arXiv subjects

Sergey Volkov

Publications and source records attributed to Sergey Volkov.

18 recordsLinked to original sources

arXiv2026

From Evidence to Effect: Authority Semantics and Runtime Infrastructure for Stateful Agents

Stateful agents reuse artifacts after producing executions and permissions change. We formalize authority-sufficient observations and durable effects bound to execution and material identities. WTB implements this interface through runtime adapters, shared evidence, and transactional publication/recovery. Six study families separate the mechanism from its integration. Raw and typed evidence both solve 32/32 authority cases, with model-dependent planning effects. Fixed-intent enforcement blocks six unsafe proposals and executes 12 eligible authorized intents. Complete controls match WTB's capability. Paid integration yields 176/210 accepted benchmark-source stages, including 19/30 publication stages, recovery on 8/8 primary SWE repositories, and the most complete continuous trajectories on each of three source tasks. The findings connect authority information, effect admission, and infrastructure reuse in stateful agents.

cs.MA↗

arXiv2026

ASCT: Attentive Search over Counterfactual Trees for Credit Assignment in Agentic Reinforcement Learning

Terminal utility evaluates a complete agentic workflow, but learning requires credit for the decisions within it. We introduce Attentive Search over Counterfactual Trees (ASCT), a framework that turns training-time multi-step search into local action credit. At actor-visited states, an auxiliary tree evaluates alternative legal actions from the same recoverable prefix. Its action-value table is centered by the frozen actor's probabilities and supplies credit for PPO on actor-sampled trajectories. This protocol connects counterfactual evaluation to policy learning while deploying the actor alone. Uniform, UCT, and cost-aware AgentUCT instantiate the framework. On HotpotQA agentic retrieval-augmented generation, all three improve mean held-out utility over trajectory-return PPO and workflow-adapted VinePPO. Across three seeds, ASCT-AgentUCT reaches 0.6187 utility versus 0.5939 for VinePPO, with gains in answer F1 and execution cost, and uses 50.3% fewer recorded auxiliary Qwen tokens. Transfer and component-description studies examine the learned policies beyond the training setting.

cs.AI↗

arXiv2026

Agent-UCT: Upper Confidence Bounds Applied to Trees for Agentic Workflow Optimization with Cost-Awareness

Optimizing agentic workflows, such as retrieval-augmented generation (RAG) pipelines, requires navigating a combinatorial space of discrete component choices under tight evaluation budgets. Existing approaches - heuristic search, black-box optimization, and standard tree search methods - do not explicitly exploit the compositional structure of these workflows, leading to redundant computation and inefficient budget allocation. We introduce Agent-UCT (Agent-based Cost-Aware Upper Confidence Bounds Applied to Trees), a tree search algorithm that extends UCT with a reuse-aware regularization term derived from a bipartite prefix reuse graph. Agent-UCT biases selection toward branches that leverage previously materialized configuration prefixes, reducing redundant execution while maintaining effective exploration. Our framework, RAGSpace, unifies heterogeneous RAG components from LongRAG, LightRAG, and Self-RAG into a five-dimensional configuration space, enabling systematic cross-framework recombination. WTB (Workflow Test Bench) provides deterministic replay, content-addressable caching, and transactional consistency, ensuring that intermediate states are materialized once and reused across the search. Experiments on HotpotQA and UltraDomain demonstrate that Agent-UCT identifies configurations with the highest out-of-sample performance among the evaluated fixed framework presets. Under full-pool evaluation, bipartite prefix reuse reduces logical search cost by 73.6% relative to the no-prefix-sharing cost upper bound. Compared with full-pool evaluation, sampling-based evaluation further achieves a 4.2x wall-clock speedup. Agent-UCT, RAGSpace, and WTB together provide a unified framework for cost-aware, reproducible, and compositionally efficient agentic workflow optimization.

