NOTE-708: Vekilov — Nucleation

Contribution

The paper is a perspective, built mostly on the author’s group’s own work on lysozyme and other proteins. It sets classical nucleation theory (CNT) against measured rates of homogeneous nucleation of lysozyme crystals and lists four features CNT cannot explain. It explains them with two ideas. The first is a solution–crystal spinodal: a line in the phase diagram past which the nucleus is one molecule and there is no barrier. The second is the two-step mechanism: a crystalline nucleus forms inside a pre-existing metastable cluster of dense liquid, and that second step is the slow one. It then surveys evidence that the two-step mechanism operates in other proteins, small molecules, colloids, biominerals and polymers. It draws one further conclusion: near the spinodal, heterogeneous substrates must speed nucleation by helping clusters order, not by lowering the barrier.

Key insight

Forming a crystal from a dilute solution needs two changes: the molecules must get denser and they must get ordered. CNT assumes both happen at once, in one fluctuation. The paper argues they happen in sequence. Dense disordered clusters form first, quickly and in large numbers. Ordering inside a cluster comes second, slowly, and only that step sets the rate. Because clusters fill only about 10⁻⁷–10⁻⁶ of the volume and are about 100 times more viscous than the solution, most of the solution never hosts a nucleus, and the rate is far below the classical estimate. At the high supersaturations used in practice, the barrier to the second step is negligible. What limits nucleation is then the kinetics of ordering, not a threshold.

Assumptions

• CNT’s barrier (Eqs. 1–2). A three-dimensional nucleus with a macroscopic surface free energy α; the paper takes a cube of side a, Ω = a³, and notes that other shapes change the coefficient but keep the n^{2/3} scaling (The classical nucleation theory: Thermodynamics). The paper does not discuss whether a macroscopic α holds for nuclei of 1–10 molecules, which is the size it then measures.
• CNT’s rate (Eq. 3). “A major assumption in the derivation of Eq. (3) is that the solute molecules exchange directly with the crystalline embryo” (The rate of crystal nucleation). The replacement partition function of the nucleus is set to one, which the paper says is reasonable in a viscous solution but not in the gas phase.
• The rate data are homogeneous. This rests on the measurement method of refs. 36–37, which separates homogeneous from heterogeneous nuclei, and on the rates being several orders below less careful measurements (Experimental data on the rate of nucleation of crystals).
• The CNT estimate. ν* ≈ 300 s⁻¹ is taken from step velocities on large lysozyme crystals (~1 µm s⁻¹). The paper calls it “approximate”. Z = 0.01–0.1 is assumed for any protein condensed phase, and n = 2 × 10¹⁸ cm⁻³. ΔG* ≈ 10⁻¹⁹ J is taken from the data, not from theory (The classical theory overestimates the crystal nucleation rate by 10 orders of magnitude).
• The two-step rate law (Eq. 5) assumes step one is fast and step two is rate-determining, that the clusters are in equilibrium with the solution, and that the cluster number is steady (The rate determining step in the two-step nucleation mechanism). It also assumes the concentration inside the clusters, C1, rises as temperature falls, “in agreement with the phase diagram in Fig. 6 and the likely similarity between the dense liquid in the clusters and the stable [dense] liquid” (The rate law for the two-step mechanism).
• The setting is narrower than the title. “Nucleation” here is the nucleation of crystals from solution. Melts and vapour are set aside in the Introduction. The detailed case is one protein, lysozyme, at one buffer and pH.

Key results

Classical nucleation theory (Eqs. 1–4)

• Free energy of a cubic cluster of n molecules (Eq. 1): ΔG(n) = −nΔμ + 6a²n^{2/3}α, with Δμ = μ_solute − μ_crystal > 0.
• Critical size and barrier (Eq. 2): n* = 64Ω²α³/Δμ³ and ΔG* = 32Ω²α³/Δμ² = ½ nΔμ, with Ω = a³. Both fall as supersaturation rises; n as Δμ⁻³, ΔG* as Δμ⁻².
• Rate (Volmer): J = J₀ exp(−ΔG*/k_BT). In the statistical-mechanical form (Eq. 3): J = ν* Z n exp(−ΔG*/k_BT), with ν* the attachment rate to the nucleus, Z the Zeldovich factor and n the number density of solute. (The manuscript prints the exponent as −G*/k_BT, without the Δ.) External parameters act “mostly through ΔG*”; effects on J₀ are “weaker”.
• Nucleation theorem (Eq. 4): n* − n₀ = k_BT ∂lnJ/∂Δμ + α₁, with 0 ≤ α₁ ≤ 1 and n₀ < 1. It is called “a universal, model-independent nucleation law”.

