Not quite. In the U.S. the primary concern of anti-trust review for mergers and acquisitions is if it benefits or harms the consumer. In the EU, the primary concern is if the merger adversely impacts competition in the market. The failed acquisition of Honeywell by GE was blocked by the EU after being approved by the U.S. The U.S. review stated that bundling of avionics and engines, with some oversight of aircraft leasing by GE would reduce costs for aircraft and, as such, would not harm the consumer. The EU ruled that the merger would provide GE an unfair advantage against European jet engine manufacturers and blocked it.
The beating that NSF, NIST, and DOE Office of Science took in the FY08 budget should be evidence enough that we need to make a better case about the value of science. Despite the rhetoric and the America COMPETES act, science is not a budget priority. Washington listens to voters, and if the voters seem indifferent to science funding then congress is indifferent to science funding.
Learning perl as a first language is a bit like turning 21 in Las Vegas with a big wad of cash in your pocket.
There are simply too many options and not enough experience to deal with it.
Perl allows everything from OO to shell script type syntax and you do see everything in between.
I love perl, but for a first language, you are much better off with a simpler language that has strong type checking and strongly encourages structured programming practices.
I am sure the government and large media companies would love the right to designate who is and who is not a journalist.
I think we are safer regarding all citizens as journalists.
Of course, you just need to be able to sample fast enough.
A record is a physical device as is the needle and player; it cannot reproduces all frequencies, only a finite range. You simply cannot create fine enough details in the vinyl to capture very high frequencies when the track is moving relatively slowly (70 RPM or whatever it is). If you know the frequency limits of the physical system, you could then sample at a high enough rate to exactly capture and reproduce the signal digitally.
For its intended purposes, MPI is incredibly successful and for good reason. Virtually all large scale computational fluid dynamics codes are built on MPI and scale well to thousands of processors. MPI is successful because it isn't so ambitious. It is pragmatic and straightforward. Each process has its own address space so we don't have to put locks on every data structure. Current MPI implementations work transparently whether you are on a single SMP machine or across a network. It is also portable so you don't have to rewrite the parallel communication every time you are fortunate enough to get hours on a new machine. It is misleading to say parallel programming has gone nowhere, when with only a little effort you can take a serial numerical code, add domain decomposition and a few MPI calls and you'll have a code that will scale reasonably to 64 or more processors while portable to multiple clusters. There hasn't been a silver bullet, but parallel scientific computing is now easily in reach of most practitioners. It wasn't that way a decade ago. It would be nice if there was a portable, efficient language that would automatically parallelize the code and scale linearly to thousands of processors, but until that shows up, MPI allows us to run on large machines without extraordinary amounts of effort.
No, MPI is not a reasonable way to do fine scale parallelism. However, efficient fine scale parallelism is mostly architecture dependent, so it is a natural to have compilers and languages manage those aspects, which they have been doing for years with some limited success. That is going to be increasingly important, but that wasn't the problem domain MPI was designed for. I suspect the large scale parallelism with still be done with MPI in another decade.
The second order wave equation can be recast as a first order hyperbolic system. The literature on numerical approximation of hyperbolic systems is extensive. An explicit finite volume implementation should work and can handle irregular geometries. The books by R. Leveque, "Finite Volume Methods for Hyperbolic Problems" and
"Numerical methods for conservation laws" provide a good introduction. For implementations on unstructured grids, there are a couple of review articles by T. Barth that appear in the VKI Lecture series (look for his web page at NASA). The book "Numerical Methods for Wave Equations in Geophysical Fluid Dynamics" is well regarded, but I don't have any experience with it.
If the CFL constraint is so problematic, you will have to go to an implicit time integration scheme, which brings linear solvers, preconditioners, and (possibly) inexact Newton solves into the picture. If you want a quick implementation, stick to explicit time integration and live with the CFL constraint.
If the wave speed varies abruptly you are going to have problems with any method. Try a smoothly varying wave speed and see if that makes a difference.
It is good news, but it shows the value of opening up scientific research. NASA has not disclosed the statistical correction algorithms (or source code) which they use to correct the raw data. Had they opened the source code or at least published, in detail, the algorithms used with the raw data, than the problem would have likely been caught much earlier.
There are strong connections between mimetic finite differences and mixed (Raviart-Thomas) finite elements. I wouldn't say it has "it has nothing to do with the finite element/difference methods at all".
With the disclaimer that I only gave the paper a cursory reading...
It is similar. They use what is referred to as a consistent or mimetic discretization where the discrete operators have some of the same properties as the continuous operators, i.e. div curl v = 0 in the continuous case becomes
DIV CURL v = 0 where DIV and CURL are the discrete version of the divergence and curl (i.e. matrices). It makes it easier to get local conservation. In this case, they focus on conserving circulation which is an integral of the vorticity in the fluid.
