6
$\begingroup$

Often I am dealing with an integral of let's say:

$$\int\frac{dt}{(t-2)(t+3)}$$

or

$$\int \frac{dt}{t(t-4)}$$

or to make this a more general case in which I am interested the most:

$$\int \frac{dt}{(t+\alpha)(t+\beta)} \quad \quad \alpha, \beta \in \mathbb{R}$$

Basically any integral with decomposed polynomial of second degree in the denominator. Obviously this leads to sum of two natural logarithms.

What I do is write down partial fractions equation and then solve system of linear equations (2 variables here):

$$\frac{1}{(t+\alpha)(t+\beta)} = \frac{A}{t+\alpha} + \frac{B}{t+\beta}$$

After solving this I end up with some $A, B$ coefficients and I can solve the integral.

Is there faster way to find $A, B$ ? Some algorithm or anything that I can follow and would always work for such case? Surely, solving that does not take much time but I just wonder if it could be done even faster.

(bonus question: what if there were more variables, like 3 variables?)

I would greatly appreciate all feedback as it could help me save countless number of minutes in the future.

$\endgroup$
  • $\begingroup$ Also when I think about it, I wonder if there's even a general method (beyond the one you referred to). I can see very easily a general solution for when (i) the numerator is $1$ (ii) $\alpha \ne \beta$ (iii) we only have a product of two factors in the denominator, but that's such a narrow set of circumstances that I wonder if it's worth it. Maybe someone has a more general form though. $\endgroup$ – Eevee Trainer Apr 12 at 1:25
6
$\begingroup$

Here's your answer for general $n$.

$\dfrac1{\prod_{k=1}^n (x-a_k)} =\sum_{k=1}^n \dfrac{b_k}{x-a_k} $.

Therefore $1 =\sum_{k=1}^n \dfrac{b_k\prod_{j=1}^n (x-a_j)}{x-a_k} =\sum_{k=1}^n b_k\prod_{j=1, j\ne k}^n (x-a_j) $.

Setting $x = a_i$ for each $i$, all the terms except the one with $k=i$ have the factor $a_i-a_i$, so $1 = b_i\prod_{j=1, j\ne i}^n (a_i-a_j) $ so that $b_i =\dfrac1{\prod_{j=1, j\ne i}^n (a_i-a_j)} $.

For $n=2$, $b_1 =\dfrac1{a_1-a_2} $, $b_2 =\dfrac1{a_2-a_1} $.

For $n=3$, $b_1 =\dfrac1{(a_1-a_2)(a_1-a_3)} $, $b_2 =\dfrac1{(a_2-a_1)(a_2-a_3)} $, $b_3 =\dfrac1{(a_3-a_1)(a_3-a_2)} $.

$\endgroup$
8
$\begingroup$

If your fraction is in form of $$ \frac {1}{(t-\alpha_1)(t-\alpha_2)(t-\alpha_3)...(t-\alpha_k)}$$ where the $\alpha s$ are different, there is an easy way to find the partial fraction decomposition.

Let $t=\alpha_i$ and cover $(t-\alpha_i)$ while evaluating the above fraction to get your $A_i$

For example $$ \frac{1}{(t-1)(t-3)(t+4)} = \frac {A_1}{(t-1)} + \frac{ A_2}{(t-3)} +\frac { A_3}{(t+4)}$$ Where $$A_1 = \frac {1}{(1-3)(1+4)}=\frac{-1}{10}$$ $$A_2 = \frac {1}{(3-1)(3+4)} = \frac{1}{14}$$

$$A_3=\frac {1}{(-4-1)(-4-3)}=\frac{1}{35}$$

$\endgroup$
7
$\begingroup$

Unless I am misunderstanding the question, by "root rotation" (when $\beta \neq \alpha$):

$$\frac{1}{(t + \alpha)(t + \beta)} = \frac{A}{t + \alpha} + \frac{B}{t + \beta}$$

