Abstract
A common approach to study nucleation rates is the estimation of freeenergy barriers. This usually requires knowledge about the shape of the forming droplet, a task that becomes notoriously difficult in macromolecular setups starting with a proper definition of the cluster boundary. Here we demonstrate a shapefree determination of the free energy for temperaturedriven cluster formation in particle as well as polymer systems. Combined with rigorous results on equilibrium droplet formation, this allows for a welldefined finitesize scaling analysis of the effective interfacial free energy at a fixed density. We first verify the theoretical predictions for the formation of a liquid droplet in a supersaturated particle gas by generalizedensemble Monte Carlo simulations of a LennardJones system. Going one step further, we then generalize this approach to cluster formation in a dilute polymer solution. Our results suggest an analogy with particle condensation, when the macromolecules are interpreted as extended particles.
Introduction
The formation of equilibrium droplets from a supersaturated gas is a longstanding subject of interest, being an essential phase transition in nature^{1,2,3}. More importantly, the underlying mechanism is relevant for a multitude of nucleationlike processes from statistical mechanics to material science. These include crystallization in colloidal suspensions^{4,5} and cluster formation in protein solutions^{5,6}, as well as domain formation in socalled phasechange materials^{7,8,9} and glassy solids^{10}. It is even connected to field theory^{11} and nuclear reactions^{12}. The formal framework of freeenergy calculations is straightforward, for example, in terms of reaction coordinates in phase space, but the application in computer simulations is diverse with an ongoing demand for further methodological developments^{13}. A seminal application was the parameterfree estimate of crystal nucleation rates from equilibrium freeenergy barriers^{4}. It seems natural that the estimation of nucleation barriers becomes increasingly difficult when considering more complex systems, such as polymer or protein solutions^{5}.
The rate of nucleation R is related by classical nucleation theory^{1,3} to the freeenergy cost βΔF of a nucleus on top of the nucleation barrier:
with the inverse temperature β=1/k_{B}T and the Boltzmann constant k_{B}. The kinetic prefactor κ includes the kinetic details of the nucleation process, such as diffusion and nucleusattachment rates. The freeenergy barrier may be related to the suppression in the equilibrium probability distribution. Physically relevant barriers for liquidvapour condensation are supposed to be in the range of 20k_{B}T to 100k_{B}T (see, for example, ref. 14). The typical setup for the study of freeenergy barriers is at fixed temperature by variation of the density or degree of supersaturation or directly in the grand canonical ensemble. The barrier is then associated with the suppression in the dropletsize probability distribution^{4} and is clearly temperature dependent^{15,16}. This usually requires estimating the droplet size and interface, a task that introduces systematic uncertainties and strongly depends on the droplet definition. Instead, the freeenergy barrier may be directly related to the volume of the critical nucleus and the pressure difference^{17}, exploiting a thorough understanding of the underlying phenomenon in a clever way. In this context, the problem can be reduced to conformational phase space, knowing that canonical expectation values typically do not depend on the kinetic energy.
In the following, we address the question of how to easily obtain dependable results on nucleation barriers without invoking elaborate thermodynamic reasoning or estimating nucleus shapes. This opens the door to more complex systems with nucleationlike mechanisms such as selfassembly and aggregation, where the nucleus shapes are a priori unknown. Importantly, we consider a setup at fixed density with varying temperature—an intuitive approach from a condensed matter perspective. We focus on aggregation of polymers in a dilute setup^{18,19,20} guided by the canonical case of droplet formation in a particle gas. For the canonical case, we analyse a twodimensional freeenergy landscape and identify the energy as a suitable reaction coordinate. This allows us to formulate the problem in the microcanonical ensemble of either fixed total energy E or fixed potential energy E_{p}. The first is the usual textbook definition, while the latter has been frequently applied in recent computer simulation studies. This enables us to directly discuss the effect of kinetic energy when changing between the two formulations E↔E_{p}. If kinetic energy matters, only the first one allows a direct physical interpretation.
Results
Droplet formation freeenergy barrier
We begin with the paradigm of nucleation and dissolution, the equilibrium droplet formation in a supersaturated particle gas^{21,22,23,24,25,26,27,28,29}. The firstorder condensation–evaporation transition separates a homogeneous gas phase from an inhomogeneous phase, where a single macroscopic droplet of size N_{D} is in equilibrium with the remaining vapour^{21,22,23,24,25,26}. In fact, the probability for intermediatesized droplets was shown to vanish^{23,24}. In the vicinity of the transition, the energydominated inhomogeneous condensed phase coexists with the entropydominated homogeneous gas phase. A transition between both phases may only occur by energy variation upon nucleation or dissolution. In reality, this of course refers to the total energy E. For systems where the momentum phase space may be integrated out explicitly (see Methods section), the problem simplifies in terms of computability. We hence begin with an illustration in the potential energy formulation (denoted by a hat, for example, ) as a direct result of computer simulations before we go over to comparing both energy approaches.
