Randomized Rounding Research Articles

Investigating mixed-precision for AGATA pulse-shape analysis

Investigating mixed-precision for AGATA pulse-shape analysis

PyTracer: Automatically Profiling Numerical Instabilities in Python

Numerical stability is a crucial requirement of reliable scientific computing. However, despite the pervasiveness of Python in data science, analyzing large Python programs remains challenging due to the lack of scalable numerical analysis tools available for this language. To fill this gap, we developed PyTracer, a profiler to quantify numerical instability in Python applications. PyTracer transparently instruments Python code to produce numerical traces and visualize them interactively in a Plotly dashboard. We designed PyTracer to be agnostic to numerical noise model, allowing for numerical profiling through Monte-Carlo Arithmetic, random rounding, random data perturbation, or structured noise for a particular application. We illustrate PyTracer's capabilities by testing the numerical stability of key functions in both SciPy and Scikit-learn, two dominant Python libraries for mathematical modeling. Through these evaluations, we demonstrate PyTracer as a scalable, automated, and generic framework for numerical profiling in Python.

Continuous Influence-Based Community Partition for Social Networks

Community partition is of great importance in social networks because of the rapid increasing network scale, data and applications. We consider the community partition problem under LT model in social networks, which is a combinatorial optimization problem that divides the social network to disjoint $m$ communities. Our goal is to maximize the sum of influence propagation through maximizing it within each community. As the influence propagation function of community partition problem is supermodular under LT model, we use the method of Lov{$\acute{a}$}sz Extension to relax the target influence function and transfer our goal to maximize the relaxed function over a matroid polytope. Next, we propose a continuous greedy algorithm using the properties of the relaxed function to solve our problem, which needs to be discretized in concrete implementation. Then, random rounding technique is used to convert the fractional solution to integer solution. We present a theoretical analysis with $1-1/e$ approximation ratio for the proposed algorithms. Extensive experiments are conducted to evaluate the performance of the proposed continuous greedy algorithms on real-world online social networks datasets and the results demonstrate that continuous community partition method can improve influence spread and accuracy of the community partition effectively.

Mixed-Projection Conic Optimization: A New Paradigm for Modeling Rank Constraints

Many central problems throughout optimization, machine learning, and statistics are equivalent to optimizing a low-rank matrix over a convex set. However, although rank constraints offer unparalleled modeling flexibility, no generic code currently solves these problems to certifiable optimality at even moderate sizes. Instead, low-rank optimization problems are solved via convex relaxations or heuristics that do not enjoy optimality guarantees. In “Mixed-Projection Conic Optimization: A New Paradigm for Modeling Rank Constraints,” Bertsimas, Cory-Wright, and Pauphilet propose a new approach for modeling and optimizing over rank constraints. They generalize mixed-integer optimization by replacing binary variables z that satisfy z2 =z with orthogonal projection matrices Y that satisfy Y2 = Y. This approach offers the following contributions: First, it supplies certificates of (near) optimality for low-rank problems. Second, it demonstrates that some of the best ideas in mixed-integer optimization, such as decomposition methods, cutting planes, relaxations, and random rounding schemes, admit straightforward extensions to mixed-projection optimization.

The smallest singular value of heavy-tailed not necessarily i.i.d. random matrices via random rounding

We show the existence of a net near the sphere, such that the values of any matrix on the sphere and on the net are compared via a regularized Hilbert-Schmidt norm, which we introduce. This allows to construct an efficient net which controls the length of Ax for any random matrix A with independent columns (no other assumptions are required). As a consequence we show that the smallest singular value σn (A) of an N × n random matrix A with i.i.d. mean zero, variance one entries enjoys the following small ball estimate, for any ϵ > 0 $$P({\sigma _n}(A) < \epsilon (\sqrt {N + 1} - \sqrt n )) \le {(C\epsilon \,\log \,1/\epsilon )^{N - n + 1}} + {e^{ - cN}}.$$ The proof of this result requires working with matrices whose rows are not independent, and, therefore, the fact that the theorem about discretization works for matrices with dependent rows, is crucial. Furthermore, in the case of the square n×n matrix A with independent entries having concentration function separated from 1, i.i.d. rows, and such that $$\mathbb{E}\left\| A \right\|_{HS}^2 \le c{n^2}$$ , one has $$P({\sigma _n}(A) < {\epsilon \over {\sqrt n }} \le C\epsilon + {e^{ - cn}},$$ for any ϵ > 0. In addition, for $$\epsilon > {c \over {\sqrt n }}$$ the assumption of i.i.d. rows is not required. Our estimates generalize the previous results of Rudelson and Vershynin [29], [30], which required the sub-gaussian mean zero variance one assumptions, as well as the work of Rebrova and Tikhomirov [25], where mean zero variance 1 and i.i.d. entries were required.

