betaFeedback

WorksMathematics

Descending Everest, Part II: Oracle Complexity of Minimizing a Sum of Convex (L0,L1)(L_0,L_1)-Smooth Functions

Abstract

We study the minimization of the average of nn convex (L0,L1)(L_0,L_1)-smooth functions: the Hessian norm of each function is bounded by an affine function of its gradient norm. We assume known upper bounds RR on the distance from the starting point to a minimizer and δδ on the difference between the minimum of the average and the average of the infima of the components. We derive upper and lower bounds on the oracle complexity of reaching accuracy ε\varepsilon in function value in expectation. Up to logarithmic factors, both bounds are of order n(1+(L1R)2/3)+n(L0+L12δ)R2ε+n(L1R)2n(1+(L_1R)^{2/3}) +\sqrt{\frac{n(L_0+L_1^2δ)R^2}{\varepsilon}+n(L_1R)^2}. The upper bound is attained by the proposed method YETI. The method combines ideas of accelerated proximal methods and variance reduction methods with a heuristic for choosing additional points. The number of oracle calls made by the heuristic is bounded, which preserves the upper bound on the total number of queries needed to reach the target accuracy. In the experiments, YETI is compared with first-order methods on synthetic problems and Poisson regression problems. We also examine the stopping criterion and the effect of the heuristic separately. The lower bound is obtained by an explicit construction of families of instances that are hard for any first-order method.

Use of language models. The models Codex Astra 6 and Claude Opus 5.5 were used in obtaining the theoretical results of this work. They were also used to prepare and typeset the manuscript and to translate the original text into English. Nevertheless, every line of the paper, including every statement, every proof, and every computation, has been personally edited and checked by the authors. The authors bear sole responsibility for the correctness of all mathematical statements, proofs, and conclusions.

Introduction

We consider the problem of minimizing a finite sum

f(x)=1n∑i=1nfi(x)→min⁡x∈Rd.f(x)=\frac{1}{n}\sum_{i=1}^{n}f_i(x)\to\min_{x\in\mathbb{R}^d}.

Problems of this form arise in empirical risk minimization in machine learning, where fif_i is the loss on the ii-th example of the training set [31, 6]. In federated and distributed learning, the component fif_i is the loss on the data of the ii-th device or computing node [24]. The gradient of a single component is often much cheaper to compute than the full gradient. Therefore, we measure the complexity of methods by the number of calls to an oracle that returns the value and the gradient of a single component.

For a sum of LL-smooth functions, variance reduction methods [16, 38] make it possible to reduce the number of oracle calls. They periodically compute the full gradient of ff, and at the remaining steps they use the gradients of individual components fif_i. Lower bounds for arbitrary randomized methods show which dependencies on the number of components and on the accuracy cannot be improved [37].

However, the LL-smoothness condition fails, for example, for exponential loss functions. The (L0,L1)(L_0, L_1)-smoothness condition was introduced by Zhang et al. [41] to explain the success of methods with gradient clipping. When training an LSTM-based language model, the authors observed that the local smoothness constant grows together with the gradient norm. A similar dependence was also found when training transformers [10]. The (L0,L1)(L_0, L_1)-smoothness condition admits precisely this kind of curvature growth. Its generalizations and first-order methods for such functions were studied in [19], [14], [36], Tyurin [34]. For finite sums, one has to account simultaneously for the number of components and for the growth of curvature. For example, in Poisson regression with the log link (Stan Development Team) fi(x)=exp⁡(aiTx)−biaiTxf_i(x) = \exp(a_i^\mathsf{T}x) - b_i a_i^\mathsf{T}x, where bi≥0b_i \ge0, the Hessian is unbounded on the whole space, but

∥∇2fi(x)∥≤bi∥ai∥2+∥ai∥∥∇fi(x)∥.\lVert\nabla^2 f_i(x)\rVert\le b_i\lVert a_i\rVert^2 + \lVert a_i\rVert\lVert\nabla f_i(x)\rVert.