Figure 1a shows the freeenergy landscape of droplet condensation–evaporation of LennardJones particles (see Methods section) at the finitesize transition temperature. We define , where Ω(E_{p}, N_{D}) is a generalization of the (conformational) density of states to the twodimensional E_{p}–N_{D} reaction coordinate space. For E_{p} fixed, resembles a parabola with a single local minimum. The resulting transition path is shown as a black line and its relative maximum along this path (around ) is an estimate of the transition freeenergy barrier. Instead, however, one may consider the projection along the droplet size^{4} or equivalently along the energy, connected to the corresponding probability distributions and .
We follow the latter approach and derive the freeenergy barrier from the suppression of transition states in the canonical energy probability distributions (see Methods section) for both reaction coordinates E and E_{p}. The probability distributions are shown in Fig. 1b,d and are clearly asymmetric, with a narrow peak for the gas phase and a broad peak for the droplet phase. Methodologically, both ensembles are analogous so that we limit in the following the notation to the case of total energy E. At the equalheight inverse temperature β_{eqh}, (E) has two peaks at E^{±} of equal height and in between a minimum at E^{0}. The resulting freeenergy barrier is βΔF=ln((E^{±})/(E^{0})). Equivalently, one may perform the analysis entirely in the microcanonical frame^{30} and consider the enclosed area by the microcanonical inverse temperature β(E) and the canonical inverse temperature (see Methods section)
shown in Fig. 1c,e. Demanding areas of equal size yields the equalarea inverse temperature β_{eqa}, which is in fact identical to β_{eqh} (ref. 31).
We notice that the energy probability distributions P_{β}(E) and are related by a convolution involving the Maxwell–Boltzmann distribution P_{MB}(x) as (E) (see Methods section). This in turn corresponds to a physical smoothing, which diminishes the ratio between maxima and minimum. As a consequence, we expect a lower barrier in the total energy formulation due to the kinetic contribution.
Finitesize scaling of freeenergy barrier
In the following, we discuss the freeenergy barrier of droplet formation in a particle gas and a dilute polymer solution as a function of system size. We here extend the notion of droplet formation to the formation of clusters or aggregates in polymeric systems. In particular, we consider linear beadspring polymers (see Methods section), each consisting of 13 monomers. The resulting polymer cluster or aggregate is coexisting with a polymer ‘gas’, see Fig. 2, a first sign for the analogy to particle droplet formation.
The freeenergy barrier is commonly assumed to be proportional to the occurring interface. Here the surface of the droplet δV_{D} separates the liquid droplet from the surrounding gas and consequently we expect βΔF=σδV_{D}, with the interface tension σ. For any nonfractal shape, the surface area is related to the droplet volume V_{D} as δV_{D}∝. Since nucleation shows no sign of critical behaviour, this is a physically valid assumption. However, at the condensation–evaporation transition V_{D} itself does not scale trivially with system size V. At a fixed temperature, general arguments exploiting the equivalence to the Ising model imply that droplet formation is triggered by insertion of particles until a single macroscopic droplet of size V_{D}∝V^{3/4} coexists with the surrounding vapour^{23,24,25,26}. This result may be translated to a fixeddensity setup using Taylor expansions, where directly at the finitesize transition temperature the analogue scaling V_{D}∝N^{3/4} was verified for both lattice and offlattice particle models^{29}.
Putting everything together and introducing an effective interfacial free energy τ_{eff} then yields to leading order βΔF∝τ_{eff}N^{1/2}. It is common for the study of interface tensions to consider logarithmic corrections^{32,33,34,35}, dating back to early field theoretic results^{11}. The physical origin are translational invariance as well as capillary waves at the interface. Altogether we use for our final scaling ansatz
where α and c are constants. This is the leadingorder exponent in equation (1). Neglecting the prefactor κ for now, we obtain from equation (3) to leading order the rate of equilibrium droplet formation as R∝. Thus for increasing system size the probability that a single macroscopic droplet forms decreases exponentially.