Assortment Optimization under a Single Transition Choice Model

In this study, we consider a new customer choice model which we call the single transition choice model. In this model, there is a universe of products and customers arrive at each product with a certain probability. If the arrived product is unavailable, then the seller can recommend a subset of available products and the customer will purchase one of the recommended products or choose not to purchase with certain transition probabilities. The distinguishing features of the model are that the seller can control which products to recommend depending on the arrived product, and each customer either purchases a product or leaves the market after one transition. We study the assortment optimization problem under this model. Particularly, we show that it is NP‐Hard even if the customer can transition from each product to at most two products. Despite the computational complexity, we provide polynomial time algorithms or approximation algorithms for several special cases, such as when the customer can only transition from each product to at most a given number of products and the size of each recommended set is bounded. Our approximation algorithms are developed by invoking the submodularity arguments, or connecting the problem with maximum constraint satisfaction problem and applying randomized rounding techniques to its semidefinite programming relaxation. We also provide a tight worst‐case performance bound for revenue‐ordered assortments. In addition, we propose a compact mixed‐integer program formulation, which is efficient for moderate size problems. Finally, we conduct numerical experiments to demonstrate the effectiveness of the proposed algorithms.

Ranking an Assortment of Products Via Sequential Submodular Optimization

We study a submodular maximization problem motivated by applications in online retail. A platform displays a list of products to a user in response to a search query. The user inspects the first k items in the list for a k chosen at random from a given distribution, and decides whether to purchase an item from that set based on a choice model. The goal of the platform is to maximize the engagement of the shopper defined as the probability of purchase. This problem gives rise to a less-studied variation of submodular maximization in which we are asked to choose an ordering of a set of elements to maximize a linear combination of different submodular functions. First, using a reduction to maximizing submodular functions over matroids, we give an optimal (1-1/e)-approximation for this problem. We then consider a variant in which the platform cares not only about user engagement, but also about diversification across various groups of users, that is, guaranteeing a certain probability of purchase in each group. We characterize the polytope of feasible solutions and give a bi-criteria ((1-1/e)^2,(1-1/e)^2)-approximation for this problem by rounding an approximate solution of a linear programming relaxation. For rounding, we relay on our reduction and the particular rounding techniques for matroid polytopes. For the special case in which underlying submodular functions are coverage functions -- which is practically relevant in online retail -- we propose an alternative LP relaxation and a simpler randomized rounding for the problem. This approach yields to an optimal bi-criteria (1-1/e,1-1/e)-approximation algorithm for the special case of the problem with coverage functions.

Concurrent Multipath Routing Optimization in Named Data Networks

To achieve maximum throughput in named data networking (NDN) with consideration of fairness among each source-sink pair, we model the routing problem in NDN as a mixed-integer maximum concurrent flow problem. Taking advantage of the smart forwarding framework in NDN, we obtain an approximate optimal routing and forwarding solution with our improved Fleischer and random rounding (FRR) algorithm. We analyze the complexity of the Fleischer algorithm more comprehensively and give the best scaling method to achieve the least iterations. In addition, we further reduce the complexity of the Fleischer algorithm so that it iterates only once, thus getting a heuristic algorithm which is called multicommodity multipath (MCMP). Using ndnSIM-based simulation, we compare our proposed routing algorithms with the existing routing algorithms, including flooding, k-shortest path, and best single path routing, and the simulation results prove that FRR and MCMP can outperform other algorithms both in throughput and latency, and it can also be concluded that the best number of paths for k-shortest path routing should be close to and less than the average node degree of the network. For studying routing optimization with caching, we conduct simulations with different in-network cache capacity and different caching policies. The simulation results show that although the LRU can achieve the best performance under the same cache capacity, the cache capacity plays a decisive role. All routing algorithms produce better performance with caching, but the FRR still achieves the best performance, and follows by MCMP.