In phase retrieval from intensity measurements, one uses the losses fi(x)=(⟨ai,x⟩2−bi)2f_i(x) = (\langle a_i,x\rangle^2 - b_i)^2 [7]. Here, too, one has to minimize a finite sum of components with unbounded curvature. Another example is the exponential loss in classification [14 Example 1.5]. In all these problems, when the number of observations is large, computing the full gradient may be expensive, and the gradient itself is not globally Lipschitz. Therefore, one needs methods that account for the growth of curvature and do not compute the gradients of all components at every step. Below we assume that ff attains its minimum, and that the distance to the minimizer, the initial optimality gap, and the heterogeneity of the components are bounded.

The peculiarity of finite sums. In the problem under consideration, each component satisfies the inequality

0⪯∇2fi(x)⪯(L0,i+L1∥∇fi(x)∥)I.0 \preceq\nabla^2 f_i(x) \preceq(L_{0,i} + L_1\lVert\nabla f_i(x)\rVert)I.

Even at a minimizer of the average, the gradients of individual components can be large: the equality ∇f(x∗)=0\nabla f(x^*) = 0 only means that their sum is zero. Therefore, a small norm of the full gradient does not imply small smoothness constants of the components, on which the bounds of variance reduction methods depend. What is the oracle complexity of minimizing a finite sum in this situation? We answer this question in this work.

Notation. We denote by ε>0\varepsilon> 0 the required accuracy of the solution in function value: the method must output a point x^\hat{x} such that f(x^)−f∗≤εf(\hat{x}) - f^* \le\varepsilon. Throughout, ∥⋅∥\lVert\cdot\rVert denotes the Euclidean norm of a vector and the spectral norm of a matrix, [x,y][x,y] is the segment with endpoints x,y∈Rdx, y \in\mathbb{R}^d, and B(x,r)B(x,r) is the closed ball of radius rr centered at xx. We denote the sum of the first kk terms of a sequence by the corresponding capital letter: for example, Ak=a1+⋯+akA_k = a_1 + \cdots+ a_k and A0=0A_0 = 0.

For randomized methods, the stated accuracy is required in expectation. All projections are Euclidean; ΠC\Pi_C denotes the projection onto a nonempty closed convex set CC.

Main result. Let the components be convex, twice continuously differentiable, and bounded from below, and let n−1∑iL0,i≤L0n^{-1}\sum_i L_{0,i} \le L_0. We assume that the starting point x0x_0 is zero: this can be achieved by a shift, with respect to which the problem class is invariant. We assume that ff attains its minimum at a point x∗x^* with ∥x∗∥≤R\lVert x^* \rVert\le R and f(0)−f∗≤Δf(0)-f^* \le\Delta. Under the conditions of Theorem 2.9, the upper and lower complexity bounds are, up to logarithmic factors, of order

n(1+(L1R)2/3)+n(L0+L12δ)R2ε+n(L1R)2.n\left(1+(L_1R)^{2/3}\right)+\sqrt{\frac{n(L_0+L_1^2\delta)R^2}{\varepsilon}+n(L_1R)^2}.

Here, complexity means the worst-case expected number of queries of values and gradients of individual components that suffices to guarantee E[f(x^)−f∗]≤ε\mathbb{E}[f(\hat{x})-f^*] \le\varepsilon. The lower bound holds for arbitrary randomized methods; the dimension of the problem is not bounded. The precise definition of the class and the logarithmic factors are given in § 2.

Note on the logarithmic factors. In the upper bound, the hidden factor is log⁡2N\log^2 N, where NN from (5) grows linearly in Δ/ε\Delta/\varepsilon. In the worst case, when Δ\Delta is exponentially large in L1RL_1R, this logarithm can be of order L1RL_1R, so the factor log⁡2N\log^2 N can be of order (L1R)2(L_1R)^2, that is, it can give an additional polynomial factor.

The term nL0R2/ε\sqrt{nL_0R^2/\varepsilon} is known from the theory of minimizing a sum of LL-smooth functions ([37]. The term n(L1R)2/3n(L_1R)^{2/3} reflects the number of oracle calls required to approach the minimum from the region where the gradients are exponentially large. The term

nL12R2δε+n(L1R)2\sqrt{\frac{nL_1^2R^2\delta}{\varepsilon}+n(L_1R)^2}

arises because the curvature of the components near the minimum is determined by their gradients, which can remain large even at the minimizer of the average.

