pith. sign in

arxiv: 2606.28105 · v1 · pith:4CAXCU4Pnew · submitted 2026-06-26 · ❄️ cond-mat.dis-nn · cond-mat.stat-mech· cs.CL

Scaling limit of the Random Language Model

Eric De Giuli This is my paper

Pith reviewed 2026-06-29 01:53 UTC · model grok-4.3

classification ❄️ cond-mat.dis-nn cond-mat.stat-mechcs.CL
keywords random language modelcondensation transitionscaling limitstochastic context-free grammarsrandom energy modellarge deviation principlelanguage statisticscorpus length dependence
0
0 comments X

The pith

The Random Language Model condenses at a critical scaled temperature x=1/8, making language statistics depend on corpus length below that point.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a quantitative theory for an ensemble of stochastic context-free grammars in the scaling limit where the number of hidden symbols grows to infinity while a grammar temperature shrinks proportionally to one over the log of that number. It shows that this limit is controlled by a large-deviation principle over rule-usage patterns, which maps under a semi-annealed approximation onto a class of random energy models. The central result is a condensation transition at x_c=1/8 below which rule usage concentrates sharply and generated language acquires explicit dependence on the length of the observed corpus, together with a second scale at x=1/2 where entropy begins to drop from its maximum. Explicit scaling laws are derived for the number of distinct rules used, the entropy, and related observables across the resulting regimes.

Core claim

In the scaling limit defined by N to infinity and tilde-epsilon_d to zero at fixed x equals tilde-epsilon_d times log N, the Random Language Model is described by a large-deviation principle over rule-usage patterns; a semi-annealed approximation then maps it to random energy models with nontrivial combinatorics, revealing a condensation transition at x_c=1/8 below which rule usage concentrates and language statistics depend nontrivially on corpus length, plus a second characteristic scale at x=1/2 marking the onset of entropy reduction, with explicit scaling laws for observables in the scaling, saturation, and critical regimes.

What carries the argument

The scaled temperature x equals tilde-epsilon_d log N together with the large-deviation principle over rule-usage patterns and its semi-annealed mapping to random energy models.

If this is right

  • Below x=1/8 rule usage concentrates and language statistics acquire a nontrivial dependence on corpus length.
  • Explicit scaling laws hold for the number of distinct rules, entropy, and related observables in distinct regimes controlled by grammar size, corpus length, and temperature.
  • The slow approach to the large-N limit is explained by the dependence on log N.
  • Universal statistical properties of language emerge from typical realizations of generative grammars.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the condensation transition controls natural language, then statistics extracted from corpora of different sizes should show the same nontrivial length dependence predicted for x below 1/8.
  • The same scaling framework could be used to analyze whether finite training data in other generative models produces analogous concentration of rule-like structures.
  • Numerical checks at moderate N could confirm or rule out the location of the transition by measuring how the effective number of rules scales with log N at fixed x.

Load-bearing premise

The semi-annealed approximation maps the problem to random energy models and the large-deviation principle over rule-usage patterns gives a controlled description in the scaling limit.

What would settle it

Simulations of finite but large N grammars that check whether the fraction of distinct rules used and the entropy exhibit the predicted sharp change exactly at x=1/8 when corpus length is varied.

Figures

Figures reproduced from arXiv: 2606.28105 by Eric De Giuli.

Figure 1
Figure 1. Figure 1: shows an example derivation of English phrase structure. The grammar G is defined by two objects G = {Mabc, OaB} over sets of hidden (”nonterminal”) symbols indexed by a = 1, . . . , N and observable (”ter￾minal”) symbols indexed by B = 1, . . . , T such that the rule a → bc has the weight Mabc and the rule a → B has the weight OaB. Conventionally, each derivation be￾gins with a distinguished symbol, calle… view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8 [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
read the original abstract