cs.AI↗

arXiv2026

StateFuse: Deterministic Conflict-Preserving Memory for Multi-Agent Systems

Agent systems accumulate conflicting observations across branches, retries, and replicas, yet many practical memory layers still collapse disagreement behind overwrite rules that are difficult to inspect or correct. We present StateFuse, a conflict-aware replicated memory contract built on standard OpSet/CRDT merge. StateFuse does not introduce a new join algebra; it defines an agent-facing semantics layer with immutable history, explicit conflict objects, exact and semantic correction handles (claim_id / claim_ref), deterministic predicate contracts, and projection-time resolution that cannot rewrite replicated state. We evaluate StateFuse against flat multi-value, raw-log, provenance-style, and collapsed baselines under matched resolver and verification policies. On a 282-question official conflict-bearing MemoryAgentBench slice, the compared methods tie on answer accuracy, but conflict-preserving surfaces keep contradictions visible while collapsed surfaces do not. In a controlled agent loop with uniform verification, preserving ambiguity enables safer abstention and correction than early collapse. A correction-handle ablation further shows that semantic handles matter when exact prior identifiers are unavailable. The resulting claim is narrow: StateFuse is best supported as a safer public memory contract for contradiction surfacing, abstention, and auditable correction, not as a universal accuracy gain.

cs.AI↗

arXiv2026

QED vacuum polarization in the Coulomb field of a nucleus: a method of high-order calculation

A calculation of the QED vacuum polarization potential in the Coulomb field of a pointlike nucleus was presented in an earlier publication by the author and his collaborators. Corrections up to order $α^2 (Zα)^7$ were evaluated, where $Z$ is the nuclear charge number and $Zα$ is treated as an independent variable. These corrections correspond to two-loop Feynman graphs with proper propagators of fermions in the external field. The calculation employed a reduction to free QED, leading to free QED Feynman graphs with up to eight independent loops. The method of calculation is described here in detail.

physics.atom-ph↗

arXiv2024

Calculation of the total 10-th order QED contribution to the electron magnetic moment

The total 5-loop quantum electrodynamics universal contribution to the anomalous magnetic moments of the leptons was calculated by the author. The obtained value $A_1^{(10)}=5.891(61)$ provides the first complete verification of the previously known value obtained by T. Aoyama, M. Hayakawa, T. Kinoshita, M. Nio (AHKN). The discrepancy is $5 σ$. The computation includes the recalculation of the part that the author calculated in 2019 using a slightly different method and the calculation of the remaining part of the coefficient. A comparison with the AHKN values in 32 gauge-invariant classes is provided. In addition, the results are divided into 95 small gauge-invariant classes that subdivide the former ones. Such a detailzation is provided for the first time. The method described in previous works of the author was used in general. However, the Monte Carlo integration method is new and is described in detail. Some useful technical information and information on numerical cancellations is also given.

hep-ph↗

arXiv2023

Calculation of lepton magnetic moments in quantum electrodynamics: a justification of the flexible divergence elimination method

The flexible method of reduction to finite integrals, briefly described in earlier publications of the author, is described in detail. The method is suitable for the calculation of all quantum electrodynamical contributions to the magnetic moments of leptons. It includes mass-dependent contributions. The method removes all divergences (UV, IR and mixed) point-by-point in Feynman parametric space without any usage of limit-like regularizations. It yields a finite integral for each individual Feynman graph. The subtraction procedure is based on the use of linear operators applied to the Feynman amplitudes of UV-divergent subgraphs; a placement of all terms in the same Feynman parametric space is implied. The final result is simply the sum of the individual graph contributions; no residual renormalization is required. The method also allows us to split the total contribution into the contributions of small gauge-invariant classes. The procedure offers a great freedom in the choice of the linear operators. This freedom can be used for improving the computation speed and for a reliability check. The mechanism of divergence elimination is explained, as well as the equivalence of the method and the on-shell renormalization. For illustrative purposes, all 4-loop contributions to the anomalous magnetic moments of the electron and muon are given for each small gauge-invariant class, as well as their comparison with previously known results. This also includes the contributions that depend on the ratios of the tau-lepton mass to the electron and muon mass.

