Taylor series for pi: 35 Practice MCQs with Answers and Explanation

Q1. A student approximates $\sin(3^\circ)$ using a Maclaurin series but obtains a result significantly different from the calculator value. Upon review, you notice they substituted $x = 3$ directly into the series $x - x^3/3! + x^5/5!$ . Beyond the obvious unit error, which conceptual misunderstanding about Taylor series convergence does this specific mistake reveal regarding the domain of validity?

A.The student assumes the series converges for all real numbers regardless of the function's periodicity.

B.The student fails to recognize that trigonometric derivatives used to derive the series are defined strictly in radians, making the coefficients invalid for degree inputs. ✅

C.The student incorrectly believes that Maclaurin polynomials only approximate functions near

x=1

rather than

x=0

.

D.The student assumes the remainder term

R_n(x)

vanishes instantly for any integer input, ignoring the factorial growth in the denominator.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This question targets a fundamental conceptual prerequisite for approximation. While substituting degrees is a common calculation error, the deeper issue is understanding that the calculus operations (differentiation) generating the series coefficients $f^{(n)}(0)/n!$ rely on radian measure. If degrees were used, the derivative of sine would include a conversion factor, altering every coefficient. Thus, the series form itself is intrinsically tied to the radian definition, not just the numerical evaluation.

Q2. When approximating $\int_0^1 e^{-x^2} dx$ using a power series, a student integrates the first four terms and claims the error is exactly equal to the fifth term because the series is alternating. However, the actual error differs slightly. Which condition of the Alternating Series Estimation Theorem might be subtly violated or misapplied if the integration limits were changed to $[0, 2]$ ?

A.The terms of the integrated series no longer alternate in sign over the interval

[0, 2]

.

B.The absolute values of the terms do not monotonically decrease for all

n

when evaluated at the upper limit

x=2

. ✅

C.The function

e^{-x^2}

is not continuous on

[0, 2]

, invalidating term-by-term integration.

D.The radius of convergence for the Maclaurin series of

e^{-x^2}

is less than 2, so the series diverges at the endpoint.

💡 Difficulty: hard | ✅ Correct: B

📖 Explanation: This problem tests error analysis and the precise conditions of estimation theorems. The Alternating Series Estimation Theorem requires that the magnitude of terms $|a_n|$ be non-increasing. For $\int e^{-x^2} dx$ , the terms involve powers of $x$ . At $x=1$ , factorials dominate quickly, ensuring monotonic decrease. However, at $x=2$ , the numerator $2^{2n+1}$ grows rapidly before the factorial takes over. If the sequence of term magnitudes increases initially before decreasing, the simple bound $|R_n| \leq |a_{n+1}|$ is invalid until monotonicity is established.

Q3. Consider two methods to approximate $\ln(2)$ : Method A uses the standard Maclaurin series $\sum (-1)^{k+1}/k$ at $x=1$ . Method B uses the transformed series $2\sum \frac{1}{2k+1}(\frac{1}{3})^{2k+1}$ . Why is Method B computationally superior for achieving high-precision approximations, despite being algebraically more complex?

A.Method B converges absolutely while Method A converges conditionally, eliminating round-off errors entirely.

B.Method B utilizes a geometric decay rate of

1/9

per term pair compared to Method A's harmonic decay of

1/n

, drastically reducing truncation error for the same number of terms. ✅

C.Method A suffers from catastrophic cancellation due to alternating signs, whereas Method B has all positive terms.

D.Method B avoids the use of factorials in the denominator, making individual term computation faster on digital hardware.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This compares convergence rates, a critical HOTS concept in numerical analysis. The standard alternating harmonic series converges extremely slowly (error $\approx 1/n$ ). To get 5 decimal places, one needs $\approx 100,000$ terms. The transformed series derives from $\ln((1+x)/(1-x))$ with $x=1/3$ . The terms shrink by a factor of roughly $x^2 = 1/9$ each step. This geometric convergence means only ~6-7 terms are needed for the same accuracy. Understanding rate of convergence is essential for efficient algorithm design.

Q4. You are modeling a physical system where the restoring force is proportional to $\sin(\theta)$ . For small angles, you use the linear approximation $\theta$ . If the maximum allowable relative error in the force model is 1%, what is the most rigorous way to determine the valid range of $\theta$ ?

A.Set

|\sin(\theta) - \theta| < 0.01

and solve numerically.

B.Use the Lagrange error bound with

n=1

and

M=1

to find

|R_1(\theta)| \leq \theta^2/2 < 0.01

. ✅

C.Set the next term in the series

\theta^3/6

equal to 0.01 and solve for

\theta

.

D.Graph

y=\sin(\theta)

and

y=\theta

and visually estimate the intersection point where the gap exceeds 1%.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This applies error bounds to physical modeling constraints. Visual estimation is imprecise. Solving $|\sin \theta - \theta| < 0.01$ gives absolute error, not relative error, and requires numerical methods. Using the next term is an approximation of the error, not a guaranteed bound. The Lagrange Remainder Theorem provides a mathematically rigorous upper bound on the error $|R_n(x)| \leq M|x-a|^{n+1}/(n+1)!$ . Since $|f''(z)| = |-\sin z| \leq 1$ , setting $\theta^2/2 < 0.01$ guarantees the error constraint is met without solving transcendental equations.

