📝 Euler's method differential equations (38 MCQs)

📖 Explanation: Euler's method is strictly a first-order Taylor approximation. Simply adding the second derivative term requires computing y'' = f_x + f_y f via the chain rule. If a student adds this term without performing this specific differentiation, they are not using a valid numerical scheme for the given ODE. This distinguishes Euler's method from higher-order Taylor series methods, which require analytical derivation of higher derivatives, whereas Runge-Kutta methods achieve higher order without explicit derivative evaluation.

📖 Explanation: For the test equation y&#039; = \lambda y, Euler's method yields $y_{n+1} = (1+h\lambda)y_n$. Stability requires the amplification factor $|1+h\lambda| \leq 1$. With $\lambda = -100$ and $h=0.05$, the factor is $|1 - 5| = 4$, causing exponential growth of errors. This demonstrates that for stiff equations, step size is limited by stability constraints, not just accuracy requirements, making explicit Euler unsuitable unless $h < 2/|\lambda|$.

📖 Explanation: At the specific point $(0,2)$, the slope field indicates horizontal tangents, meaning $f(0,2) = 0$. Euler's formula $y_1 = y_0 + h f(x_0, y_0)$ evaluates to $2 + h(0) = 2$. Even though nearby points might have steep or undefined slopes, Euler's method only samples the derivative at the current node. Therefore, the first step remains stationary at $y=2$, potentially masking singularities or rapid changes immediately adjacent to the starting point.

📖 Explanation: The logistic equation has stable equilibrium at $P=L$. Euler's method approximates the curve with tangent lines. If $h$ is too large, the tangent line at a point near $L$ may project significantly above $L$. In the next step, P&#039; becomes negative, and a large $h$ can project the value below zero. This is a geometric failure of linear approximation violating the invariant region $[0, L]$, highlighting the need for adaptive step sizes or positivity-preserving integrators in biological modeling.

📖 Explanation: Standard Euler matches the Taylor series up to $O(h)$, resulting in local truncation error $O(h^2)$ and global error $O(h)$. Improved Euler uses a predictor-corrector approach equivalent to matching the Taylor series up to $O(h^2)$, yielding local error $O(h^3)$ and global error $O(h^2)$. The averaging of the slope at the beginning and predicted end of the interval effectively captures the curvature of the solution, eliminating the dominant linear error component that plagues basic Euler.

📖 Explanation: Euler's method follows the tangent line. If the solution curve is concave up (y&#039;&#039; > 0), the tangent line lies below the curve, making the Euler approximation an underestimate. Differentiating y&#039; = x^2 + y^2 gives y&#039;&#039; = 2x + 2y(x^2+y^2). For $x \geq 0, y \geq 0$, y&#039;&#039; \geq 0. Thus, the true solution curves upward away from the tangent, confirming $y_{Euler} < y_{exact}$. This connects calculus concepts of concavity directly to numerical error analysis.

📖 Explanation: Theoretical global truncation error is $O(h)$, so halving $h$ should halve the error asymptotically. However, total error = truncation error + round-off error. Round-off error accumulates as $O(\epsilon/h)$ where $\epsilon$ is machine precision. As $h \to 0$, round-off grows. If $h$ is already small, halving it may double the round-off contribution, negating the reduction in truncation error. This highlights the practical limitation of numerical methods versus theoretical convergence rates.

📖 Explanation: The exact flow of Hamiltonian systems preserves the symplectic form and energy $E = (y^2+z^2)/2$. Explicit Euler is not a symplectic integrator; it introduces artificial dissipation or excitation depending on formulation. For y&#039;=z, z&#039;=-y, Euler produces $y_{n+1}^2 + z_{n+1}^2 = (1+h^2)(y_n^2+z_n^2)$, causing energy to grow as $(1+h^2)^n$. This spiraling outward violates energy conservation, demonstrating why geometric integrators are preferred for oscillatory systems over long time intervals.

📖 Explanation: The factor $e^{L(b-a)}$ arises from Gronwall's inequality. For stiff problems with large $L$, this factor is enormous, making the theoretical bound useless for predicting actual error or choosing $h$. While mathematically correct, it fails to capture the asymptotic behavior where error depends on the smoothness of the solution rather than just the Lipschitz constant. Practitioners must rely on local error estimation and adaptive stepping rather than global a priori bounds for stiff systems.

