Papers

We study the problem of learning nearest-neighbor maps from adaptive queries, which is equivalent to the following problem of reconstructing a hidden set $H$ via a nearest-neighbor query oracle. Let $K \subset \mathbb{R}^d$ be a compact domain in a normed space $(\mathbb{R}^d,\| \cdot\|)$ and let $H \subset K$ be a hidden set of $n$ points. Upon querying $q \in K$, the oracle returns some $h \in H$ with minimum distance from $q$. How many queries are required to exactly recover $H$? Previous work has studied this question in specific domains, namely the Boolean hypercube and the $\ell_2$-unit sphere. We generalize previous work and prove the tight worst-case query complexity bound of $\Theta(n\kappa)$, where $\kappa$ is the kissing number of the underlying norm. In the Euclidean norm, obtaining tight asymptotic bounds on $\kappa$ is a significant open question, although it is known that $\kappa = \exp(\Theta(d))$. Our second set of results shows that an exponential dependence on $d$ is required even in natural Euclidean domains: $\exp(\Omega(d))$ queries are needed in the ball, even when $n=2$, and $n\exp(\Omega(d))$ queries are needed in the cone. Lastly, we prove a sharper upper bound in the Euclidean sphere. Here, $d$ can be replaced by $\min(n,d)$ via a dimension reduction preprocessing step. This is a randomized version of a procedure due to Prabhu-Woodruff (ICML 2024) where we improve the query complexity from $O(nd)$ to $O(\min(n,d))$. This reveals a striking contrast between the sphere and the ball: when $n = O(1)$, the sphere admits an $O(1)$ query algorithm, whereas the ball requires $\exp(\Omega(d))$.

We consider the following problems: Given two $n \times n$ tables defining binary operations $+$ and $\cdot$ on a set $S$ of $n$ elements, decide whether $(S,+,\cdot)$ forms a ring or, respectively, a field. Recently, Dudek, Fischer, Gokaj, Jin, Künnemann, Mao, and Redzic (STOC 2026) obtained the following two (near-)optimal results: (1) A randomized $O(n^2\log(1/\delta))$-time algorithm for verifying rings. (2) A deterministic $O(n^2)$-time algorithm for verifying fields. Their algorithms build on machinery of Evra, Gadot, Klein, and Komargodski (FOCS 2024), which relies on Classification of Finite Simple Groups (CFSG). In this work, we give a deterministic $O(n^2)$-time algorithm for ring verification, resolving the deterministic complexity of this problem. As a corollary, we also obtain a deterministic $O(n^2)$-time algorithm for field verification. Our algorithms are elementary and avoid CFSG machinery entirely.

Let $G$ be an undirected unweighted planar graph and let $S=(s_0,\dots,s_{k-1})$ be the vertices of a designated face, listed in cyclic order. Consider a vector that stores the distances from an arbitrary vertex $v$ to all vertices of $S$. The pattern of $v$ is obtained by taking the difference between every pair of consecutive values in this vector. Li and Parter [STOC'19] proved an upper bound of $O(k^3)$ on the number of unique patterns over all vertices of $G$. We improve this to $O(k^2)$, matching a known lower bound and settling a conjecture in [ISAAC'22]. The simple proof was found by OpenAI's GPT 5.6-Sol model. Plugging this new bound into known results has the following three immediate implications for undirected unweighted planar graphs: (1) it gives an improved compression of the Okamura-Seymour metric (2) it improves the space required by constant-time exact distance oracles, and (3) it improves the fastest distributed algorithm for computing the diameter. We further present a previously unknown and nontrivial implication: a (centralized) $\tilde{O}(n^{8/5})$-time algorithm for computing the diameter, improving over the $\tilde{O}(n^{5/3})$ algorithm of [SODA'18] which works for weighted directed planar graphs. Thus, there is currently a gap between the time for computing the diameter between weighted and unweighted planar graphs.

Heaps (priority queues) are among the best-studied data structures in computer science. In this paper, we critically revisit the textbook assumption that in the comparison model at least one of the two standard heap operations of inserting an element and deleting the minimum must take logarithmic time. By decoupling the deletion itself from the act of revealing the identity of the deleted element to the user, we avoid the sorting barrier and obtain a novel trade-off between the complexities of heap operations. This shows that the logarithmic barrier is not inherently the cost of deleting the minimum but rather the information cost of immediately learning which element was deleted. In the special case when the user inspects the heap state only constantly many times, we show that both insertions and deletions can be supported in constant amortized time. As an application, this yields a runtime improvement from $\mathcal{O}(n \log n)$ to the optimal $\mathcal{O}(n)$ for a textbook unit-time scheduling problem. We obtain our results by designing a new data structure, the sync heap, which gains speed by rearranging and compacting its operations until queries force it to synchronize and reveal its state. As a key component, we use the soft heap introduced by Chazelle as part of his minimum spanning tree algorithm. Our data structure is simple, comparison-based, and deterministic, and our results are asymptotically optimal.

