2026 National Graduate Entrance Examination Mathematics I Past Paper

I. Multiple Choice

I. Multiple choice: questions 1 to 10, 5 points each, 50 points in total.

1. Partial derivatives of an implicit function

Let z=z(x,y)z=z(x,y) be determined by x−az=ey+azx-az=e^{y+az}, where a≠0a\ne0. Find the linear combination of zxz_x and zyz_y.

implicit differentiationpartial derivatives

Approach:

This question tests taking partial derivatives of an implicit function. The key is to differentiate the implicit relation with respect to xx and yy in turn, and then eliminate the part containing the exponential term by subtracting.

Steps:

  1. The implicit relation is x−az=ey+azx-az=e^{y+az}; differentiate both sides with respect to xx (treating zz as a function of x,yx,y): 1−azx=ey+az⋅azx1-a z_x=e^{y+az}\cdot a z_x.
  2. Rearranging gives (1+ey+az)azx=1(1+e^{y+az})a z_x=1, and hence zx=1a(1+ey+az)z_x=\dfrac{1}{a(1+e^{y+az})}.
  3. Differentiate both sides with respect to yy: −azy=ey+az(1+azy)-a z_y=e^{y+az}(1+a z_y); rearranging gives a(1+ey+az)zy=−ey+aza(1+e^{y+az})z_y=-e^{y+az}, and hence zy=−ey+aza(1+ey+az)z_y=-\dfrac{e^{y+az}}{a(1+e^{y+az})}.
  4. Subtracting the two: zx−zy=1+ey+aza(1+ey+az)=1az_x-z_y=\dfrac{1+e^{y+az}}{a(1+e^{y+az})}=\dfrac1a, so the answer is (A).

Answer:

A, that is, zx−zy=1az_x-z_y=\dfrac1a.

Differentiating with respect to xx and yy in turn gives

zx=1a(1+ey+az),zy=−ey+aza(1+ey+az).z_x=\frac{1}{a(1+e^{y+az})},\qquad z_y=-\frac{e^{y+az}}{a(1+e^{y+az})}.

Subtracting the two makes the exponential term cancel.

2. The interval of convergence of a power series

Determine the interval of convergence of

∑n=1∞(3+(−1)n4)nx2n\sum_{n=1}^{\infty}\left(\frac{3+(-1)^n}{4}\right)^n x^{2n}
power seriesinterval of convergence

Approach:

This question tests the interval of convergence of a power series. First make a substitution to turn the power series into a one-variable power series in a new variable, then use the ratio test to find the radius of convergence and discuss the endpoints separately.

Steps:

  1. Put u=x2u=x^2, and the series becomes ∑n=1∞(3+(−1)n4)nun\sum\limits_{n=1}^{\infty}\left(\dfrac{3+(-1)^n}{4}\right)^n u^n.
  2. The root limit of the coefficients an=(3+(−1)n4)na_n=\left(\dfrac{3+(-1)^n}{4}\right)^n is lim⁡n→∞∣an∣n=lim⁡n→∞3+(−1)n4\lim\limits_{n\to\infty}\sqrt[n]{|a_n|}=\lim\limits_{n\to\infty}\dfrac{3+(-1)^n}{4}; this limit does not exist, but its upper limit is 11, so the radius of convergence in uu is Ru=1R_u=1.
  3. That is, the series converges for ∣u∣<1|u|<1 and diverges for ∣u∣>1|u|>1.
  4. Since u=x2≥0u=x^2\ge0, it converges for ∣x∣<1|x|<1; at x=±1x=\pm1 we have u=1u=1, where the coefficients are 11 (for odd nn) and (12)n\left(\dfrac12\right)^n (for even nn), and the terms do not tend to zero, so the series diverges.
  5. Hence the interval of convergence is (−1,1)(-1,1), and the answer is (D).

Answer:

D, (−1,1)(-1,1).

Put u=x2u=x^2. The root limit of the even-indexed coefficients is 11, so the radius of convergence in uu is 11; it converges for ∣x∣<1|x|<1, while at x=±1x=\pm1 the even-indexed terms do not tend to zero.

3. Concave functions and the slope of a secant

When the function is concave on [−1,1][-1,1], determine the monotonicity of the secant slope

f(x)−f(1)x−1\frac{f(x)-f(1)}{x-1}

with the right endpoint 11 fixed, and its converse.

concave functionsecant slope

Approach:

This question tests the properties of a concave function and the monotonicity of the secant slope. For a concave function the secant slope varies monotonically as the other endpoint moves; but the converse does not follow — knowing only that one family of secant slopes is monotone does not imply concavity.

