sigma}. We consider Σ‾,Σ1\overline{\Sigma},\Sigma_1, and Σ2\Sigma_2 in microstate mk,mk1m_k,m_{k_1}, and mk2m_{k_2} of energy Ek,Ek1E_k,E_{k_1}, and Ek2E_{k_2}, respectively. We also introduce n1=N1/Nn_1=N_1/N and n2=N2/Nn_2=N_2/N. If we neglect the energy δEk\delta E_k due to the interface between Σ1\Sigma_1 and Σ2\Sigma_2, the Hamiltonians of the three bodies (we do not show them their arguments xk\mathbf{x}_k, xk1\mathbf{x}_{k_1} and xk2\mathbf{x}_{k_2}, respectively, but use microstate suffixes) are related by

Hk(w,ξ)≃Hk1(w1)+Hk2(w2),(14)\mathcal{H}_k(\mathbf{w},\boldsymbol{\xi}) \simeq\mathcal{H}_{k_1}(\mathbf{w}_1) + \mathcal{H}_{k_2}(\mathbf{w}_2), \tag*{(14)}

which ensures microenergy quasi-additivity

Ek≃E1k1+E2k2,(15)E_k \simeq E_{1k_1} + E_{2k_2}, \tag*{(15)}

which can be justified only if we restrict the minimum sizes of Σ1\Sigma_1 and Σ2\Sigma_2 to be some ΣE\Sigma_E that itself is determined by the range of inter-particle interactions. Under this conditions, we denote Σ\Sigma, Σ1\Sigma_1, and Σ2\Sigma_2 by ΣE\Sigma_E, Σ1E\Sigma_{1E}, and Σ2E\Sigma_{2E}, and we have mk≃mk1⊗mk2\mathbf{m}_k \simeq\mathbf{m}_{k_1} \otimes\mathbf{m}_{k_2}. However, for the sake of notational simplicity, we suppress E in the suffix for various quantities except for bodies.

The discussion in (A) applies to any ΣE\Sigma_E and ΣM\Sigma_M. In general, ΣE\Sigma_E is much smaller in size than ΣM\Sigma_M. The additivity of extensive (which tt is not) work parameters w1\mathbf{w}_1 and w2\mathbf{w}_2 in (16) is not affected by the presence or absence of δEk\delta E_k so that we can introduce the extensive work parameter set

w=˙w1+w2.(16)\mathbf{w} \mathrel{\dot=} \mathbf{w}_1 + \mathbf{w}_2. \tag*{(16)}

We introduce a new extensive internal variable set ξk=˙(ξkE,ξ)\boldsymbol{\xi}_k \mathrel{\dot=} (\xi_{kE},\boldsymbol{\xi}), where

ξkE=˙E1k1/n1−E2k2/n2,ξ=˙w1/n1−w2/n2,(17)\xi_{kE} \mathrel{\dot=} E_{1k_1}/n_1 - E_{2k_2}/n_2,\quad\boldsymbol{\xi} \mathrel{\dot=} \mathbf{w}_1/n_1 - \mathbf{w}_2/n_2, \tag*{(17)}

and the extensive work parameter

W=˙(w,ξ)(18)\mathbf{W} \mathrel{\dot=} (\mathbf{w},\boldsymbol{\xi}) \tag*{(18)}

are independent variables for ΣE\Sigma_E in that (Ek,ξkE)(E_k,\xi_{kE}) and (wl∈w,ξl∈ξ)(w_l \in\mathbf{w},\xi_l \in\boldsymbol{\xi}) form pairs of independent variables. The introduction of ξk\xi_k is to ensure that the number of variables on both sides in (14) are equal as they must be for an equality. We also have

E1k1,2k2=n1,2(Ek±n2,1ξkE), w1,2=n1,2(w±n2,1ξ).(19)E_{1k_1,2k_2} = n_{1,2}(E_k \pm n_{2,1}\xi_{kE}),\ \mathbf{w}_{1,2} = n_{1,2}(\mathbf{w} \pm n_{2,1}\boldsymbol{\xi}). \tag*{(19)}

We can easily extend the above discussion to any numbers mm of subsystems {Σl}\{\Sigma_l\}, l=1,2,⋯ ,ml = 1, 2, \cdots, m forming ΣE\Sigma_E, each specified by its own observable set wl\mathbf{w}_l to allow for a complex form of nonuniformity in terms of uniform subsystems; Eq. (15) remains valid for any mm. It is easy to verify that all internal variables in ξ\boldsymbol{\xi}, whose number we denote by ι\iota, can be expressed as a linear combinations of {wl}\{\mathbf{w}_l\} [8, 13], with mm and ι\iota increasing with the degree of nonuniformity of mk(ξ)\mathbf{m}_k(\boldsymbol{\xi}). Even though the definition of ξk\xi_k is not unique, we choose it for convenience so that it vanishes when ΣE\Sigma_E is uniform i.e., in EQ as defined in Eq. (14). The set (Ek,ξk)(E_k,\xi_k) forms the complete set of variables to uniquely specify mk\mathbf{m}_k of ΣE\Sigma_E in GZ\mathcal{G}_Z. It is easily verified that

FkE=−∂Ek/∂ξkE=0,(20)F_{kE} = -\partial E_k/\partial\xi_{kE} = 0, \tag*{(20)}

a consequence of independent EkE_k and ξkE\xi_{kE} so the variation of ξkE\xi_{kE} does not generate any SI-microwork for any {wm}\{\mathbf{w}_m\}. This explains why it is not included in W\mathbf{W}, in which w\mathbf{w} remains constant as Σ\Sigma evolves in time; however, ξ˙k\dot{\xi}_k continues to change due to internal flows.

