Active Subspaces: Emerging Ideas for Dimension Reduction in by Paul G. Constantine

By Paul G. Constantine

Scientists and engineers use computing device simulations to check relationships among a model's enter parameters and its outputs. even if, thorough parameter reviews are tough, if now not very unlikely, whilst the simulation is dear and the version has numerous inputs. To permit stories in those cases, the engineer may well try to lessen the size of the model's enter parameter area. lively subspaces are an rising set of measurement aid instruments that establish vital instructions within the parameter house. This e-book describes concepts for locating a model's energetic subspace and proposes equipment for exploiting the diminished measurement to permit another way infeasible parameter stories. Readers will locate new principles for measurement aid, easy-to-implement algorithms, and several other examples of lively subspaces in action.

Parameter reports are all over the place in computational technology. complicated engineering simulations needs to run a number of instances with various inputs to successfully examine the relationships among inputs and outputs. stories like optimization, uncertainty quantification, and sensitivity research produce subtle characterizations of the input/output map. yet thorough parameter reviews are tougher whilst every one simulation is pricey and the variety of parameters is big. In perform, the engineer may possibly try and restrict a examine to crucial parameters, which successfully reduces the size of the parameter examine.

Show description

Read or Download Active Subspaces: Emerging Ideas for Dimension Reduction in Parameter Studies PDF

Best analysis books

Analysis 1

Dieses Lehrbuch ist der erste Band einer dreiteiligen Einf? hrung in die research. Es ist durch einen modernen und klaren Aufbau gepr? gt, der versucht den Blick auf das Wesentliche zu richten. Anders als in den ? blichen Lehrb? chern wird keine ok? nstliche Trennung zwischen der Theorie einer Variablen und derjenigen mehrerer Ver?

Fouriertransformation fuer Fussgaenger

Dieses unterhaltsame Lehrbuch wendet sich an alle, die in der Ausbildung und in ihrer beruflichen Praxis mit Fouriertransformationen zu tun haben. Das Buch behandelt sowohl Fourierreihen als auch kontinuierliche und diskrete Fouriertransformationen. Außerdem werden Fensterfunktionen ausführlich diskutiert.

The Challenges of Privatization: An International Analysis

This e-book offers the 1st systematic empirical research of privatization strategies world wide to give an explanation for how and why governments privatize. Privatization is proven to be a tricky method, formed through political personal tastes and budgetary constraints, frequently pursued within the absence of appropriate economics and felony associations.

Exploratory Data Analysis Using Fisher Information

The fundamental objective of a learn scientist is to appreciate a given, unknown procedure. This leading edge ebook develops a scientific strategy for reaching this objective. All technology is finally established upon remark which, in flip, calls for a move of knowledge. Fisher details, specifically, is located to supply the most important to figuring out the procedure.

Extra info for Active Subspaces: Emerging Ideas for Dimension Reduction in Parameter Studies

Sample text

To answer these questions, we apply recent work by Tropp [121] and Gittens and Tropp [59] on the spectrum of sums of random matrices. The gradient vector ∇x f (x) is a deterministic function of x, so what is random? Randomness comes from the independent samples of x from the density ρ—a standard interpretation of Monte Carlo integration. We use the following notation to develop the results. For two symmetric matrices A and B, A B means that B − A is positive semidefinite. The notation λk (·) and λmax (·) with matrix arguments denote the kth eigenvalue and the maximum eigenvalue of the matrix, respectively.

There is a strong bias in the estimates of the subspace error when the phantom eigenvalues appear. 4, the estimates of the error for subspaces of dimension 4 through 6 are biased for h = 10−3 and significantly biased for h = 10−1 . 3. 3. Eigenvalues, estimates, and bootstrap intervals using finite difference gradients with h = 10−1 ((a)–(c)), h = 10−3 ((d)–(f)), and h = 10−5 ((g)–(i)). The horizontal black lines indicate the value of h in each plot. In general, estimates of eigenvalues smaller than h are less accurate than those larger than h.

J Choose V+ as an orthogonal matrix of size m × m − k + 1 that satisfies V+T ( X j )V+ , μk = λmax j and define V+T Σ2j V+ . σk2 = λmax j Then for any t ≥ 0, ⎡ ⎤ ⎣λk X j ≥ μk + t ⎦ ≤ j 1 (m − k + 1) · exp{− 4 t 2 /σk2 }, 1 (m − k + 1) · exp{− 4 t /B}, t ≤ σk2 /B, t ≥ σk2 /B. 28 Chapter 3. Discover the Active Subspace Proof. 21). First, note that ⎡ ⎤ M ⎣λk ˆ ≥ λ (C) + t = λk (C) k j =1 ≥ M λk + M t ⎦ . 4. We check that the bound on the gradient’s norm implies that the matrix ∇x f ∇x f T satisfies the subexponential growth condition: p ∇x f ∇x f T (∇x f T ∇x f ) p−1 ∇x f ∇x f T ρ d x ρ dx = p−1 L2 p!

Download PDF sample

Active Subspaces: Emerging Ideas for Dimension Reduction in by Paul G. Constantine
Rated 4.13 of 5 – based on 37 votes