Minimal Power Dissipation Principle

• The principle of minimal power dissipation defines how systems optimize performance by minimizing irreversible energy losses under specified constraints.
• In digital electronics and stochastic thermodynamics, models employ limits like Landauer’s bound and Wasserstein distances to quantify and design minimal dissipation processes.
• This principle guides the development of reversible computing, adiabatic circuits, and continuum flow models, balancing trade-offs between speed, accuracy, and energy efficiency.

Searching arXiv for the cited papers and closely related work on minimal dissipation principles. I’m checking whether the arXiv search tool is available in this environment. The principle of minimal power dissipation denotes a family of variational and thermodynamic statements according to which an admissible computation, transport process, or flow is selected—or optimally designed—by minimizing irreversible loss subject to specified constraints. In digital electronics, it is tied to Landauer’s observation that logically irreversible operations dissipate heat; in stochastic thermodynamics, it becomes a finite-time bound on dissipated work; in linear and nonlinear irreversible thermodynamics, it is formulated through Onsager- or Rayleigh-type functionals; and in continuum mechanics, related minimum-dissipation statements characterize Stokes flow and other dissipative media (Singla et al., 2012, Gao et al., 2021, Oikawa et al., 3 Mar 2025).

1. Thermodynamic meaning and lower bounds

A canonical starting point is Landauer’s limit for bit erasure. Erasing one bit reduces the Shannon–Boltzmann entropy of the computational device by

$\Delta S=-k_B\ln 2,$

so the second law requires at least

$Q\ge T|\Delta S|=k_B T\ln 2$

of heat to be transferred to the environment. The corresponding lower bound on energy dissipation is

$E_{\min}=k_B T\ln 2.$

In the detailed derivation, the number of accessible states is reduced from $2$ to $1$, yielding

$$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$

and Clausius’s inequality then gives the heat bound. In this formulation, the essential thermodynamic event is not switching per se, but the irreversible merging of distinguishable logical states, as in bit erasure or in an AND gate whose inputs cannot be reconstructed from its output (Singla et al., 2012).

A stochastic-thermodynamic formulation expresses the same idea at trajectory level. In the minimal stochastic model for complementary logic gates, each transistor–reservoir junction is treated as a two-state Markov jump process satisfying local detailed balance,

$$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$

The total entropy production over an observation time $\tau_{\rm obs}$ is written as

$$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$

and the average dissipated power is identified with

$$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$

Within this framework, one-bit erasure again obeys $Q\ge T|\Delta S|=k_B T\ln 2$0, while finite-time switching produces additional entropy beyond that lower bound (Gao et al., 2021).

A central point across these formulations is that the “minimal” quantity depends on the problem specification. In logical operations it is the unavoidable heat associated with entropy reduction; in stochastic dynamics it is entropy production or irreversible work; and in other fields it may be viscous dissipation or a generalized excess-dissipation functional. This suggests that the principle is best understood as a constrained extremum principle rather than a single universal formula.

2. Logical reversibility, physical reversibility, and adiabatic electronics

Logical reversibility means that the input–output map is one-to-one. Gates such as NOT, the $Q\ge T|\Delta S|=k_B T\ln 2$1 Feynman gate, the $Q\ge T|\Delta S|=k_B T\ln 2$2 Toffoli gate, and the Fredkin gate do not lose information, whereas NAND, NOR, and AND are logically irreversible because the inputs cannot be reconstructed from the output alone. Physical reversibility is stricter: it requires the physical implementation to generate no entropy and dissipate no heat in the ideal limit, which in turn requires quasi-static switching. Conventional CMOS can therefore be logically reversible yet physically irreversible. Even a logically reversible inverter implemented with a standard CMOS pair dissipates

$Q\ge T|\Delta S|=k_B T\ln 2$3

on each charging and discharging event because abrupt rail-to-rail switching dumps energy into channel resistance and ground (Singla et al., 2012).

Adiabatic logic addresses this contradiction by replacing abrupt switching with slow, energy-recovering switching. In the adiabatic CMOS inverter, the fixed $Q\ge T|\Delta S|=k_B T\ln 2$4 rail is replaced by a slowly varying trapezoidal or multi-phase clock waveform $Q\ge T|\Delta S|=k_B T\ln 2$5. If the output capacitance $Q\ge T|\Delta S|=k_B T\ln 2$6 is charged through a resistance $Q\ge T|\Delta S|=k_B T\ln 2$7 using a ramp rather than a step, then

$Q\ge T|\Delta S|=k_B T\ln 2$8

and the cycle dissipation obeys

$Q\ge T|\Delta S|=k_B T\ln 2$9

The reversible-inverter example uses a four-phase trapezoidal power clock; by stretching the linear rise and fall times far beyond the $E_{\min}=k_B T\ln 2.$0 time constant, the conventional $E_{\min}=k_B T\ln 2.$1 loss can in principle be reduced arbitrarily close to zero (Singla et al., 2012).