Figure 3a shows the freeenergy barrier for droplet formation in a particle gas as a function of system size for both reaction coordinates E and E_{p}, obtained via equation (2). Both estimates yield barriers up to about 42k_{B}T, showing at close sight an almost constant shift. Only the total energy E includes the kinetic contribution, which is here reflected in a smaller barrier, whereas the interfacial free energies are expected to be identical, . In fact, fits to equation (3) yield τ_{eff}=0.939(4) and , each for N≥N_{min}=320 with goodnessoffit parameter Q≈0.3, for reaction coordinate E and E_{p}, respectively. This remains consistent within error bars under variation of N_{min}∈[224, 1,280]. The accordance of data and fit is demonstrated in the inset. In order to test the significance of the logarithmic corrections, we considered in addition a restricted fit to βΔF=τ_{eff}N^{1/2}+c. We obtain τ_{eff}=0.973(2) and for N_{min}=768 with Q≈0.2 and Q≈0.1, respectively. However, the estimate of τ_{eff} gradually decreases with increasing N_{min}. Thus the most probable scenario remain small logarithmic corrections competing with a constant shift. The leading scaling behaviour was also observed in ref. 17 directly as a linear function of the droplet surface. The advantage of the present approach is that it avoids difficulties and uncertainties coming from the ‘correct’ identification of the cluster surface.
For the cluster formation in a dilute polymer solution shown in Fig. 3b, the situation remains qualitatively similar. Fits to equation (3) yield τ_{eff}=1.06(3) and , each for N_{min}=16 with Q≈0.2, for reaction coordinate E and E_{p}, respectively. Note that this is on the same scale as for droplet formation of particles in Fig. 3a. Compared with the particle case, the system sizes are, however, much smaller, making quantitative predictions for the polymer case less reliable. Still, the overall behaviour supports the hypothesis that cluster formation in a dilute polymer solution shows a strong analogy to droplet formation in a particle gas.
We observe that considering the kinetic contribution results in a shifted barrier. For the considered examples and relevant system sizes, the shift is about , that is, of the order . This additive contribution, while much smaller than the leading behaviour, leads to a multiplicative relation between the nucleation rates R∝e^{−βΔF}≈∝3. Neglecting the momentum phase space thus underestimates the rates. In common situations, however, the deviations between experiment and simulations are of the order of several magnitudes, such that the effect of the kinetic contribution may be considered subleading.
Finitesize scaling of transition temperature
The evaluation of the freeenergy barrier via equal areas provides us with a definition of the finitesize transition temperature. At fixed density, we showed for the condensation–evaporation transition^{29} that the transition temperature, obtained from specificheat peaks, scales as inverse power of the critical droplet radius R_{D}∝∝N^{1/4}. The same is expected for all other transition temperature definitions. Figure 4 shows the equalarea definition together with a fit including higherorder corrections of the form
for cluster formation in both particle gas and polymer solution. In the case of particle condensation, leastsquare fits for N_{min}=192 yield β_{0}=1.436(2) and each with Q≈0.5, for reaction coordinate E and E_{p}, respectively. The excellent fit results show that the empirical, yet physically motivated, higherorder corrections describe the finitesize scaling very well. In addition, the strong finitesize deviations open a possibility to study finitesize scaling in experiments on the nanoscale (for a conversion, see Methods section). The finitesize transition temperature of the largest system (N=2,048) is , which still deviates from the thermodynamic limit by (10%).
It is worth noting that typical canonical finitesize transition temperatures, for example, the peak location of the specific heat, do not depend on the kinetic contribution to the partition function. In the canonical expectation values, the kinetic prefactor simply cancels. Here, however, we observe a finitesize difference in the transition temperature depending on whether we consider the kinetic contribution or not. Of course, the thermodynamic limit coincides. This is illustrated in the inset of Fig. 4a with a finitesize scaling of the transition temperature difference . It shows a prominent power law scaling of the form , which interestingly is the same scaling as the inverse transition droplet volume. The finitesize difference arises from the convolution of an asymmetric energy probability distribution with the Maxwell–Boltzmann distribution, compare Fig. 1 and equation (8) in Methods section, which manifests in the geometric differences of the microcanonical inverse temperature and the enclosed areas in equation (2). It appears that correlations between the ensemble definitions account for compatible leadingorder scaling corrections in equation (4), which further supports this ansatz and explains the observed difference.
For polymer aggregation in Fig. 4b, fits of equation (4) yield β_{0}=0.64(2) and for N_{min}=14 (guided by the inset) each with Q≈0.8, for reaction coordinate E and E_{p}, respectively. Qualitatively, the fit ansatz describes the data already well when including the smallest system sizes. Also the finitesize ensemble deviation in the inset shows a clear N^{−3/4} trend for N≥14. Again, this is an indication for the analogy between cluster formation in polymer solutions and droplet formation in a particle gas.
