Quantum Thermodynamical Engines Nanodevices Landauer Limit Thermoelectric Power Dissipation

# Quantum Thermodynamics and Nanoscale Energy Conversion

Quantum thermodynamics asks how heat, work, information, and irreversibility behave when the working medium has only a few energy levels. At this scale fluctuations are not corrections to an average trajectory: they are part of the operating principle. A useful analysis therefore connects three descriptions. Hamiltonian dynamics determines energy levels and coherent strokes, open-system master equations describe contact with reservoirs, and stochastic thermodynamics assigns work and heat to individual realizations. This framework covers information erasure, quantum Otto engines, thermoelectric nanodevices, and fluctuation theorems without treating any one formula as a universal efficiency claim.

## Energy, heat, and work for an open quantum system

Let a system have density operator ρ(t) and a controlled Hamiltonian H(t). Its internal energy is

Differentiation separates the change into

The first term is heat associated with redistribution of populations by an environment; the second is work associated with changing the energy landscape. This split depends on the declared system boundary and driving protocol. It is clean in the weak-coupling, Markovian limit, but strong coupling requires including interaction energy or using a Hamiltonian of mean force.

A common reservoir model is the Lindblad master equation

For a harmonic oscillator coupled to a bath with occupation n_th, choose L_- = a and L_+ = a†, with rates γ(n_th + 1) and γn_th. Then ⟨n⟩ relaxes exponentially toward the Bose occupation [exp(ℏω/k_BT) - 1]⁻¹. For a two-level medium, replace it with the Fermi-like excitation probability [1 + exp(ℏω/k_BT)]⁻¹. Confusing these working media reverses signs and changes predicted work.

The von Neumann entropy is S = -k_B Tr(ρ ln ρ). For a thermal reservoir at temperature T, entropy production obeys

Coherence can store ergotropy, the portion of energy extractable by a cyclic unitary, but coherence generation and disposal must be included in a complete accounting. Apparent performance beyond a thermal bound generally signals an uncounted resource.

## Landauer erasure from phase-space compression

One bit with equiprobable logical states has entropy S_i = k_B ln 2. Resetting it to a known state has S_f = 0, so the memory entropy changes by ΔS_mem = -k_B ln 2. The second law requires the environment to gain at least the missing entropy:

For an isothermal quasistatic protocol, the minimum supplied work equals the nonequilibrium free-energy increase,

At 300 K, the bound is about 2.87 × 10⁻²¹ J, or 17.9 meV, per unbiased bit. It is an asymptotic lower bound, not the switching energy of a practical transistor. Finite-time control, leakage, clocking, error correction, and interconnect charging make real dissipation larger.

For a bit whose initial probability of logical one is p, its information entropy is k_B h(p), where

Resetting it reversibly costs at least k_B T h(p), not automatically k_B T ln 2. If the reset error is ε, the final entropy is k_B h(ε), reducing the ideal cost to k_B T[h(p) - h(ε)]. The controller must still know and maintain the distribution; exploiting biased data is a form of compression.

A concrete double-well protocol lowers the barrier, tilts the potential so probability moves to the chosen well, raises the barrier, and removes the tilt. In the quasistatic limit every intermediate distribution is near equilibrium. Driving faster creates lag and excess work. Linear-response theory often gives W_ex ∝ 1/τ for duration τ, while very abrupt driving can create a much larger penalty.

Landauer's principle applies to logically irreversible maps. Reversible gates may ideally avoid this specific entropy cost, yet eventually resetting ancillas, syndromes, or output storage restores the accounting. Measurement does not intrinsically require k_BTln2 per outcome; resetting the measurement record does.

## Quantum Otto engine

Consider a harmonic oscillator whose controllable frequency switches between ω_c and ω_h. An ideal Otto cycle has four strokes:

1. Cold isochore: at ω_c, the medium thermalizes to occupation n_c.
2. Isentropic compression: isolated driving changes ω_c→ω_h, preserving populations.
3. Hot isochore: at ω_h, thermalization produces occupation n_h.
4. Isentropic expansion: isolated driving returns ω_h→ω_c, again preserving populations.

Ignoring zero-point terms that cancel over the cycle, heat absorbed from the hot reservoir is

and heat delivered to the cold reservoir has magnitude

The positive work output is therefore

With the sign convention that work done on the system is positive, W_net=-W_out. Stating the convention prevents a frequent false conclusion that the same engine both consumes and produces positive work.

The ideal efficiency is

Positive work requires n_h>n_c. For thermal harmonic reservoirs this implies ω_h/T_h<ω_c/T_c, hence

Equality is approached only at the boundary where output vanishes. Maximum efficiency and maximum power are different design points.

