You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Reported from an AI-assisted test of ALPS 3.0.0 on Ubuntu 24.04, then re-checked against the v3.0.0 sources and by targeted reruns. No human has independently reviewed it yet — please
verify before acting on it, and corrections are very welcome.
We ran the four quickstarts under https://alps.comp-phys.org/start/ with the pyalps==3.0.0 wheel,
keeping the pages' simulation parameters and adding diagnostic prints (and savefig where we wanted
a file), and we built v3.0.0 from source following /install/source/ plus the interpreter workaround
reported in #145.
How much each claim is checked:
Independently verified: selected Ising |m| values against Onsager; DMRG against TeNPy; ED for
L ≤ 12 against a separate Lanczos code; the analytic high-temperature limit of the CT-HYB quantity.
Not independently verified: agreement of the CT-HYB curve with the page's own Kondo.png (that is
a consistency check against the same solver, not a validation), the 162/162 test result (one run,
one environment), and the HDF5 version boundary below.
Checked once, not repeated during the follow-up: the full test suite (162/162), all ED sizes, and all CT-HYB temperatures.
Everything below is page text that does not match what the software produces. Website pointers are
paths in ALPSim/alpsim.github.io at 712b572 (2026-09-18), so the wording is easy to find. If it
helps triage, I am happy to split this into per-page issues.
Quickstart pages
content/en/start/dmrg.md — three statements do not match the run.
[wrong number] "The ground-state energy should be close to E0 ≈ −14.3 J for this 32-site open chain." The code gives E0 = −13.99731562: ALPS −13.997315618223, TeNPy (χ = 300) −13.997315618224. Against −14.3 a correct result looks 2 % off.
[wrong number] Truncation errors "around 1e−8–1e−7 … are typical". Our final Truncation error was 4.5e−14, while the 337-point iteration history peaked near 1.6e−5. A final value alone does not settle a claim about "typical" errors, so the page should say which of the two it means (final value, per-sweep maximum, or something else) — 1e−7 does not describe either of ours.
[wording] loadEigenstateMeasurements is described as returning "energy, magnetization, correlation functions, and any other quantities ALPS computed". With the page's parameters and no MEASURE_* setting, only Energy and Truncation error are returned. The example above it prints 'Energy', so the prose over-states it.
content/en/start/ed.md — the stated extrapolation is not what the page's own code produces.
[wrong number] "Result" claims the fit gives "Δ ≈ 0.411 J — in good agreement with … 0.41048". Running the page's code (L = 4…16, fit on L ≥ 8) gives Δ = 0.4191 with an effective ξ = 4.04. scipy.optimize.curve_fit reaches the same optimum from several initial guesses, the page's own ED_spin.png shows the same ≈ 0.42 intercept, and the raw spectra are right (L ≤ 12 matches an independent Lanczos code to 1e−10). Either the number should be ≈ 0.419 as stated, or the recipe needs larger L or subleading corrections — the effective ξ = 4.04 against the physical ξ ≈ 6 quoted higher up the page suggests one exponential is not enough at these sizes.
[minor] plt.ylabel('Triplet gap $\Delta/J$') raises SyntaxWarning: invalid escape sequence '\D' on Python ≥ 3.12 (reproduced). A raw string fixes it.
[missing] No runtime hint. The script took ~35 min on one core, ~32 min of it in the L = 16 sectors.
content/en/start/qmc.md — the prose contradicts the page's own figure in two places.
[wrong physics] "At high temperature it approaches 1 (free spin, S = 1/2)". The data go to 0.5 for T ≫ U (D = 2), which is the uncorrelated-electron value; a value near 1 needs T_K ≪ T ≪ U, a regime U = 2 does not reach. The U = 2 curve peaks at ≈ 0.62 near T ≈ 1 and is then screened.
[wrong physics] "For U = 0, the effective moment is approximately constant." It falls from 0.50 at T = 100 to 0.04 at T = 0.05, which is Pauli behaviour.
[naming] pyalps.cthyb here is the segment hybridization solver in applications/dmft/qmc/hybridization, not ALPSCore/CT-HYB. Worth saying so on the page, since the name is shared.