Discussion
We have presented a shapefree approach to the estimation of canonical freeenergy barriers in equilibrium droplet formation. The finitesize scaling is dominated by the predicted N^{1/2} behaviour but we identified additional logarithmic corrections from precise numerical estimates. Somewhat surprisingly, the absolute freeenergy barrier is sensitive to the consideration of the kinetic contribution. It is well known that the restriction to the conformational phase space does not influence finitesize transition points determined from canonical expectation values. These are evaluated on the level of the canonical partition function. The freeenergy barrier and the associated equalheight or equalarea transition temperature, however, are determined from the energy probability distribution. Here the two formulations are related by a convolution with the Maxwell–Boltzmann distribution, which explains the finitesize differences. At the same time, the probability distributions are the integrands of the respective partition functions. In the end, this boils down to the trivial fact that equality of integrals does not imply equality of the integrands. This may become relevant once theoretical predictions and experimental measurements become precise enough. Still, a restriction to the conformational phase space retains intensive parameters in the thermodynamic limit. As a numerical advantage, considering the total energy as reaction coordinate leads to less fluctuations in the microcanonical partition function and canonical probability distribution, since the underlying convolution is a smoothing procedure of physical origin.
We provided evidence that the derived finitesize scaling of canonical droplet formation is applicable to cluster formation in dilute polymer solutions as well, despite the a priori nontrivial shape of the polymer cluster. This is a clear indication of an analogy between particle condensation and polymer aggregation. In particular, we showed that polymer clusters are in equilibrium with nonattached (free) polymers: an inhomogeneous or mixed phase of aggregate and solute polymers. This is intuitively clear when the polymers are interpreted as extended particles. The leadingorder corrections then follow from the interplay between energy minimization by forming a local cluster and entropy maximization by retaining freely movable constituents. Of course, additional corrections should follow from the explicit geometry and internal behaviour of the constituents.
It is expected that the energy remains a suitable reaction coordinate for general nucleationlike mechanisms. In this case, generalized ensemble methods may unfold their full power. Moreover, our approach at fixed density provides the possibility for experiments to perform heating–cooling studies in order to probe transition rates. The presented results for polymer aggregation suggest that this approach may be generalized to studies of protein cluster formation^{6}. In a wider scope, it may also find potential application for temperaturedriven selfassembly^{36,37}, crystallization in phasechange materials^{7,8,9} and glassy solids^{10}, dislocation nucleation^{38} or the study of surface nanobubbles and nanodroplets^{39}. Of course, experimental observations commonly include the formation of multiple clusters. Reasons for this include heterogeneities or impurities acting as nucleation seeds. We suppose that this leads to a local quasiequilibrium on the respective length scales. Here a proper combination of the canonical droplet formation with the effect of nucleation seeds^{40} seems to be a fruitful approach to a systematic understanding. With further developments, simulations may provide reliable estimates for finitesize systems and meet experiments on the nanometer scale.
Methods
Microcanonical ensembles
Recently, there has been some ambiguity with the definition of a ‘microcanonical ensemble’ in computer simulations^{41}. This is a crucial aspect relevant for physical interpretations that appears to be unwittingly softened in the past decade. The microcanonical ensemble (NVE) describes an isolated system in which the number of constituents N, the volume V and the total energy E are conserved. Here the transfer of potential energy E_{p} into kinetic energy E_{k} and vice versa is a valid and relevant mechanism, where E=E_{k}+E_{p}. The microcanonical (Boltzmann) entropy is defined as S(E)=k_{B} ln Γ(E), with the partition function , where denotes the integration over state space and over momentum space.
The other common definition is the conformational microcanonical ensemble (NVE_{p}), describing instead a system with fixed potential energy E_{p}. The conformational microcanonical entropy is , where is the density of states or the conformational microcanonical partition function. The consequences are drastic: a (physical) interpretation of energy transfer from potential to kinetic energy is no longer valid. This is natural for spin systems, where a kinetic contribution is not defined in the first place (but may be exploited for numerical purposes^{42}). On the contrary, it is particularly relevant for situations in soft condensed matter, for example, for particles and polymers, where interpretations of energy transfer become natural. However, there are good reasons for this choice: Ω(E_{p}) is a fundamental property of statistical mechanics. It encodes the full information about the conformational space and allows for identification of (structural) phase transitions^{30,31,43,44}. Furthermore, it may be exploited for reweighting techniques and flathistogram Monte Carlo methods^{18,45,46}.
