[HN Gopher] Physics Informed Neural Networks
       ___________________________________________________________________
        
       Physics Informed Neural Networks
        
       Author : nchagnet
       Score  : 89 points
       Date   : 2025-02-16 21:14 UTC (1 days ago)
        
 (HTM) web link (nchagnet.pages.dev)
 (TXT) w3m dump (nchagnet.pages.dev)
        
       | westurner wrote:
       | From "Physics-Based Deep Learning Book" (2021)
       | https://news.ycombinator.com/item?id=28510010 :
       | 
       | > _Physics-informed neural
       | networks:https://en.wikipedia.org/wiki/Physics-
       | informed_neural_networ... _
        
       | Jordanpomeroy wrote:
       | Very well explained to a lay person.
       | 
       | Are PINNs the current state of the art in ML methods for solving
       | PDEs? What are their limitations?
        
         | nchagnet wrote:
         | Thank you!
         | 
         | As far as I can tell, PINNs are promising and an active
         | research area, but they are also young and far from being as
         | widely adopted as finite element methods (at least that's my
         | experience academic environments).
         | 
         | I do see great improvements are being made both on the
         | performance level but also on the applications.
         | 
         | One aspect I didn't discuss in the post is the use for inverse
         | solution search, where you fit experimental data to your
         | equation, and where your parameters and your initial conditions
         | can also be trainable parameters. This has great potential to
         | improve the methodology of experimental results analysis.
        
         | cherryteastain wrote:
         | > Are PINNs the current state of the art in ML methods for
         | solving PDEs? What are their limitations?
         | 
         | I guess in a way they are. They aren't new, they have been
         | around since the 90s [1]. The problem with them is, you
         | typically need to train them on a specific problem (boundary
         | conditions, domain, equation, PDE coefficients etc). Compared
         | to a traditional solver, the training is much slower, and on
         | top of that the results are typically much less accurate. The
         | PDE + NN community has a bit of a problem dealing with this in
         | general [2], there are tons of papers that make NNs look much
         | better at solving PDEs than they are compared to traditional
         | solvers.
         | 
         | [1] https://www.cs.uoi.gr/~lagaris/papers/TNN-LLF.pdf
         | 
         | [2] https://www.nature.com/articles/s42256-024-00897-5
        
         | hansvm wrote:
         | It depends on the PDE and what you want to do with it. A PINN
         | requires:
         | 
         | 1. Some example data or other way to add boundary conditions
         | 
         | 2. Autograd over PDE constraints
         | 
         | 3. A training loop incorporating both of those
         | 
         | And it produces
         | 
         | a. An approximate, differentiable, mesh-free solution
         | 
         | PINNs are most applicable when (1) is expensive (since that
         | expense will apply more to traditional solvers, especially with
         | fine meshes) and when the error in (a) is acceptable.
         | 
         | Regarding the error, PINNs are still extremely useful in
         | generating an initial state to pass to a traditional solver
         | even when the error is not tolerable, so that's not _really_ a
         | concern. The main consideration is how expensive a particular
         | problem is to solve classically. If it's too cheap, the PINN
         | will never beat it.
         | 
         | You have a secondary consideration with (2) and (3). The
         | training loop is a fixed cost which you can amortize over many
         | executions, but you have to use the network enough times for
         | that to actually pay off.
         | 
         | The last point I want to bring up is that you can sometimes get
         | value from the extra features in (a). Perhaps you want to use
         | the PINN to figure out where your mesh should be finer, or you
         | have a derived field you want to inspect. Neural net gradients
         | in general tend to poorly approximate real gradients if you
         | only train on the function itself, but PINNs have the gradients
         | you're likely to care about baked into their definition (and
         | can thus approximate them well), and they'll model those much
         | more cheaply than traditional solvers will.
         | 
         | We used them for a few things at my last job, and they were
         | definitely worth it. We erred toward smaller (faster) nets with
         | higher errors just to accelerate convergence with a classical
         | solver.
        
           | quanto wrote:
           | I get that PINN is a less expensive approximate solution
           | method. If so, how does it perform superior to many
           | approximate, coarse numerical methods?
        