Q5. A graph shows a function $f(x)$ and its Taylor polynomials $P_1, P_2, P_3$ centered at $x=0$ . Near the origin, all polynomials track $f(x)$ closely. As $x$ increases, $P_1$ diverges first, followed by $P_2$ , while $P_3$ maintains accuracy longest. However, at a specific point $x=c$ , $P_3$ suddenly crosses $f(x)$ while $P_2$ remains below it. What does this crossing behavior imply about the remainder term $R_3(c)$ ?

A.The remainder

R_3(c)

must be zero, indicating exact convergence at that point.

B.The sign of the fourth derivative

f^{(4)}(z)

changes within the interval

(0, c)

, causing the error bound to shift direction.

C.The third-degree polynomial has overshot the true value, meaning the next non-zero term in the series has the opposite sign of the current cumulative error. ✅

D.The radius of convergence for the Taylor series has been exceeded at

x=c

, rendering the polynomial invalid.

💡 Difficulty: hard | ✅ Correct: C

📖 Explanation: This interprets graphical behavior through the lens of series theory. Taylor polynomials approximate functions by matching derivatives. The error $R_n(x)$ behaves like the first omitted term $\frac{f^{(n+1)}(c)}{(n+1)!}x^{n+1}$ for small $x$ . If $P_3$ crosses $f(x)$ , the error changes sign. This typically indicates that the dominant term in the remainder (the $n+1$ term) has a sign opposite to the accumulated error of lower-order terms, or that the function's concavity/jerk characteristics change such that the cubic fit transitions from overestimate to underestimate. It reflects the local nature of Taylor approximations.

Q6. A student attempts to approximate $f'(x)$ for $f(x) = \sum_{k=0}^\infty c_k x^k$ by differentiating the series term-by-term. They correctly obtain $\sum k c_k x^{k-1}$ but claim this new series always has the same interval of convergence as the original. Construct a counter-example scenario where this claim fails specifically at the endpoints.

A.The original series converges at

x=R

but the differentiated series diverges at

x=R

because the extra factor of

k

prevents conditional convergence. ✅

B.The differentiated series always has a smaller radius of convergence, so the claim fails everywhere.

C.Differentiation introduces a constant of integration that shifts the interval of convergence.

D.The original series must be absolutely convergent for differentiation to preserve the interval; otherwise, the radius shrinks.

💡 Difficulty: medium | ✅ Correct: A

📖 Explanation: This addresses endpoint behavior nuances. While the Radius of Convergence $R$ remains identical upon differentiation, the behavior *at* the endpoints $\pm R$ can change. Differentiation multiplies the $k$ -th term by $k$ . A series might converge conditionally at an endpoint (e.g., alternating harmonic), but multiplying by $k$ could cause the terms to fail the divergence test or comparison tests, leading to divergence. Integration, conversely, divides by $k$ , potentially improving convergence at endpoints. Students must distinguish between the open interval $(-R, R)$ and the closed interval of convergence.

Q7. In approximating $\pi$ using the Leibniz series $4(1 - 1/3 + 1/5 - \dots)$ , a student computes 10,000 terms and still lacks 3-decimal accuracy. They propose accelerating convergence by averaging consecutive partial sums $S_n$ and $S_{n+1}$ . Why is this heuristic effective for this specific series but potentially dangerous for a general non-alternating series?

A.Averaging works for any convergent series because it reduces variance.

B.For alternating series with monotonically decreasing terms, the true sum lies strictly between consecutive partial sums; averaging exploits this bracketing property, which does not exist for non-alternating series. ✅

C.Averaging eliminates the truncation error entirely by cancelling the remainder term.

D.It is only effective because

\pi

is irrational; for rational sums, averaging produces incorrect results.

💡 Difficulty: hard | ✅ Correct: B

📖 Explanation: This mixes concepts of series properties and numerical heuristics. The Leibniz series is alternating with decreasing terms. By the Alternating Series Test, $S_{2k} < S < S_{2k+1}$ . The error is bounded by the next term. Averaging $(S_n + S_{n+1})/2$ effectively centers the estimate within this error bracket, often yielding accuracy comparable to using twice as many terms. For non-alternating series, partial sums may approach the limit monotonically from one side; averaging two underestimates simply yields another underestimate, providing no structural advantage and potentially masking divergence or slow convergence patterns.

Q8. When deriving the Maclaurin series for $\tan^{-1}(x)$ , it is standard practice to integrate the geometric series for $1/(1+x^2)$ rather than computing derivatives of $\tan^{-1}(x)$ directly. If a student instead computes derivatives and finds $f'(0)=1, f''(0)=0, f'''(0)=-2$ , they struggle to find a general pattern. What is the primary strategic advantage of the integration method beyond computational ease?

A.Integration guarantees absolute convergence whereas differentiation only guarantees conditional convergence.

B.The geometric series provides a known closed-form general term

(-1)^k x^{2k}

that survives integration, whereas higher derivatives of inverse trig functions become combinatorially complex without a clear sigma notation pattern. ✅

C.Integration automatically determines the constant of integration

C

without needing initial conditions.

D.The integration method extends the radius of convergence from 1 to infinity, unlike the derivative method.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This evaluates method selection strategy. Direct differentiation of composite/inverse functions leads to increasingly complex expressions involving products and chains, making the general $n$ -th derivative formula obscure. Recognizing $\tan^{-1} x$ as an integral of a rational function allows leveraging the simple geometric series structure. Term-by-term integration preserves the clean $(-1)^k$ and power structure, immediately yielding $\sum (-1)^k x^{2k+1}/(2k+1)$ . This highlights the importance of connecting new functions to known series representations via calculus operations rather than brute-force definition application.

