I have just started with my Masters thesis in Organic chemistry. All the administrative meetings now, and in a week or two I can start with my experiments. Really looking forward to it, I love doing research with (nearly) a carte blanche.

>>1204 What kind of research are you working on? I'm in engineering, so unfortunately our research tends to be "build something, prove it's useful or better in some way". I'm curious how novel and out-there you can get with chemistry while doing a Masters.

>>1211 Catalysis of halogenated compounds in water. Basically: halogenated compound synthesis need solvents that are expensive and rather unsustainable. However, the synthesis does not really work in water, so using all kinds of polymers to stabilize a catalyst in a suspension does allow the reaction to take place in water, which in turn allows for easy separation. I can't go into detail since NDA, but I will be using some fancy coordinated catalysts which I will modify to suit my needs. It is a pretty new field - even in chemistry, where a certain kind of supermolecular interaction will hopefully make this possible. In terms of how novel I can go depends completely on the group you are working in - I worked myself into a pretty high-standing research group, so I am actually working on completely new concepts. Some of my yearmates are currently working on projects that have a more simple layout, but not less "on the frontier". Depends on group and university, I chose both pretty well and worked hard for it.

Which programming languages should I learn if I'm getting into theoretical physics? I already know LaTeX, which is basically a requirement and not a programming language, can wrangle some libraries in Python to plot and do numerical calculations a la MATLAB, and have started learning C. I have a feeling it's going to be either Python, Julia, and maybe R, but that is falling out of fashion. Fortran or C if I really need to write a library for doing expensive ass calculations, but that seems unlikely.

>>1231 Over in the programming thread I had a discussion about these programs, but theoretically, C+GSL is probably the most optimized. If you don't care about speed or if you just need to do simple and repetitive calculation Python may be a better option because of it's huge internal library and the fact that graphing, file IO and complex calculations have a fuckton of libraries. For my courses, I used MATLAB because our uni has a licence. It's a piece of closed source shit through. Depending on what level of calculations you'll be doing, you should try to find out what you need. Physical simulations, ODEs, lots of graphs? High-level might be better for easy modifications of your environments. Photophysical calculations? try something more low-level. Subatomics? You'll have to write your own programs, because there aren't really any libraries for you to choose from.

>>1231 This depends on whether you need high-performance or not. I know a lot of physics labs use Python. Many of them use Jupyter Notebooks. This has some upsides but I'd definitely encourage getting good with the core language. A lot of physicists I've helped over the years would struggle with basic issues and just tinker with lines until they got an expected result. I'd take a look and figure it out in a few second, and the reliance on Notebooks to let you just re-run a section didn't do them any favors. It's a nice tool for scientific computer once you know Python well enough, though. I'd also recommend figuring out how to self-host. At this stage, Python with Numpy and SciKit is the de facto standard for scientific computing. Python has libraries to make data management, statistical analysis, and machine learning easy. Most labs don't need high-performance, they need something easy to write. R is effectively dead because Python has all the same capabilities with easier syntax and it has more libraries you can shove the data into. Want a graph? Matplotlib. Want to train a model? SciKit. Need to do matrix math? Numpy. "Batteries Included" is most relevant for scientific computing. However, if you do need to write a high-performance model to crunch a lot of numbers (common in astrophysics and the like) then many physicists are still using FORTRAN. You can also get away with MATLAB, which is (or used to be, at least) a clever frontend for FORTRAN. However, I'd recommend using Octave. It's GNU, offers a few more features, and the syntax is identical excepting those few extra features. So it's 100% interoperable with your colleagues paying out the nose for a license. In my experience, these are the two most common cases. Statistical (AKA big data and machine learning) go with Python, for complex high-performance simulations go with FORTRAN/MATLAB.

