betaFeedback

WorksMathematics

Setting Things Straight: On the Straightness and Non-Straightness of Two-Step Rectified Flow

Abstract

Rectified flow (RF) is an ODE-based generative modeling framework that has achieved remarkable success in practice over the past few years. By design, RF seeks to produce flows that have a high degree of straightness, in the sense that the resulting ODE trajectories can be obtained with relatively few (ideally, one) sampling steps. However, a rigorous analysis of straightness or near-straightness of RF flows remains missing in the literature. In this paper, we study settings in which the coupling generated by two-step RF is provably straight at the population level. We provide a complete characterization in two popular classes of data distributions: (i) smooth, strongly log-concave densities and (ii) Gaussian mixture models. Our main results yield sharp conditions for the coupling generated by two-step RF to be straight. Non-straight couplings are constructed when these conditions fail, which demonstrate the sharpness of these conditions. Using the theory developed on the straightness of two-step RF, we propose a notion of approximate straightness that measures the straightness of the learned couplings and can be estimated from the data. We introduce a procedure for constructing asymptotic confidence intervals for our proposed notion of approximate straightness and propose a data-driven method for selecting the number of reflow rounds in multiple iterations of RF.

  • Equal contribution

Corresponding authors: Tuan Pham ([email protected]) and Saptarshi Roy ([email protected]).

Introduction

Generative models have gained substantial attention over the last few years due to their impressive performance in drawing approximate samples from complex probability distributions. Early successes in deep generative modeling include variational autoencoders (VAE) [27] and generative adversarial networks (GAN) [20], which learn data distributions through latent-variable modeling and adversarial training, respectively. Diffusion and score-based models subsequently emerged as powerful approaches that generate samples by learning to reverse a process that gradually transforms data into noise [53, 25]. Their continuous-time formulations connect generative modeling to stochastic and ordinary differential equations (ODEs) that transport a simple reference distribution to the target data distribution [54].

In the past few years, Rectified Flow or Flow matching [36, 38, 33] has emerged as a prominent sampling methodology due to its state-of-the-art generative performance across various modalities including image generation [39, 60], audio generation [59, 34], and video generation [9, 35]. Apart from applications, RF has also attracted several strands of theoretical research including low-dimensional adaptivity [5, 50, 28], sampling performance [2, 58, 21, 29], alignment [17, 48], optimal distillation [49] and many more. The success of RF stems from two key factors: (1) RF is a deterministic ordinary differential equation (ODE)-based sampler that is easier to simulate compared to other stochastic samplers, and (2) RF learns almost straight trajectories between source and target data distributions via multiple reflow steps, leading to faster sampling. In particular, the straightness or near-straightness of its ODE trajectories is critical: flows with nearly straight trajectories require fewer sampling steps to generate new samples and are thus more efficient. Though this feature of RF is generally recognized as a main reason for its empirical successes and wide applicability, the theoretical understanding of the conditions under which the RF trajectories are guaranteed to be (near) straight or not has remained limited. In this paper we make a step forward in filling this gap by providing a rigorous framework for understanding the straightness of RFs. The key message is that our framework yields sharp conditions for straightness and identifies precisely when non-straightness can happen. We demonstrate the use of our framework on two common classes of data distributions: smooth, strongly log-concave distributions and Gaussian mixture models (GMMs).

Rectified flows Let us briefly recall the RF framework. In this generative modeling framework, we observe X1,…,Xn∼pdata\mathbf{X}_1,\ldots,\mathbf{X}_n \sim p_{\mathrm{data}}, where pdatap_{\mathrm{data}} is an unknown distribution on Rd\mathbb{R}^d, and aim to generate new samples approximately from this distribution. The RF framework learns a velocity field v(x,t)\mathbf{v}(x,t) by minimizing the following objective:

v=arg min⁡f∫01E∥X1−X0−f(Xt,t)∥2dt,(1)\mathbf{v}=\underset{f}{\operatorname{arg\,min}}\int_{0}^{1}\mathbb{E}\left\|\mathbf{X}_1-\mathbf{X}_0-f(\mathbf{X}_t,t)\right\|^2\mathrm{d}t, \tag*{(1)}