The same adiabatic principle was applied to a $E_{\min}=k_B T\ln 2.$2T SRAM cell in $E_{\min}=k_B T\ln 2.$3 nm CMOS. There, replacing the DC rail with a multi-phase power clock reduces average power dissipation in simulation by up to $E_{\min}=k_B T\ln 2.$4, and a detailed comparison reports reductions of $E_{\min}=k_B T\ln 2.$5 for Write “0”/“1”, $E_{\min}=k_B T\ln 2.$6 for Write + Hold, and $E_{\min}=k_B T\ln 2.$7 for Write + Read. The associated trade-off is visible in the static noise margin, which drops from $E_{\min}=k_B T\ln 2.$8 V in the conventional cell to $E_{\min}=k_B T\ln 2.$9 V in the adiabatic cell (Jadav et al., 2012).

A complementary CMOS-design perspective decomposes total power into switching, short-circuit, and leakage components, often writing

$2$0

with

$2$1

Subthreshold leakage is approximated by

$2$2

so lowering $2$3 yields quadratic dynamic-power savings while lowering $2$4 restores speed but raises leakage exponentially. This gives a second, device-level interpretation of minimal dissipation: power is minimized not by arbitrary voltage reduction, but by co-optimizing $2$5 and $2$6 under a timing constraint and then using leakage-control techniques such as device stacking, multi-threshold CMOS, and reverse body-biasing (Kaur et al., 2013).

3. Finite-time dissipation and stochastic design principles

In finite-time stochastic thermodynamics, the problem is no longer only to identify a quasi-static lower bound but to determine the minimal excess dissipation at fixed duration. For an overdamped Brownian particle in a time-dependent potential $2$7, the probability density obeys the Fokker–Planck equation

$2$8

If the system is driven between $2$9 and $1$0 in time $1$1, the average work splits as

$1$2

and the minimal dissipated work is bounded by

$1$3

where $1$4 is the $1$5-Wasserstein distance. The bound is achievable by transporting the probability density along the geodesic path in distribution space at uniform speed (Oikawa et al., 3 Mar 2025).

For information erasure, this refines Landauer’s bound. If the initial state is a symmetric double-well equilibrium encoding one bit and the final state is a single-well equilibrium storing logic “0,” then

$1$6

and the finite-time erasure bound becomes

$1$7

The excess dissipation beyond $1$8 is therefore exactly the geometric term $1$9. The same framework yields a speed–dissipation–accuracy hierarchy, including

$$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$0

and

$$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$1

where $$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$2 is an error-distance relative to the perfect reset distribution $$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$3 (Oikawa et al., 3 Mar 2025).

The stochastic circuit model for logic gates makes the trade-off explicit at gate level. For a NOT gate, the reversible charging cost is

$$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$4

the error probability decays asymptotically as

$$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$5

the propagation delay scales as

$$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$6

and the total entropy production over an observation time satisfies

$$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$7

with leakage dissipation rate $$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$8. Higher $$\Delta S = k_B\ln(W_{\rm final}/W_{\rm initial})=k_B\ln(1/2)=-k_B\ln2,$$9 suppresses error but slows the gate and raises reversible charging cost; lower $$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$0 speeds switching but degrades accuracy and may increase leakage. The resulting design rules favor smooth protocols, leakage suppression, shallow logic depth, and reuse of residual gate charge across successive operations (Gao et al., 2021).

A related finite-time formulation in linear response recasts irreversible work as a quadratic functional of the driving speed,

$$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$1

or, with $$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$2 defined by $$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$3, $$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$4,

$$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$5

The minimizing protocols develop initial and final steps, and in the fast or underdamped regime their derivatives acquire sharply peaked boundary structures approaching $$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$6-functions. In this setting, minimal dissipation is a protocol-design problem for nonlocal response kernels rather than a static bound alone (Bonança et al., 2018).

