Mingji Yang (杨铭基)

Authors are ordered alphabetically with equal contributions by default.

1. ITCS 2026

   On Solving Asymmetric Diagonally Dominant Linear Systems in Sublinear Time

   Tsz Chiu Kwok, Zhewei Wei, and Mingji Yang

   In Proceedings of the 17th Innovations in Theoretical Computer Science Conference, Jan 2026

   TLDR Abstract BibTex arXiv Video Slides

   We initiate a study of solving row/column diagonally dominant (RDD/CDD) linear systems in sublinear time, which includes estimating (Personalized) PageRank and effective resistance on graphs as special cases. We characterize the problem's mathematical structure via a novel quantity called the maximum \(p\)-norm gap, and develop a collection of algorithmic results by adapting techniques from local graph algorithms.

   We initiate a study of solving a row/column diagonally dominant (RDD/CDD) linear system \(\mathbf{M}\boldsymbol{x} = \boldsymbol{b}\) in sublinear time, with the goal of estimating \(\boldsymbol{t}^{\top}\boldsymbol{x}^{\ast}\) for a given vector \(\boldsymbol{t} \in \mathbb{R}^n\) and a specific solution \(\boldsymbol{x}^{\ast}\). This setting naturally generalizes the study of sublinear-time solvers for symmetric diagonally dominant (SDD) systems [Andoni-Krauthgamer-Pogrow, ITCS 2019] to the asymmetric case, which has remained underexplored despite extensive work on nearly-linear-time solvers for RDD/CDD systems.
   Our first contributions are characterizations of the problem's mathematical structure. We express a solution \(\boldsymbol{x}^{\ast}\) via a Neumann series, prove its convergence, and upper bound the truncation error on this series through a novel quantity of \(\mathbf{M}\), termed the maximum \(p\)-norm gap. This quantity generalizes the spectral gap of symmetric matrices and captures how the structure of \(\mathbf{M}\) governs the problem's computational difficulty.
   For systems with bounded maximum \(p\)-norm gap, we develop a collection of algorithmic results for locally approximating \(\boldsymbol{t}^{\top}\boldsymbol{x}^{\ast}\) under various scenarios and error measures. We derive these results by adapting the techniques of random-walk sampling, local push, and their bidirectional combination, which have proved powerful for special cases of solving RDD/CDD systems, particularly estimating PageRank and effective resistance on graphs. Our general framework yields deeper insights, extended results, and improved complexity bounds for these problems. Notably, our perspective provides a unified understanding of Forward Push and Backward Push, two fundamental approaches for estimating random-walk probabilities on graphs.
   Our framework also inherits the hardness results for sublinear-time SDD solvers and local PageRank computation, establishing lower bounds on the maximum \(p\)-norm gap or the accuracy parameter. We hope that our work opens the door for further study into sublinear solvers, local graph algorithms, and directed spectral graph theory.

   @inproceedings{kwok2026solving,
     author       = {Tsz Chiu Kwok and
                     Zhewei Wei and
                     Mingji Yang},
     title        = {On solving asymmetric diagonally dominant linear systems in sublinear time},
     booktitle    = {Proceedings of the 17th Innovations in Theoretical Computer Science Conference},
     pages        = {89:1--89:25},
     year         = {2026},
     doi          = {10.4230/LIPIcs.ITCS.2026.89}
   }
     

2. SODA 2026

   PageRank Centrality in Directed Graphs with Bounded In-Degree

   Mikkel Thorup, Hanzhi Wang, Zhewei Wei, and Mingji Yang

   In Proceedings of the 2026 Annual ACM-SIAM Symposium on Discrete Algorithms, Jan 2026

   TLDR Abstract BibTex arXiv

   For estimating single-node PageRank on directed graphs, if we parameterize the complexity by the maximum in-degree \(\Delta_{\mathrm{in}}\) of the graph, the upper bound in our STOC 2024 paper is optimal only when \(\Delta_{\mathrm{in}} = n^{\Omega(1)}\). In this paper, we improve the upper bound to match the lower bound (up to polylog factors) for all regimes of \(\Delta_{\mathrm{in}}\). Our key technique is a bidirectional algorithm that employs a novel randomized local exploration method.

   We study the computational complexity of locally estimating a node's PageRank centrality in a directed graph \(G\). For any node \(t\), its PageRank centrality \(\pi(t)\) is defined as the probability that a random walk in \(G\), starting from a uniformly chosen node, terminates at \(t\), where each step terminates with a constant probability \(\alpha \in (0,1)\).
   To obtain a multiplicative \(\bigl(1 \pm O(1)\bigr)\)-approximation of \(\pi(t)\) with probability \(\Omega(1)\), the previously best upper bound is \(O\bigl(n^{1/2}\min\bigl\{ \Delta_{\mathrm{in}}^{1/2},\ \Delta_{\mathrm{out}}^{1/2},\ m^{1/4} \bigr\}\bigr)\) from [Wang, Wei, Wen, Yang, STOC 2024], where \(n\) and \(m\) denote the number of nodes and edges in \(G\), and \(\Delta_{\mathrm{in}}\) and \(\Delta_{\mathrm{out}}\) upper bound the in-degrees and out-degrees of \(G\), respectively. Using a refinement of the proof in the same paper, we establish a lower bound of \(\Omega\bigl(n^{1/2}\min\bigl\{ \Delta_{\mathrm{in}}^{1/2} \big/ n^{\gamma},\Delta_{\mathrm{out}}^{1/2} \big/ n^{\gamma},m^{1/4}\bigr\}\bigr)\), where \(\gamma = \frac{1}{2} \bigl(2\max\bigl\{\log_{1/(1-\alpha)}\Delta_{\mathrm{in}},1\bigr\}-1\bigr)^{-1}\). As \(\gamma\) only depends on \(\Delta_\mathrm{in}\) and \(n^{\gamma} = O(1)\) for \(\Delta_{\mathrm{in}} = n^{\Omega(1)}\), the known upper bound is tight if we only parameterize the complexity by \(n\), \(m\), and \(\Delta_{\mathrm{out}}\). However, there remains a gap of \(\Omega\bigl(n^{\gamma}\bigr)\) when considering the maximum in-degree \(\Delta_{\mathrm{in}}\), and this gap is large when \(\Delta_{\mathrm{in}}\) is small. In the extreme case where \(\Delta_{\mathrm{in}} \le 1/(1 - \alpha)\), we have \(\gamma = 1/2\), leading to a gap of \(\Omega\bigl(n^{1/2}\bigr)\) between the bounds \(O\bigl(n^{1/2}\bigr)\) and \(\Omega(1)\).
   In this paper, we present a new algorithm that achieves the above lower bound (up to logarithmic factors). The algorithm assumes that \(n\) and the bounds \(\Delta_{\mathrm{in}}\) and \(\Delta_{\mathrm{out}}\) are known in advance. Our key technique is a novel randomized backwards propagation process that only propagates selectively based on Monte Carlo estimated PageRank scores.