We develop a quantitative theory of the Random Language Model (RLM), an ensemble of stochastic context-free grammars, in a scaling limit where the number of hidden symbols $N \to \infty$ while the grammar temperature $\tilde{\epsilon}_d \to 0$ at fixed $x = {\tilde\epsilon}_d \log N$. In this limit, the model admits a controlled description based on a large-deviation principle over rule-usage patterns. A semi-annealed approximation maps the problem to a class of Random Energy Models with nontrivial combinatorics. We show that the RLM exhibits a condensation transition at a critical value $x_c=1/8$, below which rule usage concentrates and language statistics acquire a nontrivial dependence on corpus length. A second characteristic scale at $x=1/2$ marks the onset of entropy reduction from its maximal value. Across these regimes, we derive explicit scaling laws for the number of distinct rules, entropy, and related observables, identifying distinct scaling, saturation, and critical regimes controlled by the interplay of grammar size, corpus length, and temperature. The theory resolves previous ambiguities regarding the existence of a thermodynamic transition and explains the slow approach to the large-$N$ limit as a consequence of the dependence on $\log N$. It further provides a unified framework in which universal statistical properties of language emerge from typical realizations of generative grammars, with implications for both natural language statistics and the behavior of large language models.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper develops a quantitative theory of the Random Language Model (RLM), an ensemble of stochastic context-free grammars, in the scaling limit N→∞ with grammar temperature ilde{\epsilon}_d→0 at fixed x= ilde{\epsilon}_d log N. It employs a large-deviation principle over rule-usage patterns together with a semi-annealed approximation that maps the problem to a class of Random Energy Models with nontrivial combinatorics. The central results are a condensation transition at x_c=1/8 (below which rule usage concentrates and language statistics depend nontrivially on corpus length) and a second scale at x=1/2 marking the onset of entropy reduction; explicit scaling laws are derived for the number of distinct rules, entropy, and related observables across scaling, saturation, and critical regimes.

Significance. If the semi-annealed mapping and LDP control are valid, the work supplies a controlled analytic framework that resolves prior ambiguities about the existence of a thermodynamic transition in the RLM and accounts for the slow approach to the large-N limit via explicit log N dependence. It identifies universal statistical properties emerging from typical realizations of generative grammars, with direct implications for natural-language statistics and the behavior of large language models. The parameter-free character of the derived transition points and scaling laws (once the approximation is accepted) is a notable strength.

major comments (2)
  1. [Mapping to Random Energy Models and LDP control] The precise location of the condensation transition at x_c=1/8 and the distinction among the three regimes (x<1/8, 1/8<x<1/2, x>1/2) rest on the semi-annealed approximation and the assertion that the large-deviation rate function remains controlled in the joint limit. The manuscript must supply explicit error bounds or a dominance argument showing that combinatorial prefactors and approximation corrections do not shift the saddle or the critical value; without this, the claimed exact value x_c=1/8 is not yet secured.
  2. [Derivation of scaling laws] The abstract states that the description is “controlled,” yet the justification that the semi-annealed saddle remains dominant across the relevant scaling regimes is not accompanied by uniform error estimates. This is load-bearing for the claim that language statistics acquire a nontrivial corpus-length dependence only below x_c=1/8.
minor comments (2)
  1. Notation for the scaled temperature x and the hidden-symbol count N should be introduced with a single consistent definition early in the text to avoid repeated re-explanation.
  2. The abstract refers to “explicit scaling laws” for entropy and rule counts; a brief table summarizing the three regimes and their leading scalings would improve readability.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the thorough review and for identifying key points regarding the rigor of our approximations. We respond to the major comments below.

read point-by-point responses

  1. Referee: [Mapping to Random Energy Models and LDP control] The precise location of the condensation transition at x_c=1/8 and the distinction among the three regimes (x<1/8, 1/8<x<1/2, x>1/2) rest on the semi-annealed approximation and the assertion that the large-deviation rate function remains controlled in the joint limit. The manuscript must supply explicit error bounds or a dominance argument showing that combinatorial prefactors and approximation corrections do not shift the saddle or the critical value; without this, the claimed exact value x_c=1/8 is not yet secured.

    Authors: We acknowledge that the manuscript does not contain explicit error bounds or a full dominance argument for the semi-annealed mapping and LDP control. The value x_c=1/8 is obtained directly from the saddle-point equation of the mapped REM after applying the large-deviation rate function; combinatorial prefactors enter only as sub-exponential corrections that do not shift the leading saddle in the scaling limit. Nevertheless, a rigorous uniform bound on the approximation error across the joint limit is absent from the present analysis. revision: no

  2. Referee: [Derivation of scaling laws] The abstract states that the description is “controlled,” yet the justification that the semi-annealed saddle remains dominant across the relevant scaling regimes is not accompanied by uniform error estimates. This is load-bearing for the claim that language statistics acquire a nontrivial corpus-length dependence only below x_c=1/8.