The "breaking problems into smaller pieces" here is simply the spatial descritization. In this case, tets in 3D or triangles in 2D. They are partitioning the physical domain into a set of small regions, basically like finite elements or finite volume methods. The divergence free constraint and viscous terms still result in a global linear algebra problem to be solved at each time step. So, in this case, it doesn't really have anything to do with multi-core processors. It isn't any more (or less) amenable to parallel computation than most other methods.
America has always relied on foreign scientists. Many of the premier scientists of the Manhattan project were immigrants from Europe (i.e. the famed Hungarian Conspiracy of Wigner, Szilard, Teller, and von Neuman). It hasn't changed. Los Alamos National Laboratory relies more on foreign scientists now than ever. Over the last five years, the percentage of postdocs at Los Alamos that are U.S. citizens has decreased from 52% to 40% while the total postdoc population has increased from 423 to 515. The perception is that it is more difficult to attain a visa now, but I'm not convinced that is true. There is no shortage of applicants for positions.
I've talked with some of the foreign born researchers, and they claim that to attain an academic position in Europe typically requires a few years of post graduate research in the U.S. As long as the U.S. maintains premier scientific institutions and funding, this won't change. We will lead in science, but we will do it with a lot of foreign scientists as we always have. Developing the next generation is important, but getting the current generation of scientists to come to the U.S. is equally important. I'm sure some are discouraged by the
by U.S. immigration policies (real or perceived), but overall we are still doing reasonably well.
Definitely look at the national labs. I have worked at three different DOE laboratories and each has a different culture. The smaller, DOE labs (Argonne, PNNL, Lawrence-Berkeley, etc) have a very academic feel to them. The larger labs (Los Alamos, Sandia, Livermore) are more industrial, but, in general, have better facilities and funding. It is important to remember that even within one lab there is dramatic variation between groups. It depends on the mission of the group and the people that work there. I have found the group I am currently doing my postdoc in to be the best compromise between adacemia and industry. We have a lot of freedom to pursue the research we want, there just has to be funding for it.
I have also found the labs to be more open to interdisciplinary research. Our group has mathematicians, physicists, engineers and even a chemist or two around.
This may help, however, there are other promising approaches. We no longer have to rely on low order approximations to curves for finite element analysis. In some applications, like thin shells, low order representations of surfaces tend to give unreliable results. Prof. Tom Hughes and his research group at U. of Texas have been working on finite element methods using NURBS. The obvious advantage is the easier and more accurate representation of surfaces, as you are more closely tied to what CAD programs are actually doing. I don't know if they are doing anything with regards to CAD to mesh software, but a NURBS based method would seem to have an advantage is this context.
Sam
Disclaimer: I come from a spectral/high-order methods background, so I have a bit of a bias.
That is a horrible idea. Microsoft has a responsibility to shareholders and the EU commission has a responsibility to EU citizens. The proceedings should be much more open. Imagine that the proceedings in SCO v. IBM, SCO v. Novell, and SCO v. everything that is good in the world were regarded as confidential. We wouldn't have Groklaw and the non-technical world might still believe there is something to the case. The accusations are made public, but it could be years before case is resolved.
The whole process should be more open. This isn't a case between individuals. It is between a government and a large corporation. I don't trust MS and I don't trust government. I want to know what they are up to. Obviously Microsoft's motives are not sincere, but openness in government should always be the goal. Furthermore, there are some basic free speech concerns. MS and its shills have a right to present their case to the public. Similarly, MS critics have a right to present the overwhelming evidence of MS wrongdoing. Do you really want the government to file a case against you and then tell you are not allowed to disclose anything they say in court?
I'm starting a postdoc at Los Alamos in a month. From discussions I have had with people there, this is a good thing. After Lockheed Martin took over Sandia National Lab, publications declined as they focused research on core programs. In addition, the increased focus on classified programs at Sandia had several foreign nationals concerned about their future at LANL. As a postdoc, it doesn't affect me much either way, but permanent employees are likely better off with UC stll in charge.
The root problem is our public school system. This fight will continue and become increasingly ugly. It is incredible that a court has to decide what our children will or will not be taught in schools. Ultimately, the educational decisions should be made by parents, not judges or activists. Unfortunately, the most vocal groups can force the entire community to be educated of their beliefs. In my opinion, ID has no place in a science class, but I'm not going to be calling the school board to let them know. In the future, the court rulings may go the other way. The only way to prevent these sort of battles is to change the system. Until parents have the freedom to choose among several schools, curriculum battles will go on.
