The 3/4 size exponent for honeycomb self-avoiding walks

← All digests

Family 237 · Probability and statistical mechanics
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-11

1. The question

A self-avoiding walk cannot visit the same vertex twice. It is a simple model of a polymer whose different parts cannot occupy the same location. If the walk has $n$ steps, how large a region does it occupy?

In two dimensions, the predicted answer is a linear size of order $n^{3/4}$, intermediate between the square-root scale of an ordinary random walk and the linear scale of a straight path. The reciprocal exponent $4/3$ describes how the number of visited sites grows with spatial scale.

Family 237 contains thirteen manuscripts developing this prediction on the regular honeycomb lattice. Its strongest fixed-length statement concerns the diameter, local mass and covering numbers of a uniform walk at every sufficiently large integer length. Other papers treat critically weighted bridges, prescribed boundary chords, polygons, infinite walks and thermally discounted walks.

The sampling law matters throughout. A mean length in a strip is not automatically a theorem about a uniform walk of one exact length, and a diameter lower bound does not force the two endpoints to be far apart.

2. Background and essential definitions

Walks, size and critical weights

Fix a vertex $o$ of the honeycomb graph. An $n$-step self-avoiding walk is a sequence $\gamma=(\gamma_0,\ldots,\gamma_n)$, starting at $o$, with adjacent consecutive vertices and no repeated vertex. Let $c_n$ be their number and choose one uniformly.

Three size statistics are useful:

$$D(\gamma)=\max_{i,j}|\gamma_i-\gamma_j|, \qquad A(\gamma)=|\gamma_n-o|, \qquad M_\gamma(z,s)=\#\{\gamma_i:|\gamma_i-z|\leq s\}.$$

They are respectively diameter, endpoint distance and local mass in a ball. Let $N_\gamma(s)$ be the fewest radius-$s$ balls covering the visited vertices. Distances are Euclidean in a fixed regular embedding.

An alternative ensemble allows every length and gives a path of length $L$ the critical weight

$$\rho^L,\qquad \rho=(2+\sqrt2)^{-1/2}.$$

The reciprocal $1/\rho$ is the honeycomb connective constant, the exponential growth rate of $c_n$. A weight sum is a mass, not a probability until divided by its total.

Boundary paths use ports, midpoints of boundary edges. Their length counts visited vertices, and ports have no weight. This convention lets adjoining path pieces multiply weights exactly. A bridge goes from one strip boundary to the other without another boundary visit; an arch returns to its starting boundary. A polygon is an unoriented simple cycle. Polygon sums modulo translations count each translated shape once.

Reading exponent statements

For a positive deterministic quantity, $f(R)=R^{a+o(1)}$ means $\log f(R)/\log R\to a$. It permits subpower corrections and does not assert a limiting amplitude. The stronger notation $f(R)\asymp R^a$ means upper and lower fixed positive multiples.

For random variables, an exponent statement in probability means that, for every $\varepsilon>0$, the quantity lies between the two powers with exponents $a-\varepsilon$ and $a+\varepsilon$ with probability tending to one. An almost-sure statement is stronger and is identified separately below.

3. Development history and previous partial results

Nienhuis’s analysis of the dilute loop model predicted the honeycomb critical point and the two-dimensional polymer exponents, including $3/4$. Lawler, Schramm and Werner later formulated the conjectural connection between planar self-avoiding walk and SLE₈/₃, a conformally invariant random curve. That picture explains a coherent set of exponent predictions, but a prediction from a conjectured scaling limit is not a lattice theorem. Nienhuis’s work, scaling-limit conjectures.

Duminil-Copin and Smirnov proved the exact connective constant using a parafermionic observable: a weighted sum of paths with complex phases recording their winding. Its local cancellation yields a positive boundary identity after summing over a domain. This determines the critical activity, but not the full size exponent. Connective-constant theorem.

