Four frameworks that normally describe unrelated physics, meeting in one molecule
A protein is a chain of amino acids that folds into a specific three-dimensional shape. The shape determines the function: hemoglobin carries oxygen because of how it folds, not because of which amino acids it contains. Understanding how a linear chain reliably finds its correct shape out of an astronomically large number of possible configurations is one of biology’s deepest problems.
What makes protein folding extraordinary from a physics perspective is not the folding itself. It is that four independent phase-transition frameworks — each developed in a completely different area of physics — all converge in this single system. No other system in this series contains more independent transitions.
Protein folding is a first-order phase transitionAn abrupt, discontinuous jump between two states — like water boiling. The system doesn’t gradually blend between states; it snaps from one to the other. Contrast with second-order transitions, which are smooth and continuous. with cooperative behavior.2 The protein goes from unfolded to folded in a single step — not gradually. For small proteins, this transition is two-state: unfolded or folded, nothing in between. The enthalpy-entropy compensation at the folding temperature Tf produces a sharp transition, and the cooperativity (measured by the van’t Hoff enthalpyA measure of how “all-or-nothing” a transition is. If the van’t Hoff enthalpy matches the calorimetric enthalpy, the transition is perfectly two-state — like a light switch, not a dimmer. ratio) approaches a true thermodynamic critical point for larger proteins.
The interior of a folded protein is packed. Amino acid side chains fill space with a packing fractionThe fraction of space actually occupied by particles. A packing fraction of 0.55 means 55% of the volume is filled with amino acid side chains and 45% is empty. Like measuring how much of a jar is marbles vs. air. φc ≈ 0.55 — denser than random loose packing (~0.55) but less dense than crystalline close packing (~0.74). This is not coincidental. It is a jamming transitionThe point where randomly packed particles become mechanically rigid — they can no longer rearrange. Like when a traffic jam locks up: each car is individually mobile, but collectively they’re stuck. A phase transition from fluid-like to solid-like.: the point where random packing becomes mechanically rigid.
In 2025, Grigas and colleagues published a finding in PRX Life: the jamming transition in protein cores has an anomalous critical exponentA number that describes how a physical quantity changes as a system approaches a tipping point. Different universality classes have different exponents — they’re the mathematical fingerprint of the transition type. δ = 9/2.1 This value appears nowhere else in jamming physics. Standard frictionless spheres give δ = 2. Standard frictional particles give δ = 3/2. The protein value δ = 9/2 defines a new universality class — one that appears to be specific to the geometry and interaction potential of amino acid side chains.
And this exponent appears to be universal across all protein fold classes.
How long a protein takes to fold is not random. Folding time distributions follow power lawsA mathematical relationship where one quantity varies as a fixed power of another — like y = x³. Power-law distributions have no characteristic scale: small events are common, large events are rare, and the ratio between them is constant. Exponential distributions have a typical size; power laws don’t., not exponentials. This is a signature of self-organized criticalityA system that drives itself to a critical state without anyone tuning the parameters. Think of a sand pile: you keep adding grains, and it automatically maintains itself at the angle where avalanches of all sizes occur. No thermostat needed — it just happens. (SOC) — the system organizes itself to a critical state without fine-tuning.
Three kinetic universality classes have been identified, each with its own power-law exponent, corresponding to three distinct folding mechanisms (two-state, multi-state, and downhill folding). The power-law structure means that folding is scale-free: there is no characteristic time scale. Fast folders and slow folders follow the same distribution, just at different scales.
Not all proteins fold. About 30% of the human proteome consists of intrinsically disordered proteinsProteins that never settle into a fixed 3D shape. They exist as a constantly shifting ensemble of conformations — like a piece of wet spaghetti rather than a rigid tool. They function through flexibility, not structure. (IDPs)3 — sequences that never adopt a fixed three-dimensional structure. IDPs are not broken or incomplete. They function because they are disordered: flexible, adaptive, capable of binding multiple partners.