Here, we focus on two (m=2m = 2) subsystems Σ1E\Sigma_{1E} and Σ2E\Sigma_{2E} of ΣE\Sigma_E, for simplicity, each of which is uniform (no internal variables for them) but ΣE\Sigma_E is not. We let Ekl,wl=˙(Nl,Vl)E_{k_l},\mathbf{w}_l \mathrel{\dot=} (N_l,V_l) specify uniform microstates mkl\mathbf{m}_{k_l}, l=1,2l = 1,2 in GX\mathcal{G}_X. We now focus on the nonuniform microstate mk\mathbf{m}_k of ΣE\Sigma_E, which require internal variables along with EkE_k and w=˙(N,V)\mathbf{w} \mathrel{\dot=} (N,V) to be specified in GZ\mathcal{G}_Z. We keep observables EkE_k and w=˙(N,V)\mathbf{w} \mathrel{\dot=} (N,V) of ΣE\Sigma_E fixed along with N1N_1 and N2N_2 of Σ1E\Sigma_{1E} and Σ2E\Sigma_{2E}. Following (17), we have

ξkE=˙Ek1/n1−Ek2/n2,ξV=˙V1/n1−V2/n2,(21)\xi_{kE} \mathrel{\dot=} E_{k_1}/n_1 - E_{k_2}/n_2,\quad\xi_V \mathrel{\dot=} V_1/n_1 - V_2/n_2, \tag*{(21)}

to identify Z=(Ek,N,V,ξkE,ξV)\mathbf{Z} = (E_k,N,V,\xi_{kE},\xi_V) for ΣE\Sigma_E.

Densities in Σ1E\Sigma_{1E} and Σ2E\Sigma_{2E} are equal in a uniform microstate mk(ξk)\mathbf{m}_k(\xi_k) of ΣE\Sigma_E so ξk=0\xi_k = 0 and need not be considered. This is consistent with the fact that uniform mkeq\mathbf{m}_{keq} is uniquely specified in GX\mathcal{G}_X. In this case, we have a trivial additivity of the Hamiltonians in (14) given by

Hk(w)≃Hk1(w1)+Hk2(w2)(22)\mathcal{H}_k(\mathbf{w}) \simeq\mathcal{H}_{k_1}(\mathbf{w}_1) + \mathcal{H}_{k_2}(\mathbf{w}_2) \tag*{(22)}

with no internal flows between subsystems. In a nonuniform microstate, ξk≠0\xi_k \ne0. We recall that w\mathbf{w} is fixed for mk(ξk)\mathbf{m}_k(\xi_k) but w1\mathbf{w}_1 and w2\mathbf{w}_2 can change due to possible transfers (internal flows) between Σ1E\Sigma_{1E} and Σ2E\Sigma_{2E} with dw1=−dw2d\mathbf{w}_1 = -d\mathbf{w}_2. It follows from (19) that

E˙1k1,2k2=n1,2(E˙k±n2,1ξ˙kE), w˙1=−w˙2=n1n2ξ˙(23)\dot{E}_{1k_1,2k_2} = n_{1,2}(\dot{E}_k \pm n_{2,1}\dot{\xi}_{kE}),\ \dot{\mathbf{w}}_1 = -\dot{\mathbf{w}}_2 = n_1n_2\dot{\boldsymbol{\xi}} \tag*{(23)}

for mk(ξk)\mathbf{m}_k(\xi_k).

As W\mathbf{W}, i.e., ξ\boldsymbol{\xi} in (18) for ΣE\Sigma_E is the work variable in Hk(ξk)\mathcal{H}_k(\xi_k) (we suppress w\mathbf{w} as it is constant), Ek(ξk)E_k(\xi_k) corresponding to mk(ξk)\mathbf{m}_k(\xi_k) is only a function of ξ\xi due to (20). This is shown in Figure 1 in main text. As ξk\xi_k continuously changes due to transfers between Σ1E\Sigma_{1E} and Σ2E\Sigma_{2E}, this causes variations in Ek(ξ)E_k(\xi). The variation is similar to the variation in Ek(t)E_k(t) for a Σ\Sigma in GX\mathcal{G}_X as discussed in (A).

In summary, as long as an internal variable is used, the system is restricted to be at least ΣE\Sigma_E in size and requires GZ\mathcal{G}_Z for unique specification. There is no explicit time dependence in this state space. If considerations of subsystems are not important in any discussion then quasi-additivity is not an issues as is the case for the discussion of (9). In that case, Σ\Sigma can be considered as a whole in GX\mathcal{G}_X; there is no need to consider δEk\delta E_k separately, which is included in Hk(w,t)\mathcal{H}_k(\mathbf{w},t). Thus, the mechanical situation here is that of (A).