Subsequent work improved control of critical strip crossings. Beaton, Bousquet-Mélou, de Gier, Duminil-Copin and Guttmann proved that crossing mass tends to zero. Glazman and Manolescu obtained further bounds and boundary two-point invariance for a specified class of integrable rhombic deformations. Krachun and Panagiotis proved polynomial decay of strip mass and a quantitative sub-ballistic bound: uniform honeycomb walks remain within order $n/\log n$ of their origin with high probability. Strip decay, rhombic boundary results, quantitative sub-ballisticity.

Kesten’s bridge renewal method supplies another foundation. A bridge splits at lines crossed exactly once into irreducible pieces; critically weighted pieces can be concatenated independently. Later work developed infinite half-plane and strip laws through this representation. Kesten’s paper, strip and infinite-walk laws.

This family claims the sharper quantitative inputs and changes of ensemble needed to reach the predicted exponents. Its geometric conclusions rely on the accompanying analytic manuscripts, rather than following from the previously known connective constant alone.

4. What the new manuscripts claim

Every fixed length: diameter, mass and covering

M4 claims that for every $\delta>0$ and $q>0$, with probability at least $1-C_{\delta,q}n^{-q}$, a uniform $n$-step walk simultaneously satisfies

$$n^{3/4-\delta}\leq D(\gamma)\leq n^{3/4+\delta},$$

$$n^{-\delta}\min\{n,s^{4/3}\} \leq M_\gamma(z,s) \leq n^\delta\min\{n,s^{4/3}\},$$

and

$$n^{-\delta}(1+ns^{-4/3}) \leq N_\gamma(s) \leq n^\delta(1+ns^{-4/3}).$$

The last two bounds hold for every visited vertex $z$ and every real $1\leq s\leq n$. They apply to all sufficiently large integer $n$, with no subsequence or parity restriction for this ordinary vertex-rooted walk.

The paper also claims the same size exponent for the radius of gyration, the root mean square distance of visited vertices from their average position. Every fixed positive moment of diameter or radius of gyration has exponent $3p/4$ for moment order $p$. Spans in each of the three lattice-normal directions have matching lower exponents.

These are discrete geometric statements. They do not assert that the rescaled path converges to a continuum curve or identify the Hausdorff dimension of such a limit.

Critical crossings and polygons

M11 claims critical strip-crossing mass $B_h\asymp h^{-1/4}$, with a fixed source port and terminal port summed over the opposite wall. It also controls a first horizontal-displacement moment of returning paths, of order $h^{3/4}$.

M12 sharpens finite bridge information: the first-length mass is at most $Ch^{13/12}$, and mass at least $ch^{-1/4}$ is carried by bridges of length at least $ch^{4/3}$. Together with the total mass, this gives a bridge mean comparable to $h^{4/3}$. M8 obtains the exponent for both unrestricted bridges and bridges confined to a sufficiently wide fixed corridor. M9 gives another construction with logarithmically widening confinement.

For unrooted polygon classes, M1 claims

$$\sum_{D(\gamma)\geq R}\rho^{L(\gamma)}=R^{-2+o(1)}, \qquad \sum_{D(\gamma)\geq R}L(\gamma)\rho^{L(\gamma)}=R^{-2/3+o(1)},$$

and a truncated squared-length bound

$$\sum_{D(\gamma)\leq R}L(\gamma)^2\rho^{L(\gamma)} \leq R^{2/3+o(1)}.$$

Under either unweighted or length-weighted critical polygon mass, conditioning on $L\geq n$ then gives diameter $n^{3/4+o(1)}$ in probability. This is a length-cutoff ensemble, not a polygon of one prescribed length.

Chords and other walk laws

M2 claims length $R^{4/3+o(1)}$ in probability for critical chords between prescribed, macroscopically separated side-interior ports of regular-hexagon approximations. It also treats specified half-plane, parallel-strip and nonparallel-cut geometries with nondegenerate endpoint clearances. Half-plane arches additionally have the matching mean exponent. Its more specialized strip endpoint means retain their density-one-height or critically weighted even-height-mixture conventions.