    Authors: The term “controlled” in the abstract refers to the fact that all scaling laws follow from a single saddle-point evaluation once the semi-annealed mapping is accepted. We agree that uniform error estimates justifying dominance of this saddle in every regime are not supplied. A brief clarifying paragraph can be added to the discussion section stating the assumptions under which the mapping holds, but deriving the requested bounds lies outside the scope of the current work. revision: partial

standing simulated objections not resolved
  • Explicit error bounds or dominance arguments for the semi-annealed approximation and LDP control in the joint scaling limit

Circularity Check

0 steps flagged

No circularity; central claims follow from stated semi-annealed mapping and LDP without reduction to inputs

full rationale

The derivation proceeds by introducing a semi-annealed approximation that maps the RLM to a REM class, then invoking an LDP over rule-usage patterns to locate x_c=1/8 and the x=1/2 scale. These steps are presented as controlled approximations whose outputs (condensation transition, scaling laws) are derived quantities rather than inputs. No self-definitional loop, fitted parameter renamed as prediction, or load-bearing self-citation chain appears in the provided text. The result is therefore independent of its own conclusions and scores at the low end of the range.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The theory rests on the large-deviation principle over rule-usage patterns and the semi-annealed approximation mapping to Random Energy Models, both introduced for the scaling limit analysis.

axioms (2)
  • domain assumption Large-deviation principle over rule-usage patterns holds in the scaling limit
    Invoked to provide controlled description of the model
  • ad hoc to paper Semi-annealed approximation is valid for mapping to Random Energy Models with nontrivial combinatorics
    Used to derive explicit scaling laws for observables

pith-pipeline@v0.9.1-grok · 5792 in / 1533 out tokens · 48844 ms · 2026-06-29T01:53:57.323398+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

63 extracted references · 3 canonical work pages · 1 internal anchor

  1. [1]

    = 2 ℓ − m, so that Ztree ≡ Q i P(Ti|G) = pℓ E(1 − pE)2ℓ−mZ −m tree just gives a trivial factor. The first result of [ 19] is that Z(G; m, ℓ) has a field- theoretic representation in terms of two complex scalar 2 The idea to relate belief propagation equations to the Random Energy Model is due to Alessio Giorlandino and Sebastian Goldt. ‘fields’ La, Ra, a ...

  2. [2]

    Entropy in the frozen region: In the frozen regime we have log z = log k∗ + 3 2k∗ − log(p/N) = log(3/2) − log log α + log α − log(p/N) = C − log a + Fz(log G2/3) with C = 2 log(3/2) − 1 2 log(2πe3) and Fz(L) = −2 log log α + log α − 3L 1 + α (59) We also have p n log e/v = 2 3 a √ 2πe3 log(α)Gβ(G) N n 1 − 1 1 + α log G = Fp(log G2/3) with Fp(L) = 2 3 √ 2π...

  3. [3]

    We have log z = − log a − 3 k∗ − a +

    Entropy in the critical region: For x < 1/8 but close to 1/8 we obtain another con- tribution from the singularity at x = 1/8. We have log z = − log a − 3 k∗ − a + . . . with k∗ ≈ 3 4 + 1 2 a + 1 2 q a2 + a + 9 4 . Then we get (with a ∼ √ 8x/β) Hd(ℓ∞) ≈ 1 4 log 8x + 3k∗/2 (k∗ − √ 8x)2 + const x ≈ 1/8 (64) Call this prediction Hhi(x). It predicts the obser...

  4. [4]

    babbling

    Entropy in the transition region: In the transition regime we need both log z and the p/n terms. We have, in the scaling limit, − log z + (p/n) log(e/v) = 1 + χ log N + . . . so that Hd(ℓ∞) ≈ 1 − 1 8x log N + 1 2 1/8 < x < 1/2 independent of ℓ∞ in this approximation. This result is very interesting as it implies a change of scaling Hd ∼ log N for x > 1/8 ...