The lysozyme data (Figs. 4–5)

Homogeneous nucleation rates against σ ≡ Δμ/k_BT at 12.6 °C and NaCl at 2.5, 3 and 4 %. J rises exponentially with σ, as Eq. 3 expects, with four “peculiarities” (Experimental data on the rate of nucleation of crystals):

• (i) at 4 % NaCl, J(σ) breaks at σ = 3.1 and is “practically steady” above it;
• (ii) above σ = 3.45 the scatter grows;
• (iii) J is 0.1–1 cm⁻³ s⁻¹, “about ten orders of magnitude less than the prediction of the classical nucleation theory”;
• (iv) as T is lowered, J rises exponentially, passes “a sharp maximum” and then falls.

The solution–crystal spinodal

• Nucleus sizes from Eq. 4, read off the four linear segments of Fig. 4b, are 10, 4, 5 and 1 molecules. At 4 % NaCl the size drops from five to one at σ = 3.1, C = 33.5 mg/ml (The nucleus size and solution-crystal spinodal).
• “Nucleus size n* − n0 = 1 means that every molecule in the solution can be an embryo of the crystalline phase, and the growth to dimer and larger clusters occurs with a free energy gain.” ΔG* is then below k_BT.
• The spinodal is defined kinetically, as the transition to a nucleus of one molecule, because “a thermodynamic definition of the solution-crystal spinodal is impossible”: no mean-field free energy covers a solution and a solid.
• Past the spinodal ΔG* ≈ 0 and J does not rise with supersaturation, which explains (i). At the transition J has a sharp maximum and is very sensitive to conditions, which explains (ii).

CNT overestimates the rate by ten orders

With ν* ≈ 300 s⁻¹, Z = 0.01–0.1 and n = 2 × 10¹⁸ cm⁻³, the prefactor is 10¹⁹–10²⁰ cm⁻³ s⁻¹. ΔG* ≈ 10⁻¹⁹ J from the slopes of Fig. 4b gives, by Eq. 2 with Ω ≅ 3 × 10⁻²⁰ cm³, α = 0.5–0.6 erg cm⁻², close to other protein crystals. Eq. 3 then predicts J ≈ 10⁸–10⁹ cm⁻³ s⁻¹. Because ΔG* came from the data, “the difference between the experimentally determined J and the prediction of the classical nucleation theory is due to an overestimate of the pre-exponential factor by the classical theory.” Holds for: this one system, at the stated conditions, with an order-of-magnitude ν*.

Metastable dense liquid clusters

Seen by dynamic light scattering in solutions of lysozyme, three hemoglobin variants and lumazine synthase, and by AFM for lumazine synthase (Dense liquid clusters in the homogenous region of the phase diagram):

• they appear within seconds of preparing the solution and the population is then steady for hours;
• they occupy ϕ ≈ 10⁻⁶–10⁻³ of the volume, at 10⁵–10¹⁰ cm⁻³, and hold up to 10⁶ molecules;
• lifetimes are > 15 ms for lysozyme and Hb (a lower bound, from Γ₂ = Γ₀ + D₂q² with Γ₀ ≪ 70 s⁻¹) and ~10 s for lumazine synthase. Both are far longer than the ~10⁻⁵ s equilibration time at sub-micrometre scales, so “the compact clusters represent a metastable phase separated from the bulk dilute solution by a free energy barrier”;
• their size is not explained by short-range attraction with screened repulsion, which needs charges proteins do not carry. A cited study (ref. 67) puts it down to long-lived but unstable protein complexes.