During finite isochores, a scalar occupation model is

Solving the periodic steady state yields the actual corner occupations and cycle power P=W_out/τ_cyc. Shorter cycles increase repetition rate but reduce thermalization. Fast frequency ramps also generate quantum friction: nonadiabatic transitions leave excess excitations that reduce useful work. Counterdiabatic control can suppress them, but the control field's energetic cost belongs in the denominator.

## Fluctuation relations

For a system initially equilibrated at inverse temperature β=1/k_BT, driven by an arbitrary protocol, the Jarzynski equality is

Jensen's inequality immediately gives ⟨W⟩ ≥ ΔF. The equality is stronger because it constrains the entire work distribution, including rare trajectories with W < ΔF. In a quantum two-projective-measurement protocol, one measures initial energy, applies the drive, measures final energy, and defines W = E_m^f - E_n^i. This operational definition destroys initial energy coherence; alternative quasiprobability formulations must be identified explicitly.

The Crooks relation compares forward and time-reversed work distributions:

Their crossing occurs at W = ΔF, providing an experimental estimator. Exponential averaging is dominated by rare low-work events, so naive finite samples are biased. Bidirectional estimators and uncertainty analysis are essential.

For nonequilibrium steady transport, integral fluctuation theorems often take ⟨e^(-Σ/k_B)⟩ = 1. They imply average nonnegative entropy production while allowing negative values on individual short trajectories. Such events do not violate the second law; their probability falls exponentially with magnitude and observation time.

## Quantum thermoelectric nanodevices

A coherent two-terminal conductor is described by a transmission function T(E). Landauer transport gives charge and heat currents

Define moments

Then G=e²L_0, S=-L_1/(eTL_0), and the electronic thermal conductance is κ_e=(L_2-L_1²/L_0)/T. The power factor is GS², while

A narrow transmission window can raise thermopower and electronic ZT, but narrowing it also reduces conductance and power. Phonon heat leakage, contact resistance, Coulomb blockade, and lifetime broadening decide the practical optimum. Efficiency at maximum power is not determined by ZT alone outside linear response.

For a single resonant level,

Gate voltage moves E_0 through the Fermi window. Thermopower changes sign as the level crosses μ, a useful diagnostic of electron-like versus hole-like transport.

## Numerical simulator

The following self-contained Python model evaluates the Landauer limit, an ideal and finite-time oscillator Otto engine, a Jarzynski consistency test for Gaussian work, and linear-response thermoelectric coefficients. Frequencies are angular frequencies.

import numpy as np
import matplotlib.pyplot as plt
from scipy.constants import hbar, h, k, e
from scipy.integrate import trapezoid

def bose(omega, temperature):
    x = hbar * omega / (k * temperature)
    return 1.0 / np.expm1(np.clip(x, 1e-12, 700.0))

def landauer_heat(temperature, bits=1.0, p=0.5, error=0.0):
    def entropy(q):
        q = np.clip(q, 1e-15, 1 - 1e-15)
        return -q*np.log(q) - (1-q)*np.log(1-q)
    return bits * k * temperature * max(entropy(p)-entropy(error), 0.0)

def periodic_corners(nc_eq, nh_eq, gc_tc, gh_th):
    ac, ah = np.exp(-gc_tc), np.exp(-gh_th)
    # x: occupation after cold isochore; y: after hot isochore
    x = (nc_eq*(1-ac) + ac*nh_eq*(1-ah)) / (1-ac*ah)
    y = nh_eq + (x-nh_eq)*ah
    return x, y

def otto(omega_c, omega_h, tc, th, gamma_c, gamma_h, cycle_overhead=0):
    nc_eq, nh_eq = bose(omega_c, tc), bose(omega_h, th)
    nc, nh = periodic_corners(nc_eq, nh_eq, gamma_c*tc*0 + gamma_c,
                              gamma_h*th*0 + gamma_h)
    # Arguments gamma*t are supplied below explicitly for clarity.
    ac, ah = np.exp(-gamma_c*tc), np.exp(-gamma_h*th)
    nc = (nc_eq*(1-ac)+ac*nh_eq*(1-ah))/(1-ac*ah)
    nh = nh_eq+(nc-nh_eq)*ah
    qh = hbar*omega_h*(nh-nc)
    work = hbar*(omega_h-omega_c)*(nh-nc)
    eta = work/qh if qh > 0 else np.nan
    power = work/(tc+th+cycle_overhead)
    return work, eta, power, nc, nh

Tc, Th = 0.08, 0.20
wc = 2*np.pi*2.0e9
ratios = np.linspace(1.02, Th/Tc*0.995, 250)
results = np.array([otto(wc, r*wc, 2e-6, 2e-6, 8e5, 6e5, 0.5e-6)[:3]
                    for r in ratios])

temperatures = np.geomspace(0.01, 400, 300)
qbit = k*temperatures*np.log(2)

