WorksPhysics and Astronomy
Only the immortal can (almost) flock while turning: Two-dimensional chiral immortal flocks, time crystals and non-universal sub-logarithmic roughness
Abstract
We study two-dimensional chiral dry'' (non-momentum conserving) immortal'' (number conserving) flocks'' (polar ordered active fluids). Noise driven fluctuations about the noise-free time crystal'' state, in which the velocity rotates uniformly in time, obey coupled Kardar-Parisi-Zhang-like equations for the local velocity orientation and a conserved concentration field; the latter field stabilizes order. Orientational fluctuations grow only sub-logarithmically, and number fluctuations are suppressed (``hyperuniform''); both exhibit nonuniversal scaling exponents. There are three different universality classes of disordering transitions.
Flocking - the collective, coherent motion of a large number of active particles - is the canonical example of active matter [1, 2, 3, 4, 5, 6, 7]. The most common examples occur in biological systems, where they are found over an enormous range of length scales: from herds of caribou, through flocks of birds and schools of fish, down to bacteria colonies and even the biomolecular level, e.g., the mitotic spindle [8, 9]. Synthetic examples (e.g., “flocks” of quincke rotators [10]) have also been studied.
Flocking is particularly striking in two dimensions (2D), since, in order to move coherently, a flock must spontaneously break continuous rotation invariance, an impossibility in 2D equilibrium systems [11, 12]. Thus, the order in 2D flocks is an intrinsically non-equilibrium phenomenon [1, 4, 5, 6, 7].
Besides activity, chirality, i.e., the absence of left-right symmetry, is also widespread in biology. Most earlier studies[1, 2, 3, 4, 5, 6, 7] have focused only on achiral flocks, which retain left-right symmetry. Among the few exceptions are numerical studies of simple chiral modifications of the Vicsek model, which find chiral flocking states [13], and recent analytic studies [14] of non-number-conserving (“Malthusian”) flocks, in which active agents are continuously born and die. The latter study showed, by mapping these systems to the 2D KPZ equation [15], that chirality destroys long-range orientational order in chiral Malthusian flocks, leaving only short-range order, rather than either the quasi-long-range order seen in 2D equilibrium XY models [16], or the long-range order found in achiral 2D flocks.
In this Letter, we study immortal chiral dry flocks, with a fixed number of active agents moving on a frictional substrate, i.e., momentum is not conserved. Number conservation is the system’s only conservation law.
We find that dry chiral immortal flocks, while still not truly long-ranged orientationally ordered, have a phase in which they are better ordered than 2D equilibrium XY models. This phase is also “hyperuniform”; i.e., it has miniscule number fluctuations.
This “better ordered” chiral flock can be understood by comparing it to an ideal, perfectly ordered one. Because chirality causes all agents to spontaneously turn in the same direction (left or right) as they move, the ideal state [13, 17, 18] has a coarse-grained velocity field that is uniform in space but periodic in time [14]:
Here the parameter is an intrinsic frequency characteristic of the system, whose sign determines whether the agents spontaneously turn to the left () or the right (). An achiral system must have . Hence, may be used as a measure of the chirality of the system. Here, is the mean speed of the flockers, and a phase which is spontaneously chosen by the system. Since in (1) is spatially uniform at any , the state (1) has infinite-ranged orientational order. This state can be thought of as a “time crystal”; or, more precisely, as a “time cholesteric”, since the velocity rotates in time in the same way that the molecular axis rotates in space in a three-dimensional cholesteric liquid crystal [19]. Thus the active chiral flock resembles a cholesteric with the “pitch axis” being time; see Figure 1.

Figure 1. The analogy between 2D chiral flocks (left) and three-dimensional cholesteric liquid crystals (right). The single-headed arrows represent polar active particles with their velocities pointing in the directions of the arrows. The double-headed arrows represent apolar rod-like long molecules of cholesteric liquid crystals.
In the presence of noise, is no longer a constant. However, we find that there is a substantial range of parameters, which we will hereafter refer to as the “strongly quasi-ordered phase”, over which is a slowly varying function of and . The fluctuations of diverge, and, as a result, the order parameter decays as the spatial extent of the flock goes to infinity according to the law
where the coefficient , the microscopic length , and, most significantly, the exponent are all non-universal parameters of the flock. Though non-universal, is bounded below: . The fact that does diverge as for all ’s allowed by (4) implies that 2D chiral flocks do not have long-range order, but can be infinitely more ordered (for ) than 2D equilibrium XY models.