The relation between Γ(E) and Ω(E_{p}) is given by a convolution with the kinetic energy contribution (see, for example, ref. 47): If momenta and positions are independent, one may separate the kinetic energy contribution , explicitly perform the momentum integration and obtain for N particles in three dimensions^{48}
where Γ() is the Gamma function. We define Ω(E_{p})=0 ∀ E_{p}<E_{p,min} in order to extend the integral over the full energy range (−∞, ∞). In this way, the total energy surface entropy S(E) appears as a (weighted) potential energy volume entropy. Notice that, since all Ω(E_{p})≥0, the classical NVE entropy increases with E and the microcanonical inverse temperature k_{B}β(E)=δS(E)/δE cannot become negative, opposed to its conformational counterpart . This may be an interesting aspect for a recent debate on the correct definition of entropy when connected with the phenomenological thermodynamic entropy, for example, in refs 49, 50, 51 and references therein.
Numerically, we determine the microcanonical inverse temperatures as follows. In the conformational microcanonical ensemble, we have direct access to an estimate of ln Ω(E_{p}) (see below) such that (E_{p}) is obtained by a numerical fivepoint derivative. In the full microcanonical ensemble, we may estimate the inverse temperature in terms of microcanonical expectation values for N independent particles
where the explicit prefactor in equation (5) cancels. Then we may express .
Canonical ensembles
The canonical ensemble is defined in terms of the partition function , where each phasespace point is weighted with the Boltzmann factor according to the total energy. Again, the kinetic part may be explicitly integrated for generic systems. Each degree of freedom contributes with a Gaussian integral, and one obtains for N particles,
where is the partition function of the conformational canonical ensemble. Both partition functions may be expressed as integrals in terms of the respective energies, namely, and . The corresponding canonical energy probability distributions are defined as P_{β}(E)=Γ(E)e^{−βE}/Z_{β} and .
We may now relate the two energy probability distributions by starting with the definition of P_{β}(E) and inserting equations 5 and 7:
We identify the latter part of the integrand as the Nparticle Maxwell–Boltzmann distribution P_{MB}(x) and may write equation (8) as a convolution .
LennardJones particles
We consider a system of LennardJones particles in a dimensionless periodic box of length L with fixed density ρ=N/L^{3}=10^{−2}. Mutual avoidance and shortrange attraction are modelled by the 12–6 LennardJones potential with =1 and σ=2^{−1/6} such that r_{min}=1. For computational efficiency, the potential is cutoff at r_{c}=2.5σ and shifted by −V_{LJ}(r_{c}). System sizes range up to N=2,048, which is competitive with stateoftheart Molecular Dynamics simulations such as welltempered metadynamics^{52}. For the chosen density ρ=10^{−2}, a system with 2,048 particles requires a box of linear dimension L′≈59 =59 × 2^{1/6} σ′. For argon, σ′≈3.4 Å such that L′≈22.5 nm is on the nanoscale. Of course, for a comparison to an experimental setup one should include both the explicit geometric constraints and the full LennardJones potential.
Beadspring polymers
The considered dilute polymer solution is modelled by N linear beadspring polymers, consisting of 13 monomers each, again in a dimensionless periodic box with monomer density ρ=13N/L^{3}=10^{−2}. Bonds are modelled between neighbouring monomers by the FENE potential V_{FENE}(r)=−(KR^{2}/2)ln[1−(r−r_{0})^{2}/R^{2}] with K=40, R=0.3 and r_{0}=0.7. Nonbonded monomers interact with the same LennardJones potential as above but with σ=r_{0} 2^{−1/6} such that r_{min}=r_{0}^{18,19,20}. The total number of monomers is 13N, which yields 3 × 13N total momentum degrees of freedom in equation (5) and successive relations. The bounded bond length [r_{0}−R, r_{0}+R] from the FENE potential formally introduces constraints on these degrees of freedom. However, for practical applications in ordinary temperature ranges this effect is negligible and reweighting to the full microcanonical and canonical ensemble is feasible^{41}.
Multicanonical Monte Carlo simulations
Parallel multicanonical Monte Carlo simulations^{53,54,55,56,57,58} allow us to efficiently sample the suppressed states, by iteratively adapting an auxiliary weight function W(E_{p}) to yield a flat histogram H(E_{p}). The final weight function is related to the density of states up to a multiplicative factor: Ω(E_{p})∝H(E_{p})/W(E_{p}). This gives direct access to microcanonical estimates and canonical expectation values at any temperature. Using equation (5), this even provides an estimate of Γ(E). Monte Carlo updates for the particle case include short and longrange particle displacements. For updates of the polymer configurations, we employed local singlemonomer shifts, bondrotation and doublebridging moves, as well as longrange polymer displacements We measure the conformational (potential) energy E_{p} and the number of particles in the largest cluster N_{D} as in ref. 29. Error bars are obtained by the Jackknife method^{59,60}.