where

Xt=tX1+(1−t)X0,(X0,t)∼N(0,Id)×Unif⁡(0,1),\mathbf{X}_t=t\mathbf{X}_1+(1-t)\mathbf{X}_0,\qquad(\mathbf{X}_0,t)\sim\mathcal{N}(0,\mathbf{I}_d)\times\operatorname{Unif}(0,1),

with (X0,t)(\mathbf{X}_0,t) independent of the data. In the objective (1) above, Xt\mathbf{X}_t is formed by interpolating the same pair (X0,X1)(\mathbf{X}_0,\mathbf{X}_1) that appears in the displacement X1−X0\mathbf{X}_1-\mathbf{X}_0. For each tt, the above objective in (1) is minimized at

Schematic of the iterative reflow procedure for constructing $K$-RF

Figure 1. Schematic of the iterative reflow procedure for constructing KK-RF.

(x,t)↦vt(x):=v(x,t)=E[X1−X0∣Xt=x].(2)(x,t) \mapsto v_t(x) := v(x,t) = \mathbb{E}[\mathbf{X}_1-\mathbf{X}_0 \mid X_t=x]. \tag*{(2)}

Under appropriate regularity conditions (investigated in Bansal et al. [ [2]; Mena et al. [ [42]) the velocity is differentiable and the ODE

dZt=vt(Zt) dt,Z0∼p0=N(0,Id)(3)\mathrm{d}Z_t = v_t(Z_t)\,\mathrm{d}t,\qquad\mathbf{Z}_0 \sim p_0 = \mathcal{N}(\mathbf{0},I_d) \tag*{(3)}

has a unique solution and satisfies the marginal preserving property, i.e., Law⁡(Zt)=Law⁡(Xt)=pt\operatorname{Law}(\mathbf{Z}_t) = \operatorname{Law}(\mathbf{X}_t) = p_t. Under a mild non-explosivity assumption on the ODE and the second moment assumption on pdatap_{\mathrm{data}}, Bansal et al. [ [2] showed that ODE (3) admits a unique solution. Throughout this paper, we assume that the above ODE has a unique solution, which is also necessary for a coupling to be rectifiable [36, 38].

The initial fitting procedure yields a learned velocity field, which we denote by v(1)v^{(1)}. More generally, after kk rectification rounds we write v(k)v^{(k)} for the learned velocity field at round kk. To simplify notation, when discussing the first round we will write v:=v(1)v := v^{(1)}. We refer to the endpoint coupling induced by v(1)v^{(1)} as the 1-RF coupling:

1-RF:=(X0(1),Z1(1)(X0(1))),X0(1)∼N(0,Id),\mathrm{1\text{-}RF} := \left(\mathbf{X}_0^{(1)},\mathbf{Z}_1^{(1)}\left(\mathbf{X}_0^{(1)}\right)\right),\qquad\mathbf{X}_0^{(1)} \sim\mathcal{N}(\mathbf{0},I_d),

where the flow map Zt(1)\mathbf{Z}_t^{(1)} solves

ddtZt(1)(x)=v(1)(Zt(1)(x),t),Z0(1)(x)=x.\frac{\mathrm{d}}{\mathrm{d}t}\mathbf{Z}_t^{(1)}(x) = v^{(1)}\left(\mathbf{Z}_t^{(1)}(x),t\right),\qquad\mathbf{Z}_0^{(1)}(x)=x.

This coupling is deterministic in the sense that, conditional on the learned velocity field, the terminal state is a deterministic function of the initial state X0(1)\mathbf{X}_0^{(1)}. We can iterate this procedure to obtain KK-RF by performing K−1K-1 additional reflow rounds, as illustrated below: At each subsequent round, a new velocity field is fitted using the paired initial and terminal states generated by the preceding learned flow. We refer readers to Liu et al. [ [38] for a comprehensive description of the procedure and its implementation. Once the procedure stops after the KK-th rectification, new samples are generated by

drawing X0(K)∼N(0,Id) and computing Z1(K)(X0(K)).\text{drawing }\mathbf{X}_0^{(K)} \sim\mathcal{N}(\mathbf{0},I_d)\text{ and computing }\mathbf{Z}_1^{(K)}\left(\mathbf{X}_0^{(K)}\right).

