Gather a quotient of powers whose exponents a rewrite has already moved (#740)

[Rafael-SOWNet]

Closes #740.

What was wrong

a^p / b^p is gathered into (a/b)^p, and stopped being gathered as soon as one of the bases was itself a power:

"(a ^ 2) ^ x / (b ^ 2) ^ x".Simplify()      // (a / b) ^ (2 * x)     ✅
"(a ^ 2) ^ x / b ^ x".Simplify()            // unchanged             ❌
"(a ^ 2 + 1) ^ x / (a ^ 2) ^ x".Simplify()  // unchanged             ❌

It is not the shape of the quotient but the order the rules run in. The rule that rewrites (b^c)^p as b^(c*p) applies to the child, on the way up, so by the time the pair of powers is looked at it has already happened:

• where both bases are powers it moves both exponents together, the exponents still match, and the pair survives — which is why the first line works, and why this looked like a question of shape;
• where one base is a power it moves one exponent and not the other, they stop matching, and there is nothing left to pair on.

What the fix does

Reads the pair back: a^p / b^(c*p) is (a / b^c)^p, and the same the other way up.

Restricted to a whole c, so b^c moves into the base and nothing gains a root it did not have. A fractional c would have to divide the exponent instead, turning (sqrt(x) + 1)^x / sqrt(x)^x into ((sqrt(x) + 1)^2 / x)^(x/2) — a gathered form bought with a squared numerator. That is a judgement about output rather than the gap this fixes, so the issue's second example is deliberately left as it stands, and pinned as such.

	was	is
(a^2 + 1)^x / (a^2)^x	unchanged	(1 + 1/a^2)^x
(x^3 + 1)^x / (x^3)^x	unchanged	(1 + 1/x^3)^x
(a^2)^x / b^x	unchanged	(a^2 / b)^x
x^(2a) / y^a	unchanged	(x^2 / y)^a
2^(2x) / 3^x	unchanged	(4/3)^x
x^4 / y^2	unchanged	unchanged — a numeric exponent is not a product, so it matches nothing here

Why it is worth gathering

The limit machinery reads a 1^oo off a single power and cannot see one in a quotient, so the same function was answered or not according only to how it had been written:

	was	is
lim x->+oo (x^2 + 1)^x / (x^2)^x	unevaluated after 5.5 s	1 in 31 ms
lim x->+oo (x^3 + 1)^x / (x^3)^x	unevaluated	1
lim x->+oo ((x^2 + 1) / x^2)^x	1 (unchanged)	1

Measured

• Suite 4979 → 4986 passed / 0 failed, F# 130/130.
• Corpus 112/117, 0 wrong / 0 error / 0 timeout, every verdict and answer byte-identical.
• work/rootcheck 595/596; its one incomplete case is A power of a polynomial is solved by inverting into itself, so the roots come back containing x #744, fixed in Only invert an equation whose variable occurs once (#744) #745, not here.

The tests pin the value as well as the shape — the rewrite is checked numerically at a point where every base is positive, which is where a^p / b^(c*p) = (a/b^c)^p holds without a branch argument — and the limits #739 fixed are pinned alongside, since they go through this same gathering.

🤖 Generated with Claude Code