Is Numerical Analysis Used in Non-Financial Industries?

[lagrange476]

Are there any uses of numerical analysis in industry (outside of financial firms)? I get the impression that a lot of industrial software has already been written and standardized so you don't really have people coming up with innovative ways to e.g. numerically solve PDEs so much as using pre-made software and calibrating it to the problem at hand.

 

[UltrafastPED]

You still have to understand what is going on to setup the problems and verify the validity of the model.

Just because there are good packages for statistics does not mean that there is no need for statisticians any more!

 

[ModusPwnd]

Just because there are good packages for statistics does not mean that there is no need for statisticians any more!

But it does mean there is much less need for statisticians. This is basically the "smart cow problem".

 

[UltrafastPED]

The "smart cow" problem doesn't apply here ... what a stupid and demeaning expression! ... it would only apply when a particular problem only needs to be solved once. This is hardly true for computer programming, engineering, or any other technical field.

In general technological "progress" generates more problems which require technical solutions - that's why there are more programmers, engineers, and statisticians working today than ever in the past.

See projections by US Dept. of Labor: http://www.bls.gov/ooh/Math/Statisticians.htm

 

[StatGuy2000]

The notion that good packages for statistics somehow reduces the need for statisticians is a clear indication of the lack of understanding of the role of statisticians. A package only serves as a tool to build your model and analyze data. There is still the need to set up the problem, to perform exploratory analysis to understand the nature of the problem at hand (to help build the model), to test the model and interpret the results of the analysis. All of these require an in-depth understanding that cannot be done solely by a software package (no, not even those that incorporate "automated" algorithms such as neural networks, etc.)

 

[SteamKing]

Are there any uses of numerical analysis in industry (outside of financial firms)? I get the impression that a lot of industrial software has already been written and standardized so you don't really have people coming up with innovative ways to e.g. numerically solve PDEs so much as using pre-made software and calibrating it to the problem at hand.

You have a mistaken impression of the state of things. There is innovation constantly occurring in math and the sciences. I don't know what you mean by 'industrial software', but there is no lack of jobs for programmers in industry and academia. There are many problems in math (specifically solving PDEs) where the current solutions are non-optimal or don't exist at all.

 

[AlephZero]

You probably won't find many people in industry who call themselves "numerical analysts", because to make any significant contribution you need a deep understanding of the application are, not just the math. But you will find plenty of people with job titles like mechanical engineers, fluid dynamicists, control systems specialists, robotics engineers, computer software engineers, etc who are doing numerical analysis, if you look in the right places.

You didn't say what your own background is, but from what I see of questions asked on this forum, most of what is being taught in numerical analysis courses was "state of the art" say 50 or 60 years ago. Things have moved on just a little bit since then [/IRONY] but at the "first degree course" level that message might not be getting transmitted, or received.

 

[ModusPwnd]

The notion that good packages for statistics somehow reduces the need for statisticians is a clear indication of the lack of understanding of the role of statisticians. A package only serves as a tool to build your model and analyze data. There is still the need to set up the problem, to perform exploratory analysis to understand the nature of the problem at hand (to help build the model), to test the model and interpret the results of the analysis. All of these require an in-depth understanding that cannot be done solely by a software package (no, not even those that incorporate "automated" algorithms such as neural networks, etc.)

Your paragraph doesn't support your claim. If the packages do not reduce statistician's work load, why are they ever invented and used in the first place? If the packages do reduce statisticians work load then we need less statisticans because each one is more productive. Of course there is still need to set up the problem, perform exploratory analysis, etc. It was never claimed otherwise.

The same is true with respect to numerical analysis in general. Sure it may be an increasing field. But many people get to use pre-made code and software and thus they don't have to be experts in numerical analysis themselves. If there was no pre-made code and software then everybody would have to make their own, this means more hours of work and more people hired. Its the smart cow problem. Generally speaking, its a good thing. It allows large groups of people to get more done with less training. But if you are the second smartest cow and you want a job opening the pen, you may be out of luck. Only one is needed.

But more to the OP's point, I agree with other posters. Numerical analysis is a great skill to have and employers seem to value it. I wish my physics curriculum had more of it, rather than the ancient history that is usually taught in physic's math methods classes.

 

[StatGuy2000]

Your paragraph doesn't support your claim. If the packages do not reduce statistician's work load, why are they ever invented and used in the first place? If the packages do reduce statisticians work load then we need less statisticans because each one is more productive. Of course there is still need to set up the problem, perform exploratory analysis, etc. It was never claimed otherwise.