This can be done through numerical integration of the ODE associated with the final learned velocity field v^(K)\hat{v}^{(K)}. In practice, 2-RF or 3-RF can already produce high-quality samples [38]. Interestingly, repeated rectifications seem to increase the straightness; see Theorem D.7 in Liu et al. [ [38].

Despite the impressive empirical success of RF, theoretical understanding and guarantees regarding its straightness remained limited until recently, when Bansal et al. [ [2] developed a framework for studying RF straightness. In particular, they showed that 2-RF produces straight flows under certain technical assumptions on the target distribution and the velocity field. A substantial limitation of these results is that they provide only a partial characterization of the conditions under which RF produces straight flows. Furthermore, they offer limited practical guidance for assessing the straightness of trajectories generated by a learned velocity field, which is an important consideration in determining how many rectification steps are needed and whether RF is an appropriate generative modeling framework for a given dataset.

The goal of the present article is to provide a general, simple framework for understanding the straightness property of RF under some common classes of data distributions via the concept of “maximum rotation” which borrows ideas from analysis. In particular, we will be concerned with the following questions:

(i) Given a data distribution pdatap_{\mathrm{data}}, when does RF produce straight couplings after two rectification steps, as one observes in practice? If not, can it be non-straight even for very regular data distributions?

(ii) Based on the data X1,…,Xn∈RdX_1,\ldots,X_n \in\mathbb{R}^d, how many rectification steps K≥1K \ge1 are needed so that the learned velocity fields after KK steps give approximately straight couplings that are good enough to use for sampling?

Our results make both theoretical and methodological contributions, which can be summarized as follows.

  1. We develop a general theory for proving the straightness of two-step RF by bounding the maximum angular rotation of the endpoint map. We apply our theory to two concrete, popular classes of distributions: (a) smooth, strongly log-concave distributions and (b) Gaussian mixture models with isotropic covariance matrices.

Using the maximum rotation framework, geometric conditions under which 2-RF produces exactly straight couplings (at the population level) are obtained for these two classes of distributions. We find that the condition number of the log-concave distribution and the maximum cluster mean separation in the case of GMM are the crucial quantities for characterizing the straightness of RF. Additionally, we show that our conditions are sharp by constructing non-straight flows when these geometric conditions fail. This answers question (i) in full generality. To the best of our knowledge, this is the first construction of non-straight couplings obtained from 2-RF.

  1. We consider the problem of assessing the straightness of a learned velocity field, which is crucial for determining the number of rectification steps upon observing the data. We propose a notion of “approximate straightness” and construct a consistent estimator to quantify it. Using this estimator, we construct valid confidence intervals and develop a data-driven stopping rule based on a prescribed straightness tolerance.

  1. On the technical level, the framework developed here is useful for understanding both straightness and non-straightness of the couplings produced by RF. In particular, our results provide the first sharp characterization of RF straightness. Proposition 4.1 below provides a route to proving that a coupling is non-straight, and we expect it to be relevant to other settings beyond those considered in this paper.

Notation. Throughout the paper, with some abuse of notation, we will identify a probability distribution with its Lebesgue density, which we will assume exists unless otherwise noted. For a distribution (Lebesgue density) pp, write supp⁡(p)\operatorname{supp}(p) to denote the support of pp. For a random variable XX we use Law⁡(X)\operatorname{Law}(X) to denote its distribution. For a function f=(f1,…,fd):Rd→Rd\boldsymbol{f} = (f_1,\ldots,f_d): \mathbb{R}^d \to\mathbb{R}^d differentiable at xx, we denote by Df(x)Df(x) the Jacobian matrix of ff at xx

Df(x):=(∂fi∂xj(x))i,j=1d.Df(x) := \left(\frac{\partial f_i}{\partial x_j}(x)\right)_{i,j=1}^{d}.

When there are many variables, we will also write DxD_x to indicate that the Jacobian matrix is taken with respect to xx. In the case where the function takes real values, we use ∇\nabla to indicate its gradient and ∇2\nabla^2 to indicate its Hessian matrix.