(C) MECHANICAL EQUILIBRIUM PRINCIPLE OF ENERGY (MEC-EQ-P) FOR mk\mathbf{m}_k: Time-independent microstates {mkeq}\{\mathbf{m}_{keq}\} of a Hamiltonian system, being uniform as demonstrated by the Uniformity Theorem below, play an important role in our approach as we explain now. Any nonuniformity in Σˉ\bar{\Sigma} endows {mk}\{\mathbf{m}_k\} with explicit time dependence as {mk(t)}\{\mathbf{m}_k(t)\} for Σ\Sigma or implicit time dependence through ξk\xi_k as mk(ξk)\mathbf{m}_k(\xi_k) for ΣE\Sigma_E or ΣM\Sigma_M as shown by {mks\{\mathbf{m}_k^s and {mku}\{\mathbf{m}^{u}_{k}\}; see dashed-dotted blue and red curves, respectively; the directions of blue and red arrows on them denote increasing tt during their temporal SI-evolution controlled by internally generated processes that are mechanically spontaneous and generate micropower Pk(t)P_k(t). In analytical mechanics, there is no dissipation in evolution but mks\mathbf{m}^{s}_{k} is special in that its evolution will normally undergo oscillations about mkseq\mathbf{m}_{k\mathrm{seq}} that will persist forever as manifested by the dashed-dot blue curve in Figure 1 in main text; in contrast, the evolution of mku\mathbf{m}^{u}_{k} has no oscillation, see the dashed-dot red curve there, as mku\mathbf{m}^{u}_{k} runs away from mkueq\mathbf{m}_{k\mathrm{ueq}} but terminates is a catastrophe in which mk\mathbf{m}_{k} becomes extremely nonuniform, which for ΣE\Sigma_{E} or ΣM\Sigma_{M} corresponds to an extremely large l=lcatsl = l_{\mathrm{cats}}; see (E) below. Thus, mkueq\mathbf{m}_{k\mathrm{ueq}} is the source for the SI-evolution of mku\mathbf{m}^{u}_{k}.

As we are eventually interested in mks\mathbf{m}^{s}_{k} and mku\mathbf{m}^{u}_{k} in a thermodynamic setting, there will be macroheat to be be simultaneously considered that we discuss in (D) below. As is well known, its presence gives rise to dissipation so all stable thermodynamic processes in Ms\mathcal{M}^{s} terminate asymptotically at Mseq\mathcal{M}_{\mathrm{seq}} as shown by the solid blue curve in the figure. Keeping this in hindsight, we intentionally overlook oscillations in mks\mathbf{m}^{s}_{k} as they do not affect the change in EksE^{s}_{k} during its SI-evolution to mkseq\mathbf{m}_{k\mathrm{seq}} only, making the latter as the sink for the SI-evolution of mks\mathbf{m}^{s}_{k}. This is a useful strategy, since oscillations become thermodynamically irrelevant as Ms\mathcal{M}^{s} asymptotically approaches Mseq\mathcal{M}_{\mathrm{seq}}; see below. This also clarifies the importance of the uniform microstate mkueq\mathbf{m}_{k\mathrm{ueq}}; see the UniformityTheorem below.

We first treat ΣE\Sigma_{E}, whose results pave the way for a clear understanding of what to expect for Σ\Sigma. In analytical mechanics, the temporal evolution of ξk\boldsymbol{\xi}_{k} and mk(ξk)\mathbf{m}_{k}(\boldsymbol{\xi}_{k}) in SZ\mathbb{S}_{\mathcal{Z}} is controlled by mkeq\mathbf{m}_{k\mathrm{eq}}, and is governed by the emergent SI-microforce Fk\mathbf{F}_{k} in Eq. (10) by controlling internal flows within mk(ξk)\mathbf{m}_{k}(\boldsymbol{\xi}_{k}). The emergent processes in ΣE\Sigma_{E} resulting in the SI-microwork in Eq. (11) are commonly identified as spontaneous since they are internally controlled by SI-microforce Fk\mathbf{F}_{k} and not from the outside by any nonsystem microforce. The mechanical equilibrium (Mec-EQ) point at EkeqE^{\mathrm{eq}}_{k} is the equilibrium point [11] in the SI-evolution under the SI-microforce Fk\mathbf{F}_{k}, where not only the SI-internal velocity ξ˙keq\dot{\boldsymbol{\xi}}^{\mathrm{eq}}_{k} but also the SI-microforce Fkeq\mathbf{F}^{\mathrm{eq}}_{k} vanish:

ξ˙keq=0,(24)\dot{\boldsymbol{\xi}}^{\mathrm{eq}}_{k}=0, \tag*{(24)}
Fkeq=−∂Ek∂ξkeq=0.(25)\mathbf{F}^{\mathrm{eq}}_{k}=-\frac{\partial E_{k}}{\partial\boldsymbol{\xi}^{\mathrm{eq}}_{k}}=0. \tag*{(25)}

It follows from Eq. (21b) that this point represents an extremum in EkE_k; see Figure 1 in main text. Its minimum at Ekseq=EkeqE^{\mathrm{seq}}_{k}=E^{\mathrm{eq}}_{k} representing a mechanically stable EQ point enunciates the mechanical asymptotic stability principle of minimum energy. In thermodynamics, stability refers to asymptotic stability in that the system must eventually approach stable equilibrium so that ξk(t)→0\boldsymbol{\xi}_{k}(t)\to0 as t→∞t\to\infty [3]. In this sense, we see that the absolute value ∣ξk(t)∣|\boldsymbol{\xi}_{k}(t)| behaves similar to 1/t1/t in the figure. The asymptotic approach is a stronger requirement than just imposing stability in which the system never strays far away from stable EQ point. In contrast, the maximum of EkE_k at Eueq=EkeqE_{\mathrm{ueq}}=E^{\mathrm{eq}}_{k} representing a mechanically unstable EQ point enunciates the mechanical instability principle of maximum energy. In this case, ∣ξk(t)∣|\boldsymbol{\xi}_{k}(t)| behaves similar to tt in the figure. Both kinds of points are determined by the curvature of EkE_k at EkeqE^{\mathrm{eq}}_{k}. The two principles are collectively called Mec-EQ-P in this investigation.