📖 Explanation: Picard-Lindelöf theorem guarantees uniqueness only if $f$ is Lipschitz. $\sqrt{|y|}$ has infinite slope at $y=0$, allowing multiple solutions. Euler's method, starting exactly at the singularity with zero slope, stays trapped at the equilibrium $y=0$. Any perturbation or different discretization might trigger the parabolic branch. This illustrates that numerical methods assume well-posedness; when uniqueness fails, the algorithm selects one solution arbitrarily based on initialization and arithmetic.

📖 Explanation: Clipping (Option A) destroys accuracy and conservation. Implicit Euler (B) preserves positivity for certain classes but requires solving nonlinear equations. Option C, often called the Patankar trick or exponential Euler, inherently maintains positivity because the exponential is always positive, and it reduces to standard Euler as $h \to 0$. This is crucial in chemistry/biology where negative concentrations are unphysical. It modifies the discrete map to respect the invariant set $C \geq 0$ structurally.

📖 Explanation: Expanding $e^{\lambda h} = 1 + \lambda h + (\lambda h)^2/2 + \dots$. Euler's amplification factor is exactly $1+\lambda h$. Thus, Euler's method matches the exact amplification only to first order. The discrepancy $O(h^2)$ per step accumulates to global $O(h)$ error. This comparison directly links the algebraic stability function of the numerical method to the analytic properties of the exponential function, explaining the origin of truncation error in linear problems.

📖 Explanation: The exact solution $y = 1/(1-x)$ has a vertical asymptote at $x=1$. Euler's method, being a linear projection, steps across the singularity without detecting the blow-up, producing finite but completely erroneous values for $x>1$. This is a critical failure mode: numerical methods do not automatically detect domain boundaries or singularities. Users must independently verify solution existence; blind trust in the output leads to catastrophic misinterpretation of physical models.

📖 Explanation: Adaptive control targets the local truncation error per step, which for Euler scales as h^2 y&#039;&#039;/2. To achieve a target local error $\tau$, we solve $C h_{new}^2 = \tau \Rightarrow h_{new} \propto \sqrt{\tau}$. If we were controlling global error (which scales as $h$), the exponent would be 1. Confusing local vs. global error scaling is a common mistake. The square root arises directly from the Taylor remainder term governing the single-step deviation.

📖 Explanation: Euler's method applied to y&#039; = -y + g(x) acts as a low-pass filter with gain $|1/(1+h)|$ for the homogeneous part and modified response for forcing. For $h>0$, the numerical damping is stronger than exact damping $e^{-h} \approx 1-h+h^2/2$. Since $1-h < e^{-h}$, Euler dissipates energy faster, reducing the amplitude of the forced response. This artificial viscosity distorts long-term dynamics, making Euler poor for capturing accurate steady-state amplitudes even when stable.

📖 Explanation: The definition of Euler's method $y_{n+1} = y_n + h f(x_n, y_n)$ can be rearranged as $(y_{n+1} - y_n)/h = f(x_n, y_n)$. This is precisely the forward difference approximation of the derivative at $x_n$. This connection explains why Euler is first-order accurate: the forward difference has truncation error $O(h)$. Understanding this link helps students see numerical ODE solvers as discrete analogues of differential operators.

📖 Explanation: Asymptotic $O(h)$ convergence holds only as $h \to 0$. For moderate $h$, the error expansion is $E = C_1 h + C_2 h^2 + \dots$. If $C_1 h \ll C_2 h^2$ initially, the error behaves quadratically. Only when $h$ is sufficiently small does the linear term dominate. Observing quadratic reduction suggests the solver hasn't reached the asymptotic regime yet. This warns against estimating order from coarse grids; true order verification requires progressively finer steps until the ratio stabilizes.

📖 Explanation: The stability condition is $|1 + z| < 1$ where $z = h\lambda$. This inequality describes the interior of a circle of radius 1 centered at $z = -1$ in the complex plane. For purely imaginary eigenvalues ($\alpha=0$), $z$ lies on the imaginary axis, which never enters this disk (except at origin). Thus, explicit Euler is unconditionally unstable for undamped oscillators. This geometric insight explains why Euler fails for wave equations and orbital mechanics regardless of step size.

📖 Explanation: Lotka-Volterra systems are conservative with a first integral. The exact flow is area-preserving (divergence-free). Explicit Euler expands area: the Jacobian determinant of the map $I + hJ$ is $1 + h \text{tr}(J) + O(h^2)$. Even if $\text{tr}(J)=0$, higher-order terms cause area change. This artificial source/sink of phase volume drives the spiral. Symplectic or volume-preserving integrators are required to maintain qualitative correctness. Euler's failure here is structural, not just quantitative.

