Low-discrepancy point sets and digital sequences underpin quasi-Monte Carlo (QMC) methods for high-dimensional integration. We cast two long-standing QMC design problems as program synthesis and solve them with an LLM-guided evolutionary loop that mutates and selects code under task-specific fitness: (i) constructing finite 2D/3D point sets with low star discrepancy, and (ii) choosing Sobol' direction numbers that minimize randomized QMC error on downstream integrands. Our two-phase procedure combines constructive code proposals with iterative numerical refinement. On finite sets, we rediscover known optima in small 2D cases and set new best-known 2D benchmarks for N >= 40, while matching most known 3D optima up to the proven frontier (N <= 8) and reporting improved 3D benchmarks beyond. On digital sequences, evolving Sobol' parameters yields consistent reductions in randomized quasi-Monte Carlo (rQMC) mean-squared error for several 32-dimensional option-pricing tasks relative to widely used Joe--Kuo parameters, while preserving extensibility to any sample size and compatibility with standard randomizations. Taken together, the results demonstrate that LLM-driven evolutionary program synthesis can automate the discovery of high-quality QMC constructions, recovering classical designs where they are optimal and improving them where finite-N structure matters. Data and code are available at https://github.com/hockeyguy123/openevolve-star-discrepancy.git.
Low-discrepancy point sets and digital sequences underpin quasi-Monte Carlo (QMC) methods for high-dimensional integration. We cast two long-standing QMC design problems as program synthesis and solve them with an LLM-guided evolutionary loop that mutates and selects code under task-specific fitness: (i) constructing finite 2D/3D point sets with low star discrepancy, and (ii) choosing Sobol' direction numbers that minimize randomized QMC error on downstream integrands. Our two-phase procedure combines constructive code proposals with iterative numerical refinement. On finite sets, we rediscover known optima in small 2D cases and set new best-known 2D benchmarks for N >= 40, while matching most known 3D optima up to the proven frontier (N <= 8) and reporting improved 3D benchmarks beyond. On digital sequences, evolving Sobol' parameters yields consistent reductions in randomized quasi-Monte Carlo (rQMC) mean-squared error for several 32-dimensional option-pricing tasks relative to widely used Joe--Kuo parameters, while preserving extensibility to any sample size and compatibility with standard randomizations. Taken together, the results demonstrate that LLM-driven evolutionary program synthesis can automate the discovery of high-quality QMC constructions, recovering classical designs where they are optimal and improving them where finite-N structure matters. Data and code are available at https://github.com/hockeyguy123/openevolve-star-discrepancy.git.
Quasi-Monte Carlo (QMC) methods are known to achieve faster convergence rates than Monte Carlo (MC), but their effectiveness in high dimensions often relies on additional structure, such as low effective dimension or carefully chosen coordinate weights. Moreover, in many applications one has access only to random samples rather than deterministic QMC constructions. In this work, we extend the recent method of N. Bansal and H. Jiang and construct high-dimensional QMC point sets from random samples via combinatorial discrepancy. We establish error bounds for these constructions in weighted function spaces, including settings with low effective dimension in both the superposition and truncation senses. We also implement the resulting Subgaussian Transference algorithm and present numerical experiments to assess empirically the performance of these constructions.
The quasi-Monte Carlo (QMC) methods use deterministic, usually low-discrepancy sequences, to improve the order of convergence when compared with Monte Carlo methods dealing with the same problem. Although this improved convergence depends on the smoothness properties of the functions under consideration and sometimes cannot be established theoretically, good practical results were obtained in many areas, notably in Financial Mathematics. Most of the QMC methods do not depend on particular properties of the sequences used beyond their good distribution in a suitable multidimensional space. Many families of low-discrepancy sequences have been studies, with the family of the Sobol sequences being not only one of the oldest but also the most popular in practical applications. Since these sequences have many free parameters, it is desirable to have methods to obtain a good set of parameters, suitable for a wide class of problems. In this work we propose a fully deterministic method for optimisation of the direction numbers of the Sobol sequences that is flexible in taking into account smoothness properties of the potential sub-integral functions. We demonstrate how by implementing this algorithm on modern GPUs it is possible to obtain good sets of direction numbers within a reasonable timeframe. The numerical tests demonstrate the significant advantage in terms of accuracy that is observed, which makes the corresponding QMC methods even more competitive.
