Initialize UCJ from CISD by davidomanovic · Pull Request #595 · qiskit-community/ffsim

davidomanovic · 2026-03-16T10:38:43Z

Initialization of UCJ from CISD as described in https://arxiv.org/pdf/2511.22476

Added as a method from.cisd() in UCJOpSpinBalanced, and any necessary tests. Closes issue #592.

kevinsung · 2026-03-17T12:33:05Z

+    def from_cisd(
+        cisd,
+        *,
+        civec: np.ndarray | None = None,


Instead of accepting cisd, only accept civec, and rename it to cisd_vec. Also, rename the method to from_cisd_vec.

kevinsung · 2026-03-17T12:34:17Z

+                "or use a converged PySCF CISD object with a `ci` attribute."
+            )
+
+        c0, c1, c2 = cisd.cisdvec_to_amplitudes(civec, copy=False)


Use pyscf.ci.cisd.cisdvec_to_amplitudes instead, since you won't have the cisd object anymore.

kevinsung · 2026-03-17T17:51:06Z

+                "UCJOpSpinBalanced.from_cisd requires amplitudes from a restricted "
+                "closed-shell CISD calculation."
+            )
+        if np.isclose(c0, 0.0):


Add a keyword-only argument c0_tol: float = 1e-8 to use for the absolute tolerance here. Also, this is a nit but prefer math.isclose to np.isclose when comparing numbers because it avoids numpy function dispatch overhead.

kevinsung · 2026-03-17T19:53:05Z

+        ]
+        | None = None,
+        tol: float = 1e-8,
+        c0_tol: float = 1e-8,


Let's put c0_tol after nocc and before n_reps.

kevinsung · 2026-03-17T19:53:53Z

+        :meth:`from_t_amplitudes`.
+
+        Args:
+            civec: CISD coefficient vector.


Need to update this to cisd_vec and also add document norb, nocc, and c0_tol.

kevinsung · 2026-03-17T20:10:57Z

+        if cisd_vec is None:
+            raise ValueError("CISD coefficient vector `cisd_vec` cannot be None.")
+


This check is unnecessary.

kevinsung · 2026-03-17T20:11:05Z

+            cisd_vec, nmo, nocc, copy=False
+        )
+
+        if math.isclose(c0, 0.0, rel_tol=0.0, abs_tol=c0_tol):


Since we're comparing to zero, it's actually unnecessary to specify rel_tol.

kevinsung · 2026-03-17T20:14:46Z

+                f"CISD reference coefficient c0={c0} is too close to zero "
+                f"(abs_tol={c0_tol})."


Suggested change

f"CISD reference coefficient c0={c0} is too close to zero "

f"(abs_tol={c0_tol})."

f"CISD reference coefficient c0={c0} is smaller than the specified threshold, c0_tol={c0_tol}."

kevinsung · 2026-03-17T20:15:23Z

+    t2 = c2 / c0 - 0.5 * np.einsum("ia,jb->ijab", t1, t1)
+
+    operator = ffsim.UCJOpSpinBalanced.from_cisd_vec(
+        cisd.ci, nmo=cisd.nmo, nocc=cisd.nocc


You should also test that cisd.ci was not modified, since you chose to pass copy=False in cisdvec_to_amplitudes.

kevinsung

Thank you! Just a few more minor changes.

davidomanovic · 2026-03-19T10:18:37Z

The current CI failures appear to be unrelated to this change.

The failing jobs are invoking tox with environment names like py3.14, but tox.ini defines the test environments as py310, py311, py312, py313, and py314:

env_list =
py{310,311,312,313,314}

For example, one failing job reports:
tox run -e py3.14
and tox returns:
provided environments not found in configuration file: py3.14 - did you mean py314?

So at least these failures are due to a workflow/tox environment naming mismatch @kevinsung

kevinsung · 2026-03-19T15:08:19Z

The current CI failures appear to be unrelated to this change.

The failing jobs are invoking tox with environment names like py3.14, but tox.ini defines the test environments as py310, py311, py312, py313, and py314:

env_list = py{310,311,312,313,314}

For example, one failing job reports: tox run -e py3.14 and tox returns: provided environments not found in configuration file: py3.14 - did you mean py314?

