Iterative linear solvers as metaphor

Gaussian elimination is systematic way to solve systems of linear equations in a finite number of steps. Iterative methods for solving linear systems require an infinite number of steps in theory, but may find solutions faster in practice.

Gaussian elimination tells you nothing about the final solution until it’s almost done. The first phase, factorization, takes O(n^3) steps, where n is the number of unknowns. This is followed by the back-substitution phase which takes O(n^2) steps. The factorization phase tells you nothing about the solution. The back-substitution phase starts filling in the components of the solution one at a time. In application n is often so large that the time required for back-substitution is negligible compared to factorization.

Iterative methods start by taking a guess at the final solution. In some contexts, this guess may be fairly good. For example, when solving differential equations, the solution from one time step gives a good initial guess at the solution for the next time step. Similarly, in sequential Bayesian analysis the posterior distribution mode doesn’t move much as each observation arrives. Iterative methods can take advantage of a good starting guess while methods like Gaussian elimination cannot.

Iterative methods take an initial guess and refine it to a better approximation to the solution. This sequence of approximations converges to the exact solution. In theory, Gaussian elimination produces an exact answer in a finite number of steps, but iterative methods never produce an exact solution after any finite number of steps. But in actual computation with finite precision arithmetic, no method, iterative or not, ever produces an exact answer. The question is not which method is exact but which method produces an acceptably accurate answer first. Often the iterative method wins.

Successful projects often work like iterative numerical methods. They start with an approximation solution and iteratively refine it. All along the way they provide a useful approximation to the final product. Even if, in theory, there is a more direct approach to a final product, the iterative approach may work better in practice.

Algorithms iterate toward a solution because that approach may reach a sufficiently accurate result sooner. That may apply to people, but more important for people is the psychological benefit of having something to show for yourself along the way. Also, iterative methods, whether for linear systems or human projects, are robust to changes in requirements because they are able to take advantage of progress made toward a slightly different goal.

More linear algebra posts

Multiple zeta

The Riemann zeta function, introduced by Leonard Euler, is defined by

\zeta(k) = \sum n^{-k}

where the sum is over all positive integers n.

Euler also introduced a multivariate generalization of the zeta function

\zeta(k_1, \ldots, k_r) = \sum n_1^{-k_1}\cdots n_r^{-k_r}

where the sum is over all decreasing k-tuples of positive integers. This generalized zeta function satisfies the following beautiful identity:

 \zeta(a)\,\zeta(b) = \zeta(a, b) + \zeta(b, a) + \zeta(a+b)

The multivariate zeta function and identities such as the one above are important in number theory and are the subject of open conjectures.

Source

Benchmarking C++, Python, R, etc.

The other day Travis Oliphant pointed out an interesting paper: A Comparison of Programming Languages in Economics. The paper benchmarks several programming languages on a computational problem in economics.

All the usual disclaimers about benchmarks apply, your mileage may vary, etc. See the paper for details.

Here I give my summary of their summary of their results. The authors ran separate benchmarks on Mac and Windows. The results were qualitatively the same, so I just report the Windows results here.

Times in the table below are relative to the fastest C++ run.

Language Time
C++ 1.00
Java 2.10
Julia 2.70
CPython 155.31
Python with Numba 1.57
R 505.09
R using compiler package 243.38

 

The most striking result is that the authors were able to run their Python code 100x faster, achieving performance comparable to C++, by using Numba.

Weak static type systems

Comment by Simon Peyton Jones in an interview:

People often dislike static type systems because they’ve only met weak ones. A weak or not very expressive type system gets in your way all the time. It prevents you from writing functions you want to write that you know are fine. … The solution is not to abandon the type system but to make the type system more expressive.

In particular, he mentions Haskell’s polymorphic types and type inference as ways to make strong static typing convenient to use.