    2020

  5. [5]

    G. K. Zipf, The psycho-biology of language: An intro- duction to dynamic philology (Routledge, Milton Park, 2013)

    2013

  6. [6]

    H. S. Heaps, Information retrieval: Computational and theoretical aspects (Academic Press, Inc., 1978)

    1978

  7. [7]

    Shannon and W

    C. Shannon and W. Weaver, The Mathematical Theory of Communication , 1st ed. (University of Illinois Press, Urbana, Illinois, USA, 1964)

    1964

  8. [8]

    C. E. Shannon, Bell system technical journal 30, 50 (1951)

    1951

  9. [9]

    Schürmann and P

    T. Schürmann and P. Grassberger, Chaos: An Interdis- ciplinary Journal of Nonlinear Science 6, 414 (1996)

    1996

  10. [10]

    A. M. Petersen, J. N. Tenenbaum, S. Havlin, H. E. Stan- ley, and M. Perc, Scientific reports 2, 943 (2012)

    2012

  11. [11]

    Ferrer i Cancho and R

    R. Ferrer i Cancho and R. V. Solé, Journal of Quantita- tive Linguistics 8, 165 (2001)

    2001

  12. [12]

    Corominas-Murtra and R

    B. Corominas-Murtra and R. V. Solé, Physical Review E 82, 011102 (2010)

    2010

  13. [13]

    Corral, G

    A. Corral, G. Boleda, and R. Ferrer-i Cancho, PloS one 10, e0129031 (2015)

    2015

  14. [14]

    Bentz, D

    C. Bentz, D. Alikaniotis, M. Cysouw, and R. Ferrer-i Cancho, Entropy 19, 275 (2017). 18

    2017

  15. [15]

    Chomsky, Syntactic structures (Walter de Gruyter, Berlin, 2002)

    N. Chomsky, Syntactic structures (Walter de Gruyter, Berlin, 2002)

    2002

  16. [16]

    Chomsky, Aspects of the Theory of Syntax , Vol

    N. Chomsky, Aspects of the Theory of Syntax , Vol. 11 (MIT press, Cambridge, 2014)

    2014

  17. [17]

    Chomsky, The minimalist program (MIT press, 2014)

    N. Chomsky, The minimalist program (MIT press, 2014)

    2014

  18. [18]

    De Saussure, Course in general linguistics 1, 65 (1916)

    F. De Saussure, Course in general linguistics 1, 65 (1916)

    1916

  19. [19]

    Jurafsky and J

    D. Jurafsky and J. H. Martin, Computational Linguistics, and Speech Recognition, Prentice Hall 2 (2000)

    2000

  20. [20]

    Ferrer-i Cancho, C

    R. Ferrer-i Cancho, C. Gómez-Rodríguez, J. L. Esteban, and L. Alemany-Puig, Physical Review E 105, 014308 (2022)

    2022

  21. [21]

    DeGiuli, Phys

    E. DeGiuli, Phys. Rev. Lett. 122, 128301 (2019)

    2019

  22. [22]

    J. C. Kieffer and E.-H. Yang, IEEE Transactions on In- formation Theory 46, 737 (2000)

    2000

  23. [23]

    De Giuli, Journal of Physics A: Mathematical and Theoretical 52, 504001 (2019)

    E. De Giuli, Journal of Physics A: Mathematical and Theoretical 52, 504001 (2019)

    2019

  24. [24]

    Mézard, G

    M. Mézard, G. Parisi, and M. Virasoro, Spin glass the- ory and beyond: An Introduction to the Replica Method and Its Applications , Vol. 9 (World Scientific Publishing Company, 1987)

    1987

  25. [25]

    Nakaishi and K

    K. Nakaishi and K. Hukushima, Physical Review Re- search 4, 023156 (2022)

    2022

  26. [26]

    Lalegani and E

    F. Lalegani and E. De Giuli, Physical Review E 109, 054313 (2024)

    2024

  27. [27]

    Derrida, Physical Review Letters 45, 79 (1980)

    B. Derrida, Physical Review Letters 45, 79 (1980)

    1980

  28. [28]

    Derrida, Phys

    B. Derrida, Phys. Rev. B 24, 2613 (1981)

    1981

  29. [29]

    Phase struc- ture of the random language model,

    A. Giorlandino, E. D. Giuli, and S. Goldt, “Phase struc- ture of the random language model,” (2026)

    2026

  30. [30]

    T. L. Booth, in 10th annual symposium on switching and Automata Theory (swat 1969) (IEEE, 1969) pp. 74–81

    1969

  31. [31]

    C. S. Wetherell, ACM Computing Surveys (CSUR) 12, 361 (1980)

    1980

  32. [32]

    Mézard and A

    M. Mézard and A. Montanari, Information, physics, and computation (Oxford University Press, 2009)

    2009

  33. [33]

    De Giuli, Journal of Physics A: Mathematical and Theoretical 55, 489501 (2022)

    E. De Giuli, Journal of Physics A: Mathematical and Theoretical 55, 489501 (2022)

    2022

  34. [34]