So at least these failures are due to a workflow/tox environment naming mismatch @kevinsung

CI was broken because of a change in tox 4.50.1. I've fixed this in #603 , please merge main.

davidomanovic · 2026-03-19T16:29:22Z

The current CI failures appear to be unrelated to this change.

The failing jobs are invoking tox with environment names like py3.14, but tox.ini defines the test environments as py310, py311, py312, py313, and py314:

env_list = py{310,311,312,313,314}

For example, one failing job reports: tox run -e py3.14 and tox returns: provided environments not found in configuration file: py3.14 - did you mean py314?

So at least these failures are due to a workflow/tox environment naming mismatch @kevinsung

CI was broken because of a change in tox 4.50.1. I've fixed this in #603 , please merge main.

Hmmm still something wrong with tox.

kevinsung · 2026-03-19T16:31:02Z

Just need to fix the sphinx warnings now. Unfortunately, I've never figured out a good way to debug those, besides tracking down the offending line through binary search.

kevinsung · 2026-03-28T16:22:27Z

+        *,
+        norb: int,
+        nocc: int,
+        c0_threshold: float = 1e-8,


Suggested change

c0_threshold: float = 1e-8,

c0_threshold: float = 1e-12,

Let's make this smaller actually.

kevinsung · 2026-03-28T16:23:49Z

+            cisd_vec: The CISD vector. This is a one-dimensional array storing the
+            reference coefficient :math:`c_0` in the first entry, followed by the
+            singles and then doubles coefficients.
+            norb: The number of spatial orbitals.
+            nocc: The number of occupied orbitals.
+            c0_threshold: Absolute value threshold for the reference coefficient.
+            An error is raised if the absolute value of the reference coefficient is
+            smaller than this threshold.
+            n_reps: The number of ansatz repetitions.


The indentation here doesn't match the other arguments, where lines after the first one are indented. Is this related to the Sphinx issues you were having?

kevinsung · 2026-03-31T13:57:09Z

Hmm the CI failure seems unrelated, I will look into it...

kevinsung

Thank you!

kevinsung requested changes Mar 17, 2026

View reviewed changes

kevinsung requested changes Mar 18, 2026

View reviewed changes

davidomanovic force-pushed the feature/ucj_from_cisd branch 2 times, most recently from c4d1e16 to 1e3131f Compare March 19, 2026 10:13

davidomanovic requested a review from kevinsung March 19, 2026 10:19

Initialization of UCJ from CISD vector

1e3131f

Merge branch 'main' into feature/ucj_from_cisd

4b7f0c1

davidomanovic added 3 commits March 20, 2026 12:54

Merge branch 'main' into feature/ucj_from_cisd

fa55996

Merge branch 'main' into feature/ucj_from_cisd

88583ae

Merge branch 'main' into feature/ucj_from_cisd

f48bb25

kevinsung reviewed Mar 28, 2026

View reviewed changes

refactor: indentation and c0_threshold=1e-12 by default

06730ee

kevinsung approved these changes Mar 31, 2026

View reviewed changes

kevinsung merged commit 250d579 into qiskit-community:main Mar 31, 2026
27 of 28 checks passed

kevinsung mentioned this pull request Mar 31, 2026

Add .from_cisd() method for initializing UCJ from CISD #592

Closed

		if cisd_vec is None:
		raise ValueError("CISD coefficient vector `cisd_vec` cannot be None.")

		f"CISD reference coefficient c0={c0} is too close to zero "
		f"(abs_tol={c0_tol})."

Conversation

davidomanovic commented Mar 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kevinsung left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

davidomanovic commented Mar 19, 2026

Uh oh!

kevinsung commented Mar 19, 2026

Uh oh!

davidomanovic commented Mar 19, 2026

Uh oh!

kevinsung commented Mar 19, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kevinsung commented Mar 31, 2026

Uh oh!

kevinsung left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

davidomanovic commented Mar 16, 2026 •

edited

Loading