61bb119 | Manuel Kauers | 28 February 2017, 13:34:58 UTC | ready for packaging | 28 February 2017, 13:34:58 UTC |
2a947f9 | Manuel Kauers | 24 February 2017, 11:38:57 UTC | one more doctest updated | 24 February 2017, 11:38:57 UTC |
ecde797 | Manuel Kauers | 24 February 2017, 11:31:49 UTC | Merge branch 'master' of http://marc.mezzarobba.net/code/ore_algebra-analytic | 24 February 2017, 11:31:49 UTC |
c23b16f | Manuel Kauers | 24 February 2017, 11:29:15 UTC | * doctests updated for 7.5.1 * multivariate case: intersection, symmetric product, composition, associates, elimination, creative telescoping * various bugfixes throughout the code | 24 February 2017, 11:29:15 UTC |
672cbec | Marc Mezzarobba | 07 February 2017, 10:28:35 UTC | teach numerical_solution() to compute derivatives of solutions | 07 February 2017, 10:28:35 UTC |
49dcbcb | Marc Mezzarobba | 06 February 2017, 19:38:20 UTC | analytic: improve+test handling of ops in K(x)[Dx] | 06 February 2017, 19:41:36 UTC |
8790e62 | Marc Mezzarobba | 06 February 2017, 13:56:37 UTC | analytic continuation: accept more types of initial values | 06 February 2017, 19:41:36 UTC |
d5a961c | Marc Mezzarobba | 06 February 2017, 13:19:01 UTC | binsplit: add a test | 06 February 2017, 19:41:36 UTC |
3e23fed | Marc Mezzarobba | 06 February 2017, 10:51:19 UTC | local_basis_monomials: clarify that (1, x, ...) =/=> analytic | 06 February 2017, 13:19:17 UTC |
553cf68 | Marc Mezzarobba | 20 January 2017, 15:54:43 UTC | Merge analytic -> sage7 | 20 January 2017, 18:08:13 UTC |
a71c861 | Marc Mezzarobba | 18 January 2017, 18:03:04 UTC | numerical_solution: fix regression for deqs of order 0 from 402cde9edae4653ed60706c2a48a4895e50e5671 | 18 January 2017, 18:03:04 UTC |
44a41b7 | Marc Mezzarobba | 13 January 2017, 14:06:02 UTC | EvaluationPoint.__repr__() | 15 January 2017, 15:13:59 UTC |
ff28bee | Marc Mezzarobba | 12 January 2017, 13:37:09 UTC | polapprox: robustify | 15 January 2017, 15:13:59 UTC |
dbf4109 | Marc Mezzarobba | 09 January 2017, 15:03:10 UTC | binsplit: fix doctests | 09 January 2017, 15:12:03 UTC |
00cacdc | Marc Mezzarobba | 09 January 2017, 14:52:52 UTC | naive_sum: fix speed regression from b6e8bfa4704c09156bffd916436797c71e06e00c | 09 January 2017, 15:12:03 UTC |
850f17d | Marc Mezzarobba | 07 January 2017, 15:29:45 UTC | binsplit: limit overhead of pol/mat ops over number fields | 09 January 2017, 15:12:03 UTC |
1b81171 | Marc Mezzarobba | 06 January 2017, 13:26:41 UTC | binsplit: get rid of _eval_template even a naïve version now runs faster | 06 January 2017, 13:38:48 UTC |
5c8280d | Marc Mezzarobba | 06 January 2017, 12:01:44 UTC | binplit: comment out costly(?!) assertions | 06 January 2017, 12:01:44 UTC |
33a14d7 | Marc Mezzarobba | 06 January 2017, 12:01:06 UTC | binsplit: minor rewrite to ease profiling | 06 January 2017, 12:01:06 UTC |
268a017 | Marc Mezzarobba | 06 January 2017, 11:42:56 UTC | binsplit: mark long test as such | 06 January 2017, 11:43:20 UTC |
47bd785 | Marc Mezzarobba | 06 January 2017, 11:42:11 UTC | remove outdated comment | 06 January 2017, 11:42:11 UTC |
4ed3a9f | Marc Mezzarobba | 05 January 2017, 16:17:54 UTC | get rid of superseded features from oaa.ui | 05 January 2017, 16:38:51 UTC |
402cde9 | Marc Mezzarobba | 05 January 2017, 16:09:46 UTC | teach numerical_solution() to compute limits at regsing points | 05 January 2017, 16:38:51 UTC |
f841635 | Marc Mezzarobba | 05 January 2017, 16:08:44 UTC | bounds: fix bug(s) with imaginary local exponents | 05 January 2017, 16:08:44 UTC |
51c1381 | Marc Mezzarobba | 05 January 2017, 14:13:05 UTC | minor change to the ordering of local solutions | 05 January 2017, 14:55:44 UTC |