Q9. A student uses the Remainder Estimation Theorem to bound the error of approximating $e^x$ at $x=2$ using a degree-4 polynomial. They set $M = e^2$ because $f^{(5)}(x) = e^x$ is increasing on $[0,2]$ . Another student argues $M=e^2$ is too conservative and suggests using $M=3^2=9$ since $e<3$ . How does this substitution affect the rigor and utility of the error bound?

A.Using

M=9

makes the bound tighter and equally rigorous because

e<3

is a proven fact.

B.Using

M=9

provides a computable rational bound that avoids circular reasoning (using

e

to approximate

e

), though it is slightly looser than the theoretical minimum. ✅

C.Using

M=9

invalidates the theorem because

M

must be the exact maximum of the derivative.

D.Both bounds are useless because the Taylor series for

e^x

converges too slowly at

x=2

.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This analyzes the practical application of error theorems. Theoretically, $M=e^2$ is the least upper bound. However, if the goal is to approximate $e^2$ , using $e^2$ in the error bound is circular. Replacing $e$ with a known upper bound like 3 (or 2.72) breaks the circularity and yields a usable numerical guarantee. While $9 > e^2 \approx 7.39$ , making the bound looser, it transforms an abstract expression into a concrete verification tool. This distinction between theoretical tightness and computational utility is crucial in numerical analysis.

Q10. Consider the binomial series expansion for $(1+x)^m$ where $m$ is not an integer. A student applies this to approximate $\sqrt{1+x}$ at $x=1.5$ and gets a nonsensical result. They argue that since $\sqrt{1+x}$ is defined and smooth at $x=1.5$ , the series should work. What fundamental constraint on power series representations explains this failure despite the function's smoothness?

A.The function must be infinitely differentiable at the center, which

\sqrt{1+x}

is not.

B.Power series represent functions only within their radius of convergence determined by singularities in the complex plane; for

(1+x)^{1/2}

, the singularity at

x=-1

limits

R

to 1, excluding

x=1.5

. ✅

C.Binomial series only converge for negative values of

x

.

D.The student failed to include enough terms; the series eventually converges at

x=1.5

after 1000 terms.

💡 Difficulty: hard | ✅ Correct: B

📖 Explanation: This tests deep conceptual understanding of convergence domains. Real-valued smoothness is insufficient for power series representation globally. The radius of convergence of a Taylor/Maclaurin series centered at 0 is the distance to the nearest singularity in the complex plane. For $(1+x)^{1/2}$ , the branch point is at $x=-1$ . Distance from 0 to -1 is 1. Thus $R=1$ . Even though the real function exists at $x=1.5$ , the power series centered at 0 cannot reach it. One must re-center the series (e.g., at $x=1$ ) to approximate values outside $(-1, 1)$ .

Q11. You need to evaluate $\int_0^{0.1} \frac{\sin x}{x} dx$ to 6 decimal places. You decide to use the Maclaurin series. Which of the following represents the most efficient and error-controlled workflow?

A.Substitute

x=0.1

into the series for

\sin x

, divide by 0.1, then integrate the resulting constant.

B.Integrate the series

\sum (-1)^k x^{2k}/(2k+1)!

term-by-term to get

\sum (-1)^k x^{2k+1}/[(2k+1)(2k+1)!]

, substitute limits, and use the alternating series error bound on the resulting numerical series. ✅

C.Use Simpson's Rule with

n=100

subintervals since numerical integration is always more accurate than series for definite integrals.

D.Differentiate the series for

\sin x

to improve convergence speed before integrating.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This applies multi-step reasoning to algorithm selection. Option A is mathematically incorrect (cannot treat variable as constant). Option C ignores the prompt's context and series advantages for high precision/small intervals. Option D is nonsensical. Option B correctly performs term-by-term integration to generate a new series for the antiderivative, evaluates it at bounds to create a numerical alternating series, and applies the appropriate error test. This leverages the rapid convergence of the sine series near 0 and provides a guaranteed error bound, which is superior to blind numerical quadrature for this specific task.

Q12. A physics model uses the approximation $\cos \theta \approx 1 - \theta^2/2$ . A critic points out that for $\theta = 0.5$ rad, the error is non-negligible. If you must maintain this second-order model form but improve accuracy without adding higher-order terms, which modification is theoretically justified?

A.Replace

\theta^2/2

with a fitted coefficient

c\theta^2

where

c

minimizes mean-square error over the operating range. ✅

B.Add a constant offset to force the curve through the endpoint.

C.Multiply the entire expression by

\sec(0.5)

to normalize.

D.No modification is possible; one must strictly use the Taylor coefficient

1/2

.

💡 Difficulty: hard | ✅ Correct: A

📖 Explanation: This bridges pure math and applied modeling. Taylor series minimize error locally at a point (derivatives match). In engineering, we often care about error over an interval. Minimizing mean-square error or Chebyshev norm over a specific domain yields coefficients that differ from Taylor coefficients. While $1-\theta^2/2$ is optimal at 0, a coefficient like $0.48$ might yield lower average error over $[0, 0.5]$ . This distinguishes 'Taylor Approximation' from 'Best Uniform Approximation', a key HOTS concept in applied mathematics and signal processing.

Q13. When approximating $e$ using $\sum 1/k!$ , a student stops when the next term $1/(n+1)! < 10^{-6}$ . They assume this guarantees 6-decimal accuracy. Why is this assumption technically flawed for a non-alternating positive series, even though it often works in practice?