Let $L$ and $f$ be positive integers. An $(L,f)$-replacement path covering (RPC) for a graph $G$ is a family $\mathcal{G}$ of subgraphs such that, for every set $F$ of at most $f$ edges, there is a subfamily $\mathcal{G}_F \subseteq \mathcal{G}$ with the following properties. (1) No subgraph in $\mathcal{G}_F$ contains an edge of $F$. (2) For each pair of vertices $s,t$ that have a shortest path in $G{-}F$ with at most $L$ edges, one such path also exists in some subgraph in $\mathcal{G}_F$. The total number $|\mathcal{G}|$ of subgraphs is called the covering value. RPCs are an important tools in the design of fault-tolerant data structures. Weimann and Yuster [TALG 2013] presented an RPC with covering value $\widetilde{O}(f L^f)$. Karthik and Parter [TALG 2024] showed that $\Omega( (L/f)^f )$ subgraphs are necessary. Recently, Bilò, Chechik, Choudhary, Cohen, and Schirneck [ICALP 2026] devised a new approach for very small sensitivities $f = o(\log L)$ with covering value $\widetilde{O}(f e^f (L/f)^{f+o(1)})$. They also showed that any RPC in the complementary range $f = \Omega(\log L)$ must contain $\Omega( (\sqrt{f e^f}/L) \cdot (L/f)^f)$ subgraphs. This left open the question of what is the true covering value. We give two surprisingly simple constructions that improve both the upper and lower bound. This results in a near-tight covering value of $\widetilde{\Theta}(\frac{(L+f)^{L+f}}{L^L f^f}) \cdot \mathsf{poly}(f)$ for the much wider range of $f = O(L)$.

Consider a poset - or equivalently an $n$-vertex DAG $G=(V, E)$ - and a boolean function $f: V \rightarrow \{0, 1\}$ on its vertex set. We say $f$ is monotone if $f(u) \leq f(v)$ for all $(u, v) \in E$. While there is extensive literature on the query complexity of testing monotonicity, we focus instead on the space complexity and initiate the study of this problem in the streaming setting. Namely, the edges of $G$ arrive in an arbitrary order, and the goal is to estimate distance to monotonicity of a given function $f$ using $\widetilde{O}(n)$ space. Note that while this space allows receiving and storing $f$, it is much smaller than the input graph $G$ which could have up to $\Omega(n^2)$ edges. Our main result is an algorithm that $(1+\epsilon)$-approximates distance to monotonicity in $\sqrt{n}^{1+o(1)}$ passes. We also prove that this is the best pass-complexity one can hope for, for any $O(1)$-approximation, short of improving the state-of-the-art streaming algorithm for $st$-reachability, which is a very well-studied problem. On the technical side, our algorithm approximates the size of maximum matching in (a subgraph of) the transitive closure of $G$. While the maximum matching problem has received significant attention in the streaming setting, the fact that we are computing it in the transitive closure requires very different ideas. In fact, a main contribution of our work is to connect sublinear time algorithms for estimating the maximum matching size to the streaming setting for the first time. While existing off-the-shelf sublinear time algorithms only result in an $n\sqrt{n}^{1+o(1)}$ pass algorithm in our setting, we show how to significantly improve upon them by allowing stronger queries (such as vertex and subset queries) that can be implemented just as efficiently as more standard adjacency matrix and list queries for our problem.