Closed-world expertise

Venkat Rao has an interesting take on the ideas of deliberate practice, flow, and the 10,000 hour rule. In The Deliberate Practice of Disruption he points out that these ideas of expertise are always presented in closed worlds.

The real problem is that research on expertise focuses on fields where “expertise” is a well-posed and objectively codified notion. This means mature fields that are closed and bounded, and can be easily observed, modeled and studied under laboratory conditions. So it is not surprising that the work … is based on fields like “medicine, music, chess and sports” … all sharply circumscribed and regulated domains.

Rao’s essay may be a bit too harsh on closed-world domains and a bit too romantic about open-world domains, but the distinction between the domains is important. You don’t become a successful entrepreneur, for example, the same way you become a successful violinist. Closed worlds place a much higher emphasis on error-elimination, at least initially, than do open worlds. In an open world, the concept of an error may not even make sense. Where there is no law there is no sin.

Consulting is a more open world than academia. As Rao notes, academia can close off an otherwise open world through “bureaucratic productivity measures like publications and citations.” Clients are happy if you solve their problems. They could not care less whether your solutions are publishable and all that implies. Original and thoroughly footnooted work that doesn’t solve their problems is not appreciated.

Clients are not going to give you an oral exam to see whether you’ve mastered some canon. And they don’t care if you cross academic boundary lines to use something “outside your field.” They do care about credentials sometimes, but in a pragmatic way: they may need someone with the right credentials to review something. In that case, your credentials are part of the solution.

Related posts

Every exercise in the book

When I did an independent study course with Ted Odell, he told me to get a copy of De Vito’s Functional Analysis and work every exercise. He said that if I were taking a class with him, he wouldn’t assign every exercise, but that I needed to work more problems to make up for not having the structure of a class.

I don’t recall whether I actually worked every problem, though I believe I at least did most of them. I know of someone who learned algebraic geometry by working every problem in Hartshorne, and I’ve heard other people recommend the same.

Doing all the exercises in a book isn’t a bad way to learn something, though it depends on the book, what you’re trying to accomplish, and on the quality and quantity of the exercises.

Have you ever gone through a book working every exercise? If so, what book? How was your experience?

Swedish superellipse

Sergels torg, Stockholm, Sweden. Photo via Wikipedia.

The fountain in Sergels torg (English: Sergel’s Square) in Sockholm is in the shape of a “superellipse.” Piet Hein proposed this shape as a practical and aesthetically pleasing compromise between a circle and a rectangle.

What Piet Hein dubbed a superellipse is a curve satisfying

\left| \,\frac{x}{a}\,\right|^{5/2} + \left| \,\frac{y}{b}\,\right|^{5/2} = 1

In the case of Sergels torg, a/b = 6/5. A superellipse is what an analyst would call an Lp ball with p = 2.5 and rescaled axes.

Incidentally, I used curves of this form in a dose-finding method a few years ago. A clinical trial design would pick a, b, and p to match a physician’s desired trade-off between probability of efficacy and probability of toxicity.

Update: A few minutes after I posted this, I realized that there was a bowl on my desk shaped like a superellipse:

The bowl came from Ikea, so there’s another connection to Sweden.

Related post: Volumes of Lp balls

The door was unlocked

From the dedication of C. S. Lewis’ A Preface to Paradise Lost:

Apparently the door of the prison was really unlocked all the time; but it was only you who thought of trying the handle.

In context he was saying that Charles Williams had recovered a clear understanding of Milton that had been obfuscated for over a century.

That line made me think of a quote from Alexander Grothendieck (that I haven’t been able to find this morning) to the effect that progress in mathematics has often waited for someone to be bold enough to ask a simple question or introduce a simple concept.

Update: Thanks to Roland Elliot for providing the quote I was missing. Ronald Brown says that Grothendieck said in a letter to him in 1982:

The introduction of the cipher 0 or the group concept was general nonsense too, and mathematics was more or less stagnating for thousands of years because nobody was around to take such childish steps …