A.For positive series, the remainder is the sum of ALL remaining terms, not just the next term; however, for factorial denominators, the tail is dominated by the first omitted term. ✅

B.The student should have used the Ratio Test to determine stopping criteria.

C.Positive series never allow error estimation based on individual terms.

D.The remainder for positive series is exactly equal to the integral of the function from

n

to infinity.

💡 Difficulty: medium | ✅ Correct: A

📖 Explanation: This targets error analysis misconceptions. The Alternating Series Estimation Theorem ( $|R_n| \leq |a_{n+1}|$ ) does NOT apply to positive series. For $\sum 1/k!$ , $R_n = \sum_{k=n+1}^\infty 1/k!$ . Strictly speaking, this sum is greater than $1/(n+1)!$ . However, due to super-exponential factorial growth, $R_n \approx a_{n+1}(1 + \epsilon)$ . The student's logic is formally wrong (applying alternating test to positive series) but numerically salvageable due to specific series properties. Recognizing why it's formally wrong is the HOTS component.

Q14. Given the Maclaurin series for $f(x)$ is $\sum a_k x^k$ and for $g(x)$ is $\sum b_k x^k$ , you wish to approximate $h(x) = f(x)g(x)$ . You compute the Cauchy product. If both original series have radius of convergence $R=1$ , what can be definitively stated about the radius of convergence $R_h$ of the product series?

A.

R_h

must be exactly 1.

B.

R_h

must be at least 1, but could be larger if singularities cancel. ✅

C.

R_h

must be less than 1 due to coefficient multiplication.

D.

R_h

is unrelated to the original radii.

💡 Difficulty: hard | ✅ Correct: B

📖 Explanation: This tests mixed concepts of series algebra and analytic continuation. The product of two power series converges absolutely within the intersection of their disks of convergence, so $R_h \geq \min(R_f, R_g) = 1$ . However, if $f$ and $g$ have singularities at the boundary that are removable in the product (e.g., $f(x)=(1-x)^{-1}, g(x)=(1-x)$ ), the product function may be analytic on a larger domain. Thus, the product series *can* have a larger radius. Students often mistakenly assume operations preserve boundaries rigidly without considering function-level simplifications.

Q15. A graph displays $y=f(x)$ and three approximating polynomials $A, B, C$ near $x=0$ . Polynomial $A$ intersects $f$ at 0 with same slope but different curvature. Polynomial $B$ matches value, slope, and curvature at 0 but diverges quickly. Polynomial $C$ matches value and slope but has opposite curvature. Rank these by degree of contact at $x=0$ .

A.A > B > C

B.B > A > C ✅

C.B > C > A

D.C > A > B

💡 Difficulty: easy | ✅ Correct: B

📖 Explanation: This is a graph-based conceptual check. Degree of contact corresponds to matching derivatives. Matching value = order 0. Matching slope = order 1. Matching curvature (2nd deriv) = order 2. Polynomial B matches up to 2nd derivative (highest contact). A matches up to 1st. C matches 0th and 1st but fails 2nd (opposite sign implies mismatch). Wait, if C has opposite curvature, $f''(0) \neq C''(0)$ , so contact is order 1. A also has different curvature, so contact is order 1. Between A and C, both are order 1. The ranking B > A=C is ideal, but given options, B is clearly highest. If forced to rank A vs C, usually 'different' vs 'opposite' are both failures of 2nd order. Assuming standard hierarchy: B (2nd) > A (1st) > C (1st with sign error). Option B reflects B as best.

Q16. In approximating $\ln(1.1)$ , why is it preferable to use the series for $\ln(1+x)$ with $x=0.1$ rather than the series for $\ln(x)$ centered at $x=1$ with $x=1.1$ , even though both are valid Taylor expansions?

A.They are mathematically identical series; there is no preference.

B.The series for

\ln(1+x)

at

x=0.1

avoids computing powers of

(x-1)

which are numerically unstable.

C.Actually, expanding

\ln(x)

about

x=1

yields

\sum (-1)^{k-1}(x-1)^k/k

, which IS the same as substituting

u=x-1

into

\ln(1+u)

; the question tests recognition of equivalence. ✅

D.Expanding about

x=1

has a smaller radius of convergence.

💡 Difficulty: medium | ✅ Correct: C

📖 Explanation: This challenges students to see connections. Many treat 'Maclaurin of ln(1+x)' and 'Taylor of ln(x) at 1' as distinct formulas to memorize. In reality, letting $u = x-1$ transforms one into the other. The computational steps are identical. Recognizing this equivalence reduces cognitive load and prevents rote memorization. Distractors exploit fears of numerical instability or convergence differences that don't exist here. This promotes structural understanding over procedural isolation.

Q17. A student calculates the 3rd degree Maclaurin polynomial for $\cos(x)$ as $1 - x^2/2 + x^4/24$ . They then attempt to bound the error using $n=3$ in the Lagrange formula. What subtle error exists in their setup regarding the choice of $n$ ?

A.They should use

n=4

because the

x^3

term is zero, making the polynomial effectively degree 4. ✅

B.Lagrange error bound cannot be used for cosine.

C.They must use

n=2

because the highest non-zero term is quadratic.

D.The error bound requires

n

to match the number of non-zero terms, not the polynomial degree.

💡 Difficulty: hard | ✅ Correct: A

📖 Explanation: This targets technical precision in error estimation. The polynomial $P_3(x)$ for cosine is indeed $1-x^2/2$ because the cubic term vanishes. However, since the quartic term also exists in the function's expansion and the student included it (or if they consider the next non-zero term), the effective approximation order matters. Strictly, $P_3(x) = P_4(x)$ for cosine. Using $n=3$ gives bound $M|x|^4/4!$ . Using $n=4$ gives $M|x|^5/5!$ . Since $P_3=P_4$ , the tighter bound from $n=4$ is valid and preferable. Failing to exploit the zero term leads to loose error estimates.