📖 Explanation: Using $x_{n+1}$ with $y_n$ creates an inconsistent scheme. Taylor expanding $f(x_n+h, y_n) = f + h f_x + \dots$ shows this differs from standard Euler by $O(h^2)$ locally, but since the base method is $O(h)$, it doesn't raise the order. It merely changes the error constant. Crucially, it doesn't gain the stability benefits of true implicit methods because $y_{n+1}$ isn't used in $f$. This is a common coding bug that produces plausible but incorrect results.

📖 Explanation: Unconditional stability means stable for all $h > 0$. Implicit Euler has this property for $\text{Re}(\lambda) < 0$. Explicit Euler requires $|1 - h\lambda| < 1 \Rightarrow h < 2/\lambda$. If $h$ exceeds this, numerical solutions grow despite physical decay. This conditional nature forces small steps for fast-decaying modes (stiffness), making explicit Euler inefficient for multiscale problems. The distinction is fundamental in selecting solvers for transient simulations.

📖 Explanation: The factor $e^{LT}$ is the Gronwall constant. It quantifies how much an error introduced at any point can grow by the final time due to the system's dynamics. For chaotic or stiff systems with large $L$, this factor is huge, meaning tiny local errors explode. This explains why high precision per step doesn't guarantee global accuracy for sensitive systems. It links numerical analysis to dynamical systems theory: error bounds depend on intrinsic system properties, not just algorithm parameters.

📖 Explanation: Standard Euler adds energy to Hamiltonian systems. Backward Euler removes energy. Symplectic Euler (updating position with new velocity or vice versa) preserves the symplectic structure, bounding energy error over exponentially long times despite being only first-order. RK4 conserves energy better short-term but drifts long-term. For orbital mechanics, preserving geometric structure is more important than local order. Symplectic Euler is the minimal fix for Euler's energy drift in conservative systems.

📖 Explanation: Numerical ODE theory assumes sufficient smoothness. At discontinuities, derivatives are undefined, and Taylor-based error estimates fail. The rigorous approach is event detection: locate the discontinuity exactly, stop integration, apply transition conditions, and restart. Smoothing introduces modeling error; tiny steps don't fix undefined derivatives; implicit methods still struggle with non-uniqueness. Domain splitting respects the mathematical structure. This highlights that numerical methods are tools for well-posed problems; preprocessing is often required for real-world nonsmooth models.

📖 Explanation: Richardson extrapolation relies on the asymptotic error expansion where coefficients $C_k$ depend on the solution and its derivatives but NOT on $h$. If $h$ is too large, higher-order terms contaminate the estimate, or $C_1$ varies effectively with $h$. The formula $2y_{h/2} - y_h$ cancels the $O(h)$ term only if $C_1$ is constant between runs. This technique boosts Euler to second-order accuracy post-hoc but fails outside the asymptotic regime, requiring validation.

📖 Explanation: Deterministic Euler assumes y(t+h) \approx y(t) + h y&#039;(t). In SDEs, the Wiener increment $\Delta W$ scales as $\sqrt{h}$, not $h$. The Ito-Taylor expansion includes terms of order $\sqrt{h}$. Standard Euler-Maruyama retains only drift $h$ and diffusion $\Delta W$, missing mixed terms. Since the leading neglected term involves $\Delta W \cdot h$ or similar, the strong error is dominated by the roughness of Brownian paths, limiting convergence to 0.5. This contrasts sharply with smooth ODE theory.

📖 Explanation: Truncation error $\sim 0.1$ is indeed vastly larger than double precision $\sim 10^{-16}$. For well-conditioned problems, extra digits are wasted. However, if the problem is ill-conditioned (large Lipschitz constant or long integration interval), round-off can be amplified by $e^{LT}$. If $e^{LT} > 10^{15}$, double precision matters even at coarse $h$. Thus, the student is usually right but dangerously wrong for sensitive systems. Safe practice considers both truncation and conditioning.

📖 Explanation: This is another non-Lipschitz example. The zero solution is technically stable but not isolated. Numerical round-off or intentional perturbation acts as a selector among the infinite family of solutions branching from the origin. Euler's method with exact zero arithmetic stays at zero, but real computers introduce noise, pushing the state onto the non-zero manifold. This demonstrates that numerical solutions to non-unique problems are artifacts of implementation details, not unique mathematical truths.

