Quadratic Probing Revisited: Smoothed Analysis and the Fall of Robin Hood

Abstract: Quadratic probing is one of the most widely used open-addressing hash-table schemes in practice, but after more than half a century, even its most basic performance guarantees remain poorly understood. In this paper, we revisit quadratic probing through the lens of a smoothed variant in which each key follows a random probe sequence where its $k$th probe is expected at offset $Θ(k<sup>2)$. This is simultaneously a toy model for better understanding regular quadratic probing and a natural hashing scheme in its own right. We analyse smoothed quadratic probing for both Robin Hood ordering and anti-Robin Hood ordering and reveal a surprising separation: At load factor $1-\varepsilon$, anti-Robin Hood achieves an expected query time of $Θ(\log \varepsilon<sup>{-1})$, which matches the conjectured expected average successful query time for regular quadratic probing, while Robin Hood falls short at $Θ(\varepsilon<sup>{-1/2})$. Our analysis generalises to degree-$d$ probing for any $d \ge 1$ with expected query time $$O(\max(\log \varepsilon^{-1}, \varepsilon<sup>{1-2/d}))$$ for anti-Robin Hood and $Θ(\varepsilon<sup>{-1/d})$ for Robin Hood. Finally, we go beyond smoothed analysis: using the probabilistic method, we show that for every $d \ge 2$, almost every random fixed-offset degree-$d$ probing sequence achieves expected query time $O(\log \varepsilon<sup>{-1})$ under anti-Robin Hood ordering, simultaneously over all admissible table sizes and load factors. Thus, while quadratic probing itself remains elusive, we prove that essentially all quadratic-probing-like fixed-offset schemes achieve the ideal performance under the anti-Robin Hood ordering.

• The paper introduces a smoothed variant of quadratic probing that rigorously establishes expected query and insertion time bounds, confirming optimal O(log(1/ε)) performance under anti-Robin Hood ordering.
• It reveals that Robin Hood ordering, contrary to popular belief, leads to increased displacement and suboptimal query times compared to anti-Robin Hood in quadratic and degree‑d probing.
• The study extends the analysis to degree‑d polynomial probing using probabilistic methods, providing tight lower bounds and practical implications for high-performance hash table implementations.

Smoothed Analysis of Quadratic Probing and the Performance of Robin Hood

Introduction

Quadratic probing is a classic open addressing scheme for hash tables designed to mitigate primary clustering prevalent in linear probing. Despite its widespread use in high-performance and production-grade hash tables, rigorous analysis of its performance—particularly at high load factors—remains elusive decades after its introduction. Practitioners often rely on conjectured time bounds (notably, $O(\log \varepsilon^{-1})$ expected query time at load factor $1-\varepsilon$), but theoretical guarantees are lacking for the regime $\varepsilon \ll 1$. This stagnation in analysis impedes progress on key algorithmic questions, such as the optimality of organizational policies (Robin Hood vs. anti-Robin Hood), generalization to other polynomial-based probing schemes, and the effects of probe sequence design.

This paper, "Quadratic Probing Revisited: Smoothed Analysis and the Fall of Robin Hood" (2607.13247), introduces a smoothed variant of quadratic probing that enables analytical advances while retaining algorithmic naturalness. By leveraging this smoothed model, the authors resolve longstanding questions about expected query and insertion times, the implications of Robin Hood and anti-Robin Hood tie-breaking policies, and, crucially, the generalization to degree-$d$ polynomial probing. The results yield new performance separations, optimality results for specific orderings, and a near-complete picture of open-addressed hash table asymptotics under polynomial probing.

Model and Methods

The smoothed quadratic probing model is constructed as a tractable surrogate for standard quadratic probing. Instead of utilizing the deterministic probe sequence $h(x), h(x) + 1^2, h(x) + 2^2, \ldots$, each key $x$ is given a random sequence with the $k$-th offset present with probability $\Theta(k^{-1/2})$. This construction admits two interpretations: as a model for the effect of random perturbations in the probe sequence (smoothed analysis), and as a valid randomized algorithm in its own right.

The analysis focuses on two ordering policies: Robin Hood (where, upon collision, the element with the longest probe path retains the slot) and anti-Robin Hood (where the element closest to its origin prevails). Both policies produce history-independent table states given a fixed set of inserted keys. The authors generalize both policies and probe sequences to degree-$d$ polynomial probing, where offsets scale as $i^d$.

For theoretical completeness, the analysis is extended from smoothed-probing (key-independent, randomized sequences) to fixed-offset randomized degree-$1-\varepsilon$0 probe sequences, achieved via the probabilistic method.