Improved approximation algorithms for minimum power covering problems

Given an undirected graph with edge costs, the power of a node is the maximum cost of an edge incident to it, and the power of a graph is the sum of the powers of its nodes. Motivated by applications in wireless networks, we consider two network design problems under the power minimization criteria. In both problems we are given a graph G=(V,E) with edge costs and a set T⊆V of terminals. The goal is to find a minimum power edge subset F⊆E such that the graph H=(V,F) satisfies some prescribed requirements. In the Min-Power Edge-Cover problem, H should contain an edge incident to every terminal. Using the Iterative Randomized Rounding (IRR) method, we give an algorithm with expected approximation ratio 1.41; the ratio is reduced to 73/60<1.217 when T is an independent set in G. In the case of unit costs we also achieve ratio 73/60, and in addition give a simple efficient combinatorial algorithm with ratio 5/4. For all these NP-hard problems the previous best known ratio was 3/2. In the related Min-Power Terminal Backup problem, H should contain a path from every t∈T to some node in T∖{t}. We obtain ratio 3/2 for this NP-hard problem, improving the trivial ratio of 2.

Numerical validation of compensated algorithms with stochastic arithmetic

Compensated algorithms consist in computing the rounding errors of individual operations and then adding them later on to the computed result. This makes it possible to increase the accuracy of the computed result efficiently. Computing the rounding error of an individual operation is possible through the use of a so-called error-free transformation. In this article, we show that it is possible to validate the result of compensated algorithms using stochastic arithmetic. We study compensated algorithms for summation, dot product and polynomial evaluation. We prove that the use of the random rounding mode inherent to stochastic arithmetic does not change much the accuracy of compensated methods. This is due to the fact that error-free transformations are no more exact but still sufficiently accurate to improve the numerical quality of results.

A Unified Approach to Volatility Estimation in the Presence of Both Rounding and Random Market Microstructure Noise

Despite the coexistence of random noise and rounding in practice, existing widely used volatility estimation methods focus only on one of the two sources of error. In this paper, we propose a unified approach to volatility estimation in the presence of both rounding and random market microstructure noise. The proposed method enjoys superior properties in terms of bias correction and convergence rate. Simulation and empirical studies based on both stock and currency exchange data show clear advantages of the proposed method.

Energy-efficient scheduling and routing via randomized rounding

We propose a unifying framework based on configuration linear programs and ran-domized rounding, for different energy optimization problems in the dynamic speed-scaling setting. We apply our framework to various scheduling and routing problems in heterogeneous computing and networking environments. We first consider the energy minimization problem of scheduling a set of jobs on a set of parallel speed scalable processors in a fully heterogeneous setting. For both the preemptive-non-migratory and the preemptive-migratory variants, our approach allows us to obtain solutions of almost the same quality as for the homogeneous environment. By exploiting the result for the preemptive-non-migratory variant, we are able to improve the best known approximation ratio for the single processor non-preemptive problem. Furthermore, we show that our approach allows to obtain a constant-factor approximation algorithm for the power-aware preemptive job shop scheduling problem. Finally, we consider the min-power routing problem where we are given a network modeled by an undirected graph and a set of uniform demands that have to be routed on integral routes from their sources to their destinations so that the energy consumption is minimized. We improve the best known approximation ratio for this problem.

Numerical Validation of Compensated Summation Algorithms with Stochastic Arithmetic

Compensated summation algorithms are designed to improve the accuracy of ill-conditioned sums. They are based on algorithms, such as FastTwoSum, which are proved to provide, with rounding to nearest, the sum of two floating-point numbers and the associated rounding error. Discrete stochastic arithmetic enables one to estimate rounding error propagation in numerical codes. It requires a random rounding mode which consists in rounding each computed result toward −∞ or +∞ with the same probability. In this paper we analyse the impact of this random rounding mode on compensated summations based on the FastTwoSum algorithm. We show the accuracy improvement obtained using such compensated summations in numerical simulations controlled with discrete stochastic arithmetic.