We study the numerical approximation of high-dimensional integrals using randomized quasi-Monte Carlo (RQMC) methods, with a focus on scrambled Sobol' sequences. While asymptotically faster than Monte Carlo, classical RQMC suffers from error bounds that grow exponentially in the dimension s, and its root mean squared error (RMSE) convergence rate is generally no better than O(N−3/2). To overcome these limitations, we combine two recent developments: Suzuki's coarse scrambling and the median trick. Coarse scrambling randomizes Sobol' sequences according to their generating base polynomials and significantly reduces the maximal gain coefficient. By appropriately choosing the base polynomials, we show that the coefficient can be made uniformly bounded in s, yielding an O(N−1/2) RMSE for L2 integrands with a dimension-independent constant. The median trick then enables near-optimal convergence for function classes beyond L2: for Lp integrands with p∈(1,2), the median of independent coarsely scrambled estimates achieves an error of O(N−1+1/p) with high probability; for integrands in the Haar wavelet space Hwav,α,s,p,q with αp:=α−(1/p−1/2)+>0, we prove a high-probability error bound of O(N−αp−1/2+ε) for any ε>0; the bound improves to O(N−r−αp−1/2+ε) when the integrand has dominating mixed derivatives of order r≥1 that belong to Hwav,α,s,p,q. The latter two bounds are uniform in s under suitable conditions on the ANOVA components of the integrands. Numerical experiments confirm the predicted convergence rates and demonstrate the robustness of the proposed approach.
We survey basic ideas and results on randomized quasi-Monte Carlo (RQMC) methods, discuss their practical aspects, and give numerical illustrations. RQMC can improve accuracy compared with standard Monte Carlo (MC) when estimating an integral interpreted as a mathematical expectation. RQMC estimators are unbiased and their variance converges at a faster rate (under certain conditions) than MC estimators, as a function of the sample size. Variants of RQMC also work for the simulation of Markov chains, for function approximation and optimization, for solving partial differential equations, etc. In this introductory survey, we look at how RQMC point sets and sequences are constructed, how we measure their uniformity, why they can work for high-dimensional integrals, and how can they work when simulating Markov chains over a large number of steps.
Monte Carlo (MQ) method is a powerful tool to approximate high dimensional integrals. The disadvantage of ordinary MQ is the slow convergence rate by cause of the essential randomness of this method. Computations of this convergence can lead to pure time consuming. Quasi-Monte Carlo (QMC) method yields considerably better results since this method is a deterministic alternative which uses Low-Discrepancy sequences instead of random samples . There are a multiplicity of big open problems in QMC-methods, problems partly arising from applications and partly arising from theory. QMC are developed to integrate over unite cube, where it has much more accuracy than MC for integrands of bounded variation. Integration over more general spaces such as triangles, disks and Cartesian products of such spaces is more challenging for QMC. Nevertheless in real-world applications various problems are defined over such spaces. The aim of this thesis is to provide a survey of a solution of such problems of numerical integration defined over non-cubical spaces. We present QMC and randomised QMC (RQMC) constructions in the triangle with a vanishing discrepancy based on the recent work of Basu 2016. The QMC construction is a version of the Van der Corput-Halton sequences specially made to the unit triangle. The attraction of scrambled net is replication based error estimation for QMC with slightly the same accuracy as QMC, and for smooth enough integrands.
Quantum LDPC code discovery requires searching large algebraic design spaces while reliably certifying the parameters and equivalence classes of any candidates found. We introduce an LLM-guided evolutionary workflow in which language models mutate Python programs that generate bivariate-bicycle and perturbed bivariate-bicycle code ansätze. Across five campaigns, the system performed approximately 1{,}650 evolutionary iterations, screened about 2×105 candidate codes, and required ∼140 hours of computation and ∼US$400 in LLM inference cost. Candidate codes are evaluated through a staged validation pipeline combining GF(2) rank computation, distance estimation and certification, mixed-integer linear programming, BLISS Tanner-graph deduplication, decomposability analysis, and local-Clifford equivalence checks. At block length n≤360, the workflow identifies 465 distinct candidate codes: 97 CSS bivariate-bicycle codes and 368 non-CSS perturbed variants. The CSS search recovers known high-performing codes and finds new finite-length representatives, including an indecomposable [[288,16,12]] code and higher-weight codes with up to k=50 at distance d=8. The non-CSS search produces perturbed codes matching the gross-code figure of merit at [[144,12,12]], along with additional high-distance candidates reported as certified values or upper bounds according to MILP status. Overall, these results show that LLM-guided program evolution can serve as a practical tool for structured quantum-code discovery when paired with independent evaluation.