Steps:

  1. A concave function (the concavity inequality f(tx1+(1−t)x2)≥tf(x1)+(1−t)f(x2)f(tx_1+(1-t)x_2)\ge tf(x_1)+(1-t)f(x_2)) has its graph below every secant.
  2. With the right endpoint 11 fixed, as the left endpoint xx increases the secant slope f(x)−f(1)x−1\dfrac{f(x)-f(1)}{x-1} increases monotonically.
  3. The reverse reasoning does not hold: knowing only that the slopes of the secants with a fixed right endpoint are monotone does not guarantee that the function is concave on every interval, so the converse is false.
  4. So the correct statement is “the secant slope increases as the left endpoint increases, and its converse does not hold” — choose (C).

Answer:

C.

For a concave function the secant slope increases as the left endpoint xx increases; knowing only that one family of fixed-right-endpoint secant slopes is monotone does not give back concavity of the function as a whole.

4. The triple integral bounded by a sphere and a cone

The region is bounded by z=4−x2−y2z=\sqrt{4-x^2-y^2} and z=x2+y2z=\sqrt{x^2+y^2}; determine the correct form of the spherical-coordinate integral for ∭Ωf(x2+y2+z2) dV\iiint_\Omega f(x^2+y^2+z^2)\,dV.

triple integralspherical coordinates

Approach:

This question tests the spherical-coordinate transformation of a triple integral. The key is to write the sphere and the cone as constraints on ρ\rho and φ\varphi respectively, and then to write out the volume element.

Steps:

  1. The sphere z=4−x2−y2z=\sqrt{4-x^2-y^2} is x2+y2+z2=4x^2+y^2+z^2=4, which in spherical coordinates gives ρ=2\rho=2.
  2. The cone z=x2+y2z=\sqrt{x^2+y^2} is z=rz=r, and in spherical coordinates ρcos⁡φ=ρsin⁡φ\rho\cos\varphi=\rho\sin\varphi, so φ=π4\varphi=\dfrac\pi4.
  3. The region lies inside the sphere and above the cone (close to the zz-axis), so 0≤ρ≤20\le\rho\le2, 0≤φ≤π40\le\varphi\le\dfrac\pi4, 0≤θ≤2π0\le\theta\le2\pi.
  4. The spherical volume element is dV=ρ2sin⁡φ dρ dφ dθdV=\rho^2\sin\varphi\,d\rho\,d\varphi\,d\theta.
  5. Hence ∭Ωf(x2+y2+z2)dV=∫02πdθ∫0π/4sin⁡φ dφ∫02f(ρ2)ρ2 dρ\displaystyle\iiint_\Omega f(x^2+y^2+z^2)dV=\int_0^{2\pi}d\theta\int_0^{\pi/4}\sin\varphi\,d\varphi\int_0^2f(\rho^2)\rho^2\,d\rho, which is (C).

Answer:

C.

The sphere gives ρ=2\rho=2 and the cone gives φ=π/4\varphi=\pi/4; the region corresponds to 0≤ρ≤20\le\rho\le2, 0≤φ≤π/40\le\varphi\le\pi/4, 0≤θ≤2π0\le\theta\le2\pi, and the volume element is ρ2sin⁡φ dρ dφ dθ\rho^2\sin\varphi\,d\rho\,d\varphi\,d\theta.

5. Permutation matrices

Let AA be a permutation matrix obtained from the identity matrix by swapping several rows. Determine the properties of its adjoint matrix and its inverse.

permutation matrixinverse matrix

Approach:

This question tests the properties of a permutation matrix. A permutation matrix satisfies AT=A−1A^T=A^{-1}, so its inverse is again a permutation matrix; the adjoint matrix, however, is affected by the sign of the determinant.

Steps:

  1. A permutation matrix is obtained from the identity matrix by swapping several rows, and satisfies AAT=ATA=EAA^T=A^TA=E, that is, A−1=ATA^{-1}=A^T.
  2. ATA^T is still a permutation matrix (the rows of AA rearranged in the same way), so A−1A^{-1} is still a permutation matrix.
  3. The adjoint matrix is A∗=sgn⁡(A)A−1A^*=\operatorname{sgn}(A)A^{-1}; when sgn⁡(A)=−1\operatorname{sgn}(A)=-1 (that is, AA is an odd permutation) the entries of A∗=−A−1A^*=-A^{-1} are no longer all 0,10,1, so one cannot assert unconditionally that A∗A^* is a permutation matrix.
  4. The correct statement is therefore that ”A−1A^{-1} is still a permutation matrix” — choose (B).

Answer:

B, A−1A^{-1} is still a permutation matrix.

The inverse of a permutation matrix is its transpose, which is still a permutation matrix; the adjoint matrix is also affected by the sign of the determinant, and cannot be declared a permutation matrix without qualification.