For a set {xj}j∈[K]⊂R\{x_j\}_{j\in[K]} \subset\mathbb{R} and probability vector π:=(π1,…,πK)\boldsymbol{\pi} := (\pi_1,\ldots,\pi_K), we denote the variance of the set by

Var⁡π({xj}j∈[K])=∑j∈[K]πjxj2−(∑j∈[K]πjxj)2.\operatorname{Var}_{\pi}(\{x_j\}_{j\in[K]}) = \sum_{j\in[K]} \pi_j x_j^2 - \left(\sum_{j\in[K]} \pi_j x_j\right)^2.

For two vectors u,v\boldsymbol{u},\boldsymbol{v}, we denote by Angle⁡(u,v)∈[0,π]\operatorname{Angle}(\boldsymbol{u},\boldsymbol{v}) \in[0,\pi] the angle between u\boldsymbol{u} and v\boldsymbol{v}. Given a matrix M∈Rd×dM \in\mathbb{R}^{d\times d}, we also define the maximum angle

ϕM:=sup⁡u∈Rd:Mu≠0Angle⁡(u,Mu),(4)\phi_M := \sup_{\boldsymbol{u}\in\mathbb{R}^d: M\boldsymbol{u}\ne0} \operatorname{Angle}(\boldsymbol{u},M\boldsymbol{u}), \tag*{(4)}

which represents the maximum rotation achieved by the matrix MM. For two square matrices MM and NN, we write their commutator as [M,N]=MN−NM[M,N] = MN - NM. Finally, for a symmetric matrix MM, we identify its maximum and minimum eigenvalues as λmax⁡(M)\lambda_{\max}(M) and λmin⁡(M)\lambda_{\min}(M), respectively.

We use GLk(R)\mathrm{GL}_k(\mathbb{R}) to denote the group of real invertible matrices of size kk, i.e.

GLk(R):={A∈Rk×k:det⁡(A)≠0}.\mathrm{GL}_k(\mathbb{R}) := \left\{A \in\mathbb{R}^{k\times k}: \det(A) \ne0\right\}.

Similarly, the space of positive-definite matrices of size kk will be denoted by S++k\mathbb{S}_{++}^k:

S++k:={A:∃ε>0 s.t. A⪰ε⋅Ik}.\mathbb{S}_{++}^k := \left\{A : \exists\varepsilon> 0\ \text{s.t.}\ A \succeq\varepsilon\cdot I_k\right\}.

Both these spaces are regarded as open subsets of Rk2\mathbb{R}^{k^2} and Rk(k+1)/2\mathbb{R}^{k(k+1)/2}, respectively, when we talk about differentiability. For example, a map TT is said to be in C1([0,1],GLk(R))C^1([0,1],\mathrm{GL}_k(\mathbb{R})) if TT takes values in GLk(R)\mathrm{GL}_k(\mathbb{R}) and is C1C^1 as a map from [0,1][0,1] to Rk2\mathbb{R}^{k^2}.

Background and Preliminaries

Straight coupling and straight flow

Rectified flow interpolates between the two distributions in straight line paths as Xt=tX1+(1−t)X0X_t = tX_1 + (1-t)X_0 for t∈[0,1]t \in[0,1], giving vt(x)v_t(x) as in (2). Therefore, the resulting sampling ODE (3) also approximates a geodesic path between p0p_0 and p1p_1 that allows faster sampling. In the context of RFs, the optimization step (1) is known as a rectification step, and the solution path