hep-ph↗

arXiv2022

A Flexible Divergence Elimination Method for Calculating Lepton Magnetic Moments in Quantum Electrodynamics

A precise calculation of the lepton anomalous magnetic moments (AMM) requires an evaluation of the quantum electrodynamics (QED) Feynman diagrams up to five independent loops. The complicated structure of ultraviolet (UV), infrared (IR) and mixed divergences in the corresponding integrals makes it difficult to calculate these high-order contributions in reasonable computer time frame. We demonstrate a method that eliminates all divergences point by point in Feynman parametric space (before integration) and possesses a flexibility that can be used for improving the precision of the numerical integration. This flexibility is especially actual for calculating the contributions of the Feynman diagrams with electron loops to the muon AMM. 3-loop and 4-loop numerical test results are provided. The subtraction procedure is based on a forest formula with linear operators applied to the Feynman amplitudes of UV divergent subdiagrams. It is similar to BPHZ; the difference is in the linear operators used and in the way of combining them. It is equivalent to the on-shell renormalization after summation over diagrams: no residual renormalization is required.

hep-ph↗

arXiv2021

A Way of Fast Calculating Lepton Magnetic Moments in Quantum Electrodynamics

A new method of divergence subtraction in Feynman parametric integrals is presented. The method is suitable for calculating the lepton anomalous magnetic moments (AMM) in quantum electrodynamics (QED). The subtraction procedure eliminates all divergences before integration and leads to a finite Feynman parametric integral for each individual Feynman diagram. It is based on a forest formula with linear operators applied to the Feynman amplitudes of ultraviolet-divergent subdiagrams. The formula is similar to BPHZ; the difference is only in the linear operators used and in the way of combining them. The subtraction is equivalent to the on-shell renormalization from the beginning: for obtaining the final result we should only sum up the contributions of all Feynman diagrams after subtraction. The developed method is an improvement of the method presented by the author in 2016. The modification is specifically designed for calculating the contributions dependent on the relations of particle masses. In comparison with the old version, the new subtraction formula does not contain redundant terms and possesses some flexibility that can be used for improving the precision of calculations. Numerical test results are presented up to four loops.

hep-ph↗

arXiv2020

Infrared and Ultraviolet Power Counting on the Mass Shell in Quantum Electrodynamics

A power counting rule is provided that allows us to obtain upper bounds for the absolute values of Feynman parametric integrands. The rule reflects both the ultraviolet and infrared behavior taking into account that the external momenta are on the mass shell. It gives us the ability to rigorously prove the absolute convergence of the corresponding integrals. The consideration is limited to the case of the quantum electrodynamics Feynman graphs contributing to the lepton magnetic moments and not containing either lepton loops or ultraviolet divergent subgraphs. However, a rigorous mathematical proof is given for all Feynman graphs satisfying these restrictions. The power counting rule is formulated in terms of Hepp's sectors, ultraviolet degrees of divergence and so-called I-closures. The obtained upper bound can not be substantially improved: the illustrative example is provided. The paper provides the first mathematically rigorous treatment of the ultraviolet behavior together with the on-shell infrared behavior with some kind of generality. Practical applications of this rule are explained.

hep-th↗

arXiv2019

Calculating the five-loop QED contribution to the electron anomalous magnetic moment: Graphs without lepton loops