>>1232 >For my courses, I used MATLAB because our uni has a licence. It's a piece of closed source shit through. In my degree, there was a course called Statistical and Numerical Methods which basically taught us to write just that in MATLAB. Nobody liked the course because the professors were garbage, and I didn't exactly like MATLAB either, so I wouldn't be giddy about having to use that again. As a matter of fact, I remember tinkering with Octave to practice a bit on my own, but I didn't really dig it. >Depending on what level of calculations you'll be doing, you should try to find out what you need. I will be studying quantum field theories, so if I end up doing something bleeding edge I might need something performant. I suppose that beside all the basics and some more, I'll need to figure out multithreading, right? Actually, might even have to suck up to Nvidia, I think there's a bunch of calculations that can be quite a bit parallelized, since it's nothing but matrices sometimes... >>1233 >I'd also recommend figuring out how to self-host. Selfhosting the Jupyter server, you mean? I already figured how to run it locally a while ago, honestly do not understand why they don't make it more simple for a man to just do that instead of having to run something in the c l o u d. Didn't figure out how to let people access the server though. >At this stage, Python with Numpy and SciKit is the de facto standard for scientific computing. SciKit and not Scipy? Scipy might just bundle a few libraries and provide constants, but it did have an ODR package that saved my ass countless times. Haven't found anything similar in SciKit. I did see SciKit being used heavily in machine learning and probably statistics. >However, if you do need to write a high-performance model to crunch a lot of numbers (common in astrophysics and the like) then many physicists are still using FORTRAN. You can also get away with MATLAB, which is (or used to be, at least) a clever frontend for FORTRAN. >However, I'd recommend using Octave. It's GNU, offers a few more features, and the syntax is identical excepting those few extra features. I heard Octave's slower than MATLAB though. Is Octave sort of a clever Fortran frontend as well? Or does it at least use the GSL? Thanks for the feedback. I'll keep hammering away at C and start looking at Fortran. I only have a cursory understanding of Python's syntax, I really only chained together packages to do my bidding. I suppose I'll look how it goes when I need to do something with it.

>>1237 I'm actually not sure where I heard about MATLAB being a FORTRAN frontend. As far as I can tell both it and Octave are implemented in themselves and C. But I haven't looked that closely. Either way, be warned that I am not an expert on this topic (since I just use Python for all scientific computing) and may be speaking out of my ass. I found a StackOverflow post on the issue of performance (https://archive.vn/wip/T4OPD) and it seems like MATLAB is indeed faster due to JIT compilation and hooks into low-level non-free libraries. As far as I know the guys doing particle physics calculations still write raw FORTRAN. But in my experience in a physics lab (also cutting-edge stuff) they were primarily using Jupyter Notebook Python, and I was fixing their scripts and delivering machine learning models in pure Python. The truth is that Python is slow on its own but its machine learning libraries are quite robust and if they have support for GPUs then that means they support CUDA. I wouldn't recommend learning CUDA just to do this sort of thing. Libraries for other languages, especially Python, handle the busywork. Implementing backprop from scratch is a pain in the ass and hard to get right. Scaling that to CUDA is another great leap. Or you could load your data with Pandas, run a bunch of models on it, see which ones converge, and then try to get the best results. You need to do research and pick which library you use carefully but at this stage it's becoming the standard for machine learning. Again, just because your lab is "cutting edge" doesn't mean you'll be writing simulations of particle physics. In many cases, aggregating data and building a predictive model is sufficient for the purposes of research. High-performance applications are more applicable at NASA than in a research lab. Pick what workflow works best for you and your colleagues. Just beware that C and C++ are systems languages and require a lot of development time. If you aren't well-versed in writing neural network training algorithms then I wouldn't go down that road. It's a deep rabbit hole with tons of poorly-understood empirical results and almost no theory, much to the chagrin of researchers in this area. I've had the pleasure of working with one of the top machine learning researchers in the world and even he says "just use a pre-trained model and stick a softmax on it". Don't reinvent the wheel. Find a Python library or a MATLAB library that works for your needs and learn the ins and outs.

>>1238 You fucked up your hyperlink https://archive.vn/wip/T4OPD

Isn't it great when your polymerization only proceeds to 40% instead of 70%? 6 hours and a day wasted, really agitates my skeleton. Image related: me looking at my NMR spectrum.

>>1202 /sci/ now exists

>>1367 I wish you luck. Not sure there's enough activity for a whole board, though. /sci/ was already pretty slow/dead before CF pulled the plug. You'll need to find somewhere to pull anons from who are into science and math.

>>1367 Nice, maybe I'll give it some traffic some time soon.

>>1375 Done, come at a look at >>>/sci/.

Aricebo fell apart completely. Dish is completely shattered and all three towers, and the sensors, are down. Total loss. The worst part? It's had no funding for years, the US probably won't rebuild it, and the closest thing to a replacement is a Chinese one that isn't even as good. This would be the perfect excuse to fund a replacement. https://archive.is/B539k