Data availability
The data that support the findings of this study are available from the corresponding author upon request. The computer code required to generate the data as well as the analysis scripts that lead to our conclusions are available from the corresponding author upon reasonable request.
Additional information
How to cite this article: Zierenberg, J. et al. Canonical freeenergy barrier of particle and polymer cluster formation. Nat. Commun. 8, 14546 doi: 10.1038/ncomms14546 (2017).
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
 1.
Feder, J., Russell, K. C., Lothe, J. & Pound, G. M. Homogeneous nucleation and growth of droplets in vapours. Adv. Phys. 15, 111–178 (1966).
 2.
Oxtoby, D. W. Homogeneous nucleation: theory and experiment. J. Phys. Condens. Matter 4, 7627–7650 (1992).
 3.
Kashchiev, D. Nucleation: basic Theory with Applications ButterworthHeinemann (2000).
 4.
Auer, S. & Frenkel, D. Prediction of absolute crystalnucleation rate in hardsphere colloids. Nature 409, 1020–1023 (2001).
 5.
Sear, R. P. Nucleation: theory and applications to protein solutions and colloidal suspensions. J. Phys. Condens. Matter 19, 033101 (2007).
 6.
Stradner, A. et al. Equilibrium cluster formation in concentrated protein solutions and colloids. Nature 432, 492–495 (2004).
 7.
Kalb, J. A., Spaepen, F. & Wuttig, M. Kinetics of crystal nucleation in undercooled droplets of Sb and Tebased alloys used for phase change recording. J. Appl. Phys. 98, 054910 (2005).
 8.
Wuttig, M. & Yamada, N. Phasechange materials for rewriteable data storage. Nat. Mater. 6, 824–832 (2007).
 9.
Lee, T. H. & Elliott, S. R. Ab initio computer simulation of the early stages of crystallization: application to Ge_{2} Sb_{2} Te_{5} phasechange materials. Phys. Rev. Lett. 107, 145702 (2011).
 10.
Lee, B.S. et al. Observation of the role of subcritical nuclei in crystallization of a glassy solid. Science 326, 980–984 (2009).
 11.
Langer, J. S. Theory of the condensation point. Ann. Phys. NY 41, 108–157 (1967).
 12.
Goodmann, A. L., Kapusta, J. I. & Mekjian, A. Z. Liquidgas phase instabilities and droplet formation in nuclear reactions. Phys. Rev. C 30, 851–865 (1984).
 13.
Hansen, N. & van Gunsteren, W. F. Practical aspects of freeenergy calculations: a review. J. Chem. Theory Comput. 10, 2632–2647 (2014).
 14.
Schrader, M., Virnau, P. & Binder, K. Simulation of vaporliquid coexistence in finite volumes: a method to compute the surface free energy of droplets. Phys. Rev. E 79, 061104 (2009).
 15.
Hale, B. N. & Thomason, M. Scaled vaportoliquid nucleation in a LennardJones system. Phys. Rev. Lett. 105, 046101 (2010).
 16.
Tanaka, K. K., Diemand, J., Angélil, R. & Tanaka, H. Free energy of cluster formation and a new scaling relation for the nucleation rate. J. Chem. Phys. 140, 194310 (2014).
 17.
Statt, A., Virnau, P. & Binder, K. Finitesize effects on liquidsolid phase coexistence and the estimation of crystal nucleation barriers. Phys. Rev. Lett. 114, 026101 (2015).
 18.
Zierenberg, J., Marenz, M. & Janke, W. Dilute semiflexible polymers with attraction: collapse, folding and aggregation. Polymers 8, 333 (2016).
 19.
Zierenberg, J., Mueller, M., Schierz, P., Marenz, M. & Janke, W. Aggregation of thetapolymers in spherical confinement. J. Chem. Phys. 141, 114908 (2014).
 20.
Zierenberg, J. & Janke, W. From amorphous aggregates to polymer bundles: the role of stiffness on structural phases in polymer aggregation. Europhys. Lett. 109, 28002 (2015).
 21.
Binder, K. & Kalos, M. H. ‘Critical clusters’ in a supersaturated vapor: theory and Monte Carlo simulation. J. Stat. Phys. 22, 363–396 (1980).
 22.
Furukawa, H. & Binder, K. Twophase equilibria and nucleation barriers near a critical point. Phys. Rev. A 26, 556–566 (1982).