Q18. To approximate $\sqrt{2}$ , a student uses the binomial series for $(1+x)^{1/2}$ at $x=1$ . The series converges very slowly. A peer suggests rewriting $\sqrt{2} = \sqrt{1.96 + 0.04} = 1.4(1 + 0.04/1.96)^{1/2}$ and expanding around the new small parameter. What principle justifies this dramatic improvement in convergence speed?

A.Shifting the center of expansion closer to the target value reduces the argument magnitude

|x/a|

, exploiting the geometric nature of binomial convergence. ✅

B.Factoring out 1.4 converts the series into a geometric series with ratio 1.

C.The binomial series only converges for arguments less than 0.05.

D.Rewriting eliminates the square root singularity entirely.

💡 Difficulty: hard | ✅ Correct: A

📖 Explanation: This applies modeling strategies to accelerate convergence. Binomial series $(1+u)^m$ converges for $|u|<1$ , with rate governed by $|u|$ . Original $u=1$ is at boundary (slow/conditional). Transformed $u = 0.04/1.96 \approx 0.02$ . Convergence is now geometric with ratio $\approx 0.02$ , yielding massive speedup. This technique—scaling/shifting to minimize expansion parameter—is fundamental in computational mathematics and perturbation theory. It demonstrates that series representation is flexible, not fixed to standard forms.

Q19. When approximating $f(x) = e^{-x^2}$ using Taylor polynomials, a student notices that for large $x$ , high-degree polynomials oscillate wildly and deviate significantly from the function, even though the series converges everywhere. What explains this counter-intuitive behavior?

A.The series has a finite radius of convergence that was exceeded.

B.Taylor polynomials are local approximations; for fixed degree

n

, error grows as

|x|

increases because the remainder term involves

x^{n+1}

, overwhelming factorial suppression until

n

is sufficiently large relative to

x

. ✅

C.The function

e^{-x^2}

is not analytic for large

x

.

D.Round-off error in coefficient calculation causes instability.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This addresses the distinction between pointwise convergence of the infinite series and uniform approximation by finite polynomials. While $\sum a_k x^k \to f(x)$ for all $x$ , any *fixed* partial sum $P_n(x)$ behaves like a polynomial of degree $n$ at infinity, while $e^{-x^2} \to 0$ . Thus $|f-P_n| \to \infty$ . To maintain accuracy at larger $x$ , $n$ must increase. This highlights that 'convergence' is a limit process $n \to \infty$ , not a property of any single approximant.

Q20. You are evaluating $\int_0^1 \frac{1}{1+x^4} dx$ via series. You expand $\frac{1}{1+x^4} = \sum (-1)^k x^{4k}$ and integrate to get $\sum \frac{(-1)^k}{4k+1}$ . To guarantee error $< 10^{-4}$ , you check terms. If you had instead expanded $\frac{1}{1+x^4}$ as a geometric series in $(1-x^4)$ near $x=1$ , how would the complexity change?

A.It would simplify because powers of

(1-x)

integrate easily.

B.It would become significantly harder because the series would involve powers of

(x-1)

requiring binomial expansion of each term or re-indexing, losing the simple monomial integration structure. ✅

C.The convergence rate would improve dramatically.

D.There is no difference in complexity.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This compares expansion strategies. Expanding in powers of $x$ (Maclaurin) yields simple monomials $x^{4k}$ that integrate to $1/(4k+1)$ . Expanding near $x=1$ involves $(1-(1-x))^4$ or similar, leading to messy coefficients or nested sums. While centering near 1 might help convergence at the upper limit, the loss of simple term structure makes it impractical for hand calculation. This reinforces that 'best center' balances convergence rate against algebraic tractability.

Q21. A student claims that since $\sin(x)$ is odd, its Maclaurin series contains only odd powers, and therefore the error in approximating $\sin(x)$ with $P_3(x) = x - x^3/6$ is exactly $x^5/120$ . Why is this statement incorrect despite the symmetry argument?

A.Symmetry only dictates zero coefficients, not the exact value of the remainder.

B.The error is

R_3(x)

, which equals

f^{(4)}(c)x^4/4!

or

f^{(5)}(c)x^5/5!

depending on formulation; it involves an unknown

c

, not just the next coefficient. ✅

C.Odd functions cannot be approximated by polynomials.

D.The next term is actually

-x^5/120

, so the sign is wrong.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This corrects a common misconception conflating 'next term' with 'exact error'. The series is $x - x^3/6 + x^5/120 - \dots$ . The error of $P_3$ is indeed dominated by $x^5/120$ for small $x$ , but strictly equals $\sin^{(5)}(c) x^5 / 5!$ for some $c \in (0,x)$ . Since $\sin^{(5)}(c) = \cos(c) \neq 1$ generally, the error is not *exactly* the next term. Only for alternating series with specific properties do we have inequalities, not equalities. Precision in language ('bounded by' vs 'equals') is key.

Q22. In computing $\pi$ via Machin's formula $\pi/4 = 4\arctan(1/5) - \arctan(1/239)$ , why is this preferred over $\pi/4 = \arctan(1/2) + \arctan(1/3)$ despite both using valid identities?