Method and experiments. The upper bound is attained by the method YETI. When the local curvature is small, it performs an accelerated step, solving the auxiliary subproblem by SVRG, and when it is large, it projects the auxiliary point onto a halfspace containing the minimizer. In addition, YETI uses a heuristic with a bounded number of oracle queries: it makes it possible to take advantage of good proposed points without worsening the theoretical bound. In the experiments, YETI is compared with other methods, and its stopping criterion is tested.

Relation to known methods. YETI uses the accelerated proximal scheme of Monteiro–Svaiter ([25]) and the idea of adaptive choice of regularization without solving an implicit equation at every step ([9]). The local subproblem is solved by a projected variant of SVRG. Acceleration via approximate solution of regularized subproblems was also studied in the Catalyst scheme ([20, 21]) and in Fros­tig et al. ((2015)). In YETI, the choice of the step depends on the computed curvature estimate of the sum: if this estimate is too large, the method constructs a separating halfspace instead of a step. The complexity bound accounts for the queries of both the local solver and the heuristic.

The lower bound relies on Yao’s minimax principle [39]: a hard instance is chosen at random from a finite family, after which we average over the random variables of the algorithm. Hidden coordinates and the distribution of information across different components continue the known approach to constructing lower bounds for finite-sum minimization [37]. For the class under consideration, we apply exponential transformations of convex chains and choose their parameters so that the constraints on L1L_1, δ\delta, and the initial optimality gap are satisfied.

The present paper continues our previous work Rudomanov et al. [30], where the oracle complexity of this class was found for a single function (n=1n=1): its order is L0R2/ε+L1R\sqrt{L_0R^2/\varepsilon}+L_1R, and for a small initial optimality gap it is L0R2/ε+(L1R)2/3\sqrt{L_0R^2/\varepsilon}+(L_1R)^{2/3}. Here we consider a finite sum, and the oracle returns the value and the gradient of only one component; therefore, to obtain the upper bound, we use variance reduction. The results for a single function [36, 35] concern an oracle that computes the full gradient. In particular, the optimality of the method of Tyurin [35] for sufficiently small ε\varepsilon is based on the classical lower bound for the smooth subclass and concerns only the leading term L0R2/ε\sqrt{L_0R^2/\varepsilon}. The optimality of the term n(L1R)2/3n(L_1R)^{2/3} for an oracle returning the gradient of a single component does not follow from it. Similarly, the bounds for stochastic methods in the case of a common minimizer of all components [14] do not cover an arbitrary positive heterogeneity δ\delta.

Organization of the paper. The problem setting and the main result are given in § 2, the method YETI and its analysis in § 3–§ 6, the main experiments in § 7 (further experiments and their details are in Appendix § 18), and the lower bound in § 8. The main text contains the results and a proof sketch of the lower bound, while auxiliary proofs and variants of the method are deferred to the appendices. A summary table of known bounds is given in Appendix § 19.

Problem setting and main bounds

We consider the function f=n−1∑i=1nfif=n^{-1}\sum_{i=1}^{n}f_i on Rd\mathbb{R}^d. We are given n≥2n\ge2, L0,δ≥0L_0,\delta\ge0, and L1,R,Δ,ε>0L_1,R,\Delta,\varepsilon>0. We assume that the starting point is zero; this can be achieved by a shift of coordinates. All general assumptions are collected below.

Assumption 2.1 (components). The functions fi:Rd→Rf_i:\mathbb{R}^d\to\mathbb{R}, 1≤i≤n1\le i\le n, are convex, twice continuously differentiable, and bounded from below.

Assumption 2.2 (generalized smoothness). Numbers L0,i≥0L_{0,i}\ge0 are known such that

∥∇2fi(x)∥≤L0,i+L1∥∇fi(x)∥(x∈Rd, 1≤i≤n),\left\|\nabla^2 f_i(x)\right\|\le L_{0,i}+L_1\left\|\nabla f_i(x)\right\| \qquad(x\in\mathbb{R}^d,\ 1\le i\le n),
1n∑i=1nL0,i≤L0.(1)\frac{1}{n}\sum_{i=1}^{n}L_{0,i}\le L_0. \tag*{(1)}

Assumption 2.3 (minimizer). There exists a minimizer x∗∈arg min⁡fx^*\in\operatorname*{arg\,min} f such that ∥x∗∥≤R\left\|x^*\right\|\le R. We denote f∗=f(x∗)f^*=f(x^*).