The sub-logarithmic decay of order implied by (3) for is so slow that it is not surprising that Ref. [13] found an ordered state like (1). Their system was, presumably, not big enough for them to observe the asymptotic decay of order implied by (3), since “big enough” could be very big indeed (indeed, it could be literally astronomical) if in (2) is small.
We derived the result (2) in the so-called “spin-wave” approximation [20], in which we neglect the possibility of topological defects - specifically, vortices [20] - in the velocity field.
We strongly suspect, and will present heuristic arguments to support this below, that the effects of vortices are negligible, at least for sufficiently small noise, in systems with ; that is, in chiral flocks that are better ordered than equilibrium XY models. The same arguments suggest that vortices are always important, even at arbitrarily small noise, if . If this argument is correct, then the locus in parameter space on which is a phase transition boundary between the strongly quasi-ordered phase with phase fluctuations given by (2), and a much less ordered (specifically short-range ordered) phase riddled with a proliferation of unbound vortices. It also implies that the only ’s that will be observed in practice are those with bounded both below and above:
We find that number fluctuations are also unusual in chiral flocks. Specifically, if we consider an open region of such an area as to contain, on average, active particles, then the standard deviation of the number of particles in this area scales with as
From this, we see that, if , as we suspect will always be the case, these fluctuations are always infinitely less, in the limit of mean number , than the usual ”law of large numbers” relation . We therefore follow [21, 22], who found the same phenomenon in diffusing active XY models, and refer to these by the appealing phrase ”miniscule number fluctuations” (hereafter ”MNF”), or “hyperuniform” [23].
If our conjecture that regimes with are unstable against vortex proliferation is incorrect, then regions in
which will have ”Giant number fluctuations” with being infinitely larger, in the limit of , than the usual ”law of large numbers” relation , as can be seen directly from (5).
Even if we are correct that vortices always unbind for , in systems with small noise, there could be a wide range of length scales over which giant number fluctuations could be seen. In that range of scales, “weakly quasi-ordered states” with should be visible.
Our theory also predicts the phase diagram shown in Figure 2 and Figure 3. The axes in that figure are two parameters which we call and of the hydrodynamic theory that we will present below. Since we expect the parameters of our hydrodynamic theory to be smooth continuous single-valued functions of the control parameters in experiments or simulations on flocks, the topology of the phase diagram of a chiral active flock in the space of any two control parameters should have the topology of Figure 2 and Figure 3. Furthermore, if we tune just one experimental control parameter (call it ), since we expect and to both be functions of , tuning will move the system along some locus in the plane.

Figure 2. Phase diagram of chiral flocks in the - plane for modest values of and . Our strongly quasi-ordered phase occurs in two lobes which touch at the point , where , . The large lobe actually extends out to infinity in the negative and quadrant, in which its boundary is asymptotically bounded by the straight lines and . The boundaries of these two regions are the loci of vortex unbinding phase transitions of the novel type described below, except for the portion of the boundary between the point and the point which lies at , . Along that section, the chiral flock disorders via a completely different type of continuous phase transition which we call the ”instability” transition. See the text for a more detailed description.
Figure 3. Phase diagram of chiral flocks in the – plane for large positive values of and . The ”better ordered” phase occurs within the narrow sliver inside the curve, which also extends out to infinity in the negative and quadrant, in which its boundary is asymptotically bounded by the straight lines and . The transition along this locus is a vortex unbinding transition, in the same universality class as that along the boundaries in figure (2). Again, see the text for a more detailed description.
There are three distinct types of phase transitions in our model:
An “instability” transition occurs when the system crosses the segment PC in figure (2). As the transition is approached from the disordered side, by tuning an experimental control parameter which we will call (this is the analog of tuning temperature in an equilibrium system), orientational correlations decay beyond a correlation length . This correlation length grows extraordi-
narily rapidly as the control parameter approaches its critical value :
where is a non-universal length and a non-universal dimensionless constant.
A ”diffusion constant driven” instability transition, which occurs when one tunes a third parameter: the ratio of the diffusion constant for the density of active particles to the diffusion constant for the phase . These diffusion constants are defined more precisely below.