M10 claims endpoint distance $n^{3/4+o(1)}$ almost surely for the infinite irreducible-bridge walk. Uniform strict bridges have terminal distance and diameter with that exponent for every sufficiently large compatible even vertex length. For unrestricted full-plane and half-plane walks, its endpoint lower law holds on a common density-one set of lengths: the fraction of excluded integers up to $N$ tends to zero.

Thermally discounted walks have weight $\rho^L e^{-L/N}$. At every large discount scale $N$, M10 claims $L=N^{1+o(1)}$ and endpoint distance and diameter $N^{3/4+o(1)}$, with matching positive-moment exponents. It also treats prescribed thermal endpoints at that spatial scale. Its near-critical inverse correlation scale is $t^{3/4+o(1)}$, and the small-force free energy is $s^{4/3+o(1)}$, with force response $s^{1/3+o(1)}$. These conclusions have their own orders of limits.

How the thirteen manuscripts fit together

Several papers develop overlapping constructions and stronger versions of particular inputs. Their roles can be summarized without treating them as thirteen independent fixed-length theorems.

Manuscript	Main contribution in the family
M1	Radial transfer estimates, polygon tails and length-cutoff laws
M2	Chords with prescribed boundary ports and qualified endpoint means
M3	Separating-loop partition exponents for every fixed positive fugacity; planar nesting
M4	Exact-length diameter, local mass, covering and gyration statements
M5	Signed cylinder propagation, free-boundary hexagon means and marked-polygon bounds
M6	Logarithmic length windows after averaging bridge mass over heights
M7	Squared-length polygon estimates and bounds for one marked arc
M8	Disk transfer construction and bridge means in a fixed corridor
M9	Polynomial transfer states, bridge means and free-boundary hexagon visits
M10	Renewal estimates and transfers to infinite, thermal and fixed-length laws
M11	Bounded-factor strip-crossing exponent
M12	Uniform two-mark bounds, sharp finite bridge moments and positive long-bridge mass
M13	Cylinder amplitudes selected by boundary caps and free-boundary triangle means

For instance, at loop fugacity two—an extra factor two per polygon—M3 and M5 obtain cylinder nesting exponent $1/6$ and planar exponent $1/12$. M3 gives a parameter-dependent exponent for every fixed positive fugacity. M12 obtains bounded-factor planar nesting estimates in its stated domains. M13’s triangle chord mean is $R^{4/3+o(1)}$, with both ports summed and a macroscopic diameter cutoff; it is not the prescribed-port law of M2.

5. Why the advance matters

Importance: 5/5. If correct, the fixed-length size and mass exponents would settle central geometric predictions for a canonical two-dimensional polymer model. The surrounding boundary and polygon results would provide a substantial quantitative theory supporting those exponents.

The local mass statement gives more than a single overall radius: roughly $s^{4/3}$ visited sites lie near each point at scale $s$, subject to the stated power slack. The covering statement expresses the corresponding number of spatial pieces needed to describe the trace.

The significance also lies in connecting distinct ensembles carefully. Fixed-length counting, infinite concatenation and critical boundary weights are useful for different purposes. Establishing the normalization and rare-event controls needed to move between them is a major part of the claimed contribution.

6. The main idea

The analytic papers start from exact lattice identities. Finite transfer matrices encode how partial paths and loops connect across a moving cut. Integrable local relations allow changes in those cuts and yield algebraic expressions for selected boundary states and marked observables.

These transformed expressions can have complex or signed terms even though the original walk weights are positive. A formal leading power is therefore insufficient: its coefficient could cancel. The papers develop normalization, nonvanishing and uniform remainder estimates to connect the transformed calculations back to positive path sums. Marking two points on a polygon records a squared-length quantity after summing the marks, providing control over unusually long paths packed into a small region.

