How Science Works | How Scientists Tell Equilibrium from a Steady Flow

Imagine a molecule switching between three observable conformational states, A, B and C. At each instant it occupies one state. Over a long record, scientists estimate the fraction of time spent in each state and count the transitions between them.

The occupancy probabilities p_A, p_B and p_C describe where the molecule is likely to be. The transition rates describe how it moves. Different rate networks can have the same occupancy probabilities, so a static histogram does not uniquely reveal the dynamics.

In a continuous-time Markov model, k_AB is the rate of jumping from A to B conditional on currently being in A. It has units of inverse seconds. The probability flux from A to B is p_A k_AB, also measured per second for a normalised single-system probability distribution.

The net current on that edge is J_AB = p_A k_AB − p_B k_BA. Positive J_AB means more probability flows from A to B than back, in the ensemble or long-time stationary sense. Individual trajectories still jump stochastically in both directions.

A rate is not a probability. For a sufficiently short interval dt, k_AB dt approximates the conditional jump probability to first order. Multiplying a large rate by a long interval can exceed one and is not a valid finite-time transition probability calculation.

Contents · Next chapter

The probability of A changes according to dp_A/dt = p_B k_BA + p_C k_CA − p_A(k_AB + k_AC). The first two terms are arrivals and the last two departures. Analogous equations hold for B and C.

A stationary distribution π has all these derivatives equal to zero. This requires the sum of incoming fluxes to equal the sum of outgoing fluxes at every state. It does not require each individual incoming flux to match its paired reverse.

Detailed balance imposes π_i k_ij = π_j k_ji for every connected pair. In our state model, that means J_AB = J_BC = J_CA = 0 at stationarity. Detailed balance implies stationarity because every edge cancels separately.

The converse fails when the network contains a cycle. Current can enter A from C and leave A for B at equal rates, leaving A’s occupancy unchanged. The same can happen at B and C, producing a circulating steady flow.

For this lesson, the states are ordinary configurational labels that do not change sign under time reversal, and transitions resolve the relevant channels. Systems involving velocities, magnetic fields or other time-reversal-odd variables require the appropriate reversed variables and conditions. A visible phase-space current is not universally a dissipative current without that physical interpretation.

Contents · Previous chapter · Next chapter

In the first fictional network, set every directed rate between neighbouring states on the triangle to 4.5 per second. Thus A→B and B→A both have rate 4.5, as do B↔C and C↔A.

The stationary occupancy is π_A = π_B = π_C = 1/3. On every directed edge, probability flux is (1/3) × 4.5 = 1.5 per second. Subtracting the reverse gives zero net current on all three edges.

The system is still moving. From each state the total escape rate is 4.5 + 4.5 = 9 per second. The mean residence time is 1/9 second, approximately 0.1111 seconds. At each jump, the next state is selected with equal probability from the two available neighbours.

The total expected jump activity is Σ_i π_i Σ_j k_ij = 9 jumps per second. Equilibrium-like detailed balance therefore does not mean every forward and reverse event happens simultaneously or that a molecule stops changing state.

To interpret this as a physical equilibrium model, assume the states have equal effective free energies, interact with an equilibrium bath and have no hidden driven channels. The rate symmetry is compatible with those assumptions. A transition matrix alone cannot certify that an arbitrary physical apparatus meets them.

Contents · Previous chapter · Next chapter

In the second fictional network, choose a clockwise orientation A→B→C→A. Set each clockwise rate to 6 per second and each anticlockwise rate to 3 per second. Every state still has total escape rate 9 per second.

Uniform occupancy remains stationary. At A, incoming flux is (1/3) × 3 from B plus (1/3) × 6 from C, giving 3 per second. Outgoing flux is (1/3) × (6 + 3), also 3 per second. The other states satisfy the same balance.

But each clockwise edge now carries flux 2 per second, while its reverse carries 1 per second. The net clockwise current is therefore 1 per second on each edge. This is a stationary circulating probability current.

The two networks have identical one-third occupancies, identical total jump activity of 9 per second and identical exponential residence-time distributions with mean 1/9 second. None of those measurements alone distinguishes them. The direction of successive transitions does.