One can think of as a third axis coming out of the plane of Figure 2 or (3). Defining the distance to the transition by analogy with that used for transitions of type (1): , we find that the associated correlation length also diverges strongly as from the disordered side; specifically as
where is a constant, non-universal length and another non-universal dimensionless constant. The exponent is also non-universal, and can vary from 0 to .
A vortex unbinding transition as passes through . Our analysis of this transition is more speculative and heuristic than the above two. Making what we believe are reasonable assumptions, outlined below and discussed in details in the associated long paper (ALP) [24], we find that the correlation length in this case diverges extremely strongly as the transition is approached from the disordered side:
where, as before, is the difference between whatever experimental parameter is being tuned to move through the phase diagram, an its value when the transition line is crossed. In addition, (here “v” stands for “vortex”) is a constant, non-universal length and another non-universal dimensionless constant.
In addition to being the meeting point of three different phase transition lines, the very busy point is also a sort of ”accumulation point” for different values of . More precisely, in both lobes, contours of constant for all values of in the allowed range meet at the point . In fact, in the lower lobe, these contours are a set of nested closed loops, starting and ending at the point , with the contours for smaller values of nested inside those for larger values. The size of these loops goes to zero linearly with the difference ; see Figure 4.
Figure 4. Contours of constant in the plane. These form two lobes of nested contour loops, with contours of smaller always lying inside those of larger . The size of the loops goes to zero as . The contours with also form the phase boundaries of the two inner lobes of our ”strongly quasi-ordered phase”.
In the upper lobe, the structure is very similar; the sole difference is that the contours do not completely close. Instead, for , they start at , and end on the locus PC (not shown in figure Figure 4), while for ’s in the narrow range they extend out to reach . For these values of , a second, disconnected branch comes in from infinity and ends on the locus PC.
In the disconnected outer sliver of strongly quasi-ordered phase shown in Figure 3, the contours of constant are a set of near hyperbolae, nested inside the boundary of that region, with the contours for smaller ’s again inside those of larger .
We derive the hydrodynamic equations for dry, immortal chiral flocks at large spatial and temporal scales. In an isotropic, homogeneous frictional medium ensuring no momentum conservation, the key slow variables are the local phase and local number density of the flockers of mean concentration . Symmetry specifically, rotation invariance and conservation law considerations imply that the hydrodynamic equations for and number fluctuations take the form
where, as usual in the spirit of hydrodynamic approaches, we have retained only the symmetry permitted terms to lowest order in fields and gradients. The parameters are the aforementioned phase and density diffusivities; are the nonlinear coupling constants in the theory, which may be individually positive or negative; and reflect the fact that the spontaneous rotation rate of the flockers can depend on the local density. The parameter reflects the fact that particle currents can be driven by spatial variations of the phase . The product proves to be the square of the ”speed of sound” of a pair of propagating coupled phase-concentration waves, which together are the hydrodynamic modes of chiral flocks. Linear stability requires , which is the region on which we focus our attention here.
In (10), is the anti-symmetric rank-2 tensor: . The – and – terms in (10), respectively, contribute to the particle current in response to a local phase difference along and perpendicular i.e., chiral to .
The additive noises and are Gaussian distributed, zero mean noises. Considering that and are, respectively, nonconserved and conserved variables, the variances of and are chosen as
where the noise strengths are phenomenological parameters of the hydrodynamic theory. See ALP [24] for more details.
In the linearized approximation, the equal-time correlations of and in Fourier space can be calculated straightforwardly. In the long wavelength limit they are given by (with , )
where the subscript “0” identifies these as linear theory results.
To go beyond the linear theory, we perform the perturbative dynamical Renormalization Group (hereafter ”RG”) on Eqs. (9) and (10) to one-loop order. It turns out that only , , , and receive relevant fluctuation-corrections at one-loop order due to the nonlinear terms. These corrections are log-divergent in 2D. However, there are no diverging fluctuation corrections to , or , , , at one-loop order. Furthermore, none of the relevant fluctuation corrections of , , , involves .
We give the details of this RG calculation in the ALP [24], where we also show that our one-loop results become asymptotically exact in the long-distance, long-time limit.
The one-loop differential RG recursion relations for , , , are
where we have defined the two dimensionless parameters
Since , , and do not renormalize at one-loop order, the ratios do not either.
