I tried to do convolution in R directly and using FFTs then taking inverse. But it seems from simple observation it is not correct. Look at this example:

# DIRECTLY
> x2$xt
[1] 24.610 24.605 24.610 24.605 24.610
> h2$xt
[1] 0.003891051 0.003875910 0.003860829 0.003845806 0.003830842
> convolve(h2$xt,x2$xt)
[1] 0.4750436 0.4750438 0.4750435 0.4750437 0.4750435

# USING INVERSE FOURIER TRANSFORM
> f=fft(fft(h2$xt)*fft(x2$xt), inv=TRUE)
> Re(f)/length(f) 
[1] 0.4750438 0.4750435 0.4750437 0.4750435 0.4750436
>

Lets take the index 0. At 0, the convolution should simply be the last value of x2$xt (24.610) multiplied by first value of h2$xt (0.003891051) which should give convolution at index 0 = 24.610*0.003891051 = 0.09575877 which is way off from 0.4750436.

Am I doing something wrong? Why is the values so different from expected?