The two-step rate law (Eqs. 5–7)

From a phenomenological theory (ref. 50):

• Eq. 5: J = k₂ C₁ T exp(−ΔG₂*/k_BT) / { η(C₁,T) [1 + (U₁/U₀) exp(ΔG°_C/k_BT)] }, with C₁ ≈ 300 mg ml⁻¹ the concentration inside the clusters, ΔG₂* the barrier to a crystal nucleus inside a cluster, η the viscosity inside the clusters, U₁ and U₀ the effective rates of cluster decay and formation, and ΔG°_C ≈ 10 k_BT the excess standard free energy of a molecule in a cluster. The bracket is about ϕ₂⁻¹.
• Eq. 6 (after Kashchiev, ref. 79), near the spinodal: ΔG₂*(T) = E*/(T_e − T)² · [1 − (T_e − T)²/(T_e − T_sp)²], which vanishes at T = T_sp.
• Eq. 7: η = η₀ {1 + [η]C₁ exp(k_η[η]C₁)} exp(−E_η/k_BT).
• Fitted with “literature values or independently determined parameters”, Eqs. 5–7 reproduce Figs. 4 and 5 “with high fidelity”. The maximum in J(T) falls at T = T_sp. The fall below T_sp comes from smaller cluster volume and higher viscosity inside the clusters. A one-step mechanism would give a fall of only ~16 % over the 5–6 K probed. The paper calls the steep fall “a crucial part in the proof of the validity of the two-step mechanism”.
• Comparing Eq. 5 with Eq. 3, k₂ϕ₂C₁T/η replaces ν*Zn. With ϕ₂ ≈ 10⁻⁷–10⁻⁶ and η ≈ 100 cP, “the nucleation rate should be ~ 10⁹× lower than the prediction of the classical theory”. That accounts for about nine of the ten orders.

The rate-determining step

• Step one is fast: clusters appear within seconds. Stable dense-liquid droplets nucleate at ~10⁸ cm⁻³ s⁻¹, “about ten orders of magnitude faster” than crystals, and the paper uses that as a proxy for the unstable clusters (The rate determining step in the two-step nucleation mechanism).
• So the nucleation theorem applied to the macroscopic J measures the crystalline nucleus inside the clusters, and the spinodal corresponds to ΔG₂* = 0.
• The two-step path is chosen because it is faster. The surface free energy between crystal and solution “is likely significantly higher” than between crystal and dense liquid, so direct nucleation would be slower.

Heterogeneous substrates

Of the four parameters J₀₁, J₀₂, ΔG₁*, ΔG₂*, the answer lies between J₀₂ and ΔG₂*, and since ΔG₂* is very small near the spinodal, “the most important parameter is J02” (The role of heterogeneous nucleation substrates). Hence, “in contrast to the generally accepted viewpoint heterogeneous nucleation centers assist nucleation not by lowering ΔG2*, but by assisting the growth of the ordered clusters”. Suggested mechanisms are templating, stabilizing an intermediate structure, and catalysing bond formation. Polarized-light nucleation of α- and γ-glycine is offered as structuring of the dense liquid by a field.

Other systems

The paper grades its own evidence (Other systems for which the two-step nucleation mechanism applies). Lysozyme allows “a rather confident conclusion”. For other proteins the evidence “is less direct”. Colloids are where the evidence “is the strongest”, from particle tracking by confocal microscopy. It also cites HbS polymers, amyloid (with a molten globule as the intermediate), urea and glycine (inferred from laser-induced nucleation), calcium carbonate prenucleation clusters, and one block copolymer system (theory only). For organic “oils” it says the two-step route “has not been tested”.