Sam
There is still a lot of tweaking going on in scientific computing. I work on large scale parallel computational fluid dynamics codes where performance can be tricky, especially when dealing with parallel iterative solvers. We often have to build and try a variety of preconditioners and simply measure which gives us the best performance. One preconditioner may lower the iteration count substantially, but it is quite expensive to apply; another may require excessive communication. You may also find that what works great on 32 processors deteroriates quickly once you scale to 256 processors. The point is that you rarely know in advance what the optimal strategies are.
Of course, these "tweaks" are related the global numerical scheme. Reordering a loop here and there then running to see if it made a difference is simply not practical as you point out.
The point of the legislation was to remove the ambiguity regarding in kind contributions with respect to internet sites. If you believe that government should not allow people to spend too much money to advocate their position, then you can argue that this particular exemption was overly broad, but it does not alleviate the obvious need for free speech protections for individual bloggers. You should read the brief submitted by the EFF in regards to FEC regulation of the internet.
You are missing the point. If you are paid to shill by a candidate, you should disclose it. The problem with campaign finance restrictions is when you are paying to speak your mind. Say you run a popular blog and pay $500 a month for bandwidth. By endorsing a candidate you may violate campaign finance rules since you are contributing money, in the form of advertising and bandwidth, to a candidate. This is the problem. It becomes more pronounced for the larger blogs (Daily Kos, Instapundit, etc) whose bandwidth costs are substantial and require outside funding to pay the bills. They can easily pass monetary thresholds required for enforcement of the laws.
A recent case in Washington illustrates exactly this problem. Radio personalities in Seattle had been speaking out against a gax tax proposal. The local government decided that amount to an in-kind contribution to the anti-gas tax campaign, so they violated campaign finance rules. They were expressing their own opinion and were not paid to by gas tax foes. This same scenario applies to internet speech. If you pay for bandwidth and express an opinion, you may be violating election laws.
I don't think this is what free speech in a democracy is: say what you want, just don't spend too much money doing it, and be careful in which format you choose to speak, and try not to mention candidates by name.
Slashdot Mirror
Not quite. In the U.S. the primary concern of anti-trust review for mergers and acquisitions is if it benefits or harms the consumer. In the EU, the primary concern is if the merger adversely impacts competition in the market. The failed acquisition of Honeywell by GE was blocked by the EU after being approved by the U.S. The U.S. review stated that bundling of avionics and engines, with some oversight of aircraft leasing by GE would reduce costs for aircraft and, as such, would not harm the consumer. The EU ruled that the merger would provide GE an unfair advantage against European jet engine manufacturers and blocked it.
The beating that NSF, NIST, and DOE Office of Science took in the FY08 budget should be evidence enough that we need to make a better case about the value of science. Despite the rhetoric and the America COMPETES act, science is not a budget priority. Washington listens to voters, and if the voters seem indifferent to science funding then congress is indifferent to science funding.
Perl allows everything from OO to shell script type syntax and you do see everything in between. I love perl, but for a first language, you are much better off with a simpler language that has strong type checking and strongly encourages structured programming practices.
20% reduction in fuel used + 20% of fuel costs sold as carbon offsets = 40% reduction in fuel costs.
I am sure the government and large media companies would love the right to designate who is and who is not a journalist. I think we are safer regarding all citizens as journalists.
A record is a physical device as is the needle and player; it cannot reproduces all frequencies, only a finite range. You simply cannot create fine enough details in the vinyl to capture very high frequencies when the track is moving relatively slowly (70 RPM or whatever it is). If you know the frequency limits of the physical system, you could then sample at a high enough rate to exactly capture and reproduce the signal digitally.
No, MPI is not a reasonable way to do fine scale parallelism. However, efficient fine scale parallelism is mostly architecture dependent, so it is a natural to have compilers and languages manage those aspects, which they have been doing for years with some limited success. That is going to be increasingly important, but that wasn't the problem domain MPI was designed for. I suspect the large scale parallelism with still be done with MPI in another decade.
If the CFL constraint is so problematic, you will have to go to an implicit time integration scheme, which brings linear solvers, preconditioners, and (possibly) inexact Newton solves into the picture. If you want a quick implementation, stick to explicit time integration and live with the CFL constraint.
If the wave speed varies abruptly you are going to have problems with any method. Try a smoothly varying wave speed and see if that makes a difference.
Sam
It is good news, but it shows the value of opening up scientific research. NASA has not disclosed the statistical correction algorithms (or source code) which they use to correct the raw data. Had they opened the source code or at least published, in detail, the algorithms used with the raw data, than the problem would have likely been caught much earlier.
There are strong connections between mimetic finite differences and mixed (Raviart-Thomas) finite elements. I wouldn't say it has "it has nothing to do with the finite element/difference methods at all".
