Limit Techniques
The toolkit for actually evaluating limits — factoring, conjugate rationalising, standard limits, and Taylor expansion — drilled on real GATE DA NATs from 2024 and 2025.
What you'll learn
- Factor and cancel to resolve a 0/0 that comes from a common factor
- Multiply by the conjugate to handle √-differences and ∞ − ∞ forms
- Standard limits: sin x / x → 1 and (eˣ − 1)/x → 1
- Taylor-expand the numerator and denominator to read off a limit
Before you start
The previous lesson ended at a wall. You tried to evaluate lim_{x → 1} (x² − 1)/(x − 1) by
substituting x = 1, and out came 0/0 — not an answer, just a shrug. Yet the curve plainly
heads somewhere definite. Knowing what a limit means did not tell you how to compute one,
and closing that gap is the whole of this lesson.
That 0/0 is an indeterminate form: the limit may well exist, but the expression has to
be rewritten before it will show itself. ∞ − ∞ is another such form, and so is ∞/∞. None
of them is a value — each is a signpost reading rewrite me first. This lesson is the small
toolkit for that rewrite, and GATE DA leans on it hard: it recurs as a NAT, in 2024 and again
in 2025. The same rewrites earn their keep in real machine-learning code, where a raw 0/0
or ∞ − ∞ becomes a NaN that silently poisons an entire training run.
Think of an indeterminate form as a locked door. 0/0 is not a dead end — it is a lock, and
each lock takes a particular key. The craft is reading which lock you face, then reaching
for the matching key rather than rattling all of them at once.
The four techniques
- Factor and cancel. When direct substitution gives
0/0because numerator and denominator share a root, factor both and cancel the common piece. The classic(x² − 1)/(x − 1)becomes(x − 1)(x + 1)/(x − 1) = x + 1 → 2asx → 1— exactly the wall from last lesson, now walked straight through. - Rationalise with the conjugate. When a
√difference produces0/0or∞ − ∞, multiply top and bottom by the conjugate (same terms, opposite sign) to turn the square-root difference into a clean difference of squares. - Standard limits. Memorise these two — they appear constantly:
sin x / x → 1asx → 0, and(eˣ − 1)/x → 1asx → 0. - Taylor expansion. Replace each function by its first few series terms near the point
(e.g.
sin x ≈ x,cos x ≈ 1 − x²/2,ln(1 + u) ≈ u − u²/2,eˣ ≈ 1 + x), then cancel and read off the ratio. This dissolves messy0/0forms quickly.
How GATE asks this
This is a NAT favourite: you are handed a limit that substitution cannot crack and must
type the numeric value. GATE DA 2025 asked lim_{t → ∞} (√(t² + t) − t) — a ∞ − ∞ form
cracked by the conjugate (worked below). GATE DA 2024 asked
lim_{x → 0} ln((x² + 1)·cos x) / x² — a 0/0 form cracked by Taylor expansion. Occasionally
the form appears as an MCQ asking which technique applies. Either way, the skill tested
is the same: recognise the form, then choose the right rewrite.
Worked example 1 — a real GATE DA 2025 question
Evaluate
lim_{t → ∞} (√(t² + t) − t).
Substituting t = ∞ gives ∞ − ∞, which is indeterminate — do not stop here. The form is
a √-difference, so reach for the conjugate key: multiply by √(t² + t) + t over itself.
√(t²+t) − t = (√(t²+t) − t) · (√(t²+t) + t)
──────────────────────────────
√(t²+t) + t
= (t² + t) − t² t
────────────────────── = ───────────
√(t²+t) + t √(t²+t) + t
Now divide top and bottom by t (pulling t out of the root as √(t²·…) = t·√…):
t 1
─────────────── = ──────────────────── → 1 / (√1 + 1) = 1/2
√(t²+t) + t √(1 + 1/t) + 1
As t → ∞, the 1/t inside the root vanishes, leaving 1/(√1 + 1) = 1/2. So the limit is
0.5. (This is GATE DA 2025.)
Sanity-check it by feeding in growing values of t: at t = 10 the expression is already
0.488…, at t = 100 it is 0.498…, and at t = 10⁵ it is 0.499999. The march toward
0.5 is exactly what the algebra promised.
Worked example 2 — a real GATE DA 2024 question
Evaluate
lim_{x → 0} ln((x² + 1)·cos x) / x².
Substituting x = 0 gives ln(1·1)/0 = 0/0 — indeterminate. This one is messy, so reach for
the Taylor key: split the log of a product into a sum, then expand each piece near x = 0.
ln((x²+1)·cos x) = ln(1 + x²) + ln(cos x)
ln(1 + x²) ≈ x² (since ln(1+u) ≈ u, with u = x²)
ln(cos x) ≈ ln(1 − x²/2) ≈ −x²/2 (since cos x ≈ 1 − x²/2)
Add the two expansions and divide by x²:
ln((x²+1)·cos x) x² − x²/2 x²/2 1
───────────────── ≈ ─────────── = ────── = ── = 0.5
x² x² x² 2
The higher-order terms vanish faster than x², so the limit is exactly 0.5. (This is
GATE DA 2024.) Two functions, two short series, and the 0/0 dissolves.
A question to carry forward
Each technique here works, but each also asks you to spot the trick — the right factor, the
right conjugate, the right series. That cleverness is exactly what makes these problems feel
slow under exam pressure. So a fair question arises: is there one mechanical rule that
cracks any 0/0 or ∞/∞ without the hunt for an algebraic move — a rule you could turn
almost without thinking? Here is the thread onward: what if, on an indeterminate quotient, you
simply differentiated the top and the bottom separately and tried again?
In one breath
- An indeterminate form (
0/0,∞ − ∞,∞/∞) is not a value — it is a signpost to rewrite first, then take the limit. - Factor and cancel a
0/0from a shared root:(x²−1)/(x−1) = x+1 → 2. - Conjugate a
√-difference /∞ − ∞:√(t²+t) − t → 1/2(GATE DA 2025). - Standard limits, quotable on sight:
sin x / x → 1and(eˣ − 1)/x → 1asx → 0. - Taylor-expand messy forms (
sin x ≈ x,cos x ≈ 1 − x²/2,ln(1+u) ≈ u): the 2024ln((x²+1)cos x)/x² → 1/2falls in two lines. - Reflex: read the form, then pick the matching key — never treat
∞ − ∞as plain subtraction.
Practice
Quick check
Practice this in an interview
All questionsThe core toolkit is: system prompts (role and constraints), few-shot examples (format and tone anchoring), chain-of-thought (step-by-step reasoning), and output constraints (JSON schema, stop sequences). Combining these predictably closes the gap between a capable base model and a production-ready feature.
Reasoning models are trained to produce an extended chain of thought before answering, often via reinforcement learning, so they spend more computation deliberating on hard problems. Test-time compute is the idea of improving answer quality by allocating more inference-time compute, for example longer reasoning chains, sampling multiple solutions, or self-verification, rather than only scaling parameters.