Z:={Zt}t∈[0,1](5)\mathcal{Z} := \{Z_t\}_{t\in[0,1]} \tag*{(5)}

References

  1. [1]Luigi Ambrosio, Nicola Gigli, and Giuseppe Savaré. Gradient flows: in metric spaces and in the space of probability measures. Springer, 2005. URL https://doi.org/10.1007/b137080.
  2. [2]Vansh Bansal, Saptarshi Roy, Alessandro Rinaldo, and Purnamrita Sarkar. On the convergence and straightness of rectified flow. In Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, pages 4636–4644. PMLR, 2026. URL https://openreview.net/forum?id=qKxKzIugsy.
  3. [3]Rajendra Bhatia and Rajesh Sharma. Some inequalities for positive linear maps. Linear algebra and its applications, 436(6):1562–1571, 2012. URL https://doi.org/10.1016/j.laa.2011.12.017.
  4. [4]Christopher M. Bishop. Pattern Recognition and Machine Learning. Springer, 2006. URL https://link.springer.com/book/9780387310732.
  5. [5]Nicholas Boffi, Arthur Jacot, Stephen Tu, and Ingvar Ziemann. Shallow diffusion networks provably learn hidden low-dimensional structure. In International Conference on Learning Representations, volume 2025, pages 52889–52923, 2025. URL https://openreview.net/forum?id=K1xK4ncqWZ.arxiv.org/abs/2410.11275
  6. [6]Stéphane Boucheron, Gábor Lugosi, and Pascal Massart. Concentration Inequalities: A Nonasymptotic Theory of Independence. Oxford University Press, 2013. URL https://global.oup.com/academic/product/concentration-inequalities-9780199535255.
  7. [7]Antonin Chambolle and Johannes Hertrich. Limit points of reflow with minibatch optimal transport. arXiv preprint arXiv:2608.07042, 2026. URL https://arxiv.org/abs/2608.07042.
  8. [8]Fan Chen, Sinho Chewi, Constantinos Daskalakis, and Alexander Rakhlin. High-accuracy sampling for diffusion models and log-concave distributions. In Proceedings of the 43rd International Conference on Machine Learning, volume 306 of Proceedings of Machine Learning Research, pages 13835–13869. PMLR, 2026. URL https://proceedings.mlr.press/v306/chen26p.html.arxiv.org/abs/2602.01338
  9. [9]Shoufa Chen, Chongjian Ge, Yuqi Zhang, Yida Zhang, Fengda Zhu, Hao Yang, Hongxiang Hao, Hui Wu, Zhichao Lai, Yifei Hu, et al. Goku: Flow based video generative foundation models. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 23516–23527, 2025a. URL https://openaccess.thecvf.com/content/CVPR2025/html/Chen_Goku_Flow_Based_Video_Generative_Foundation_Models_CVPR_2025_paper.html.arxiv.org/abs/2502.04896
  10. [10]Sitan Chen, Vasilis Kontonis, and Kulin Shah. Learning general Gaussian mixtures with efficient score matching. In Proceedings of the Thirty-Eighth Conference on Learning Theory, volume 291 of Proceedings of Machine Learning Research, pages 1029–1090. PMLR, 2025b. URL https://proceedings.mlr.press/v291/chen25e.html.arxiv.org/abs/2404.18893
  11. [11]Sinho Chewi. Log-concave sampling, 2026. URL https://chewisinho.github.io/main.pdf. Book draft, version of August 20, 2026.
  12. [12]Sinho Chewi and Aram-Alexandre Pooladian. An entropic generalization of Caffarelli’s contraction theorem via covariance inequalities. Comptes Rendus. Mathématique, 361:1471–1482, 2023. doi: 10.5802/crmath.486. URL https://doi.org/10.5802/crmath.486.
  13. [13]Earl A. Coddington and Norman Levinson. Theory of Ordinary Differential Equations. McGraw-Hill, New York, 1955. URL https://archive.org/details/theoryofordinaryd00codd.
  14. [14]Arnak S. Dalalyan. Theoretical guarantees for approximate sampling from smooth and log-concave densities. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 79(3): 651–676, 2017. URL https://doi.org/10.1111/rssb.12183.
  15. [15]Alain Durmus, Szymon Maje wski, and Błażej Miasojedow. Analysis of Langevin Monte Carlo via convex optimization. Journal of Machine Learning Research, 20(73):1–46, 2019. URL https://www.jmlr.org/papers/v20/18-173.html.