[headless] All quickstarts end with plt.show(). On a headless machine (SSH, HPC) this only prints FigureCanvasAgg is non-interactive, and thus cannot be shown and no figure appears. One line pointing at plt.savefig(...) would cover it.
Installation pages
[stale requirement] CMake minimum.install/source.md lists CMake ≥ 3.18 (twice), but CMakeLists.txt:20 requires 3.22. README-py.md already states 3.22.
[unclear requirement] HDF5 minimum.install/source.md says ≥ 1.10.0, while codes/install/alps-install-agent.md says ≥ 1.14 and states that 1.10 fails 8 tests. With HDF5 1.10.10 (Ubuntu 24.04 libhdf5-dev) the v3.0.0 build passed 162/162 tests. The teardown check behind that claim ("Not all resources closed in file", src/alps/hdf5/archive.cpp:332) is still present in the source, and the guide's report is about distro HDF5 1.10.5. We have not isolated whether the difference is the patch level, distribution patches, build options, or something else — naming the failing version and configuration would be more useful than excluding all of 1.10.
[unclear requirement] System Boost warning.install/source.md warns that "if a system Boost is present CMake will find it and the build fails". On Ubuntu 24.04 with libboost1.83-dev installed, the default source build used only the Boost 1.87 sources — none of the 644 object dependency files referenced /usr/include/boost — and the build succeeded. The warning could name the configurations that actually fail.
[missing instruction] Binary install (install/_index.md, install/binary.md). The overview says the binary package needs "no special system privileges", and binary.md gives pip install pyalps with no environment setup. On the tested Ubuntu 24.04 machine the system /usr/bin/python3 is EXTERNALLY-MANAGED (PEP 668) and has neither pip nor ensurepip, and python3 -m venv then fails with "ensurepip is not available … apt install python3.12-venv … You may need to use sudo". A short venv, uv or conda route would get a first-time user unstuck.
The scripts, logs and independent reference calculations can be shared on request. Related bug reports
from the same test: #145, #146, #147. The same test also found that the CI Boost
matrix always built with Boost 1.87; that appears to be fixed already in open PR #107 (commit 08ad15d), so it is not filed separately. Several install-page items will change anyway if #107 lands.
Note
Reported from an AI-assisted test of ALPS 3.0.0 on Ubuntu 24.04, then re-checked against the
v3.0.0sources and by targeted reruns. No human has independently reviewed it yet — pleaseverify before acting on it, and corrections are very welcome.
Category
ALPS Documentation (errors, missing content, clarity)
Describe your question or issue
We ran the four quickstarts under https://alps.comp-phys.org/start/ with the
pyalps==3.0.0wheel,keeping the pages' simulation parameters and adding diagnostic prints (and
savefigwhere we wanteda file), and we built
v3.0.0from source following/install/source/plus the interpreter workaroundreported in #145.
How much each claim is checked:
L ≤ 12 against a separate Lanczos code; the analytic high-temperature limit of the CT-HYB quantity.
Kondo.png(that isa consistency check against the same solver, not a validation), the 162/162 test result (one run,
one environment), and the HDF5 version boundary below.
Everything below is page text that does not match what the software produces. Website pointers are
paths in
ALPSim/alpsim.github.ioat712b572(2026-09-18), so the wording is easy to find. If ithelps triage, I am happy to split this into per-page issues.
