Deprecate adjoint in favor of adjugate

[1d7ec08f-60ae-4512-91a6-8324c06eab9f]

Matrix methods named adjoint and _adjoint are renamed adjugate and _adjugate and replacements are added that raise deprecation warnings.

This is part of the program at #10465.

CC: @williamstein @sagetrac-mvngu @sagetrac-kohel @tornaria @orlitzky

Component: linear algebra

Keywords: notation, adjugate, matrices, determinants

Author: Kwankyu Lee

Branch/Commit: bf44252

Reviewer: Darij Grinberg

Issue created by migration from https://trac.sagemath.org/ticket/10501

[1d7ec08f-60ae-4512-91a6-8324c06eab9f]

Attachment: trac_10501-deprecate-adjoint.patch.gz

[1d7ec08f-60ae-4512-91a6-8324c06eab9f]

comment:1

Three files caused doctest errors on a full run with only the necessary changes in sage/matrix. I've made changes in these other places to fix those failures, and the affected files now pass their tests. I'm running the full suite right now.

I've cc'ed folks who I think might be able to double-check that no complications have crept in. If you want to sneak a quick look at the patch, here's a quick guide:

Minh, David: sage/crypto/classical.py, inverse_key() for a Hill Cryptosystem

Gonzalo: sage/quadratic_forms/quadratic_form_ternary_Tornaria.py, adjoints of a form

William: sage/quadratic_forms/quadratic_form.py, adjoint_primitive()

[tornaria]

comment:2

I already stated some objections to this on the mailing list, but I'll repeat:

On deprecating "adjoint" meaning "matrix of cofactors"

1. it's standard terminology and has meant this in sage for long
2. "adjugate" is newer and (IMO) less standard terminology -- in particular it has no obvious translations

On using "adjoint" meaning "conjugate transpose"

3. "conjugate transpose" is easy to say, and it's really what is meant
4. the "adjoint operator" for a matrix seems ill-defined, because a matrix is not an operator but only a representation of an operator in some basis.

Moreover, if there are two colliding usages of the name "adjoint", I would find it more reasonable to keep the usage that is already traditional in Sage.

The usage of "adjoint" is ubiquitous in relation to quadratic forms afaict (and, as John Cremona pointed out, is where the term originates with Gauss on ternary quadratic forms)

---

Reference for "Adjoint of a matrix":

Bourbaki, Elements, book 2, chapter III, section 11, exercise 9:

The adjoint of a square matrix X of order n over A is the matrix X = (det (A'")) of minors of A" of order n — 1.

(Note that the term also shows at the index of terminology of the book)

PS: searching for

"The adjoint of a square matrix" bourbaki

in books.google.com, yields the above passage.

[1d7ec08f-60ae-4512-91a6-8324c06eab9f]

comment:3

Replying to @tornaria:

Hi Gonzalo,

I certainly read your postings to the mailing list carefully and appreciated the points you raised. However, I had not realized you were so opposed to the change.

After some discussion, I asked 'Is there any objection to deprecating the current .adjoint() function (which returns a matrix of cofactors) and renaming it as the "adjugate"?'

It was not meant to be an official vote, but I got +1 replies from Grout, Cremona, Loeffler and Stein. Dima P and Karl Crisman had earlier voiced support. There were no objections stated once I asked the question carefully. So I have been proceeding on the assumption that there was strong support.

I do not believe I changed any of the names of the commands for quadratic forms, though I can see that causing confusion if the adjoint of a matrix becomes the conjugate transpose.

I have written a patch (#10471) with the conjugate_transpose(), which I find a really clumsy command, but workable in the interim. William has suggested a more general adjoint function, which I would need to think about some more, but maybe that does not help with any of your objections (sounds like maybe that is worse in your view).

I have twice now taught a "matrix analysis" course and it seems to me that adjoint gets used regularly (but maybe not consistently) for the conjugate transpose. I am in the middle of making a major push to add significant amount of Sage code to my introductory linear algebra text, which is going very nicely. But I need to also fix my "complex inner product" since I defined it with the conjugation on the "wrong" half. So I would really like to keep Sage, my text, and the word "adjoint" all consistent with each other when I get to that point in a few weeks.

Do you have some suggestions for a way forward?