  16. [16]Bradley Efron. Tweedie’s formula and selection bias. Journal of the American Statistical Association, 106(496):1602–1614, 2011. URL https://doi.org/10.1080/01621459.2011.622401.
  17. [17]Jiajun Fan, Shuaike Shen, Chaoran Cheng, Yuxin Chen, Chumeng Liang, and Ge Liu. Online reward-weighted fine-tuning of flow matching with wasserstein regularization. In The Thirteenth International Conference on Learning Representations, 2025. URL https://openreview.net/forum?id=2IoFFexvuw.arxiv.org/abs/2502.06061
  18. [18]Gerald B. Folland. Real Analysis: Modern Techniques and Their Applications. John Wiley & Sons, 2 edition, 1999. URL https://www.wiley.com/en-us/Real+Analysis%3A+Modern+Techniques+and+Their+Applications%2C+2nd+Edition-p-9780471317166.
  19. [19]Khashayar Gatmiry, S itan Chen, and Adil Salim. High-accuracy and dimension-free sampling with diffusions. In Proceedings of the 43rd International Conference on Machine Learning, volume 306 of Proceedings of Machine Learning Research, pages 34242–34274. PMLR, 2026. URL https://proceedings.mlr.press/v306/gatmiry26a.html.arxiv.org/abs/2601.10708
  20. [20]Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in Neural Information Processing Systems, volume 27, pages 2672–2680, 2014. URL https://arxiv.org/abs/1406.2661.
  21. [21]Yunrui Guan, Krishna Balasubramanian, and Shiqian Ma. Mirror flow matching with heavy-tailed priors for generative modeling on convex domains. In International Conference on Learning Representations, volume 2026, pages 130098–130124, 2026. URL https://openreview.net/forum?id=aZK17uc0XQ.
  22. [22]Godfrey Harold Hardy, John Edensor Littlewood, and George Pólya. Inequalities. Cambridge University Press, 1952. URL https://openlibrary.org/books/OL7715912M/Inequalities.
  23. [23]Johannes Heiny and Carolin Kleemann. Maximum interpoint distance of high-dimensional random vectors. Bernoulli, 31(1):537–560, 2025. URL https://doi.org/10.3150/24-BEJ1738.
  24. [24]Johannes Hertrich, Antonin Chambolle, and Julie Delon. On the relation between rectified flows and optimal transport. Advances in Neural Information Processing Systems, 38:172578–172610, 2026. URL https://papers.nips.cc/paper_files/paper/2026/hash/2f6c89b3d2f1f1696b5b2dd55f0cb0e4-Abstract-Conference.html.
  25. [25]Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. URL https://arxiv.org/abs/2006.11239.
  26. [26]Shin Seong Kim, Mingi Kwon, Jaeseok Jeong, and Youngjung Uh. Balanced conic rectified flow. In Advances in Neural Information Processing Systems, volume 38, 2025. URL https://openreview.net/forum?id=kqHzgTV9AU.
  27. [27]Diederik P. Kingma and Max Welling. Auto-encoding variational Bayes. In International Conference on Learning Representations, 2014. URL https://arxiv.org/abs/1312.6114.
  28. [28]Shivam Kumar, Yixin Wang, and Lizhen Lin. Flow matching is adaptive to manifold structures. arXiv preprint arXiv:2602.22486, 2026a. URL https://arxiv.org/pdf/2602.22486.
  29. [29]Syamantak Kumar, Dheeraj Nagaraj, Saptarshi Roy, and Purnamrita Sarkar. Conditional flow matching for transport between markov processes. arXiv preprint arXiv:2610.07229, 2026b. URL https://arxiv.org/abs/2610.07229.
  30. [30]Raz Kupferman and Asaf Shachar. A geometric perspective on the Piola identity in Riemannian settings. Journal of Geometric Mechanics, 11(1):59–76, 2019. doi: 10.3934/jgm.2019004.
  31. [31]Holden Lee, Andrej Risteski, and Rong Ge. Beyond log-concavity: Provable guarantees for sampling multi-modal distributions using simulated tempering Langevin Monte Carlo. In Advances in Neural Information Processing Systems, volume 31, 2018. URL https://proceedings.neurips.cc/paper/2018/hash/c6ede20e6f597abf4b3f6bb30cee16c7-Abstract.html.