We will show later that this failure to renormalize at one-loop order rather than having a universal fixed point in the - plane (which is the sort of behavior usually found under the RG), means that a portion of the - plane, in which the strongly quasi-ordered phase exists, is a fixed plane under the RG. This is analogous in the sense of being a fixed plane, rather than a single fixed point, to the well-known [20] ”fixed line” that controls the Kosterlitz-Thouless ordered phase of the equilibrium two-dimensional XY model. As mentioned above, however, the properties of the system in the strongly quasi-ordered phase controlled by this fixed plane are quite different from those of the Kosterlitz-Thouless ordered phase.
We further define the additional dimensionless ratios: , . These do renormalize. Their recursion relations can be derived from the full recursion relations (14)-(17), combined with the fact that and do not renormalize. They imply that and flow, in the region in Figure 2 and Figure 3 of the - plane occupied by the strongly quasi-ordered phase, to fixed point values and which depend only on the values of and . (The rather complicated analytic expressions for and in terms of and are given in the ALP [24].
The one-loop perturbation theory is accurate when the effective dimensionless coupling constant is small.
Once and have flown to their fixed point values and , the recursion relations (14)-(17) imply that obeys a closed recursion relation
where and are defined by the equally closed recursion relations for the mean noise and mean damping :
Equations (19) and (20) follow from (16) and (17), after using the fixed point values and defined above. Detailed (and complicated!) analytic expressions for , and are given in the ALP [24].
We focus on the case (i.e., ), because in this regime, which is a well defined and non-vanishing portion of the - plane, . This implies that our perturbation theory, which is valid for small , becomes asymptotically exact in this regime. In this regime, the solution of (19) for large RG ”time” is
Using this in (20), we find, again for large ,
where
Here, and are non-universal constants. As usual in the RG, we can define an ”effective” values and of the noise and damping at wavevector . These have the property that the true two-point equal-time Fourier transformed correlation functions , at wavevector of the full non-linear system are given by the expressions (13) obtained from the linear theory, except that everywhere the noise strength and damping appear in those linear expressions, they must be replaced by and respectively. Here with a phenomenological ultraviolet cutoff (i.e., an upper limit on ) of order the inverse of some microscopic length (e.g., the inter-flocker distance). Using this value of in (22) and the resultant and in (13), we obtain the renormalized correlations (which are the ones observed in experiments) are given by
The real space phase fluctuations in a system of linear spatial extent can be calculated by Fourier transforming (26) back to real space:
where we have defined
In (28), , and the -integration is from to . Since both and vary continuously with and , whose values vary from system to system, the exponent is not universal. Therefore, the real-space phase fluctuations (28) imply continuously varying logarithmic growth of fluctuations. This is quite an unusual feature of this problem; usually [16], scaling laws are universal for all systems of a given symmetry.
The variability of has limits, however. In the region of the - plane we are considering (which is the only regime in which equation (28) applies) is bounded below. To see this, we use our expressions (24) and (25) for and in (29), to obtain
where the last inequality follows from the fact that, in the regime of we are discussing, is positive, since and both are.
There is also an upper bound on ; specifically, . This bound holds because systems with are unstable to the unbinding of vortices - that is, topological defects in the phase field , like those that disorder the equilibrium 2D XY model at sufficiently high temperatures[20].
This bound can be understood as arising from the fact that for , the ”effective reduced temperature” of a chiral flock renormalizes to infinity, thereby passing the Kosterlitz-Thouless threshold for vortex unbinding.
To make this idea more precise, we note that, if we compare our result (28) for the fluctuations in real space with the analogous result[20]
for the equilibrium XY model (whose Goldstone mode variable is also a phase ), we see that the two are equivalent if we define an effective, length scale dependent “reduced temperature” given by
Thus we see that the effective reduced temperature diverges as if and vanishes as if . The former result implies that vortices will always unbind when , since they do[20] for sufficiently large . Thus, we can only find a strongly quasi-ordered phase, whose existence depends on the absence of vortices, for . Indeed, the locus in the - plane is the locus of a vortex unbinding transition at which the order of the strongly quasi-ordered phase disappears. More on this later.
Hence for the fluctuations in this system are always smaller than, or comparable to, those of the equilibrium XY model. Note, however, that, since our lower bound on is greater than 0, the fluctuations always diverge in the limit of system size . Hence, the system never has true long-ranged orientational order, even though it is generically better ordered than the well-known QLRO of the 2D equilibrium XY model.