6. A containment relation between column spaces

The column vectors of a 3×3 matrix AA can be expressed as linear combinations of the column vectors of a matrix BB. Determine the relation between the solvability of Ax=βAx=\beta and Bx=βBx=\beta.

matrixsystem of linear equations

Approach:

This question tests the column space of a matrix and the solvability of a system of linear equations. The system has a solution if and only if the right-hand side vector belongs to the column space of the coefficient matrix.

Steps:

  1. “The column vectors of AA can be expressed as linear combinations of the column vectors of BB” is equivalent to Col⁡(A)⊆Col⁡(B)\operatorname{Col}(A)\subseteq\operatorname{Col}(B).
  2. If Ax=βAx=\beta has a solution, then β\beta is a linear combination of the columns of AA, that is, β∈Col⁡(A)\beta\in\operatorname{Col}(A).
  3. From Col⁡(A)⊆Col⁡(B)\operatorname{Col}(A)\subseteq\operatorname{Col}(B) it follows that β∈Col⁡(B)\beta\in\operatorname{Col}(B), so Bx=βBx=\beta has a solution as well.
  4. The converse does not hold (a solution of Bx=βBx=\beta does not imply a solution of Ax=βAx=\beta), so choose (A).

Answer:

A.

The condition is equivalent to Col⁡(A)⊆Col⁡(B)\operatorname{Col}(A)\subseteq\operatorname{Col}(B). If Ax=βAx=\beta has a solution then β∈Col⁡(A)\beta\in\operatorname{Col}(A), hence β∈Col⁡(B)\beta\in\operatorname{Col}(B), so Bx=βBx=\beta has a solution.

7. A quadratic form representing a cylinder

The diagonal coefficients of a quadratic form are all aa and the cross-term coefficients are all 44. If the equation f(x1,x2,x3)=−1f(x_1,x_2,x_3)=-1 represents a cylinder, determine aa and the orthogonal standard form.

quadratic formorthogonal transformation

Approach:

This question tests the orthogonal standard form of a quadratic form and the type of surface it represents. The equation f=−1f=-1 representing a cylinder means the matrix of the quadratic form has a zero eigenvalue, which fixes the parameter aa.

Steps:

  1. The matrix of the quadratic form is A=(a222a222a)A=\begin{pmatrix}a&2&2\\2&a&2\\2&2&a\end{pmatrix} (the diagonal entries are aa and the cross-term coefficients are 44).
  2. A=(a−2)E+2JA=(a-2)E+2J (where JJ is the all-ones matrix), so the eigenvalues are a+4a+4 (corresponding to 1=(1,1,1)T\bm{1}=(1,1,1)^T) and a−2a-2 (twice, corresponding to the subspace orthogonal to 1\bm{1}).
  3. The equation f(x1,x2,x3)=−1f(x_1,x_2,x_3)=-1 representing a cylinder means the matrix of the quadratic form has a zero eigenvalue, that is, a+4=0a+4=0 or a−2=0a-2=0.
  4. If a=2a=2, then f=4(x1+x2+x3)2f=4(x_1+x_2+x_3)^2, which represents a plane rather than a cylinder; so take a=−4a=-4.
  5. The eigenvalues are then 0,−6,−60,-6,-6, the orthogonal standard form is −6y12−6y22-6y_1^2-6y_2^2, and the equation becomes y12+y22=16y_1^2+y_2^2=\dfrac16, which is indeed a circular cylinder — choose (B).

Answer:

B, a=−4a=-4, and the orthogonal standard form is −6y12−6y22-6y_1^2-6y_2^2.

The eigenvalues of the matrix of the quadratic form are a+4,a−2,a−2a+4,a-2,a-2. A cylinder requires one zero eigenvalue, and with a=−4a=-4 the eigenvalues are 0,−6,−60,-6,-6 and the equation becomes y12+y22=1/6y_1^2+y_2^2=1/6.

8. A mean square error function

Let X∼N(1,2)X\sim N(1,2) and f(t)=E[(X+t)2]f(t)=E[(X+t)^2]. Find the minimiser of f(t)f(t) and the minimum value.

normal distributionmathematical expectation

Approach:

This question tests the arithmetic of expectation and the mean square error. Expand (X+t)2(X+t)^2, simplify it with the linearity of expectation and the definition of the variance, and then complete the square to find the minimum.

Steps:

  1. f(t)=E(X+t)2=E(X2)+2tE(X)+t2=D(X)+E(X)2+2tE(X)+t2f(t)=E(X+t)^2=E(X^2)+2tE(X)+t^2=D(X)+E(X)^2+2tE(X)+t^2.
  2. Substituting E(X)=1, D(X)=2E(X)=1,\ D(X)=2 gives f(t)=2+(1+t)2f(t)=2+(1+t)^2.
  3. Since (1+t)2≥0(1+t)^2\ge0, f(t)f(t) attains its minimum at t=−1t=-1, and the minimum value is 22.
  4. So choose (C).