  32. [32]Sangyun Lee, Zinan Lin, and Giulia Fanti. Improving the training of rectified flows. In Advances in Neural Information Processing Systems, volume 37, 2024. URL https://papers.neurips.cc/paper_files/paper/2024/hash/7343a5c976f8399880b695267f1f9e9f-Abstract-Conference.html.arxiv.org/abs/2405.20320
  33. [33]Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling. arXiv preprint arXiv:2210.02747, 2022. URL https://arxiv.org/abs/2210.02747.
  34. [34]Huadai Liu, Jialei Wang, Rongjie Huang, Yang Liu, Heng Lu, Zhou Zhao, and Wei Xue. Flaushadio: Rectified flow for fast and high-fidelity text-to-audio generation. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13694–13710, 2025a. URL https://aclanthology.org/2025.acl-long.666/.
  35. [35]Jie Liu, Gongye Liu, Jiajun Liang, Ziyang Yuan, Xiaokun Liu, Mingwu Zheng, Xiele Wu, Quilin Wang, Menghan Xia, Xintao Wang, et al. Improving video generation with human feedback. arXiv preprint arXiv:2501.13918, 2025b. URL https://arxiv.org/abs/2501.13918.
  36. [36]Qiang Liu. Rectified flow: A marginal preserving approach to optimal transport. arXiv preprint arXiv:2209.14577, 2022. URL https://arxiv.org/abs/2209.14577.
  37. [37]Qiang Liu, Jason Lee, and Michael Jordan. A kernelized Stein discrepancy for goodness-of-fit tests. In Proceedings of the 33rd International Conference on Machine Learning, volume 48, pages 276–284, 2016. URL https://proceedings.mlr.press/v48/liub16.html.
  38. [38]Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. In The Eleventh International Conference on Learning Representations, 2023. URL https://arxiv.org/abs/2209.03003.
  39. [39]Xingchao Liu, Xiwen Zhang, Jianzhu Ma, Jian Peng, et al. Instaflow: One step is enough for high-quality diffusion-based text-to-image generation. In The Twelfth International Conference on Learning Representations, 2024. URL https://arxiv.org/abs/2309.06380.
  40. [40]Wilhelm Magnus. On the exponential solution of differential equations for a linear operator. Communications on pure and applied mathematics, 7(4):649–673, 1954. URL https://doi.org/10.1002/cpa.3160070404.
  41. [41]Geoffrey McLachlan and David Peel. Finite Mixture Models. Wiley, 2000. URL https://doi.org/10.1002/0471721182.
  42. [42]Gonzalo Mena, Arun Kumar Kuchibhotla, and Larry Wasserman. Statistical properties of rectified flow. arXiv preprint arXiv:2511.03193, 2025. URL https://arxiv.org/abs/2511.03193.
  43. [43]Chenlin Meng, Yang Song, Wenzhe Li, and Stefano Ermon. Estimating high order gradients of the data distribution by denoising. Advances in Neural Information Processing Systems, 34:25359–25369, 2021. URL https://openreview.net/forum?id=YTkQqrqSyE1.arxiv.org/abs/2111.04726
  44. [44]T. Tin Nguyen, Hien D. Nguyen, Faicel Chamroukhi, and Geoffrey J. McLachlan. Approximation by finite mixtures of continuous density functions that vanish at infinity. Cogent Mathematics & Statistics, 7(1):1750861, 2020. URL https://doi.org/10.1080/25742558.2020.1750861.
  45. [45]Emile Pierret, Johannes Hertrich, Samuel Hurault, and Julie Delon. Tessellations of semi-discrete flow matching. arXiv preprint arXiv:2605.07513, 2026. URL https://arxiv.org/abs/2605.07513.
  46. [47]Tiberiu Popoviciu. Sur les équations algébriques ayant toutes leurs racines réelles. Mathematica, 9(129–145):20, 1935. URL https://www.ictp.acad.ro/popoviciu/articles/sur_les_equations_algebriques_ayant_toutes_leurs_racines_reelles.pdf.
  47. [48]Peter Potaptchik, Adhi Saravanan, Abbas Mammadov, Alvaro Prat, Michael Samuel Albergo, and Yee Whye Teh. Meta flow maps enable scalable reward alignment. In Tong Zhang, Miroslav Dudik, Martin Jaggi, Alekh Agarwal, Sharon Li, Dale Schuurmans, Jerry Zhu, Felix Berkenkamp, Hanze Dong, and Alberto Bietti, editors, Proceedings of the 43rd International Conference on Machine Learning, volume 306 of Proceedings of Machine Learning Research, pages 99255–99295. PMLR, 06–11 Jul 2026. URL https://proceedings.mlr.press/v306/potaptchik26b.html.arxiv.org/abs/2601.14430