IDPs operate at the edge of chaosThe narrow boundary between two regimes: too ordered (frozen, rigid, unable to adapt) and too chaotic (random, unstable, unable to remember). Systems at this boundary can both store information and respond flexibly — the computational sweet spot. — the boundary between ordered dynamics (stable folded structure) and chaotic dynamics (random coil with no persistent interactions). This boundary is itself a critical phenomenon, producing long-range correlations along the protein chain and sensitivity to perturbation that enables the rapid conformational switching IDPs require for function.
The jamming result deserves special attention. When you compress random spheres together, they reach a packing fraction where they can no longer rearrange — the jamming point. At this point, the bulk modulusA measure of how much a material resists being uniformly compressed. High bulk modulus = hard to squeeze. It describes the material’s response to pressure applied equally from all directions. diverges as a power law with an exponent that depends on the particle interactions. For frictionless spheres: δ = 2. For frictional particles: δ = 3/2.
For amino acids in protein cores: δ = 9/2.
This value has no precedent in jamming physics. It suggests that the effective interaction potential between amino acid side chains — a combination of van der Waals attraction, hydrogen bonding, and the hydrophobic effectThe tendency of nonpolar molecules to cluster together in water — not because they attract each other, but because water molecules prefer bonding with each other. It’s the main force driving protein folding: oily side chains get pushed to the interior. — produces a fundamentally different type of mechanical rigidity than any previously studied granular system.
More remarkably, this exponent may be universal across all fold classes.◆ our resultAn original, not-yet-peer-reviewed result from this research program — distinct from established consensus. The δ = 9/2 exponent itself is from Grigas et al.; the cross-fold-class universality test below is ours. A preliminary test on 30 protein structures across all major fold classes (all-alpha, all-beta, alpha/beta, alpha+beta) shows no statistically significant variation in packing fraction: Kruskal-WallisA statistical test that checks whether several groups come from the same distribution, without assuming the data is bell-shaped. A high p-value (like 0.31) means no significant difference between groups — they look statistically identical. p = 0.31 with a mean φ = 0.590 ± 0.042 (N = 30 — too small to be definitive). The fold class — the overall architecture of the protein — does not change the packing. The jamming transition is the same regardless of how the protein is arranged.
Click a fold class to see its measured packing fractions. If the jamming transition is universal, all classes should cluster near φc ≈ 0.55.
This is where protein folding connects to the glass transition. Below the Kauzmann temperatureThe temperature at which a supercooled liquid’s extra entropy over the crystal would vanish — a lower bound on where the glass transition must intervene. Named after Walter Kauzmann. TK, the glass undergoes a Gardner transitionA transition within the glass phase where each metastable state shatters into a fractal hierarchy of sub-states. Named after Elizabeth Gardner (1957–1988). It connects glass physics to the physics of tightly packed particles.4 — each metastable state fractures into a fractal hierarchy of marginally stableBalanced on the edge of instability. The smallest push can trigger a rearrangement. Not unstable (falling apart), not stable (firmly locked) — right at the boundary. Like a stack of blocks that stands but wobbles at the slightest touch. sub-states. The Gardner transition produces the same scaling laws that govern jamming.
The chain runs:
glass bicriticality → Gardner transition → jamming scalings → protein packing fraction
Each arrow represents a known mathematical connection, not an analogy. The Gardner transition is a consequence of glass physics. The jamming scalings are a consequence of the Gardner transition. The protein packing fraction sits at a jamming point. The chain connects a condensed-matter phase transition to the geometry of living molecules through three steps, each grounded in established physics.
If glass bicriticality is real (Tutorial 01), then the Gardner transition is where the two glass transitions most strongly couple. And the jamming that emerges from that coupling is the same physics that packs protein cores. The most basic feature of protein structure — how tightly the amino acids are packed — may be a downstream consequence of a codimension-two pointA point in parameter space you can only reach by tuning two independent knobs at once — where two critical lines cross. Rarer and more constrained than an ordinary critical point. in glass parameter space.