Given a heterogeneous information network (HIN) and a query meta-path P of length i, the densest P-partite subgraph problem finds the subgraph, spanning the i typed layers of P, that maximizes a parameter-free density: the number of meta-path instances over the geometric mean of the layer sizes. It has applications across bibliographic, e-commerce, and biomedical networks. The state-of-the-art approximation linearizes the geometric-mean objective by fixing per-layer weights, but solves one subproblem for every feasible weight set, of which there are $O((n/i)^i)$, and on each achieves only a $1/i$ approximation. We show that neither the exhaustive enumeration nor the loose guarantee is necessary. First, we replace enumeration by covering: polylogarithmically many representative weight sets, localized further by a data-dependent bound, cover all feasible ones while losing only a tunable factor $1+\eta$ in density. Second, we cast each fixed-weight subproblem as a weighted supermodular densest-subgraph instance and solve it near-optimally, lifting the overall guarantee to $(1-\delta)/(1+\eta)$. To our knowledge, this is the first near-optimal density approximation beyond the bipartite ($i=2$) case, and it yields a PTAS for every fixed i. Algorithmically, our solver is an adaptive peeling scheme that never materializes the meta-path instances, whose number can exceed the graph size by orders of magnitude. An incumbent-driven reduction further discards representative weight sets before their subproblems are solved. Experiments on five real HINs show that our algorithms achieve substantial speedups over enumeration-based baselines and can further certify the near-optimality of the returned subgraph.

For a finite hypergraph H, the complements of the models of its associated definite Horn CNF are exactly the stopping sets of H; hence the complements of its coatoms are the inclusion-minimal nonempty stopping sets. We study their output-sensitive enumeration from the hypergraph incidence representation. Our main result is a representation of arbitrary Horn model posets whose incidence size is linear in the incidence length of the normalized Horn input. Given a Horn CNF $\Gamma$, we construct a hypergraph $C(\Gamma)$ of rank at most three whose proper-model poset is inclusion-order isomorphic to the model poset of $\Gamma$; equivalently, each source model has a unique extension to a proper target model. Thus maximal models of $\Gamma$ correspond bijectively to target coatoms. Combining this representation with the maximal-model construction of Kavvadias, Sideri, and Stavropoulos shows that coatom enumeration is not in OutputP unless P=NP, even when all hyperedges have size two or three. Incidence splitting reduces maximum element frequency to three while preserving the stopping-set poset, and a local replacement of two-element hyperedges yields the same lower bound for three-uniform hypergraphs of maximum element frequency at most three. These thresholds are conditionally sharp for arbitrary-order enumeration: rank at most two and maximum element frequency at most two both admit output-linear total-time generation; in the frequency-two case, a polynomial-delay, polynomial-space algorithm is also available. In contrast, coatom extension is NP-complete already for three-uniform hypergraphs of exact element frequency two.

We give a new $O(\log n)$ approximation algorithm for Node Weighted Steiner Tree and Node Weighted Steiner Forest. Our algorithm matches the bounds of Klein & Ravi [J. Algorithms '95] which are best possible unless P = NP, but have the advantage that they work in the online setting when the terminal pairs are revealed in random order. To obtain our results, we combine the LearnOrCover framework due to Gupta, Kehne, Levin [FOCS '21] with the Augmented Greedy algorithm of Berman & Coulston [STOC '97] for online edge-weighted Steiner Forest. Neither algorithm suffices on its own, but the analyses dovetail to imply our guarantee. Run offline, the algorithm reduces to a very simple randomized rounding scheme that (in spirit) reduces Node Weighted Steiner Forest to Edge Weighted Steiner Forest, and we hope this idea finds further applications.

We study online multi-level aggregation on finite rooted trees with a per-batch maximum-delay objective. A service pays for a rooted subtree and for the maximum waiting time among the requests cleared by that service. We show that the offline optimum admits a consecutive-arrival-block normal form and can be computed by a polynomial-time dynamic program. The same dynamic program defines the deadlines of a family of online algorithms, which we call DP-Envelope. Its deterministic endpoint is $2$-competitive. Sampling one global parameter with density $e^\theta/(e-1)$ leads to an $e/(e-1)$-competitive randomized algorithm against an oblivious adversary. The deterministic guarantee matches the known fixed-node lower bound, and we prove a matching randomized lower bound. Thus, both guarantees are optimal on every nondegenerate rooted tree. We first develop the line metric as a warm-up, where the algorithm and its nested block partitions have a direct geometric interpretation. Finally, we show that the upper bounds extend to every realizable static service system with a normalized, nondecreasing, submodular joint service cost.

Let \(G\) be a connected simple undirected graph. A vertex \(w\) is said to strongly resolve a pair of distinct vertices \(u, v \in V(G)\) if either there exists an isometric path (i.e.~a shortest path) from \(w\) to \(u\) that contains \(v\), or there exists an isometric path from \(w\) to \(v\) that contains \(u\). A subset \(S \subseteq V(G)\) is said to strongly resolve \(G\) if every pair of distinct vertices of \(G\) is strongly resolved by at least one vertex in \(S\). In the Strong Metric Dimension problem, the input consists of a graph \(G\) and a positive integer \(k\), and the objective is to determine whether there exists a subset \(S \subseteq V(G)\) of size at most \(k\) that strongly resolves \(G\). In this article, we show that Strong Metric Dimension is \NP-complete even on \((i)\) graphs of diameter two, and \((ii)\) graphs of constant pathwidth and constant feedback vertex set number.

