Jay Zhuang

Came across the same issue. Note that [`pyamg.aggregation.smoothed_aggregation_solver`](https://pyamg.readthedocs.io/en/latest/_modules/pyamg/aggregation/aggregation.html#smoothed_aggregation_solver) allows near-nullspace candidates `B` of shape `(N, K)`. As long as the SA setup logic follows PyAMG, passing a 2-D matrix `B`...

Actually `smoothed_aggregation` also diverges, while `pyamg.smoothed_aggregation_solver` with exactly the same setting works fine. ```julia ml_sa = AMG.smoothed_aggregation(A) AMG._solve(ml_sa, b, verbose=true, maxiter=20) ``` gives: ``` Norm of residual at iteration 1...

It is actually the smoother's fault. Swapping out the built-in smoother by `IterativeSolvers: gauss_seidel!` solves the problem. ```julia using MatrixMarket import AlgebraicMultigrid as AMG import IterativeSolvers: gauss_seidel! A = MatrixMarket.mmread("poisson3Db.mtx")...

Hi @pghysels I'd like to better understand the steps to develop symmetric LDLT from the existing general LU. Thanks! From the book [Multifrontal Methods Parallelism, Memory Usage and Numerical Aspects](https://graal.ens-lyon.fr/~jylexcel/hdr.pdf)...

Thanks for the explanation! > Keeping the low rank algorithms symmetric will be more complicated I'm afraid. Could you elaborate more on this point? From the [MUMPS BLR paper](https://epubs.siam.org/doi/abs/10.1137/120903476), the...

> then replace the `_trsm` calls (I'm not sure by what). I think `trsm` is the only LAPACK triangular solve routine -- the triangular factors should not be aware of...

I quickly browsed through NumPyro source code to locate JAX-heavy code. For example, the most frequently used APIs include: - `numpyro.sample` - `numpyro.infer.{MCMC, NUTS}` - `numpyro.distributions.{Normal, Exponential}` 1. `sample` is...

> none of which are fully compatible with NEP-47 yet This is correct, at this moment. My question is, once those frameworks get fully-compatible with NEP-47, what would be the...

> especially higher-order forward- and reverse-mode AD and a large library of probability distributions with [Tensorflow-style shape semantics and reparameterized samplers](https://arxiv.org/abs/1711.10604) Thanks, this is very useful information. I will read...

Thanks @cheshmi ! Let me try manually reversing the iteration and see how it works.