A.Machin's formula uses smaller arguments (1/5 and 1/239 vs 1/2 and 1/3), causing the arctangent series to converge geometrically much faster, requiring far fewer terms for equivalent precision. ✅

B.Machin's formula avoids alternating signs, reducing round-off error.

C.The identity

\arctan(1/2)+\arctan(1/3)

is mathematically incorrect.

D.Machin's formula allows parallel computation of terms.

💡 Difficulty: medium | ✅ Correct: A

📖 Explanation: This evaluates algorithmic efficiency in historical/computational context. Arctan series $\sum (-1)^k x^{2k+1}/(2k+1)$ converges geometrically with ratio $x^2$ . For $x=1/2$ , ratio is 0.25. For $x=1/5$ , ratio is 0.04. For $x=1/239$ , ratio is negligible. Machin's formula trades one slowly converging series for two rapidly converging ones. This optimization was crucial for pre-computer era calculations and illustrates how mathematical identities serve computational goals.

Q23. You approximate $f(x) = \ln(1+x)$ using $P_2(x) = x - x^2/2$ . At $x=0.1$ , the actual error is $\approx 3.3 \times 10^{-4}$ . The Lagrange bound with $M=\max|f'''|$ on $[0, 0.1]$ gives $\approx 3.33 \times 10^{-4}$ . At $x=-0.1$ , actual error is $\approx 3.4 \times 10^{-4}$ , but the bound using same $M$ gives $\approx 3.33 \times 10^{-4}$ , which is LESS than actual error. What went wrong?

A.Lagrange bound doesn't work for negative x.

B.The maximum of

|f&#039;&#039;&#039;(x)| = 2/(1+x)^3

on

[-0.1, 0]

occurs at

x=-0.1

, not

x=0

; using

M=f&#039;&#039;&#039;(0)=2

underestimates the true max derivative on the interval. ✅

C.The remainder formula changes sign for negative x.

D.Actual error calculation was incorrect.

💡 Difficulty: hard | ✅ Correct: B

📖 Explanation: This tests careful application of error bounds on intervals containing negative values. $f'''(x) = 2(1+x)^{-3}$ is increasing. On $[0, 0.1]$ , max is at 0.1 (or 0 if decreasing, wait: $(1+x)^{-3}$ decreases as x increases. So max on $[0, 0.1]$ is at $x=0$ , $M=2$ . Correct.). On $[-0.1, 0]$ , $(1+x)^{-3}$ is largest at left endpoint $x=-0.1$ . Value is $2/(0.9)^3 \approx 2.74$ . Using $M=2$ is invalid for the negative interval. Students often blindly use $M$ at center or right endpoint without checking monotonicity over the specific interval.

Q24. A graph shows $y=e^x$ and a polynomial $P(x)$ that touches $e^x$ at $x=0$ and stays below it for all $x \neq 0$ in the viewing window. Another polynomial $Q(x)$ crosses $e^x$ at $x=0$ . Based solely on this geometric behavior, what can be inferred about the degrees and remainders?

A.P is likely an even-degree Taylor polynomial (since

e^x

is convex and even partial sums stay below), while Q is odd-degree (crossing inflection-like behavior). ✅

B.P must be the exact function.

C.Q has a larger radius of convergence.

D.P represents a Padé approximant rather than Taylor.

💡 Difficulty: medium | ✅ Correct: A

📖 Explanation: This links geometry to series parity. $e^x = \sum x^k/k!$ . Even partial sums $P_{2n}$ are always positive and lie below $e^x$ for all real $x$ (property of exponential series). Odd partial sums $P_{2n+1}$ cross $e^x$ at 0 and go to $-\infty$ as $x \to -\infty$ . Observing 'stays below' vs 'crosses' allows identifying parity of approximation without seeing coefficients. This visual diagnostic is valuable for quick validation of series implementations.

Q25. A student models air resistance using $F(v) = kv^2$ . For low velocities, they want a linear approximation but insist on using the Maclaurin series of $v^2$ , getting $F \approx 0$ . Why is this physically useless despite being mathematically correct, and what alternative series approach respects the physics?

A.Maclaurin series fails because

v^2

is not differentiable.

B.The function is already a monomial; its Maclaurin series IS

v^2

. Linearizing at 0 gives 0 because drag vanishes quadratically. To get a useful linear model, one must expand about a non-zero cruising velocity

v_0

, yielding

F \approx kv_0^2 + 2kv_0(v-v_0)

. ✅

C.Linear approximations are never valid for quadratic functions.

D.The student should have used

F=kv

instead.

💡 Difficulty: hard | ✅ Correct: B

📖 Explanation: This integrates modeling sense with series mechanics. Mathematically, $T_1(v^2)$ at 0 is 0. Physically, saying 'drag is zero' near 0 is true but unhelpful for dynamics near operational speed. Taylor series are local. Choosing expansion point $v_0 \neq 0$ adapts the math to the physical regime of interest. This underscores that 'approximation' is contextual; mathematical correctness at a point doesn't imply modeling utility in a region.

Q26. When using the Alternating Series Estimation Theorem for $\sum (-1)^k a_k$ , a student verifies $a_k \to 0$ but forgets to check monotonicity. The series actually has $a_3 > a_2$ . What is the consequence for the error bound $|R_n| \leq a_{n+1}$ if they stop at $n=2$ ?

A.The bound remains valid because convergence is all that matters.

B.The bound may be invalid because the proof relies on nesting intervals

[s_{2k}, s_{2k+1}]

; non-monotonicity breaks this nesting, so

s_3

might not lie between

s_2

and the true sum. ✅

C.The series definitely diverges.