We now prove the following important theorem emphasizing the physical significance of the uniformity of the Mec-EQ point mkeq\mathbf{m}_{k\mathrm{eq}} for thermodynamic stability, which must exist only in SX\mathbb{S}_{\mathcal{X}} at ξk=0\xi_k=0, where Meq\mathbb{M}_{\mathrm{eq}} exists.

Theorem 3 Uniformity Theorem of mkeq\mathbf{m}_{k\mathrm{eq}}: The Mec-EQ point mkeq\mathbf{m}_{k\mathrm{eq}} of mk\mathbf{m}_{k} is stationary and uniform microstate with no internal flows between the microstates of its uniform subsystems.

Proof. We begin by considering a microstate mk(t)\mathbf{m}_{k}(t) for Σ\Sigma in SX\mathbb{S}_{\mathcal{X}} to establish that mkeq:(mkseq,mkueq)\mathbf{m}_{k\mathrm{eq}}:(\mathbf{m}_{k\mathrm{seq}},\mathbf{m}_{k\mathrm{ueq}}) is stationary. The asymptotic convergence of mks\mathbf{m}^{s}_{k} to mkseq\mathbf{m}_{k\mathrm{seq}} as t→∞t\to\infty means that mkseq\mathbf{m}_{k\mathrm{seq}} must be stationary so it must be a uniform microstate of the time-independent Hamiltonian H(x)\mathcal{H}(\mathbf{x}). The instability of mkueq\mathbf{m}_{k\mathrm{ueq}} is due to the instability of the H(x)\mathcal{H}(\mathbf{x}) in this case so it is also uniform.

As Σ\Sigma contains no information about its internal structure, it cannot be used to understand internal flows for which we need to at least consider ΣE\Sigma_{E} to justify the remainder part of the theorem. From (24), we obtain ξ˙kEeq=0\dot{\xi}^{\mathrm{eq}}_{kE}=0 and ξ˙keq=0\dot{\xi}^{\mathrm{eq}}_{k}=0. Using first ξ˙keq=0\dot{\xi}^{\mathrm{eq}}_{k}=0 in (23), we conclude that E˙keq=0\dot{E}^{\mathrm{eq}}_{k}=0, which follows directly from (22). Using then ξ˙kEeq=0\dot{\xi}^{\mathrm{eq}}_{kE}=0 in (23) along with E˙keq=0\dot{E}^{\mathrm{eq}}_{k}=0, we conclude that E˙1k1,2k2eq=0\dot{E}^{\mathrm{eq}}_{1k_{1},2k_{2}}=0. Together, they show that all flows cease at EkeqE^{\mathrm{eq}}_{k}, which makes mkeq\mathbf{m}_{k\mathrm{eq}} stationary and, therefore, uniform as above. Any nonzero ξk\xi_k is due to nonuniformity, and affects Ek(ξ)≡Ek(ξk)E_k(\boldsymbol{\xi})\equiv E_k(\xi_k) as shown in Figure 1 in main text. This completes the proof of the theorem. ▪

It should be clear from the above mechanical proof of the theorem that any time-dependent mk:(mks,mku)\mathbf{m}_{k}:(\mathbf{m}^{s}_{k},\mathbf{m}^{u}_{k}) must be uniform and stationary only at the equilibrium point mkeq:(mkseq,mkueq)\mathbf{m}_{k\mathrm{eq}}:(\mathbf{m}_{k\mathrm{seq}},\mathbf{m}_{k\mathrm{ueq}}) for Σ\Sigma. Away from it, mk\mathbf{m}_{k} must be nonuniform.

The First Law and General Thermodynamics (Gen-Th) for Σ\Sigma

So far, we have only considered the step (S1), see main text. We now take the step (S2) and introduce stochasticity by considering an ensemble of Σ~\widetilde{\Sigma} in which mk\mathbf{m}_{k} appears with probability pkp_k, but without invoking Mec-EQ-P discussed in (C). It should be stressed that the deterministic Hamiltonian H(x(t))\mathcal{H}(\mathbf{x}(t)) is oblivious to pkp_k so mk(t)\mathbf{m}_{k}(t) is also oblivious to pkp_k in that all its microquantities such as Ek(t)E_k(t) do not change with pkp_k. Despite this, they together define various macrostates M:{mk}\mathbb{M}:\{\mathbf{m}_{k}\} for the same set {mk}\{\mathbf{m}_{k}\} but different sets {pk}\{p_k\}. We put no restrictions on possible sets {pk}\{p_k\} so the resulting macrostates may have nothing to do with what we encounter in classical thermodynamics (Cl-Th), see the table above, and their thermodynamics, to be called general thermodynamics (Gen-Th), may not always satisfy the second law (dS≥0dS\ge0) that is a fundamental axiom (or assumption) in Cl-Th [3] as we will see below. By taking {pk}\{p_k\} arbitrary will allow us to determine the root cause of the second law of Cl-Th and what will cause its violation in the violation thermodynamics (Viol-Th), both of them are contained in Gen-Th.

The macroenergy EE of Σˉ\bar{\Sigma} in Gen-Th is given by the ensemble average ⟨E⟩\langle E\rangle

E=⟨E⟩≔∑kpkEk,E=\langle E\rangle\coloneqq\sum_k p_k E_k,

which is valid for any N≥1N\ge1. It follows from this that

dE=∑kEk dpk+∑kpk dEk.dE=\sum_k E_k\,dp_k+\sum_k p_k\,dE_k.