   @inproceedings{thorup2026pagerank,
     author       = {Mikkel Thorup and
                     Hanzhi Wang and
                     Zhewei Wei and
                     Mingji Yang},
     title        = {{PageRank} centrality in directed graphs with bounded in-degree},
     booktitle    = {Proceedings of the 2026 Annual {ACM-SIAM} Symposium on Discrete Algorithms},
     pages        = {3820--3840},
     year         = {2026},
     doi          = {10.1137/1.9781611978971.140}
   }
     

3. STOC 2024

   Revisiting Local Computation of PageRank: Simple and Optimal

   Hanzhi Wang, Zhewei Wei^*, Ji-Rong Wen, and Mingji Yang

   In Proceedings of the 56th Annual ACM Symposium on Theory of Computing, Jun 2024

   TLDR Abstract BibTex arXiv Video Slides by Hanzhi Comments by Seshadhri

   We establish nearly-tight complexity upper and lower bounds on estimating PageRank contributions and single-node PageRank on directed graphs. Our techniques and analyses are surprisingly simple, where the upper bounds are derived by revisiting classic algorithms.

   We revisit the classic local graph exploration algorithm ApproxContributions proposed by Andersen, Borgs, Chayes, Hopcroft, Mirrokni, and Teng [WAW 2007, Internet Math. 2008] for computing an \(\epsilon\)-approximation of the PageRank contribution vector for a target node \(t\) on a graph with \(n\) nodes and \(m\) edges. We give a worst-case complexity bound of ApproxContributions as \(O\bigl(n\pi(t)/\epsilon \cdot \min\bigl\{\Delta_{\mathrm{in}},\Delta_{\mathrm{out}},\sqrt{m}\bigr\}\bigr)\), where \(\pi(t)\) is the PageRank score of \(t\), and \(\Delta_{\mathrm{in}}\) and \(\Delta_{\mathrm{out}}\) are the maximum in-degree and out-degree of the graph, resp. We also give a lower bound of \(\Omega\bigl(\min\bigl\{\Delta_{\mathrm{in}}/\delta,\Delta_{\mathrm{out}}/\delta,\sqrt{m}/\delta,m\bigr\}\bigr)\) for detecting the \(\delta\)-contributing set of \(t\), showing that the simple ApproxContributions algorithm is already optimal.
   As ApproxContributions has become a cornerstone for computing random-walk probabilities, our results and techniques can be applied to derive better bounds for various relevant problems. In particular, we investigate the computational complexity of locally estimating a node's PageRank centrality. We improve the best-known upper bound of \(\widetilde{O}\bigl(n^{2/3}\min\bigl\{\Delta_{\mathrm{out}}^{1/3},m^{1/6}\bigr\}\bigr)\) given by Bressan, Peserico, and Pretto [SICOMP 2023] to \(O\bigl(n^{1/2}\min\bigl\{\Delta_{\mathrm{in}}^{1/2},\Delta_{\mathrm{out}}^{1/2},m^{1/4}\bigr\}\bigr)\) by simply combining ApproxContributions with Monte Carlo simulation. We also improve their \(\Omega\bigl(\min\bigl\{n^{1/2}\Delta_{\mathrm{out}}^{1/2},n^{1/3}m^{1/3}\bigr\}\bigr)\) lower bound to \(\Omega\bigl(n^{1/2}\min\bigl\{\Delta_{\mathrm{in}}^{1/2},\Delta_{\mathrm{out}}^{1/2},m^{1/4}\bigr\}\bigr)\) if \(\min\{\Delta_{\mathrm{in}},\Delta_{\mathrm{out}}\}=\Omega\bigl(n^{1/3}\bigr)\). Our matching upper and lower bounds resolve the open problem of whether one can tighten the bounds given by Bressan, Peserico, and Pretto [FOCS 2018, SICOMP 2023]. Remarkably, the techniques and analyses for proving all our results are surprisingly simple.

   @inproceedings{wang2024revisiting,
     author       = {Hanzhi Wang and
                     Zhewei Wei and
                     Ji{-}Rong Wen and
                     Mingji Yang},
     title        = {Revisiting local computation of {PageRank}: Simple and optimal},
     booktitle    = {Proceedings of the 56th Annual {ACM} Symposium on Theory of Computing},
     pages        = {911--922},
     year         = {2024},
     doi          = {10.1145/3618260.3649661}
   }