d789a81 | Marc Mezzarobba | 05 January 2017, 10:40:14 UTC | analytic: port ex/tests away from .ui except for those from .ui itself, and a few examples that involve evaluations at regular singular points (supported by eval_diffeq(), but not by numerical_solution()) | 05 January 2017, 10:51:09 UTC |
e439d10 | Marc Mezzarobba | 05 January 2017, 10:13:18 UTC | move most examples from anaytic.ui to .__init__ and port them to the "new" interface | 05 January 2017, 10:15:22 UTC |
f40ba8c | Marc Mezzarobba | 04 January 2017, 15:06:41 UTC | binsplit: fix(?) bugs when ordrec < orddeq For now, we artificially increase the order, because our current strategy of using vectors of coeffs of the form [u(n-s'), ..., u(n+r-1)] to make the initial values as simple as possible would break otherwise. But note that when s' < 0, the functions 1, x, ..., x^(-s'+1) are local solutions. In the longer term, it would probably be better to always normalize the recurrence to Sn-valuation zero, and add the columns corresponding to 1, x, ... in partial_sums(). | 04 January 2017, 16:17:44 UTC |
45d7271 | Marc Mezzarobba | 04 January 2017, 14:40:19 UTC | binsplit: reorganize code for partial sums | 04 January 2017, 16:03:44 UTC |
9aab946 | Marc Mezzarobba | 04 January 2017, 14:12:54 UTC | rename orddiff to orddelta | 04 January 2017, 14:12:54 UTC |
e198349 | Marc Mezzarobba | 04 January 2017, 11:10:58 UTC | update comment on how to run the tests | 04 January 2017, 11:20:25 UTC |
68b66ee | Marc Mezzarobba | 04 January 2017, 10:59:26 UTC | add variants of a few tests using binary splitting | 04 January 2017, 11:20:25 UTC |
50e620c | Marc Mezzarobba | 04 January 2017, 11:07:33 UTC | binsplit: make logging levels consistent w/ ordinary case | 04 January 2017, 11:20:25 UTC |
fbf34c3 | Marc Mezzarobba | 04 January 2017, 10:29:46 UTC | make it a easier to force a particular algorithm | 04 January 2017, 10:29:46 UTC |
f322c2d | Marc Mezzarobba | 22 December 2016, 12:38:43 UTC | generalized_series: simplify element_class and coercion mechanics | 22 December 2016, 12:43:28 UTC |
9fab0e5 | Marc Mezzarobba | 22 December 2016, 12:07:05 UTC | GeneralizedSeriesMonoid: use UniqueRepresentation... ...instead of reimplementing a cache | 22 December 2016, 12:09:07 UTC |
370d2cb | Marc Mezzarobba | 22 December 2016, 11:18:16 UTC | improve/fix doctests - add imports (this is necessary to run the doctester with --force-lib, and relative imports seem not to work without --force-lib); - fix whitespace; - robustify some tests by sorting the output. | 22 December 2016, 11:45:24 UTC |
6fa30b1 | Marc Mezzarobba | 22 December 2016, 09:10:13 UTC | analytic: fix imports in doctests so that tests can be run with --force-lib | 22 December 2016, 11:45:24 UTC |
8d31bfc | Marc Mezzarobba | 22 December 2016, 11:41:58 UTC | analytic: clean up imports - switch to relative imports - fix a few "import *" without __all__ | 22 December 2016, 11:45:24 UTC |
a09dc87 | Marc Mezzarobba | 22 December 2016, 11:14:00 UTC | misc fixes for sage 7+ | 22 December 2016, 11:45:24 UTC |
86d2bb6 | Marc Mezzarobba | 22 December 2016, 07:36:32 UTC | clean up imports | 22 December 2016, 11:45:24 UTC |
b6e8bfa | Marc Mezzarobba | 22 December 2016, 11:07:21 UTC | naive_sum: fix regressions for recs of order zero | 22 December 2016, 11:45:24 UTC |
2842905 | Marc Mezzarobba | 22 December 2016, 10:31:05 UTC | path: minor fix | 22 December 2016, 11:45:24 UTC |
f21ed8d | Marc Mezzarobba | 14 December 2016, 15:03:53 UTC | path: fix usage of comparisons of objects of different types | 22 December 2016, 06:33:22 UTC |
fadb9c7 | Marc Mezzarobba | 20 November 2016, 15:44:15 UTC | polapprox: combine_radii | 20 November 2016, 15:44:15 UTC |
080f9d9 | Marc Mezzarobba | 20 November 2016, 11:40:51 UTC | clarify acceptable types of Point.value | 20 November 2016, 11:40:51 UTC |
