信息热动力学 Part II
Thermodynamics of Information II

报告人:凌峰

亥姆霍兹慕尼黑 / 慕尼黑工业大学 - 亥姆霍兹先锋园区、生物医学成像研究中心

Helmholtz Munich / TUM - Helmholtz Pioneer Campus / IBMI

非平衡统计物理读书会 - 信息热动力学​​

  • 包含信息的涨落定理​(Fluctuation Theorem)

    • 涨落定理基本推导简述

  • 稳定记忆与Landauer原理
  • 测量过程的代价

  • 连续信息流

  • 研究前景与展望

  • 题外话
    • 信息热力学能否用于区分微观环境下的闭环操控与开环涌现?

分享大纲

非平衡统计物理读书会 - 信息热动力学​​

Recall that without explicitly writing the information component we have

[Clausius Inequality] \(\qquad\qquad\quad\;\; W - \Delta \mathcal F\geq 0\)

[Crooks Fluctuation Theorem] \(\,\quad P(x(t))/P(x_{\text{reverse}}(t)) = \exp[(w-\Delta F)/k_BT]\)

[Integral Fluctuation Theorem] \(\;\;\; \langle e^{-(w-\Delta F)/k_BT}\rangle=1\)

Analogously, from our information-aware Clausius-like inequality (eq.5)

\[W-\Delta \mathcal F\geq -k_BT I\]

One can derive (eq.6)

\[\frac{\mathcal P(\text{forward trajectory},m)}{\mathcal P(\text{reverse trajectory},m)}=e^{(w-\Delta F)/k_BT + \mathcal I}\]

and (eq.7)

\[\langle e^{-(w-\Delta F)/k_BT - \mathcal I} \rangle=1\]

Fluctuation Theorem with Information

非平衡统计物理读书会 - 信息热动力学​​

Consider a finite, classical system in contact with a thermal bath of temperature \(T\)

let \(\lambda\) denote an externally controllable parameter

\([\lambda,T]\) specifies an equilibrium state

system evolves from equilibrium \([A,T]\) at \(t=0\)

to \(\lambda=B\) at \(t=\tau\) in non-equilibrium,

then relax to equilibrium \([B,T]\) at \(t=\tau*\) (no work performed)

Clausius Inequality \(\leftrightarrow\) Jarzynski Equality

[Clausius Inequality]

\[W=\int_0^\tau\mathrm d t\,\dot{\lambda} \frac{\partial \mathcal H(x(t);\lambda(t))}{\partial \lambda}\geq\Delta F\equiv F_{[B,T]} - F_{[A,T]}\]

Zarzynski (2010), 10.1146/annurev-conmatphys-062910-140506

非平衡统计物理读书会 - 信息热动力学​​

non-equilibrium

equilibrium bath T

[A,T]

[B,T]

[B]

\(t=0\)

\(t=\tau*\)

\(t=\tau\)

\(W\)

relax

\(\Delta F\)

non-equilibrium work relation \(W = \mathcal H(x_\tau; B) -\mathcal{H} (x_0 ; A)\) gives
 \[\langle e^{-W/k_BT}\rangle=\int \mathrm d\bm{x}_0\,p_{[A,T]}(\bm x_0)e^{-W/k_BT}\]

\[=\frac{1}{Z_{[A,T]}}\int \mathrm d\bm{x}_\tau \,e^{-\mathcal H(x_\tau;B)/k_BT}\cancel{\left|\frac{\partial \bm{x}_\tau}{\partial \bm{x}_0}\right|^{-1}}\]
\[=\frac{Z_{[B,T]}}{Z_{[A,T]}}=e^{-\Delta F/k_BT}\]

Clausius Inequality \(\leftrightarrow\) Jarzynski Equality

[Jensen's Inequality] \(\langle exp(\cdot)\rangle\geq exp(\langle\cdot\rangle)\) recovers

\[-\Delta F/k_BT\geq-W/k_BT\text{ or } W\geq\Delta F\]

[Louville's theorem]

Zarzynski (2010), 10.1146/annurev-conmatphys-062910-140506

非平衡统计物理读书会 - 信息热动力学​​

[Jarzynski Equality]

Note: Jarzynski Equality can be seen as a specific case of the [Integral Fluctuation theorem]

\[\langle e^{-(W-\Delta V)/k_BT} \, \frac{p_1(\bm x_\tau)}{p_0(\bm x_0)}\rangle=1\]

with a specific choice of the arbitrary normalized distribution \(p_1,p_0\) that cancels out the potential term and introduce microstate independent free energy.

Jarzynski Equality provide us a non-equilibirum counterpart to the following equality as a result of reversible work relation \(W=\Delta F\)

\[-k_BT\log\langle e^{-W/k_BT}\rangle=\Delta F\]

Clausius Inequality \(\leftrightarrow\) Jarzynski Equality

Seifert (2008), 10.1140/epjb/e2008-00001-9

Seifert (2005), 10.1103/PhysRevLett.95.040602

非平衡统计物理读书会 - 信息热动力学​​

非平衡统计物理读书会 - 信息热动力学​​

The non-equilibrium free energy of memory M is (eq.8)

\[\mathcal F(M) = \langle \mathcal H\rangle_\rho - TS(\rho) = \sum_m p_m F_m - k_BT H(M)\]

Manipulating memory should result in a loop in terms of Hamiltonian.