Claims

id	claim	strength	support
C1	In CNT, n* = 64Ω²α³/Δμ³ and ΔG* = 32Ω²α³/Δμ² for a cubic nucleus, so both fall steeply with supersaturation	strong	textbook derivation from Eq. 1 (The classical nucleation theory: Thermodynamics)
C2	Homogeneous nucleation of lysozyme crystals is about ten orders of magnitude slower than CNT predicts, and the error is in the prefactor, not the barrier	moderate	one system; ν* is an order-of-magnitude estimate and Z is assumed; ΔG* is taken from the data, so the barrier cannot be wrong by construction (The classical theory overestimates …)
C3	The lysozyme nucleus is 1–10 molecules, and at 4 % NaCl it drops from five molecules to one at σ = 3.1	moderate	nucleation theorem (Eq. 4) applied to slopes in Fig. 4b, with n₀ “roughly estimated as less than 1”
C4	A solution–crystal spinodal exists, past which there is no barrier, and it explains the plateau, the scatter and the maximum in J	moderate	the spinodal is defined kinetically by C3; the explanation of the scatter and the maximum leans on refs. 34, 50
C5	Metastable clusters of dense liquid, long-lived and mesoscopic, exist in homogeneous protein solutions	moderate	DLS for five proteins, AFM for one; networks excluded by a viscosity argument; the lysozyme lifetime is a lower bound
C6	Lysozyme crystals nucleate inside those clusters	moderate	inferred, not observed: direct imaging exists “only for the case of stable dense protein liquid” (Fig. 7); rests on C5 and on fitting Eqs. 5–7 to Figs. 4–5, chiefly the steep fall below T_sp
C7	The two-step mechanism accounts for ~10⁹ of the ~10¹⁰ shortfall	weak	order-of-magnitude arithmetic with ϕ₂ and η inside the clusters
C8	Cluster formation is fast and ordering inside clusters is the rate-determining step	moderate	clusters appear within seconds; stable-droplet nucleation rates used as a proxy for unstable clusters
C9	Near the spinodal, heterogeneous substrates speed nucleation through the prefactor, not by lowering the barrier	weak	inference from C4 and C8; no substrate data are presented, and it runs against the accepted view, as the paper says
C10	The two-step mechanism operates in many systems beyond lysozyme	weak to moderate, by system	the paper’s own grading: strongest for colloids, “less direct” for other proteins, untested for organic oils, theory only for the copolymer
C11	Systems take the two-step route because it is faster	weak	rests on a surface energy that “is likely significantly higher”; Lutsko and Nicolis (ref. 107) are cited in support
C12	Most industrial and laboratory crystallization runs at or near the spinodal, with nuclei of a few molecules	assertion	stated in the Conclusions as “more straightforward”; no data beyond lysozyme

Concepts

• Nucleus: a cluster of size n*, with “equal probabilities of growth and decay”. “[B]y this definition all nuclei are critical and the term ‘critical nuclei’ is redundant” (Thermodynamics). A nucleus is the point of even odds, not a guaranteed crystal.
• Supersaturation: σ ≡ Δμ/k_BT, the excess chemical potential of the solute over the crystal, in units of k_BT.
• Order parameters: concentration (dilute solution against dense liquid) and structure (dense liquid against crystal). The dilute solution and the crystal differ in both. Crystallization from solution is a transition along both.
• Dense liquid: a liquid phase much denser in solute than the solution. It can be stable, forming macroscopic droplets, or unstable, in which case it exists only as metastable mesoscopic clusters. The paper says the two have distinct causes and occur independently (Conclusions and perspectives).
• Two-step mechanism: a dense liquid cluster forms (step one); a crystalline nucleus forms inside it (step two). Fig. 3c gives the two barriers, ΔG₁* and ΔG₂*.
• Solution–crystal spinodal: the line at which n* − n₀ = 1 and the barrier vanishes. It is kinetic by definition, not thermodynamic.
• Spinodal regime: past that line, where “the rate of generation of the new phase is only limited by the kinetics of growth of its clusters”.
• Nucleation theorem: Eq. 4, which gives the nucleus size from the slope of ln J against Δμ without a model of the nucleus.
• Zeldovich factor: Z, which accounts for the width of ΔG(n) around its maximum.

Connections

The thermodynamics is Gibbs’s, adapted to a cubic crystal; the rate law is Volmer’s, in the form of Kashchiev’s Nucleation (2000, ref. 18). The two-step idea was “first suggested by simulations and analytical theory”: ten Wolde and Frenkel 1997, Talanquer and Oxtoby 1998, Soga et al. 1999 (refs. 27–29). The paper reports that this theory predicted the density and structure fluctuations “are only separated near the critical point for liquid-liquid (L-L) separation”, and coincide away from it. The experiments the paper reviews are claimed to show two steps “under a broad range of conditions”. So the paper’s claim goes beyond the simulations it credits. The paper does not say the author proposed the mechanism.