D.The error becomes exactly zero.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This probes the 'why' behind theorem conditions. The alternating series error bound is derived from the fact that partial sums oscillate around the limit with decreasing amplitude. If $a_3 > a_2$ , the oscillation amplitude increases temporarily. $s_3$ could overshoot the limit further than $s_2$ did, placing the true sum outside $[s_2, s_3]$ . Thus $|S-s_2|$ could exceed $a_3$ . Checking monotonicity isn't pedantry; it's structural verification of the error geometry.

Q27. To compute $e$ , a student uses $\sum_{k=0}^n 1/k!$ . They worry about round-off error accumulating. Compared to approximating $\pi$ via Leibniz series, why is round-off less of a concern for $e$ despite both being infinite sums?

A.Computers handle factorials better than pi.

B.The series for

e

has all positive terms and converges super-exponentially, requiring very few terms (~15 for double precision), minimizing accumulation opportunities. Leibniz converges slowly with alternating signs, requiring millions of terms and suffering catastrophic cancellation. ✅

C.e is rational so it terminates.

D.Round-off is identical for both.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: This contrasts numerical stability. Positive, fast-converging series are numerically benign. Slow, alternating series suffer from two issues: massive term count (accumulation) and cancellation (loss of significance when subtracting nearly equal large partial sums). $e$ 's factorial denominator ensures terms drop below machine epsilon quickly. Leibniz's harmonic decay keeps significant terms active for millions of iterations. Understanding this guides choice of algorithm in finite-precision environments.

Q28. You are given $f(x) = \int_0^x \sin(t^2) dt$ (Fresnel integral). No elementary antiderivative exists. You approximate $f(0.5)$ using series. After integrating term-by-term, you get an alternating series. If you need error $< 10^{-8}$ , and the 4th term of the integrated series is $\approx 10^{-9}$ , can you stop?

A.Yes, because it's alternating.

B.Only if you verify the terms are decreasing in magnitude at

x=0.5

. Since

x

is small, they likely are, but verification is mandatory. ✅

C.No, you must always sum 10 terms.

D.Yes, provided the original series converged absolutely.

💡 Difficulty: medium | ✅ Correct: B

📖 Explanation: Reinforces conditional application of estimation theorems. Integrated Fresnel series: $\sum (-1)^k x^{4k+3}/[(2k+1)!(4k+3)]$ . At $x=0.5$ , terms decrease rapidly. But the rule is universal: Alternating Series Test conditions must be verified. Blind application is a common failure mode. The explanation emphasizes that 'alternating' alone is insufficient; monotonic decrease of magnitudes at the specific evaluation point is the operative constraint.

Q29. A student approximates $(1.02)^{10}$ using binomial expansion $(1+x)^n \approx 1+nx$ . Result: 1.2. Actual: ~1.219. Error ~1.5%. They try quadratic: $1+nx+n(n-1)x^2/2$ . Result: 1.218. Much better. Why does adding the quadratic term disproportionately improve accuracy here compared to approximating $e^{0.2}$ ?

A.Binomial coefficients grow initially for

n=10

, making higher terms significant even for small

x

. Exponential coefficients

1/k!

decay immediately. ✅

B.Quadratic terms are always more important for polynomials.

C.The linear approximation for exponentials is inherently worse.

D.Calculator error affected the binomial result.

💡 Difficulty: hard | ✅ Correct: A

📖 Explanation: Compares series behaviors. For $(1+x)^{10}$ , terms are $\binom{10}{k} x^k$ . $\binom{10}{2}=45$ . Term size: $45(0.0004)=0.018$ . Significant! For $e^{0.2}$ , term 2 is $(0.2)^2/2 = 0.02$ . Similar magnitude actually. Wait. Let's re-eval. $e^{0.2} \approx 1+0.2+0.02 = 1.22$ . Actual $e^{0.2}=1.2214$ . Error similar. Perhaps the premise implies $n$ is large? If $n=100, x=0.02$ , linear=3, quad=3+198(0.0004)=3.079. Actual $3^2=9$ ? No $(1.02)^{100} \approx 7.3$ . Linear fails badly. The key is that binomial coefficients $\binom{n}{k}$ can be large, delaying decay, whereas $1/k!$ decays universally. For moderate $n$ , binomial series behaves more like a polynomial with significant higher-order content.

Q30. To evaluate $\int_0^1 x^x dx$ , you use the series $x^x = e^{x \ln x} = \sum (x \ln x)^k / k!$ . Integrating term-by-term requires $\int_0^1 x^k (\ln x)^k dx$ . Why is this valid despite $\ln x$ being singular at 0?

A.It isn't valid; the integral diverges.

B.The singularity is removable/integrable;

x^k (\ln x)^k \to 0

as

x \to 0

for

k \geq 0

, and the series converges uniformly on [0,1], permitting term-wise integration. ✅

C.Logarithms are never singular in series.

D.The integral is improper but the series handles it automatically.

💡 Difficulty: hard | ✅ Correct: B

📖 Explanation: Tests advanced justification of operations. $x \ln x \to 0$ as $x \to 0$ . Function $x^x$ is continuous on [0,1] (defining $0^0=1$ ). Uniform convergence of the power series in $u=x \ln x$ combined with integrability of each term justifies the swap. This connects real analysis (uniform convergence, improper integrals) with series manipulation. Students must recognize that apparent singularities in components may cancel or be integrable, allowing valid series methods where naive inspection suggests failure.