$$1 = A(t + \beta) + B(t + \alpha)$$

Evaluating $-\beta$ for $t$:

$$1 = B(\alpha - \beta)$$

$$B = \frac{1}{\alpha - \beta}$$

Similarly, for $A$, sub in $-\alpha$:

$$1 = A(\beta - \alpha)$$

$$A = \frac{1}{\beta - \alpha}$$

$\endgroup$
  • $\begingroup$ I'll be coming back to this post. This is what I was looking for. $\endgroup$ – weno Apr 12 at 1:45
  • $\begingroup$ Unless that takes you too much time, would you be able to derive formulas for $A, B, C$, please? Function of form $\frac{1}{(t+\alpha)(t+\beta)(t+\gamma)}$ and $\alpha \neq \beta \neq \gamma$ $\endgroup$ – weno Apr 12 at 1:47
1
$\begingroup$

Let $n$,$m$ be integers such that $n\ge 2$ and $n-1 \ge m \ge 0$ .Then if the roots of the polynomial in question are all distinct the following formula holds: \begin{equation} \frac{x^m}{\prod\limits_{j=1}^n (x-b_j)} = \sum\limits_{i=1}^n \frac{1}{x-b_i} \cdot \frac{b_i^m}{\prod\limits_{j=1,j\neq i}^n (b_i-b_j)} \end{equation} Now, if the roots are not distinct then it gets more complicated but even at the outset it is clear that we can generate a similar formula by differentiating both sides with respect to the particular root a certain number of times. For example if $n=2$ and $m_1 \ge 1$, $m_2 \ge 1$ and $m\ge 0$ the following formula holds true: \begin{eqnarray} \frac{x^m}{(x-b_1)^{m_1} (x-b_2)^{m_2}}=\sum\limits_{j=0}^m \left( \sum\limits_{l_1=1}^{m_1} \binom{m_1+m_2-1-l_1}{m_2-1}(-1)^{m_2} b_1^j \frac{1_{l_1=1} \binom{m-1}{j-1} + 1_{l_1>1} \binom{m}{j}}{(x-b_1)^{l_1-m+j} (-b_1+b_2)^{m_1+m_2-l_1}}+ \sum\limits_{l_1=1}^{m_2} \binom{m_1+m_2-1-l_1}{m_1-1}(-1)^{m_1} b_2^j \frac{1_{l_1=1} \binom{m-1}{j-1} + 1_{l_1>1} \binom{m}{j}}{(x-b_2)^{l_1-m+j} (-b_2+b_1)^{m_1+m_2-l_1}} \right) \end{eqnarray}

In[5732]:= ll = {}; x =.; b1 =.; b2 =.; m = RandomInteger[{0, 10}];
For[count = 1, count <= 100, count++,
  {m1, m2} = RandomInteger[{1, 5}, 2]; x =.; b1 =.; b2 =.;
  xx1 =
      Sum[ 
     Binomial[m1 + m2 - 1 - l1, m2 - 1] ((-1)^
       m2 b1^j If[l1 == 1, Binomial[m - 1, j - 1], 
        Binomial[m, j]])/((x - b1)^(l1 - m + j) (-b1 + b2)^(
       m1 + m2 - l1)), {l1, 1, m1}, {j, 0, m}] + 
    Sum[ Binomial[m1 + m2 - 1 - l1, m1 - 1] ((-1)^
       m1 b2^j If[l1 == 1, Binomial[m - 1, j - 1], 
        Binomial[m, j]])/((x - b2)^(l1 - m + j) (-b2 + b1)^(
       m1 + m2 - l1)), {l1, 1, m2}, {j, 0, m}];
  xx2 = x^m/((x - b1)^m1 (x - b2)^m2);
  ll = Join[ll, {Simplify[xx1 - xx2]}];
  ];
ll

Out[5734]= {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, \
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, \
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, \
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, \
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}

It would be interesting to derive similar formulae in the case $n > 2$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.