The cluster and rate-law results come from the author’s group: Pan, Kolomeisky and Vekilov 2005 for Eqs. 5–7 (ref. 50); Filobelo, Galkin and Vekilov 2005 for the spinodal (ref. 81); Pan et al. 2007 and 2010 and Gliko et al. 2005 and 2007 for the clusters (refs. 66–69). The author’s own 2004 paper on a dense liquid precursor (ref. 33) is cited. The Nanoscale paper of the same year, LIT-926, is not cited, and neither is De Yoreo and Vekilov 2003, LIT-924, although De Yoreo is thanked in the acknowledgement. Lutsko and Nicolis 2006 (ref. 107) and Sear 2009 (ref. 106) are the theory offered for why the two-step route is general. The calcium carbonate prenucleation clusters of Gebauer et al. 2008 (ref. 104) are cited as a biomineral case.

The paper does not cite Ostwald’s rule of stages (LIT-920), though the two-step route, through a metastable intermediate before the stable crystal, is close to it in spirit.

Bearing on the record

The paper is physical chemistry. Nothing in it is about arguments, records or recognition. What follows is what the physics says about the analogies the record draws, not evidence for them.

• CLAIM-176 (recurrence is the nucleation threshold). The paper complicates the analogy. It does not undercut it.
  • In the physics the threshold is a size, n* = 64Ω²α³/Δμ³, and it moves with the driving force as Δμ⁻³. A fixed count of two has no counterpart. If the analogy is taken seriously, the threshold count should depend on how “supersaturated” the exchange is. CLAIM-176 already leaves “whether ‘second’ is the right number” untested, and this is a reason to expect it is not fixed.
  • At high supersaturation the threshold disappears. Past the spinodal “every molecule in the solution can be an embryo”, and the paper says most practical crystallization runs at or near that regime. Read across, a fast, prolific exchange (the high-supersaturation case of CLAIM-186) would be one where coding at first occurrence is the natural rule. That fits the claim’s own exceptions (“a decision, or a source”) but makes them the regime, not the exception. This is an inference from the analogy, not from anything the paper says.
  • A nucleus is a point of even odds of growth and decay, not a settled crystal. Coding at the threshold, on this picture, gives a thing an even chance of persisting, no more.
  • The claim’s second job, keeping capture from flooding the record, has a counterpart only in the Introduction. Fast nucleation gives many crystals of similar size; nucleation hindered “everywhere in the growth container but at a few selected spots” gives few crystals that “grow large before the solution is depleted of nutrient”. This is stated qualitatively, with no data.
  • The code as nucleation site: the paper argues (C9, weak) that near the spinodal a site works by helping clusters order, through templating or stabilizing an intermediate, and not by lowering a barrier. If that transfers, a code helps by giving structure to material already gathered, which is closer to CLAIM-181’s lattice reading than to a threshold reading.
• The two-step reading of LIT-926 and LIT-919 (an idea first recurs as a loose cluster; the threshold is when the cluster is recognized). The paper supports the shape of this reading and qualifies its details.
  • Separability. The LIT-926 standing asks whether the two steps are separable in the physics. The paper’s answer is: in the order-parameter plane and in rate, yes, for lysozyme. Cluster formation is ~10 orders faster than crystal nucleation, and the clusters persist for milliseconds to seconds. But the early theory separated the steps only near a liquid–liquid critical point, and the Lutsko–Nicolis result, which the paper calls “an intriguing hypothesis”, has the intermediate “just a well developed density fluctuation”, so “the two-step mechanism may in fact operate in systems where no intermediate is independently found”. The first step need not leave a lasting object.
  • What transfers: step one is fast and common; step two, ordering inside the cluster, is slow and sets the rate. That matches the reading’s picture, in which loose recurrence is common and recognizing it as one thing is the bottleneck. Most of the volume never hosts nucleation (ϕ₂ ≈ 10⁻⁷–10⁻⁶), which matches most loose recurrence never being coded.
  • What does not transfer: the ordering step is a spontaneous fluctuation inside the cluster. Nothing in the physics corresponds to an outside agent recognizing the cluster. And in the two-step picture the threshold is still a critical nucleus, ΔG₂*, inside the cluster, which near the spinodal is negligible. The rate is then set by the prefactor: how much clustered material there is (ϕ₂) and how mobile it is (η inside the cluster). Read across, the limit would be how much loosely gathered material exists and how easily it can be rearranged, not a moment of recognition.
  • One detail transfers as a caution. Denser clusters are more viscous and order more slowly, and some organic “oils” are “so viscous that no crystals can form in it”. An idea accreted into a dense tangle might be harder to order, not easier. This is a suggestion from the analogy only.
• QUESTION-037. The question’s framing says a supersaturated solution may not crystallize “because crystals need a site to form on”. The paper’s own data are homogeneous rates: crystals do form without sites, at 0.1–1 cm⁻³ s⁻¹. What holds them back is the barrier and, more so, the slow ordering kinetics. Fig. 1’s caption also shows that a supersaturated solution can yield amorphous precipitate or dense liquid droplets instead of crystals, and Fig. 6 has a gelation line; the paper does not discuss these outcomes.
• CLAIM-186 (growth rate). The paper is about nucleation and says nothing about morphological instability or dendrites. Its only growth content is that crystals growing from solution have smooth surfaces and molecules attach only at steps covering 10⁻³–10⁻² of the surface. It neither supports nor tests the claim.
• CLAIM-181. C9 is the only point of contact: substrates act by structuring, through templating, consistent with the objection’s view that a seed passes on its lattice. The paper’s evidence for C9 is inference, not measurement.
• No THEORY should be filed from this reading. No instruction for machine-learning practice; the paper does not belong in the Anthology of the SOTA.