Thanks, Rob

[orlitzky]

comment:4

+1 from me. I hit this today, and just checked a handful of books:

• Atkinson, An Introduction to Numerical Analysis, 1989. Section 7.1.
• Axler, Linear Algebra Done Right, 1997. Ch. 6.
• Marcus & Minc, Introduction to Linear Algebra, 1988. Section 1.4.
• Meyer, Matrix Analysis and Applied Linear Algebra, 2000. Section 3.2.
• Rudin, Functional Analysis, 1991. Chapter 4.
• Shilov, Linear Algebra, 1977. Section 7.6.

All of which use the "new" meaning. In the interest of fairness, I also found,

• Edwards, Elementary Linear Algebra, 2000. Section 3.4.

Which uses the cofactor definition.

[jhpalmieri]

comment:5

Hmm. Given two completely different uses of the word "adjoint" in this situation, I wonder if the right solution is to avoid it completely (with a deprecation warning for a while). If we use the "new" meaning, there will still be people who type A.adjoint() expecting the old meaning, and vice versa. Something like A.conjugate_transpose() can be found by tab completion; is that good enough? Is A.adjugate() the right name for the other version?

[orlitzky]

comment:6

Replying to @jhpalmieri:

Hmm. Given two completely different uses of the word "adjoint" in this situation, I wonder if the right solution is to avoid it completely (with a deprecation warning for a while). If we use the "new" meaning, there will still be people who type A.adjoint() expecting the old meaning, and vice versa. Something like A.conjugate_transpose() can be found by tab completion; is that good enough? Is A.adjugate() the right name for the other version?

Did someone seriously implement m.conjugate_transpose() as a shortcut for m.conjugate().transpose()? =)

I never thought to look for another method, I just did the operations individually.

From what I understand, the terms "adjoint" and "adjunct" come from higher algebra, most of which is over my head. If that's the case, books written after e.g. category theory became popular will probably gravitate towards the new terminology. Although it does suck to have to deprecate adjoint, give it a new name, and then give something else the old name.

Most of us have access to math departments; maybe we could do a survey of people working in linear algebra? If the result is overwhelming, rename it.

[1d7ec08f-60ae-4512-91a6-8324c06eab9f]

comment:7

Replying to @orlitzky:

Did someone seriously implement m.conjugate_transpose() as a shortcut for m.conjugate().transpose()? =)

Yep, that was me. ;-) But the BDFL suggested it. Required reading:

https://groups.google.com/forum/?fromgroups=#!topic/sage-devel/YjImMWVVwo4

You will see a lot of support for changes. You'll see one conscientious objector. I dropped it. If someone else wants to carry the torch, I'll have their back.

Rob

[kcrisman]

comment:8

Did someone seriously implement m.conjugate_transpose() as a shortcut for m.conjugate().transpose()? =)

It's not as bad as you think, because tab-completion doesn't work on m.conjugate(), though it would be awesome if Sage could magically know that...

Rob, so what does the latest version of your book do?

[1d7ec08f-60ae-4512-91a6-8324c06eab9f]

comment:9

Replying to @kcrisman:

Rob, so what does the latest version of your book do?

conjugate-transpose has always been called "adjoint," in line with my experience teaching numerical linear algebra. I even have my inner product conjugating the correct vector now. ;-)

See: http://linear.ups.edu/html/section-MO.html#subsection-AM

I almost never have need to reference the matrix of cofactors (proposed as adjugate here), but do use it one exercise about building a matrix inverse this way.

See: Exercise PDM.T20 in http://linear.ups.edu/html/section-PDM.html

Rob

[kwankyu]

comment:14

I want to revive this ticket. So here is the needed patch.

One thing not found in Rob's original patch is alias adjoint_classical of adjugate. The alias is used in quadratic form code in Sage.

---

New commits:

0669f64

Deprecate adjoint for adjugate and adjoint_classical

[kwankyu]

Branch: u/klee/10501

[kwankyu]

Author: Kwankyu Lee

[kwankyu]

Commit: 0669f64

[kwankyu]

Description changed:

--- 
+++ 
@@ -1,3 +1,3 @@
 Matrix methods named `adjoint` and `_adjoint` are renamed `adjugate` and `_adjugate` and replacements are added that raise deprecation warnings.

-This is part of the program at #10465.  Patch coming once tests conclude.
+This is part of the program at #10465.  

