Optimization and Control (math.OC)

Abstract: Algorithmic differentiation (AD) has become increasingly capable and straightforward to use. However, AD is inefficient when applied directly to solvers, a feature of most engineering analyses. We can leverage implicit differentiation to define a general AD rule, making adjoints automatic. Furthermore, we can leverage the structure of differential equations to automate unsteady adjoints in a memory efficient way. We also derive a technique to speed up explicit differential equation solvers, which have no iterative solver to exploit. All of these techniques are demonstrated on problems of various sizes, showing order of magnitude speed-ups with minimal code changes. Thus, we can enable users to easily compute accurate derivatives across complex analyses with internal solvers, or in other words, automate adjoints using a combination of AD and implicit differentiation.

Abstract: If a sparse semidefinite program (SDP), specified over $n\times n$ matrices and subject to $m$ linear constraints, has an aggregate sparsity graph $G$ with small treewidth, then chordal conversion will frequently allow an interior-point method to solve the SDP in just $O(m+n)$ time per-iteration. This is a significant reduction over the minimum $\Omega(n^{3})$ time per-iteration for a direct solution, but a definitive theoretical explanation was previously unknown. Contrary to popular belief, the speedup is not guaranteed by a small treewidth in $G$, as a diagonal SDP would have treewidth zero but can still necessitate up to $\Omega(n^{3})$ time per-iteration. Instead, we construct an extended aggregate sparsity graph $\overline{G}\supseteq G$ by forcing each constraint matrix $A_{i}$ to be its own clique in $G$. We prove that a small treewidth in $\overline{G}$ does indeed guarantee that chordal conversion will solve the SDP in $O(m+n)$ time per-iteration, to $\epsilon$-accuracy in at most $O(\sqrt{m+n}\log(1/\epsilon))$ iterations. For classical SDPs like the MAX-$k$-CUT relaxation and the Lovasz Theta problem, the two sparsity graphs coincide $G=\overline{G}$, so our result provide a complete characterization for the complexity of chordal conversion, showing that a small treewidth is both necessary and sufficient for $O(m+n)$ time per-iteration. Real-world SDPs like the AC optimal power flow relaxation have different graphs $G\subseteq\overline{G}$ with similar small treewidths; while chordal conversion is already widely used on a heuristic basis, in this paper we provide the first rigorous guarantee that it solves such SDPs in $O(m+n)$ time per-iteration. [Supporting code at https://github.com/ryz-codes/chordalConv/]

Abstract: The focus of this article is on shape and topology optimization of transient vibroacoustic problems. The main contribution is a transient problem formulation that enables optimization over wide ranges of frequencies with complex signals, which are often of interest in industry. The work employs time domain methods to realize wide band optimization in the frequency domain. To this end, the objective function is defined in frequency domain where the frequency response of the system is obtained through a fast Fourier transform (FFT) algorithm on the transient response of the system. The work utilizes a parametric level set approach to implicitly define the geometry in which the zero level describes the interface between acoustic and structural domains. A cut element method is used to capture the geometry on a fixed background mesh through utilization of a special integration scheme that accurately resolves the interface. This allows for accurate solutions to strongly coupled vibroacoustic systems without having to re-mesh at each design update. The present work relies on efficient gradient based optimizers where the discrete adjoint method is used to calculate the sensitivities of objective and constraint functions. A thorough explanation of the consistent sensitivity calculation is given involving the FFT operation needed to define the objective function in frequency domain. Finally, the developed framework is applied to various vibroacoustic filter designs and the optimization results are verified using commercial finite element software with a steady state time-harmonic formulation.

Abstract: In this paper we study set convergence aspects for Banach spaces of vector-valued measures with divergences (represented by measures or by functions) and applications. We consider a form of normal trace characterization to establish subspaces of measures that directionally vanish in parts of the boundary, and present examples constructed with binary trees. Subsequently we study convex sets with total variation bounds and their convergence properties together with applications to the stability of optimization problems.

Abstract: In this work, we propose a robust optimization approach to mitigate the impact of uncertainties in particle precipitation. Our model incorporates partial differential equations, more particular nonlinear and nonlocal population balance equations to describe particle synthesis. The goal of the optimization problem is to design products with desired size distributions. Recognizing the impact of uncertainties, we extend the model to hedge against them. We emphasize the importance of robust protection to ensure the production of high-quality particles. To solve the resulting robust problem, we enhance a novel adaptive bundle framework for nonlinear robust optimization that integrates the exact method of moments approach for solving the population balance equations. Computational experiments performed with the integrated algorithm focus on uncertainties in the total mass of the system as it greatly influence the quality of the resulting product. Using realistic parameter values for quantum dot synthesis, we demonstrate the efficiency of our integrated algorithm. Furthermore, we find that the unprotected process fails to achieve the desired particle characteristics, even for small uncertainties, which highlights the necessity of the robust process. The latter consistently outperforms the unprotected process in quality of the obtained product, in particular in perturbed scenarios.

Abstract: Non-asymptotic convergence analysis of quasi-Newton methods has gained attention with a landmark result establishing an explicit superlinear rate of O$((1/\sqrt{t})^t)$. The methods that obtain this rate, however, exhibit a well-known drawback: they require the storage of the previous Hessian approximation matrix or instead storing all past curvature information to form the current Hessian inverse approximation. Limited-memory variants of quasi-Newton methods such as the celebrated L-BFGS alleviate this issue by leveraging a limited window of past curvature information to construct the Hessian inverse approximation. As a result, their per iteration complexity and storage requirement is O$(\tau d)$ where $\tau \le d$ is the size of the window and $d$ is the problem dimension reducing the O$(d^2)$ computational cost and memory requirement of standard quasi-Newton methods. However, to the best of our knowledge, there is no result showing a non-asymptotic superlinear convergence rate for any limited-memory quasi-Newton method. In this work, we close this gap by presenting a limited-memory greedy BFGS (LG-BFGS) method that achieves an explicit non-asymptotic superlinear rate. We incorporate displacement aggregation, i.e., decorrelating projection, in post-processing gradient variations, together with a basis vector selection scheme on variable variations, which greedily maximizes a progress measure of the Hessian estimate to the true Hessian. Their combination allows past curvature information to remain in a sparse subspace while yielding a valid representation of the full history. Interestingly, our established non-asymptotic superlinear convergence rate demonstrates a trade-off between the convergence speed and memory requirement, which to our knowledge, is the first of its kind. Numerical results corroborate our theoretical findings and demonstrate the effectiveness of our method.

Abstract: The increasing challenges to the grid stability posed by the penetration of renewable energy resources urge a more active role for demand response programs as viable alternatives to a further expansion of peak power generators. This work presents a methodology to exploit the demand flexibility of energy-intensive industries under Demand-Side Management programs in the energy and reserve markets. To this end, we propose a novel scheduling model for a multi-stage multi-line process, which incorporates both the critical manufacturing constraints and the technical requirements imposed by the market. Using mixed integer programming approach, two optimization problems are formulated to sequentially minimize the cost in a day-ahead energy market and maximize the reserve provision when participating in the ancillary market. The effectiveness of day-ahead scheduling model has been verified for the case of a real metal casting plant in the Nordic market, where a significant reduction of energy cost is obtained. Furthermore, the reserve provision is shown to be a potential tool for capitalizing on the reserve market as a secondary revenue stream.