    Ben Arous, L

    G. Ben Arous, L. V. Bogachev, and S. A. Molchanov, Probability theory and related fields 132, 579 (2005)

    2005

  35. [35]

    J. M. Kosterlitz and D. J. Thouless, Journal of Physics C: Solid State Physics 6, 1181 (1973)

    1973

  36. [36]

    Kosterlitz, Journal of Physics C: Solid State Physics 7, 1046 (1974)

    J. Kosterlitz, Journal of Physics C: Solid State Physics 7, 1046 (1974)

    1974

  37. [37]

    Zinn-Justin, Quantum field theory and critical phe- nomena (Clarendon Press, 1996)

    J. Zinn-Justin, Quantum field theory and critical phe- nomena (Clarendon Press, 1996)

    1996

  38. [38]

    P. M. Chaikin and T. C. Lubensky, Principles of Con- densed Matter Physics (Cambridge University Press, Cambridge, U.K., 2000)

    2000

  39. [39]

    T. M. Cover and J. A. Thomas, Elements of information theory (John Wiley & Sons, 1999)

    1999

  40. [40]

    R. F. i Cancho and R. V. Solé, Proceedings of the Na- tional Academy of Sciences 100, 788 (2003)

    2003

  41. [41]

    Mazzolini, J

    A. Mazzolini, J. Grilli, E. De Lazzari, M. Osella, M. C. Lagomarsino, and M. Gherardi, Physical review E 98, 012315 (2018)

    2018

  42. [42]

    Mazzolini, A

    A. Mazzolini, A. Colliva, M. Caselle, and M. Osella, Physical Review E 98, 052139 (2018)

    2018

  43. [43]

    T. OLMo, P. Walsh, L. Soldaini, D. Groeneveld, K. Lo, S. Arora, A. Bhagia, Y. Gu, S. Huang, and M. Jordan, arXiv preprint arXiv:2501.00656 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  44. [44]

    Scheibner, L

    C. Scheibner, L. M. Smith, and W. Bialek, arXiv preprint arXiv:2512.24969 (2025)

    work page arXiv 2025

  45. [45]

    Nakaishi and K

    K. Nakaishi and K. Hukushima, Physical Review Re- search 6, 033216 (2024)

    2024

  46. [46]

    Y. Toji, J. Takahashi, V. Roychowdhury, and H. Miya- hara, Physical Review E 113, 015305 (2026)

    2026

  47. [47]

    Longobardi, The handbook of contemporary syntactic theory , 562 (2001)

    G. Longobardi, The handbook of contemporary syntactic theory , 562 (2001)

    2001

  48. [48]

    Longobardi and A

    G. Longobardi and A. Treves, in A Cartesian dream: A geometrical account of syntax. In honor of Andrea Moro. , Vol. Rivista di Grammatica Generativa/Research in Gen- erative Grammar., edited by M. Greco and D. Mocci (Lingbuzz Press, 2023)

    2023

  49. [49]

    Yarahmadi, K

    H. Yarahmadi, K. I. Ryom, G. Longobardi, and A. Treves, Physical Review E 113, 034312 (2026)

    2026

  50. [50]

    Berthier and G

    L. Berthier and G. Biroli, Reviews of Modern Physics 83, 587 (2011)

    2011

  51. [51]

    Mosam, D

    F. Mosam, D. Vidaurre, and E. De Giuli, Physical Re- view E 104, 024305 (2021)

    2021

  52. [52]

    Mosam and E

    F. Mosam and E. De Giuli, arXiv preprint arXiv:2602.04905 (2026)

    work page arXiv 2026

  53. [53]

    Bernard, J.-P

    M. Bernard, J.-P. Bouchaud, and P. Le Doussal, Physical Review E 113, L032101 (2026)

    2026

  54. [54]

    Lalegani, Insights from Random Language Model , Ph.D

    F. Lalegani, Insights from Random Language Model , Ph.D. thesis, Toronto Metropolitan University (2024)

    2024

  55. [55]

    I. J. Good, The Annals of Mathematical Statistics 34, 911 (1963)

    1963

  56. [56]

    Barvinok, Combinatorics, Probability and Computing 19, 517 (2010)

    A. Barvinok, Combinatorics, Probability and Computing 19, 517 (2010). Appendix A: Counting derivations Suppose we have p rules ar → brcr, r = 1, . . . , p and q rules tr → Br. Write Mabc = X r δ(abc),(ar,br,cr), O aB = X r δ(aB),(tr,Br). (A1) The partition function Z(G; m, ℓ) then counts how many configurations can be made with this grammar. We will evalu...