In the driven network, a jump goes clockwise with probability 6/9 = 2/3 and anticlockwise with probability 1/3. In the balanced network those probabilities are both 1/2. A labelled trajectory contains information about directed flow that its occupancy histogram and unlabelled jump times discard.

Contents · Previous chapter · Next chapter

For the three-state cycle, compare the product of forward rates with the product of reverse rates. In the balanced network the ratio is one. In the circulating network it is (6 × 6 × 6)/(3 × 3 × 3) = 8.

If detailed balance held, each rate ratio would equal the corresponding stationary probability ratio: k_AB/k_BA = π_B/π_A, and similarly around the loop. Multiplying those probability ratios makes them telescope to one. A cycle product of eight therefore rules out detailed balance for this resolved Markov network.

The logarithm of the cycle rate-product ratio is the dimensionless cycle affinity: A_cycle = ln 8 = 3 ln 2, approximately 2.07944. It quantifies the directed bias around the whole loop in this model.

There is also a trajectory interpretation. Starting from a given state and considering the next three jumps, the probability of three clockwise steps is (2/3)³ = 8/27. Three anticlockwise steps have probability (1/3)³ = 1/27. Their ratio is eight. Because all states have the same escape rate here, the waiting-time factors do not introduce an additional asymmetry for the matched paths.

A single edge can have unequal rates even at equilibrium if its states have unequal free energies and occupancies. The criterion concerns probability fluxes, or equivalently the appropriate complete cycle conditions, not a blanket demand that every pair of bare rate constants be equal.

A second balanced example makes the unequal-rate point concrete. Choose stationary occupancies A = 1/2, B = 1/3 and C = 1/6. Set A→B and A→C to 2 per second, B→A and B→C to 3, and C→A and C→B to 6. Every directed flux is then 1 per second, despite the unequal rates. The clockwise rate product is 2 × 3 × 6 = 36, exactly matching the reverse product. Less occupied states leave more rapidly in this construction, while pairwise fluxes remain balanced. This example demonstrates why comparing raw transition rates without their source occupancies can falsely suggest a nonequilibrium bias.

Contents · Previous chapter · Next chapter

Why can the clockwise direction remain preferred around a closed loop? Assigning only state free energies cannot explain it at thermal equilibrium: free-energy differences sum to zero on returning to the starting state. A nonzero cycle affinity requires additional driving in a thermodynamically consistent physical realisation.

For example, a conformational cycle could be coupled to reservoirs that maintain a chemical free-energy difference. Fuel conversion can bias transitions while the molecule returns to its initial conformation. A mechanically driven system or multiple thermal reservoirs can provide other forms of nonequilibrium forcing.

These are physical interpretations, not invented chemical reaction protocols. The numerical triangle does not identify a particular molecule or fuel. To connect rates quantitatively to heat or chemical work, the transitions and reservoir exchanges must obey an appropriate local detailed-balance relation.

For an isothermal realisation with equal state free energies and the stated rate ratios, each clockwise step can be associated with a dimensionless reservoir-driving contribution ln 2. A completed clockwise cycle has affinity 3 ln 2. The reverse cycle is still possible, but less likely.

Maintaining the reservoirs is part of the account. If a finite fuel supply is allowed to run down, the rates and occupancies may change and circulation can cease. A steady state is maintained under specified conditions; it is not necessarily a state that persists after every external supply is removed.

Contents · Previous chapter · Next chapter

For the resolved stationary Markov network, the standard entropy-production expression is k_B times the sum over each unordered edge once of J_ij ln[(π_i k_ij)/(π_j k_ji)], using a consistent edge orientation. The flux ratio is dimensionless, and the current supplies inverse-time units.

In our clockwise orientation, every current is 1 per second and every forward-to-reverse stationary flux ratio is 2. The total rate is therefore 3k_B ln 2 per second, approximately 2.07944 k_B per second, or 2.87098 × 10⁻²³ joules per kelvin per second.

For the stated single-temperature thermodynamic interpretation at 300 kelvin, T times this entropy-production rate is approximately 8.61294 × 10⁻²¹ watts of dissipated free-energy rate. This is per model system with the given rates, not a macroscopic power measurement.