The Square Root Multipliers Algorithm for Discrete Capacity and Buffer Assignment problems in Elastic Traffic Networks

This paper discusses a novel approach for the dimensioning of packet networks, under the constraints of end-to-end Quality-of-Service (QoS) requirements. The network modeling also considers the dynamic behavior of today's networks traffic. In particular, the proposed approach considers multilink networks where discrete capacities are the decision variables. This Discrete Capacity Assignment (DCA) problem can be classified as a constrained combinatorial optimization problem and therefore it is NP-complete. In order to solve the problem, we propose the Square Root Multipliers Algorithm (SRMA) followed by a Randomized Rounding (RR) method. We also enforce loss probability constraints by properly choosing buffer sizes, therefore facing the Buffer Assignment (BA) problem. To evaluate the performance of our approach, SRMA/RR results are compared with the optimal solutions found by an exhaustive search (ES) procedure. Analytical results for a variety of problem instances are reported, and a verification is performed based on simulations conducted with the software NS-2. The results suggest that the proposed approach provides a quite efficient method to obtain near-optimal solutions with small computational effort.

Upper and lower bounds for dynamic cluster assignment for multi-target tracking in heterogeneous WSNs

In this paper, we consider the problem of cluster task assignment to maximize total utilities of nodes for target coverage in heterogeneous Wireless Sensor Networks. We define this problem as assigning the tasks of Cluster Head (CH) and Cluster Members (CM) to nodes for each target and requiring communication connectivity between every CH with its members. The utility of each node for each target is defined as a function of its distance to the target and its remaining energy. We propose an upper bound based on Lagrangian Relaxation (LR) and a lower bound by Linear Programming (LP) relaxation with a combination of Randomized Rounding (RR) and a greedy-based heuristic. Furthermore, we propose a distributed heuristic algorithm based on matching and a general assignment problem. Dynamic movements of targets are taken into account by intra/inter-cluster task reassignments. Simulation results, compared with optimal values, reveal that the LR upper bound performs better than the bound reached by pure LP relaxation. The lower bound obtained by LP relaxation combined with the RR technique provides close results in comparison with the distributed heuristic algorithm. Furthermore, the results of the distributed heuristic algorithm remain between the upper and lower bounds and close to optimal values.

Statistical disclosure control using random rounding and quadratic programming

The most common method of providing data to the public is through statistical tables. The problem of protecting confidentiality in statistical tables containing sensitive information has been of great concern during the recent years. Rounding methods are perturbation techniques widely used by statistical agencies for protecting the confidential data. Random rounding is one of these methods. In this paper, using the technique of random rounding and quadratic programming, we introduce a new methodology for protecting the confidential information of tabular data with minimum loss of information. The tables obtained through the proposed method consist of unbiasedly rounded values, are additive and have specified level of confidentiality protection. Some numerical examples are also discussed to demonstrate the superiority of the proposed procedure over the existing procedures.

Decoding by Sampling: A Randomized Lattice Algorithm for Bounded Distance Decoding

Despite its reduced complexity, lattice reduction-aided decoding exhibits a widening gap to maximum-likelihood (ML) performance as the dimension increases. To improve its performance, this paper presents randomized lattice decoding based on Klein's sampling technique, which is a randomized version of Babai's nearest plane algorithm (i.e., successive interference cancelation (SIC)). To find the closest lattice point, Klein's algorithm is used to sample some lattice points and the closest among those samples is chosen. Lattice reduction increases the probability of finding the closest lattice point, and only needs to be run once during pre-processing. Further, the sampling can operate very efficiently in parallel. The technical contribution of this paper is two-fold: we analyze and optimize the decoding radius of sampling decoding resulting in better error performance than Klein's original algorithm, and propose a very efficient implementation of random rounding. Of particular interest is that a fixed gain in the decoding radius compared to Babai's decoding can be achieved at polynomial complexity. The proposed decoder is useful for moderate dimensions where sphere decoding becomes computationally intensive, while lattice reduction-aided decoding starts to suffer considerable loss. Simulation results demonstrate near-ML performance is achieved by a moderate number of samples, even if the dimension is as high as 32.

Metric Clustering via Consistent Labeling

We design approximation algorithms for a number of fundamental optimization problems in metric spaces, namely computing separating and padded decompositions, sparse covers, and metric triangulations. Our work is the first to emphasize relative guarantees that compare the produced solution to the optimal one for the input at hand. By contrast, the extensive previous work on these topics has sought absolute bounds that hold for every possible metric space (or for a family of metrics). While absolute bounds typically translate to relative ones, our algorithms provide significantly better relative guarantees, using a rather different algorithm. Our technical approach is to cast a number of metric clustering problems that have been well studied---but almost always as disparate problems---into a common modeling and algorithmic framework, which we call the consistent labeling problem. Having identified the common features of all of these problems, we provide a family of linear programming relaxations and simple randomized rounding procedures that achieve provably good approximation guarantees.