For the $n\times n$ lower-triangular all-ones matrix $Q$, we prove a near-optimal lower bound \[ \gamma_{2,1}(Q) := \inf_{Q=AB} \|A\|_{2\to\infty}\|B\|_{1\to1} = \Omega\!\left( \frac{\log^{3/2}n}{(\log\log n)^{3/2}} \right), \] where the infimum ranges over real factorizations of arbitrary finite inner dimension. This cost is a central parameter in space bounds for factorization-based rank and quantile estimation in turnstile streams and in error bounds for matrix mechanisms for continual counting under pure differential privacy. The proof combines right-sided Haar projections with a scale-dependent numerical-sparsity decomposition of the rows of $B$. At each scale, a rank–Frobenius argument shows that the numerically sparse rows cannot account for all of the required Schatten $2/3$ mass, while a Haar projection estimate bounds the contribution of the remaining rows. Summing these bounds over the dyadic scales yields the result. The proof was obtained using a fully automated Gemini-based agentic system developed internally at Google. The authors verified the proof and made minor revisions.

The standard online version of correlation clustering is prohibitively hard, as even randomized algorithms cannot achieve competitive ratio better than $\Omega(n)$. Prior works bypass this lower bound by relaxing the online model through recourse, random arrival order, or seeding the algorithm with an offline sample of the underlying input. We instead ask whether additional structure in the input itself can overcome this lower bound. We study weighted correlation clustering under probability constraints, where $w^+_{uv}+w^-_{uv}=1$ for every $uv$ edge, and triangle inequality constraints, where the negative weights $w^-$ satisfy triangle inequality. While this version of correlation clustering is well-studied in the offline setting, we initiate its online study and give a deterministic online algorithm that maintains a clustering whose total weighted disagreement cost is within an $O(1)$ factor of the offline optimum, against adversarial arrival order. This is the first constant-competitive online algorithm for a natural minimization variant of correlation clustering in the fully online model, and shows that metric consistency on the edge weights separates tractable from intractable online instances.

Many local search methods for problems in artificial intelligence can be viewed as an uphill climb on a corresponding discrete fitness landscapes. Finding even local peaks in these fitness landscapes is computationally intractable in theory, but often works in practice. So what features of fitness landscapes allow for efficient local search? Since any fitness landscapes can be represented by a (hyper)graph of valued constraints, I re-frame this as a question of parameterized complexity: what structural parameter of a valued constraint graph guarantees that a strict local search will find a local peak in the corresponding fitness landscape efficiently? Given a valued constrain graph of vertex cover number k, I prove that greedy local search will find the a local fitness peak in at most $2^{2k}\cdot(n - k + 1)$ steps and random uphill local search will find a local fitness peak in an expected number of at most $2^k\cdot n(n - k)$ steps. I also show that these results are asymptotically good for strict local search because there are valued constraint graphs of vertex cover number $k$ where every ascent from some initial assignment has a length of $\frac{9}{128} \cdot 2^k \cdot (n - k)$ or greater. This suggests vertex cover number as a good structural parameter for the complexity of local search.

For every $\varepsilon > 0$, it is NP-hard to $n^{1-\varepsilon}$-approximate Max Independent Set in $n$-vertex graphs [Hastad '96, Zuckerman '07]. In triangle-free graphs, a simple argument gives a polynomial-time $n^{1/2}$-approximation algorithm, whereas, for every $\varepsilon > 0$, an $n^{1/4-\varepsilon}$-approximation algorithm would imply that NP $\subseteq$ BPP [Bonnet, Thomassé, Tran, Watrigant; ESA '20]. In this note, we close this gap by proving the corresponding hardness against $n^{1/2-\varepsilon}$-approximation algorithms. The reduction is very simple and uses the Moser-Tardos resampling algorithm to make the constructed graphs triangle-free. The soundness uses a result of Haeupler, Saha, and Srinivasan building on the proof of Moser and Tardos, to upper-bound the probability that a fixed relatively large subset is an independent set after the Moser-Tardos algorithm terminates. We generalize this scheme and show that, for any nonempty finite family $\mathcal F$ of graphs, each containing at least one cycle, for any $\varepsilon > 0$, an $n^{\mu(\mathcal F)-\varepsilon}$-approximation algorithm for Max Independent Set in graphs excluding every member of $\mathcal F$ as a subgraph implies that NP $\subseteq$ BPP, where $\mu(\mathcal F) := 1 - \max\limits_{H \in \mathcal F}~\min\limits_{U \subseteq V(H), H[U] \text{contains a cycle}} (|U|-2)/(|E(H[U])|-1)$.