Answer:

C, the minimiser is t=−1t=-1 and the minimum value is 22.

f(t)=Var⁡(X)+(EX+t)2=2+(1+t)2f(t)=\operatorname{Var}(X)+(EX+t)^2=2+(1+t)^2.

9. A transformation of distribution functions

The distribution function of XX is F(x)F(x) with mean μ\mu and variance σ2\sigma^2; the distribution function of YY is F(ay+b)F(ay+b), and EY=0, DY=1EY=0,\ DY=1. Find aa and bb.

distribution functiontransformation of a random variable

Approach:

This question tests determining the relationship between two random variables from their distribution functions, and then recovering the coefficients of the transformation from the moments.

Steps:

  1. The distribution function of YY is F(ay+b)F(ay+b); comparing it with the distribution function FF of XX shows that YY has the same distribution as aX+baX+b, that is, XX and YY satisfy X=aY+bX=aY+b.
  2. By the linearity of expectation: EX=a EY+bEX=a\,EY+b, that is, μ=b\mu=b.
  3. By the property of the variance: DX=a2DYDX=a^2DY, that is, σ2=a2\sigma^2=a^2, so ∣a∣=σ|a|=\sigma.
  4. Taking a>0a>0 as the options do, a=σ, b=μa=\sigma,\ b=\mu — choose (A).

Answer:

A, a=σ,b=μa=\sigma,b=\mu.

The relation between the distribution functions gives X=aY+bX=aY+b. Hence μ=b\mu=b and σ2=a2\sigma^2=a^2; taking a>0a>0 as the options do gives a=σa=\sigma and b=μb=\mu.

10. A conditional probability for a discrete distribution

Let P(X=k)=2−(k+1)+3−kP(X=k)=2^{-(k+1)}+3^{-k}, k=1,2,…k=1,2,\ldots. Compare P(X>m+n∣X>m)P(X>m+n\mid X>m) with P(X>n)P(X>n).

discrete random variableconditional probability

Approach:

This question tests the tail probability and conditional probability of a discrete random variable. Write the conditional probability in terms of tail probabilities as a weighted average, and then compare it with the unweighted average.

Steps:

  1. First the tail probability: P(X>k)=∑j=k+1∞(2−(j+1)+3−j)=2−(k+1)+12⋅3kP(X>k)=\sum\limits_{j=k+1}^{\infty}\left(2^{-(j+1)}+3^{-j}\right)=2^{-(k+1)}+\dfrac{1}{2\cdot3^k}.
  2. Hence P(X>m+n∣X>m)=P(X>m+n)P(X>m)=2−(m+n+1)+12⋅3m+n2−(m+1)+12⋅3mP(X>m+n\mid X>m)=\dfrac{P(X>m+n)}{P(X>m)}=\dfrac{2^{-(m+n+1)}+\frac{1}{2\cdot3^{m+n}}}{2^{-(m+1)}+\frac{1}{2\cdot3^m}}.
  3. Divide numerator and denominator by 2−(m+1)2^{-(m+1)}: this gives 2−n+3−n⋅2m+13m⋅131+2m+13m⋅13\dfrac{2^{-n}+3^{-n}\cdot\dfrac{2^{m+1}}{3^m}\cdot\dfrac13}{1+\dfrac{2^{m+1}}{3^m}\cdot\dfrac13}, that is, a weighted average of 2−n2^{-n} and 3−n3^{-n}.
  4. Since the weight leans towards the larger 2−n2^{-n} (because 2−n>3−n2^{-n}>3^{-n}), this weighted average is greater than the unweighted average 12(2−n+3−n)\dfrac12(2^{-n}+3^{-n}).
  5. And P(X>n)=2−(n+1)+12⋅3n=12(2−n+3−n)P(X>n)=2^{-(n+1)}+\dfrac{1}{2\cdot3^n}=\dfrac12(2^{-n}+3^{-n}), so P(X>m+n∣X>m)>P(X>n)P(X>m+n\mid X>m)>P(X>n) — choose (D).

Answer:

D.

The tail probability is

P(X>k)=2−(k+1)+12⋅3k.P(X>k)=2^{-(k+1)}+\frac{1}{2\cdot3^k}.

The conditional probability is a weighted average of 2−n2^{-n} and 3−n3^{-n} whose weight leans towards the larger 2−n2^{-n}, and so it exceeds the corresponding unweighted average P(X>n)P(X>n).

