… further help for intuition

statistical mechanics

antibody binding

Boltzmann distributtion

spin glass

Random Energy Model

An even deeper intuition about the Boltzmann distribution relation to the underlying distribution of states and what it really means in the case of antibodies.

Published

June 21, 2026

Modified

June 21, 2026

An even deeper intuition about the Boltzmann distribution relation to the underlying distribution of states and what it really means in the case of antibodies.

The Gas: Trivial Preexisting Distribution

Start from scratch. The Boltzmann distribution for a gas does not require that the system is “random” or “unselected.” It requires only that the system is in thermal equilibrium with a heat bath at temperature \(T\). The exponential form ¹ ²

\[ p_i \propto e^{-E_i / k_B T} \]

falls out from one universal principle: the heat bath can explore its own microstates freely, and the most probable joint state is the one where the combined entropy (system + bath) is maximized. That is all. Selection — meaning that certain microstates are systematically favored by history — is not forbidden. It just changes which states exist, not the Boltzmann weighting of the ones that do. An iron crystal is “selected” into a lattice; at finite temperature its phonons still obey the Bose-Einstein / Planck distribution, which is the quantum version of the same logic.³ ⁴

So: the Boltzmann weighting of accessible states is a statement about equilibrium, not about the absence of selection. This is the first important clarification.

The Ensemble of Epitopes is Not Trivial

A real and important difference, though - in the ideal gas, the energy levels are not given from outside — they emerge from kinetic theory (particle in a box), and the density of states (how many microstates have a given energy) itself has a specific form. The distribution we observe in the gas is the product of:

\[ P(E)\,dE = \underbrace{g(E)}_{\text{density of states}} \times \underbrace{e^{-E/k_BT}}_{\text{Boltzmann weight}} \]

For a 3D ideal gas, \(g(E) \propto \sqrt{E}\), and the product gives the Maxwell-Boltzmann distribution. The shape of \(g(E)\) comes from pure geometry (phase space volume) — no external assumptions about the distribution of energies are needed. Everything follows from first principles.⁵ ⁶

For antibodies, the analogy breaks here: you do not automatically know what the density of epitope states \(g(\Delta G)\) looks like. That’s a concern, and it is legitimate. The epitope universe is not an ideal gas — it has structure given by chemistry, evolution, and the specific antibody’s binding site. The Boltzmann weighting applies, but it must be multiplied by some distribution of available epitopes, which is not derivable from first principles alone.

The Random Energy Model: Where First Principles Re-enter

Here is where the deep analogy, our intuition craves, actually lives, and it comes from spin glass theory — specifically Derrida’s Random Energy Model (REM).⁷ ⁸

The setup: suppose an antibody paratope contacts \(N\) positions on a peptide, and the contribution of each contact is an independent random variable (drawn from some distribution, e.g. Gaussian). The total binding energy is a sum of many independent random contributions:

\[ \Delta G = \sum_{k=1}^{N} \epsilon_k \]

By the Central Limit Theorem, if each \(\epsilon_k\) is drawn independently with finite variance, the total \(\Delta G\) across many random sequences (the epitope ensemble) is Gaussian-distributed:⁹

\[ p(\Delta G) \propto \exp\!\left(-\frac{(\Delta G - \mu)^2}{2\sigma^2}\right) \]

This is the first-principles analog of the energy level structure in the gas — except here, \(g(\Delta G)\) is a Gaussian density of states arising from the randomness and additivity of contacts, not from geometric phase space. It requires no knowledge of the actual epitopes — only that the contacts are many, approximately independent, and drawn from some distribution.

The Extreme-Value Distribution: What Selection Does

Now here is the key step. We are not observing the full Gaussian. We observe binding only when \(\Delta G\) is sufficiently negative (i.e., strong binding). Asking about the best binders in a Gaussian landscape of epitopes is asking about the extreme tail of a Gaussian — and the statistics of extremes of a Gaussian (or any distribution with a faster-than-polynomial tail) are governed by the Gumbel distribution, the type I extreme value distribution.¹⁰ ¹¹

So the chain of logic is:

Random contacts → Gaussian density of epitope energies (Central Limit Theorem / REM)¹²
Sampling the tail (strong binders only) → Gumbel extreme value distribution¹³
Affinity maturation / clonal selection → further deformation of this distribution toward even deeper energies¹⁴

The Boltzmann weighting \(e^{-\Delta G / k_B T}\) then operates on top of this density of states. For the observable binding probability across epitopes you get:

\[ P(\text{observe binding to epitope }j) \propto \underbrace{g(\Delta G_j)}_{\text{density of epitopes with energy }\Delta G_j} \times \underbrace{e^{-\Delta G_j / k_B T}}_{\text{Boltzmann occupancy}} \]

This is structurally identical to the gas formula, with \(g(\Delta G)\) playing the role of the phase space density of states.¹⁵ ¹⁶

Putting It All Together: The Deep Analogy Table

Concept	Ideal gas	Antibody–epitope system
“States”	Momentum/position microstates	Epitope sequences in sequence space
Density of states \(g\)	Geometric (phase space volume)	Statistical (random contact energies → Gaussian by CLT)
Weighting	\(e^{-E/k_BT}\) from heat bath equilibrium	\(e^{-\Delta G/k_BT}\) from solution equilibrium
Selection role	None (equilibrium)	Shapes \(g(\Delta G)\) via affinity maturation; Boltzmann weighting unchanged
Observable distribution	Maxwell-Boltzmann	Gumbel-modified Boltzmann (extreme-value regime)
“Temperature” meaning	Physical temperature of bath	Same physical \(T\); also a metaphor for binding promiscuity

The One-Sentence Intuition

In both cases, the exponential Boltzmann weighting is imposed by thermal equilibrium (not by randomness), and the shape of the full distribution is the product of that weighting with a density of available states — which, in the antibody case, is Gaussian by the Central Limit Theorem (many random contacts), so the observed distribution of tight binders is the extreme-value tail of a Boltzmann-weighted Gaussian.

Further reading: ¹⁷ ¹⁸ ¹⁹ ²⁰

⁂

This is a note, further explaining Specificity is a shape, not a switch and From the Boltzmann distribution to receptor–ligand kinetics.