We develop R2N, a modified quasi-Newton method for minimizing the sum of a $\mathcal{C}^1$ function $f$ and a lower semi-continuous prox-bounded $h$. Both $f$ and $h$ may be nonconvex. At each iteration, our method computes a step by minimizing the sum of a quadratic model of $f$, a model of $h$, and an adaptive quadratic regularization term. A step may be computed by a variant of the proximal-gradient method. An advantage of R2N over trust-region (TR) methods is that proximal operators do not involve an extra TR indicator. We also develop the variant R2DH, in which the model Hessian is diagonal, which allows us to compute a step without relying on a subproblem solver when $h$ is separable. R2DH can be used as standalone solver, but also as subproblem solver inside R2N. We describe non-monotone variants of both R2N and R2DH. Global convergence of a first-order stationarity measure to zero holds without relying on local Lipschitz continuity of $\nabla f$, while allowing model Hessians to grow unbounded, an assumption particularly relevant to quasi-Newton models. Under Lipschitz-continuity of $\nabla f$, we establish a tight worst-case complexity bound of $O(1 / \epsilon^{2/(1 - p)})$ to bring said measure below $\epsilon > 0$, where $0 \leq p < 1$ controls the growth of model Hessians. The latter must not diverge faster than $|\mathcal{S}_k|^p$, where $\mathcal{S}_k$ is the set of successful iterations up to iteration $k$. When $p = 1$, we establish the tight exponential complexity bound $O(\exp(c \epsilon^{-2}))$ where $c > 0$ is a constant. We describe our Julia implementation and report numerical experience on a basis-pursuit problem, image denoising, minimum-rank matrix completion, and a nonlinear support vector machine. In particular, the minimum-rank problem cannot be solved directly at this time by a TR approach as corresponding proximal operators are not known analytically.

This is the first such exact algorithm that is guaranteed to B. Background have a runtime that is linear in the number of stages, as well as linear in the number of both state-only constraints as well Discrete-time optimal control problems are ubiquitous in as mixed state-and-control constraints, without imposing any the fields of robotics, motion planning, and control theory, restrictions on the problem instances. We also show how to often being solved at high frequencies as part of real-time easily parallelize this algorithm to run in parallel runtime systems. A key requirement for a solution mechanism to be practical is that it should depend only linearly on the number of II. In the case of unconstrained problems, this is typically A. Constrained LQR Without loss of generality, we assume LQR problems. Clearly, constrained LQR problems are of paramount Note that these regularity conditions ensure that, even in practical importance.

We introduce a new algorithm for solving unconstrained discrete-time optimal control problems. Our method follows a direct multiple shooting approach, and consists of applying the SQP method together with an $\ell_2$ augmented Lagrangian primal-dual merit function. We use the LQR algorithm to efficiently solve the primal-dual Newton-KKT system. As our algorithm is a specialization of NPSQP, it inherits its generic properties, including global convergence, fast local convergence, and the lack of need for second order corrections or dimension expansions, improving on existing direct multiple shooting approaches such as acados, ALTRO, GNMS, FATROP, and FDDP. The solutions of the LQR-shaped subproblems posed by our algorithm can be be parallelized to run in time logarithmic in the number of stages, states, and controls. Moreover, as our method avoids sequential rollouts of the nonlinear dynamics, it can run in $O(1)$ parallel time per line search iteration. Therefore, this paper provides a practical, theoretically sound, and highly parallelizable (for example, with a GPU) method for solving nonlinear discrete-time optimal control problems. An open-source JAX implementation of this algorithm can be found on GitHub (joaospinto/primal_dual_ilqr).

We consider the problem of training a deep neural network with nonsmooth regularization to retrieve a sparse and efficient sub-structure. Our regularizer is only assumed to be lower semi-continuous and prox-bounded. We combine an adaptive quadratic regularization approach with proximal stochastic gradient principles to derive a new solver, called SR2, whose convergence and worst-case complexity are established without knowledge or approximation of the gradient's Lipschitz constant. We formulate a stopping criteria that ensures an appropriate first-order stationarity measure converges to zero under certain conditions. We establish a worst-case iteration complexity of $\mathcal{O}(\epsilon^{-2})$ that matches those of related methods like ProxGEN, where the learning rate is assumed to be related to the Lipschitz constant. Our experiments on network instances trained on CIFAR-10 and CIFAR-100 with $\ell_1$ and $\ell_0$ regularizations show that SR2 consistently achieves higher sparsity and accuracy than related methods such as ProxGEN and ProxSGD.