Next we calculate the number fluctuation correlations in the real space and the variance of the number fluctuations in an open finite square area of size . Inverse Fourier transforming (27), we find
for large . Then, as shown in ALP [24],
which leads to our result (5) for . This means that the number fluctuations are strongly suppressed for compared to those in equilibrium systems. We call this “miniscule number fluctuations” (MNF), or “hyperuniformity”.
The recursion relations (14)-(17) above can also be used to show that there are three distinct types of phase transitions out of the strongly quasi-ordered phase in chiral flocks.
The first of these is the “instability transition”, which occurs along the segment PC in Figure 2. This transition occurs when and/or are tuned into a region in which the stable fixed point for disappears. When this happens, the system flows under renormalization into a regime of negative diffusion constants, in which the ordered state becomes dynamically unstable. In the ALP [24], we explain in detail how we obtain this locus numerically, and how we analytically obtain the associated correlation length (6).
The second, “diffusion constant driven”, transition occurs if and are in the region in which a stable fixed point for does exist, but the bare value of the parameter is on the ”wrong” side of an unstable value ; that is, on the side that flows away from both and . We show in the ALP that such a point always exists for any value of and such that a stable fixed point for does exist.
When this happens, then the system also flows into a regime in which the ordered state becomes dynamically unstable. Tuning from the “right” to the “wrong” side of therefore leads to another type of order-disorder transition. This is the transition of the type “2” described above.
The third transition, which is the vortex unbinding transition, can, as we explained earlier, be understood as being driven by the renormalization of the “effective temperature” to infinity, which occurs when the exponent increases through 1. As also discussed above, this results in the unbinding of vortices. The recursion relations (14)-(17) and their analysis up to now has ignored the possibility of vortices. Including them systematically is an extremely difficult problem, since the system is nonlinear. As a result, the fields of vortices, in contrast to those of the equilibrium XY model, do not simply add. This makes the problem much more difficult.
However, our earlier intuitive argument, based on identifying as the scale-dependent analog of the reduced temperature in the 2D XY model strongly suggests that ; vortices do unbind. Consequently, only short range order is expected. In contrast for , for , suggesting a vortex-free ordered state, at least for low noises.
To address this transition more quantitatively, in the ALP [24], we use a “kludge”: we add to our recursion relations for the chiral flock without vortices terms from the recursion relations for the equilibrium XY model with vortices. More details are given in the ALP [24]; the resultant recursion relations lead to the prediction (8) for the behavior of the correlation length . Obviously, that result is somewhat speculative, in contrast to our results for the other two transitions, of which we are quite confident.
In summary, we have shown that chirality radically changes flocking in two dimensions. It reduces the long-ranged order present in achiral flocks to a type of order that is better than the quasi-long-ranged order found in the equilibrium 2DXY model, but still not long ranged. Nonetheless, this order is infinitely stronger than the short-range order found in mortal or Malthusian chiral flock [14]. This order is characterized, as illustrated in (2), by a non-universal exponent giving the scaling of the decay of orientational order with the logarithm of the system size; always in the equilibrium XY model, but here varies continuously between 2/3 and 1. The number density fluctuations are also suppressed in these nearly ordered states, giving hyperuniformity, being characterized by the same non-universal exponent . We have also found three different types of phase transitions out of this nearly ordered state, all of which belong to heretofore undiscovered universality classes, whose universal properties we have worked out. Number density fluctuations help to stabilize order in immortal chiral flocks; this mechanism is absent in Malthusian chiral flocks, which are therefore disordered [14].
A.B. and S.M. thank Alexander von Humboldt Stiftung (Germany) for partial financial support through their research group linkage programme (2024). A.B. thanks ANRF (India) for partial financial support through the ARG (MATRICS) programme (file no.: ANRF/ARGM/2025/000461/TS). S.M. thanks ANRF (India) for partial financial support through the ARG (MATRICS) programme (file no.: ANRF/ARGM/2025/000748/TS). J.T. thanks the Max-Planck Institut für Physik Komplexer Systeme, Dresden, and the Higgs Center at the University of Edinburgh, for their hospitality and support while much of this work was underway.
