Detailed Balance · aiwiki.page

Detailed balance is a pairwise balance condition for transitions in a stochastic process or for forward and reverse reactions in a kinetic system. For a Markov chain, it requires that the probability flow from each state to another equal the reverse flow under a specified probability distribution. It implies stationarity and characterizes reversibility of the stationary chain. In chemical kinetics, the corresponding condition requires each elementary reaction to be individually balanced by its reverse at equilibrium, rather than merely requiring unchanged concentrations. (stat.berkeley.edu)

Mathematical definition

Consider a discrete-time Markov chain on a finite or countable state space $S$ , with transition matrix $P$ . Write

P_{ij}=\Pr(X_{n+1}=j\mid X_n=i).

A normalized distribution $\pi=(\pi_i)_{i\in S}$ satisfies detailed balance with $P$ if

\boxed{\pi_iP_{ij}=\pi_jP_{ji} \qquad\text{for every }i,j\in S.}

The quantity $\pi_iP_{ij}$ is the joint probability of occupying $i$ and then moving to $j$ when the current distribution is $\pi$ . Detailed balance therefore concerns probability-weighted transitions, not equality of the transition probabilities themselves. (stat.berkeley.edu)

For a continuous-time Markov chain with transition rates $q_{ij}$ , the condition is

\pi_iq_{ij}=\pi_jq_{ji}.

For a general-state-space transition kernel $K$ , its measure-theoretic form is

\pi(dx)K(x,dy)=\pi(dy)K(y,dx),

meaning that the joint measure of successive states is symmetric under their interchange. This formulation includes continuous distributions without requiring a probability density function. (stat.berkeley.edu)

Detailed balance and stationarity

A stationary distribution satisfies the global balance equations

\pi_j=\sum_i\pi_iP_{ij}.

Detailed balance implies these equations: summing the pairwise equalities gives

\sum_i\pi_iP_{ij} =\pi_j\sum_iP_{ji} =\pi_j,

because the entries in row $j$ sum to one. The distinction is that global balance constrains total incoming probability, whereas detailed balance constrains each individual pair of flows. (stat.berkeley.edu)

The converse fails. As an illustrative calculation, consider three states arranged in a ring. From each state, let the chain move clockwise with probability $a$ , counterclockwise with probability $b$ , and remain in place with probability $1-a-b$ , where $a,b>0$ and $a+b<1$ . The uniform distribution is stationary, since every state has the same total incoming probability. Yet the clockwise current across each edge is

J=\frac{a-b}{3}.

Unless $a=b$ , the stationary chain has a persistent circulation and does not satisfy detailed balance. This example illustrates how unchanged state probabilities can coexist with unbalanced transitions. (stat.berkeley.edu)

Detailed balance alone also does not guarantee convergence from every initial distribution. For finite chains, irreducibility guarantees uniqueness of the stationary distribution, while aperiodicity is additionally needed for ordinary discrete-time convergence. A chain that alternates deterministically between two states satisfies detailed balance with the uniform distribution but remains periodic. (dpmms.cam.ac.uk)

Reversibility and time reversal

For a stationary Markov chain with $\pi_i>0$ , define the reversed transition matrix by

P^{\mathrm{rev}}_{ij} =\frac{\pi_jP_{ji}}{\pi_i}.

Detailed balance is precisely the condition $P^{\mathrm{rev}}=P$ : the stationary process obeys the same transition law when observed backward in time. (stat.berkeley.edu)

The consequence extends beyond one-step transitions. Under detailed balance, a stationary trajectory and its reversed trajectory have equal probabilities:

\pi_{i_0}\prod_{r=0}^{n-1}P_{i_ri_{r+1}} = \pi_{i_n}\prod_{r=0}^{n-1}P_{i_{r+1}i_r}.

Thus

(X_0,X_1,\ldots,X_n) \overset{d}{=} (X_n,X_{n-1},\ldots,X_0).

This is statistical reversibility of trajectories, not the thermodynamic definition of a quasistatic reversible process. The stationarity assumption matters: initializing a reversible transition mechanism away from its invariant distribution does not make its transient trajectory law time-reversal invariant. (dpmms.cam.ac.uk)

Kolmogorov’s cycle criterion

Kolmogorov’s cycle criterion tests reversibility directly from transition probabilities. For a finite irreducible chain, detailed balance holds with its stationary distribution if and only if, for every closed sequence $i_0,i_1,\ldots,i_m=i_0$ ,

\prod_{r=0}^{m-1}P_{i_ri_{r+1}} = \prod_{r=0}^{m-1}P_{i_{r+1}i_r}.

In words, the product of transition probabilities around every cycle must equal the product around the reversed cycle. The continuous-time version uses transition rates instead. (statslab.cam.ac.uk)

Necessity follows by multiplying detailed-balance equations around the cycle: all factors of $\pi$ cancel. Conversely, cycle consistency permits stationary-weight ratios to be reconstructed along paths. For example, where both transition probabilities are positive,

\frac{\pi_j}{\pi_i}=\frac{P_{ij}}{P_{ji}}.

The cycle condition ensures that different paths produce consistent ratios. On infinite state spaces, however, a reversible invariant measure obtained this way need not have finite total mass, so it need not be normalizable into a stationary probability distribution. (statslab.cam.ac.uk)

Linear-algebraic consequences

Detailed balance connects Markov-chain analysis with linear algebra. For a finite chain with strictly positive $\pi$ , let

D=\operatorname{diag}(\pi_1,\ldots,\pi_N).

The balance equations can be written

DP=P^{\mathsf T}D.