19cf571 | Marc Mezzarobba | 18 November 2016, 17:11:24 UTC | naive_sum: stopping criterion -> more accurate pol approx | 18 November 2016, 17:12:24 UTC |
1d519d8 | Marc Mezzarobba | 18 November 2016, 15:56:16 UTC | polapprox: quick fix for bug with real series on complex disks polapprox.on_disk() used to return polynomials with real coefficients and hence real tail intervals in some cases. Evaluating these polynomials on complex points without special precautions would lead to incorrect enclosures. | 18 November 2016, 16:02:13 UTC |
14aefda | Marc Mezzarobba | 18 November 2016, 13:56:30 UTC | don't reinvent utilities.is_real_parent() | 18 November 2016, 16:02:04 UTC |
f39dfb0 | Marc Mezzarobba | 10 November 2016, 15:17:23 UTC | path: support translating variable by exact ball Support "translating" an operator to a dyadic real or complex number represented by an exact ball. Probably too ad hoc, but will do for now. | 10 November 2016, 15:17:23 UTC |
7cc3fc1 | Marc Mezzarobba | 10 November 2016, 15:01:40 UTC | path: support a few more point types on input | 10 November 2016, 15:01:40 UTC |
63ff726 | Marc Mezzarobba | 11 October 2016, 16:36:55 UTC | fix bug in Chebyshev economization The economization process used to "forget" the higher-degree terms when it found a way to remove a monomial other than that of highest degree. | 07 November 2016, 10:13:50 UTC |
1ffdc84 | Marc Mezzarobba | 08 September 2016, 14:27:54 UTC | timings | 21 October 2016, 16:06:17 UTC |
36d2080 | Marc Mezzarobba | 13 September 2016, 09:14:35 UTC | naive_sum: improve handling of working precision - Move PrecisionError and related logic to naive_sum.py. - Make it possible for the summation function to suggest a new working precision. No nontrivial use for now, as I couldn't come up with a good heuristic. In particular, multiplying the working precision by something like log(target_accuracy)/log(current_error_estimate) doesn't seem to work well. - Use a much larger initial precision. I have no really satisfying explanation for this formula, but it results in significant running time improvements, while starting with a more optimistic choice and increasing the precision if necessary doesn't seem to work as well. | 21 October 2016, 16:06:17 UTC |
e16c0a1 | Marc Mezzarobba | 12 September 2016, 14:35:02 UTC | naive_sum: minor change to debug message | 21 October 2016, 16:06:17 UTC |
1c1874b | Marc Mezzarobba | 12 September 2016, 12:38:38 UTC | naive_sum: improve heuristic convergence estimate The heuristic convergence estimate used to look only at one of the derivatives, leading to false positives, which themselves led to the majorants being refined a bit too aggressively--and to misleading debugging output suggesting very loose bounds when in fact the bounds were acceptable. | 21 October 2016, 16:06:17 UTC |
1719e73 | Marc Mezzarobba | 12 September 2016, 08:38:50 UTC | avoid re-evaluating rec coeffs to get residuals When unrolling a recurrence, avoid evaluating the coefficients that are used in residual computations twice. | 21 October 2016, 16:06:17 UTC |
3c9de8e | Marc Mezzarobba | 08 September 2016, 13:55:03 UTC | naive_sum: quick fix for prec increase pb... ...exposed by f76187ca433f75a2e60d08ff8d5d3f3a3a68b676 Precision tracking TBI in any case! | 21 October 2016, 16:06:17 UTC |
16ea9e7 | Marc Mezzarobba | 05 September 2016, 14:32:22 UTC | update timings for long tests | 21 October 2016, 16:06:17 UTC |
ce71c90 | Marc Mezzarobba | 07 September 2016, 11:52:05 UTC | naive_sum: a data structure for recurrences - Significantly faster than using polynomials directly in the case of recurrences over ℚ[i]. - Basic implementation for now; further speedup and code simplification should be possible... | 21 October 2016, 16:06:17 UTC |