This paper describes a computation of a part of the QED contribution to the electron anomalous magnetic moment that was performed by the author with the help of a supercomputer. The computed part includes all 5-loop QED Feynman graphs without lepton loops. The calculation has led to the result $A_1^{(10)}[\text{no lepton loops}]=6.793(90)$ that is slightly different than the value $7.668(159)$ presented by T. Aoyama, T. Kinoshita, and M. Nio in 2018. The discrepancy is about $4.8σ$. The computation gives the first independent check for that value. A shift in the fine-structure constant prediction is revealed in the paper. The developed calculation method is based on (a) a subtraction procedure for removing all ultraviolet and infrared divergences in Feynman parametric space before integration; (b) a nonadaptive Monte Carlo integration that uses the probability density functions that are constructed for each Feynman graph individually using its combinatorial structure. The method is described briefly in the paper (with the corresponding references to the previous papers). The values for the contributions of nine gauge-invariant classes splitting the whole set are presented in the paper. Moreover, the whole set of all 5-loop graphs without lepton loops is split into 807 subsets for comparison (in the future) of the calculated values with the values obtained by another methods. These detailed results are presented in the supplemental materials. Also, the supplemental materials contain the contribution values for each of 3213 individual Feynman graphs. An "oscillating" nature of these values is discussed. Technical details of the realization are described.

hep-ph↗

arXiv2019

Numerical calculation of 5-loop QED contributions to the electron anomalous magnetic moment

Preliminary numerical results of calculating the 5-loop QED contributions to the electron (and muon) anomalous magnetic moment are presented. The results include the total contribution of the 5-loop Feynman diagrams without lepton loops and the contributions of nine gauge-invariant classes that form that set. A discrepancy with known results is revealed. The contributions of the gauge-invariant classes are presented for the first time. The method of the computation is briefly described (with the corresponding references). The calculation is based on the following elements: 1) a subtraction procedure for removing infrared (IR) and ultraviolet (UV) divergences point by point in Feynman parametric representation before integration; 2) a nonadaptive Monte Carlo integration method that is founded on probability density functions (PDF) that are constructed for each Feynman diagram individually using its combinatorial structure; 3) a GPU-based numerical integration with the help of the supercomputer "Govorun" (JINR, Dubna, Russia).

hep-ph↗

arXiv2018

Numerical calculation of high-order QED contributions to the electron anomalous magnetic moment

This paper describes a method of numerical evaluating high-order QED contributions to the electron anomalous magnetic moment. The method is based on subtraction of infrared and ultraviolet divergences in Feynman-parametric space before integration and on nonadaptive Monte Carlo integration that is founded on Hepp sectors. A realization of the method on the graphics accelerator NVidia Tesla K80 is described. A method of removing round-off errors that emerge due to numerical subtraction of divergences without losing calculation speed is presented. The results of applying the method to all 2-loop, 3-loop, 4-loop QED Feynman graphs without lepton loops are presented. A detailed comparison of the 2-loop and 3-loop results with known analytical ones is given in the paper. A comparison of the contributions of 6 gauge invariant 4-loop graph classes with known analytical values is presented. Moreover, the contributions of 78 sets of 4-loop graphs for comparison with the direct subtraction on the mass shell are presented. Also, the contributions of the 5-loop and 6-loop ladder graphs are given as well as a comparison of these results with known analytical ones. The behavior of the generated Monte Carlo samples is described in detail, a method of the error estimation is presented. A detailed information about the graphics processor performance on these computations and about the Monte Carlo convergence is given in the paper.

hep-ph↗

arXiv2017

New method of computing the contributions of graphs without lepton loops to the electron anomalous magnetic moment in QED

This paper presents a new method of numerical computation of the QED contributions to the electron anomalous magnetic moment which arises from Feynman graphs containing electron and photon lines and not containing electron loops. The method consists of two parts. The first part is the forest-like subtraction formula that removes all ultraviolet and infrared divergences in each Feynman graph point-by-point in Feynman-parametric space. The second part is the importance sampling Monte-Carlo algorithm with the probability density function that is constructed for each Feynman graph individually. The method is fully automated at any order of the perturbation series. The results of applying the method to 2-loop, 3-loop, 4-loop Feynman graphs and to some individual 5-loop graphs are presented.

hep-ph↗

arXiv2017

Subtraction Procedure for Calculation of Anomalous Magnetic Moment of Electron in QED and its Application to Numerical Computation at 3-loop Level