 23.
Biskup, M., Chayes, L. & Kotecký, R. On the formation/dissolution of equilibrium droplets. Europhys. Lett. 60, 21–27 (2002).
 24.
Biskup, M., Chayes, L. & Kotecký, R. Critical region for droplet formation in the twodimensional Ising model. Commun. Math. Phys. 242, 137–183 (2003).
 25.
Neuhaus, T. & Hager, J. S. 2D crystal shapes, droplet condensation, and exponential slowing down in simulations of firstorder phase transitions. J. Stat. Phys. 113, 47–83 (2003).
 26.
Binder, K. Theory of the evaporation/condensation transition of equilibrium droplets in finite volumes. Physica A 319, 99–114 (2003).
 27.
Nußbaumer, A., Bittner, E., Neuhaus, T. & Janke, W. Monte Carlo study of the evaporation/condensation transition of Ising droplets. Europhys. Lett. 75, 716–722 (2006).
 28.
Nußbaumer, A., Bittner, E. & Janke, W. Monte Carlo study of the droplet formationdissolution transition on different twodimensional lattices. Phys. Rev. E 77, 041109 (2008).
 29.
Zierenberg, J. & Janke, W. Exploring different regimes in finitesize scaling of the droplet condensationevaporation transition. Phys. Rev. E 92, 012134 (2015).
 30.
Gross, D. H. E. Microcanonical Thermodynamics World Scientific (2001).
 31.
Janke, W. Canonical versus microcanonical analysis of firstorder phase transitions. Nucl. Phys. B (Proc. Suppl.) 63 AC, 631–633 (1998).
 32.
Ryu, S. & Cai, W. Validity of classical nucleation theory for Ising models. Phys. Rev. E 81, 030601R (2010).
 33.
Nußbaumer, A., Bittner, E. & Janke, W. Freeenergy barrier at droplet condensation. Prog. Theor. Phys. Suppl. 184, 400–414 (2010).
 34.
Prestipino, S., Laio, A. & Tosatti, E. Systematic improvement of classical nucleation theory. Phys. Rev. Lett. 108, 225701 (2012).
 35.
Prestipino, S., Laio, A. & Tosatti, E. A fingerprint of surfacetension anisotropy in the freeenergy cost of nucleation. J. Chem. Phys. 140, 094501 (2014).
 36.
Jonkheijm, P., van der Schoot, P., Schenning, A. P. H. J. & Meijer, E. W. Probing the solventassisted nucleation pathway in chemical selfassembly. Science 313, 80–83 (2006).
 37.
Huisman, B. A. H., Bolhuis, P. G. & Fasolino, A. Phase transition to bundles of flexible supramolecular polymers. Phys. Rev. Lett. 100, 188301 (2008).
 38.
Ryu, S., Kang, K. & Cai, W. Entropic effect on the rate of dislocation nucleation. Proc. Natl Acad. Sci. USA 108, 5174–5178 (2011).
 39.
Lohse, D. & Zhang, X. Surface nanobubbles and nanodroplets. Rev. Mod. Phys. 87, 981–1035 (2015).
 40.
Cacciuto, A., Auer, S. & Frenkel, D. Onset of heterogeneous crystal nucleation in colloidal suspensions. Nature 428, 404–406 (2004).
 41.
Schierz, P., Zierenberg, J. & Janke, W. Molecular dynamics and Monte Carlo simulations in the microcanonical ensemble: quantitative comparison and reweighting techniques. J. Chem. Phys. 143, 134114 (2015).
 42.
MartinMayor, V. Microcanonical approach to the simulation of firstorder phase transitions. Phys. Rev. Lett. 98, 137207 (2007).
 43.
Kastner, M. & Pleimling, M. Microcanonical phase diagrams of shortrange ferromagnets. Phys. Rev. Lett. 102, 240604 (2009).
 44.
Schnabel, S., Seaton, D. T., Landau, D. P. & Bachmann, M. Microcanonical entropy inflection points: key to systematic understanding of transitions in finite systems. Phys. Rev. E 84, 011127 (2011).
 45.
Janke, W. in Order, Disorder and Criticality: advanced Problems of Phase Transition Theory Vol. 3 (ed. Holovatch, Y.) 93–166 (World Scientific, 2012).
 46.
Janke, W. & Paul, W. Thermodynamics and structure of macromolecules from flathistogram Monte Carlo simulations. Soft Matter 12, 642–657 (2016).
 47.
