0
$\begingroup$

So I want to obtain the following form: For $x, p \in R^n$

$$\nabla f(x+p) = \nabla f(x) + \int^1_0 \nabla^2 f(x+tp)p \,dt$$ for $f$ twice continuously differentiable and $t \in (0,1)$.

Taylor's theorem in integral form is:

$$f(x) = f(a) + \int^x_a \nabla f(t) (x-t)\,dt$$ from here: https://www.math.upenn.edu/~kazdan/361F15/Notes/Taylor-integral.pdf

My approach is:

$$f(x+p) = f(x) + \int^{x+p}_x \nabla f(t) (x+p-t)\,dt$$ Reformulated as: $$f(x+p) = f(x) + \int^1_0 \nabla f(x+tp)(1-t)p\,dt$$ $$= f(x) + \int^1_0 \nabla f(x+tp) p \,dt - \int^1_0 \nabla f(x+tp) t p \,dt$$

But I am not sure how to take the derivatives of the integrals with respect to x; am I at a dead end? Should I instead, just begin with $\nabla f(x)$ in place of $f(x)$ when I apply Taylor's theorem?

$\endgroup$
7
  • $\begingroup$ Why do you write $\nabla f$? $\endgroup$ Commented Jul 9, 2016 at 22:28
  • $\begingroup$ It's meant to be the gradient of $f$. Am I writing the notation wrong? Please let me know. $\endgroup$ Commented Jul 9, 2016 at 22:31
  • $\begingroup$ Are we on the real line? $\endgroup$ Commented Jul 9, 2016 at 22:32
  • $\begingroup$ Oh sorry. $x$ and $p$ are in $R^n$. My bad. I will specify that next time. $\endgroup$ Commented Jul 9, 2016 at 22:37
  • $\begingroup$ Your reference specifies the real line however. If you are in higher dimensions then your approach is not correct. It is correct if $\nabla f$ means $f'$ on the line. $\endgroup$ Commented Jul 9, 2016 at 22:38

1 Answer 1

-1
$\begingroup$

Okay, I figured out the problem. Taylor's theorem in integral form is actually $$f(x) = f(a) + \int^x_a \nabla f(t) \,dt$$

From there, we have $$f(x+p) = f(x) + \int_0^1 \nabla f(x+tp)p\,dt$$

And we can substitute $\nabla f(x)$ for f(x) into the above statement.

EDIT:

The above is incorrect. Using https://www.math.washington.edu/~folland/Math425/taylor2.pdf (page 3), (with the same notation regarding $\alpha$)

$$f(x+p) = f(x) + \sum_{|\alpha| = 1} \frac{p^{\alpha}}{\alpha!} \int_0^1 \partial^{\alpha} f(x+pt) \,dt $$

$$f(x+p) = f(x) + \sum_{j=1}^n p_j \int_0^1 \partial_j f(x+pt) \,dt = f(x) + p \cdot \int_0^1 \nabla f(x+pt)\,dt $$

Taking the gradient of the above: $$\frac{\partial}{\partial_i} f(x+p) = \frac{\partial}{\partial_i} f(x) + \frac{\partial}{\partial_i} p \cdot \int_0^1 \nabla f(x+pt) \, dt = \frac{\partial}{\partial_i} f(x) + p \cdot \int^1_0 \frac{\partial}{\partial_i} \nabla f(x+pt) \,dt $$

We have overall that

$$\nabla f(x+p) = \nabla f(x) + \int^1_0 \nabla^2 f(x+pt) p \, dt$$

$\endgroup$
0

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.