4. Onsager, Rayleigh, and generalized variational formulations

In linear irreversible thermodynamics, minimal dissipation is expressed in terms of conjugate fluxes and forces. With $$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$7 and $$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$8, two Rayleigh dissipation functions are introduced: $$\ln\!\bigl[k_{ji}/k_{ij}\bigr]=-\beta(E_j-E_i), \qquad \beta\equiv 1/(k_B T).$$9

$\tau_{\rm obs}$0

Mauri’s formulation defines a Hamiltonian time-rate

$\tau_{\rm obs}$1

along the minimizing path, with $\tau_{\rm obs}$2 constant in time on that path. The action

$\tau_{\rm obs}$3

is extremized under fixed endpoints, producing Euler–Lagrange equations that, for constant $\tau_{\rm obs}$4, reduce to

$\tau_{\rm obs}$5

At steady state, maximizing $\tau_{\rm obs}$6 with fixed generalized forces yields the largest possible flux and maximal entropy production, while stationarity at fixed flux minimizes the required force and hence the entropy production. Within its assumptions—linearity, near-equilibrium, conservative forces, and Gaussian fluctuations—this formulation unifies least-dissipation and minimum-entropy-production statements (Mauri, 2015).

A more specialized two-force linear-response treatment considers forces $\tau_{\rm obs}$7, fluxes $\tau_{\rm obs}$8, entropy production

$\tau_{\rm obs}$9

power

$$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$0

and efficiency

$$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$1

Optimizing with respect to the load force $$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$2 gives distinct regimes of maximum power, maximum efficiency, and minimum dissipation. For minimum dissipation,

$$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$3

with

$$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$4

$$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$5

Under Onsager symmetry $$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$6, the relations simplify to

$$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$7

In that symmetric case, reversible operation coincides with zero power at minimum dissipation (Proesmans et al., 2016).

The same linear-response logic underlies the thermoelectric “small dissipation” limit. For a two-terminal thermoelectric device, the ideal or strong-coupling condition is

$$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$8

which yields Carnot efficiency but vanishing cooling power. Writing

$$\Sigma[\tau_{\rm obs}] = \sum_r\int_0^{\tau_{\rm obs}} dt\, J_r(t)F_r,$$9

one finds that efficiency and power deviate linearly in $$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$0, while total dissipation scales quadratically: $$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$1 Equivalently, delivering a fixed small cooling power $$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$2 requires a minimal dissipation

$$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$3

This is a particularly clear example of a recurrent theme: zero dissipation is compatible with useful operation only in a singular limit, and finite power requires controlled departure from that limit (Entin-Wohlman et al., 2013).

Nonlinear transport generalizes Onsager’s principle by allowing the transport coefficients to depend on the forces: $$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$4 The dissipation functional is written as

$$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$5

and the thermodynamic forces are decomposed into a boundary-fixed subspace and its metric-orthogonal complement. The stationary nonequilibrium state then minimizes the dissipation functional only with respect to variations in the free subspace. This is not the full linear Onsager extremum principle, but a restricted one adapted to locally equilibrated systems with nonlinear transport coefficients (Sonnino et al., 2015).

5. Continuum-mechanical analogues: viscous, granular, and collective systems

In incompressible Stokes flow, Helmholtz’s dissipation theorem states that the physical velocity field minimizes viscous dissipation among admissible incompressible fields that satisfy the imposed velocity boundary conditions. With

$$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$6

the Stokes solution $$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$7 satisfies

$$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$8

for every divergence-free $$P=\langle \dot Q\rangle =\lim_{\tau_{\rm obs}\to\infty}\Sigma(\tau_{\rm obs})/\tau_{\rm obs}.$$9 with the same prescribed boundary velocity. For mixed velocity–traction conditions, the relevant functional is the excess dissipation

$Q\ge T|\Delta S|=k_B T\ln 2$00

and the Stokes solution minimizes $Q\ge T|\Delta S|=k_B T\ln 2$01. This extension supports low-dimensional trial-function approximations for conductance in prismatic channels and establishes comparison principles such as the monotonic increase of conductance when additional slip boundary conditions are introduced (Ruangkriengsin et al., 2022).

Granular mechanics provides a different steady-state use of the same variational logic. In quasistatic true biaxial tests with idealized infinitely thin shear bands, each band element dissipates power

$Q\ge T|\Delta S|=k_B T\ln 2$02

and the admissible shear-band structure $Q\ge T|\Delta S|=k_B T\ln 2$03 is selected by minimizing

$Q\ge T|\Delta S|=k_B T\ln 2$04

For a single straight band at angle $Q\ge T|\Delta S|=k_B T\ln 2$05, stationarity yields the classical Mohr–Coulomb angle

$Q\ge T|\Delta S|=k_B T\ln 2$06

and the associated stress ratio is

$Q\ge T|\Delta S|=k_B T\ln 2$07

With wall friction, the degeneracy of admissible optimal patterns is reduced and an X-shaped arrangement is favored over a wide range of parameters (Stegmann et al., 2011).