Labastie, P. & Whetten, R. L. Statistical thermodynamics of the cluster solidliquid transition. Phys. Rev. Lett. 65, 1567–1570 (1990).
 48.
Calvo, F., Neirotti, J. P., Freeman, D. L. & Doll, J. D. Phase changes in 38atom LennardJones clusters. II. A parallel tempering study of equilibrium and dynamic properties in the molecular dynamics and microcanonical ensembles. J. Chem. Phys. 112, 10350–10357 (2000).
 49.
Dunkel, J. & Hilbert, S. Consistent thermostatistics forbids negative absolute temperatures. Nat. Phys. 10, 67–72 (2014).
 50.
Hilbert, S., Hänggi, P. & Dunkel, J. Thermodynamic laws in isolated systems. Phys. Rev. E 90, 062116 (2014).
 51.
Swendsen, R. H. & Wang, J. S. Gibbs volume entropy is incorrect. Phys. Rev. E 92, 020103(R) (2015).
 52.
Salvalaglio, M., Tiwary, P., Maggioni, G. M., Mazzotti, M. & Parrinello, M. Overcoming timescale and finitesize limitations to compute nucleation rates from small scale well tempered metadynamics simulations. J. Chem. Phys. 145, 211925 (2016).
 53.
Berg, B. A. & Neuhaus, T. Multicanonical algorithms for first order phase transitions. Phys. Lett. B 267, 249–253 (1991).
 54.
Berg, B. A. & Neuhaus, T. Multicanonical ensemble: a new approach to simulate firstorder phase transitions. Phys. Rev. Lett. 68, 9–12 (1992).
 55.
Janke, W. Multicanonical simulation of the twodimensional 7state Potts model. Int. J. Mod. Phys. C 03, 1137–1146 (1992).
 56.
Janke, W. Multicanonical Monte Carlo simulations. Physica A 254, 164–178 (1998).
 57.
Zierenberg, J., Marenz, M. & Janke, W. Scaling properties of a parallel implementation of the multicanonical algorithm. Comput. Phys. Commun. 184, 1155–1160 (2013).
 58.
Zierenberg, J., Wiedenmann, M. & Janke, W. Application of the parallel multicanonical method to lattice gas condensation. J. Phys. Conf. Ser. 510, 012017 (2014).
 59.
Efron, B. The Jackknife, the Bootstrap and Other Resampling Plans Society for Industrial and Applied Mathematics (1982).
 60.
Young, P. Everything You Wanted to Know About Data Analysis and Fitting but Were Afraid to Ask Springer Briefs in Physics Springer International Publishing (2015).
Acknowledgements
The project was funded by the European Union, the Free State of Saxony and the Deutsche Forschungsgemeinschaft (DFG) under Grant No. JA 483/311 and Sonderforschungsbereich/Transregio SFB/TRR 102 (Project B04). We gratefully acknowledge the computing time provided by the John von Neumann Institute for Computing (NIC) on the supercomputer JURECA at Jülich Supercomputing Centre (JSC) under Grant No. HLZ24. Part of this work has been financially supported by the Leipzig Graduate School of Natural Sciences ‘BuildMoNa’ and by the DeutschFranzösische Hochschule (DFHUFA) through the Doctoral College ‘’ under Grant No. CDFA0207.
Author information
Affiliations
Institut für Theoretische Physik, Universität Leipzig, Postfach 100 920, D04009 Leipzig, Germany
 Johannes Zierenberg
 , Philipp Schierz
 & Wolfhard Janke
Authors
Search for Johannes Zierenberg in:
Search for Philipp Schierz in:
Search for Wolfhard Janke in:
Contributions
All authors contributed equally to the theory and J.Z. performed the simulations. J.Z. and W.J. analysed the data and wrote the paper. All authors discussed the results and commented on the manuscript.
Competing interests
The authors declare no competing financial interests.
Corresponding authors
Correspondence to Johannes Zierenberg or Philipp Schierz or Wolfhard Janke.
Rights and permissions
This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/
About this article
Further reading

Diagrams of States of Single FlexibleSemiflexible MultiBlock Copolymer Chains: A FlatHistogram Monte Carlo Study
Polymers (2019)

Massively parallel multicanonical simulations
Computer Physics Communications (2018)

Crystallization in melts of short, semiflexible hard polymer chains: An interplay of entropies and dimensions
Physical Review E (2018)

The StressDependent Activation Parameters for Dislocation Nucleation in Molybdenum Nanoparticles
Scientific Reports (2018)

Recyclable OilAbsorption Foams via Secondary Phase Separation
ACS Sustainable Chemistry & Engineering (2018)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.