  48. [49]Saptarshi Roy, Debepsita Mukherjee, and Pratik Patil. Optimal self-distillation for rectified flow via linear probing. In The First Conference on Statistics and Trustworthy AI for Cross (X)-Domain Acceleration, 2026a. URL https://arxiv.org/abs/2607.14947.
  49. [50]Saptarshi Roy, Alessandro Rinaldo, and Purnamrita Sarkar. Low-dimensional adaptation of rectified flow: A diffusion and stochastic localization perspective. arXiv preprint arXiv:2601.15500, 2026b. URL https://arxiv.org/abs/2601.15500.
  50. [51]Michael Ruzhansky and Mitsuru Sugimoto. On global inversion of homogeneous maps. Bulletin of Mathematical Sciences, 5(1):13–18, 2015. URL https://doi.org/10.1007/s13373-014-0059-1.
  51. [52]Christian G. Simader. A homotopy argument and its applications to the transformation rule for bi-Lipschitz mappings, the Brouwer fixed point theorem and the Brouwer degree, 2005. URL https://epub.uni-bayreuth.de/id/eprint/813/. preprint, University of Bayreuth.
  52. [53]Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International conference on machine learning, pages 2256–2265. PMLR, 2015. URL https://arxiv.org/abs/1503.03585.
  53. [54]Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=St1giarCHLP.arxiv.org/abs/2010.02502
  54. [55]Gerald Teschl. Ordinary differential equations and dynamical systems, volume 140. American Mathematical Soc., 2012. URL https://bookstore.ams.org/gsm-140/.DOI
  55. [56]Panos Tsimpos, Daniel Sharp, and Youssef Marzouk. One-shot generative flows: Existence and obstructions. arXiv preprint arXiv:2604.15439, 2026. URL https://arxiv.org/abs/2604.15439.
  56. [57]Fu-Yun Wang, Ling Yang, Zhaoyang Huang, Mengdi Wang, and Hongsheng Li. Rectified diffusion: Straightness is not your need in rectified flow. In International Conference on Learning Representations, 2025. URL https://proceedings.iclr.cc/paper_files/paper/2025/hash/4df9a5e6bad9e64ebcea453e031142bb-Abstract-Conference.html.arxiv.org/abs/2410.07303
  57. [58]Leda Wang, Zehao Xu, Qiang Liu, and Harrison H Zhou. Computational and statistical guarantees of the\textit {c}-rectified flow. arXiv preprint arXiv:2608.02487, 2026. URL https://arxiv.org/abs/2608.02487.
  58. [59]Yongqi Wang, Wenxiang Guo, Rongjie Huang, Jiawei Huang, Zehan Wang, Fuming You, Ruiqi Li, and Zhou Zhao. Frieren: Efficient video-to-audio generation network with rectified flow matching. Advances in Neural Information Processing Systems, 37:128118–128138, 2024. URL https://papers.nips.cc/paper_files/paper/2024/hash/743db1fd01fe1c4795f4b1aa11fc9c2d-Abstract-Conference.html.arxiv.org/abs/2406.00320
  59. [60]Hanshu Yan, Xingchao Liu, Jiachun Pan, Jun Hao Liew, Qiang Liu, and Jiashi Feng. Perflow: Piecewise rectified flow as universal plug-and-play accelerator. In Advances in Neural Information Processing Systems, volume 37, 2024. URL https://papers.nips.cc/paper_files/paper/2024/hash/8f9d1362a386e80a723e98451a8c7564-Abstract-Conference.html.arxiv.org/abs/2405.07510
  60. [61]Huminhao Zhu, Fangyikang Wang, Tianyu Ding, Qing Qu, and Zhihui Zhu. Analyzing and mitigating model collapse in reflow methods. In Conference on Parsimony and Learning, volume 328 of Proceedings of Machine Learning Research, pages 314–340. PMLR, 2026. URL https://proceedings.mlr.press/v328/zhu26a.html.

Paper details

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

views · 21

downloads · 4

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

  • Mathematics

GPT 6-Astra

View statement
Looking for related papers…
Contents