At the jamming transition, the bulk modulus K scales as K ~ (φ − φc)δ−2 and the shear modulusA measure of how much a material resists being deformed sideways — pushed so the top slides relative to the bottom, like a deck of cards. Related to but distinct from bulk modulus, which resists uniform compression. G scales as G ~ (φ − φc)δ−3/2. These relations connect the macroscopic elastic response to the microscopic packing geometry.
Known universality classes for jamming:
• Frictionless spheres (Hertzian contact): δ = 2, K ~ Δφ0 (constant), G ~ Δφ1/2
• Frictional particles: δ = 3/2, with hysteresis and shear-jamming
• Protein cores (Grigas et al. PRX Life 2025): δ = 9/2, K ~ Δφ5/2, G ~ Δφ3
The protein exponent implies a much stiffer divergence of elastic moduli near the jamming point. Physically: amino acid side chains interlock more rigidly than hard spheres. The combination of directional hydrogen bonds, shape complementarity, and the hydrophobic effect creates a packing geometry with qualitatively different mechanical properties.
The connection to the Gardner transition: in infinite-dimensional mean-field theory (Parisi & Zamponi 2010), the Gardner transition produces jamming with δ = 2 (frictionless hard spheres). The protein value δ = 9/2 suggests either a different starting glass phase or a finite-dimensional correction that preserves the Gardner mechanism but changes the exponent. This is an open question.
Protein folding kinetics divide into three universality classes based on the energy landscapeAn abstract surface where each point represents a possible arrangement of all the atoms, and the height represents the system’s energy. Valleys are stable configurations; ridges are barriers between them. Folding is a journey across this landscape to the deepest valley. topology:
Two-state folders: A single barrier separates the unfolded and folded states. Folding is an all-or-none transition. Small proteins (< 100 residues) typically fold this way. The folding time distribution follows a power law with an exponent related to the barrier height.
Multi-state folders: Multiple intermediate states create a rough landscape with sequential barriers. Larger proteins fold through obligatory intermediates. The folding time distribution has a different power-law exponent reflecting the multi-step process.
Downhill folders: No barrier — the protein slides continuously from unfolded to folded. The landscape is smooth with a single funnel. Found in ultra-fast folders (< 10 μs). The folding time distribution is yet another power law.
The power-law structure across all three classes is the SOC signature: the system self-organizes to a critical point in the energy landscape, producing scale-free behavior regardless of the specific folding mechanism. This is not fine-tuned — it emerges from the evolutionary optimization of the sequence to its landscape.
The chain connecting glass physics to protein packing runs through three established results:
Step 1: Glass → Gardner. Below TK, the Random First-Order Transition (RFOT) framework predicts that each glass state undergoes a Gardner transition: the basin of attraction fractures into a fractal hierarchy of sub-basins separated by vanishing barriers. This is a replica-symmetry-breaking transition of the Parisi type. Confirmed in mean-field theory; finite-dimensional simulations show consistent signatures.
Step 2: Gardner → Jamming. Recent results (arXiv 2410.23797) demonstrate that the Gardner transition produces the same scaling exponents that govern jamming. The marginally stable states of the Gardner phase sit at the edge of mechanical stability — exactly the jamming point. The connection is not analogical; the Gardner free energy landscape contains the jamming critical point as a specific limit.
Step 3: Jamming → Protein. Protein cores pack at φc ≈ 0.55, which is the random loose packing / jamming onset density for polydisperse spheres. The anomalous exponent δ = 9/2 indicates a specific type of jamming, but the packing fraction itself is the universal value. The protein “solves” its packing problem by evolving sequences that jam at the universal density.