Let $H\subseteq\{-1,+1\}^X$ be a class of finite VC dimension $d\ge1$. Writing $L$ for the binary risk and $L^*=\min_{h\in H}L(h)$, we construct a learner achieving the statistically optimal risk bound: from an i.i.d.\ sample of size $n$, for every $0<\delta\le 1/2$, with probability at least $1-\delta$, \[ L(\widehat h) \le L^*+ 7\cdot10^8\left( \sqrt{\frac{L^*(d+\log(1/\delta))}{n}} +\frac{d+\log(1/\delta)}{n} \right). \] This settles the sample complexity of agnostic PAC learning up to universal constants at every fixed $L^*$, matching the lower bounds of Devroye, Györfi, and Lugosi [A Probabilistic Theory of Pattern Recognition, Springer, 1996].

A monotone adversary observes an i.i.d. labeled sample and appends a finite number of further examples of its choice, every one of them labeled correctly by the target hypothesis. The learner sees a uniform shuffle of the combined sample and is scored on the original distribution. Every example is correctly labeled, but the insertions depend on the clean sample, so the combined sample is not exchangeable. Larsen, Pabbaraju, and Shetty, who introduced this model, showed that empirical risk minimization attains expected error $O((d/n)\log(n/d))$ for classes of VC dimension $d$, and that every known optimal learner can be pushed away from the $\Theta(d/n)$ rate, optimal for PAC learning. They asked whether the extra logarithm is an artifact of those particular algorithms or an inherent consequence of the lack of exchangeability. We show that this additional cost is inherent beyond VC dimension one. In the worst case over classes of VC dimension $d$ and over known finite insertion budgets, the minimax expected error is $\Theta(1/n)$ at $d=1$ and $\Theta((d/n)\log(n/d))$ for $d\geq 2$. The same rates hold with Littlestone dimension $d_{\mathrm L}$ in place of $d$, so the clean online-to-batch rate $O(d_{\mathrm L}/n)$ is unattainable as well. Thus, somewhat counterintuitively, adding correctly labeled examples can make learning harder by a logarithmic factor, even for classes that admit finite mistake bounds in online learning. The dimension-one upper bound is achieved by a simple improper learner whose analysis adapts the leave-one-out argument underlying the one-inclusion graph. All of our lower bounds are elementary and come from a single construction: an explicit class and prior on which two target hypothesis, which differ a point of nonnegligible mass, produce the same sample.

The low-degree heuristic has become a widely used framework for predicting computational thresholds in average-case planted-versus-null problems. However, a recent sequence of counterexamples shows that low-degree indistinguishability does not, in general, rule out efficient noise-tolerant distinguishers; see Buhai et al. (2025) and Mao (2026). Motivated by these developments, Hsieh et al. (2026) initiated the study of rigorous consequences of the low-degree heuristic. In this work, we continue this program for planted-graph problems. Let $Q_n=G(n,c/n)$, and let $P_n$ be obtained by planting a uniformly random copy of a deterministic graph $\Gamma_n$ into an independent sample from $Q_n$. In the supercritical regime $c>1$, we show that if $P_n$ is degree-$D_n$ indistinguishable from $Q_n$ and $\operatorname{tw}(\Gamma_n)=o(D_n/\log n)$, then a noisy version of $P_n$ is asymptotically indistinguishable from $Q_n$. Here $\operatorname{tw}(\Gamma_n)$ denotes the treewidth of $\Gamma_n$, a measure of how efficiently the graph can be decomposed into tree-like pieces. In the critical and subcritical regimes $0<c\leq 1$, the same conclusion holds whenever $D_n=\omega(\log n)$, without any treewidth assumption. Our proof has two main ingredients. First, we uncover a correspondence between the subgraph-count and automorphism factors in the Fourier expansion and counts of isomorphism triples. Second, we cut the decomposition tree into subtrees, breaking each large Fourier support into low-degree pieces that meet at only a few interface vertices, and use noise to absorb the cost of reassembling them. At and below criticality, the low-degree assumption rules out short cycles, while noise destroys the remaining long cycles.