From Eq. (11) in (A), we observe that the second sum gives the negative of the SI-macrowork dWdW

dW≔∑kpk dWk(26)dW\coloneqq\sum_k p_k\,dW_k \tag*{(26)}

as the ensemble average of SI-microwork dWkdW_k. We identify the first sum, which is dE+dWdE+dW, with SI-macroheat dQdQ as the ensemble average of SI-microheat dQkdQ_k

dQ≔∑kpk dQk=∑kpkEk dηk,(27)dQ\coloneqq\sum_k p_k\,dQ_k=\sum_k p_k E_k\,d\eta_k, \tag*{(27)}

where ηk≔ln⁡pk\eta_k\coloneqq\ln p_k is the Gibbs probability index [9]. It should be stressed that the above identification of dQdQ does not require any size restriction on Σ\Sigma and any notion of temperature or entropy; see below. We thus obtain the statement of the first law

dE=dQ−dW(28)dE=dQ-dW \tag*{(28)}

for Σˉ\bar{\Sigma} in terms of dQdQ and dWdW, both defined for any NN. This completes the step (S3). It should be clear that dQdQ and dWdW are the primary concepts in the first law and Gen-Th, making them equivalent as both are independent of SL (dS≥0dS\ge0).

We observe that dE=0dE=0 in Σˉ\bar{\Sigma} so that

E=∑kpk(t)Ek(t)=∑kpkeqEkeq=Eeq,E=\sum_k p_k(t)E_k(t)=\sum_k p_k^{\mathrm{eq}}E_k^{\mathrm{eq}}=E_{\mathrm{eq}},

where pkeqp_k^{\mathrm{eq}} is the probability of mkeq\mathfrak{m}_{k\mathrm{eq}}, and EeqE_{\mathrm{eq}} is the macroenergy of Meq\mathcal{M}_{\mathrm{eq}}. We use this fact in the first law in Eq. (28) to obtain a simple but very remarkable and extremely profound result

dQ=dW≥0.(29)dQ=dW\ge0. \tag*{(29)}

which provides the mechanical formulation by dWdW of the stochasticity inherent in the thermodynamic process through dQdQ. We conclude that the stochasticity, i.e., the change {dpk}\{dp_k\} in dQdQ is strongly constrained by the mechanical work dWdW and its signature for any P\mathcal{P}. Because of this constraint, we call Eq. (29) the irreversibility principle (Irr-P) under steps (S1-S3), which is valid for an isolated Σ\Sigma of any size, and is a direct consequence of the first law.

We show in Lemma 1 in the main text that the inequality dW≥0dW\ge0 is satisfied in any spontaneous infinitesimal process controlled by Mec-EQ-P. As a consequence, dW<0dW<0 must only happen in any nonspontaneous infinitesimal process that violates Mec-EQ-P.

This completes the discussion of Gen-Th and the first law in it.

As pkp_k remains constant in dWdW, it represents an isentropic macroquantity to justify it as a mechanical quantity, the average of the change dWk=−dEkdW_k=-dE_k; see Eq. (22a). On the other hand, pkp_k does not remain constant in dQdQ so it justifies dQdQ as a stochastic quantity undergoing entropy change dSdS but the two are not simply related in all cases as discussed below.

We are interested in the microwork done during the mkeq\mathfrak{m}_{k\mathrm{eq}}-controlled evolution of mk(t)m_k(t) along a trajectory γk\gamma_k in a thermodynamic process P\mathcal{P} in Σˉ\bar{\Sigma}. The accumulated microwork along γk\gamma_k in step (S1) follows from Eq. (9), and is given by

ΔWk=−ΔEk(30)\Delta W_k=-\Delta E_k \tag*{(30)}

by the microenergy change ΔEk\Delta E_k along γk\gamma_k; recall Eq. (20). In step (S2), its ensemble average over P\mathcal{P} yields the SI-macrowork

ΔW=∫PdW(31)\Delta W=\int_{\mathcal{P}}dW \tag*{(31)}

in Gen-Th; it remains valid even if pkp_k remains constant over γk\gamma_k. In the cumulative form of Eq. (29), we have the identity

ΔQ=ΔW,(32)\Delta Q=\Delta W, \tag*{(32)}

where ΔQ\Delta Q is given by

ΔQ=∑k∫γkpkEk dηk,(33)\Delta Q=\sum_k\int_{\gamma_k}p_k E_k\,d\eta_k, \tag*{(33)}

and exists if and only if pkp_k does not remain constant during P\mathcal{P}. This is consistent with the conventional wisdom that the concept of heat does not apply to mechanical bodies for which ΔQ≡0\Delta Q\equiv0. While ΔW\Delta W is determined by the instantaneous value of pkp_k along γk\gamma_k, it does not determine how it changes along it. The latter is determined by γk∈P\gamma_k\in\mathcal{P} and determines dQdQ.

The macroworks dWdW and ΔW\Delta W play an important role in understanding the function EwE_w shown in Figure 1. We first introduce the cumulative macrowork

Weq′=∫Peq′dW≡⟨ΔW⟩Peq′(34)W_{\mathrm{eq}}'=\int_{\mathcal{P}_{\mathrm{eq}}'}dW\equiv\langle\Delta W\rangle_{\mathcal{P}_{\mathrm{eq}}'} \tag*{(34)}

along a process Peq′\mathcal{P}_{\mathrm{eq}}' starting from EQ point Meq\mathcal{M}_{\mathrm{eq}} to some point M\mathcal{M} on solid curves in Figure 1. We immediately see that Peq′\mathcal{P}_{\mathrm{eq}}' is in the direction opposite to the blue arrow for Ms\mathcal{M}^{\mathrm{s}} but in the direction of the red arrow for Mu\mathcal{M}^{\mathrm{u}}. We use it to introduce a macrowork function

Ew=Eeq−Weq′(35)E_w=E_{\mathrm{eq}}-W_{\mathrm{eq}}' \tag*{(35)}

of tt or ξ\xi. It is this average EwE_w that is shown in Figure 1, and differs from E=EeqE=E_{\mathrm{eq}} by the cumulative macrowork Weq′W_{\mathrm{eq}}'. We see that

dEw=−dW,(36)dE_w=-dW, \tag*{(36)}

a result that provides an equivalent justification of GSL, which is thermodynamically more intuitive regarding the role of the extremum of EwE_w in the evolution of M\mathcal{M}. We have focused on microstates to emphasize the role of mechanics for deriving GSL.