II. Fill in the Blanks

II. Fill in the blanks: questions 11 to 16, 5 points each, 30 points in total.

  1. Let v1=(0,x,z)\boldsymbol v_1=(0,x,z), v2=(y,0,1)\boldsymbol v_2=(y,0,1) and F=v1×v2\boldsymbol F=\boldsymbol v_1\times\boldsymbol v_2. Then
div⁡F=‾.\operatorname{div}\boldsymbol F=\underline{\hspace{2em}}.
vector algebradivergence

Approach:

This question tests the cross product of vectors and the divergence. First compute the three components of F=v1×v2\boldsymbol F=\boldsymbol v_1\times\boldsymbol v_2 from the determinant, and then take the divergence.

Steps:

  1. By the definition of the cross product, F=∣ijk0xzy01∣=xi+yzj−xyk=(x, yz, −xy)\boldsymbol F=\begin{vmatrix}\bm i&\bm j&\bm k\\0&x&z\\y&0&1\end{vmatrix}=x\bm i+yz\bm j-xy\bm k=(x,\ yz,\ -xy).
  2. Hence div⁡F=∂x∂x+∂(yz)∂y+∂(−xy)∂z=1+z+0=1+z\operatorname{div}\boldsymbol F=\dfrac{\partial x}{\partial x}+\dfrac{\partial(yz)}{\partial y}+\dfrac{\partial(-xy)}{\partial z}=1+z+0=1+z.

Answer:

1+z1+z.

lim⁡x→0[1x−ln⁡(1+x)xsin⁡x]=‾.\lim_{x\to0}\left[\frac1x-\frac{\ln(1+x)}{x\sin x}\right]=\underline{\hspace{2em}}.
limitequivalent infinitesimals

Approach:

This question tests a limit involving equivalent infinitesimals. Separate out the two 1/x1/x-type principal terms, and then use the Taylor expansion to find their difference.

Steps:

  1. From ln⁡(1+x)=x−x22+O(x3)\ln(1+x)=x-\dfrac{x^2}{2}+O(x^3) and sin⁡x=x+O(x3)\sin x=x+O(x^3), we get ln⁡(1+x)xsin⁡x=x−x22+O(x3)x2+O(x4)=1x−12+O(x)\dfrac{\ln(1+x)}{x\sin x}=\dfrac{x-\frac{x^2}{2}+O(x^3)}{x^2+O(x^4)}=\dfrac1x-\dfrac12+O(x).
  2. Hence the original expression =lim⁡x→0[1x−(1x−12+O(x))]=lim⁡x→0(12+O(x))=12=\lim\limits_{x\to0}\left[\dfrac1x-\left(\dfrac1x-\dfrac12+O(x)\right)\right]=\lim\limits_{x\to0}\left(\dfrac12+O(x)\right)=\dfrac12.

Answer:

12\dfrac12.

  1. The curve determined by x=2sin⁡2tx=2\sin^2t, y=t+cos⁡ty=t+\cos t has, at t=π/4t=\pi/4,
d2ydx2∣t=π/4=‾.\left.\frac{d^2y}{dx^2}\right|_{t=\pi/4}=\underline{\hspace{2em}}.
parametric equationssecond derivative

Approach:

This question tests the second derivative of the curve determined by parametric equations. Substitute directly into d2ydx2=x′y′′−y′x′′(x′)3\dfrac{d^2y}{dx^2}=\dfrac{x'y''-y'x''}{(x')^3}.

Steps:

  1. Differentiate: x′=4sin⁡tcos⁡t=2sin⁡2tx'=4\sin t\cos t=2\sin2t and y′=1−sin⁡ty'=1-\sin t.
  2. Differentiate again: x′′=4cos⁡2tx''=4\cos2t and y′′=−cos⁡ty''=-\cos t.
  3. Substitute into the formula d2ydx2=x′y′′−y′x′′(x′)3\dfrac{d^2y}{dx^2}=\dfrac{x'y''-y'x''}{(x')^3}; at t=π4t=\dfrac\pi4 we have x′=2x'=2, y′=1−22y'=1-\dfrac{\sqrt2}{2}, x′′=0x''=0 and y′′=−22y''=-\dfrac{\sqrt2}{2}.
  4. Hence d2ydx2∣t=π/4=2⋅(−22)−023=−28\left.\dfrac{d^2y}{dx^2}\right|_{t=\pi/4}=\dfrac{2\cdot\left(-\frac{\sqrt2}{2}\right)-0}{2^3}=-\dfrac{\sqrt2}{8}.

Answer:

−28-\dfrac{\sqrt2}{8}.

∫1+∞ln⁡(1+x)x2 dx=‾.\int_1^{+\infty}\frac{\ln(1+x)}{x^2}\,dx=\underline{\hspace{2em}}.
improper integralintegration by parts

Approach:

This question tests the computation of an improper integral. Use integration by parts to replace ln⁡(1+x)\ln(1+x) with its derivative, and turn the integral into a computable partial fraction.