We show that the hard-core partition function admits a fully polynomial-time randomised approximation scheme (FPRAS) on planar graphs when the activity is a sufficiently small constant. In contrast, we show that for any constant $q\ge 4$, approximately counting $q$-colourings in planar graphs is NP-hard. We also give a complete characterisation of when an FPRAS exists for a sufficiently small external field for 2-spin systems on planar graphs. The main ideas of all proofs were found using GPT-5.6 Sol Ultra.

Stone Game III admits a standard backward dynamic program using $O(n)$ time and $O(1)$ auxiliary space. The upper bound is immediate, but its optimality raises a deceptively simple question: must a correct algorithm really inspect a linear number of input values? For the original problem, an all-zero instance gives a short indistinguishability proof that every position must be inspected. This argument appears to depend strongly on the possibility of a tie. We show that it does not. Even under the promise that every input has a winner, an adversary can force any deterministic algorithm to make $\Omega(n)$ inspections by combining modular move control with indistinguishable input completions. We also extend the argument to positive but unbounded values, obtaining the same linear lower bound without zeros or ties. Together these results establish the asymptotic optimality of the standard $O(n)$-time, $O(1)$-space solution in several increasingly restrictive variants.

We give a query-optimal algorithm for simulating a general $n$-qubit time-dependent Hamiltonian $H(t)$ on $[0,T]$, assuming that $H$ is Lipschitz continuous and $\|H(t)\|\leq\alpha$. In the standard $\mathrm{HAM\mbox{-}T}$ access model, the algorithm approximates the time-ordered propagator $U_H(T)$ to error $\varepsilon$ using $$ O\left( \alpha T+\frac{\log(1/\varepsilon)} {\log(e+\log(1/\varepsilon)/(\alpha T))} \right) $$ $\mathrm{HAM\mbox{-}T}$ queries. This matches the known query lower bound for time-independent Hamiltonians, showing that time dependence incurs no asymptotic query overhead. Our method first constructs a one-query transducer that, given an auxiliary state, implements an approximation to $U_H(T)$ and returns the state unchanged. A weighted combination of circuits that apply the transducer different numbers of times makes the error caused by omitting this state decay factorially, yielding the stated optimal precision dependence. For time-independent Hamiltonians, the same method also gives a query-optimal alternative to qubitization.

The dynamic dictionary is a fundamental data structure that maintains a set $S\subset [U]$ of size $n$ (we assume $n=U^{1-\Theta(1)}$), supporting insertions, deletions and membership queries. Previous works mostly focused on constructing dictionaries that support operations in $O(1)$ time and use space as close to the information-theoretic bound of $\log\binom{U}{n}$ bits as possible. In this paper, we study difference-encoded dictionaries, which are dictionaries that use space close to the gap entropy $\text{gap}(S):=\sum_{i=2}^{|S|}\left(\lceil\log(x_i-x_{i-1}+1)\rceil+1\right)$ bits to store the set $S=\{x_1<\dots<x_n\}$. On many real-world datasets where the keys are clustered, we have $\text{gap}(S)\ll \log\binom{U}{n}$, making difference-encoded dictionaries more favorable than standard dictionaries in practice. Prior to this work, the best dynamic difference-encoded dictionary is by Blandford and Blelloch [SODA'04], whose construction supports operations in $O(\log n)$ time and uses $O(\text{gap}(S))$ bits of space. In the static case, Gupta, Hon, Shah and Vitter [DCC'06] presented a dictionary that uses $$ \text{gap}(S)+O(n\log\log U) $$ bits of space and supports membership queries in $O(\log\log n)$ time. In this work, we go beyond these bounds and fully settle the optimal time-space tradeoff for difference-encoded dictionaries. For an arbitrary parameter $0<\varepsilon<1/4$, we construct a dynamic dictionary that supports operations in $O(\log\varepsilon^{-1}/\log\log\varepsilon^{-1})$ expected amortized time and uses $$ \text{gap}(S)\cdot(1+O(\varepsilon))+O\left(n\log\frac{\text{gap}(S)}{n}\right) $$ bits of space. We also prove a matching lower bound, showing that our time-space tradeoff is optimal even in the static case.