Thus work to shift one memory state \(p_m\) to another \(p'_m\) satisfies (eq.9)

\[W-\Delta \mathcal F_\text{memory} = W - (\mathcal F(M') - \mathcal F(M))\geq0\]

for symmetric storage devices, without energy differential only the Shannon entropy/information changes,

\[W\geq -k_BT[ H(M') - H(M)]\]

Thus resetting memory back from zero-information state requires (Landauer's principle)

\[W_{erase}\geq k_BT H(M)\]

Stable Memory & Landauer's Principle

非平衡统计物理读书会 - 信息热动力学​​

Information Reservoirs - Maxwell's Refrigerator

Mandal, Zarzynski (2012), 10.1073/pnas.1204263109

Mandal, Quan, Zarzynski (2013), 10.1103/PhysRevLett.111.030602

非平衡统计物理读书会 - 信息热动力学​​

Non-equilibrium free energy without interaction is (eq.11)

\[\mathcal F(XM)=\mathcal F(X) + \mathcal F(M) + k_BT\,I(X;M)\]

Measuring non interacting system requires (eq.13)

\[W_\text{measurement}\geq\Delta\mathcal F_{total} =\Delta\mathcal F_{memory} + kT\,I(X;M)\]

RHS cancels out for symmetric memory set-up discussed before! => Error-free measurement can be done with \(W\geq0\)!

Cost of Measurement

非平衡统计物理读书会 - 信息热动力学​​

Feedback that depends on the memory bounds free energy change to be 

\[\Delta \mathcal F_\text{feedback} = \Delta \mathcal F - k_BT\,I(X;M)\]

Thus the extracted work W over the full cycle satisfies

\[W_\text{measurement} + W\geq\Delta\mathcal F_\text{memory} + \Delta \mathcal F\]

Since resetting reverses the memory effort \(\Delta\mathcal F_\text{reset} = -\Delta\mathcal F_\text{memory}\),

\[W_\text{measurement} + W_\text{reset}\geq k_BT\,I(X;M)\]

Cost of Measurement

非平衡统计物理读书会 - 信息热动力学​​

with Continuous Information Flow

If information exchange between subsystems
continuously, we can analyze entropy
production rate instead.

Consider subsystems \(X,Y\) as in figure (eq.15)

\[\dot S_{XY} = \dot S_X + \dot S_Y\]

\[\dot S_X = \mathrm d_t S_X + \dot S_X^\text{reservoir} - k_B \dot I_X \geq 0\]

\[\dot S_Y = \mathrm d_t S_Y + \dot S_Y^\text{reservoir} - k_B \dot I_Y \geq 0\]

Here \(\dot S^\text{reservoir}\) indicating rates of entropy change in reservoir: e.g., \(\dot S^\text{res} = -\dot Q / T\) for isothermal process.

Horowitz, Esposito (2014), 10.1103/PhysRevX.4.031015

非平衡统计物理读书会 - 信息热动力学​​

with Continuous Information Flow

Importantly, \(\dot I\) represent the time variation of the
mutual information

\[\dot I_X = \sum_{x\geq x'; y} p({x\rightarrow x'})\log\frac{p(y|x)}{p(y|x')}\]

\[\dot I_Y = \sum_{x;y\geq y'} p({y\rightarrow y'})\log\frac{p(x|y)}{p(x|y')}\]

Here \(p(a\rightarrow b)\) represents the transition current (see reference).

\(\dot I_X>0\) implies X is measuring Y; \(\dot I_X<0\) implies X is erasing information about Y, i.e. resetting.

Horowitz, Esposito (2014), 10.1103/PhysRevX.4.031015

非平衡统计物理读书会 - 信息热动力学​​

Example Flow Systems

Horowitz, Esposito (2014), 10.1103/PhysRevX.4.031015

非平衡统计物理读书会 - 信息热动力学​​

(a) Quantum-dot Maxwell demon

(b) Transmembrane sensor with
      adaptive rate/stability

(c) Transmembrane sensor that
      promotes protein with switch

(d) Brownian particle in switchable
      energy landscapes

Outlook and Perspectives

非平衡统计物理读书会 - 信息热动力学​​

Generalizations to Consider

  • Key quantity relevant for repeated measurement & feedback:
    • Transfer entropy: captures new information per measurement
  • Thermodynamic costs as a function of memory architecture:
    • Single-memory multiple-rewrite vs. multiple-memory interactions.
  • Fluctuation theorems extended to systems without detailed balance:
    • autonomous Maxwell's demon and ratchets.
    • optimal information motors can give rise to different entropy production compared to chemical motors[25].

  • Quantum Szilárd engine models and quantum fluctuation theorems.

非平衡统计物理读书会 - 信息热动力学​​

Applications for Biological Sensors

  • Copolymerization (DNA/RNA replication/transcription):

    • Thermodynamic accuracy linked to entropic forces.

    • Kinetic proofreading:

      • Trade-off between accuracy and dissipation.

      • Biological implications for transcription kinetics.

  • Sensing & adaptation:

    • Energy required to gather environmental information.

    • Minimum information needed for robust adaptation.

    • Predictive information sets minimal work for forecasting environmental fluctuations.

非平衡统计物理读书会 - 信息热动力学​​

Open Questions

  • Fundamental debates:

    • Connection between thermodynamic and psychological arrows of time [93].

  • Unifying Framework:

    • A comprehensive theory that works for diverse processes.

  • Revisiting Maxwell’s original challenge:

    • Role of information in macroscopic physics and statistical mechanics.

    • Subjectivity of entropy in information-driven systems.

非平衡统计物理读书会 - 信息热动力学​​

open-loop phenomenon vs close-loop control

  • Inspiration related to my own work: for microscopic biological machinery (e.g. cilia/flagella molecular motor machines), can information-aware theory distinguish between open-loop 'mechanics-only' phenomenon and tightly regulated close-loop feedback mechanisms?

非平衡统计物理读书会 - 信息热动力学​​

信息热动力学 - 集智

By Feng Ling

信息热动力学 - 集智

  • 0