             | hansvm wrote:
             | 1. Those methods are coarse. The interpolation they provide
             | is worse than what a PINN provides, meaning that
             | equivalently performing PINNs (compard to coarse numerical
             | methods) can easily and cheaply serve as better
             | initializations for your finer numerical methods.
             | 
             | 2. Go back to (1) from my previous message. For some
             | intuition, fiddly solutions take a long time to optimize.
             | Your only options (aside from spending more time and money)
             | are tailoring the initial conditions and the algorithm for
             | your particular problem. You see that a lot in, e.g., 1-3
             | atom quantum chemistry, where a good choice of basis
             | functions is worth several papers. A neural network allows
             | you to automagically bake everything that's hard about your
             | problem into the training step and amortize those hard
             | calculations across many experiments. It's not superior to
             | enough man-centuries of human intuition, but it's dead
             | simple to deploy, and for those sorts of hard problems it
             | definitely beats a single human century of effort. Once you
             | have a neural network output, the problem is well
             | conditioned and suitable for refinement by a classical
             | solver.
             | 
             | For a somewhat concrete example, imagine a problem where
             | the space is largely uninteresting but there are a few
             | tight swirls here and there. Coarse numerical methods can't
             | really do anything with those. Adaptive-precision numerical
             | methods can, but they're slow, and you have to re-run an
             | intensive solving step for every new input. The PINN
             | solution bakes everything that's hard about that into the
             | neural net structure, and it solution will have
             | approximately the right swirls in approximately the right
             | places. If you want to refine them further, the fact that
             | your solver doesn't have to dynamically handle resolution
             | anymore and doesn't have to deal with any major phase
             | shifts makes it much easier to iterate on via the normal
             | classical methods.
        
       | hazrmard wrote:
       | Good read! I am developing PINNs at work and this certainly
       | helped me recall important concepts. This post used deepxde
       | library [2] to compose the PINN. Can anyone comment on how
       | NVIDIA's modulus [2] compares to this? Modulus appears to be much
       | more verbose and poorly documented.
       | 
       | [1]: https://github.com/lululxvi/deepxde [2]:
       | https://github.com/nvidia/modulus
        
       | joshpopelka20 wrote:
       | It's from the future. Must be really good :)
       | 
       | Physics informed neural networks 16 Feb 2026
        
       | SnooSux wrote:
       | Around a month ago there was a PINN post[1] on here and there was
       | a healthy amount of skepticism in the comments. Even in the toxic
       | positivity of LinkedIn, commentors say they're overhyped when a
       | ML "Influencer" posts that one GIF with a MLP and PINN fitting to
       | an oscillator. I would be interested to see what they're actively
       | being used for.
       | 
       | [1] https://news.ycombinator.com/item?id=42769623
        
         | nchagnet wrote:
         | I fully agree with the comments in that post. I started
         | studying them because, well, they sound really cool, but my
         | first impression was definitely that this sounded like a lot of
         | effort (computationally) to solve a single equation. But then I
         | only tested on simple equations where traditional solvers have
         | no difficulties.
         | 
         | In my previous position, we studied the behaviour of black
         | holes with exotic geometries, and we never could make our
         | solvers work with the added time dependence. I would be very
         | curious to see how a PINN would have fared on this (given
         | enough compute time of course).
        
       | lagrange77 wrote:
       | Neural ODEs are also interesting.
        
       | lagrange77 wrote:
       | I can also recommend Steve Brunton's playlist [0] on the topic of
       | physics informed machine learning, as well as the book 'Data-
       | driven Science & Engineering' [1] by him and Nathan Kutz.
       | 
       | [0]
       | https://www.youtube.com/playlist?list=PLMrJAkhIeNNQ0BaKuBKY4...
       | 
       | [1] https://databookuw.com
        
       | quanto wrote:
       | The key intuition is calculating the loss function without
       | actually knowing the exact solution ("labels" in supervised
       | learning parlance). Note that this is not unique to PINN: there
       | are existing numerical methods that do exactly this.
       | 
       | I used to solve PDEs for a living; and my academic background is
       | in numerical solutions to PDEs before going into ML. In my
       | industry and academic experience, PINN is a novel curiosity with
       | perhaps niche applications that I am not as familiar with. Yes, I
       | am aware of works of Bruton, Duvenaud et al (and was even in the
       | same lab group with some of them). I am happy to be corrected and
       | learn if PINN has found a strong application.
       | 
       | A better introduction to this approach and its critique here:
       | https://arxiv.org/pdf/2206.02016
        
       | sundarurfriend wrote:
       | I've never clearly understood the relationship/difference between
       | PINNs and SciML (Scientific Machine Learning). The "how do they
       | work" section here sounds pretty similar to how I've heard SciML
       | described in the past.
       | 
       | From some searcing around, it sounds like maybe SciML is a
       | broader concept with PINNs being a particular implementation of
       | it? Maybe SciML started with PINN related ideas, but has
       | broadened beyond that over time? Would appreciate an explanation
       | from someone who's actively in this field.
        
       ___________________________________________________________________
       (page generated 2025-02-17 23:01 UTC)