Steps:

  1. Take u=ln⁡(1+x)u=\ln(1+x) and dv=dxx2dv=\dfrac{dx}{x^2}, so du=dx1+xdu=\dfrac{dx}{1+x} and v=−1xv=-\dfrac1x.
  2. Integration by parts gives ∫1+∞ln⁡(1+x)x2dx=[−ln⁡(1+x)x]1+∞+∫1+∞dxx(1+x)\displaystyle\int_1^{+\infty}\dfrac{\ln(1+x)}{x^2}dx=\left[-\dfrac{\ln(1+x)}{x}\right]_1^{+\infty}+\int_1^{+\infty}\dfrac{dx}{x(1+x)}.
  3. The boundary term is ln⁡2\ln2; and 1x(1+x)=1x−11+x\dfrac{1}{x(1+x)}=\dfrac1x-\dfrac1{1+x}, so ∫1+∞dxx(1+x)=[ln⁡x−ln⁡(1+x)]1+∞=ln⁡2\displaystyle\int_1^{+\infty}\dfrac{dx}{x(1+x)}=\left[\ln x-\ln(1+x)\right]_1^{+\infty}=\ln2.
  4. Hence the original integral =ln⁡2+ln⁡2=2ln⁡2=\ln2+\ln2=2\ln2.

Answer:

2ln⁡22\ln2.

  1. For the matrices A,BA,B given in the question, let m(X)m(X) denote the largest real eigenvalue of XX; from m(A)<m(B)m(A)<m(B) it follows that
a∈‾.a\in\underline{\hspace{3em}}.
matrixeigenvalue

Approach:

This question tests comparing the largest real eigenvalues of two matrices. First find the eigenvalues of each matrix, then solve the inequality m(A)<m(B)m(A)<m(B) for the parameter aa.

Steps:

  1. For the matrix AA given in the question, find the characteristic equation and solve it for all its real eigenvalues, and denote the largest of them by m(A)m(A).
  2. Do the same for the matrix BB to get its real eigenvalues, and denote the largest by m(B)m(B).
  3. Write m(A)m(A) and m(B)m(B) as expressions in the parameter aa, and then solve the inequality m(A)<m(B)m(A)<m(B).
  4. Solving gives that the parameter must satisfy a<0a<0, that is, a∈(−∞,0)a\in(-\infty,0).

Answer:

(−∞,0)(-\infty,0).

  1. If X∼P(1)X\sim P(1) and Y∼P(3)Y\sim P(3), and XX is independent of X−YX-Y, then
E(XY)=‾.E(XY)=\underline{\hspace{2em}}.
Poisson distributioncovarianceindependence

Approach:

This question tests the moments, the covariance and the relation of independence for the Poisson distribution.

Steps:

  1. From X∼P(1)X\sim P(1) and Y∼P(3)Y\sim P(3) we have EX=1, DX=1, EY=3EX=1,\ DX=1,\ EY=3.
  2. Since XX is independent of X−YX-Y, Cov⁡(X,X−Y)=0\operatorname{Cov}(X,X-Y)=0, that is, DX−Cov⁡(X,Y)=0DX-\operatorname{Cov}(X,Y)=0, so Cov⁡(X,Y)=DX=1\operatorname{Cov}(X,Y)=DX=1.
  3. Hence E(XY)=EX⋅EY+Cov⁡(X,Y)=1⋅3+1=4E(XY)=EX\cdot EY+\operatorname{Cov}(X,Y)=1\cdot3+1=4.

Answer:

44.

III. Problems to Solve

III. Problems to solve: questions 17 to 22, 70 points in total.

17. An extremum of a function of two variables

Find the extrema of f(x,y)=(2x2−y2)exf(x,y)=(2x^2-y^2)e^x.

extremum of a function of two variablesthe second derivative test

Approach:

This question tests the extrema of a function of two variables. First solve the equations for the stationary points, and then use the Hessian determinant formed from the second partial derivatives to classify them.

Steps:

  1. Partial derivatives: fx=ex(2x2+4x−y2)f_x=e^x(2x^2+4x-y^2) and fy=−2yexf_y=-2ye^x.
  2. From fy=0f_y=0 we get y=0y=0; substituting into fx=0f_x=0 gives 2x2+4x=2x(x+2)=02x^2+4x=2x(x+2)=0, so the stationary points are (0,0)(0,0) and (−2,0)(-2,0).
  3. Second partial derivatives: fxx=ex(2x2+8x+2−y2)f_{xx}=e^x(2x^2+8x+2-y^2), fxy=−2yexf_{xy}=-2ye^x and fyy=−2exf_{yy}=-2e^x.
  4. At (0,0)(0,0) we have A=2, B=0, C=−2A=2,\ B=0,\ C=-2, and AC−B2=−4<0AC-B^2=-4<0, so it is not an extremum.
  5. At (−2,0)(-2,0) we have A=e−2(8−16+2)=−6e−2<0A=e^{-2}(8-16+2)=-6e^{-2}<0, B=0B=0 and C=−2e−2C=-2e^{-2}, and AC−B2=12e−4>0AC-B^2=12e^{-4}>0, so it is a local maximum, with maximum value f(−2,0)=8e−2f(-2,0)=8e^{-2}.