Assumption 2.4 (initial optimality gap). We have f(0)−f∗≤Δf(0)-f^* \le\Delta, and Δ≥2ε\Delta\ge2\varepsilon.

Assumption 2.5 (heterogeneity).

f∗−1n∑i=1ninf⁡x∈Rdfi(x)≤δ.(2)f^*-\frac{1}{n}\sum_{i=1}^{n}\inf_{x\in\mathbb{R}^d} f_i(x) \le\delta. \tag*{(2)}

Assumption 2.6 (range of accuracies).

L12R2δε≥3.(3)\frac{L_1^2 R^2 \delta}{\varepsilon} \ge3. \tag*{(3)}

Definition 2.7 (problem class). We denote by Pd\mathcal{P}_d the set of collections of functions f1,…,fn:Rd→Rf_1,\ldots,f_n:\mathbb{R}^d\to\mathbb{R} and known numbers L0,1,…,L0,nL_{0,1},\ldots,L_{0,n} satisfying Assumptions 2.1–2.5.

Assumption 2.2 bounds only the average of the numbers L0,iL_{0,i}; individual L0,iL_{0,i} may exceed L0L_0. It suffices to construct the lower bound on the subclass L0,i=L0L_{0,i}=L_0. For C2C^2 functions, the Hessian bound from Assumption 2.2 is equivalent to symmetric (L0,i,L1)(L_{0,i},L_1)-smoothness [14 Assumption 1.3, Eq. (5]:

References

  1. [1]Artem Agafonov, Aslan Ashabokov, Alexander D’yakonov, Martin Takáč, Alexander Gasnikov, and Dmitry Kamzolov. Optimal parameter-free second-order acceleration with inexact hessians under generalized smoothness. arXiv:2609.32977, 2026. URL https://arxiv.org/abs/2609.32977.
  2. [2]Zeyuan Allen-Zhu. Katyusha: The first direct acceleration of stochastic gradient methods. Journal of Machine Learning Research, 18(221):1–51, 2018. URL https://www.jmlr.org/papers/v18/16-410.html.
  3. [3]Jean-Bernard Baillon and Georges Haddad. Quelques propriétés des opérateurs angle-bornés et n-cycliquement monotones. Israel Journal of Mathematics, 26(2):137–150, 1977. doi: 10.1007/BF03007664.DOI
  4. [4]Heinz H. Bauschke and Patrick L. Combettes. The Baillon–Haddad theorem revisited. Journal of Convex Analysis, 17(3–4):781–787, 2010. URL https://pcombet.math.ncsu.edu/jca3.pdf.arxiv.org/abs/0906.0807
  5. [5]Ekaterina Borodich and Dmitry Kovalev. Nesterov finds GRAAL: Optimal and adaptive gradient method for convex optimization. In International Conference on Learning Representations, 2026. URL https://openreview.net/forum?id=vPSiCA3CkD.arxiv.org/abs/2507.09823
  6. [6]Léon Bottou, Frank E. Curtis, and Jorge Nocedal. Optimization methods for large-scale machine learning. SIAM Review, 60(2):223–311, 2018.arxiv.org/abs/1606.04838
  7. [7]Emmanuel J. Candès, Xiaodong Li, and Mahdi Soltanolkotabi. Phase retrieval via wirtinger flow: Theory and algorithms. IEEE Transactions on Information Theory, 61(4):1985–2007, 2015. doi: 10.1109/TIT.2015.2399924.DOI
  8. [8]Yair Carmon, John C. Duchi, Oliver Hinder, and Aaron Sidford. Lower bounds for finding stationary points I. Mathematical Programming, 184:71–120, 2020. URL https://arxiv.org/abs/1710.11606.
  9. [9]Yair Carmon, Danielle Hausler, Arun Jambulapati, Yujia Jin, and Aaron Sidford. Optimal and adaptive Monteiro–Svaiter acceleration. In Advances in Neural Information Processing Systems, volume 35, 2022. URL https://arxiv.org/abs/2205.15371.
  10. [10]Michael Crawshaw, Mingrui Liu, Francesco Orabona, Wei Zhang, and Zhenxun Zhuang. Robustness to unbounded smoothness of generalized SignSGD. arXiv:2208.11195, 2022.arxiv.org/abs/2208.11195
  11. [11]Aaron Defazio, Francis Bach, and Simon Lacoste-Julien. SAGA: A fast incremental gradient method with support for non-strongly convex composite objectives. In Advances in Neural Information Processing Systems, volume 27, 2014. URL https://proceedings.nips.cc/paper_files/paper/2014/hash/937964195d6fb3a55cd7cc578165f058-Abstract.html.
  12. [12]Elizabeth D. Dolan and Jorge J. Moré. Benchmarking optimization software with performance profiles. Mathematical Programming, 91(2):201–213, 2002. doi: 10.1007/s101070100263.DOI
  13. [14]Eduard Gorbunov, Nazarii Tuptitsa, Sayantan Choudhury, Alen Aliev, Peter Richtárik, Samuel Horváth, and Martin Takáč. Methods for convex (L₀, L₁)-smooth optimization: Clipping, acceleration, and adaptivity. In International Conference on Learning Representations, 2025. arXiv:2409.14989.arxiv.org/abs/2409.14989
  14. [15]Martin Jaggi. Revisiting Frank–Wolfe: Projection-free sparse convex optimization. In Proceedings of the 30th International Conference on Machine Learning, volume 28 of Proceedings of Machine Learning Research, pages 427–435, 2013. URL https://proceedings.mlr.press/v28/jaggi13.html.
  15. [16]Rie Johnson and Tong Zhang. Accelerating stochastic gradient descent using predictive variance reduction. In Advances in Neural Information Processing Systems, volume 26, pages 315–323, 2013. URL https://proceedings.neurips.cc/paper/2013/hash/ac1dd209cbbc5e5d1c6e28598e8cbbe8-Abstract.html.
  16. [17]Dmitry Kovalev. Stochastic non-smooth convex optimization with unbounded gradients. arXiv:2605.15522, 2026. URL https://arxiv.org/abs/2605.15522.
  17. [18]Guanghui Lan, Zhize Li, and Yi Zhou. A unified variance-reduced accelerated gradient method for convex optimization. In Advances in Neural Information Processing Systems, volume 32, 2019. URL https://arxiv.org/abs/1905.12412.
  18. [19]Haochuan Li, Jian Qian, Yi Tian, Alexander Rakhlin, and Ali Jadbabaie. Convex and non-convex optimization under generalized smoothness. In Advances in Neural Information Processing Systems, volume 36, 2023.arxiv.org/abs/2306.01264
  19. [20]Hongzhou Lin, Julien Mairal, and Zaid Harchaoui. A universal catalyst for first-order optimization. In Advances in Neural Information Processing Systems, volume 28, 2015. URL https://proceedings.neurips.cc/paper_files/paper/2015/hash/c164bbc9d6c72a52c599bbb43d8db8e1-Abstract.html.
  20. [21]Hongzhou Lin, Julien Mairal, and Zaid Harchaoui. Catalyst acceleration for first-order convex optimization: from theory to practice. Journal of Machine Learning Research, 18(212):1–54, 2018. URL https://arxiv.org/abs/1712.05654.