Results

Main Quantitative Findings

• Anti-Robin Hood Ordering (Smoothed Quadratic Probing, $1-\varepsilon$1):
  • Expected query time: $1-\varepsilon$2.
  • Amortized expected insertion time: $1-\varepsilon$3.
  • This matches the conjectured (but previously unproven) bounds for classical quadratic probing and provides the first proof of a natural scheme with optimal asymptotics under anti-Robin Hood ordering.
• Robin Hood Ordering (Smoothed Quadratic Probing, $1-\varepsilon$4):
  • Expected query time: $1-\varepsilon$5 for all keys, with corresponding amortized insertion time $1-\varepsilon$6.
  • This demonstrates that, contrary to folklore, Robin Hood ordering does not yield the conjectured optimal query time for quadratic probing, but significantly increases displacement and query cost.
• Generalization to Degree-$1-\varepsilon$7 Probing:
  • Anti-Robin Hood: Expected query time $1-\varepsilon$8; insertion time $1-\varepsilon$9.
  • Robin Hood: Query time $\varepsilon \ll 1$0; insertion time $\varepsilon \ll 1$1.
  • For $\varepsilon \ll 1$2, anti-Robin Hood achieves the optimal $\varepsilon \ll 1$3 query time.
• Beyond Smoothed Analysis—Probabilistic Method on Fixed-Offset Sequences:
  • For every $\varepsilon \ll 1$4, with high probability over the random choice of offset sequence, all load factors $\varepsilon \ll 1$5 and table sizes experience $\varepsilon \ll 1$6 query time and $\varepsilon \ll 1$7 insertion time under anti-Robin Hood.
  • The lower bounds prove that these are optimal for any fixed-offset scheme under anti-Robin Hood.

Technical Mechanisms

• The analysis is anchored on smoothed rather than deterministic probe sequences, enabling the use of concentration inequalities and negative feedback loops to derive tight tail bounds on element displacement and query cost.
• For anti-Robin Hood, the negative feedback argument ensures that, as long as some displacement tail is large, there will quickly be enough free slots examined, causing the tail to decay rapidly.
• For Robin Hood, clustering effects re-emerge, and attempts to “equalize” displacement to the origin inadvertently result in longer probe sequences overall.
• The probabilistic method—selecting random fixed-offset sequences—demonstrates that ideal performance is typical, and any poor behavior of the classic quadratic sequence would be highly anomalous.

Table: Summary of Key Results

Probing Scheme	Ordering	Query Time	Insertion Time
Smoothed quadratic ($\varepsilon \ll 1$8)	Anti-Robin Hood	$\varepsilon \ll 1$9	$d$0
Smoothed quadratic ($d$1)	Robin Hood	$d$2	$d$3
Smoothed degree-$d$4	Anti-Robin Hood	$d$5	$d$6
Smoothed degree-$d$7	Robin Hood	$d$8	$d$9
Fixed-offset ($h(x), h(x) + 1^2, h(x) + 2^2, \ldots$0)	Anti-Robin Hood	$h(x), h(x) + 1^2, h(x) + 2^2, \ldots$1	$h(x), h(x) + 1^2, h(x) + 2^2, \ldots$2

Implications

Algorithmic and Theoretical Consequences

• The anti-Robin Hood tie-breaking policy is provably optimal in expected query and insertion time for smoothed and typical fixed-offset degree-$h(x), h(x) + 1^2, h(x) + 2^2, \ldots$3 probing, with $h(x), h(x) + 1^2, h(x) + 2^2, \ldots$4.
• Robin Hood ordering, while advantageous for query uniformity in linear probing, is empirically and theoretically suboptimal in quadratic and higher-degree settings, yielding query and insertion times with worse scaling.
• The performance gap between Robin Hood and anti-Robin Hood policies in quadratic probing is substantial and persists as $h(x), h(x) + 1^2, h(x) + 2^2, \ldots$5 increases.
• The smoothed model provides deep analytical insight into why quadratic probing is effective and pinpoints the failure modes of Robin Hood ordering which resurrect linear-clustering phenomena.
• The tightness of lower bounds in the smoothed and fixed-offset settings underscores the optimality of the presented algorithms and analysis.

Practical Considerations

• For hash table implementations, these results justify the adoption of anti-Robin Hood ordering in quadratic- and higher-degree polynomial hashing schemes when seeking low expected query/insertion times under high load.
• The anti-Robin Hood version of quadratic probing is highly local—making it attractive for cache performance—and attains the best known tradeoff between probe locality and query/insertion asymptotics.
• The robustness of the results under the random fixed-offset construction means hash table practitioners can use randomly generated probe sequences without risking large performance degradation.

Open Questions and Future Directions

• The analysis does not resolve the classical deterministic quadratic probing sequence, leaving the (now even more pressing) question of whether $h(x), h(x) + 1^2, h(x) + 2^2, \ldots$6 is pathological open.
• The negative result for Robin Hood in quadratic probing challenges standard recommendations and motivates further empirical study of probe sequence/ordering interactions and hybrid policies.
• The general methodology—analyzing via smoothed versions and probabilistic methods—may extend to understanding other hard-to-analyze randomized data structures and variants, including high-independence hash functions and variants of cuckoo hashing.

Conclusion

This work makes significant advances in the theoretical understanding of open-addressed hash tables, particularly quadratic probing and its generalizations. By introducing and exploiting a smoothed quadratic probing model alongside a detailed treatment of ordering policies, the authors demonstrate that anti-Robin Hood is uniquely suited to achieving optimal query and insertion times with quadratic probe growth. The striking negative result for Robin Hood ordering reveals deep intrinsic differences from the linear probing analogy. The main technical results establish that almost all degree-$h(x), h(x) + 1^2, h(x) + 2^2, \ldots$7 (or higher) fixed-offset quadratic schemes, when using anti-Robin Hood ordering, are optimal for all admissible load factors, essentially closing the problem for a large class of hash table designs.

The tools and conclusions in this work will influence both future theoretical investigations and practical high-performance hash table implementations. The open question regarding the classical quadratic probing sequence remains, but the demonstrated optimality of almost all related schemes provides a compelling foundation for modern hash table algorithmics.