Different components of ξk\xi_k take different times when they vanish. They are called relaxation times so the components can be ordered according to them [10] as fast and slow. As ΣE\Sigma_E or ΣM\Sigma_M evolves in time, they appear or disappear at different times, making mkm_k more or less uniform, a point that is discussed in (C). For a single ξ\xi in the figure in the main text, Ms\mathcal{M}^s (Mu\mathcal{M}^u) becomes more and more (less and less) uniform as tt increases so that dS>(<)0dS > (<) 0 during any infinitesimal change dtdt.

Restriction Thermodynamics (Rest-Th)

We now put a particular form of restriction on {pk}\{p_k\} for M\mathcal{M} in GZ\mathcal{G}_Z of ΣM\Sigma_M that ensures that its EE becomes a state function of its SS and ξk\xi_k, which in turn means that SS is a state function of EE and ξk\xi_k in GZ\mathcal{G}_Z, whereas it is not a state function for ΣE\Sigma_E. Using EE as a state function E(S,ξE,ξV)E(S,\xi_E,\xi_V), NN and VV fixed, we obtain the Gibbs fundamental relation

dE=TdS−dW,dE = T dS - dW,

where T≐(∂E/∂S)ξT \doteq\left(\partial E/\partial S\right)_{\xi} is the thermodynamic temperature of ΣM\Sigma_M, FE≐−(∂E/∂ξE)S,ξVF_E \doteq-\left(\partial E/\partial\xi_E\right)_{S,\xi_V}, FV≐−(∂E/∂ξV)S,ξEF_V \doteq-\left(\partial E/\partial\xi_V\right)_{S,\xi_E} are SI-macroforces, and dW=FEdξE+FVdξVdW = F_Ed\xi_E + F_Vd\xi_V. It is the first law now in terms of dSdS and dWdW and yields the restriction thermodynamics (Rest-Th), which is nothing but Gen-Th for ΣM\Sigma_M. The first term is nothing but dQdQ in Eq. (28), which immediately establishes

dQ=TdS,T≐(∂E/∂S)ξ.(37)dQ = T dS, \quad T \doteq\left(\partial E/\partial S\right)_{\xi}. \tag*{(37)}

It should be noted that the relationship [8] is simply a the mathematical consequence of the state function E(S,ξk)E(S,\xi_k). However, the functional dependence E(S,ξk)E(S,\xi_k) still does not enforce SL (TdS≥0T dS \ge0) as was also the case for Gen-Th, so Rest-Th allows us to make direct connection with Cl-Th and Viol-Th. the justification for keeping TT in the new formulation TdS≥0T dS \ge0 will be justified in the main text within the context of generalized SL (GSL). We thus note that dQdQ and dSdS have the same signature for positive TT, but opposite signature for negative TT.

From Eq. (37) follows a simple but very remarkable and extremely profound relation

dQ∣E=TdS=dW∣E(38)dQ|_E = T dS = dW|_E \tag*{(38)}

for ΣM\Sigma_M in Gen-Th (we add an extra suffix EE as a reminder that EE is fixed, and should not be confused the suffix in ξE\xi_E) so it remains valid both in Cl-Th and Viol-Th for ΣM\Sigma_M:

TdS≥0⇒dW∣E≥0,TdS<0⇒dW∣E<0.(39)\begin{aligned} T dS &\ge0 \Rightarrow dW|_E \ge0,\\ T dS &< 0 \Rightarrow dW|_E < 0. \tag*{(39)} \end{aligned}

But what is most remarkable about Eq. (38) is that it provides a purely mechanical definition of stochastic entropy change by using the first law that holds in all kinds of thermodynamics discussed here. In addition, we also see from this equation that T≐(∂E/∂S)ξT \doteq(\partial E/\partial S)_{\xi} for ΣM\Sigma_M here is no different than TT in Eq. (7) for Σ\Sigma of any size so we use the latter as the definition of TT for any system of any size.

This completes the introduction of the Rest-Th (we remove the extra suffix EE in Eq. (39) for simplicity), which is applied below to a macrostate M\mathcal{M} of ΣM\Sigma_M used for Eq. (21), in which ξE\xi_E must be replaced by its ensemble average ξˉE≐E1/n1−E2/n2\bar{\xi}_E \doteq E_1/n_1 - E_2/n_2. A simple calculation using the state function S(E,ξk)S(E,\xi_k) and dE=0dE = 0 yields

dS=n1n2[(β1−β2)dξE+(β1P1−β2P2)dξV],(40)dS = n_1n_2\left[(\beta_1-\beta_2)d\xi_E + (\beta_1P_1-\beta_2P_2)d\xi_V\right], \tag*{(40)}