  21. [22]Dong C. Liu and Jorge Nocedal. On the limited memory BFGS method for large scale optimization. Mathematical Programming, 45:503–528, 1989. doi: 10.1007/BF01589116.
  22. [23]Yura Malitsky and Konstantin Mishchenko. Adaptive gradient descent without descent. In Proceedings of the 37th International Conference on Machine Learning, volume 119 of Proceedings of Machine Learning Research, pages 6702–6712, 2020. URL https://proceedings.mlr.press/v119/malitsky20a.html.arxiv.org/abs/1910.09529
  23. [24]H. Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Agüera y Arcas. Communication-efficient learning of deep networks from decentralized data. In Proceedings of the 20th International Conference on Artificial Intelligence and Statistics, volume 54 of Proceedings of Machine Learning Research, pages 1273–1282, 2017.arxiv.org/abs/1602.05629
  24. [25]Renato D. C. Monteiro and Benar F. Svaiter. An accelerated hybrid proximal extragradient method for convex optimization and its implications to second-order methods. SIAM Journal on Optimization, 23(2):1092–1125, 2013. doi: 10.1137/110833786. URL https://w3.impa.br/~benar/docs/SIAM/accHPE_Optimp.pdf.DOI
  25. [26]Yurii Nesterov. Lectures on Convex Optimization, volume 137 of Springer Optimization and Its Applications. Springer, 2018.
  26. [27]Jorge Nocedal and Stephen J. Wright. Numerical Optimization. Springer, 2 edition, 2006. doi: 10.1007/978-0-387-40065-5.DOI
  27. [28]Irina Podlipnova, Maxim Mashtaler, Artem Agafonov, Yuriy Dorn, and Alexander Gasnikov. Smooth or separable? sparse block acceleration for entropy-regularized linear programs. arXiv:2609.32558, 2026. URL https://arxiv.org/abs/2609.32558.
  28. [29]Amirhossein Reisizadeh, Haochuan Li, Subhro Das, and Ali Jadbabaie. Variance-reduced clipping for non-convex optimization. arXiv:2303.00883, 2023.arxiv.org/abs/2303.00883
  29. [30]M. Rudomanov, I. Fomin, A. V. Gasnikov, and F. S. Stonyakin. Descending Everest with first-order oracles: Oracle complexity of convex optimization under (l₀,l₁)-smoothness. arXiv:2610.21461, 2026. URL https://arxiv.org/abs/2610.21461.
  30. [31]Shai Shalev-Shwartz and Shai Ben-David. Understanding Machine Learning: From Theory to Algorithms. Cambridge University Press, 2014.
  31. [33]Artem Tsedenov and Dmitry Kovalev. An optimal second-order algorithm for finite-sum optimization under average smoothness and convexity. arXiv:2610.10495, 2026. URL https://arxiv.org/abs/2610.10495.
  32. [34]Alexander Tyurin. Toward a unified theory of gradient descent under generalized smoothness. In Proceedings of the 42nd International Conference on Machine Learning, volume 267 of Proceedings of Machine Learning Research, pages 60493–60514, 2025. URL https://proceedings.mlr.press/v267/tyurin25a.html.arxiv.org/abs/2412.11773
  33. [35]Alexander Tyurin. Near-optimal convergence of accelerated gradient methods under generalized and (L₀,L₁)-smoothness. In Proceedings of the 43rd International Conference on Machine Learning, volume 306 of Proceedings of Machine Learning Research, pages 122876–122905, 2026. URL https://proceedings.mlr.press/v306/tyurin26b.html.
  34. [36]Daniil Vankov, Anton Rodomanov, Angelia Nedić, Lalitha Sankar, and Sebastian U. Stich. Optimizing (L₀,L₁)-smooth functions by gradient methods. In International Conference on Learning Representations, 2025. arXiv:2410.10800.
  35. [37]Blake Woodworth and Nathan Srebro. Tight complexity bounds for optimizing composite objectives. In Advances in Neural Information Processing Systems, volume 29, 2016. URL https://arxiv.org/abs/1605.08003.
  36. [38]Lin Xiao and Tong Zhang. A proximal stochastic gradient method with progressive variance reduction. SIAM Journal on Optimization, 24(4):2057–2075, 2014. doi: 10.1137/140961791. URL https://arxiv.org/abs/1403.4699.
  37. [39]Andrew Chi-Chih Yao. Probabilistic computations: Toward a unified measure of complexity. In 18th Annual Symposium on Foundations of Computer Science, pages 222–227, 1977. doi: 10.1109/SFCS.1977.24. URL https://doi.org/10.1109/SFCS.1977.24.
  38. [40]Dingzhi Yu, Wei Jiang, Hongyi Tao, Yuanyu Wan, and Lijun Zhang. Mirror descent under generalized smoothness. In Proceedings of the 43rd International Conference on Machine Learning, 2026. URL https://arxiv.org/abs/2502.00753. arXiv:2502.00753v4.
  39. [41]Jingzhao Zhang, Tianxing He, Suvrit Sra, and Ali Jadbabaie. Why gradient clipping accelerates training: A theoretical justification for adaptivity. In International Conference on Learning Representations, 2020. arXiv:1905.11881.arxiv.org/abs/1905.11881

Paper details

vlnc:W_ctnu7rvy
Version
v2 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 19

downloads · 4

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Codex Astra 6, Claude Opus 5.5

View statement
Looking for related papers…
Contents