References
- [1]J. Toner, The Physics of Flocking (Cambridge University Press, 2024).DOI
- [2]S. Ramaswamy, The Mechanics and Statistics of Active Matter, Ann. Rev. Cond. Matt. Phys. 1, 323 (2010).
- [3]M. C. Marchetti, J. F. Joanny, S. Ramaswamy, T. B. Liverpool, J. Prost, M. Rao, and R. A. Simha, Hydrodynamics of soft active matter, Rev. Mod. Phys. 85, 1143 (2013).DOI
- [4]T. Vicsek, A. Czirók, E. Ben-Jacob, I. Cohen, and O. Shochet, Novel Type of Phase Transition in a System of Self-Driven Particles, Phys. Rev. Lett. 75, 1226 (1995).pubmed.ncbi.nlm.nih.gov/10060237
- [5]J. Toner and Y. Tu, Long-range order in a two-dimen- sional dynamical XY model: How birds fly together, Phys. Rev. Lett. 75, 4326 (1995).
- [6]J. Toner and Y. Tu, Flocks, herds, and schools: A quantitative theory of flocking, Phys. Rev. E 58, 4828 (1998).DOI
- [7]J. Toner, Y. Tu and S. Ramaswamy, Hydrodynamics and phases of flocks, Ann. Phys. 318, 170 (2010).
- [8]S. Gadde and R. Heald, Mechanisms and Molecules of the Mitotic Spindle, Current Biology 14, R797 (2004).pubmed.ncbi.nlm.nih.gov/15380094
- [9]D. Oriola, D. J. Needleman, and J. Brugues, The Physics of the Metaphase Spindle, Ann. Rev. Biophys., 47, 655 (2018).DOI
- [10]A. Bricard, J.-B. Caussin, N. Desreumaux, O. Dauchot, and D. Bartolo, Emergence of macroscopic directed motion in populations of motile colloids, Nature (London) 503, 95 (2013).DOI
- [11]N. D. Mermin and H. Wagner, Absence of Ferromagnetism or Antiferromagnetism in One- or Two-Dimensional Isotropic Heisenberg Models, Phys. Rev. Lett. 17, 1133 (1966).
- [12]P. C. Hohenberg, Existence of long-range order in one and two dimensions, Phys. Rev. 158, 383 (1967).DOI
- [13]Y. Wang, B. Ventéjou, H. Chaté, X. Q. Shi, Condensation and Synchronization in Aligning Chiral Active Matter, Phys. Rev. Lett. 133, 258302 (2024).DOI
- [14]L. Chen, C. F. Lee and J. Toner, Only the ambidextrous can flock: Two-dimensional chiral Malthusian flocks, time cholesterics, and the Kardar-Parisi-Zhang equation, DOI
- [15]M. Kardar, G. Parisi, and Y.-C. Zhang, Dynamic scaling of growing interfaces, Phys. Rev. Lett. 56, 889 (1986).pubmed.ncbi.nlm.nih.gov/10033312
- [16]P. M. Chaikin and T. C. Lubensky, Principles of Condensed Matter Physics (Cambridge University Press, Cambridge, UK, 2000).
- [17]B. Liebchen and D. Levis, Chiral active matter, Europhys. Lett. 139, 67001 (2022).DOI
- [18]B. Liebchen and D. Levis, Collective behavior of chiral active matter: Pattern formation and enhanced flocking, Phys. Rev. Lett. 119, 058002 (2017).DOI
- [19]P. G. de Gennes and J. Prost, The Physics of Liquid Crystals (Oxford University Press, Oxford, 1995).
- [20]J. M. Kosterlitz and D. J. Thouless, Ordering, metastability and phase transitions in two-dimensional systems, J. Phys. C: Solid State Phys. 6, 1181 (1973).
- [21]A. Haldar, A. Sarkar, S. Chatterjee and A. Basu, Mobility-induced order in active XY spins on a substrate, Phys. Rev. E (Lett.) 108, 032101 (2023).DOI
- [22]A. Haldar, A. Sarkar, S. Chatterjee and A. Basu, Active XY model on a substrate: Density fluctuations and phase ordering, Phys. Rev. E 108, 034114 (2023).DOI
- [23]S. Torquato, Hyperuniform states of matter, Phys. Rep. 745, 1 (2018).DOI
- [24]S. Mukherjee, A. Basu, L. Chen and J. Toner, Associated long paper.