where β1\beta_1, P1P_1, and β2\beta_2, P2P_2 are the inverse temperature and pressure of Σ1M\Sigma_{1M} and Σ2M\Sigma_{2M}, respectively, and β≐1/T=n1β1+n2β2\beta\doteq1/T = n_1\beta_1+n_2\beta_2 is the inverse temperature of Σ\Sigma. The two terms on the right side in Eq. (40) represent entropic contributions due to the two internal variables ξE\xi_E and ξV\xi_V in Rest-Th; each must be nonnegative for SL or negative for its violation. In terms of dE1=−dE2dE_1 = -dE_2 as the macroenergy change and dV1=−dV2dV_1 = -dV_2 as the volume change of Σ1M\Sigma_{1M}, we have

dξE=1n1n2dE1,dξV=1n1n2dV1.d\xi_E = \frac{1}{n_1n_2}dE_1,\quad d\xi_V = \frac{1}{n_1n_2}dV_1.

Using γ=−1/T\gamma= -1/T introduced by Ramsey [15], whose numerical values define the ”hotness” of ΣM\Sigma_M as it increases from −∞-\infty to +∞+\infty covering positive and negative temperatures, Eq. (40) becomes

dQ=TdS=−(Δγ/γ)dE1−(Δ(γP)/γ)dV1(41)dQ = T dS = -\left(\Delta\gamma/\gamma\right)dE_1 - \left(\Delta(\gamma P)/\gamma\right)dV_1 \tag*{(41)}

for ΣM\Sigma_M in Rest-Th, where Δγ≐(γ2−γ1)\Delta\gamma\doteq(\gamma_2-\gamma_1), and Δ(γP)≐(γ2P2−γ1P1)\Delta(\gamma P) \doteq(\gamma_2P_2-\gamma_1P_1); Δγ>0\Delta\gamma> 0 means that Σ2M\Sigma_{2M} is hotter than Σ1M\Sigma_{1M}, and vice-versa. For GSL to hold, we require dSdS in dQ=TdSdQ = T dS to be ≥0\ge0 for T>0T > 0 (γ<0\gamma< 0) and <0< 0 for T<0T < 0 (γ>0\gamma> 0). For GSL violation, we require dS<0dS < 0 for T>0T > 0 and dS>0dS > 0 for T<0T < 0. Both situations are considered in the main text. Let us consider just the first term above to be specific by setting dV1=0dV_1 = 0. We consider various scenarios; (a) and (b) refer to Ms\mathcal{M}^s having T>0T > 0, and (c) and (d) refer to Mu\mathcal{M}^u having T<0T < 0. The analysis here is more extensive compared to an earlier preliminary and incomplete investigation [13], where the issue of catastrophic evaluation was first discussed.

  • (a) For dQ=dW>0dQ = dW > 0 and T>0T > 0, we must have dE1>0dE_1 > 0 for Ms\mathcal{M}^s so that macroheat flows from hot to cold as expected in which it converges to Mseq\mathcal{M}_{seq} due to an attractive SI-macroforce Fseq\mathbf{F}^{seq} pointing towards SEQSE_Q. The SI-evolution of Ms\mathcal{M}^s is spontaneous due to Fseq\mathbf{F}^{seq} pointing towards its sink Mseq\mathcal{M}_{seq}. Therefore, as expected in this case, dS>0dS > 0 so Cl-Th and Gen-GSL-Th remain valid. This is the most common situation.

  • (b) For dQ=dW<0dQ = dW < 0 and T>0T > 0, dE1<0dE_1 < 0 so that macroheat flows from cold to hot, and $\mathcal{M}^s

runs away from Mseq\mathcal{M}_{\mathrm{seq}} due to some repulsive macroforce Frepus\mathbf{F}^{\mathrm{s}}_{\mathrm{repu}}, distinct from the SI-macroforce Fseq\mathbf{F}^{\mathrm{seq}}, to eventually converge to Mcatas\mathcal{M}^{\mathrm{s}}_{\mathrm{cata}} by becoming more and more nonuniform. The evolution of Ms\mathcal{M}^{\mathrm{s}} is not spontaneous as is in (a) and Mseq\mathcal{M}_{\mathrm{seq}} is no longer the sink. In this case, dS<0dS < 0 (Viol-Th), but we also violate GSL (Gen–GSL-Th).

(c) For dQ=dW>0dQ = dW > 0 but T<0T < 0 for Mu\mathcal{M}^{\mathrm{u}}, dE1<0dE_{1} < 0 so that macroheat flows from cold to hot and ΣM\Sigma_{\mathrm{M}} becomes more and more nonuniform because of the instability in it as discussed in the main text. The spontaneous evolution of Mu\mathcal{M}^{\mathrm{u}} from its source Mueq\mathcal{M}_{\mathrm{ueq}} under the repulsive SI-macroforce Fueq\mathbf{F}^{\mathrm{ueq}} is catastrophic in that it converges to a catastrophic macrostate Mcatau\mathcal{M}^{\mathrm{u}}_{\mathrm{cata}}. In this case, dS<0dS < 0 so SL is violated (Viol-Th) but GSL (Gen–GSL-Th) remains valid.

(d) For dQ=dW<0dQ = dW < 0 and T<0T < 0, we must have dE1>0dE_{1} > 0 so that macroheat flows from hot to cold in Mu\mathcal{M}^{\mathrm{u}}. In this case, Mu\mathcal{M}^{\mathrm{u}} nonspontaneously converges to Mueq\mathcal{M}_{\mathrm{ueq}} due to an attractive macroforce, which is distinct from the repulsive SI-macroforce Fueq\mathbf{F}^{\mathrm{ueq}}, with dS>0dS > 0 so CI-Th remains valid but not Gen–GSL-Th. Thus, Mueq\mathcal{M}_{\mathrm{ueq}} is no longer the source for Mu\mathcal{M}^{\mathrm{u}}-evolution.