A new version of the CADNA library for estimating round-off error propagation in Fortran programs

The CADNA library enables one to estimate, using a probabilistic approach, round-off error propagation in any simulation program. CADNA provides new numerical types, the so-called stochastic types, on which round-off errors can be estimated. Furthermore CADNA contains the definition of arithmetic and relational operators which are overloaded for stochastic variables and the definition of mathematical functions which can be used with stochastic arguments. On 64-bit processors, depending on the rounding mode chosen, the mathematical library associated with the GNU Fortran compiler may provide incorrect results or generate severe bugs. Therefore the CADNA library has been improved to enable the numerical validation of programs on 64-bit processors. New version program summary Program title: CADNA Catalogue identifier: AEAT_v1_1 Program summary URL: http://cpc.cs.qub.ac.uk/summaries/AEAT_v1_1.html Program obtainable from: CPC Program Library, Queen's University, Belfast, N. Ireland Licensing provisions: Standard CPC licence, http://cpc.cs.qub.ac.uk/licence/licence.html No. of lines in distributed program, including test data, etc.: 28 488 No. of bytes in distributed program, including test data, etc.: 463 778 Distribution format: tar.gz Programming language: Fortran NOTE: A C++ version of this program is available in the Library as AEGQ_v1_0 Computer: PC running LINUX with an i686 or an ia64 processor, UNIX workstations including SUN, IBM Operating system: LINUX, UNIX Classification: 6.5 Catalogue identifier of previous version: AEAT_v1_0 Journal reference of previous version: Comput. Phys. Commun. 178 (2008) 933 Does the new version supersede the previous version?: Yes Nature of problem: A simulation program which uses floating-point arithmetic generates round-off errors, due to the rounding performed at each assignment and at each arithmetic operation. Round-off error propagation may invalidate the result of a program. The CADNA library enables one to estimate round-off error propagation in any simulation program and to detect all numerical instabilities that may occur at run time. Solution method: The CADNA library [1–3] implements Discrete Stochastic Arithmetic [4,5] which is based on a probabilistic model of round-off errors. The program is run several times with a random rounding mode generating different results each time. From this set of results, CADNA estimates the number of exact significant digits in the result that would have been computed with standard floating-point arithmetic. Reasons for new version: On 64-bit processors, the mathematical library associated with the GNU Fortran compiler may provide incorrect results or generate severe bugs with rounding towards −∞ and +∞, which the random rounding mode is based on. Therefore a particular definition of mathematical functions for stochastic arguments has been included in the CADNA library to enable its use with the GNU Fortran compiler on 64-bit processors. Summary of revisions: If CADNA is used on a 64-bit processor with the GNU Fortran compiler, mathematical functions are computed with rounding to the nearest, otherwise they are computed with the random rounding mode. It must be pointed out that the knowledge of the accuracy of the stochastic argument of a mathematical function is never lost. Restrictions: CADNA requires a Fortran 90 (or newer) compiler. In the program to be linked with the CADNA library, round-off errors on complex variables cannot be estimated. Furthermore array functions such as product or sum must not be used. Only the arithmetic operators and the abs, min, max and sqrt functions can be used for arrays. Additional comments: In the library archive, users are advised to read the INSTALL file first. The doc directory contains a user guide named ug.cadna.pdf which shows how to control the numerical accuracy of a program using CADNA, provides installation instructions and describes test runs. The source code, which is located in the src directory, consists of one assembly language file (cadna_rounding.s) and eighteen Fortran language files. cadna_rounding.s is a symbolic link to the assembly file corresponding to the processor and the Fortran compiler used. This assembly file contains routines which are frequently called in the CADNA Fortran files to change the rounding mode. The Fortran language files contain the definition of the stochastic types on which the control of accuracy can be performed, CADNA specific functions (for instance to enable or disable the detection of numerical instabilities), the definition of arithmetic and relational operators which are overloaded for stochastic variables and the definition of mathematical functions which can be used with stochastic arguments. The examples directory contains seven test runs which illustrate the use of the CADNA library and the benefits of Discrete Stochastic Arithmetic. Running time: The version of a code which uses CADNA runs at least three times slower than its floating-point version. This cost depends on the computer architecture and can be higher if the detection of numerical instabilities is enabled. In this case, the cost may be related to the number of instabilities detected.

