holonomic question related to lazy term generation from an unexpanded expression

43 views
Skip to first unread message

Chris Smith

unread,
Oct 2, 2026, 6:06:05 PMOct 2
to sympy

I've been experimenting with lazy leading-term generation for unexpanded polynomials after working on "yet another method of computing degree without relying on full expansion". The prototype aims to generate leading terms without expanding the expression. I have a basic system working but am now investigating how to handle larger products. For example, (and I let AI take over from here):

After reversing \(x\) via \(t=1/x\), products such as

\[ \prod_i(a_i+b_it)^{n_i} \]

have a first-order differential annihilator, and converting that to a coefficient recurrence lets us generate only the leading polynomial terms that are requested.

I discovered that this is essentially the same transformation already implemented by HolonomicFunction.to_sequence(): for a two-factor example it produces exactly the same recurrence as the prototype.

I am wondering what the intended architectural boundary of the holonomic machinery is. HolonomicSequence currently seems mainly to store an annihilator recurrence and initial conditions, while recurrence evaluation is done by HolonomicFunction.series().

Would it make sense to factor out a lightweight “differential annihilator -> coefficient recurrence” and/or recurrence-coefficient iterator that could be reused outside holonomic? In particular, polys cannot reasonably depend on holonomic since holonomic already depends on polys.

Is HolonomicSequence intended eventually to be the general P-recursive sequence abstraction for this sort of use, or would a lightweight lower-level recurrence utility be more appropriate?


/c

lkcl

unread,
Nov 15, 2026, 9:35:42 PM (4 days ago) Nov 15
to sympy
On Saturday, October 3, 2026 at 3:36:05 AM UTC+5:30 smi...@gmail.com wrote:

I am wondering what the intended architectural boundary of the holonomic machinery is. HolonomicSequence currently seems mainly to store an annihilator recurrence and initial conditions, while recurrence evaluation is done by HolonomicFunction.series().

Would it make sense to factor out a lightweight “differential annihilator -> coefficient recurrence” and/or recurrence-coefficient iterator that could be reused outside holonomic? In particular, polys cannot reasonably depend on holonomic since holonomic already depends on polys.

hi, i have been dealing with holonomic math for some time now, please bear with me
as this is written *without* the symbiotic AI assistance i normally use to compensate
for stroke damage.

1. holonomic balance automatically means you are looking at a Golden Spiral situation
2. that implies you are looking at an RCFT modular torus
3. any polynomial series must converge at the limit and that implies superexponentiation
    and/or Cauchy product convergence

basically you need to prove that the lazy-evaluated terms will not go haywire unstable
(non converging) and i don't mean by way of the usual issues associated with FP
inaccuracy (dividing by smaller and smaller numbers introducing larger and
larger rounding errors)

on Cauchy Product convergence imagine you have two commutative lazy-evaluated polynomials
you need to prove that a term moving from (a+b)^2 has b^2 converging to zero much faster
than the a^2 + 2ab part. standard stuff but the lazy evaluation angle is new to me.
appreciate you introducing me to it.

l.

Reply all
Reply to author
Forward
0 new messages