We analyze nonspontaneous processes further. We first consider (b). As Ms\mathcal{M}^{\mathrm{s}} runs away from its sink Mseq\mathcal{M}_{\mathrm{seq}} to a new macrostate Ms′\mathcal{M}^{\mathrm{s}\prime}, the latter further runs away from Ms\mathcal{M}^{\mathrm{s}} by its ξE\xi_{\mathrm{E}} deviating further from their values in Ms\mathcal{M}^{\mathrm{s}}. Thus, we get successive macrostates Ms(p)\mathcal{M}^{\mathrm{s}(p)}, p=0,1,2,⋯p = 0, 1, 2, \cdots which run away from the sink farther and farther, during which dS(p)dS^{(p)} remains non-positive as pp increases. Therefore, the evolution to Mcatas\mathcal{M}^{\mathrm{s}}_{\mathrm{cata}} is catastrophic in that it makes ΣM\Sigma_{\mathrm{M}} highly nonuniform due to unexplained nonsystem repelling macroforce Frepus\mathbf{F}^{\mathrm{s}}_{\mathrm{repu}} that mutilates Mseq\mathcal{M}_{\mathrm{seq}}.

It follows from the stability of Ms\mathcal{M}^{\mathrm{s}} considered here that Σ1M\Sigma_{1\mathrm{M}} and Σ2M\Sigma_{2\mathrm{M}} are also stable so their specific heats at constant volume are nonnegative. As dE1=−dE2<0dE_{1} = -dE_{2} < 0, the disparity Δγ\Delta\gamma continues to increase with Σ2M\Sigma_{2\mathrm{M}} getting more hot and Σ1M\Sigma_{1\mathrm{M}} getting more cold, until Δγ\Delta\gamma takes its maximum value Δγcata\Delta\gamma_{\mathrm{cata}} in Mcatas\mathcal{M}^{\mathrm{s}}_{\mathrm{cata}}.

We now consider (d), where a similar discussion can also be carried out for Mu\mathcal{M}^{\mathrm{u}} but different conclusions. Here, Mu\mathcal{M}^{\mathrm{u}} gets more uniform but the uniformity is not due to SI-macroforce so the evolution is not governed by SL. On the other hand, the catastrophe in (c) occurs spontaneously at negative TT so dS<0dS < 0 should be treated as being governed by SL. Thus, GSL seems to capture spontaneous processes at both positive and negative temperatures. In this sense, GSL subsumes SL.

The discussion is easily extended to include the second term in Eq. (41) with same conclusions, which is consistent with Lemma in the main text.

We clearly see from (a) and (d) that dS>0dS > 0 is not always a consequence of spontaneous processes. In (d), it is a consequence not only of negative TT but also of negative dWdW performed by nonsystem forces that result in a nonspontaneous process. This process is not controlled by SL so dS>0dS > 0 has no significance. Similarly, dS<0dS < 0 in (b) and (c) shows that it is not always a consequence of nonspontaneous processes. In (b), it is a consequence only of negative dWdW performed by nonsystem forces that result in a nonspontaneous process such as during the creation of internal constraints as explained by Callen [3]. Again, this process is not controlled by SL, while SI-macro­work dW>0dW > 0 or the removal of the internal constraint is controlled by SL so dS<0dS < 0 in (d) has no significance for SL-violation. From (a) and (c), we observe that GSL is always a consequence of spontaneous processes, but fails for nonspontaneous processes in (b) and (c).

References

  1. [1]L.D. Landau and E.M. Lifshitz, Statistical Physics, Third Ed., Pergamon Press, Oxford (1980).
  2. [2]The prefix micro- is used for quantities pertaining to microstates, while macro- is used for quantities pertaining to macrostates.
  3. [3]H.B. Callen, Thermodynamics and an introduction to thermostatis, Second Ed, John Wiley & Sons, New York (1985).
  4. [4]J.C. Maxwell, Theory of Heat, Longmans, Green, and Co., London (1902).
  5. [5]S.R. de Groot and P. Mazur, Nonequilibrium Thermodynamics, First Edition, Dover, New York (1984).
  6. [6]D. Kondepudi and I. Prigogine, Modern Thermodynamics, John Wiley and Sons, West Sussex (1998).
  7. [7]G.A. Maugin, The Thermodynamics of Nonlinear Irreversible Behaviors: An Introduction, World Scientific, Singapore (1999).
  8. [8]P.D. Gujrati, Entropy, 23, 1584 (2021).
  9. [9]J.W. Gibbs, Elementary Principles in Statistical Mechanics, Scribner’s Sons, N.Y. (1902).
  10. [10]P.D. Gujrati, Entropy 20, 149 (2018).
  11. [11]A.I. Arnold, Mathematical Methods of Classical Mechanics, Second Ed, Springer-Verlag, New York (1989).
  12. [12]L.D. Landau and E.M. Lifshitz, Mechanics, Third Ed., Pergamon Press, Oxford (1976).
  13. [13]P.D. Gujrati, Foundations, 1, 1 (2023).
  14. [14]P.D. Gujrati, Phys. Rev. E 102, 012140 (2020).
  15. [15]N.F. Ramsey, Phys. Rev. 103, 20 (1956).

Paper details

vlnc:W_37rqeuu5
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

Not recorded for this paper

View statement
Looking for related papers…
Contents