Limitations

• One system carries the argument. Every quantitative result is for lysozyme in 0.05 M Na acetate, pH 4.5, at 2.5–4 % NaCl. The other systems are cited, not analysed, and the paper grades their evidence as weaker.
• The two-step mechanism is inferred for the common case. For metastable clusters the evidence is that clusters exist plus a fit of a phenomenological model with several fitted or imported parameters (k₂, E*, the viscosity constants, U₁/U₀). The fit’s quality is shown only in Fig. 5, which I could not see.
• The CNT estimate is order-of-magnitude. ν* is borrowed from step growth on large crystals and Z is assumed. The size of the discrepancy, and so how much of it the two-step model explains, carries that uncertainty. The two-step estimate leaves about one order unexplained, which the paper does not mention.
• CNT’s barrier formula is used on nuclei of 1–10 molecules to extract α, without comment on whether a macroscopic surface free energy applies at that size.
• The substrate conclusion (C9) is argued, not shown. No measurement of substrate action is presented.
• The spinodal is a definition. It is defined by nucleus size one, so the statement that the nucleus is one molecule past the spinodal is true by construction. The content is that this happens at a definite line, measured for one system (ref. 81).
• Slips in the manuscript. Eq. 3 prints the exponent as −G*/k_BT, without the Δ. ΔG°_C is defined twice in the text after Eq. 5, once as “the barrier for nucleation of crystals inside the clusters” (which is ΔG₂*) and once as the excess free energy of a molecule in a cluster. E* is said to be found by “fitting Eq. (7)”, but Eq. 6 is meant. The maxima in J(T) are put “in Fig. 4”, but they are in Fig. 5. The calcite free energy landscape is “in Fig. 2c”, but there is no Fig. 2c; Fig. 3c is meant. “Fig. 4c” is cited and does not exist. Eq. 7’s last factor, exp(−E_η/k_BT), would make viscosity fall as temperature falls, which contradicts the text around it; the sign is probably a slip. α is described as the surface free energy “between the dense liquid and the solution”, though Eq. 2 is applied to a crystal nucleus. I did not check whether the published version corrects these.
• Some evidence is unpublished. The point about viscous organic oils rests on “unpublished evidence from the pharmaceutical industry”.

Open questions

• Is two-step nucleation general, or limited to systems with suitable clusters? The paper says both: “not all solutions would support the existence of such clusters”, yet the mechanism “may in fact operate in systems where no intermediate is independently found”. Direct imaging of a crystalline nucleus forming inside a metastable cluster, as Fig. 7 does for a stable droplet, would settle it for one system.
• How do heterogeneous substrates act near the spinodal? C9 predicts that substrates change J₀₂ and leave ΔG₂* alone. Rates measured on substrates of varied structure, analysed with the nucleation theorem, would test it.
• What sets the size of the clusters? The paper rules out one class of explanation and cites a mechanism based on long-lived complexes (ref. 67) without settling it.
• For the record: does the threshold count depend on the pace of an exchange, as the physics would suggest? A comparison of drift in fast and slow exchanges under first- and second-occurrence rules would test the analogy, but that test belongs to CLAIM-176 and CLAIM-186, not to this paper.