The same result is cycle current times cycle affinity: 1 per second multiplied by ln 8, then by k_B. The edge current is a net cycle rate in this simple single-loop network; the total jump activity of 9 per second is a different quantity and must not be substituted for it.

In the balanced reference, all net currents vanish, so the stationary entropy-production rate is zero under the equilibrium interpretation. Individual transitions and energy exchanges can still occur. Zero mean production at equilibrium is not a statement that every microscopic event has zero thermodynamic contribution.

Contents · Previous chapter · Next chapter

For a long stationary trajectory, estimate occupancy from residence time in each state. Estimate each transition rate from the corresponding directed count divided by total time spent in its source state. The estimated flux π_i k_ij is then closely related to directed counts divided by total observation time.

Over a fictional 300-second stationary record, the circulating model predicts expected counts of 600 clockwise and 300 anticlockwise transitions on each edge. These are expected values, not guaranteed realised counts. Their expected difference divided by 300 seconds gives the current of 1 per second.

Finite records need uncertainty estimates that respect the trajectory structure. Counts share occupancy histories and are not automatically independent Poisson samples. Missed rapid transitions, misclassified states and delayed measurements can alter the apparent direction statistics.

If B and C are merged into one observed category, the three-state loop is no longer resolved. A two-category record may conceal circulation and can develop memory that a naïve two-state Markov model misses. Failure to detect a current in the reduced record does not establish equilibrium of the full system.

Likewise, two physically distinct transition channels between the same visible states can carry opposing hidden flows. Their combined visible flux may cancel while dissipation persists in the channels. Inferring total entropy production therefore requires a justified state and channel resolution, or a carefully bounded estimate rather than an unsupported equality.

Contents · Previous chapter · Next chapter

Schnakenberg’s original 1976 network theory developed a systematic connection between master-equation networks, steady states and thermodynamic forces and fluxes. The triangle calculation illustrates that network viewpoint in a case where every term can be checked directly.

Battle and colleagues’ original 2016 study demonstrated methods for probing broken detailed balance in active biological systems at mesoscopic scales. Its significance lies in examining directed dynamics rather than equating an apparently stable distribution with equilibrium.

A 2019 original study on quantifying dissipation using fluctuating currents investigated inference from accessible fluctuations and explicitly considered data requirements and coarse-graining. Hidden degrees of freedom can carry dissipation omitted by a reduced measurement, so observable irreversibility and complete energetic accounting should not be conflated.

In a June 2026 original optical-cavity experiment, Keijsers and colleagues measured probability currents and switching asymmetries in a coherently driven bistable system. Their results connect irreversibility to the interactions between modes in that apparatus. They do not imply that merely seeing two preferred intensity levels establishes equilibrium or nonequilibrium by itself.

These studies use different observables and physical systems. The common evidential task is to compare motion with its appropriately reversed counterpart, account for the state representation and identify the maintained driving. A static image is only one part of that argument.

Contents · Previous chapter · Next chapter

Draw the two triangles with their rate labels. Verify uniform stationarity by writing the arrival and departure sums for A. Then calculate each pairwise current. Explain how both diagrams can conserve occupancy while only one satisfies detailed balance.

Calculate the mean residence time and total activity for each network. Their equality is intentional: it forces the diagnosis to use direction-resolved transitions rather than the amount of visible motion alone. Next compare the three-step clockwise and anticlockwise path probabilities.

For the driven triangle, calculate the cycle affinity and entropy-production rate, specifying that each edge is counted once. State the additional local thermodynamic assumptions required before multiplying by temperature to interpret a dissipated power.

Finally, hide one state label and ask which conclusions remain supported. Do not promote a failure to observe circulation into proof that no hidden circulation exists. Record the time resolution, state classification, stationarity checks and uncertainty needed for a real inference.

Detailed balance also appears in sampling algorithms, but a computational transition rule is not automatically a physical process exchanging heat with a bath. A nonreversible sampler can correctly preserve a target distribution. For that separate mathematical task, use How Markov Chains Work rather than assigning physical dissipation to an algorithm without a model of its implementation.

Return to Physical Chemistry and the How Science Works roadmap. Equilibrium is a statement about balanced microscopic dynamics under physical conditions, not simply a reading that has stopped changing.

Contents · Previous chapter · Continue to Physical Chemistry