Hard-to-Sample Distributions from Robust Extractors

Abstract: We provide a unified method for constructing explicit distributions which are difficult for restricted models of computation to generate. Our constructions are based on a new notion of robust extractors, which are extractors that remain sound even when a small number of points violate the min-entropy constraint. Using such objects, we show that for a broad range of sampling models (e.g., low-depth circuits, small-space sources, etc.), every output of the model has distance $1 - o(1)$ from our target distribution, qualitatively recovering essentially all previously known hardness results. Our work extends that of Viola (SICOMP '14), who developed an earlier unified framework based on traditional extractors to rule out sampling with very small error. As a further application of our technique, we leverage a recent extractor construction of Chattopadhyay, Goodman, and Gurumukhani (ITCS '24) to present the first explicit distribution with distance $1 - o(1)$ from the output of any low-degree $\mathbb{F}_2$-polynomial source. We also describe a potential avenue toward proving a similar hardness result for $\mathsf{AC^0}[\oplus]$ circuits.

• The paper establishes that robust extractors enable explicit distributions with total variation distance approaching 1, evidencing strong sampling hardness.
• It unifies prior hardness results by combining robust extractor properties with direct product techniques across various restricted computational models.
• An explicit construction for low-degree polynomial sources is provided, breaking previous bounds and advancing explicit sampling lower bounds.

Robust Extractors and Explicit Sampling Hardness

Overview and Context

The paper "Hard-to-Sample Distributions from Robust Extractors" (2604.26179) addresses unconditional sampling lower bounds for restricted computational models such as low-depth circuits, small-space sources, and communication protocols. The study focuses on the explicit construction of distributions that are provably hard to sample (i.e., cannot be closely approximated in total variation distance) by a broad class of such models. The proposed approach leverages a new notion—robust extractors—which, unlike traditional randomness extractors, maintain their pseudorandomness guarantees even when a small fraction of the domain violates the min-entropy condition. The authors establish a unified framework for explicit hardness-of-sampling results and extend them to new territory, notably low-degree $\mathbb{F}_2$-polynomial sources.

Technical Contributions

Robust Extractors: Definition and Role

A robust extractor $rExt\colon \{0,1\}^n \to \{0,1\}^m$ for a class $\mathcal{X}$ of distributions over $\{0,1\}^n$ satisfies not only the traditional extractor property (outputs close to uniform for sources with large min-entropy), but also a robustness property: the probability that a "light" point (a domain element with probability at most $2^{-k}$) maps to a specific output (e.g., $0^m$) is tightly controlled and cannot significantly exceed $2^{-m}$. This stronger requirement enables sharper separation from all distributions generated by various restricted samplers, even when the input sources slightly deviate from high min-entropy.

The paper introduces and formalizes isolators, functions constructed from robust extractors, which serve as witnesses to sampling hardness by controlling the mass assigned to light points.

Unified Sampling Lower Bounds

The main technical theorem establishes that given a sufficiently robust isolator/robust extractor for a source class, one can construct explicit distributions $D$ such that, for all samplers (e.g., depth-bounded circuits, polynomial sources, etc.), every output distribution is at total variation distance $1-o(1)$ from $D$ as parameters grow. The framework amplifies constant lower bounds to near-optimal bounds by combining robust extractors with direct product arguments and multi-bit outputs.

The authors demonstrate that most previously known sampling lower bounds for a wide array of models (local circuits, $rExt\colon \{0,1\}^n \to \{0,1\}^m$0 circuits, communication sources, small space sources, Turing machine sources) can be qualitatively recovered in this unified robust extractor framework. For these classes, explicit instances of robust extractors are constructed by modifying established extractor results, often utilizing brute-force over bounded input length or leveraging input reduction theorems.

Explicit Hardness for Low-Degree Polynomial Sources

One major advance is the first explicit distribution $rExt\colon \{0,1\}^n \to \{0,1\}^m$1 with TV distance $rExt\colon \{0,1\}^n \to \{0,1\}^m$2 from any low-degree $rExt\colon \{0,1\}^n \to \{0,1\}^m$3-polynomial source. Prior to this work, explicit distributions could only guarantee $rExt\colon \{0,1\}^n \to \{0,1\}^m$4 distance from such sources, even as classical lower bounds for function computation were long established. The constructions rely on robustifying recent polynomial extractor results [chattopadhyay2024extractors], then applying input length reduction and entropy smoothing arguments.

Extensions and Open Directions

The paper details several important open questions:

• Exhibiting explicit distributions with $rExt\colon \{0,1\}^n \to \{0,1\}^m$5 hardness for $rExt\colon \{0,1\}^n \to \{0,1\}^m$6 circuits remains unresolved. No explicit distributions are currently known to be hard for such circuits beyond certain trivial bounds.
• Improving the quantitative decay rates (e.g., exponentially small error terms) for explicit hardness against polynomial sources.
• Developing stronger notions than robust extractors (such as smooth extractors) to further amplify or generalize sampling hardness.

Strong Numerical and Structural Claims

• Total Variation Distance Amplification: For all considered models, explicit distributions are constructed for which every sampler's output distribution is at TV distance approaching 1 (e.g., $rExt\colon \{0,1\}^n \to \{0,1\}^m$7 for degree-$rExt\colon \{0,1\}^n \to \{0,1\}^m$8 $rExt\colon \{0,1\}^n \to \{0,1\}^m$9-polynomial sources).
• Unified Hardness: The robust extractor framework qualitatively matches or nearly matches prior separate sampling lower bounds for circuit sources, local sources, communication sources, small-space sources, and Turing-machine sources, unifying them in a single explicit construction paradigm.

Practical and Theoretical Implications

The results have implications across complexity theory, pseudorandomness, coding theory, streaming algorithms, and quantum computational separations:

• Data Structure Lower Bounds: Explicit hardness-to-sample distributions translate to lower bounds for high-performance data structures and streaming algorithms, by preventing efficient simulation of certain distributions in restricted computational resources.
• Quantum Advantage: Sampling lower bounds are tightly linked to quantum supremacy arguments, as certain quantum processes can efficiently sample distributions that are classically hard.
• Learning Theory and Codes: Hard-to-sample distributions provide separation against shallow circuit learners and yield explicit codes that cannot be locally sampled by $\mathcal{X}$0 circuits.
• Constructive Techniques: The robust extractor paradigm introduces practical ways to design and analyze pseudorandom generators and extractors for a variety of source classes.

Future Directions

Further progress will likely depend on new constructive techniques for isolators and robust/smooth extractors, potentially enabling hardness results for more powerful circuit classes (e.g., $\mathcal{X}$1 or $\mathcal{X}$2) and attaining optimal quantitative bounds. Improved input reduction and entropy smoothing methods may yield tighter uniform explicit hardness.

Conclusion

This paper introduces robust extractors as versatile tools for explicit sampling hardness, providing a unified framework for strong total variation distance lower bounds across a spectrum of restricted sampling models and pioneering explicit separation results for low-degree polynomial sources. The work repositions extractors as central objects in the analysis of computationally constrained sampling, both by synthesizing prior results and by breaking new ground. Future research on robust extractors and their extensions promises significant advances in complexity-theoretic separations and pseudorandomness theory.