A new subtraction procedure for removal both ultraviolet and infrared divergences in Feynman integrals is proposed. This method is developed for computation of QED corrections to the electron anomalous magnetic moment. The procedure is formulated in the form of a forest formula with linear operators that are applied to Feynman amplitudes of UV-divergent subgraphs. The contribution of each Feynman graph that contains propagators of electrons and photons is represented as a finite Feynman-parametric integral. Application of the developed method to the calculation of 2-loop and 3-loop contributions is described.

hep-ph↗

arXiv2016

Finite Bases with Respect to the Superposition in Classes of Elementary Recursive Functions, dissertation

This is a thesis that was defended in 2009 at Lomonosov Moscow State University. In Chapter 1: 1. It is proved that that the class of lower (Skolem) elementary functions is the set of all polynomial-bounded functions that can be obtained by a composition of $x+1$, $xy$, $\max(x-y,0)$, $x\wedge y$, $\lfloor x/y \rfloor$, and one exponential function ($2^x$ or $x^y$) using formulas that have no more than 2 floors with respect to an exponent (for example, $(x+y)^{xy+z}+1$ has 2 floors, $2^{2^x}$ has 3 floors). Here $x\wedge y$ is a bitwise AND of $x$ and $y$. 2. It is proved that $\{x+y,\ \max(x-y,0),\ x\wedge y,\ \lfloor x/y \rfloor,\ 2^{\lfloor \log_2 x \rfloor^2}\}$ and $\{x+y,\ \max(x-y,0),\ x\wedge y,\ \lfloor x/y \rfloor,\ x^{\lfloor \log_2 y \rfloor}\}$ are composition bases in the functional version of the uniform $\mathrm{TC}^0$ (also known as $\mathrm{FOM}$). 3. The hierarchy of classes exhausting the class of elementary functions is described in terms of compositions with restrictions on a number of floors in a formula. The results of Chapter 1 are published in: 1) Volkov S.A. An exponential expansion of the Skolem-elementary functions, and bounded superpositions of simple arithmetic functions (in Russian), Mathematical Problems of Cybernetics, Moscow, Fizmatlit, 2007, vol. 16, pp. 163-190 2) doi:10.1134/S1064562407040217 In Chapter 2 a simple composition basis in the class ${\cal E}^2$ of Grzegorczyk hierarchy is described. This result is published in DOI: 10.1515/156939206779238436 In Chapter 3 it is proved that the group of permutations $\mathrm{Gr}(Q)=\{f:\ f,f^{-1}\in Q\}$ is generated by two permutations for many classes $Q$. For example, this is proved for $Q=\mathrm{FP}$, where $\mathrm{FP}$ is the class of all polynomial-time computable functions (of the length of input). The results of chapter 3 are published in DOI: 10.1515/DMA.2008.046

cs.CC↗

arXiv2008

Improved Monotone Circuit Depth Upper Bound for Directed Graph Reachability

We prove that the directed graph reachability problem (transitive closure) can be solved by monotone fan-in 2 boolean circuits of depth (1/2+o(1))(log n)^2, where n is the number of nodes. This improves the previous known upper bound (1+o(1))(log n)^2. The proof is non-constructive, but we give a constructive proof of the upper bound (7/8+o(1))(log n)^2.

cs.CC↗

arXiv2007

P-adic arithmetic coding

A new incremental algorithm for data compression is presented. For a sequence of input symbols algorithm incrementally constructs a p-adic integer number as an output. Decoding process starts with less significant part of a p-adic integer and incrementally reconstructs a sequence of input symbols. Algorithm is based on certain features of p-adic numbers and p-adic norm. p-adic coding algorithm may be considered as of generalization a popular compression technique - arithmetic coding algorithms. It is shown that for p = 2 the algorithm works as integer variant of arithmetic coding; for a special class of models it gives exactly the same codes as Huffman's algorithm, for another special model and a specific alphabet it gives Golomb-Rice codes.

cs.DS↗