It is similar. They use what is referred to as a consistent or mimetic discretization where the discrete operators have some of the same properties as the continuous operators, i.e. div curl v = 0 in the continuous case becomes DIV CURL v = 0 where DIV and CURL are the discrete version of the divergence and curl (i.e. matrices). It makes it easier to get local conservation. In this case, they focus on conserving circulation which is an integral of the vorticity in the fluid.
The "breaking problems into smaller pieces" here is simply the spatial descritization. In this case, tets in 3D or triangles in 2D. They are partitioning the physical domain into a set of small regions, basically like finite elements or finite volume methods. The divergence free constraint and viscous terms still result in a global linear algebra problem to be solved at each time step. So, in this case, it doesn't really have anything to do with multi-core processors. It isn't any more (or less) amenable to parallel computation than most other methods.
I've talked with some of the foreign born researchers, and they claim that to attain an academic position in Europe typically requires a few years of post graduate research in the U.S. As long as the U.S. maintains premier scientific institutions and funding, this won't change. We will lead in science, but we will do it with a lot of foreign scientists as we always have. Developing the next generation is important, but getting the current generation of scientists to come to the U.S. is equally important. I'm sure some are discouraged by the by U.S. immigration policies (real or perceived), but overall we are still doing reasonably well.
Sam
Definitely look at the national labs. I have worked at three different DOE laboratories and each has a different culture. The smaller, DOE labs (Argonne, PNNL, Lawrence-Berkeley, etc) have a very academic feel to them. The larger labs (Los Alamos, Sandia, Livermore) are more industrial, but, in general, have better facilities and funding. It is important to remember that even within one lab there is dramatic variation between groups. It depends on the mission of the group and the people that work there. I have found the group I am currently doing my postdoc in to be the best compromise between adacemia and industry. We have a lot of freedom to pursue the research we want, there just has to be funding for it.
I have also found the labs to be more open to interdisciplinary research. Our group has mathematicians, physicists, engineers and even a chemist or two around.
Sam
Sam
Disclaimer: I come from a spectral/high-order methods background, so I have a bit of a bias.
The whole process should be more open. This isn't a case between individuals. It is between a government and a large corporation. I don't trust MS and I don't trust government. I want to know what they are up to. Obviously Microsoft's motives are not sincere, but openness in government should always be the goal. Furthermore, there are some basic free speech concerns. MS and its shills have a right to present their case to the public. Similarly, MS critics have a right to present the overwhelming evidence of MS wrongdoing. Do you really want the government to file a case against you and then tell you are not allowed to disclose anything they say in court?
Sam
Sam
The root problem is our public school system. This fight will continue and become increasingly ugly. It is incredible that a court has to decide what our children will or will not be taught in schools. Ultimately, the educational decisions should be made by parents, not judges or activists. Unfortunately, the most vocal groups can force the entire community to be educated of their beliefs. In my opinion, ID has no place in a science class, but I'm not going to be calling the school board to let them know. In the future, the court rulings may go the other way. The only way to prevent these sort of battles is to change the system. Until parents have the freedom to choose among several schools, curriculum battles will go on. Sam
Of course, these "tweaks" are related the global numerical scheme. Reordering a loop here and there then running to see if it made a difference is simply not practical as you point out.
Sam
The point of the legislation was to remove the ambiguity regarding in kind contributions with respect to internet sites. If you believe that government should not allow people to spend too much money to advocate their position, then you can argue that this particular exemption was overly broad, but it does not alleviate the obvious need for free speech protections for individual bloggers. You should read the brief submitted by the EFF in regards to FEC regulation of the internet.
You are missing the point. If you are paid to shill by a candidate, you should disclose it. The problem with campaign finance restrictions is when you are paying to speak your mind. Say you run a popular blog and pay $500 a month for bandwidth. By endorsing a candidate you may violate campaign finance rules since you are contributing money, in the form of advertising and bandwidth, to a candidate. This is the problem. It becomes more pronounced for the larger blogs (Daily Kos, Instapundit, etc) whose bandwidth costs are substantial and require outside funding to pay the bills. They can easily pass monetary thresholds required for enforcement of the laws.
A recent case in Washington illustrates exactly this problem. Radio personalities in Seattle had been speaking out against a gax tax proposal. The local government decided that amount to an in-kind contribution to the anti-gas tax campaign, so they violated campaign finance rules. They were expressing their own opinion and were not paid to by gas tax foes. This same scenario applies to internet speech. If you pay for bandwidth and express an opinion, you may be violating election laws.
I don't think this is what free speech in a democracy is: say what you want, just don't spend too much money doing it, and be careful in which format you choose to speak, and try not to mention candidates by name.