Reduced Bumpless Pipe Dreams

• Reduced bumpless pipe dreams (RBPD) are combinatorial tilings on n×n grids using six tile types with strict reducedness to represent canonical permutations.
• RBPDs encode monomials in Schubert polynomials by converting tile configurations into algebraic expressions tied to geometric degenerations.
• Algorithmic studies leverage local 2×2 flips and droop moves, employing MCMC sampling for rapid mixing and exploring connections to classical pipe dreams.

A reduced bumpless pipe dream (RBPD) is a combinatorial object serving as a canonical representative of permutations in $S_n$ and providing a geometric and algebraic framework for understanding Schubert polynomials. RBPDs are distinguished by their tiling rules on an $n\times n$ grid with six tile types and strict reducedness—each pair of pipes crosses at most once. RBPDs offer a fundamentally different, yet dual, approach to classical reduced pipe dreams, bridging contemporary combinatorial models and deep results in algebraic geometry, symmetric functions, and statistical mechanics.

1. Definitions and Fundamental Properties

An RBPD of size $n$ is a filling of an $n\times n$ square grid by six tile types: horizontal, vertical, two “bumpless” elbows (NW and SE turns), cross, and blank, subject to “domain wall” boundary conditions. Colored pipes labeled $k$ enter from the south at column $k$ and exit on the east at row $k$; the entire system forms $n$ nonintersecting paths. Reducedness mandates that each pair of pipes crosses at most once. For each RBPD $D$, the permutation $w \in S_n$ is determined by the exit order.

RBPDs are closely related to alternatingsign matrices (ASM) in the unreduced case. However, the reduced set fails to inherit the sublattice property inherent to ASMs: the meet of two reduced BPDs in the ASM lattice need not be reduced, with the first failures at $n\times n$0 (Anderson et al., 20 Mar 2026, Huang et al., 2023).

2. Connections to Schubert Polynomials

RBPDs encode monomials in the (single or double) Schubert polynomial $n\times n$1. Lam–Lee–Shimozono and Weigandt proved that:

$n\times n$2

where $n\times n$3 is the set of blank tiles in $n\times n$4. The ordinary Schubert polynomial arises by $n\times n$5, yielding

$n\times n$6

The principal specialization $n\times n$7 enumerates RBPDs of given boundary permutation $n\times n$8. The geometric underpinning is a diagonal Gröbner degeneration of matrix Schubert varieties, with RBPDs indexing the irreducible components and the weight recording scheme-theoretic multiplicity (Klein et al., 2021, Huang et al., 2023).

3. Enumerative Recurrences and Asymptotics

For $n\times n$9, three core recurrences exist:

• Descent formula (Macdonald):

$n$0

with $n$1.

• Transition formula (Lascoux-Schützenberger): Recursive on $n$2-patterns, reducing to $n$3 for dominant $n$4.
• Cotransition formula (Knutson): Recurrence on Bruhat covers altering a minimal position, with $n$5.

A remarkable enumerative result is that for $n$6 varying over $n$7, the maximal value of $n$8 need not be achieved for a layered permutation: counterexamples occur for $n$9 onward, disproving the Merzon–Smirnov conjecture. Yet, simulations suggest that the asymptotic growth rate of $n\times n$0 matches that for maximal layered permutations, i.e.,

$n\times n$1

(Anderson et al., 20 Mar 2026).

4. Algorithmic and Combinatorial Structures

4.1. Local Moves and Sampling

RBPDs admit two central types of local moves:

• $n\times n$2 Flips: Drip, annihilation/creation, and relocation moves analogous to six-vertex model updates, preserving reducedness.
• Droop/Undroop Moves: For rectangles $n\times n$3, a “droop” rearranges elbows and pipes globally while maintaining the permutation and reducedness. Any RBPD can be reached from its Rothe BPD using droops alone (Lam–Lee–Shimozono), and the combination of droops and $n\times n$4 flips is ergodic (Klein et al., 2021, Anderson et al., 20 Mar 2026).

4.2. Markov Chain Monte Carlo (MCMC) and Lattice Structure

Exact uniform sampling via monotone Coupling-From-The-Past (CFTP) fails for RBPDs because sublattice closure is lost: monotonicity of update maps is violated, leading to false coalescence of extremal chains. As a computational workaround, an MCMC sampler using a mixture of local ($n\times n$5) and global (droop) updates achieves rapid mixing, with empirical burn-in $n\times n$6 steps at $n\times n$7 on personal computers. For $n\times n$8, coordinated parallel sampling yields fast, scalable enumeration and statistical analysis (Anderson et al., 20 Mar 2026).

5. Canonical Bijection with Pipe Dreams and Additional Models

A canonical, weight-preserving bijection exists between reduced bumpless pipe dreams and reduced pipe dreams (Gao–Huang). This bijection arises by iteratively “popping” and column-rectifying blank tiles, extracting a reduced compatible sequence indexing the corresponding reduced pipe dream. The bijection commutes with Monk’s rule and extends to the double Schubert polynomial and back-stable settings (Gao et al., 2021, Huang et al., 2023).

A unified puzzle model encodes both RBPDs and classical pipe dreams as edge cases of a larger “master” tiling, equipping the combinatorics with Yang–Baxter moves and recasting the divided-difference recurrences for Schubert polynomials in the language of tile-moves (Xiong, 2020).

6. Broader Connections: ASMs, Plane Partitions, and Growth Diagrams

RBPDs are linked to ASMs (alternating sign matrices) via a bijection in the unreduced case, and refined bijections exist under pattern-avoidance for permutations. In particular, for $n\times n$9-avoiding $k$0, RBPDs (and their associated ASMs) are related by explicit weight- and poset-preserving bijections to totally symmetric self-complementary plane partitions (TSSCPP), with the combinatorics of “droop” and “slide” moves mediating the correspondence (Huang et al., 2023).

Further, RBPDs serve as insertion objects in new RSK-type and growth diagram algorithms tailored for Schubert calculus, yielding positive, combinatorial rules for Schubert structure constants in separated-descent cases (Huang et al., 2022).

7. Open Problems and Computational Resources

Several directions remain active:

• Connectivity under $k$1 flips without droops: the full characterization of “trapped” configurations is unknown.
• Mixing time estimates for the droop-augmented Markov chain.
• Asymptotic questions: rigorous derivation of the “Schubert permuton,” “limit shape,” and analysis of fluctuation regimes (e.g., connection to Gaussian free fields).
• Weighted specializations: extension to $k$2-specializations, Grothendieck and double Schubert polynomials, and development of corresponding fast samplers and theoretical analyses.

Code for enumeration (via all three recurrences) and sampling (including demonstration of CFTP failure and droop-augmented MCMC) is publicly available (Anderson et al., 20 Mar 2026).

---

References:

(Anderson et al., 20 Mar 2026, Klein et al., 2021, Gao et al., 2021, Xiong, 2020, Huang et al., 2023, Huang et al., 2022)