The "smart cow" problem you point out above makes 2 assumptions that are highly dubious, at least with respect to statistics (and probably to many other technical fields): (1) that the nature of the work that statisticians are involved with is static, and (2) the quantity of the work that statisticians are involved with is static.

Sure, the packages may reduce statistician's work load at one stage. But what then happens is that the statistician, having been able to more efficiently work on their given problems, will either take on more work or challenge the package to do even more, in essence pushing the use of the software to the limits. And furthermore, the increased capability of the statistician will also have a tendency to spur more demand for their services, thus leading to more work coming in and further requiring the services of more statisticians.

It's also worth keeping in mind that much of the demand for statisticians have been spurred by increased computing power which has allowed those in a variety of businesses, industries, and research organizations from collecting and analyzing ever larger quantities of data. This era of "big data" has only spurred greater demand for statisticians and those with similar cognate backgrounds, which at least in the immediate foreseeable future doesn't appear to be diminishing.

 

[chiro]

Computer programs free up the statisticians time and allow him or her to focus on higher level tasks.

Instead of wasting time with a pen, paper, and slide rule, statisticians can use more of their time to focus on the higher level parts like thinking about the context of the problem rather than performing calculations.

Its the same in an area whether it be engineering, accounting, or anything else. Accountants don't have to worry about doing double-entry ledger accounting: the computer does that for them. What they can do though is look at more higher level attributes like the statistician like how the various metrics have an effect on the organization they are dealing with.

I imagine that this will always happen in the future: as we get more and more sophisticated computational tools (numeric and symbolic), each addition will free up the mind of the analyst to think about higher and higher level issues in a deeper and deeper way and this is a good thing IMO.

 

[Locrian]

One of my favorite projects at work rebuilt a model to massively reduce prep-time and streamline analysis. As i was working on it I was excited to see how much total time the new process saved.

Zero. It saved no time because the additional time I opened up was used for further analysis. The final product was better, as a result.

The same is often true of software packages and other new models: they don't reduce the need for capable users, they allow those same users to do more.

 

[ParticleGrl]

So I think it depends on the company, and the industry. I've been working for a consulting company, for several of our projects we were essentially hired in order to replace analysts with software packages. Generally, a few statisticians and several analysts become one statistician, some IT support, and a software package.

 

[caldweab]

Your paragraph doesn't support your claim. If the packages do not reduce statistician's work load, why are they ever invented and used in the first place? If the packages do reduce statisticians work load then we need less statisticans because each one is more productive. Of course there is still need to set up the problem, perform exploratory analysis, etc. It was never claimed otherwise.

The same is true with respect to numerical analysis in general. Sure it may be an increasing field. But many people get to use pre-made code and software and thus they don't have to be experts in numerical analysis themselves. If there was no pre-made code and software then everybody would have to make their own, this means more hours of work and more people hired. Its the smart cow problem. Generally speaking, its a good thing. It allows large groups of people to get more done with less training. But if you are the second smartest cow and you want a job opening the pen, you may be out of luck. Only one is needed.

But more to the OP's point, I agree with other posters. Numerical analysis is a great skill to have and employers seem to value it. I wish my physics curriculum had more of it, rather than the ancient history that is usually taught in physic's math methods classes.

It has nothing to do with reducing workload and everything to do with calculations. Numerical calculations are calculations that just simply can't be done by hand. For example in engineering problem solving as one of our projects we went out and measured the truss on a mall bridge behind the engineering building and eventually we came up with about 50 linear equations which we then put into a matrix in Matlab, I'm really interested to know how you would solve that system by hand. Another project we were using finite element method to model heat flow in a system and that was another huge matrix. Point is you will still need the people that understand the theory and problem you are solving (engineers, statisticians,etc), because the program can only solve the problem once you give it the problem, so if you don't understand the science of what you doing the program results will be rubbish. I don't care how robust and powerful the packages get, it's not going to decrease the need for people who understand the subject or their workload. The packages are more so for doing computations that are just not possible by hand

 

[inotyce]

Are there any uses of numerical analysis in industry (outside of financial firms)? I get the impression that a lot of industrial software has already been written and standardized so you don't really have people coming up with innovative ways to e.g. numerically solve PDEs so much as using pre-made software and calibrating it to the problem at hand.

A lot, almost what you work with needs maths. To name a few,
In robot industry, people need to simulate how the robots actually work (simulation models), its movements should be 99.8-100% correct (matrices of data for input, transforming, measurement of its correctness, etc).
In network analysis, topological understandings, measurement of throughput, data transmission etc are necessary.
In game industry, AI or maths are emphasized. PDE is also used.

PDE is widely used in modeling real world objects. Modeling it helps us diagnose its current states and perform any calculations for its next state if required.