📖 Explanation: Time-stepping is serial. Multiple shooting divides $[0,T]$ into subintervals, guesses initial values at interfaces, and integrates each segment in parallel using Euler. A Newton iteration updates interface values to enforce continuity. This breaks the temporal dependency. Spatial decomposition applies to PDEs, not ODEs. Pipelining doesn't help since step $n+1$ needs result of $n$. Multiple shooting is the standard way to parallelize IVPs, trading communication for concurrency.

📖 Explanation: Linear stability analyzes behavior near equilibrium $y=0$. For y&#039;=-y^3, linearization gives y&#039;=0 (marginally stable). But nonlinear term provides strong damping for large $y$. If Euler overshoots, $y^3$ becomes huge, pulling it back violently. This nonlinear restoring force extends the effective stability region beyond linear predictions. Conversely, some nonlinearities shrink stability regions. Linear analysis is necessary but not sufficient for nonlinear stability; energy methods or Lyapunov functions give better bounds.

📖 Explanation: To reduce error by factor $K$ with Euler, steps increase by $K$, costing $K$ evaluations. With RK4 (order 4), steps increase by $K^{1/4}$, costing $4K^{1/4}$ evaluations. For $K=1000$, Euler costs 1000x, RK4 costs ~22x. Despite higher cost per step, higher-order methods are vastly more efficient for smooth problems. Only when $f$ is trivial or discontinuous might Euler compete. Efficiency = Accuracy/Cost, favoring higher order for expensive $f$.

📖 Explanation: Defective matrices have Jordan blocks. Exact solution involves polynomials times exponentials. Numerical methods approximate $e^{Ah}$. For Jordan blocks, the numerical approximation of the nilpotent part introduces additional errors that couple with the exponential. The error expansion includes terms arising from the non-commutativity or defectiveness, often leading to worse constants or transient growth even if eigenvalues suggest stability. This subtle linear algebra issue impacts numerical performance, showing that spectral analysis alone is insufficient for error prediction.

📖 Explanation: IMEX splits $f = f_{fast} + f_{slow}$. Treating $f_{fast}$ implicitly removes its stability constraint. But if coupling terms are explicit, they inherit the stiffness of the fastest scale they interact with. If coupling transfers energy from fast to slow modes explicitly, the step size is limited by the fast frequency to maintain stability. Proper IMEX requires implicit treatment of all stiff components including coupling. Partial explicit treatment defeats the purpose, illustrating the complexity of multi-scale integration.

📖 Explanation: Since $|R| > 1$ for all $h > 0$, Euler unconditionally amplifies oscillatory modes. Energy increases every step. This makes explicit Euler fundamentally unsuitable for wave propagation, acoustics, or quantum mechanics. Even infinitesimal $h$ causes eventual blow-up. This contrasts with dissipative systems where small $h$ ensures stability. Recognizing this unconditional instability for imaginary eigenvalues is a critical screening test for solver selection in hyperbolic problems.

📖 Explanation: Backward Euler defines a specific implicit relation. Solving it inexactly introduces an additional error. If Newton converges to a tolerance $\tau$, the overall error is sum of truncation + algebraic error. Poor Jacobian affects solver efficiency/robustness, not the definition of the scheme. However, if iterations stop early, the effective method is neither backward Euler nor anything well-defined. Crucially, A-stability is a property of the exact implicit map; approximate solution doesn't change the theoretical stability region, but practical instability can arise from solver divergence.

📖 Explanation: Stability ensures boundedness, not fidelity. For $h=0.01, \lambda=-1000$, $h\lambda = -10$. Amplification $|1-10|=9 > 1$. Wait, $h=0.01$ is UNSTABLE for $\lambda=-1000$. Need $h < 0.002$. Assume question implies $h$ is within stability limit, say $h=0.001$. Then $h\lambda = -1$. Stable. But tracking $\sin(x)$ requires resolving frequency $\omega=1$. $h=0.001$ resolves it well. Perhaps the point is: even if stable, Euler has $O(h)$ error. For stiff problems, accuracy requires $h$ small enough for the FAST scale during transient, but once on slow manifold, larger $h$ could suffice. Fixed $h$ chosen for stability is wasteful but accurate. If $h$ is chosen ONLY for stability (e.g., $h=0.0019$), error on slow manifold might still be acceptable. The key concept: Stability \neq Accuracy. One can be stable but inaccurate if $h$ is near stability boundary but too large for desired precision on the slow dynamics.