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Branch pushed to git repo; I updated commit sha1. New commits:

a61f618	Merge branch 'develop'
1a2f6a0	properly deprecate adjoint
43ab6df	pyflakes check and fix exception messages style

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Changed commit from 0669f64 to 43ab6df

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Changed commit from 43ab6df to 08eee09

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Branch pushed to git repo; I updated commit sha1. New commits:

08eee09

import deprecated_function_alias on global level to avoid crash

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Changed commit from 08eee09 to d1bbfb1

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Branch pushed to git repo; I updated commit sha1. New commits:

d1bbfb1

change more adjoint to adjugate

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Branch pushed to git repo; I updated commit sha1. New commits:

2f15975

fix neglected doctest failures

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Changed commit from d1bbfb1 to 2f15975

[darijgr]

comment:21

I wholeheartedly agree with "adjugate". When I see "adjoint", I look up the definition. When I see "adjugate", I immediately know what is meant.

Pushed a little commit to improve the documentation. IMHO, this is an easy ticket to review: just run all doctests. If they work, then it's fine.

---

New commits:

64b9491	Merge branch 'u/klee/10501' of git://trac.sagemath.org/sage into adj
b4994c4	actually define the adjugate in the doc

[darijgr]

Changed branch from u/klee/10501 to public/ticket/10501

[darijgr]

Changed commit from 2f15975 to b4994c4

[darijgr]

Changed keywords from none to notation, adjugate, matrices, determinants

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

356285d

Deprecate adjoint for adjugate and adjoint_classical

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Changed commit from b4994c4 to 356285d

[kwankyu]

comment:23

Squashed and rebased to sage 8.6. In addition, refined some docstrings and comments.

[kcrisman]

comment:24

If klee and darij are both agreed that each other's respective contributions are sufficient, then you can jointly set to positive review (sometimes Authors are 1, 2 and Reviewers are 2, 1).

I do wonder whether some of the py3 things like

-                raise TypeError("Oops!  The matrix must have " + str(n) + " rows. =(")
+                raise TypeError("the matrix must have {} rows".format(n))

are necessary on this ticket, since it makes it a little more of a patch bomb.

Also, don't we typically test deprecations (and then remove when deprecation is done)?

[kwankyu]

comment:25

Replying to @kcrisman:

If klee and darij are both agreed that each other's respective contributions are sufficient, then you can jointly set to positive review (sometimes Authors are 1, 2 and Reviewers are 2, 1).

Positive review on his part of the code. It is up to him to put his name to the Author field.

I do wonder whether some of the py3 things like

-                raise TypeError("Oops!  The matrix must have " + str(n) + " rows. =(")
+                raise TypeError("the matrix must have {} rows".format(n))

are necessary on this ticket, since it makes it a little more of a patch bomb.

Not necessary. But I think we don't need to freak away from making small improvements unrelated with the main issue of the ticket. Do we?

Or do you regard the small changes as no improvements?

Also, don't we typically test deprecations (and then remove when deprecation is done)?

I didn't know. I don't know.

[kcrisman]

comment:26

Or do you regard the small changes as no improvements?

No, it's just that it causes more opportunities for clashes with other tickets.

Also, don't we typically test deprecations (and then remove when deprecation is done)?

I didn't know. I don't know.

For one of many examples, see (at least right now correct link) this spot

            sage: x, y, z = var('x, y, z')
            sage: S = CoordinatePatch((x, y, z)); S
            doctest:...: DeprecationWarning: Use Manifold instead.
            See http://trac.sagemath.org/24444 for details.
            Open subset of R^3 with coordinates x, y, z

I guess this should be done for positive review.

[darijgr]

comment:27

Sorry, I don't have time for this :/

I could reread the rebased branch once the stress from the semester start has subsided, but I'm generally extremely short on time until September or so(?). Sorry.

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Branch pushed to git repo; I updated commit sha1. New commits:

bf44252

Add a test for deprecation

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Changed commit from 356285d to bf44252

[kwankyu]

comment:29

Anything else to do?

[darijgr]

comment:30

The new version looks good to me. If complete doctests pass (anyone please check), please make this a positive_review.

(I should not be listed as author; my changes were trivial.)

[darijgr]

Reviewer: Darij Grinberg

[vbraun]

Changed branch from public/ticket/10501 to bf44252