Consequently, the similar matrix

S=D^{1/2}PD^{-1/2}

is real and symmetric. Its entries satisfy

S_{ij}=\sqrt{\frac{\pi_i}{\pi_j}}P_{ij}=S_{ji}.

By the spectral theorem, $P$ has real eigenvalues and is diagonalizable. This simplifies spectral descriptions of convergence and relaxation. (stat.berkeley.edu)

Equivalently, the Markov operator is self-adjoint in the weighted inner product

\langle f,g\rangle_\pi=\sum_i\pi_i f_i g_i.

Its associated quadratic form is

\langle f,(I-P)f\rangle_\pi = \frac12\sum_{i,j}\pi_iP_{ij}(f_i-f_j)^2.

The expression measures variation of $f$ across transitions and underlies variational estimates for eigenvalues and convergence rates. Reversibility does not make all eigenvalues nonnegative: negative eigenvalues can produce alternating discrete-time behavior. (stat.berkeley.edu)

Equilibrium physics and chemical kinetics

In statistical mechanics, detailed balance provides a kinetic expression of equilibrium for many stochastic models. Suppose states have energies $E_i$ and equilibrium weights given by the Boltzmann distribution,

\pi_i=Z^{-1}e^{-\beta E_i}, \qquad \beta=\frac{1}{k_{\mathrm B}T},

where $Z$ is the partition function, $k_{\mathrm B}$ the Boltzmann constant, and $T$ the temperature. Detailed balance then requires

\frac{q_{ij}}{q_{ji}} =e^{-\beta(E_j-E_i)}.

This relation fixes rate ratios, not absolute rates: different dynamics can share the same equilibrium distribution while relaxing at different speeds. (arxiv.org)

For a reversible chemical reaction obeying mass-action kinetics,

\sum_s\alpha_{rs}A_s \rightleftharpoons \sum_s\beta_{rs}A_s,

detailed balance requires a positive equilibrium concentration vector $c^*$ such that

k_r^+\prod_s(c_s^*)^{\alpha_{rs}} = k_r^-\prod_s(c_s^*)^{\beta_{rs}}

for every reaction $r$ . This is stronger than merely requiring zero net concentration change. It also constrains rate constants through Wegscheider identities associated with dependencies among reaction stoichiometric vectors. For a simple cycle of first-order reactions, the product of forward rate constants must equal the product of reverse rate constants. (link.springer.com)

The chemical formulation dates to Rudolf Wegscheider’s work in 1901. Detailed balance is closely associated with microscopic reversibility, although deriving it for a coarse-grained kinetic model requires appropriate assumptions about equilibrium and how transitions are represented. (sites.math.duke.edu)

Monte Carlo sampling

Detailed balance is a standard construction principle in Markov chain Monte Carlo: choose transitions satisfying balance with a desired target distribution, thereby ensuring that the target is invariant. It is sufficient for invariance, but not necessary. (stats.ox.ac.uk)

In the Metropolis–Hastings algorithm, a candidate $y$ is proposed from $q(y\mid x)$ and accepted with probability

\alpha(x,y) = \min\left\{ 1,\frac{\pi(y)q(x\mid y)} {\pi(x)q(y\mid x)} \right\},

on the relevant positive support. For distinct states,

\pi(x)q(y\mid x)\alpha(x,y) = \min\{\pi(x)q(y\mid x),\,\pi(y)q(x\mid y)\},

which is symmetric in $x$ and $y$ . Rejected proposals leave the state unchanged and complete the transition kernel. (stats.ox.ac.uk)

An unknown multiplicative normalization constant in $\pi$ cancels from the acceptance ratio. This makes the method useful for sampling complicated posterior distributions and equilibrium models. Nevertheless, detailed balance does not ensure efficient exploration: proposal design and correlations between samples remain important. Nonreversible samplers can preserve the same target distribution and, in some settings, converge faster than reversible alternatives. (arxiv.org)

Limits and related notions

Local detailed balance is distinct from stationary detailed balance. In stochastic thermodynamic models, local detailed balance relates the logarithm of a forward-to-reverse transition-rate ratio to entropy transferred to the environment:

\log\frac{q_{ij}}{q_{ji}} = \frac{\Delta s_{\mathrm{env}}(i\to j)}{k_{\mathrm B}},

with suitable definitions of the transition channel and reversal. Systems coupled to multiple reservoirs can satisfy such relations while sustaining currents and failing stationary detailed balance. Thus “local” does not mean that all stationary pairwise currents vanish. (arxiv.org)

Physical time reversal also requires care when states contain quantities that change sign, such as momentum. Reversing such a trajectory involves transforming the state variables as well as reversing their temporal order. Applying the ordinary unchanged-state balance test indiscriminately to these systems can misidentify equilibrium dynamics as irreversible. Generalized balance conditions must incorporate the appropriate time-reversal transformation. (doi.org)

References

1 Introduction — Reversible Markov Chains and Random Walks on Graphsstat.berkeley.edu
Mathematical Aspects of Mixing Times in Markov Chainsstat.berkeley.edu
Modeling of Chemical Reaction Systems with Detailed Balance Using Gradient Structureslink.springer.com
Revisit of Macroscopic Dynamics for Some Non-equilibrium Chemical Reactions from a Hamiltonian Viewpointsites.math.duke.edu
How to Impose Microscopic Reversibility in Complex Reaction Mechanismspmc.ncbi.nlm.nih.gov
The Metropolis-Hastings algorithmarxiv.org
Acceleration of Convergence to Equilibrium in Markov Chains by Breaking Detailed Balancelink.springer.com
Local detailed balancearxiv.org