c3ce696 | Marc Mezzarobba | 06 September 2016, 13:22:50 UTC | unused imports | 21 October 2016, 16:06:17 UTC |
760e519 | Marc Mezzarobba | 06 September 2016, 12:34:46 UTC | bounds: micro-optimizations | 21 October 2016, 16:06:17 UTC |
b2596e4 | Marc Mezzarobba | 05 September 2016, 15:48:52 UTC | bounds: explicit shift (exponent) in HyperexpMajorant It is important to handle the x^(large) factor separately when computing series expansions and bounds, because otherwise the Taylor shift done by HyperexpMajorant.series() densifies numerators of large valuation. And since we represent the numerators by dense polynomials in order to use arb, we might as well store the shift separately. (It might make sense in the future to keep a separate shift starting with in the computation of residuals, and revert to dense polynomials for the dense part of the residual and derived quantities.) | 06 September 2016, 08:42:12 UTC |
76b4305 | Marc Mezzarobba | 05 September 2016, 15:47:55 UTC | bounds: check that we use sparse polynomials for val >> 0 | 06 September 2016, 08:42:12 UTC |
4b2403e | Marc Mezzarobba | 05 September 2016, 14:21:29 UTC | bounds: optimizations in HyperexpMajorant | 06 September 2016, 08:42:12 UTC |
12f6997 | Marc Mezzarobba | 05 September 2016, 13:03:18 UTC | simplify and speed up MajorantSeries using the new sage bindings for acb_poly | 06 September 2016, 08:42:12 UTC |
4fe2167 | Marc Mezzarobba | 05 September 2016, 14:22:56 UTC | don't check validity of intial values we built ourselves | 06 September 2016, 08:42:12 UTC |
39708d0 | Marc Mezzarobba | 04 September 2016, 14:40:20 UTC | ssw: update ref to BCHKP 2016 preprint | 04 September 2016, 14:40:20 UTC |
53338ae | Marc Mezzarobba | 04 September 2016, 14:18:51 UTC | update timings for long tests | 04 September 2016, 14:24:12 UTC |
f76187c | Marc Mezzarobba | 04 September 2016, 14:09:01 UTC | bounds: fix (significant) overestimation Fix a very significant overestimation in the computation of bounds on vectors of derivatives of HyperexpMajorants, apparently introduced while porting HyperexpMajorant._bound() from using a recurrence on the derivatives of f*exp(int(g)) to the current code based on automatic differentiation, during the prehistory of this code. | 04 September 2016, 14:24:12 UTC |
a7617f6 | Marc Mezzarobba | 03 September 2016, 17:38:46 UTC | naive_sum: don't use f._mul_trunc_(g)... ...when f, g may have different parents | 04 September 2016, 14:14:47 UTC |
cc5c004 | Marc Mezzarobba | 30 August 2016, 14:19:47 UTC | reduce overhead of bound evaluation | 30 August 2016, 14:19:47 UTC |
8a9fe00 | Marc Mezzarobba | 30 August 2016, 14:03:21 UTC | optimize usage of Factorization | 30 August 2016, 14:03:21 UTC |
abcb76a | Marc Mezzarobba | 21 July 2016, 12:44:08 UTC | arbitrarily many terms in RationalMajorant Rewrite RationalMajorant to support an arbitrary number of rational terms (instead of a rational term + a polynomial one). This was originally part of an attempt to obtain tighter bounds by partial fraction decomposition. | 30 August 2016, 12:26:12 UTC |
c25226f | Marc Mezzarobba | 21 July 2016, 16:29:23 UTC | remove outdated comment seems no longer to apply since ddcae6228186bf18328b83921f5041180e1dd548 | 30 August 2016, 10:02:09 UTC |
dde3b04 | Marc Mezzarobba | 21 July 2016, 17:33:14 UTC | bounds: two new tests | 29 August 2016, 06:57:50 UTC |
3a15585 | Marc Mezzarobba | 18 August 2016, 13:34:50 UTC | smplseqbnd: update long test timings | 29 August 2016, 05:45:33 UTC |
3a18252 | Marc Mezzarobba | 17 August 2016, 19:47:37 UTC | smplseqbnd: some more doc and tests | 18 August 2016, 10:50:08 UTC |
4740649 | Marc Mezzarobba | 17 August 2016, 19:47:37 UTC | smplseqbnd: some support for ∞ in RatSeqBound.plot() | 18 August 2016, 10:50:08 UTC |