CADNA_C: A version of CADNA for use with C or C++ programs

The CADNA library enables one to estimate round-off error propagation using a probabilistic approach. The CADNA_C version enables this estimation in C or C++ programs, while the previous version had been developed for Fortran programs. The CADNA_C version has the same features as the previous one: with CADNA the numerical quality of any simulation program can be controlled. Furthermore by detecting all the instabilities which may occur at run time, a numerical debugging of the user code can be performed. CADNA provides new numerical types on which round-off errors can be estimated. Slight modifications are required to control a code with CADNA, mainly changes in variable declarations, input and output. New version program summary Program title: CADNA_C Catalogue identifier: AEGQ_v1_0 Program summary URL: http://cpc.cs.qub.ac.uk/summaries/AEGQ_v1_0.html Program obtainable from: CPC Program Library, Queen's University, Belfast, N. Ireland Licensing provisions: Standard CPC licence, http://cpc.cs.qub.ac.uk/licence/licence.html No. of lines in distributed program, including test data, etc.: 60 075 No. of bytes in distributed program, including test data, etc.: 710 781 Distribution format: tar.gz Programming language: C++ Computer: PC running LINUX with an i686 or an ia64 processor, UNIX workstations including SUN, IBM Operating system: LINUX, UNIX Classification: 6.5 Catalogue identifier of previous version: AEAT_v1_0 Journal reference of previous version: Comput. Phys. Comm. 178 (2008) 933 Does the new version supersede the previous version?: No Nature of problem: A simulation program which uses floating-point arithmetic generates round-off errors, due to the rounding performed at each assignment and at each arithmetic operation. Round-off error propagation may invalidate the result of a program. The CADNA library enables one to estimate round-off error propagation in any simulation program and to detect all numerical instabilities that may occur at run time. Solution method: The CADNA library [1–3] implements Discrete Stochastic Arithmetic [4,5] which is based on a probabilistic model of round-off errors. The program is run several times with a random rounding mode generating different results each time. From this set of results, CADNA estimates the number of exact significant digits in the result that would have been computed with standard floating-point arithmetic. Reasons for new version: The previous version (AEAT_v1_0) enables the estimation of round-off error propagation in Fortran programs [2]. The new version has been developed to enable this estimation in C or C++ programs. Summary of revisions: The CADNA_C source code consists of one assembly language file (cadna_rounding.s) and twenty-three C++ language files (including three header files). cadna_rounding.s is a symbolic link to the assembly file corresponding to the processor and the C++ compiler used. This assembly file contains routines which are frequently called in the CADNA_C C++ files to change the rounding mode. The C++ language files contain the definition of the stochastic types on which the control of accuracy can be performed, CADNA_C specific functions (for instance to enable or disable the detection of numerical instabilities), the definition of arithmetic and relational operators which are overloaded for stochastic variables and the definition of mathematical functions which can be used with stochastic arguments. As a remark, on 64-bit processors, the mathematical library associated with the GNU C++ compiler may provide incorrect results or generate severe bugs with rounding towards −∞ and +∞, which the random rounding mode is based on. Therefore, if CADNA_C is used on a 64-bit processor with the GNU C++ compiler, mathematical functions are computed with rounding to the nearest, otherwise they are computed with the random rounding mode. It must be pointed out that the knowledge of the accuracy of the argument of a mathematical function is never lost. Additional comments: In the library archive, users are advised to read the INSTALL file first. The doc directory contains a user guide named ug.cadna.pdf and a reference guide named, ref_cadna.pdf. The user guide shows how to control the numerical accuracy of a program using CADNA, provides installation instructions and describes test runs.The reference guide briefly describes each function of the library. The source code (which consists of C++ and assembly files) is located in the src directory. The examples directory contains seven test runs which illustrate the use of the CADNA library and the benefits of Discrete Stochastic Arithmetic. Running time: The version of a code which uses CADNA runs at least three times slower than its floating-point version. This cost depends on the computer architecture and can be higher if the detection of numerical instabilities is enabled. In this case, the cost may be related to the number of instabilities detected.