We show how to implement a dynamic array $A[1, n]$ with symbols from a fixed alphabet $\Sigma$, while supporting $O(1)$-time queries and updates, and using a total space of $$ \log \binom{|\Sigma|}{m} + \left(1 + O\left(\frac{\log \log n}{\log n}\right)\right) \cdot \left(\sum_{\sigma \in \Sigma} f_\sigma \log (n / f_\sigma)\right) + n / \text{polylog } n $$ bits, where $f_\sigma$ denotes the frequency of each symbol $\sigma \in \Sigma$ and $m$ denotes the number of distinct symbols with non-zero frequencies. This resolves a long-standing open question as to whether one can achieve space bounds close to that of arithmetic coding, while supporting $O(1)$-time operations, whenever the entropy is at least $n/\text{polylog } n$. We also prove a nearly matching space lower bound: up to a factor of $O(\log \log n)$, the entropy-dependent multiplicative overhead of our construction is optimal among $O(1)$-time solutions when $|\Sigma|=O(\sqrt n)$ and the entropy $\sum_{\sigma \in \Sigma} f_\sigma \log (n / f_\sigma)$ lies between $n/\log^{O(1)}n$ and $(1/100)n\log n$. Finally, we present several applications of our results, resolving two open problems having to do with space-efficient dictionaries and filters.

Beisegel et al. [SWAT 2024] introduced the concept of simultaneous $\mathcal{C}$-numbers which associate a graph class $\mathcal{C}$ with a graph parameter. Given a graph $G$, the simultaneous $\mathcal{C}$-number is the smallest number $d$ for which there is a graph $H \in \mathcal{C}$ and a function $L : V(G) \to \mathcal{P}(\{1,\dots,d\})$ such that two vertices $u$ and $v$ are adjacent in $G$ if and only if they are adjacent in $H$ and their sets $L(u)$ and $L(v)$ are not disjoint. We study the relation of these simultaneous $\mathcal{C}$-numbers to other graph parameters. In particular, we investigate which parameters fulfill the following property: Parameter $p$ is bounded on class $\mathcal{C}$ if and only if $p$ is bounded on the class of graphs of simultaneous $\mathcal{C}$-number $d$ for any fixed $d$. We show that many well-known graph parameters have this property. Examples are cliquewidth, twin-width, mim-width, tree independence number, thinness as well as boxicity. We furthermore present some parameters, including modular-width and tree-length, that do no have this property. We also study when a parameter forms an upper bound on a simultaneous $\mathcal{C}$-number. We characterize those graph classes $\mathcal{C}$ for which the parameters treewidth, pathwidth, bandwidth, and treedepth upper bound the simultaneous $\mathcal{C}$-number. Furthermore, we present sufficient conditions on a class $\mathcal{C}$, such that $\mathcal{P}$-modular cardinality upper bounds the simultaneous $\mathcal{C}$-number, where $\mathcal{P}$ is replaced by the complete graphs, the edgeless graphs, cographs, or the class $\mathcal{C}$ itself. On the contrary, we show that modular width never forms an upper bound on a non-trivial simultaneous $\mathcal{C}$-number. Finally, we present some general algorithmic results on the clique problem and computation of simultaneous $\mathcal{C}$-numbers.

The degree sequence optimization problem is to find a subgraph of a given graph which maximizes the sum over all vertices of a given function evaluated at the subgraph degree of that vertex. Here we study this problem and its complexity for quadratic functions. In particular, we introduce the critical roots of a graph, and show they define intervals over which the optimal value of the problem, as the quadratic root varies, is convex piecewise affine.

In the Correlation Clustering problem, we are given an undirected graph and are tasked with computing a clustering (partition of the nodes) that minimizes the number of violated pairs (edges across different clusters plus non-edges within clusters). In the constrained version of this problem, the goal is to compute a clustering that satisfies additional hard constraints mandating certain pairs to be in the same cluster and certain pairs to be in different clusters. In this work, we identify Constrained Correlation Clustering as a variant of Correlation Clustering for which optimal approximations might be within reach, and make progress towards this front. Constrained Correlation Clustering is APX-Hard, and the optimal approximation factor is known to lie in $(\frac{24}{23},3]$. We significantly tighten this gap, by showing that the optimal approximation factor lies in $[2,\frac{16}{7}-\gamma)$ for a small constant $\gamma>0$. Our lower bound of $2$ shows a separation between Correlation Clustering (which admits an $1.485+\epsilon$ approximation) and Constrained Correlation Clustering\footnote{The same hardness result was obtained independently by Cao and Xu~\cite{cao2026clusterdeletionhardapproximate}.}. Our upper bound of $\frac{16}{7}-\gamma$ uses the Sherali-Adams relaxation and goes beyond straightforward Triangle-Based analysis; more precisely, our algorithm belongs to a natural class of pivoting algorithms for which we prove that a straightforward Triangle-Based analysis cannot prove a better-than-$\frac{16}{7}$ approximation. Finally, as a byproduct of our techniques, we completely resolve the approximability of Cluster Deletion. Cluster Deletion is a well-studied special case of Constrained Correlation Clustering for which a $2$-approximation algorithm is known. We show that this is optimal, as our lower bound holds even for this special case.