Quickstart pages
content/en/start/dmrg.md— three statements do not match the run.Truncation errorwas 4.5e−14, while the 337-point iteration history peaked near 1.6e−5. A final value alone does not settle a claim about "typical" errors, so the page should say which of the two it means (final value, per-sweep maximum, or something else) — 1e−7 does not describe either of ours.loadEigenstateMeasurementsis described as returning "energy, magnetization, correlation functions, and any other quantities ALPS computed". With the page's parameters and noMEASURE_*setting, onlyEnergyandTruncation errorare returned. The example above it prints'Energy', so the prose over-states it.content/en/start/ed.md— the stated extrapolation is not what the page's own code produces.scipy.optimize.curve_fitreaches the same optimum from several initial guesses, the page's ownED_spin.pngshows the same ≈ 0.42 intercept, and the raw spectra are right (L ≤ 12 matches an independent Lanczos code to 1e−10). Either the number should be ≈ 0.419 as stated, or the recipe needs larger L or subleading corrections — the effective ξ = 4.04 against the physical ξ ≈ 6 quoted higher up the page suggests one exponential is not enough at these sizes.plt.ylabel('Triplet gap $\Delta/J$')raisesSyntaxWarning: invalid escape sequence '\D'on Python ≥ 3.12 (reproduced). A raw string fixes it.content/en/start/qmc.md— the prose contradicts the page's own figure in two places.pyalps.cthybhere is the segment hybridization solver inapplications/dmft/qmc/hybridization, not ALPSCore/CT-HYB. Worth saying so on the page, since the name is shared.plt.show(). On a headless machine (SSH, HPC) this only printsFigureCanvasAgg is non-interactive, and thus cannot be shownand no figure appears. One line pointing atplt.savefig(...)would cover it.Installation pages
[stale requirement] CMake minimum.
install/source.mdlists CMake ≥ 3.18 (twice), butCMakeLists.txt:20requires 3.22.README-py.mdalready states 3.22.[unclear requirement] HDF5 minimum.
install/source.mdsays ≥ 1.10.0, whilecodes/install/alps-install-agent.mdsays ≥ 1.14 and states that 1.10 fails 8 tests. With HDF5 1.10.10 (Ubuntu 24.04libhdf5-dev) thev3.0.0build passed 162/162 tests. The teardown check behind that claim ("Not all resources closed in file",src/alps/hdf5/archive.cpp:332) is still present in the source, and the guide's report is about distro HDF5 1.10.5. We have not isolated whether the difference is the patch level, distribution patches, build options, or something else — naming the failing version and configuration would be more useful than excluding all of 1.10.[unclear requirement] System Boost warning.
install/source.mdwarns that "if a system Boost is present CMake will find it and the build fails". On Ubuntu 24.04 withlibboost1.83-devinstalled, the default source build used only the Boost 1.87 sources — none of the 644 object dependency files referenced/usr/include/boost— and the build succeeded. The warning could name the configurations that actually fail.[broken instruction] Python pinning.
-DPython3_EXECUTABLE, recommended on this page and inno_root_guide.pdf, has no effect: see [Bug]: -DPython3_EXECUTABLE (recommended in the install docs) is silently ignored; configure picks another Python and can fail #145.[stale instruction]
alpsvars.sh/PYTHONPATH. The manual workaround on this page is needed because of [Bug]: alpsvars.sh/.csh set PYTHONPATH=<prefix>/lib, but pyalps is installed in <prefix>/lib/pythonX.Y/site-packages #147, so it should stay until the template is fixed.[missing instruction] Binary install (
install/_index.md,install/binary.md). The overview says the binary package needs "no special system privileges", andbinary.mdgivespip install pyalpswith no environment setup. On the tested Ubuntu 24.04 machine the system/usr/bin/python3isEXTERNALLY-MANAGED(PEP 668) and has neitherpipnorensurepip, andpython3 -m venvthen fails with "ensurepip is not available … apt install python3.12-venv … You may need to use sudo". A short venv,uvor conda route would get a first-time user unstuck.[stale instruction] Spack (
install/spack.md).spack install alpscurrently installs 2.3.4-beta.2, which the builtin recipe markspreferred=True; 3.0.0 exists only in the open alps: add 3.0.0 and system Boost support spack/spack-packages#6489. The page could say which version it installs until that is merged (see also [ALPS]: Issues with Spack installation #117).[stale content, typos] Minor text fixes:
codes/install/alps-install-agent.md:1README-py.md(PyPI description)ALPSim/legacyCI badge and links to the 2.3.3README.txtREADME-py.md:9install/binary.md:11Environment (if installation or code‑related)
v3.0.0(commit 1950cc6) built from sourceAdditional context
The scripts, logs and independent reference calculations can be shared on request. Related bug reports
from the same test: #145, #146, #147. The same test also found that the CI Boost
matrix always built with Boost 1.87; that appears to be fixed already in open PR #107 (commit
08ad15d), so it is not filed separately. Several install-page items will change anyway if #107 lands.