The geometric papers then sew paths through corridors while keeping the joining lines recoverable. Recoverability bounds how many input constructions can produce the same output walk. First and second moments give positive mass to long bridges; independent irreducible pieces amplify these estimates through renewal arguments.

For exact-length conclusions, a generating-function estimate is not enough: one needs lower bounds on the coefficient or renewal probability at the chosen length. M10 develops such local renewal estimates for compatible bridge lengths. M4 obtains absolute bounds on the critical mass of bad geometric behavior, controls all subpaths and repeated visits to balls, and divides by a fixed-length normalization bounded below using submultiplicativity.

This last passage is why the family can state a diameter theorem for every length while retaining a weaker density-one qualification for the unrestricted endpoint lower law.

7. Scope, limitations, and open questions

The family concerns the regular honeycomb lattice and the specified critical weights. Integrable deformations used inside arguments are not a universality theorem for every planar lattice.

The ordinary fixed-length diameter theorem does not establish an endpoint-distance lower bound at every length. A long loop-like excursion can have a large diameter while ending near its starting point. The endpoint lower statement summarized from M10 retains its density-one qualification.

No manuscript here claims convergence of the uniform walk to SLE₈/₃. Consequently, continuum conformal invariance, a continuum parametrization and a continuum Hausdorff-dimension theorem should not be read into the discrete exponent conclusions.

Means, typical values and unnormalized masses must also remain distinct. M6’s logarithmic window averages over a range of heights. M13 sums both endpoints. M1 conditions polygon length from below. The prescribed-port, fixed-height and exact-length statements require their own additional estimates and cannot be substituted for these laws without justification.

Exact amplitudes and correction terms are not supplied by an $o(1)$ exponent. Removing the endpoint exceptional lengths, establishing the continuum scaling limit and extending the geometric theory beyond the honeycomb setting remain further directions beyond the stated results.

8. Sources and further reading

For the main geometric claims, begin with M4 and M10, then consult their explicitly cited finite inputs. M11 and M12 give comparatively direct statements of the strip and marked-polygon estimates. M2 is the source for prescribed boundary endpoints; the table above locates the other ensembles.

Duminil-Copin and Smirnov’s connective-constant paper explains the critical activity and boundary observable. Lawler, Schramm and Werner’s scaling-limit discussion provides the larger conjectural picture. The exponents reported here are the manuscripts’ lattice claims, rather than consequences assumed from that conjecture.

Source manuscripts

• M1. OpenAI, Radial transfer estimates and polygon length laws for honeycomb walks (September 26, 2026).
• M2. OpenAI, Critical honeycomb chords with prescribed boundary endpoints (September 26, 2026).
• M3. OpenAI, Cylinder loop weights and planar nesting (September 26, 2026).
• M4. OpenAI, Mass and covering exponents for fixed-length honeycomb walks (September 26, 2026).
• M5. OpenAI, Signed cylinder propagation and marked polygons on the honeycomb lattice (September 26, 2026).
• M6. OpenAI, Cylinder amplitudes and logarithmic bridge-length windows on the honeycomb lattice (September 26, 2026).
• M7. OpenAI, Marked polygon correlations and one-arc bounds (September 26, 2026).
• M8. OpenAI, Disk transfer representations and confined bridge mass (September 26, 2026).
• M9. OpenAI, Polynomial vacuum representations and bridge mass for honeycomb walks (September 26, 2026).
• M10. OpenAI, Renewal and changes of law for critical honeycomb walks (September 26, 2026).
• M11. OpenAI, Critical strip-crossing mass on the honeycomb lattice (September 26, 2026).
• M12. OpenAI, Uniform marked-polygon estimates and sharp finite bridge moments (September 26, 2026).
• M13. OpenAI, Cap-selected amplitudes and triangle chords for honeycomb walks (September 26, 2026).

Sources consulted on 11 October 2026 at repository revision fd4aeeb. Download the manuscript citations.