Answer:

(0,0)(0,0) is not an extremum; (−2,0)(-2,0) is a local maximum, with maximum value 8e−28e^{-2}.

18. The condition for an exact differential and a differential equation

Suppose ff is three times continuously differentiable on (0,+∞)(0,+\infty), and

dF=f(xy)x2y dx+f′′(xy)xy2 dydF=\frac{f(xy)}{x^2y}\,dx+\frac{f''(xy)}{xy^2}\,dy

is an exact differential. Prove that f′′(u)/u−f(u)/u=Cf''(u)/u-f(u)/u=C; then, given f(1)=1,f′(1)=−1,f′′(1)=0f(1)=1,f'(1)=-1,f''(1)=0, find ff.

exact differentialordinary differential equation

Approach:

This question tests the criterion for an exact differential and the solution of an ordinary differential equation with a parameter. The exact-differential condition is equivalent to the two partial derivatives being equal; first put both partial derivatives in terms of the single variable u=xyu=xy, obtain an equation for ff, integrate it, and then use the initial values to fix the constant.

Steps:

Let u=xyu=xy. The exact-differential condition gives

f′′′(u)u−f′′(u)u2−f′(u)u+f(u)u2=0.\frac{f'''(u)}u-\frac{f''(u)}{u^2}-\frac{f'(u)}u+\frac{f(u)}{u^2}=0.

Hence ddu(f′′(u)u−f(u)u)=0\dfrac{d}{du}\left(\dfrac{f''(u)}u-\dfrac{f(u)}u\right)=0, that is, f′′(u)/u−f(u)/u=Cf''(u)/u-f(u)/u=C. The initial values give C=−1C=-1, that is, f′′(u)−f(u)=−uf''(u)-f(u)=-u, and solving gives

f(u)=−eu−1+e1−u+u.f(u)=-e^{u-1}+e^{1-u}+u.

Answer:

f′′(u)/u−f(u)/u=C=−1f''(u)/u-f(u)/u=C=-1, that is, f′′(u)−f(u)=−uf''(u)-f(u)=-u, whose solution is f(u)=−eu−1+e1−u+uf(u)=-e^{u-1}+e^{1-u}+u.

19. A curve integral over an arc of an ellipse

Take, in the counterclockwise direction, the arc of the ellipse x2+3y2=1x^2+3y^2=1 from A=(−1/2,−1/2)A=(-1/2,-1/2) to B=(1/2,1/2)B=(1/2,1/2), and compute the curve integral given in the question.

curve integralGreen's theorem

Approach:

This question tests the application of Green’s theorem on a non-closed curve. Close the arc up with the straight line segment, apply Green’s theorem to the closed region, and then subtract the integral over the added segment.

Steps:

Close the curve up with the line segment from BB to AA along y=xy=x, and apply Green’s theorem to the closed region. The area of the ellipse is π/3\pi/\sqrt3, the integral over the closed curve is π/3\pi/3, and the integral over the added segment is 1/41/4, so the original integral is

I=π3−14.I=\frac\pi3-\frac14.

Answer:

I=π3−14I=\dfrac{\pi}{3}-\dfrac{1}{4}.

20. An integral and Rolle’s theorem

Let ff be differentiable and strictly increasing, with ∫−11f(x) dx=0\int_{-1}^1f(x)\,dx=0, and put a=∫01f(x) dxa=\int_0^1f(x)\,dx. Prove that a>0a>0; then, for the function FF given in the question, prove that there is a ξ∈(−1,1)\xi\in(-1,1) with F′′(ξ)=0F''(\xi)=0.

an integral inequalityRolle's theorem

Approach:

This question tests comparing the integrals of a strictly increasing function and applying Rolle’s theorem twice. For the first part, use the monotonicity to compare the two pieces of the integral; for the second part, apply Rolle’s theorem twice to the function FF constructed in the question.