Far-from-equilibrium stochastic oscillator networks introduce yet another variant. Near the synchronisation transition of driven $Q\ge T|\Delta S|=k_B T\ln 2$08-state Potts models, the stability–dissipation relation is

$Q\ge T|\Delta S|=k_B T\ln 2$09

where $Q\ge T|\Delta S|=k_B T\ln 2$10 is the change in entropy-production rate per oscillator and $Q\ge T|\Delta S|=k_B T\ln 2$11 is the change in phase-space contraction rate. For large but finite $Q\ge T|\Delta S|=k_B T\ln 2$12, the argument is that escape times scale exponentially in $Q\ge T|\Delta S|=k_B T\ln 2$13, so the state with largest $Q\ge T|\Delta S|=k_B T\ln 2$14 is visited longest; by the stability–dissipation relation, that state has the smallest dissipation. The resulting minimum-dissipation principle is therefore not an equilibrium theorem but a selection principle among nearby non-equilibrium attractors (Meibohm et al., 2024).

These continuum and collective examples show that “minimal dissipation” can refer to a true minimum principle for fields, a variational approximation scheme, or a dynamical state-selection mechanism. The common structure is the presence of admissibility constraints and a dissipation functional whose stationary point coincides with the physically realized solution.

6. Scope, extensions, and recurrent misconceptions

A recurring misconception is that logical reversibility by itself guarantees negligible power loss. The electronics literature explicitly rejects this: a standard CMOS inverter can implement a logically reversible transformation yet still dissipate $Q\ge T|\Delta S|=k_B T\ln 2$15 on each charge–discharge event. Near-minimal dissipation requires both logically reversible gates and thermodynamically reversible switching, typically via adiabatic, slowly varying power clocks (Singla et al., 2012).

A second misconception is that “minimal dissipation” implies zero dissipation at finite speed. Finite-time stochastic thermodynamics shows the opposite. Even when the quasistatic limit is dissipationless, finite duration $Q\ge T|\Delta S|=k_B T\ln 2$16 imposes a lower bound

$Q\ge T|\Delta S|=k_B T\ln 2$17

and in information erasure this adds directly to the Landauer term $Q\ge T|\Delta S|=k_B T\ln 2$18. The experimental optical-tweezer realization demonstrates that the bound can be saturated within experimental error by protocols that transport the distribution along the Wasserstein geodesic at uniform speed, but not eliminated at fixed nonzero speed (Oikawa et al., 3 Mar 2025).

A third misconception is that dissipation minimization always coincides with useful power production. Linear thermodynamics and thermoelectric theory instead show that the zero-dissipation limit is typically singular. In the strong-coupling thermoelectric limit $Q\ge T|\Delta S|=k_B T\ln 2$19, Carnot efficiency is reached but the useful cooling power vanishes. Under Onsager symmetry, the minimum-dissipation regime likewise reduces to $Q\ge T|\Delta S|=k_B T\ln 2$20. This suggests that, in many energy-conversion settings, the operational question is not how to attain zero dissipation, but how to choose the smallest admissible dissipation compatible with a prescribed throughput, error rate, or load (Proesmans et al., 2016, Entin-Wohlman et al., 2013).

Open quantum systems extend the principle beyond weak coupling. In the spin-boson model, the reduced dynamics is written as

$Q\ge T|\Delta S|=k_B T\ln 2$21

and the generator is split into Hamiltonian and dissipative parts by minimizing the Hilbert–Schmidt norm of the dissipator: $Q\ge T|\Delta S|=k_B T\ln 2$22 Stationarity $Q\ge T|\Delta S|=k_B T\ln 2$23 yields a unique dressed Hamiltonian $Q\ge T|\Delta S|=k_B T\ln 2$24, and work, heat, and entropy production are then defined with respect to that renormalized generator. In the ultra-weak coupling limit this recovers the familiar weak-coupling forms, whereas in moderate to strong coupling the method produces time-dependent renormalizations and non-Markovian corrections that alter work, heat, and entropy production in both non-adiabatic and adiabatic regimes (Gatto et al., 2024).

Taken together, these literatures indicate that the principle of minimal power dissipation is not a single theorem but a structured family of extremum principles. What remains invariant is the thermodynamic logic: dissipation is quantified by a positive functional, admissible dynamics are restricted by kinematic, logical, kinetic, or boundary constraints, and the physically relevant or optimally designed process is the one that minimizes the irreversible part of the energetic budget within those constraints.