    2010

  57. [57]

    Similarly H∗L1 ≈ 2gℓ/N

    Corrections to leading order To examine the corrections, we have H∗L1 = 2gp N [ℓ/p + log(N/2gℓ) + log(H1L∗ − 2ghL∗)] and then ℓe−ℓ/peN H1L∗/(2gp) = ℓ + 2p log(N/2gℓ) + 2p log(H1L∗ − 2ghL∗) giving H1L∗ ≈ 2gℓ/N up to logarithmic corrections. Similarly H∗L1 ≈ 2gℓ/N. To obtain the corrections we write H1L∗ = 2gℓ/N + a, H∗L1 = 2gℓ/N + b, giving b = (gℓ/N) h eN...

  58. [58]

    This means that the average sentence length ℓ/m cannot grow faster than linearly in N

    The approximation further requires a, b ≲ 1 which reduce to 4p2g2 (2p2 − ℓ2) mp N ≲ 1 which can only hold if 5 ℓ < √ 2p ≲ 4 √ 2N m. This means that the average sentence length ℓ/m cannot grow faster than linearly in N

  59. [59]

    contingency tables

    Quenched grammar average The above results can be extended to computation of log Z, the quenched average over grammars. Using the replica method log Z = lim q→0 d(Zq)/dq = lim q→0 dZq/dq we need to replicate all {η, ζ, ξ, La, Ra}, adding a replica index α = 1 , . . . , q. The key point is that the exponential coupling gets replaced by eηHaLaRa → e ∑ α ηαH...

  60. [60]

    Then ˜n is fixed by n = N en/pa X k≥1 kz k k − a ≈ N en/pa X k zk = N en/paz/(1 − z) (C1) where z = N a/2xe−˜n and we have assumed a ≪ 1, which is 8x ≪ 1

    F rozen regime To orient ourselves consider the frozen regime r = 0, which means that x < 1/8. Then ˜n is fixed by n = N en/pa X k≥1 kz k k − a ≈ N en/pa X k zk = N en/paz/(1 − z) (C1) where z = N a/2xe−˜n and we have assumed a ≪ 1, which is 8x ≪ 1. This gives z−1 = 1 + en/p/G. 25 where G = n/N a. We still have to fix p through p = X k pk = N en/pa X k≥1 ...

  61. [61]

    High temperature regime: For small trees it is convenient to make an approximation considering only the k = 1 and k = 2 terms in the sums. At high temperature and in small trees, most rules are used uniquely; if we keep only the k = 1 and k = 2 terms we have n ≈ N 3en/pe−˜n[g1 + 2g2e−˜n] (C6) p ≈ N 3en/pe−˜n[g1 + g2e−˜n] (C7) so that e−˜n = −g1 4g2 " −1 +...

  62. [62]

    Then ˜n is fixed by n = vN a X k k k + mc − 1 mc − 1 zk k − a ≈ vN a X k k + mc − 1 mc − 1 zk = vN a 1 (1 − z)mc − 1 where we assume 1 ≫ a, i.e

    F rozen regime Again start with the frozen regime r = 0, which means that x < 1/8. Then ˜n is fixed by n = vN a X k k k + mc − 1 mc − 1 zk k − a ≈ vN a X k k + mc − 1 mc − 1 zk = vN a 1 (1 − z)mc − 1 where we assume 1 ≫ a, i.e. x ≪ 1/8. This is solved z = 1 − (1 + G/v)−1/mc (D1) 27 Now we need to find p and v. Since v = e ∑ k′ pk′ N log ( 1+ k′ mc −1 ) ha...

  63. [63]

    For example if r = 1 so that 1/8 < x < 1/2, then we simply replace g1 by the other asymptotic expression

    T ransition regime The same technique can be adapted to other regions. For example if r = 1 so that 1/8 < x < 1/2, then we simply replace g1 by the other asymptotic expression. This results in z = 1 − 0 BB@1 + 1 v 2 664G − N 2 a vzm ce1/4˜ϵd e−X∗ + vzm c | {z } Gtrans 3 775 1 CCA −1/mc . 1/8 < x < 1/2 (D20) 31 For log(Gtrans/v) ≪ mc we get zmc ≈ log(Gtran...