Q31. You approximate $\sec(x)$ by dividing series: $1 / (1 - x^2/2 + x^4/24 - \dots)$ . A student performs polynomial long division and gets $1 + x^2/2 + 5x^4/24$ . They ask: 'Since secant has vertical asymptotes at $\pi/2$ , why does this polynomial not show any blow-up?' How do you explain the limitation?

A.Polynomials cannot represent singularities; this approximation is only valid within the radius of convergence

|x| < \pi/2

. Outside this disk, the polynomial diverges from the function. ✅

B.The student made a division error.

C.Secant doesn't actually have asymptotes.

D.The polynomial approximates the reciprocal, not secant.

💡 Difficulty: medium | ✅ Correct: A

📖 Explanation: Links algebraic construction to analytic limitations. Power series represent analytic functions locally. Polynomials are entire functions (no poles). They cannot capture asymptotic behavior. The series for secant has $R=\pi/2$ due to poles. Within $R$ , polynomial approximates well. Beyond, it fails catastrophically. This reinforces that series approximations are inherently local and blind to global features like singularities outside the convergence disk.

Q32. A student uses $\sum_{n=0}^\infty \frac{(-1)^n}{2n+1}$ to approximate $\pi/4$ . After 1000 terms, error is still ~0.0005. They accelerate by Euler transform or Shanks transformation. Conceptually, what do these transformations exploit that raw summation misses?

A.They predict the limit by analyzing the sequence of partial sums' asymptotic behavior, effectively extrapolating past the slow harmonic convergence. ✅

B.They add more terms faster.

C.They convert the series to an integral.

D.They remove the alternating sign.

💡 Difficulty: hard | ✅ Correct: A

📖 Explanation: Introduces advanced convergence acceleration concepts. Raw summation treats each term additively. Acceleration methods recognize that partial sums $S_n$ approach $S$ with a predictable error structure (e.g., $S_n \approx S + c/n$ ). By fitting this structure, they extrapolate to $n \to \infty$ using finite data. This shifts paradigm from 'adding terms' to 'modeling convergence'. Essential for efficient computation of slowly converging constants.

Q33. In approximating $f(x) = \sqrt{x}$ near $x=1$ , you use Taylor series. A colleague suggests using $\sqrt{x} = \sqrt{1+(x-1)}$ and binomial expansion. Are these different? If you truncate at quadratic term, do both yield identical polynomials?

A.They are fundamentally different methods yielding different results.

B.They are identical; binomial expansion of

(1+u)^{1/2}

IS the Taylor series of

\sqrt{1+u}

at 0. Truncations match exactly. ✅

C.Binomial is only for integer exponents.

D.Taylor series includes a remainder term that binomial lacks.

💡 Difficulty: easy | ✅ Correct: B

📖 Explanation: Reinforces unity of methods. Binomial series for non-integer exponent is precisely the Maclaurin series derived via derivatives. Coefficients $\binom{1/2}{k}$ equal $f^{(k)}(0)/k!$ . Students often compartmentalize 'binomial theorem' and 'Taylor series' as separate tools. Recognizing their identity simplifies the toolkit and validates using whichever form is more convenient.

Q34. You approximate $\sin(0.1)$ using $x - x^3/6$ . Error bound via Lagrange with $n=3$ uses $M=|\sin^{(4)}(c)| = |\sin c| \leq \sin(0.1) \approx 0.1$ . Bound: $0.1(0.1)^4/24 \approx 4 \times 10^{-8}$ . Actual error $\approx 8 \times 10^{-11}$ . Why is the bound so loose?

A.Lagrange bounds are worst-case; actual derivative

\sin(c)

for

c \in (0, 0.1)

is much smaller than max possible, and the alternating series bound

x^5/120 \approx 8 \times 10^{-11}

is tighter and exact for this series. ✅

B.Calculation error in bound.

C.Sine function is special.

D.Lagrange theorem is invalid for small x.

💡 Difficulty: medium | ✅ Correct: A

📖 Explanation: Contrasts general vs specific error tools. Lagrange uses global max of derivative on interval. For sine near 0, $f^{(4)}(c)=\sin c$ is tiny, but if we didn't know that and used $M=1$ (safe global bound), it would be looser. Even with $M=\sin(0.1)$ , it's an upper bound. Alternating series bound exploits specific series structure (next term) which is often sharper for alternating series. Choosing the right error tool for the series type improves estimates.

Q35. A model requires $\cos(x)$ for $x \in [0, \pi]$ . Maclaurin series needs high degree for accuracy at $\pi$ . You switch to Taylor series centered at $\pi/2$ . How does this affect required degree for uniform accuracy?

A.Centering at midpoint

\pi/2

halves the maximum distance to endpoints, drastically reducing required degree since error scales as

|x-x_0|^{n+1}

. ✅

B.Centering at

\pi/2

makes no difference.

C.Maclaurin is always optimal.

D.Taylor at

\pi/2

requires more terms because derivatives are harder.

💡 Difficulty: medium | ✅ Correct: A

📖 Explanation: Applies optimization of expansion center. Max error depends on $\max|x-x_0|$ . Maclaurin: max dist = $\pi \approx 3.14$ . Midpoint: max dist = $\pi/2 \approx 1.57$ . Since error $\propto d^{n+1}$ , halving distance reduces error exponentially for fixed $n$ , or allows lower $n$ for fixed error. Strategic center selection is key for efficient approximation over intervals.

📝 Taylor series for pi (35 MCQs)

What is Taylor series for pi?

📝 All Taylor series for pi MCQs

🔗 Related Topics (MCQs)