We consider the A/E-design problem under partition constraints: Given vectors $v_1,\ldots,v_N\in \R^d$ and a partition matroid on $[N]$, find a base $S$ of the matroid that minimizes $\tr(M(S)^{-1})$ or $\lambda_{\max}(M(S)^{-1})$ where $M(S)=\sum_{i \in S} v_i v_i^\top$. In contrast to D-design, where good estimation and approximation guarantees are known as a function of $d$, we show that no reasonable approximation exists for A/E-design. This answers a question of Brown, Laddha and Singh. The proof is based on an elementary reduction from three-dimensional matching.

When a student must learn concepts connected by prerequisite dependencies, when does the order of instruction matter, and what does it cost to find the best one? We study instructional sequencing as a stochastic shortest-path problem in which attempting a concept succeeds with a state-dependent probability and failure leaves the learner state unchanged. We first prove that this stochasticity can be eliminated exactly: the problem collapses to a deterministic shortest-path problem on the lattice of prerequisite order ideals, preserving optimal values and actions. The collapse removes stochastic complexity but not combinatorial complexity: optimal sequencing remains NP-hard – via reduction from feedback arc set in tournaments – even with no prerequisite edges, unit costs, uniform binary nonnegative transfer, and success probabilities at least $1/2$. Hardness is not uniform: when realizable transfer preferences remain jointly acyclic with the prerequisites, any topological order of the residual joint graph is optimal, and fixed prerequisite width yields polynomial-time exact dynamic programming. A computable diagnostic, $m\Delta$, bounds the value of sequencing before optimization. On 70,893 interactions from an introductory CS course, the diagnostic certifies a doubly easy regime – little value to optimize and little space to search – while constructed transfer instances realize the challenging regime, where myopic sequencing suffers large regret yet exact A* with a consistent heuristic expands only linearly many states on that family.

Consider the problem of maximizing the number of satisfied constraints of an arbitrary boolean constraint satisfaction problem with arity $k$. We obtain a polynomial time algorithm that achieves a $(k/2^k)$-approximation, improving on the previous best guarantee of $0.626612\; k/2^k$, due to Makarychev and Makarychev (arXiv:1206.3603). Assuming the Unique Games Conjecture, De and Mossel (arXiv:1202.5258) showed that achieving an approximation ratio better than $(k+1)/2^k$ for odd $k$ and $(k+2)/2^k$ for even $k$, is NP-hard. The main technical ingredient is an extension of a recently established Gaussian comparison inequality, used to resolve the Weak Simplex Conjecture in coding theory (arXiv:2607.14087).

Our results show that the existence of a short high-utility protocol already suffices for efficient communication. In particular, in a game with $n$ possible observations and $m$ actions: (1) For any achievable target utility $\alpha$, we give an algorithm with $\mathrm{poly}(n, m, 1/\epsilon)$ runtime that designs a protocol achieving utility at least $\alpha-\epsilon$ using only $2^{\mathcal O(CC_\alpha(G))}/\epsilon^2$ bits of communication. Here, $CC_\alpha(G)$ is the minimum number of bits used by any protocol, even a computationally inefficient one, to achieve utility $\alpha$. (2) We prove that this exponential dependence on $CC_\alpha(G)$ is tight up to a constant. That is, unless $\mathrm P=\mathrm{NP}$, no polynomial-time algorithm can in general find optimal protocols using fewer than $2^{CC_\alpha(G) -2}$ bits. We note that our results strictly weaken the assumptions required by prior work in the multi-agent information aggregation literature, filling a gap that had remained elusive even for games with constant $CC_\alpha(G)$. In particular, prior guarantees for agreement-based information aggregation rely on structural assumptions such as informational substitutes or weak learnability. We show that these assumptions already imply $CC_\alpha(G) = O(1)$ and are therefore more restrictive conditions than required by our protocol to succeed. On a technical level, our results involve a novel strengthening of the Frieze-Kannan weak regularity lemma and yield the following powerful polynomial-time transformation tool: for every communication game $G$, it constructs a game $\hat G$ that is a coarsening of the agents' observation spaces into constant-size partitions, such that $G$ and $\hat G$ are indistinguishable with respect to every short communication protocol. This coarsening theorem is the engine behind our algorithm and may be of independent interest.