79e6c50 | Marc Mezzarobba | 17 August 2016, 14:43:48 UTC | smplseqbnd: some documentation | 18 August 2016, 10:50:08 UTC |
2638539 | Marc Mezzarobba | 13 August 2016, 14:32:33 UTC | smplseqbnd: adapt BoundDiffopStats Minimally adapt the timings we collect when bounding differential operators to the simplified bounds on rational sequences. | 18 August 2016, 10:50:06 UTC |
23b20de | Marc Mezzarobba | 13 August 2016, 14:22:44 UTC | smplseqbnd: remove opt param 'nmin' (no longer makes sense with the simplified bounds on sequences) | 18 August 2016, 10:50:04 UTC |
6c8e50a | Marc Mezzarobba | 13 August 2016, 14:11:11 UTC | smplseqbnd: update existing tests | 18 August 2016, 10:49:16 UTC |
b4da89a | Marc Mezzarobba | 29 July 2016, 15:02:17 UTC | simplify the bounds on rational sequences Replace the bounds on sequences defined by almost everywhere rational functions (SeqBounds & co) by a significantly simpler variant. The new bounds are more readable, easier to implement (especially in the presence of interval coefficients) and optimize, and less costly to compute. Experimentally, they appear to be reasonably tight to very tight when the indicial equation only has small roots (in particular, at ordianry points). However, unlike the old version which should be tight even in pathological cases, they can also be quite loose. Thus, the old implementation may still be useful, if only for comparison purposes--but it would be too much of a hurdle to maintain it for now, I'd rather add it back in the future if necessary. | 18 August 2016, 10:47:48 UTC |
db31f28 | Marc Mezzarobba | 13 August 2016, 14:38:47 UTC | bounds: update outdated comment | 18 August 2016, 10:47:48 UTC |
7134eb5 | Marc Mezzarobba | 29 July 2016, 10:08:01 UTC | fix/clarify specification of sequence bounds | 13 August 2016, 13:25:32 UTC |
f1293fe | Marc Mezzarobba | 02 August 2016, 21:30:58 UTC | examples: dopX → fcc.dopX | 13 August 2016, 13:25:18 UTC |
53e1d35 | Marc Mezzarobba | 21 July 2016, 12:49:58 UTC | update timings for long tests | 21 July 2016, 12:54:18 UTC |
155facf | Marc Mezzarobba | 10 July 2016, 14:57:33 UTC | make Koutschan's operators available as data | 21 July 2016, 12:54:18 UTC |
ee80bfd | Marc Mezzarobba | 10 July 2016, 14:51:59 UTC | missing word in docstring | 10 July 2016, 14:51:59 UTC |
3f5fdad | Marc Mezzarobba | 20 May 2016, 13:46:19 UTC | comment | 20 May 2016, 13:46:19 UTC |
834fa8b | Marc Mezzarobba | 20 May 2016, 13:43:55 UTC | merge bound_inverse_poly into DiffOpBound Done pretty minimally at the moment. Lost some tests in the process; these may need to be salvaged depending how the new structure takes shape. | 20 May 2016, 13:45:33 UTC |
72af167 | Marc Mezzarobba | 19 May 2016, 16:16:32 UTC | bounds: coarser RatSeqBounds for small n when only interested in large n Not sure if this is worth the additional complexity... but this does make bound computations noticeably faster on the example of fcc.dop6). | 19 May 2016, 18:13:57 UTC |
40d95d6 | Marc Mezzarobba | 19 May 2016, 15:37:01 UTC | bounds: minor optimization | 19 May 2016, 15:37:01 UTC |
bbae41f | Marc Mezzarobba | 19 May 2016, 14:03:53 UTC | bound_ratio_large_n: fix incorrect formula | 19 May 2016, 14:03:53 UTC |
f45dedf | Marc Mezzarobba | 19 May 2016, 06:32:03 UTC | remove another kludge introduced for performance reasons | 19 May 2016, 06:32:03 UTC |
afe5c5e | Marc Mezzarobba | 18 May 2016, 16:46:16 UTC | some cleanup/doc | 18 May 2016, 17:03:29 UTC |
0b489f4 | Marc Mezzarobba | 18 May 2016, 15:52:37 UTC | get rid of the subclasses of Point | 18 May 2016, 16:40:46 UTC |
558258f | Marc Mezzarobba | 14 April 2016, 07:13:01 UTC | remove workaround for slow NFEq -> ball conversion | 18 May 2016, 14:44:04 UTC |
bacf63c | Marc Mezzarobba | 18 May 2016, 14:34:04 UTC | fix doctest for sage-7.2 | 18 May 2016, 14:34:04 UTC |