# Gaussian work satisfying Jarzynski when mean = dF + beta*variance/2.
rng = np.random.default_rng(7)
Tj, dF, sigma = 1.0, 2.0e-22, 0.8e-22
beta = 1/(k*Tj)
mean = dF + 0.5*beta*sigma**2
work_samples = rng.normal(mean, sigma, 2_000_000)
dF_est = -np.log(np.mean(np.exp(-beta*work_samples)))/beta
print('Jarzynski dF true, estimate:', dF, dF_est)
print('Landauer at 300 K:', landauer_heat(300), 'J/bit')

# Resonant-level thermoelectric response.
T = 10.0
mu = 0.0
energy = np.linspace(-20, 20, 20001)*k*T
gamma = 1.5*k*T
levels = np.linspace(-8, 8, 161)*k*T
G, Seebeck, ZTe = [], [], []
minus_df = 1/(4*k*T)/np.cosh(energy/(2*k*T))**2
for E0 in levels:
    transmission = (gamma/2)**2/((energy-E0)**2+(gamma/2)**2)
    moments = [2/h*trapezoid((energy-mu)**n*transmission*minus_df, energy)
               for n in range(3)]
    L0, L1, L2 = moments
    conductance = e**2*L0
    thermopower = -L1/(e*T*L0)
    kappa_e = (L2-L1**2/L0)/T
    G.append(conductance); Seebeck.append(thermopower)
    ZTe.append(conductance*thermopower**2*T/kappa_e)

fig, ax = plt.subplots(2, 2, figsize=(11, 8))
ax[0,0].loglog(temperatures, qbit)
ax[0,0].set(xlabel='temperature (K)', ylabel='minimum heat (J/bit)', title='Landauer bound')
ax[0,1].plot(ratios, results[:,1], label='Otto efficiency')
ax[0,1].axhline(1-Tc/Th, ls='--', label='Carnot')
ax[0,1].set(xlabel=r'$\omega_h/\omega_c$', ylabel='efficiency'); ax[0,1].legend()
ax[1,0].plot(ratios, results[:,2]*1e18)
ax[1,0].set(xlabel=r'$\omega_h/\omega_c$', ylabel='power (aW)', title='Finite-time power')
ax[1,1].plot(levels/(k*T), np.array(Seebeck)*1e6, label='S (uV/K)')
ax[1,1].plot(levels/(k*T), ZTe, label='electronic ZT')
ax[1,1].set(xlabel=r'$(E_0-\mu)/k_BT$'); ax[1,1].legend()
fig.tight_layout(); plt.show()

The finite-time corner calculation must be iterated or solved algebraically, as above; substituting equilibrium occupations directly silently assumes infinitely long isochores. A useful extension integrates the full Lindblad equation and includes a ramp-dependent friction factor.

## Measurement and design workflow

Begin by drawing the system boundary and declaring positive heat and work directions. Calibrate reservoir temperatures using the device's own transition rates when possible, because cryostat temperature may differ from electron or phonon temperature. Measure all four cycle corners rather than inferring work from only one spectrum. For thermoelectrics, independently measure electrical conductance, open-circuit thermovoltage, and heat flow; parasitic lead conduction can dominate the last quantity.

Report power together with efficiency, cycle time, uncertainty, and control overhead. For erasure, report logical error probability and initial bias. For fluctuation-theorem tests, publish the number of trajectories and convergence versus sample count. For quantum devices, compare energy-basis populations and coherences so that an apparent coherent resource is not mislabeled as reservoir heat.

Useful dimensionless groups are ℏω/k_BT for quantumness, Γτ for thermalization, ωτ_ramp for adiabaticity, and ZT for linear-response thermoelectric quality. No single group establishes a useful engine: high power needs adequate coupling, high efficiency needs selectivity and low leakage, and stable operation needs fluctuations small enough for the intended load.

## Common failure modes

• Treating k_BTln2 as the energy of every logic operation rather than the reversible lower bound for erasing one unbiased bit.
• Quoting Otto efficiency where the positive-work condition is not satisfied.
• Comparing an ideal cycle's work with a measured input that excludes control, reset, or refrigeration costs.
• Using Carnot efficiency as an attainable finite-power target.
• Applying Jarzynski averaging to a state that was not initially thermal or to inconsistently defined work.
• Estimating ZT while omitting phonon thermal conductance.
• Calling transient reservoir energy change heat without specifying the system boundary in a strongly coupled device.
• Reporting only averages when work and heat distributions are broad.

## Outlook

Quantum thermodynamic experiments are most valuable when they expose a controllable crossover: adiabatic to frictional driving, equilibrium to finite-time transport, classical to quantum fluctuations, or information gain to reset cost. Nanoscale engines will not evade the second law, but energy filtering, feedback, coherence, and reservoir engineering can redistribute losses and improve a chosen objective. The durable result is a complete resource ledger—one that counts information, controls, correlations, and heat leaks alongside the useful work.