The open question: does the bicritical point in glass (where TMCT = TK) produce a qualitatively different Gardner transition than the standard one? If so, the protein jamming exponent δ = 9/2 might be specific to a bicritical Gardner phase rather than the standard one.
Claim: The jamming transition in protein cores is universal across all fold classes. The packing fraction φc and the anomalous exponent δ = 9/2 are independent of the protein’s overall architecture (all-alpha, all-beta, alpha/beta, alpha+beta).
Status: Preliminary positive. N = 30 protein structures across all major fold classes. Mean packing fraction ⟨φ⟩ = 0.590 ± 0.042. Kruskal-Wallis test across fold classes: p = 0.31 — no significant fold-class dependence. Direction is consistent with universality, but the sample is too small for a definitive claim.
Next step: Expand to N ≥ 200 structures with balanced representation across SCOP/CATH fold classes. If the Kruskal-Wallis test remains non-significant at N = 200, universality is strongly supported. If fold-class effects emerge, the universality claim is weakened but the exponent itself (δ = 9/2) may still be universal within classes.
Claim: If the Gardner-jamming-protein chain is real, then the protein jamming exponent should be derivable from the glass bicritical framework. Specifically: the exponent δ = 9/2 should emerge from a Gardner transition near the bicritical point (strong-glass limit), not from the standard Gardner transition far from bicriticality.
Status: Planned. This requires computing the jamming exponent from a Gardner transition in a binary Lennard-Jones glass at varying distances from the bicritical point (varying density). If δ depends on distance from bicriticality, the chain is confirmed and the protein exponent is explained. If δ is always 2 regardless of glass type, the Gardner-protein connection exists but the bicritical structure is irrelevant.
Connected through the Gardner/jamming chain. The glass bicritical point (where TMCT = TK) sits at the top of a causal chain: bicriticality → Gardner transition → jamming scalings → protein packing fraction. The anomalous exponent δ = 9/2 in protein cores may be a downstream consequence of the specific type of Gardner transition that occurs near the bicritical point. Two systems, two millennia of independent study, one chain of mathematical connections.
Four order parameters on nested node sets. The thermodynamic transition operates on the full configurational space (all atomic positions). The jamming transition operates on the core residue contacts (a subset). The SOC kinetics operate on the folding trajectory (a time series over the configurational space). The edge-of-chaos dynamics operate on the disordered regions (another subset).
These are not four independent measurements of the same system. They are four transitions on different projections of the same high-dimensional space, with different order parameters, different symmetries, and different universality classes. The fact that they converge in one molecule suggests a constraint — the protein sequence has been optimized to sit at the intersection of four critical manifolds simultaneously.
The question: If relational gradient operators are applied to all four order parameters on their respective node sets, does the cross-component circulation detect a hierarchy? Specifically: is the thermodynamic transition the “master” that constrains where the other three can occur, or are they genuinely independent with an emergent coupling? The coupling structure — which transitions constrain which — determines whether multi-critical detection requires four independent chains or a single coupled chain.
Grigas, A. T. et al. (2025). Anomalous jamming exponent in protein cores. PRX Life.
Parisi, G. & Zamponi, F. (2010). Mean-field theory of hard sphere glasses and jamming. Rev. Mod. Phys., 82, 789–845.
Charbonneau, P. et al. (2024). Gardner transition and jamming scalings. arXiv:2410.23797.
Bryngelson, J. D. & Wolynes, P. G. (1987). Spin glasses and the statistical mechanics of protein folding. PNAS, 84, 7524–7528.
Uversky, V. N. (2019). Intrinsically disordered proteins and their “mysterious” (meta)physics. Front. Phys., 7, 10.
Thirumalai, D. & Klimov, D. K. (1999). Deciphering the timescales and mechanisms of protein folding using minimal off-lattice models. Curr. Opin. Struct. Biol., 9, 197–207.
Angell, C. A. (2002). Liquid fragility and the glass transition in water and aqueous solutions. Chem. Rev., 102, 2627–2650.