Steps:

  1. (1) Since ff is strictly increasing, for the corresponding positions in x∈[−1,0]x\in[-1,0] and x∈[0,1]x\in[0,1] we have f(x)<f(0)<f(−x)f(x)<f(0)<f(-x), and hence ∫−10f(x)dx<∫01f(x)dx=a\displaystyle\int_{-1}^0f(x)dx<\int_0^1f(x)dx=a.
  2. Also ∫−11f(x)dx=∫−10f(x)dx+∫01f(x)dx=0\displaystyle\int_{-1}^1f(x)dx=\int_{-1}^0f(x)dx+\int_0^1f(x)dx=0, so 2a>02a>0, that is, a>0a>0.
  3. (2) The function FF constructed in the question satisfies F(−1)=F(1)=0F(-1)=F(1)=0 and F′(0)=0F'(0)=0.
  4. Apply Rolle’s theorem on [−1,0][-1,0] and on [0,1][0,1] to get ξ1∈(−1,0)\xi_1\in(-1,0) and ξ2∈(0,1)\xi_2\in(0,1) with F′(ξ1)=F′(ξ2)=0F'(\xi_1)=F'(\xi_2)=0.
  5. Apply Rolle’s theorem once more on [ξ1,ξ2][\xi_1,\xi_2] to get a ξ∈(ξ1,ξ2)⊂(−1,1)\xi\in(\xi_1,\xi_2)\subset(-1,1) with F′′(ξ)=0F''(\xi)=0.

Answer:

a>0a>0; and there is a ξ∈(−1,1)\xi\in(-1,1) with F′′(ξ)=0F''(\xi)=0.

21. A system of vectors and a matrix power

Given four vectors in four dimensions, let AA have them as its columns, and let GG have the first two vectors as its columns. Prove that the first two vectors form a maximal linearly independent set, find HH such that A=GHA=GH, and compute A10A^{10}.

a system of vectorsmatrix equationmatrix power

Approach:

This question tests the criterion for a maximal linearly independent set and the solution of the matrix equation GH=AGH=A. First use row operations to find the rank of the matrix and to decide that the first two columns are linearly independent, then solve the equation for HH, and finally compute A10A^{10} from A=GHA=GH.

Steps:

Elementary row operations on the four vectors give a rank of 22, and the first two columns are linearly independent, so they form a maximal linearly independent set. Solving GH=AGH=A gives

H=(101101−1−1),A10=(18990111−191010−1788).H=\begin{pmatrix}1&0&1&1\\0&1&-1&-1\end{pmatrix},\qquad A^{10}=\begin{pmatrix}1&8&9&9\\0&1&1&1\\-1&9&10&10\\-1&7&8&8\end{pmatrix}.

Answer:

The first two vectors form a maximal linearly independent set; H=(101101−1−1)H=\begin{pmatrix}1&0&1&1\\0&1&-1&-1\end{pmatrix}, and A10=(18990111−191010−1788)A^{10}=\begin{pmatrix}1&8&9&9\\0&1&1&1\\-1&9&10&10\\-1&7&8&8\end{pmatrix}.

22. Order statistics and maximum likelihood estimation for the exponential distribution

Suppose the lives of nn components are independent and follow an exponential distribution with mean θ\theta, and the experiment stops when the kk-th failure occurs. For k=1k=1, find the density of the first failure time, an unbiased estimator and its variance; then find the maximum likelihood estimator of θ\theta from observations of kk general failure times.

exponential distributionorder statisticsmaximum likelihood estimation

Approach:

This question tests the order statistics of the exponential distribution and maximum likelihood estimation. When k=1k=1, the first failure time is the minimum of nn independent exponential variables and still follows an exponential distribution; for general kk, write the likelihood function using the joint density of the order statistics.

Steps:

When k=1k=1, the first failure time TT follows an exponential distribution with parameter n/θn/\theta:

fT(t)=nθe−nt/θ,t>0.f_T(t)=\frac n\theta e^{-nt/\theta},\quad t>0.

Hence θ^=nT\widehat\theta=nT is an unbiased estimator, and D(θ^)=θ2D(\widehat\theta)=\theta^2. For general kk failure times t1≤⋯≤tkt_1\le\cdots\le t_k, the maximum likelihood estimator is

θ^=∑i=1kti+(n−k)tkk.\widehat\theta=\frac{\displaystyle\sum_{i=1}^k t_i+(n-k)t_k}{k}.

Answer:

(1) For k=1k=1, fT(t)=nθe−nt/θ (t>0)f_T(t)=\dfrac n\theta e^{-nt/\theta}\ (t>0), θ^=nT\widehat\theta=nT is an unbiased estimator of σ\sigma and D(θ^)=θ2D(\widehat\theta)=\theta^2; (2) For general kk, θ^=∑i=1kti+(n−k)tkk\widehat\theta=\dfrac{\sum\limits_{i=1}^kt_i+(n-k)t_k}{k}.

Found an issue with this content?

Tell us what could be improved.